[{"content":"The second of four instalments. In the first I explained who these numbers describe: one seventh of registered riders, those with at least one top-five finish. Here I answer the question that started the study: at what age do race results say something about the future?\nWhen I began, I was convinced there would be nothing to see at thirteen.\nThe literature suggested as much, because at younger ages performance is hard to separate from maturation. Common sense said the same: a thirteen-year-old winning an Esordienti race, Italy\u0026rsquo;s category for thirteen- and fourteen-year-olds, or Under 15s in the literature, may be beating peers a year behind in puberty. Experience agreed. I could think of a long succession of champions at that age who were never heard of again.\nI expected to find zero, and to be able to say so precisely. That is not what happened.\nLook before modelling The simplest answer needs no statistics. Take the youngsters who later became professionals, see where they ranked at thirteen, and compare them with everyone else.\nPlacings are expressed as percentiles: 100 means top of the ranking, 50 the middle and 0 the bottom. This makes seasons and categories with different numbers of riders comparable.\nat thirteen (first-year Esordienti) typical position riders who will not become professionals 49th percentile future professionals 81st percentile ‘Typical position’ is the median, with half the group above and half below. Professional means having raced for a first- or second-tier team by twenty-five.\nAt eighteen the gap widens further, from 47 to 94.\nThere is an elegant way to summarise how far apart two groups are. Take every possible pair consisting of a future professional and a future non-professional, and count how often the professional ranks ahead. At thirteen it happens in 74% of cases, at eighteen in 89%1. On conventional scales, the separation at thirteen lies just above the boundary between medium and large. It is not an enormous signal, but it is very far from the nothing I expected.\nThe curves show group medians. Group membership is known only in retrospect: at thirteen nobody knew who was who.\nThat 74% needs two qualifications, because it is easy to misread.\nFirst, the comparison is among riders ranked that year, not among all youngsters. Anyone who never scored a point at thirteen is absent from both the professional and non-professional groups. The figure describes separation within the ranking; it says nothing about everyone else.\nSecond, the professionals in this comparison are future professionals: youngsters whose outcomes we know now, but who did not have them written on their foreheads at thirteen. It is a retrospective comparison.\nNor can 74% and 89% be compared directly, because they concern different people. At thirteen the calculation covers everyone ranked then; at eighteen it covers only those still ranked, a much smaller, already selected group. A clean comparison using the same people at every age comes later.\nMore than a comparison between two groups There is another way to examine the same data. Instead of dividing riders into those who made it and those who did not, I split them into four groups by how far they went: non-professionals; professionals who never entered the world\u0026rsquo;s top five hundred; those who reached that group; and those who reached the top hundred of the PCS Ranking on procyclingstats.com.\nAt thirteen, the medians for the first three groups are 49, 69 and 891. An ordered gradient emerges: the further someone eventually went, the higher they already ranked at thirteen. Random noise does not produce that sort of staircase, which is the main reason I do not think the 74% is an artefact.\nThe final step breaks the pattern, however. Those who reached the world\u0026rsquo;s top hundred had a median of 84 at thirteen, below the previous group. There are six riders, a very small sample in which a median can do almost anything.\nFor more detail in the technical document, in Italian: The gradient by level reached.\nIs it performance, or date of birth? This is the most reasonable doubt to raise, and it needs addressing.\nWithin a birth year, someone born in January has up to twelve months\u0026rsquo; extra development over someone born in December. The data show it clearly: in the Esordienti ranking, first-quarter births are 2.13 times as common as last-quarter births, allowing for Italy\u0026rsquo;s actual birth distribution2. If a placing at thirteen mostly measured early development, the 74% of pairs in which a future professional ranks ahead would mainly reflect that.\nTo test it, I rerun the model including relative age, measured in days between birth and year end, and see what happens to the placing effect. Nothing happens. The odds multiplier for ten percentile points is 1.40 before and 1.40 after; discrimination moves from 0.736 to 0.738, a change in the third decimal place. Relative age as the sole predictor reaches 0.513. Essentially a coin toss.\nDoes being born in January help, then? Very much, but earlier, in entering the ranking: that is what the two-to-one ratio between first and last quarters tells us. Among riders already ranked, birth date no longer distinguishes who will make it. Among professionals the ratio falls to 1.5, against 1.7 among all ranked riders, a gap indistinguishable from chance with 77 athletes. It is an advantage for entering rankings when young, not a measure of talent.\nMore Italian births occur between May and September, so the expected share is not 25% per quarter. The figure adjusts for actual birth seasonality.\nWomen\u0026rsquo;s cycling offers an independent way to test that explanation. It is one of the few questions in this series I can answer in exactly the same way for girls and boys. Prediction questions require knowing who made it, and there is no comparable female outcome, for reasons explained in the fourth instalment (forthcoming); here only birth dates are needed. Girls mature earlier, and by thirteen many have already passed through puberty, while developmental differences among boys born in January and December are at their greatest. If the advantage comes from maturation, it should be weaker among girls.\nIt is: at thirteen, using the same birth cohorts and demographic expectations, first-quarter births are 1.98 times as common as last-quarter births among boys and 1.51 times among girls. The difference between sexes passes a statistical test (p = 0.007)2. The male figure differs from the earlier 2.13 because this comparison includes only birth cohorts also observed among girls: 5,544 boys and 874 girls.\nAfter fourteen, however, the lines stop resembling each other. That part does not fit. Among boys the imbalance decreases category by category. Among girls it falls faster, becoming indistinguishable from demographic expectations among Allieve (1.20, p = 0.14), before rising among juniors to 1.42, once again distinguishable. With 681 and 370 riders, those estimates have wide intervals, and I would build nothing on them. Only the thirteen-year-old comparison supports reasonably confident conclusions: 874 girls and a clear difference from boys.\nAt thirteen the imbalance is present among girls but weaker than among boys. Later female estimates rest on a few hundred riders and follow no consistent trajectory.\nThis is indirect corroboration: men\u0026rsquo;s and women\u0026rsquo;s cycling differ in many ways beyond the timing of puberty. Nevertheless, the female population provides a check on a male finding: maturation differences explain why a January birthday matters, and why it later stops mattering.\nExactly how much the ranking matters A model lets us put a number on the advantage: what difference do ten percentile points make3?\ncategory age odds multiplier first-year Esordienti 13 ×1.40 (1.25–1.57) second-year Esordienti 14 ×1.56 first-year Allievi 15 ×1.70 second-year Allievi 16 ×1.98 first-year juniors 17 ×1.64 second-year juniors 18 ×2.28 (1.91–2.80) first-year Under 23 19 ×1.32 Each row is a separate model, with a 95% interval on a logarithmic scale: odds ratios are read as ratios, not differences.\nHow should you read this table? The multiplier applies to odds, not probability. Of two Esordienti ten percentile points apart, the higher-ranked rider has 40% higher odds of becoming professional: the ratio of their probability of making it to their probability of not making it is 1.40 times the other\u0026rsquo;s. With an outcome this rare, probability also rises by almost 40%. But that does not mean forty percentage points: in practice it rises from about 2.7% to about 3.7%, one point higher.\nThe effect grows with age, from 1.40 to 2.28, but not steadily. The two downward steps should not be read as weaker signals: the population changes at each category transition. Among first-year juniors, future professionals already make up 9.7% of the list, against 4.5% in the preceding cell; among Under 23s they are 37%. Comparing these figures as though they measured the same thing is already a misreading.\nFor more detail in the technical document, in Italian: Why the ‘% pro’ column is not a signal.\nAnother detail looks like a finding but is not. In categories with a single list, future professionals make up a larger share in the first year than the second, as though the first year selected better. As explained in the first instalment, however, first-year point-scoring places are scarce, with an average of 181 ranked riders against 320. Those present are already more selected, and a more selected group necessarily contains a larger share of future professionals. Does the first year predict better? Only comparing the same people can answer that, and the answer is no: the second year discriminates better in every category, 81% against 70% for Esordienti, 85% against 81% for Allievi and 87% against 78% for juniors.\nWhat more recent results add Allievi placings help predict future professionals. But how much do they add to what Esordienti results already told us?\nTake the 102 riders ranked in every category. They form a small, highly selected group, but allow predictions to be compared on the same athletes. Begin with birth year alone, then add results from Esordienti, Allievi, juniors and finally Under 23s, measuring the predictive gain at each step4.\nthe model knows… discrimination interval birth year alone 51% (essentially nothing) 40–62% plus Esordienti 58% 46–69% plus Allievi 66% 55–77% plus juniors 81% 73–89% plus Under 23 82% 74–90% The comparison works because it always uses the same 102 riders. Changing groups at each step would measure who remained, not added information.\nThe third column, sometimes dispensable, matters here: intervals are sixteen to twenty-three points wide, and the first three overlap extensively. Only one step clearly stands apart, the junior one. Juniors alone add as much as all earlier categories combined, and this is the only step whose improvement is distinguishable from chance.\nDo not turn non-significant steps into zeros, however. Esordienti and Allievi improve model fit beyond what chance would explain; they simply do not change the order in which riders are ranked. These are different questions: one asks how well the model agrees with the data, the other whether it puts the right rider first in a pair.\nThat might have been peculiar to this subsample, so I asked the same question in two other ways. A random forest, an algorithm that finds useful combinations for itself, received seventeen variables instead of the simple model\u0026rsquo;s two and gained 2.0 percentage points of discrimination. Junior placing ranked highest in importance. A penalised regression, putting all categories in one model and discarding those that do not earn their place, retained 2 out of 8: second-year juniors and Under 23.\nThree different routes, the same conclusion: almost all useful information lies in the latest measurement. Earlier seasons do not simply add to it; they largely show the same thing from further away.\nFor more detail in the technical document, in Italian: How similar the categories are, Would a more complicated model do better? and What if all categories were used together?.\nWhat if unranked riders counted too? This study counts only ranked riders. An alternative is to treat absence as worse performance than any ranked placing. I tried that too.\nThe answer depends on age in a way I did not expect. At eighteen, putting absent riders at the bottom raises discrimination from 89% to 94%. By then absence is a strong signal: only 3 of the 77 future professionals were unranked. At thirteen it lowers discrimination from 74% to 69%, because only 59 of the 77 were ranked then; the other eighteen would be placed at the bottom despite eventually making it5.\nAbsence is information, but only later. This is a more precise version of a point from the first instalment: a name missing from a ranking has not necessarily been rejected, and almost a third of those who disappear return.\nFor more detail in the technical document, in Italian: Do the conclusions depend on design choices?.\nWhat remains Results at thirteen carry information. Anyone denying that out of caution is saying something kind but false. When you look at your Esordienti\u0026rsquo;s placings, you are looking at something associated, on average, with their future within the observed population.\nI would not read that as ‘the literature was wrong’. Earlier studies suggested a weak signal at young ages, and both findings can coexist: this sample is Italian, large and not restricted to internationally competitive riders, precisely the kind of population in which an early signal has room to appear.\nAnd birth date does not explain it. Being born in January helps riders enter the ranking, not become professionals: selecting at thirteen partly means selecting birth dates, to no useful end.\nWhat does this mean in practice? For predicting professional careers, recent results matter far more than results three years earlier. Past results may answer another question, though: is the rider improving or declining? That is the subject of the next instalment (forthcoming).\nHow we know\nThe predictor is the percentile within each season × category × year within category cell, making rankings of different lengths comparable. The outcome is having raced for a first- or second-tier professional team by twenty-five.\nModels are logistic regressions with Firth\u0026rsquo;s correction, needed because the outcome is rare, below 3%, and uncorrected estimates would be biased upwards. They adjust for birth year because recent cohorts have had less time to make it; the adjustment changes coefficients by less than 0.01.\nPair percentages express the area under the ROC curve in plain language: the share of future-professional/non-professional pairs in which the model ranks the right rider ahead.\nCalculated in report/moduli/punteggi.py.\u0026#160;\u0026#x21a9;\u0026#xfe0e;\u0026#160;\u0026#x21a9;\u0026#xfe0e;\nCalculated in report/moduli/rae.py. Expected births come from Eurostat table demo_fmonth.\u0026#160;\u0026#x21a9;\u0026#xfe0e;\u0026#160;\u0026#x21a9;\u0026#xfe0e;\nModels in R/16_univariati.R, rendered by report/moduli/univariati.py.\u0026#160;\u0026#x21a9;\u0026#xfe0e;\nModels in R/18_annidati.R. Random forest in R/27_confronto_ml.R, penalised regression in R/17_penalizzato.R.\u0026#160;\u0026#x21a9;\u0026#xfe0e;\nSensitivity analysis in scripts/10_sensibilita.py.\u0026#160;\u0026#x21a9;\u0026#xfe0e;\n","permalink":"https://lucabontempi.com/en/blog/ciclismo-giovanile-a-che-eta-si-riconosce-un-futuro-professionista/","summary":"\u003cp\u003e\u003cem\u003eThe second of four instalments. In the \u003ca href=\"/en/blog/ciclismo-giovanile-di-chi-stiamo-parlando/\"\u003efirst\u003c/a\u003e I explained who these numbers describe: one seventh of registered riders, those with at least one top-five finish. Here I answer the question that started the study: at what age do race results say something about the future?\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eWhen I began, I was convinced there would be nothing to see at thirteen.\u003c/p\u003e\n\u003cp\u003eThe literature suggested as much, because at younger ages performance is hard to separate from maturation. Common sense said the same: a thirteen-year-old winning an Esordienti race, Italy\u0026rsquo;s category for thirteen- and fourteen-year-olds, or Under 15s in the literature, may be beating peers a year behind in puberty. Experience agreed. I could think of a long succession of champions at that age who were never heard of again.\u003c/p\u003e","title":"At what age can you recognise a future professional cyclist?"},{"content":"This series looks at Italian youth cycling rankings and the transition to professional racing. Before presenting my numbers, I explain what research already knows and, above all, who the numbers in the following instalments describe. That is where almost all the mistakes lie.\nTwenty years of research in a few paragraphs At the junior road World Championships each nation can enter at most six riders, and only the highest-ranked nations get all six places. Juniors, or Under 19s, are seventeen- and eighteen-year-olds. Anyone who has raced at that level remembers how hard it is to earn selection. Reaching that start line means passing every selection process youth cycling can devise. You could hardly look more destined for success.\nYet even winning that race guarantees nothing. Since the junior road World Championships began, only three riders have won both the junior and elite titles: Greg LeMond, junior champion in 1979 and elite champion in 1983 and 1989; Mathieu van der Poel, junior champion in 2013 and elite champion in 2023; and Remco Evenepoel. He is the rider on the podium in the cover photo, taken at the 2018 junior Worlds in Innsbruck, which he won. His elite world title would follow four years later, in 2022.\nIn 2006 a German research group1 collected results from major competitions across twenty-five years of cycling: 27,454 results from 8,004 athletes in 108 countries, between 1980 and 2004, including disciplines beyond road racing. They then examined the same story in both directions. Looking forwards: how many junior Worlds riders subsequently contested a major senior event, meaning one of the races the study includes, from the Worlds to the Grand Tours? Thirty-four in a hundred. Looking backwards: among athletes with results in major elite events, how many had competed at the junior Worlds? 29.4 in a hundred.\nThe percentages answer opposite questions, and neither goes much beyond a third: the majority is left out whichever direction you look.\nThese figures answer opposite questions and measure different things. Together, though, they say something uncomfortable: being fast as a youngster is neither essential for reaching the top nor remotely enough to guarantee it. Two in three apparent prodigies will never contest a major senior event, and seven in ten of those who reach those events never rode that junior championship. The question has since been revisited in Norway, Belgium, Spain, the Netherlands and twice in Italy, with very different samples and remarkably similar conclusions2. There are five main findings.\nFuture professionals were already performing better than their peers as youngsters. This is the most replicated finding in the field, reached by different groups using different methods.\nThe signal grows as riders get closer to the destination. The most cited Italian study, covering 1,345 riders in federation rankings, identifies the first Under 23 season, at nineteen, as the most informative point.\nBelow fifteen, by contrast, there may be little to see. A Belgian study of thirteen- and fourteen-year-olds measured maturation, coordination and motor skills. Together these explained just over a fifth of the differences in racing performance two or three years later. The interpretation was that at that age we are mostly measuring who developed earlier.\nThen there are laboratory tests, where the finding sounds odd: threshold and watts per kilogram describe a youngster\u0026rsquo;s current level well, but add little about where they will end up. One exception is a Spanish study of 65 Under 23 riders, in which an eight-minute time trial separates future professionals fairly well. How well? Of those the test identified as future professionals, 54 per cent actually made it. Just over half, among riders around twenty who had already been heavily selected.\nFinally, birth month: youth populations contain more riders born early in the year than demographics would predict, particularly at younger ages, while the effect is absent among those who make it. A study of almost three thousand Continental-team riders finds it mainly among those who did not succeed. In the second instalment I repeat that calculation on my own data, because it is the most reasonable doubt to raise about a finding at thirteen.\nOne finding recurs everywhere: the overwhelming majority of successful youngsters still do not make it. Every number needs its denominator attached; lining them up as though they measured the same thing is misleading. Two in three junior Worlds riders never contest a major elite event. Fifteen per cent of the top ten in Italy\u0026rsquo;s Allievi ranking, for fifteen- and sixteen-year-olds, subsequently scored WorldTour points, at the sport\u0026rsquo;s highest level. Five out of forty-eight riders in a complete Dutch year group reached a Continental team, the third UCI tier, or higher2. Both the starting populations, selected riders, the top ten or everyone, and the destinations differ. The scale changes each time; the direction stays the same.\nTwo gaps remain, and this series tries to fill them. Nobody has examined thirteen- and fourteen-year-olds using race results from a large population without prior selection: the Italian studies begin with Allievi, while the Belgian study includes Under 15s only if they had already achieved a top-ten finish. And no race-results study has reported how many of the youngsters identified actually make it. Yet that is precisely the number you would use to make a decision.\nThere are also two things the studies never disclose. I found neither in any of the twenty-two studies I summarised2, and noticed them while trying to reproduce the calculations.\nThe first is how many riders there were to begin with. Worlds selections, the top ten and ranked youngsters are already products of selection; nobody states how many came before. Without that number, percentages from two studies cannot be compared: they rest on different populations.\nThe second is that a youngster who stops appearing in rankings is often counted as having stopped racing. That is how the funnels in this literature are built: anyone absent from the next category is counted as lost. Nobody has checked how many disappearances are riders still racing without scoring points, or returning the following year. Later in this instalment I measure how many return. It is not a small number.\nI cannot criticise this from above, because both points apply fully to my own study. My numbers also begin with an already filtered group, and I too see youngsters disappear from rankings without knowing whether they stopped racing or simply became slower. The only difference I can claim is that I measured these limitations rather than leaving them implicit. That is what the rest of this instalment is about.\nThe ranking behind everything The study starts with the ranking published annually for every youth category by ciclismo.info. For 2007–2025 it contains 28,037 rider-season ranking entries for 11,094 boys3. Boys rather than girls, because of the numbers: female ranked riders number between a seventh and a tenth of male riders, women\u0026rsquo;s professional divisions have existed only since 2020, and there is no women\u0026rsquo;s Under 23 ranking, the most informative category among men. Where the question does not require knowing who made it, however, girls are included.\nciclismo.info is the most complete archive of Italian youth cycling available. One clarification is necessary: it is not the federation\u0026rsquo;s official archive, but an independent portal collecting and organising results. Nor does it include every young Italian cyclist. To enter, you need at least one point, and only the first five in each race score: five for the winner, down to one for fifth4.\nHow much does that filter matter? We can measure it, because the federation publishes membership totals by category each year5. The following figures are averages across several seasons.\ncategory registered riders per season ranked share Esordienti 3,214 497 15.5% Allievi 2,673 371 13.9% Juniors 1,687 262 15.5% Under 23 1,060 162 15.3% The bar is the seasonal average; the line spans the lowest and highest seasons. Membership includes riders in other disciplines who cannot appear in a road ranking, so one in seven is a lower bound.\nAbout one registered rider in seven appears in the ranking. More striking than the figure is its stability: four categories, five to seven seasons each, always the same order of magnitude, ranging from 12.2% to 17.2%. The other six in seven race without scoring points, or compete in disciplines other than road cycling. Membership is by age category, not discipline: an Esordiente racing only mountain bikes is included in the total despite being unable to appear in a road ranking.\nOne in seven underestimates the share of road cyclists who become ranked, because the denominator contains riders who cannot enter the numerator. Even so, the ranking shows only the tip of youth cycling, and every number in the next three instalments describes that tip.\nThe cohorts, and the number I will keep using The main cohorts, groups born in the same years whom I follow over time, are riders born between 1996 and 2000. All have had time to make it or not. There are 2,813 riders and 77 professionals. Throughout this series, professional means having raced for a first-tier WorldTeam or second-tier ProTeam by age twenty-five. The age window makes birth cohorts comparable; anyone who makes it later is not counted here.\nOf those 2,813, however, only 2,183 were already ranked as Esordienti, the thirteen- and fourteen-year-old category, or Under 15s in the literature. Under 23 spans four birth years, from nineteen to twenty-two. The funnel starts with Esordienti6.\ncategory ranked riders of whom ranked as Esordienti share of the 2,183 Esordienti 2,183 2,183 100% Allievi 1,741 1,303 59.7% Juniors 1,051 745 34.1% Under 23 342 224 10.3% professionals 77 66 3.0% The two middle columns differ because each category also gains riders who were not ranked as Esordienti. The percentage uses the third column, not the second.\nThe first step already consists of riders with at least one top-five finish at thirteen; there are many more young Italian cyclists. The fall to the final step is thirty-threefold, far less than three orders of magnitude.\nLook at the final row. The cohorts produced 77 professionals, but only 66 were already ranked as Esordienti; the other 11 entered later. Dividing 77 by 2,183 would put people in the numerator who are absent from the denominator. The share uses the sixty-six who actually started there: three in a hundred.\nRead backwards, this is the most useful number in this instalment: one in seven future professionals was not ranked at thirteen. When looking at an Esordienti list knowing Italy\u0026rsquo;s best are on it, remember that one in seven of those who will make it is not there yet.\nThose eleven include Christian Scaroni, a stage winner at the 2025 Giro d\u0026rsquo;Italia, and Jonathan Milan, twice winner of the Giro\u0026rsquo;s ciclamino jersey and winner of the 2025 Tour\u0026rsquo;s green jersey. Both first appear in our rankings as second-year Allievi.\nIt looks like a funnel, and partly is. But one feature defeats simple readings: it is not the same group shrinking. Between a quarter and a third of riders in each category never appeared as Esordienti; among Under 23s it is more than a third. Reading the funnel as ‘three hundred of those two thousand remained’ counts people as survivors who were never at the start.\nThe number I will use throughout the series therefore carries its label: twenty-seven per thousand among riders ever ranked, or 77 out of 2,813. The table\u0026rsquo;s three in a hundred is a different fraction: 66 out of 2,183, only those who started as Esordienti.\nThree things to know before reading a ranking First: disappearing from a ranking does not mean quitting. Some 30.6% of riders miss at least one season and then reappear; 6.7% of all riders return after at least two absent seasons.6 A return proves that the absence was not permanent withdrawal. Alongside the cohorts\u0026rsquo; 77 professionals are 108 riders still scoring points after twenty-two who never became professionals, more than the professionals themselves. Not becoming professional does not necessarily mean quitting. This matters in the second instalment, when absence from a ranking becomes information to handle carefully.\nFor more detail in the technical document, in Italian: Leaving the ranking is not quitting, When riders quit and Is moving category a break?.\nSecond: how the categories work. For Esordienti the source publishes two separate rankings, one per birth year: first- and second-year riders have separate races and do not compete against each other. From Allievi onwards there is one ranking, with both birth years racing together. The consequence appears in the point-scoring places, the five finishes in each race that earn points7.\ncategory share of places taken by first-year riders Esordienti, separate rankings 49.7% Allievi, single ranking 26.7% Juniors, single ranking 33.3% Under 23, single ranking 11.8% The juniors\u0026rsquo; 33.3% interrupts the decline, and I cannot say why: first-year riders form a larger share of that category than they do among Allievi.\nThe Esordienti\u0026rsquo;s 49.7% is what two separate rankings would lead us to expect. What matters is the share left to first-year riders when there is only one list.\nWith separate rankings, an even split between birth years is unsurprising, and holds for girls too: each races against her own year group. The interesting question is how much younger riders collect when there is one ranking and everyone races together. Just over a quarter among Allievi, a third among juniors, an eighth among Under 23s. The benchmark is equality: half with two birth years, a quarter in Under 23, which has four. Relative to an equal division, first-year riders collect roughly half their share among Allievi and Under 23s and two thirds among juniors. Top-five places do not disappear: older riders take them. Reading a fifteen-year-old first-year Allievo\u0026rsquo;s ranking means looking at someone competing against opponents with an extra year of development.\nYet some already stand out against older riders: in 2011, Filippo Ganna collected 65 points and six wins as a first-year Allievo, finishing seventh in the combined ranking. He would go on to become twice world time-trial champion.\nWomen\u0026rsquo;s cycling also supports this explanation, because the source changed its structure along the way. Until 2021 female Esordienti had a single ranking, in which first-year riders took 28.5% of the places, similar to the men\u0026rsquo;s 26.7% in a different category and population. From 2022 there were two rankings and the share rose to half8, as expected.\nThe same category and age, before and after the rule change. The informative number is the 28.5% on the left: what first-year riders collect when there is a single list.\nFor more detail in the technical document, in Italian: First-year riders do not disappear for lack of places, How concentrated the points are and A rule change as a natural experiment.\nThird: the racing calendar is shrinking. In an average season the rankings record 320 Esordienti races and 146 Under 23 races. More importantly, between 2009 and 2025 recorded races fell by 40% for Esordienti and 58% for Under 23s. Was the pandemic responsible? No: the decline had begun earlier, and after 2020 numbers never recovered to previous levels. Within the comparable window, registered Esordienti fell by around a tenth and races by almost a fifth. The rider population is shrinking, and the calendar is shrinking faster.\nFrom 2007 to 2025: Esordienti fall from 433 races in 2009 to 261, Allievi from 642 to 309, juniors from 430 to 199, and Under 23s from 222 to 88, with a shared collapse in 2020.\nFewer races mean fewer point-scoring places, so the ranking thins by itself: distant year groups cannot be compared as though nothing had changed. The decline concerns men\u0026rsquo;s cycling, however, rather than all youth cycling. Where comparison is valid, among Allievi and juniors, men\u0026rsquo;s races fell by over a third between 2011 and 2025 while women\u0026rsquo;s held steady, narrowing the gap: from one women\u0026rsquo;s race per seven men\u0026rsquo;s races to one per five among Allievi, and from one per ten to one per five among juniors8.\nFor more detail in the technical document, in Italian: How many places there used to be and Ranking attrition and real attrition.\nWhat the score measures One last point about the instrument. In 2025 two Dutch researchers published the strongest criticism one could make of this kind of study: a federation ranking that does not weight races by level treats international races and local events alike.\nFor the Italian ranking, half that criticism is addressed, because the rules do weight races. Dividing points by the number of placings shows this indirectly: 2.67 points per placing among Esordienti, 2.73 among Allievi, 3.02 among juniors and 4.17 among Under 23s9. The jump occurs exactly where the rules\u0026rsquo; multipliers start giving higher-level races more weight, from juniors onwards. For Esordienti and Allievi the scale is flat, consistent with the lack of an international calendar at those ages.\nThe other half still stands. A ranking rewards riders suited to the races most common on the calendar: a climber facing mostly fast courses has fewer opportunities to score. Correcting for that would require knowing each race\u0026rsquo;s characteristics, and those data do not exist.\nFor more detail in the technical document, in Italian: Does the same score mean the same result? and There are very many ties.\nWhat remains Every percentage in the following instalments has as its denominator one seventh of registered boys, an already selected group, rather than all of them. These results cannot therefore be extended to all registered riders. Exactly how far they differ, I do not know.\nThe funnel is not one group shrinking: each category includes a substantial share who were not ranked as Esordienti, and arriving late is entirely normal. One in seven future professionals was not ranked at thirteen. That is a different seventh from the registered-rider figure above, which measures the share who enter the ranking.\nNow you know who we are talking about. At what age does a placing begin to say something about the future? That is the subject of the next instalment.\nHow we know\nRankings come from ciclismo.info, seasons 2007–2025. Membership totals come from statistical data published by the Italian Cycling Federation for 2018–2025. The series are not perfectly aligned: membership is by category, not discipline, so estimated coverage is a lower bound.\nThe literature review summarises twenty-two studies published between 2006 and 2025, giving full references, DOIs and limitations. I read five in full; for the others I used abstracts and cross-citations.\nSchumacher Y.O. et al. (2006), Success in elite cycling: a prospective and retrospective analysis of race results, Journal of Sports Sciences 24(11), 1149-1156. DOI 10.1080/02640410500457299.\u0026#160;\u0026#x21a9;\u0026#xfe0e;\nAll studies cited in this section, with references and limitations, are in docs/literature_review.md (in Italian).\u0026#160;\u0026#x21a9;\u0026#xfe0e;\u0026#160;\u0026#x21a9;\u0026#xfe0e;\u0026#160;\u0026#x21a9;\u0026#xfe0e;\nCalculated in report/moduli/provenienza.py.\u0026#160;\u0026#x21a9;\u0026#xfe0e;\nItalian Cycling Federation, Norme Attuative 2027, Cat. Esordienti / Allievi / Donne Esordienti / Donne Allieve, article 4.4.21 for the points scale and articles 4.2.1 and 4.2.5 for separate racing by birth year. These are federation rules, distinct from the rules the source uses to construct its ranking, which weights races by level.\u0026#160;\u0026#x21a9;\u0026#xfe0e;\nCalculated in report/moduli/copertura.py, using registered-rider totals collected in riferimenti/tesserati_fci.csv.\u0026#160;\u0026#x21a9;\u0026#xfe0e;\nCalculated in report/moduli/attrito.py.\u0026#160;\u0026#x21a9;\u0026#xfe0e;\u0026#160;\u0026#x21a9;\u0026#xfe0e;\nCalculated in report/moduli/posti.py. Places are estimated from top-five finishes, five per race.\u0026#160;\u0026#x21a9;\u0026#xfe0e;\nCalculated in report/moduli/ragazze.py.\u0026#160;\u0026#x21a9;\u0026#xfe0e;\u0026#160;\u0026#x21a9;\u0026#xfe0e;\nCalculated in R/30_misura.R, rendered by report/moduli/misura.py.\u0026#160;\u0026#x21a9;\u0026#xfe0e;\n","permalink":"https://lucabontempi.com/en/blog/ciclismo-giovanile-di-chi-stiamo-parlando/","summary":"\u003cp\u003e\u003cem\u003eThis series looks at Italian youth cycling rankings and the transition to professional racing. Before presenting my numbers, I explain what research already knows and, above all, who the numbers in the following instalments describe. That is where almost all the mistakes lie.\u003c/em\u003e\u003c/p\u003e\n\u003ch2 id=\"twenty-years-of-research-in-a-few-paragraphs\"\u003eTwenty years of research in a few paragraphs\u003c/h2\u003e\n\u003cp\u003eAt the junior road World Championships each nation can enter at most \u003cstrong\u003esix\u003c/strong\u003e riders, and only the highest-ranked nations get all six places. Juniors, or Under 19s, are seventeen- and eighteen-year-olds. Anyone who has raced at that level remembers how hard it is to earn selection. Reaching that start line means passing every selection process youth cycling can devise. You could hardly look more destined for success.\u003c/p\u003e","title":"Youth cycling and professional careers: the current landscape"},{"content":"This project started with a question that has nothing to do with app stores. When I wrote the thesis, using generative AI was discouraged, and in practice that meant don\u0026rsquo;t. ChatGPT had been public for about two months, and nobody argued with the rule.\nThree years on, the same rule looks almost impossible to enforce, and harder still to justify. I\u0026rsquo;d struggle to find a student who isn\u0026rsquo;t using some AI help on a thesis, and I\u0026rsquo;d struggle more to argue they shouldn\u0026rsquo;t. That reversal, not the app store literature, is why I reopened my own work this year. I didn\u0026rsquo;t want to rewrite it. I wanted to see what a second, more sceptical pass would find in something I\u0026rsquo;d trusted untouched for three years.\nEvery figure and every number below comes from a script in the review repository.\nIn February 2023 I submitted a thesis on one narrow question. When someone lands on an app\u0026rsquo;s page in a store, what makes them decide to install it? I tested three elements: the average review rating, the number of downloads, and the developer\u0026rsquo;s brand, carried by the app name, the icon and the developer\u0026rsquo;s own name. I manipulated screenshots of a real Play Store page so each element existed in a strong and a weak version. Every respondent saw three pages, one per element, in random order, each showing one of the two versions drawn at random. Nobody saw both versions of the same element. Then I measured what the literature calls intention to download.\nWhat I published: brand matters most, reputation comes second, and popularity does essentially nothing, at least in a category with low network effects.\nThis year I reopened the files. The data held up. The randomisation worked, there are no missing values, and every regression in the thesis reproduces to four decimal places. The analysis didn\u0026rsquo;t hold up. I made three mistakes, and none of them was in the data. All three were in what I did with it.\nMistake One: When Three Regressions Can\u0026rsquo;t Be a Comparison The headline came from putting three coefficients side by side: 1.53 for brand, 0.50 for reputation, 0.36 for popularity. Brand is clearly the biggest. Done.\nExcept those numbers came from three different regressions on three different groups of respondents. The thesis never tests whether 1.53 differs from 0.50. I looked at the two numbers and concluded.\nSo I ran the test I should have run in the first place: one model, all 1,473 answers, adjusted for the fact that each person gives three of them.\nElement Effect on intention to download (1–7 scale) 95% CI Developer\u0026rsquo;s brand +1.05 0.81 to 1.29 Reputation (rating) +0.85 0.61 to 1.09 Popularity (downloads) +0.22 −0.02 to 0.46 Averaged over the three judgements each person made, brand and reputation are not distinguishable (p = 0.24). Both clearly beat popularity. My three-step hierarchy is really a two-step one, at least on that average, and that qualification matters.\nWhat the fragmentation cost me is the interesting bit. Cutting one sample into nine subsamples threw away most of the statistical power I\u0026rsquo;d collected. Reputation\u0026rsquo;s effect rose by two thirds, from 0.50 to 0.85. About half of that rise is precision: on the same first-app question, one model on all the data puts reputation at 0.68, not 0.50. The other half comes from asking a different question, because reputation gets stronger in the judgements that come later.\nThe same three effects, estimated three ways. Hollow: the thesis, one regression per element on its own subsample. Triangle: the same question (the first app seen) from one model on all the data. Solid: the average over all three judgements, which is a different question.\nI didn\u0026rsquo;t want to trust my new favourite model blindly, so I re-ran the comparison every defensible way: 24 combinations of sample, model type and controls. One failed to converge and I dropped it. Brand comes out ahead in 4 of the remaining 23. Those four aren\u0026rsquo;t a random corner of the results. They\u0026rsquo;re exactly the four that use only the first app each respondent saw.\nThat\u0026rsquo;s no coincidence. I chose the first app on purpose in 2022, because the first screen is the only one a respondent judges before they have anything to compare it with. On that quantity, the one the design was built to produce, brand does lead reputation: +0.76 points, p = 0.03, and +1.02 (p = 0.007) in the sample the thesis actually used.\nSo the headline wasn\u0026rsquo;t an artefact. But the thesis made two claims: brand wins on the first app, and brand loses its lead once the user has seen other apps. It tested neither. Now it\u0026rsquo;s tested. The first holds (+0.76, p = 0.03). The second doesn\u0026rsquo;t reach significance (p = 0.10), so the data show a tendency, not a proven change.\nThe mistake is narrower than \u0026ldquo;the ranking collapses\u0026rdquo;, and harder to shrug off. Nine regressions on nine subsamples share no parameters. That means no comparison between the three elements could be tested, and none was. One model on all 1,473 answers gives the same first-app estimate, plus the test the thesis lacked, plus what happens once the user has seen alternatives.\nEvery defensible specification of the same comparison. Brand wins clearly only where the design aimed it, on the first app each respondent saw. Everywhere else, brand and reputation can\u0026rsquo;t be told apart.\nMistake Two: My \u0026ldquo;Manipulation Check\u0026rdquo; Was Not One The result people quoted back to me was the counterintuitive one: download counts don\u0026rsquo;t matter. In a market obsessed with install numbers, that\u0026rsquo;s fun to say. It\u0026rsquo;s also not quite what I claimed. I used scanner apps because their network effects are low. A scanner isn\u0026rsquo;t more useful to you because millions of other people use it, so a download count can only work as a signal of what others think, never as a promise of extra value. My claim was that popularity is ineffective in that setting.\nI no longer think the data supports even that. There are two reasons. I should have caught one in 2022. I did catch the other, and never followed it through.\nThe first is embarrassing. I went back to the stimulus images themselves. The low-popularity version of the app shows 10,000+ downloads next to 84,000 reviews. More reviews than downloads. That\u0026rsquo;s impossible. The review count was the same in both conditions, so the flaw sits entirely in the weak version, the very one that had to carry the comparison.\nThe second is subtler, and it\u0026rsquo;s about labelling rather than analysis. The thesis reports a manipulation check, model by model: the share of respondents who confirmed they\u0026rsquo;d noticed each element. What I actually asked, once, at the very end, was which factors did you take into consideration?, with nine boxes to tick. That doesn\u0026rsquo;t check that the manipulation registered. It\u0026rsquo;s people reporting, after the fact, what they think they did.\nI knew this at the time. The thesis calls the question \u0026ldquo;a compromise (and not ideal) version\u0026rdquo;. What I did next was deliberate: measure it, report it, condition nothing on it. That was the right call and I\u0026rsquo;d make it again. What I failed to do was spell out the consequence. A study whose only perception measure is a post-treatment self-report has no verification that anyone saw the manipulation. The word check, repeated across nine tables, quietly implies otherwise.\nPopularity pays for this most, because that\u0026rsquo;s where a perception check would have mattered most. Among the 60% who ticked \u0026ldquo;number of downloads\u0026rdquo;, the popularity effect more than doubles, to +0.49, and becomes significant. I offer that as an illustration, not a correction, since the comparison is close to circular: people moved by download counts are more likely to say they looked at them. Still, with a broken stimulus on one side and an unverifiable measure of attention on the other, the null had less support than I gave it.\nSo where does that leave popularity? Before looking, I fixed the smallest effect that would matter in practice. Every reasonable way of deriving it lands between 0.26 and 0.33 scale points. Against that bar:\non the first app a respondent sees, with nothing else on screen, popularity does work: +0.45, and that one is real; on the second app it\u0026rsquo;s −0.22 and on the third +0.40. Both are inconclusive. The interval is wider than the bar in either direction, so I can\u0026rsquo;t separate \u0026ldquo;nothing\u0026rdquo; from \u0026ldquo;too small to matter\u0026rdquo;, or from something quite large. I first wrote that second bullet as a single number, +0.07 after comparison, from a model that treats the second and third app as one condition. That\u0026rsquo;s the same kind of mistake I\u0026rsquo;d just finished criticising in the thesis: collapsing −0.22 and +0.40 into one number describes a situation nobody was ever in. It also flattered a tidy story I liked, because the third app isn\u0026rsquo;t measurably different from the first (0.05 apart, p = 0.86). The dip sits on the second app alone, and no difference between positions survives a correction for having looked at all three.\nWhich nulls are really null. The shaded band is the zone of practical irrelevance, fixed before I ran the tests. An interval far wider than the band means the study couldn\u0026rsquo;t see, not that there was nothing to see.\n\u0026ldquo;Download counts do not matter\u0026rdquo; was too strong. What I can defend is smaller and less quotable. Where network effects are low, download counts move intention on the first app someone sees, and after that this study can\u0026rsquo;t tell you. It\u0026rsquo;s still enough to retire the claim I published.\nMistake Three: Silence Isn\u0026rsquo;t Evidence The thesis concludes there\u0026rsquo;s no interaction between the three elements. Only one element was manipulated per screen, so \u0026ldquo;interaction\u0026rdquo; can only mean something narrower here: does the effect of this app\u0026rsquo;s rating depend on the brand of an app seen earlier? The methods section admits that limit. The abstract doesn\u0026rsquo;t.\nThe deeper problem is that the answer is empty either way. My design could only have detected an interaction of 1.3 to 1.5 points on the 7-point scale. That\u0026rsquo;s as large as the main effects themselves (0.2 to 1.0), and interactions are usually far smaller. Finding none isn\u0026rsquo;t evidence of absence. The instrument was too coarse to see them.\nThe two moderation results I built managerial advice on are the opposite case. Here the data can say more than \u0026ldquo;not significant\u0026rdquo;: both effects fall inside the margin I\u0026rsquo;d fixed in advance as too small to matter. They aren\u0026rsquo;t unproven. They\u0026rsquo;re practically absent.\nWhat Holds Up and One Thing That Is New The brand effect survives everything. Of the 81 coefficients I estimated across nine regressions, exactly two survive a correction for the number of tests, and both are the brand effect.\nBrand and reputation are equally powerful only on average. Split by position in the sequence, the picture is sharper than anything I originally claimed:\n1st app seen 2nd 3rd Developer\u0026rsquo;s brand +1.44 +0.88 +0.92 Reputation +0.68 +0.63 +1.13 Popularity +0.45 −0.22 +0.40 Each element\u0026rsquo;s effect at each position in the sequence, with 95% intervals. Brand is strongest on the first app, reputation on the third, but no difference between positions survives a correction for multiple looks. It shows the shape of a pattern, not a demonstrated one.\nBrand dominates when there\u0026rsquo;s nothing to compare against. That gap is the one my thesis found, and it holds. Once users have seen alternatives, the two converge: 0.89 against 0.92, from a model that treats everything after the first app as one condition instead of two. As a probability, brand leads reputation with 99% before comparison and 48% after, which is a coin toss. What the data doesn\u0026rsquo;t establish is that those two situations differ from each other (p = 0.10), and no pairwise difference between the three positions survives correction, for any of the three elements. So treat the movement as the shape of something worth testing properly. It isn\u0026rsquo;t a result.\nHow large each effect plausibly is, given the data. Where two curves overlap, the data can\u0026rsquo;t tell the two elements apart, which is what “not distinguishable” looks like.\nThere\u0026rsquo;s a practical reading, and it stands or falls with that pattern. A strong brand is worth most when the user isn\u0026rsquo;t shopping around: a link from a search result, an ad, a recommendation. Ratings earn their keep when people browse and compare. For a new entrant that\u0026rsquo;s better news than it sounds. You can\u0026rsquo;t buy Adobe\u0026rsquo;s brand, but you can be the app that survives the comparison.\nThen the one genuinely new finding, which I\u0026rsquo;d predicted in writing would fail before I ran it. Respondents differ a lot in how much any of this moves them, and the difference is systematic. The people most inclined to download things in general are the least moved by what the listing shows (correlation −0.59). The cues work on the undecided. An average across everybody hides that completely.\nOne dot per respondent: baseline enthusiasm for downloading against how much the page elements move them. The cues work on the undecided. Points are model predictions, pulled towards the average where a respondent gives little information.\nFour Things I Checked That Changed Nothing A reanalysis is only worth as much as the checks that could have embarrassed it. Four did not.\nRespondents don\u0026rsquo;t treat a 1–7 scale as a ruler. An answer of 6 covers nearly twice as much underlying intention as an answer of 3: even people who are quite sure stop one short of the top, which is what avoiding the ends of a scale looks like. I re-estimated everything without assuming even spacing. The effects moved by 0.7 to 3.0 percentage points and the ranking stayed put.\nWhere each answer boundary sits on the underlying intention scale, against an evenly spaced scale.\nThe effects don\u0026rsquo;t depend measurably on who\u0026rsquo;s answering: not age, gender, education or occupation. Students respond to reputation roughly twice as strongly as employed respondents. It\u0026rsquo;s a tidy story, and it fails its own test. With 69% students in the sample, it couldn\u0026rsquo;t have passed.\nRemoving the 33 respondents who rushed, or gave a single answer to a whole scale, nudges every effect upwards, as removing noise should. Nothing else changes.\nLast, the thesis figures aren\u0026rsquo;t drawn on the raw 1–7 answers but on a mean-centred version, with one constant subtracted per element. That\u0026rsquo;s harmless as long as the model already knows which group it\u0026rsquo;s looking at, and mine did: re-estimating on the raw answers moves no coefficient by more than 0.00000000005. It does cost the reader something. The heights of the three lines in my figure aren\u0026rsquo;t comparable, only their slopes are, and in four of six panels the centred version ranks the three elements differently from the raw one. I read a slope off that figure, so the conclusion survives. The centring that would have mattered in a repeated-measures design is by respondent rather than by element. That can\u0026rsquo;t be done by subtracting a mean at all. It needs the respondent inside the model, which is exactly what my ANOVA lacked. Done the wrong way, by subtraction, it would have shrunk every effect by about a third.\nThe same cell means on the raw answer scale and on the mean-centred scale used in the thesis figures. The slopes are identical. The order of the three lines isn\u0026rsquo;t.\nWhat Three Years of Distance Taught Me Three things, none about app stores.\nProtecting an estimand and splitting a sample are different things, and I did the second while meaning the first. Measuring only the first app each respondent saw was the right instinct. It\u0026rsquo;s the only judgement made before there\u0026rsquo;s anything to compare against. Estimating it in nine disconnected regressions was wrong, because regressions that share no parameters can\u0026rsquo;t be compared with each other, and comparing them is exactly what I did. I stated both halves of the result and tested neither.\n\u0026ldquo;Not significant\u0026rdquo; is not a finding until you say what you could have detected. Every null I reported deserved a sentence on the smallest effect the design could see. Three of them dissolve under that question.\nLook at your stimuli again. The worst problem in the project, a screenshot showing more reviews than downloads, wasn\u0026rsquo;t in the models or the data. It was in an image file I\u0026rsquo;d looked at a hundred times in 2022 and never actually checked.\nA note on how this was done. The reanalysis used AI assistance. It began as a small curiosity: how would an AI have written my master\u0026rsquo;s thesis? The answer was less interesting than the question it led to, which is how an AI would review it. It found nothing I couldn\u0026rsquo;t have found myself in 2023. It did the one thing I didn\u0026rsquo;t do: check every claim against the evidence I already had, including the convenient ones. That\u0026rsquo;s a low bar, and I didn\u0026rsquo;t clear it the first time.\nIn 2023, the only thing I used AI for in this project was checking my English grammar. This year the same class of tool helped me take the analysis apart and rebuild it in under two weeks. None of the three mistakes needed a machine. A calculator and some patience would have caught them in 2022. What was missing was the decision to doubt something I already believed, and that decision is still mine.\nAI speeds up execution. Rerunning twenty-four specifications takes a few minutes. Judgement (is a null really null, does a convenient finding survive doubt) still moves at the speed it always did, and the risk is mistaking one speed for the other. I wrote this piece with the same kind of help, so it deserves the same scrutiny. If there\u0026rsquo;s a fourth mistake hiding in it, I\u0026rsquo;d like to know.\nThe full review, code, pre-set decision rules and logs are in the review repository; the original thesis, data and code are here.\n","permalink":"https://lucabontempi.com/blog/reopening-my-thesis/","summary":"\u003cp\u003eThis project started with a question that has nothing to do with app stores. When I wrote the thesis, using generative AI was discouraged, and in practice that meant don\u0026rsquo;t. ChatGPT had been public for about two months, and nobody argued with the rule.\u003c/p\u003e\n\u003cp\u003eThree years on, the same rule looks almost impossible to enforce, and harder still to justify. I\u0026rsquo;d struggle to find a student who isn\u0026rsquo;t using some AI help on a thesis, and I\u0026rsquo;d struggle more to argue they shouldn\u0026rsquo;t. That reversal, not the app store literature, is why I reopened my own work this year. I didn\u0026rsquo;t want to rewrite it. I wanted to see what a second, more sceptical pass would find in something I\u0026rsquo;d trusted untouched for three years.\u003c/p\u003e","title":"I Asked an AI to Review My Thesis. It Found Three Mistakes I Could Have Found Myself"},{"content":"Note: While I\u0026rsquo;m publishing this blog post in August 2023, I started writing it in September 2022, when I still was living in Tübingen. This means that many things may have changed since then, mainly prices, affected by last year\u0026rsquo;s inflation rates. However, I still decided to complete this post to conclude the triad of articles that I set out to do at the beginning.\nAfter one year as a student in Tübingen, I\u0026rsquo;ll try to share with you all the knowledge of the city and the surrounding area I gained.\nMy year as an international student in Tübingen is coming to an end and it would be a pity if the knowledge of the city I accumulated over these months ended up being wasted: that\u0026rsquo;s the reason why I decided to write a guide to share my experiences in this wonderful city with you all. We\u0026rsquo;ll cover multiple categories of activities, starting from natural landscapes, sports and unforgettable experiences all the way down to food and fun. Let\u0026rsquo;s start!\nExperiences and Sports There is plenty of traces of the experiences I had during the year in Tübingen also in the Tübingen Photo Collection.\nI\u0026rsquo;m a great fan of live sports and sports events, so those represented a major part of my budget allocation. Unfortunately, I wasn\u0026rsquo;t able to attend some events, such as, for example, the Chocolart, which was suspended in 2021 due to Covid limitations.\nTübingen Tigers I really like basketball and I was very happy when I discovered Tübingen had a basketball team, the Tigers. The Tübingen Tigers play in the German second league and, differently from what one might think, the atmosphere in the building was amazing!\nThey play in the Paul Horn-Arena and there are also student discounts on the tickets (about 10 euros each).\nVfB Stuttgart In order to find a major football club, we need to move to Stuttgart. VfB Stuttgart won 5 Bundesliga titles, the last in 2007. However, in the last years, they were fighting and struggling to remain in the first league.\nMatch tickets are one of the few things cheaper than in Italy (less than 30 euros, as far as I remember) and the stadium is quite nice, making it a great experience.\nATP 250 Stuttgart When I discovered there was a tennis tournament in Stuttgart, I\u0026rsquo;ve been torn for a few hours while deciding if I wanted to go or not. In the end, I was quite happy with my choice of going to see the quarter-finals, even if the price for those four matches was close to 80 euros.\nOverall, I had the opportunity to watch great tennis players live, such as Matteo Berettini, Olympic and Wimbledon champion Andy Murray and Nick Kyrgios. I think that, in the end, considering ticket prices in other tennis tournaments, I was quite lucky with the day.\nStuttgarter Frühlingsfest Between the end of April and the beginning of May, in Cannstatt, there is the Stuttgarter Frühlingsfest for about 20 days.\nThe festival is full of luna park attractions, German traditional food and, obviously, beer. A German inevitable experience, I think.\nStocherkahnrennen The traditional Stocherkahnrennen is probably one of the highlights of the summer semester in Tübingen. It is held (almost) every year since 1956 and the date to save is Corpus Christi day (on a Thursday, usually in June).\nIt is the best opportunity to feel part of the great Tübingen community!\nStadtfest Tübingen Another really funny night was the one I had at the Stadtfest in July. This is a weekend-long fest dedicated to the city of Tübingen, with many music events and a lot of food and beer. If I was still living in Tübingen, I wouldn\u0026rsquo;t miss it!\nUmbrisch-Provenzalischer Markt In September, in the Tübingen city centre, there is the Umbrisch-Provenzalischer Markt, a strange fair mixing the French and Italian food culture. It was a good opportunity to taste something different and have a nice evening.\nEuropa-Park Europa-Park is considered to be the second biggest European theme park and the first in Germany. I love theme parks, so I also really liked its theme based on different European countries. Moreover, there are at least three or four adrenaline-filled roller coasters that are above average. However, reaching Europa-Park from Tübingen is not easy: the only option to visit it on a one-day trip is taking a Flixbus very early in the morning.\nHikes I already talked extensively about my hikes in the area, you can find everything in my Tübingen Hiking Guide blog post.\nSights I also wanted to point out a few places in the Tübingen larger surroundings that I had the pleasure to visit during my stay. I\u0026rsquo;ll not include every single city and town, since you could easily track them in any tourist guide (e.g. the Baden-Württemberg Lonely Planet), but I\u0026rsquo;ll cover some castles and some panoramic viewpoints.\nCastles There are three (plus one) major castles in the larger area surrounding Tübingen. The first one, which is also the easiest to reach, is the Hohenzollernschloss Sigmaringen. I had the opportunity to visit the inside and it was worth it.\nThe second castle is the majestic Burg Hohenzollern, whose courtyard was really nice, way more than the inside. In this case, I would suggest an hike, since that would also give the opportunity to have a view of the castle from different perspectives.\nThe third and last one is the Schloss Lichtenstein, which, more than any other, gives those fairy tale vibes. The small internal courtyard is actually very nice, with a great view of the castle laying on the rocks.\nThere is actually an addition to those three castles, even if it is more a palace than a castle. North of Stuttgart there is the Residenzschloss Ludwigsburg, a really nice palace with some amazing gardens. A visit in spring or summer is highly recommended, with particular attention to the plants and flowers in the park. Moreover, in summer, they also organize some music-themed fireworks displays that were surprisingly good.\nPanoramic Views I also wanted to point out some elevated viewpoints from which you could admire the Württemberg landscapes. The first, which is also my favourite one, is the Schönbuchturm in Herrenberg, about 15/20 km from Tübingen. From the top of the tower, there is a pretty good view of Schönbuch park.\nThe second is the TV Tower (Fernsehturm) in Stuttgart, which is more than 216 m high and, from the observation decks, there is a view of Stuttgart, going from the forests and vineyards in and around Stuttgart to the Swabian Jura and the Black Forest.\nThe third and last is the Rottweil Test Tower, which is also the tallest, standing at 246 m. At 232 m, the tower contains Germany\u0026rsquo;s tallest observation deck.\nGerman Food I\u0026rsquo;ll start the food chapter with one of my favourite paragraphs: when I spent time in a place, I always try to dive deep into the local culture and, even if for some of my Italian friends was quite hard to adapt to the German dishes, I actually enjoyed them almost immediately. Therefore, I tried to create a shortlist including both affordable restaurants and some German-priced restaurants (that could be out of budget for students, but that I suggest trying at least once during your Tübingen experience). The secret, I believe, is to find a good recurring and low-priced solution that you will visit often and to choose wisely the more expensive places that deserve a shoot.\nWhile typical food and I got along pretty quickly, the same cannot be said for drinks. I honestly didn\u0026rsquo;t like Baden-Württemberg\u0026rsquo;s beers, while I was able to find some acceptable wines. Not liking the bier in Germany could be a threat to your survival.\nMy suggestions are Gasthaus Bären, Landgasthaus Waldhäuser Hof and Restaurant Mauganeschtle Tübingen. On the other side, I suggest you avoiding to insert in your everyday life places like Gasthausbrauerei Neckarmüller. While this is a pretty typical touristic choice because of its location, immediately close to the Neckarbrücke, service is terrible (a characteristic that it shares with many other German restaurants) and the quality level is quite random. My suggestion is to go there rarely and just have a beer and a Flammkuchen, avoiding a complete meal.\nTapas: Gasthaus Bären The Gasthaus Bären is an affordable restaurant and bar, which I found to be very attractive for students. Bären has an interesting \u0026ldquo;tapas\u0026rdquo; approach, since they serve small portions at a reasonable price, you will be able to taste some good dishes for less than 5 euros and I believe you could be satisfied by spending about 12/15 euros, depending on what you decide to drink.\nBären is that place on which everybody could agree when everybody is too lazy to find something new and I strongly believe it deserves to be your choice for your first dinner out in Tübingen. I had the best Käsespätzle ever here and also the Zucchini Frikadellen deserve to be tried.\nThe only downside is that, due to limited space, during winter time, you need to be there way before 7 p.m. to be sure you\u0026rsquo;ll have a table.\nMeal price: 10-15 euros. Here there are the Google Maps location and their Tripadvisor page.\nLocation: Landgasthaus Waldhäuser Hof I probably never felt so well integrated into the local culture as when I tried the Landgasthaus Waldhäuser Hof. The food was good, but the setting is probably even better, just in the middle of the Waldhäuser farms area. And, in fact, the location is probably the main highlight of this place!\nPrices here definitely rise, at least if compared to the first proposal, and become similar to those you could find in the majority of German restaurants in the area (prices similar to Gasthausbrauerei Neckarmüller, but the quality is in my opinion way higher). I would probably suggest trying meat dishes here.\nMeal price: 25-35 euros. Here there are the Google Maps location and their Tripadvisor page.\nSwabisch: Restaurant Mauganeschtle Here prices get higher, at least for students\u0026rsquo; standards, but the Restaurant Mauganeschtle was the first suggestion for tasting real Swabian cuisine by Swabian people themselves. Here you could taste Maultaschen, Käsespätzle and many other Swabian dishes, including good meat. Quality is high and it comes at a price: the main dishes go up to 30 euros.\nMeal price: 30-45 euros. Here there are the Google Maps location and their Tripadvisor page.\nInternational Food Despite being a relatively small city, a pretty good selection of international food can be experienced in Tübingen.\nPub\u0026rsquo;s Food: Saints \u0026amp; Scholars Saints \u0026amp; Scholars is an Irish pub in a very convenient location. Being so close to the main library, it has always been a valid alternative to the University canteen.\nConsidering the large portions, prices are pretty good. There, I appreciated the typical pub food, as, for example, hamburgers and the club sandwich, always served with fries. In addition, the location is big enough so that you can be confident to always find a spot.\nMeal price: 20-25 euros. Here there are the Google Maps location and their Tripadvisor page.\nGreek: Faros - an ouzo philosophy There are quite a few Greek restaurants in Tübingen and Faros might be the fanciest one. The location, in Waldhäuser Straße, close to WHO, is nice and the food is actually good too.\nPrices are a little bit higher than average, considering that you would probably want to order more than one or two dishes per person. You may also want to share different dishes to try out different kinds of Greek traditional food. For example, since I\u0026rsquo;m a great fan of Greek cheese, I enjoyed both the Feta Psiti and the Tiri Saganaki.\nIn case you just wanted to have a quick Pita, just outside the restaurant, there is the Yellow Donkey, a fast food and take-away stand, with a lean menu.\nMeal price: 30-35 euros. Here there are the Google Maps location and their Tripadvisor page.\nPizza: Da Michele In Tübingen, there are probably as many Italian restaurants as German ones. There are also many pizzeria, but the quality and the prices aren\u0026rsquo;t usually very convenient. Da Michele, in Derendingen, probably offers the best pizza in Tübingen at a price, that, in Germany makes perfect sense.\nWhile all that glitters is not gold and the service was far from perfect most of the times I was there, it often makes sense anyway to prefer Da Michele to the other pizzeria.\nMeal price: 20-30 euros. Here there are the Google Maps location and their Tripadvisor page.\nFast Food The city centre in Tübingen is littered with fast food places. Many of them, as typical in German cities, are primarily kebab places, but there is also falafel and Asian food around.\nFalafel: Kichererbse Falafel offers a suitable vegetarian alternative to kebab. Kichererbse is probably the best falafel place by democratic decision. Very often, there is a queue outside at lunchtime, demonstrating how this place is appreciated by students and locals in general. My friends and I particularly liked the falafel sandwich with eggplants\u0026hellip;\nMeal price: 8-10 euros. Here there are the Google Maps location and their Tripadvisor page.\nKebab: Lustnauer Imbiss There are so many kebab (or kebap) places in the city centre, in particular along Mühlstraße. I think I tried all of them, but, honestly, none clearly emerged as the best one.\nThe only place that surprised me in some way was Lustnauer Imbiss. I tried both the kebab and other offerings, such as the pide and lahmacun, and I liked it better than all the other places. The place is not that far away from the University Sports Center and the Österberg, so you could take some food for a picnic somewhere close.\nMeal price: 8-10 euros. Here there is the Google Maps location.\nChinese: China Restaurant San Bao While the San Bao is actually a restaurant, with table service, I mostly got take-away food from there. I liked the noodles, which were a valid alternative to the always-present kebab for those dinners in the library. Close to San Bao, there is also Hao\u0026rsquo;s Box: these two places are both good and almost exchangeable for me\u0026hellip;\nTake-away meal price: 8-12 euros. Here there are the Google Maps location and their Tripadvisor page.\nPubs, Bars and Clubs While it is pretty easy to find a good pub in Tübingen where having a beer, it is at the same time hard to find some good drinks. While highly questionable, I tried to give some indications to have an enjoyable night in the city.\nPub: Stadtpost If you asked ten different people, they would probably give at least eight different answers to the question \u0026ldquo;Which is your favourite pub in Tübingen?\u0026rdquo;.\nPersonally, I always had a preference for Stadtpost, but it is not fully rational. I actually liked the atmosphere and the low prices (a beer and fries are quite cheap), but I\u0026rsquo;m pretty sure my friends would appreciate some other pub more than this.\nHere there is the Google Maps location.\nCocktails: Schmitz Katze A great atmosphere can also be found in Schmitz Katze, along with excellent cocktails. This was my favourite choice for having a chat with a small group of people.\nHere there is the Google Maps location.\nCinema: Kino Atelier The Kino Atelier (similarly to the Kino Arsenal) is a small and single-hall cinema, projecting mostly international and German independent films. For example, there I watched Everything Everywhere All at Once when it was first released, way before it won the Academy Award for Best Picture. The small hall contributes to an intimate and cosy atmosphere, making it a really nice experience.\nClubs I will not hide my thoughts on this theme, the Tübingen clubs situation has some large margins of improvement.\nSchwarzes Schaf was funny, but only when with the right group of friends. The student-organized parties in Clubhaus are the best alternative to clubs in spring and summer, while Kuckuck is the best choice for those living in WHO. Of all the (few) other clubs I\u0026rsquo;ve been in, I actually didn\u0026rsquo;t like the atmosphere enough to suggest them.\n","permalink":"https://lucabontempi.com/blog/tuebingen_experiential_guide/","summary":"\u003cp\u003e\u003cem\u003eNote: While I\u0026rsquo;m publishing this blog post in August 2023, I started writing it in September 2022, when I still was living in Tübingen. This means that many things may have changed since then, mainly prices, affected by last year\u0026rsquo;s inflation rates. However, I still decided to complete this post to conclude the triad of articles that I set out to do at the beginning.\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eAfter one year as a student in Tübingen, I\u0026rsquo;ll try to share with you all the knowledge of the city and the surrounding area I gained.\u003c/p\u003e","title":"A Tübingen Experiential Guide for International Students"},{"content":"Premise This blog post contains the full text of my Master Thesis, written at the Eberhard Karls University of Tübingen and at the University of Pavia and submitted for the first time during February 2023.\nIn September 2026 I reopened this work and reanalyzed the original data. That second pass found three mistakes in the analysis (none of them in the data itself), and is documented in I Asked an AI to Review My Thesis. The text below is the thesis as it was submitted, left unchanged.\nThe research question investigates which characteristic out of reputation, popularity and developer\u0026rsquo;s brand, reflected in some elements in the presentation page on app stores, is the most effective in predicting user\u0026rsquo;s choice to download an app.\nFirst, Chapter 1 provides an extensive background on the mobile ecosystem, its participants and its dynamics. Then, in Chapter 2, the previous literature results on the effects of those factors are covered, while, finally, in Chapter 3 and 4 the empirical analysis and its results and discussion are presented.\nThe full document version of this thesis is available here, while a summary presentation can be found here.\nAbstract After providing an extensive background on the mobile ecosystem, its participants and its dynamics, this research aims to establish which characteristic out of reputation, popularity and developer\u0026rsquo;s brand, reflected in some elements in the presentation page on app stores, is the most effective in predicting user\u0026rsquo;s choice to download an app. Experimental results show how developer\u0026rsquo;s brand is generally the most decisive element, but, when comparison between apps is introduced in the model, it loses its effectiveness and reputation becomes more relevant. Popularity and peer influence, in the context of low network effects, appear to be unexpectedly ineffective in driving users\u0026rsquo; decisions across all models. Moreover, users\u0026rsquo; involvement in the download process reinforces the effect of reputation, while, when customers are more involved in the specific app category and perceive a higher risk, apps from established brands are more likely to benefit. Finally, it was not possible to find clear evidence of any two-way interaction between reputation, popularity and brand.\n(in Italian) Dopo aver fornito una descrizione dell\u0026rsquo;ecosistema mobile, dei suoi partecipanti e delle sue dinamiche, questa ricerca si pone l\u0026rsquo;obiettivo di stabilire quale caratteristica tra la reputazione, la popolarità e il brand dello sviluppatore, riflesse in alcuni elementi nella pagina di presentazione sugli app store, sia la più efficace nel predirre la scelta dell\u0026rsquo;utente di scaricare un\u0026rsquo;applicazione mobile rispetto ad un\u0026rsquo;altra. I risultati dell\u0026rsquo;esperimento mostrano come il brand dello sviluppatore è generalmente il fattore più influente. Tuttavia, quando la possibilità di comparare le applicazioni è introdotta nel modello, il brand perde parte della sua efficacia e la reputazione diventa più rilevante. La popolarità dell\u0026rsquo;applicazione tra gli altri utenti, in un contesto di scarsi effetti di rete, sembra essere sorprendentemente inefficace nell\u0026rsquo;influenzare le scelte degli utenti in tutti i modelli. In aggiunta, il coinvolgimento dell\u0026rsquo;utente nel processo di download delle applicazioni rinforza l\u0026rsquo;effetto della reputazione, mentre, quando l\u0026rsquo;utente è più coinvolto nella specifica categoria di applicazioni e quindi percepisce un rischio maggiore, le applicazioni sviluppate da brand consolidati hanno più probabilità di essere scelte. Infine, non è stato possibile individuare alcuna interazione bidirezionale significante tra reputazione, popolarità e brand dello sviluppatore.\nChapter 1. Introduction The mobile industry has incredibly changed throughout the years it existed. While the developments in hardware and software components have been in the public eye for many years now, the most impressive change was arguably the shift in power and influence among market participants.\nIn fact, the industry\u0026rsquo;s balance of power, which back in the 1990s and early 2000s used to favour mobile device manufacturers and, in particular, network operators, now sees a hardly disputed winner, which are mobile platform providers, like Google or Apple (Basole et. Karla, 2012, p. 35).\nThe most important year in this transition was 2008 and Apple was probably the main actor. As commonly known, in 2007, Apple launched the first version of the iPhone and it could be argued that the real innovation didn\u0026rsquo;t lay in the new hardware features of this phone: Apple\u0026rsquo;s iPhone wasn\u0026rsquo;t the first one to introduce a capacitive touchscreen, for example, and LG did that some time in advance in 2006 (Ihlwan, 2007). The real breakthrough that initiated the \u0026ldquo;app economy\u0026rdquo; was the introduction, in 2008, of Apple\u0026rsquo;s App Store, a mobile application marketplace with originally a few more than 500 apps and which now accounts for more than 1.5 million applications (Basole and Karla, 2012, p. 32; Ceci, 2022b; Friedman, 2013). The first app stores were launched at the end of the 1990s, but it wasn\u0026rsquo;t until almost a decade later, with the Apple Store, that they started to gain traction (Kouris and Piller, 2014, p. 21). Since then, the increasing availability of mobile apps, supported by enhancements in hardware features, drove the growth of the mobile industry. In fact, the number of smartphone users worldwide almost duplicated only considering the last five-year period, reaching 6.2 billion users in 2021 and being expected to grow up to 7.6 billion in 2027 (Ericsson, 2022). Meanwhile, apps became an incredibly profitable business by themselves and those downloaded from the Google Play Store, the largest app store by the number of available apps at the moment (3.5 million apps in 2022), generated in 2021 more than $ 47 billion in gross revenue, three times more than what was generated back in 2016 (Ceci, 2022a, 2022b).\nClearly, being the owner of a successful app marketplace is a position as desirable as difficult to reach for companies operating in the mobile ecosystem and many appear to fail while attempting that. The complex dynamics lying behind the interaction among market participants (e.g. app developers and customers) are shaped by the mechanics of information economics, which may raise entry barriers and uncertainty. While past literature has already provided useful insights on two-sided platforms,1 a category to which app marketplaces belong, some elements of customer behaviour, when dealing with digital products, need to be explored more deeply. However, before investigating users\u0026rsquo; preferences on mobile app stores and focusing on empirical evidence, an overview of the mobile ecosystem is necessary.\n1.1 Mobile Ecosystem The mobile phone industry is now more than forty years old. In 1978, in fact, a Motorola engineer, Martin Cooper, who is today considered the inventor of the cell phone, made the first mobile phone call using a Motorola DynaTac in New York (Zheng and Ni, 2010, p. 32). Six years later, in 1984, a more recent version of the same Motorola phone was commercialized for the first time. Those phones were also defined as \u0026ldquo;brick phones\u0026rdquo;, due to their shape and size, and, in the following ten or twenty years, the innovation efforts of this industry focused on miniaturization, better power efficiency and battery life, better displays and signal. In the meantime, generations of cellular network technologies were succeeding (from 1G to today\u0026rsquo;s 5G), improving voice communication quality, as well as introducing revolutionary features such as text messaging or Internet access. However, the real disruption in this industry came in 2008, with the introduction of the first smartphones and of the app concept, which overturned power relations between market participants.\nBasole and Karla (2012, p. 28) identified four groups of market participants, mobile platform providers, mobile device manufacturers, mobile network operators, and mobile application providers, to which a fifth group, mobile users, can be added to effectively understand the industry\u0026rsquo;s internal dynamics.\nMobile Platform Providers. Mobile platform providers (MPPs) provide operating systems and middleware solutions for smartphones and they are currently driving the mobile industry. In addition, MPPs often coincide with the owners of major mobile app stores. This double role was arguably assumed for the first time by Apple in 2008, when, while launching the iPhone, it leveraged its experience in content distribution, thanks to the previous existence of iTunes, to introduce the App Store, in order to strengthen its ecosystem and drive demand for iOS devices (Basole and Karla, 2012, p. 36; Cortimiglia et al., 2011, p. 3), before app stores became a profitable business by their own. While a couple of years after its introduction, iOS was installed on 37 percent of all smartphones, a market share similar to the one of Symbian OS (Nokia\u0026rsquo;s own operative system), and Android had just entered the market, as shown in Fig. 1.1, now the latter dominates other mobile operative systems (with about 71 percent of the global market), just followed by iOS and other operative systems with very small shares (Laricchia, 2022).\nFigure 1.1: Mobile OS market shares in 2022 and 2009 (Laricchia, 2022). Interestingly, these two systems were both able to succeed with two different approaches: iOS is considered a relatively closed ecosystem (\u0026ldquo;walled garden\u0026rdquo;), while Android is a more open one (\u0026ldquo;open garden\u0026rdquo;). The unique role of Apple as both MPP and device manufacturer enables it to apply some restrictions, as, for example, the fact that apps can only be installed from the App Store, as sole source (Basole and Karla, 2012, p.36; Cortimiglia et al., 2011, p. 3), and this approach allows more strict testing and verification, that, in combination with the existence of only a few different hardware with that operative system (OS), may lead to a better user experience, with consequently improved customer loyalty and a premium brand image. On the other hand, Android is an open-source OS and the ecosystem is more open: this may lead to higher levels of competition both on software and hardware, with lower prices as result (Müller et al., 2011, p. 74). In terms of app distribution, both Apple and Google act as gatekeepers: they decide what gets published on their stores, also covering important roles in terms of quality assurance and pre-screening of dangerous apps (Müller et al., 2011, p. 73), but, while in Google case, these limits can be circumvented thanks to the presence of other app sources, this is (legally, at least) impossible on Apple\u0026rsquo;s devices.\nOverall, MPPs took over many parts of the value chain, from OS development to payment systems, relegating other participants to more subordinate roles.\nMobile Device Manufacturers. Mobile device manufacturers\u0026rsquo; (MDMs) activity mainly consists in producing and distributing the hardware part of a smartphone. In the past, MDMs generally owned proprietary operative systems installed on their products, only a few native apps were offered with the phone and the possibility of installing third-party apps was generally limited. Their role was strictly connected to the one of network operators, on which they often depended for distributing devices, usually sold in combination with mobile subscriptions (Basole and Karla, 2012, p. 29).\nToday, the only MDMs developing a popular OS are the most important MPPs, Apple and Google, while all the other producers (Samsung, Xiaomi, Oppo, Vivo, etc.) redirected themselves towards Android as the chosen OS for their devices in most cases.\nMobile Network Operators. Mobile network operators (MNOs) have been dominant in the mobile service industry for many years. For decades, they were the main sources of innovation and added value, they supplied SIM cards and they could also dictate some pre-settings, like the browser\u0026rsquo;s homepage or types of applications installed (Basole and Karla, 2012, p. 29). Later, with the smartphone introduction and the app economy emerging, MNOs lost their traditionally central role, their margins shrunk and the risk of becoming dumb pipes is real.\nMobile Application Providers. Mobile application providers (MAPs) are those providing content in the current mobile ecosystem. They develop both web-based and native applications for specific technology platforms, which cover many different categories and use cases. Developers traditionally lacked market accessibility and app stores greatly reduced barriers and time-to-market for them, but, on the other side, they now largely depend on MPPs\u0026rsquo; enabling capabilities and developing tools (Basole and Karla, 2012, p. 30). Anyway, MAPs can now reach a large number of potential customers, at the price of leaving (on average) 30 percent of their app download revenue to gatekeepers (Cortimiglia et al., 2011, p. 5), whose role is particularly restricting in walled gardens, due to the lack of alternative channels.\nMobile Users. On the other side of a mobile app store, there are smartphone users. They search, download and pay for apps and they also later rate them. There is a tendency among users to view the mobile device and available apps as a package: they take into consideration the ecosystem as a whole and then decide which hardware and software combination to buy. In this sense, mobile devices can be considered complementary goods for apps and their price a one-time membership fee to join the ecosystem (Kouris and Piller, 2014, p. 36).\nFor customers, downloading an app and the perception of the risk involved depend more on the MPPs than MAPs: users, in fact, place their trust in platform owners as their main source of information. In addition, they use apps to perform highly fragmented everyday tasks and perceive apps not as complex as traditional desktop software. This makes their buying decision limited, habitual and impulsive, corresponding to generally low levels of involvement (Buck et al., 2014, p. 5). During the download process, users often don\u0026rsquo;t search for information actively and reputation systems (e.g. rating score), as well as peer influence (e.g. sale rankings, number of downloads), drive their decision-making (Buck et al., 2014, p. 5; Liu et al., 2014, p. 330).\n1.2 Mobile Apps as Digital Products The term \u0026ldquo;mobile application\u0026rdquo;, \u0026ldquo;mobile app\u0026rdquo; or just \u0026ldquo;app\u0026rdquo; refers to application software, which is \u0026ldquo;a type of software that allows the user to perform a specific task, that can be installed and run on a range of portable devices such as smartphones and tablets\u0026rdquo; (Liu et al., 2014, p. 327). An app usually distinguishes itself by general software because it is optimized for touch-screen-based mobile devices, it is offered free of charge or for sale and the most of the times it is available for download through a centralized online marketplace where users can also rate and review the app and access some ranking lists (Liu et al., 2014, p. 327). Apps already have a longer history than one could expect: for example, Zheng and Ni (2010, pp. 51-56) were able to individuate two generations of mobile apps already back in 2010. The first generation involved very simple apps, such as ringtones, instant messaging and personal information management apps, among which the first one was the most popular for many years. Then, the second generation included more complex apps like mobile commerce, gaming, music streaming and mobile social networking and some of these are still popular today. Later, a third generation emerged, with features like augmented reality, ubiquitous social networks and merging and seamless switching between different functions (Kouris and Piller, 2014, p. 25).\nMobile apps, exactly like all types of software, are digital goods and as such they share all the typical characteristics of this category of products. Compared to standard economic models, in a digital information setting, costs fall substantially and approach zero. In fact, Goldfarb and Tucker (2019, p. 3) were able to identify five types of cost reduction: lower search costs, lower replication costs, lower transportation costs, lower tracking costs and lower verification costs. Even if these cost reductions are intrinsic to all digital (i.e. intangible) goods and services, they are enhanced by the use of mobile app stores as the main distribution channel.\nIn fact, mobile app stores are an example of two-sided platforms. These platforms are intermediaries that enable exchanges between other players and, by doing that, they efficiently organize supply and demand allowing buyers to access the sellers (and vice versa), facilitating matchmaking and lowering transaction costs. Platforms aggregate information and provide access to that data to market participants and also provide ratings and feedbacks that help to build credibility and trust among participants, making some previously unknown seller characteristics observable, rewarding the better-quality users and creating incentives for lower-quality ones to leave the market. In exchange for easier transactions, these businesses charge a share of their value for the service (Lambrecht et al., 2014, p. 10). In addition, by aggregating market participants, intermediary platforms generate indirect externalities (or indirect network effects): in fact, they usually consist of two-sided (or multi-sided) networks with two (or multiple) types of members, where each type is looking for a network with the largest number of users of the other type. In these cases, network effects become so strong that customers tend to pick a network, instead of choosing a service or a product, since the utility on one side of the market highly depends on the number and quality of participants on the other side (Kouris and Piller, 2014, pp. 39-40; Piccoli and Pigni, 2019, p. 239). Indeed, mobile app stores are characterized by a positive feedback loop: the greater the number of users of the store (reflected in the number of downloaded apps or in the installed subscriber base), the more attractive that marketplace becomes for developers, resulting in an increase of the supply of content and applications and, as a consequence, in higher consumer benefits and increased attractiveness for additional users (Basole and Karla, 2012, p. 35).\nLower Search Costs. It is a lot easier to find and compare information about potential transactions online than it is offline. As a consequence of the decrease in search costs, prices have fallen and product variety increased (Goldfarb and Tucker, 2019, pp. 6-7). While prices online appear to be constantly lower than offline, also for tangible products (e.g. Brynjolfsson and Smith, 2000, p. 579), in purely digital contexts, such as app marketplaces, prices actually reached zero in many cases. Minimal search costs, together with other reasons that will be later discussed, had an important role in spreading and making successful the freemium strategy and increasing the offer of free apps.\nHowever, even if prices are lower, an important level of price dispersion persists online. The ease of changing prices and running price promotions on mobile apps, which also have no marginal costs of production, makes this a common characteristic of mobile app stores too. \u0026ldquo;Free App of the Day\u0026rdquo; promotions are an example of an initiative that could rise price dispersion on app marketplaces. This kind of initiative involves spotlighting in the homepage and zero-pricing an app for 24 hours and it had been used in the past, for some periods, by the Google Play Store, the Apple App Store and the Amazon Appstore, the three most popular app stores today (Chen et al., 2022, pp. 1-2). The objective of this practice is combining digital free sampling, which proved to be effective in reaching new customers, with sponsorship practices, like featuring the app in a noticeable position, consequently increasing visibility and enhancing customer awareness of the app. According to Chen et al. (2022, pp. 6-7), such a promotion may lead to adverse effects in the post-promotion phase: firstly, the removal of zero-price hurts sales, in particular, due to customers\u0026rsquo; expectations of the app being free and their perception of a loss by observing the restored (and increased) price; secondly, promotion-induced reviews, generated by the increased visibility of the app, are less effective than organic ones.\nLow search costs also mean that it is way easier to find rare and niche products online (Goldfarb and Tucker, 2019, p. 8), a phenomenon defined by Anderson (2006, p. 22) as \u0026ldquo;the long tail\u0026rdquo;. The effects of the long tail are particularly visible on app stores: apps generally require just relatively little storage space, making welcoming as many apps as possible quite inexpensive for marketplace owners, with positive benefits overall. Regardless of the number of sales, unsuccessful apps provide additional choices and options for customers, reinforcing the two-sided positive feedback loop (Basole and Karla, 2012, p. 35). In addition, the growing user-friendliness of developing tools allowed many people besides professional developers to publish their apps, further increasing the range of available options. A broader choice of apps, from a negative perspective, leads to higher choice complexity, which could cause delays in the users\u0026rsquo; app download process. This negative effect of the long tail is usually mitigated by recommendation systems, app rankings and improved research, made possible by the available data on customers\u0026rsquo; previous behaviour.\nFinally, an online marketplace can sometimes generate the superstar effect, with homogeneous customers all agreeing on which product is the best and buying it (Goldfarb and Tucker, 2019, p.8). In the case of apps, this phenomenon is extremely strengthened when software benefits from direct network effects and its utility largely depends on the size of the installed base (e.g. instant messaging apps). In these cases, the probability of being in presence of a tipping (or \u0026ldquo;winner-takes-all\u0026rdquo;) market is high.\nLower Replication Costs. In addition to being subject to marginal costs which are close to zero, software is a non-rival, meaning that it can be consumed by one person without reducing the amount of quality available to others, and, in absence of a legal or technological limit, non-excludable good, so it can be reproduced and used by anyone at approximately no cost. These characteristics often generate opportunities, like the possibility of implementing bundling strategies at a very low marginal cost, very common in the software industry, but also some threats.\nLow replication costs and non-excludability are piracy\u0026rsquo;s main reasons for existence. Piracy has been a long-discussed issue in the software industry and illegally distributed copies actually offered in some ways opportunities for free sampling and strengthening network effects in the past, leading researchers and companies to think of piracy as an advantage under certain conditions. Raising the barriers of protection causes, in fact, some users to buy the product and others to live without it or look for alternatives, consequently reducing the installed base and its value also for those users that paid for the program (Reavis Conner and Rumelt, 1991, pp. 136-137). In addition, illegal copies contribute to raising switching costs for pirates, with a higher probability that they would later buy the software. Today, the incentive not to get in the way of illegal distribution has been partially revised, also considering the effectiveness of freemium software in covering the same functions previously fulfilled by piracy, while remaining in control of intellectual property and generating some earnings from advertising, for example (Halmenschlager and Waelbroeck, 2014, pp. 30-31; Nan et al., 2018, p. 283).\nIn some way, compatibility can be seen as a limit to software duplication: in fact, software needs to be compatible with other elements of the ecosystem (the operative system, in particular) and, therefore, in some cases, it cannot be run without some efforts by developers to make this possible. Sometimes, making software compatible with an additional OS can be almost as expensive as it was creating it at first and it needs some incentives from the platform owner side, such as, for example, a large installed base, a high-quality development toolkit or low monetary and technical entry barriers: developers that choose to target two or more ecosystems try to balance between a greater potential market and the cost of porting the product for different platforms (Hyrynsalmi et al., 2012, p. 60; Koch and Kerschbaum, 2014, pp. 1432-1434; Rochet and Tirole, 2003, pp. 1015-1017). Examining the strategy of the most popular apps, multi-homing might be assumed sometimes to be a viable way to increase the circulation of an app. However, the effect could be bi-directional and popularity in a single ecosystem could actually trigger the porting of the app in a second one. In fact, on mobile app stores, the large majority of apps (about 87 to 95 percent) are developed for a single technology platform and apparently, a multi-homing strategy isn\u0026rsquo;t very effective in making an app more popular than others (Ali et al., 2017, p. 5; Hyrynsalmi et al., 2012, p. 71). While at least professional developers would be expected to develop cross-platform apps, one out of five top-rated apps are not present on both Android and iOS, for reasons which reportedly are lack of resources, platform restrictions, low expected revenue in the other platform, user fragmentation within a platform and competition being already present, and, even if developed for both platforms, in any case, they are usually not consistently released in terms of new versions and updates (Ali et al., 2017, pp. 6-7). Obviously, if multi-homing can be hard for companies with skilled developers, it gets even rarer in an ecosystem that contains a good number of app publishers doing it just for fun or lacking a clear monetisation plan (Hyrynsalmi et al., 2012, p. 68). In addition to that, platforms in some industries tend not to promote multi-homing and even to obstacle it when it reduces possibilities of differentiation. While setting conditions that raise developer attractiveness usually increases the share of cross-platform software available, it also reduces user attractiveness when compared to other platforms (Hyrynsalmi et al., 2012, p. 70).\nLower Transportation Costs. The cost of transporting information goods over the Internet is quite near zero. Software has been distributed online and in physical formats for many years now and, while, back in the days, offline distribution was certainly more widespread, the Internet started to pick up as an important distribution channel soon.\nAnyway, software distribution over the Internet used to present some drawbacks, in particular in the past. In fact, anyone could post software to the Internet, with no accountability and making malicious software downloads quite common, creating the need for a solution for trusted software distribution (Rubin, 1995, p.7). From a certain perspective, app stores answered to that problem, creating a trusted and secure channel even for less technologically advanced customers. In addition, they also provided a more efficient and faster system for automated software updates, making them almost unnoticeable to unaware customers, reducing security threats and improving user experience.\nToday, when getting information on which app to download, users place their trust in app store owners, creating a way closer relationship with intermediaries than third-party software developers (Buck et al., 2014, p. 13).\nLower Tracking Costs. In a digital domain, user activity can be easily recorded and stored (Goldfarb and Tucker, 2019, p. 19), enabling relevant opportunities to make software selling more efficient. A major example of this is personalized recommendations, present on almost every online marketplace as well as on mobile app stores.\nThe availability of information on user behaviour also opened up new price discrimination opportunities: in particular, freemium, a specific kind of versioning, became predominant. Freemium, a combination of the words \u0026ldquo;free\u0026rdquo; and \u0026ldquo;premium\u0026rdquo;, is a business model by which a service or a product is offered free of charge, but a premium version is charged for advanced features, functionality, related products or services (Liu et al., 2014, pp. 327-328). Therefore, the free basic version of the product contains limited features, less privacy, more advertising or limited time usage (Boudreau et al., 2022, p. 1378). Freemium insert itself in a context of sampling and versioning strategies, that proved, for many years, to be efficient in creating a positive brand attitude and awareness when uncertainty about the product is high (Arora et al., 2017, p. 63). In addition to that, mobile applications are largely experience goods and it can be difficult for customers to quantify the true value of the product before using it, while a free version of the software allows customers to get to know the product, expecting that a considerable share will purchase the premium version (Boudreau et al., 2022, p. 1379). The freemium strategy in the app economy is also used to exploit advantages ascribed to positive network effects, product complementarity and an accelerated diffusion process. For companies leading the market, a free version of an app may provide a method for exploiting their advantage and expanding their lead over their competitors; for new entrants, a freemium strategy may provide a way to eat a part of the leader\u0026rsquo;s advantage (Boudreau et al., 2022, pp. 1375-1376). However, unlike traditional software, apps are generally characterized by minimal learning and switching costs and a low degree of interdependence among apps from the same developer (Liu et al., 2014, p. 330), raising fragmentation and competition. Implementing this strategy remains anyway quite complex and a trade-off emerges: on one side, consumers will judge the app on the basis of the quality of the free version (Liu et al., 2014, p. 330); on the other side, if the free app is sufficient for users, it could cannibalize the paid version.\nNevertheless, collecting users\u0026rsquo; information isn\u0026rsquo;t always positive and privacy regulation puts a cost on tracking information flows (Goldfarb and Tucker, 2019, p. 23): on one hand, there are legal compliance costs and, on the other, there is a huge reputation risk. Many users report being very concerned about their privacy, but they still frequently download free apps that monetize their personal data, bringing to light a piracy paradox. Clearly, consumers on the Internet articulate their need for privacy, but they act in an opposite direction (Buck et al., 2014, p. 4).\nLower Verification Costs. As already discussed, mobile app stores, as two-sided platforms, contribute to increasing trust among participants through the rating system. In an offline setting, the company\u0026rsquo;s reputation rests on having an uninterrupted series of transactions with a positive outcome; online and on app stores in particular, the purchase is typically a one-time event, meaning that the reputation system becomes fundamental in providing insights on the app quality and making it public for users, in a way to discourage opportunistic behaviours of app developers, even if the transaction is often a unique event (Müller et al., 2011, p. 67).\nAnyway, mobile apps are continuously evolving at a rapid pace with updates being released and replacing older versions quite frequently. As a result of app updates, user satisfaction may change frequently and traditional rating systems appear sometimes unable to capture these dynamics since they are usually resilient to average rating fluctuations when the app has already a substantial number of previous reviews. This system disadvantages new emerging developers, considering that initial low quality of the app and consequently low ratings may cause the impossibility to come back to a good average score later and it could encourage developers to release a new version of a badly reviewed app as a completely new one, trying to reset its reputation (Ruiz et al., 2015, p. 11).\nMoreover, another problem of review systems on online marketplaces is \u0026ldquo;reputation inflation\u0026rdquo;. In fact, average public feedback scores on these platforms have greatly increased and it looks like customers are reluctant to give a low rating (in particular when the feedback is public) because they fear retaliation or they don\u0026rsquo;t want to harm the seller. In addition, the perceived cost of leaving a negative review increases higher the average rating, pressuring consumers to leave above-average scores (Filippas et al., 2018, p. 30; Horton and Golden, 2015, p. 2). In the long term, this phenomenon makes the reputation system less informative and also creates nonlinear rating expectations for buyers, raising the threshold above which buyers consider the average score a good one.\nAll in all, the advantages of mobile app stores as platforms and distribution channels are overwhelming, even if some aspects still need to be perfectly tuned. Their introduction contributed to making transaction costs, already decreased by digital economics, even lower and helped in democratizing software distribution.\nChapter 2. Literature Review Many authors previously covered customer behaviour on online marketplaces, comparing it with the offline domain (e.g. Brynjolfsson and Smith, 2000; Brynjolfsson et al., 2003). Recently, literature also focused on user behaviour on mobile app stores: for example, Burghers et al. (2016, p. 330) argued how apps are generally low-priced or free and this causes low effort and regret when they are unsatisfied and decide to uninstall them. This means that the perceived risk in the buying decision process is low and leads, in many cases, to low involvement. In addition, some consumers appear to be unconscious and they do not seem to search for relevant information that could reduce the information asymmetries in the app market (Buck et al., 2014, p. 12). There is evidence that decision-making when dealing with app downloading often relies on heuristics (e.g. \u0026ldquo;take the first\u0026rdquo;), which allows users to quickly navigate across offerings, but may also cause customers to overlook some important information (Dogruel et al., 2015, p. 139).\nIn this context, it can be assumed that some elements in the presentation page of an app on a store may be effective in convincing a user to download it. In particular, three main references and their influence on user\u0026rsquo;s intention to download will be investigated below: app reputation, its popularity (and the consequent peer influence) and the developer\u0026rsquo;s brand.\nProduct reputation impact on purchase decisions online has already been covered by many authors. Chevalier and Mayzlin (2006, pp. 353-354), for example, proved that an improvement in online reviews is positively correlated with an increase in sales and the impact of ratings isn\u0026rsquo;t linear, with one-star reviews being more influential than five-star reviews. Moe and Trusov (2011, pp. 455-456) testified how ratings can improve sales, but customer rating behaviour is significantly socially influenced by previously posted ratings. Previous literature investigated eWOM (electronic word-of-mouth) valence, usually reflected by the average rating score, as well as eWOM volume, which is usually measured by the number of reviews. Out of the papers that covered eWOM valence on mobile app stores, seven of them were able to find a positive impact on app success (Böhm and Schreiber, 2014, p. 13; Burgers et al., 2016, p. 340; Chen et al., 2022, pp. 6-7; Finkelstein et al., 2017, p. 38; Gokgoz et al., 2021, pp. 434-435; Jung et al., 2012, p. 936; Lee and Raghu, 2014, p. 161), one found a negative correlation (Liang et al., 2015, pp. 250-251) and four found no significant impact (Roma and Ragaglia, 2016, pp. 184-185; Sällberg et al., 2022, pp. 7-10; Timmerman and Shepherd, 2016, pp. 14-15; Wang et al., 2015, p. 5). On the other side, four authors discovered a positive relation between app success and the number of reviews (Burgers et al., 2016, p. 340; Lee and Raghu, 2014, pp. 149-151; Sällberg et al., 2022, p. 13; Wang et al., 2015, pp. 7-10), while just one found out it was negative (Liang et al., 2015, pp. 250-251). The majority, in both cases, used the number of downloads in a selected time period as a performance metric, while some others preferred app survival in the top-apps lists. Overall, reputation in the mobile ecosystem has already been studied enough to take some preliminary conclusions and the majority of past literature contributions agree that reputation signals quality and it has a generally positive effect on sales.\nThe popularity of an app is reflected in multiple elements of an app store: firstly, the number of downloads is usually shown in the app page, then the top-list presence and app sales rank also signal it is popular among users (Sällberg et al., 2022, p. 3). Popularity translates into more interest among potential customers because it reflects the previous interest of other consumers and this peer influence has been shown to be one of the primary drivers of purchase decisions in online shopping (Liu et al., 2014, p. 330). Salganik et al. (2006, p. 855) have already shown how popularity information and social influence affect other users\u0026rsquo; choices on music platforms, making popular songs even more popular. Also in the software industry, there is evidence that app popularity, reflected, for example, in the presence in top-apps ranking, has a positive impact on app success (e.g. Finkelstein et al., 2017, p. 2679). In addition, his effect may be even stronger in the case of software that generates strong network effects, since its utility increases with popularity and users are expected to consider that. Finally, the consequences of reputation aren\u0026rsquo;t independent of social influence: interestingly, less popular products in the market are more significantly influenced by online reviews (Liu et al., 2014, p. 331).\nThe third and last considered anchor is the brand of the company developing the app. Even if online marketplaces lower prices and reduce information asymmetries, brands remain sources of competitive advantage (e.g. Brynjolfsson and Smith, 2000, p. 580). In the mobile ecosystem, Roma and Ragaglia (2016, p. 184), for example, have already shown how a company\u0026rsquo;s fame positively moderates app success. On mobile app stores, the brand may be reflected in multiple elements of the presentation page: apart from the developer or publisher name, which is usually present, the company\u0026rsquo;s name is sometimes included also in the app\u0026rsquo;s name and in the icon. While, if taken singularly, each one of these elements has a relative importance on purchase decision which is low, at least if compared to average review rating and number of reviews (Böhm and Schreiber, 2014, p. 12), they could combine and consistently influence users. Furthermore, app icons are graphic elements and they are easy to understand for users (Böhm and Schreiber, 2014, p. 8) and Pol (2015, p. 51) showed how consumers prefer apps with a brand logo, well-known brands also gain an additional advantage and the brand attitude has a positive effect on the perceived app quality.\nChapter 3. Empirical Analysis The experimental approach used in this research aims to understand how users\u0026rsquo; intention to download (ITD) varies when three main variables (app reputation, popularity and developer fame) and their values change, taking also into consideration their interaction effects, the effect of comparison and of involvement in mobile apps, in the app purchase process and in the app category selected for the purpose of the experiment. Differently from previous research on the topic, this study focuses on direct user participation as its central characteristic: in fact, most of previous literature investigating the effect of the same variables used the number of downloads or an app survival in app store rankings as performance measures. However, while scraping mobile app stores as a method for data collection allows for larger samples, the algorithms behind app rankings on these stores are not public and their functioning is unknown, meaning they could not directly reflect users\u0026rsquo; preferences. For example, Google\u0026rsquo;s Play Store top apps rank is probably based on a complicated algorithm, that takes into consideration multiple factors, such as app title, description, localization, retention rate, number of reviews, app updating, number of downloads and review ratings, meaning that a high rank does not always indicate a high user satisfaction (Liu et al., 2014, p. 348).\n3.1 Method and Measurement This research relies on an online questionnaire as its data source and the experimental design consists of parallel binary manipulations of the three main variables. For each one of these variables, one out of two manipulated screenshot images of the presentation page of some apps (presented in Appendix A) was shown to participants and their ITD was measured with a 7-point scale. Since intention can be defined as \u0026ldquo;the person’s motivation in the sense of his/her conscious plan to exert effort to carry out a behaviour\u0026rdquo; (Eagly and Chaiken, 1993, p. 168) or as \u0026ldquo;an individual’s conscious plan to make an effort to purchase a brand\u0026rdquo; (Spears and Singh, 2004, p. 56), intention to download is the person’s motivation in the consciously planning to make an effort to download an app (Pol, 2015, p. 18). The screenshots images were taken from Google\u0026rsquo;s Play Store, considering it is the most common app store on the most widespread mobile OS, which is Android, as shown in Fig. 1.1, and supposing that the majority of respondents would have been familiar with it.\nBefore being exposed to the manipulated variables, survey respondents were introduced to the questionnaire and asked some questions aimed to measure their involvement in apps as products (the online survey is integrally available in Appendix B). Firstly, their involvement in mobile apps in general and in the app download process were measured using Laurent and Kapferer’s scale, as presented in Mittal (1995, p. 670). Then, the involvement in the specific app category chosen for the experimental purpose was quantified through a reduced version, presented in Celuch and Taylor (1999, p. 120), of Zaichkowsky (1994, p. 70)\u0026rsquo;s involvement scale, which was already used for mobile apps in Kang et al. (2015, p. 213).\nThe app typology used in the experiment was mobile scanner apps, which allow to digitize physical documents through the smartphone camera and usually to export them in a digital format (e.g. PDF) in just a few seconds, and it was chosen for different reasons. Firstly, this category is quite fragmented, with many apps with small market shares, and the majority of the available options share similar basic features, with just a few exceptions (e.g. better OCR): in order to empirically prove this hypothesis, a preliminary analysis of the scanner apps segment on the Play Store (reported in Appendix C) was conducted and a sample of 24 apps was taken in consideration. Clearly, these apps are relatively easy to develop and there are low entry barriers for companies with the desire to enter this segment. Secondly, mobile scanner apps are subject to really low network effects, allowing to rule out the influence of network externalities and to measure the popularity variable effect by itself. Another important element was the presence in the category of an app developed by a company, Adobe, which is known worldwide for its software. Then, mobile scanner apps are non-hedonic apps and in this case review ratings become more important than ranks (Liu et al., 2014, p. 345). Furthermore, the fact that these apps act as still, at the moment, imperfect substitutes of traditional scanner machines may lead to think they are used by consumers with low involvement in the digitizing process, in line with the quick and impulsive buying setting described by Buck et al. (2014, p. 5), even if, since they belong to the productivity apps category, there is also evidence in the opposite direction (e.g. Sällberg et al., 2022, p. 10). Finally, the vast majority of these apps are offered with the freemium model, meaning that essential features are for free and, consequently, price is not a variable.\nThe order in which the three survey slides in which the app presentation pages of a mobile scanner app were included to measure users\u0026rsquo; ITD was random and also which one out of the two screenshots corresponding to values of a variable was then randomly defined.\nAfter these three slides, a manipulation check was also included, in a compromise (and not ideal) version designed to test the effectiveness of manipulations in the experimental design, but also not to give away the purpose of the data collection. Finally, the last slide included some measures of respondents\u0026rsquo; demographics.\nIn conclusion, by conducting an analysis of variance and using three different regression models for each one of the three focal variables, the experiment aims to investigate their relation, together with the moderation of three involvement measures, with ITD. In addition, by comparing results from models across different variables, it is possible to determine which one is the most and the least effective variable. Finally, by introducing a variable that approximates comparison, the consequences of an app being shown as first or after other manipulated apps are investigated.\n3.1.1 Variables As anticipated, this research aims to determine how three variables, reflected in some elements of the app presentation page on mobile app stores, impact users\u0026rsquo; choice to download an app. In this experimental design, app reputation, popularity and developer\u0026rsquo;s brand are included in form of three binary variables. In order to make these variables manipulation effective, it is necessary to define correctly the two values for each one of the binary variables.\nApart from the three main variables, also involvement measures are included in the analysis models: definitions for involvement variables are available in the variables summary in Table 3.1, as well as for the variable reflecting comparison of an app with others. Since ITD was measured after the exposure to each one of the three main independent variables, three corresponding dependent variables (also defined in Table 3.1) resulted from the data collection and they are separately used in different analysis models.\nTable 3.1: Summary of all variables with definitions and descriptive statistics. Note: this table is also available as an image.\nVariable Definition Descriptive Statistics Mean S.D. Min Max Dependent Variables $$y_{rep}$$ Measured ITD of the apps with manipulated reputation 4.0468 1.7077 1 7 $$y_{pop}$$ Measured ITD of apps with manipulated popularity 4.5255 1.5675 1 7 $$y_{brand}$$ Measured ITD of apps with manipulated developer\u0026rsquo;s brand 4.9674 1.5863 1 7 Independent Variables $$x_{rep}$$ Binary variable, app reputation in terms of average review ratings: high (=1) vs. low (=0) 0.4847 0.4998 0 1 $$x_{pop}$$ Binary variable, app popularity in terms of number of downloads: high (=1) vs. low (=0) 0.5031 0.5000 0 1 $$x_{brand}$$ Binary variable, developer\u0026rsquo;s brand strength and fame: strong (=1) vs. weak (=0) 0.5214 0.4995 0 1 $$inv\\_app$$ Mean-centered involvement measure inmobile apps (as product category) 0 1.0523 -3.6619 1.6715 $$inv\\_dl$$ Mean-centered involvement measure in the app download process 0 1.2974 -3.6933 2.0367 $$inv\\_cat$$ Mean-centered involvement measure in the experimental app category (mobile scanner apps) 0 1.0629 -3.4802 2.5198 $$comp_i$$ Dummy variable, position in which i=rep, pop, brand was shown to users: first (=0) vs. second or third (=1) 0.6667 0.4714 0 1 App Reputation. In this research, app reputation (rep) is defined in terms of the average review rating in the five-star scale currently used on the Play Store. Taking into consideration the dynamics that affect online review ratings, such as, for example, reputation inflation and the not linear impact of good and bad reviews, the threshold above which a review rating is considered to be high cannot be set at the middle of the scale. Indeed, similarly to Roma and Ragaglia (2016, p. 180), this limit was set at the four-star level. This decision was also supported by the preliminary analysis of mobile scanner apps (reported in Appendix C), which shows how the large majority of the most popular scanner apps on the Play Store have average ratings above 4.5 stars. As a consequence, an average rating of 4.9 stars was defined as a high app reputation, while a value of 3.4 was defined as low. In both cases, the number of reviews was set to the same medium-high level, in order to give credibility to the presented rating.\nApp Popularity. As already discussed, app popularity (pop) can be reflected in many elements on an app store. In this case, since the aim was to directly investigate the specific moment in which users take the download decision, the number of downloads was chosen as a metric, also considering it is presented in a spotlighted position in the app page. The number of downloads is presented on the Play Store in form of the last downloads threshold that the app achieved and overcame (e.g. 1M+, where M stands for million), consequently the issue consists in defining above which downloads bracket an app can be considered popular. Since the analysis in Appendix C already took into consideration only popular apps, a biased sample that drives upwards the average downloads number, and the mode out that group was 10M+, the decision was to set this level as a high app popularity. In contrast, the low app popularity level was set to 10K+ (where K means thousand).\nDeveloper\u0026rsquo;s Brand. The developer\u0026rsquo;s brand (brand) affects multiple elements in the app presentation page (app icon, app name and developer\u0026rsquo;s name). In this experiment, a brand was defined as strong when it enjoyed worldwide fame, also outside the mobile ecosystem, and Adobe, which is the eighth largest software company in the world by revenue, with more than 16 billion in 2022 (Murphy and Contreras, 2022), perfectly fits the definition. On the other side, brands were considered weak when they included generic references to their products, without a strong brand identity (e.g. \u0026ldquo;Office Tools\u0026rdquo; or \u0026ldquo;Tools \u0026amp; Utilities Apps\u0026rdquo;).\n3.1.2 Sample and Data Collection In total 545 participants have taken part in the research questionnaire (available in Appendix B), while 491 of them completed it, with a completion rate of 90.09 percent, and, consequently, were included in the sample. The majority of respondents in the sample were between the 18 and 24 years old (59.27%), while the rest was between 25 and 34 (28.11%), between 35 and 44 (5.91%), between 45 and 54 (2.85%), over 55 (2.04%) or under 18 years old (1.83%). More often respondents were females (63.34%) and less frequently males (34.83%), non-binary people (0.81%) or preferred not to specify (1.02%). Regarding their occupation, students were the most common category (69.45%), followed by full-time (19.55%) and part-time employees (5.70%), self-employed workers (2.65%), unemployed (0.81%), pensioners (0.41%) and participants with other occupations (1.43%). Finally, in terms of education level, the majority in the sample had achieved a bachelor\u0026rsquo;s degree (38.70%), while others had a high school diploma (35.85%), a master\u0026rsquo;s degree (18.94%) or a doctorate (1.83%). Some participants didn\u0026rsquo;t conclude high school (2.85%) or undertook other educational paths (1.83%).\nOverall, ITD was measured three times for each one of the 491 participants, for a total of 1473 combined values of yrep, ypop and ybrand.\n3.1.3 Data Analysis Since this data collection corresponds to a repeated measures experimental model, the first step consisted in running an analysis of variance (ANOVA) among the three levels of different groups of respondents from which we are trying to achieve some results. In favour of better describing the ANOVA and also the following models, it is necessary to define i, j, k = rep, pop, brand with i ≠ j ≠ k. Then, it is useful to redefine xi as xin and yi as yin, where n = 1, 2, 3 is the random position (first, second or third) in which users were exposed to the variable i. An ANOVA investigating variance across groups defined by different values on three levels (i, xi and n) that are supposed to affect each measure of yin may help in understanding if some effects on the dependent variable exist and where they come from.\nThereafter, in order to derive more precise results, three different linear regression models were used for each one of the focal variables, to cover all possibilities and for a total of nine regressions. Now, the simplest regression model (Model 1) can be defined as:\nwhere u is the error term. The data set used for Model 1 (M1) for i includes only observations in which respondents were exposed to i before j and k (n = 1), in a way that the resulting ITD cannot be biased by comparison with other apps. In addition, it allows to verify the moderating effects of various measures of involvement on the relationship between the focal predictor and ITD. This model is the main source of evidence for the effects of the focal independent variable xi and the role of involvement as moderator.\nThen, Model 2 (M2) allows for two-way interactions also between different focal independent variables and takes into account only observations with n = 3, in the following way:\nwith m = 1, 2. The main objective of Model 2 is to explore whether there is any interaction between each couple of focal variables, reflecting comparison effects which depend on their alternatively high or low values.\nFinally, Model 3 (M3) can be defined considering all observations in the sample, independently from any specific value of n and introducing the compi variable and its interaction:\nIn the case of equation (3.3), differently from (3.1) and (3.2), n can take values which are not only 1 or 3, meaning that, while it allows to use a larger sample, it needs to include the compi variable (presented in Table 3.1), in order to take in account the possibility that results could be affected by comparison with xj and xk, which are not included in Model 3.\nConcluding, after analyzing the preliminary results of the ANOVA, the comparison of coefficients resulting from models with the same value of i, together with the coefficient of compi from Model 3, helps to reach a further objective of this research, which consists in understanding how an approximation of comparison impacts the customer choice, while differences in coefficients from models depending concurrently on i, j or k could be explained by diverse effectiveness of app reputation, app popularity and brand.\n3.2 Results From the ANOVA emerges how variance is statistically significant across groups defined by different focal variables i (F(2, 4) = 42.187, p \u0026lt; 0.01), by different levels of xi (F(1, 4) = 60.841, p \u0026lt; 0.01) and by different presentation positions n (F(2, 4) = 2.369, p \u0026lt; 0.1), meaning that these three dimensions are likely to predict differences in the dependent variables yin. In addition, also the results regarding the three-way interaction i × xi × n are statistically significant (F(4) = 2.681, p \u0026lt; 0.05) and provide further foundations for subsequent analyses.\nThen, the following step consisted of three different models and nine linear regressions. All information from these regressions, including R2 values and manipulation check ratios, are available in Table 3.2. Model 1 included in turn a focal variable (corresponding to different values of i) and involvement measures, together with their interaction terms. From the first model emerges how, when the app was shown as first (n = 1), the developer\u0026rsquo;s brand was the most effective in predicting ITD (β = 1.5289, p \u0026lt; 0.01), followed by reputation (β = 0.4986, p \u0026lt; 0.1) and popularity (β = 0.3611, p \u0026lt; 0.1). Regarding their interactions with involvement, a positive moderating effect of involvement in the download process on reputation was found (β = 0.3861, p \u0026lt; 0.1), a similar effect to the positive one operated by involvement in the specific app category on the developer\u0026rsquo;s brand (β = 0.4001, p \u0026lt; 0.1).\nThe ITD data included in Model 2 was measured after respondents were exposed to the third app (n = 3), which contained a manipulated value of a random one out of reputation, popularity and brand. The main aims of this model were to determine whether the selected focal variable could be moderated by the values of the other two to which users were previously exposed and also to verify if the effectiveness on ITD remained unchanged if compared to Model 1. In the results from Model 2, there was a reversal between the developer\u0026rsquo;s brand (β = 1.0347, p \u0026lt; 0.05) and reputation (β = 1.3351, p \u0026lt; 0.01), which became the most important element of the three, while the effect of popularity was statistically insignificant. No significant evidence of any moderation effect between one focal independent variable on the others surfaced, while there was evidence, confirmed also by Model 3, of a positive moderation of involvement in apps as generic products on all three focal variables xrep, xpop and xbrand.\nFinally, regressions based on Model 3 included all observations, independently from the position in which apps were shown. The main objective was to investigate the moderating effect of the variable compi, while also confirming the levels of effectiveness of x_i, as derived from Model 1 and 2. Out of Model 3 emerged that, when accounting for all the observations (independently from n values), the developer\u0026rsquo;s brand was the best predictor of ITD (β = 1.5028, p \u0026lt; 0.01), followed by reputation (β = 0.4576, p \u0026lt; 0.1) and popularity (β = 0.3677, p \u0026lt; 0.1). Furthermore, there is evidence of a statistically relevant and negative moderating effect of comparison on brand effectiveness (β = -0.7454, p \u0026lt; 0.05) in predicting ITD.\nTable 3.2: Resulting coefficients and standard errors of all regressions. Note: this table is also available as an image.\nModel (i) M1 (rep) M1 (pop) M1 (brand) M2 (rep) M2 (pop) M2 (brand) M3 (rep) M3 (pop) M3 (brand) (Intercept) 3.86914*** (0.18683) 4.63458*** (0.15161) 4.13417*** (0.17373) 3.69158*** (0.26951) 4.14837*** (0.30527) 4.341602*** (0.333232) 3.906095*** (0.190949) 4.64269*** (0.15454) 4.118051*** (0.178167) $$x_{rep}$$ 0.49862* (0.26259) 1.33507*** (0.41115) 0.25362 (0.37742) -0.285980 (0.360243) 0.457599* (0.268971) $$x_{pop}$$ 0.36110* (0.21478) -0.21677 (0.31337) 0.53397 (0.46800) 0.406845 (0.363070) 0.36773* (0.21838) $$x_{brand}$$ 1.52889*** (0.23554) -0.30136 (0.31366) -0.18598 (0.37590) 1.034683** (0.432264) 1.502819*** (0.240948) $$inv\\_app$$ 0.09502 (0.17658) 0.3581** (0.16063) 0.23275 (0.15521) -0.01990 (0.15111) -0.18401 (0.20203) -0.188448 (0.183269) -0.028741 (0.099168) 0.10192 (0.09446) 0.083030 (0.093585) $$inv\\_dl$$ -0.21222 (0.15674) -0.2292* (0.1235) -0.09778 (0.13349) 0.04958 (0.11528) -0.2527* (0.1505) -0.034777 (0.139169) 0.008458 (0.080334) -0.16296** (0.07627) -0.004562 (0.077751) $$inv\\_cat$$ 0.35115* (0.20204) 0.22163 (0.14454) -0.04189 (0.15636) 0.12016 (0.14256) 0.51847*** (0.19119) 0.287315 (0.173767) 0.106364 (0.099436) 0.29048*** (0.09247) 0.067097 (0.089394) $$comp_i$$ -0.353854 (0.227152) -0.28259 (0.19705) 0.473583** (0.211797) $$x_{i}×x_{rep}$$ -0.74373 (0.53923) 0.251013 (0.507205) $$x_{i}×x_{pop}$$ -0.37078 (0.46066) -0.676553 (0.511097) $$x_{i}× x_{brand}$$ 0.16603 (0.46738) 0.39159 (0.54648) $$x_{i}×inv\\_app$$ 0.2172 (0.26755) -0.0579 (0.23277) 0.24104 (0.21184) 0.28639 (0.22414) 0.57307** (0.26890) 0.540929** (0.260608) 0.278238* (0.146581) 0.29718** (0.13356) 0.214755 (0.131852) $$x_{i}× inv\\_dl$$ 0.38614* (0.23255) 0.10327 (0.16633) -0.03326 (0.17832) -0.14902 (0.17801) 0.08568 (0.22437) -0.071512 (0.193776) -0.004344 (0.116336) 0.00806 (0.10656) -0.174237 (0.105738) $$x_{i}× inv\\_cat$$ -0.10643 (0.27486) 0.06545 (0.20051) 0.40013* (0.23065) 0.18534 (0.21138) -0.42987 (0.27914) 0.006635 (0.253132) 0.200011 (0.142551) -0.11139 (0.13035) 0.158655 (0.128399) $$x_{i}× comp_i$$ 0.510013 (0.322484) -0.39837 (0.27846) -0.745391** (0.289854) R2 0.09523 0.139 0.2885 0.1874 0.1417 0.1533 0.098 0.1056 0.1555 Adj. R2 0.05031 0.1057 0.2542 0.1388 0.06958 0.08682 0.08112 0.08885 0.1397 Observations (N) 149 189 153 196 143 152 491 491 491 Manipulation Check (%) 75.84 62.43 68.63 79.08 55.94 63.16 77.19 60.49 67.41 ***p\u0026lt;0.01, **p\u0026lt;0.05, *p\u0026lt;0.1 3.3 Discussion and Managerial Implications The results indicate that the developer\u0026rsquo;s brand is generally the most effective factor in determining dl decision if compared to reputation (review ratings) and popularity (number of dls), but effectiveness varies on presentation order and information available to users (M1, M3). An unexpected outcome is the scarce (and almost zero) effectiveness of popularity, even if this setting had very low network effects. A hypothesis on the reason why popularity is not frequently taken into consideration by users may rest in the necessity for platform owners (as Google, in this case) to keep the ecosystem competitive and attractive for new developers, avoiding single-firm dominated markets and preserving information transparency on the platform (Boudreau et al., 2022, p. 1379), in particular when network effects are present: in fact, the format in which quantity of downloads is presented (e.g. 10M+ vs. 10K+) may be intentionally not very powerful in communicating the magnitude of popularity differences across apps and it may go unnoticed when users are not reviewing information carefully.\nThe high effectiveness of brands in shaping users\u0026rsquo; choices, even in an online market where a wide range of information is available, opens up relevant opportunities for established software companies. In fact, developing different apps, covering many different segments, already proved to be a successful strategy, since Lee and Raghu (2012, p. 161) pointed out how diversifying app portfolios across selling categories is a key determinant of sales performance.\nWhen users are exposed to information about multiple apps and they can consequently compare them, reputation becomes the most effective element (M2), as also shown by the parallel in Fig. 3.1 between ITD for apps shown as first (n = 1) and those shown as third (n = 3). Users may rely more on eWOM when they have the possibility to consider a wider range of options and they just may need some anchors to better understand apps\u0026rsquo; rating levels. An important cue for making an informed decision can be often found in past experience, but, when users have no previous experience with the specific product, they often rely on social experiences, as expressed in reviews (Burgers et al., 2016, p. 329). As a consequence, it is important, in particular for new entrants in an app category that are not likely to be shown in top search results, to publish an app when it is mature enough to avoid the common phenomenon of reputation backlash, in terms of initial low ratings. On the other side, brands become less predictive of ITD when comparison is introduced in the model (as evidenced by the coefficient of compi in M3): a popular brand may combine well with the \u0026ldquo;take the first\u0026rdquo; heuristic, which has already proven to be effective in the case of mobile apps (Dogruel et al., 2015, p. 139), and be penalised when users are actively collecting information.\nFigure 3.1: Mean-centered ITD (yi) depending on high vs. low values (x) of developer\u0026rsquo;s brand, reputation and popularity, with n = 1 and n = 3. Some other interesting evidence regards the moderation effects of involvement on the three focal variables. First, there is predictable and generalized evidence that higher involvement in apps as products corresponds to a higher effectiveness of all three reputation, popularity and brand levels in driving users\u0026rsquo; ITD (M2, M3). Second, when users are highly involved in the app download process, they tend to rely more on reputation, probably because they scrutinise information more closely. Last, when the involvement in the specific app category (mobile scanner apps, in this case) is high and users consequently perceive a higher risk in the app choice, also considering that scanner apps are productivity apps, frequently used for work (Sällberg et al., 2022, p. 10), they appear to preferably choose a strong and known brand (M1), which is a commonly used as a cue when attempting to mitigate it (Matzler et al., 2008, p. 158).\n3.4 Limitations This experimental approach presents some limitations. The first and major limitation lies in the experimental design: in fact, while investigating the effect of three factors on ITD, the selected one was not a traditional three-way factorial design and just one variable was manipulated at each measurement of ITD, while, at the same time, the others were set to intermediate values. As a consequence, estimates of interactions among factors were not ideal, but, on the other side, it allowed to effectively collect results on comparison effects. Consequently, despite the repeated measure design, it was not possible to include all observations of yi in each sample used for each regression, since every measure of ITD was strictly related to a manipulated value of reputation, popularity or developer\u0026rsquo;s brand (i) happening in the specific case.\nSome other limitations lie in the data collection. For example, the screenshot images used in the questionnaire were all collected from Google Play Store, the most used app store, meaning that users of other platforms, for example, Apple iOS, could be used to a different interface and to look for elements in other positions, affecting, therefore, the resulting ITD. In addition, some people, such as those with high levels of involvement, for example, may find limiting the experimental setting, in which many important elements were not considered or even included.\nFinally, the sample may represent a limitation. In fact, the large majority of respondents were students and they were aged between 18 and 24 and results could be biased by this unequal balance of composition. Nowadays, smartphones are widespread across all customer age groups, but this doesn\u0026rsquo;t mean that these groups all share the same levels of involvement and rely on the same factors when making decisions that concern downloading an app or not. Additionally, while this is not certain, participants can be assumed to be mainly from Germany and Italy and there is some evidence that behaviour on app stores and factors relevant to the purchase decision vary across countries (Kübler et al., 2018, p. 37), meaning that results may vary depending on sample geographical origin and cultural background.\nChapter 4. Conclusion 4.1 Summary As a first step, in Chapter 1, an extensive background on the mobile ecosystem, its participants and its dynamics was provided, before reviewing, in Chapter 2, previous literature on user behaviour and determinants of downloads on mobile app stores. Finally, in Chapter 3, the empirical approach of this research was discussed, together with its results. The outcome of this experiment shows how the developer\u0026rsquo;s brand is generally the most decisive element, but, when a comparison between apps is introduced in the model, it loses its effectiveness and reputation becomes more relevant. Popularity and peer influence, in the context of low network effects, appear to be unexpectedly ineffective in driving users\u0026rsquo; decisions across all models. Moreover, users\u0026rsquo; involvement in the download process reinforces the effect of reputation, while, when customers are more involved in the specific app category and perceive a higher risk, apps from established brands are more likely to benefit. Finally, it was not possible to find clear evidence of any two-way interaction between reputation, popularity and brand.\n4.2 Future Directions This research results add up to the substantial literature that already covered customer behaviour on online marketplaces, but some elements remain unexplored. For example, a better experimental design aimed to explore factorial interactions should be developed. Furthermore, peer influence effectiveness and its dynamics on app stores, both in the presence and absence of network effects, should be better investigated.\nBibliography Ali, M., Joorabchi, M. E., \u0026amp; Mesbah, A. (2017). Same app, different app stores: A comparative study. In 2017 IEEE/ACM 4th International Conference on Mobile Software Engineering and Systems (MOBILESoft) (pp. 79-90), IEEE.\nAnderson, C. (2006). The long tail: Why the future of business is selling less of more. Hachette UK.\nArora, S., Ter Hofstede, F., \u0026amp; Mahajan, V. (2017). The implications of offering free versions for the performance of paid mobile apps. Journal of Marketing, 81(6), 62-78.\nBasole, R. C., \u0026amp; Karla, J. (2012). Value transformation in the mobile service ecosystem: A study of app store emergence and growth. Service Science, 4(1), 24-41.\nBöhm, S., \u0026amp; Schreiber, S. (2014). Mobile App Marketing: A Conjoint-based Analysis on the Importance of App Store Elements. In CENTRIC 2014: The Seventh International Conference on Advances in Human-oriented and Personalized Mechanisms, Technologies, and Services.\nBoudreau, K. J., Jeppesen, L. B., \u0026amp; Miric, M. (2022). Competing on freemium: Digital competition with network effects. Strategic Management Journal, 43(7), 1374-1401.\nBrynjolfsson, E., Hu, Y., \u0026amp; Smith, M. D. (2003). Consumer surplus in the digital economy: Estimating the value of increased product variety at online booksellers. Management science, 49(11), 1580-1596.\nBrynjolfsson, E., \u0026amp; Smith, M. D. (2000). Frictionless commerce? A comparison of Internet and conventional retailers. Management science, 46(4), 563-585.\nBuck, C., Horbel, C., Germelmann, C. C., \u0026amp; Eymann, T. (2014). The Unconscious App Consumer: Discovering and Comparing the Information-seeking Patterns among Mobile Application Consumers. In ECIS.\nBurgers, C., Eden, A., de Jong, R., \u0026amp; Buningh, S. (2016). Rousing reviews and instigative images: The impact of online reviews and visual design characteristics on app downloads. Mobile Media \u0026amp; Communication, 4(3), 327-346.\nCeci, L. (2022). Google Play: annual consumer spending on mobile apps 2016-2021. Statista, https://www.statista.com/statistics/444476/google-play-annual-revenue/ (Accessed in January 2023).\nCeci, L. (2022). Number of apps available in leading app stores Q3 2022. Statista, https://www.statista.com/statistics/276623/number-of-apps-available-in-leading-app-stores/ (Accessed in January 2023).\nCeluch, K. G., \u0026amp; Taylor, S. A. (1999). Involvement with services: An empirical replication and extension of Zaichkowsky\u0026rsquo;s personal involvement inventory. The Journal of Consumer Satisfaction, Dissatisfaction and Complaining Behavior, 12, 109-122.\nChen, H., Lachaud, K., \u0026amp; Zhou, W. (2022). The sales effect of “Free App of the Day” on Amazon Appstore: An empirical study. Digital Business, 2(2), 100020.\nChevalier, J. A., \u0026amp; Mayzlin, D. (2006). The effect of word of mouth on sales: Online book reviews. Journal of marketing research, 43(3), 345-354.\nCortimiglia, M. N., Ghezzi, A., \u0026amp; Renga, F. (2011). Mobile applications and their delivery platforms. IT Professional, 13(5), 51-56.\nDogruel, L., Joeckel, S., \u0026amp; Bowman, N. D. (2015). Choosing the right app: An exploratory perspective on heuristic decision processes for smartphone app selection. Mobile Media \u0026amp; Communication, 3(1), 125-144.\nEagly, A. H., \u0026amp; Chaiken, S. (1993). The psychology of attitudes. Harcourt brace Jovanovich college publishers.\nEricsson (2022). Ericsson Mobility Report, https://www.ericsson.com/en/reports-and-papers/mobility-report/reports/november-2022 (Accessed in January 2023).\nFilippas, A., Horton, J. J., \u0026amp; Golden, J. (2018). Reputation inflation. In Proceedings of the 2018 ACM Conference on Economics and Computation (pp. 483-484).\nFinkelstein, A., Harman, M., Jia, Y., Martin, W., Sarro, F., \u0026amp; Zhang, Y. (2017). Investigating the relationship between price, rating, and popularity in the Blackberry World App Store. Information and Software Technology, 87, 119-139.\nFriedman L. (2013). The App Store turns five: A look back and forward. Macworld, https://www.macworld.com/article/221393/the-app-store-turns-five-a-look-back-and-forward.html (Accessed in January 2023).\nGokgoz, Z. A., Ataman, M. B., \u0026amp; van Bruggen, G. H. (2021). There’s an app for that! Understanding the drivers of mobile application downloads. Journal of Business Research, 123, 423-437.\nGoldfarb, A., \u0026amp; Tucker, C. (2019). Digital economics. Journal of Economic Literature, 57(1), 3-43.\nHalmenschlager, C., \u0026amp; Waelbroeck, P. (2014). Fighting free with free: Freemium vs. piracy. Piracy (November 29, 2014).\nHorton, J., \u0026amp; Golden, J. (2015). Reputation inflation in an online marketplace. New York I, 1.\nHyrynsalmi, S., Mäkilä, T., Järvi, A., Suominen, A., Seppänen, M., \u0026amp; Knuutila, T. (2012). App store, marketplace, play! an analysis of multi-homing in mobile software ecosystems. Jansen, Slinger, 59-72.\nIhlwan, M. (2007). LG ties up with PRADA for new phone. Bloomberg, https://www.bloomberg.com/news/articles/2007-01-17/lg-ties-up-with-prada-for-new-phone (Accessed in January 2023).\nJung, E. Y., Baek, C., \u0026amp; Lee, J. D. (2012). Product survival analysis for the App Store. Marketing Letters, 23(4), 929-941.\nKang, J. Y. M., Mun, J. M., \u0026amp; Johnson, K. K. (2015). In-store mobile usage: Downloading and usage intention toward mobile location-based retail apps. Computers in Human Behavior, 46, 210-217.\nKoch, S., \u0026amp; Kerschbaum, M. (2014). Joining a smartphone ecosystem: Application developers’ motivations and decision criteria. Information and Software Technology, 56(11), 1423-1435.\nKouris, I., \u0026amp; Piller, F. T. (2014). App platforms as two-sided markets: analysis and modeling of application distribution platforms for mobile devices. Lehrstuhl für Betriebswirtschaftslehre, insbesondere Technologie-und Innovationsmanagement.\nKübler, R., Pauwels, K., Yildirim, G., \u0026amp; Fandrich, T. (2018). App popularity: Where in the world are consumers most sensitive to price and user ratings?. Journal of Marketing, 82(5), 20-44.\nLambrecht, A., Goldfarb, A., Bonatti, A., Ghose, A., Goldstein, D. G., Lewis, R., Rao, A. and Sahni, N. \u0026amp; Yao, S. (2014). How do firms make money selling digital goods online?. Marketing Letters, 25(3), 331-341.\nLaricchia, F. (2022). Market share of mobile operating systems worldwide 2009-2022. Statista, https://www.statista.com/statistics/272698/global-market-share-held-by-mobile-operating-systems-since-2009/ (Accessed in January 2023).\nLee, G., \u0026amp; Raghu, T. S. (2014). Determinants of mobile apps\u0026rsquo; success: Evidence from the app store market. Journal of Management Information Systems, 31(2), 133-170.\nLiang, T. P., Li, X., Yang, C. T., \u0026amp; Wang, M. (2015). What in consumer reviews affects the sales of mobile apps: A multifacet sentiment analysis approach. International Journal of Electronic Commerce, 20(2), 236-260.\nLiu, C. Z., Au, Y. A., \u0026amp; Choi, H. S. (2014). Effects of freemium strategy in the mobile app market: An empirical study of google play. Journal of management information systems, 31(3), 326-354.\nMatzler, K., Grabner‐Kräuter, S., \u0026amp; Bidmon, S. (2008). Risk aversion and brand loyalty: the mediating role of brand trust and brand affect. Journal of product \u0026amp; brand management, 17(3), 154-162.\nMittal, B. (1995). A comparative analysis of four scales of consumer involvement. Psychology \u0026amp; marketing, 12(7), 663-682.\nMoe, W. W., \u0026amp; Trusov, M. (2011). The value of social dynamics in online product ratings forums. Journal of Marketing Research, 48(3), 444-456.\nMüller, R. M., Kijl, B., \u0026amp; Martens, J. K. (2011). A comparison of inter-organizational business models of mobile app stores: There is more than open vs. closed. Journal of theoretical and applied electronic commerce research, 6(2), 63-76.\nMurphy A., \u0026amp; Contreras I. (2022). The Global 2000. Forbes, https://www.forbes.com/lists/global2000/ (Accessed in January 2023).\nNan, G., Wu, D., Li, M., \u0026amp; Tan, Y. (2018). Optimal Freemium Strategy for Information Goods in the Presence of Piracy. Journal of the Association for Information Systems, 19(4), 266-305.\nPiccoli, G., \u0026amp; Pigni, F. (2019). Information Systems for Managers With Cases, Edition 4.0. Prospect Press, Inc..\nPol, M. (2015). App icon preferences: The influence of app icon design and involvement on quality and intention to download. Master\u0026rsquo;s thesis, University of Twente.\nReavis Conner, K., \u0026amp; Rumelt, R. P. (1991). Software piracy: An analysis of protection strategies. Management science, 37(2), 125-139\nRochet, J. C., \u0026amp; Tirole, J. (2003). Platform competition in two-sided markets. Journal of the european economic association, 1(4), 990-1029.\nRoma, P., \u0026amp; Ragaglia, D. (2016). Revenue models, in-app purchase, and the app performance: Evidence from Apple’s App Store and Google Play. Electronic commerce research and applications, 17, 173-190.\nRubin, A. D. (1995). Trusted distribution of software over the Internet. In Proceedings of the Symposium on Network and Distributed System Security (pp. 47-53), IEEE.\nRuiz, I. J. M., Nagappan, M., Adams, B., Berger, T., Dienst, S., \u0026amp; Hassan, A. E. (2015). Examining the rating system used in mobile-app stores. Ieee Software, 33(6), 86-92.\nSalganik, M. J., Dodds, P. S., \u0026amp; Watts, D. J. (2006). Experimental study of inequality and unpredictability in an artificial cultural market. science, 311(5762), 854-856.\nSällberg, H., Wang, S., \u0026amp; Numminen, E. (2022). The combinatory role of online ratings and reviews in mobile app downloads: an empirical investigation of gaming and productivity apps from their initial app store launch. Journal of Marketing Analytics, 1-17.\nSpears, N., \u0026amp; Singh, S. N. (2004). Measuring attitude toward the brand and purchase intentions. Journal of current issues \u0026amp; research in advertising, 26 (2), 53-66.\nTimmerman, J. E., \u0026amp; Shepherd, I. (2016). Does eWOM affect demand for mobile device applications?. Journal of Marketing Development \u0026amp; Competitiveness, 10(3).\nWang, Y., Song, J., \u0026amp; Aguirre-Urreta, M. (2015). An empirical investigation of factors impacting application downloads in mobile app stores. SIGHCI Proceedings, 1-5.\nZaichkowsky, J. L. (1994). The personal involvement inventory: Reduction, revision, and application to advertising. Journal of advertising, 23(4), 59-70.\nZheng, P., \u0026amp; Ni, L. (2010). Smart phone and next generation mobile computing. Elsevier.\nAppendices A App Page Screenshots This online study involves the use of manipulated screenshots images of an app presentation page on Google\u0026rsquo;s Play Store (two for each one of the three main variables). Each couple of screenshots of existing apps was manipulated in a way that the assigned variable could take different values. Only the elements directly related to that variable were admitted to change across each couple of presentation pages. As shown in Fig. A.1, in screenshot images dealing with app reputation (high vs. low) the average rating is the only element manipulated. Similarly, also in Fig. A.2, only the number of downloads is manipulated for app popularity (high vs. low). Finally, Fig. A.3 shows how the elements dealing with the developer\u0026rsquo;s brand (strong vs. weak) are three: the app\u0026rsquo;s icon, the app\u0026rsquo;s name and the developer\u0026rsquo;s name.\nFigure A.1: Screenshot images of the app presentation page with variation in app reputation. Figure A.2: Screenshot images of the app presentation page with variation in app popularity. Figure A.3: Screenshot images of the app presentation page with variation in developer’s brand. B Experimental Questionnaire Note: in this online survey some presentation pages from Google\u0026rsquo;s Play Store were included. These images are available in Appendix A.\nSlide 0. Presentation page and consent form.\nWelcome!\nThis is an online study on apps and app stores.\nThe expected duration time is less than 5 minutes.\nThrough your participation in this survey, you are making an important contribution to my research. I greatly appreciate your support.\nWhy do we collect and use your data\nThe purpose of this online study is to obtain new insights on app preferences and app store customer behavior.\nHow to contact us\nLuca Bontempi - Student at the University of Tübingen and at the University of Pavia. Email: luca.bontempi@student.uni-tuebingen.de\nIf you would like to obtain more information about the processing of your personal data, please click here\nO I agree to the processing of my personal data in accordance with the information provided herein\nO I don\u0026rsquo;t want to participate O Start the survey\nNote: respondents had to agree and select \u0026ldquo;Start the survey\u0026rdquo; in order to proceed to the next slide.\nSlide 1. Involvement in mobile apps and app download process.\nRead and answer the following questions on apps and app stores.\nWhat do you think about mobile apps?\nMobile apps are very important to me.\nStrongly disagree O 1 O 2 O 3 O 4 O 5 O 6 O 7 Strongly agree\nFor me, mobile apps do not matter.\nStrongly disagree O 1 O 2 O 3 O 4 O 5 O 6 O 7 Strongly agree\nI have a strong interest in mobile apps.\nStrongly disagree O 1 O 2 O 3 O 4 O 5 O 6 O 7 Strongly agree\nWhat do you think about downloading apps?\nI choose the mobile apps to download very carefully.\nStrongly disagree O 1 O 2 O 3 O 4 O 5 O 6 O 7 Strongly agree\nWhich mobile apps I download matters to me a lot.\nStrongly disagree O 1 O 2 O 3 O 4 O 5 O 6 O 7 Strongly agree\nChoosing an app is an important decision for me.\nStrongly disagree O 1 O 2 O 3 O 4 O 5 O 6 O 7 Strongly agree\nSlide 2. Involvement in the mobile scanner apps category.\nConsider mobile apps that allow you to digitize documents through the camera (Cam Scanner Apps). Read the following statement.\nTo me digitizing documents using a mobile scanner app is:\nSelect your personal opinion with regard to the following couples of words.\nimportant O 1 O 2 O 3 O 4 O 5 O 6 O 7 unimportant\nrelevant O 1 O 2 O 3 O 4 O 5 O 6 O 7 irrelevant\nexciting O 1 O 2 O 3 O 4 O 5 O 6 O 7 unexciting\nmeans nothing O 1 O 2 O 3 O 4 O 5 O 6 O 7 means a lot to me\nappealing O 1 O 2 O 3 O 4 O 5 O 6 O 7 unappealing\nfascinating O 1 O 2 O 3 O 4 O 5 O 6 O 7 mundane\nworthless O 1 O 2 O 3 O 4 O 5 O 6 O 7 valuable\nnot needed O 1 O 2 O 3 O 4 O 5 O 6 O 7 needed\nSlide 3, 4 and 5. Main experiment.\nImagine you are looking for a scanner app for your smartphone since you need to digitize a document. In the Android App Store (\u0026ldquo;Google Play Store\u0026rdquo;) you come across the following app. Please carefully review and take into consideration only the available information before you proceed.\nA screenshot image of the app presentation page is shown here.\nIt is very likely that I would download this app.\nPlease indicate your agreement with this statement above.\nO 1 - Fully disagree O 2 O 3 O 4 - Neutral O 5 O 6 O 7 - Fully agree\nNote: Slide 3, 4 and 5 were presented to respondents in randomized order. The screenshot image presented was randomly selected for each slide and variable out of the correspondent couple presented in Appendix A.\nSlide 6. Manipulation check.\nPlease indicate which factors you took into consideration when evaluating the apps on the previous slides.\nO Review rating O Number of downloads O Developer brand O App\u0026rsquo;s name O App\u0026rsquo;s icon O Number of reviews O App\u0026rsquo;s screenshots O PEGI rate O \u0026ldquo;About this app\u0026rdquo; section\nNote: answer options were presented in a randomized order.\nSlide 7. Demographics.\nWhat is your age?\nPlease select your age range.\nO Under 18 O 18-24 years old O 25-34 years old O 35-44 years old O 45-54 years old O Over 55\nWhat is your gender?\nPlease select your gender.\nO Female O Male O Non-binary O Prefer not to say\nWhat is the highest degree or level of school you have completed?\nPlease select your degree level.\nO Less than a high school diploma O High school degree or equivalent O Bachelor’s degree (e.g. BA, BS) O Master’s degree (e.g. MA, MS) O Bachelor’s degree Doctorate (e.g. PhD) O Other\nWhat is your current employment status?\nPlease refer to your main occupation.\nO Student O Employed full-time O Employed part-time O Self-employed O Unemployed O Retired O Other\nSlide 8. Conclusion.\nThe survey is over!\nThank you for your contribution!\nC Preliminary Analysis of Experimental App Category A preliminary analysis of a sample of popular mobile scanner apps was conducted in September 2022. Up to 24 free apps with this function were found to be ranked 300 or higher in two categories (Productivity and Business Tools) on the Play Store. All these apps presented similar names: for example, the app\u0026rsquo;s names always included the words \u0026ldquo;scan\u0026rdquo; or \u0026ldquo;scanner\u0026rdquo;, it included the word \u0026ldquo;PDF\u0026rdquo; in 21 cases and the words \u0026ldquo;doc\u0026rdquo; or \u0026ldquo;documents\u0026rdquo; in 9 cases.\nThe average review rating ranged from 3.2 out of 5 up to 4.9, with a mean value of 4.46 (S.D. = 0.42) and median value of 4.6, while the reviews number was 4.06 million at the highest and 2437 at the lowest, with a mean equal to 446850 (S.D. = 992614) and median value of 92305.\nGoogle\u0026rsquo;s app store reports the number of downloads in the form of the last threshold exceeded by that app. In this sample, apps had 100M+ downloads (where M stands for million) at their best and 50K+ downloads (where K stands for thousand) at their worst. The median value was 10M+. Since this analysis includes only apps that were among a pool of popular ones (top 300) in their categories, many other cam scanner apps with fewer downloads and review ratings are available on app stores.\nFinally, one app in this sample was published by Adobe, a developer known worldwide and mainly for its traditional software. Some other apps were developed by companies which can be considered popular in the mobile apps ecosystem, such as, for example, CamSoft, Simple Design or MobiSystems. All the others were mostly developed by unpopular brands.\nSince the word \u0026ldquo;platform\u0026rdquo; may refer to different elements of the software ecosystem, \u0026ldquo;app (or software) store or marketplace\u0026rdquo; will be used from now on referring to platforms from which software can be purchased and downloaded, as, for example, Google Play Store or Apple App Store for apps or Steam for PC videogames. On the other side, \u0026ldquo;technology platform\u0026rdquo; or just \u0026ldquo;platform\u0026rdquo; will refer in a broader sense to all the software and hardware elements that allow an application to run on a device, often including as its fundamental part the operative system (OS). Finally, \u0026ldquo;two-sided platform\u0026rdquo; may be used to generically refer to any two-sided market, beyond those dealing with software applications.\u0026#160;\u0026#x21a9;\u0026#xfe0e;\n","permalink":"https://lucabontempi.com/blog/determinants_of_download_on_mobile_app_stores/","summary":"\u003ch1 id=\"premise\"\u003ePremise\u003c/h1\u003e\n\u003cp\u003eThis blog post contains the full text of my Master Thesis, written at the \u003cstrong\u003eEberhard Karls University of Tübingen\u003c/strong\u003e and at the \u003cstrong\u003eUniversity of Pavia\u003c/strong\u003e and submitted for the first time during \u003cstrong\u003eFebruary 2023\u003c/strong\u003e.\u003c/p\u003e\n\u003cp\u003eIn \u003cstrong\u003eSeptember 2026\u003c/strong\u003e I reopened this work and reanalyzed the original data. That second pass found \u003cstrong\u003ethree mistakes in the analysis\u003c/strong\u003e (none of them in the data itself), and is documented in \u003ca href=\"https://lucabontempi.com/blog/reopening-my-thesis/\" title=\"I Asked an AI to Review My Thesis. It Found Three Mistakes I Could Have Found Myself\"\u003eI Asked an AI to Review My Thesis\u003c/a\u003e. The text below is the thesis as it was submitted, left unchanged.\u003c/p\u003e","title":"Determinants of Download on Mobile App Stores - An Empirical Analysis"},{"content":"During my stay in Tübingen (Germany), hiking was one of my favourite activities and I have to admit the surrounding area is very suitable to hike.\nTherefore, I created a collection of all the hikes I did and I also selected some of the best ones, which will be immediately presented as Top Hikes.\nEach hike will be presented through the Strava activity, which often also include some pictures, and I\u0026rsquo;ll also include a link to the Komoot activity, since it allows you to download the GPX file, which can be used to replicate the hike.\nEnjoy!\nTop Hike: around Burg Hohenzollern This hike starts at the Bisingen train station, takes you up to the Burg Hohenzollern (which you can visit), then up to the Zoller Horn hill (from where you can have an amazing view of the castle) and back to Bisingen.\nHere it is the Koomot activity.\nTop Hike: Schwarzwald Genießerpfad from Calw This is probably the easiest way to have a little taste of the Black Forest, starting in the nice town of Calw and following a loop tour that includes the city scaffold, where death sentences were carried out until the beginning of the 19th century, and some other cute passages.\nHere it is the Koomot activity.\nTop Hike: around Teckberg This hike was surprisingly good. Firstly, you go up to Teckberg, where there is a little Burg, then you visit some ruins and finally an elevated walkway over a piece of land full of snakes and frogs. This picture was taken close to that site.\nHere it is the Koomot activity.\nTop Hike: around Bad Urach Bad Urach is one of the best hiking destinations. There you will be amazed by the waterfalls and you will visit the Hohenurach ruins.\nHere it is the Koomot activity.\nalong the Danube to Sigmaringen If you have ever seen the Danube in Budapest, you\u0026rsquo;ll be easily surprised by how small it is in Sigmaringen. This hike will lead you along the newborn river, crossing the Teufelsbrücke (Devil\u0026rsquo;s Bridge) and ending in Sigmaringen. A good opportunity to visit the Sigmaringen Castle.\nHere it is the Koomot activity.\nfrom Tübingen WHO to Wurmlinger Kapelle, and back The Wurmlinger Kapelle is located at the opposite end of the Tübingen Schlossberg. The hill is quite nice, with a good view on the nearby plains.\nHere it is the Koomot activity.\nfrom Balingen to Zillhausen (almost) waterfall This hike takes you from Balingen you to the Zillhausen waterfall, which, unfortunately, were dry when I was there. I guess, it is way nicer when there actually is a waterfall.\nHere it is the Koomot activity.\nfrom Tübingen to Schönbuchturm, and back This was probably the craziest one. A 7 hours round trip crossing the whole Schönbuch twice. On the way over I took the path with a panoramic view of the nearby towns, while I got back taking a more internal one.\nHere it is the Koomot activity.\nfrom Tübingen to Herrenberg An easier solution to visit the Schönbuch park is leaving from Tübingen and finishing in Herrenberg. Herrenberg is a cute small town and close to it there is the Schönbuchturm, a panoramica tower with a view on the whole forest.\nHere it is the Koomot activity.\nfrom Tübingen to Entringen Leaving Tübingen in the Hagelloch direction (a village in Tübingen district), with this hike you will quickly explore the south-western corner of the Schönbuch park, concluding in Entringen, a small town bordering with Tübingen.\nHere it is the Koomot activity.\nfrom Tübingen to Dettenhausen via Bromberg Also this hike crosses the Schönbuch park, but, instead of going west, as in the previous ones, it goes north, ending up in Dettenhausen, with also some up and downs on the near hills.\nHere it is the Koomot activity.\naround Rottenburg am Neckar Rottenburg am Neckar is another nice and small town close to Tübingen and this hike takes you around the surrounding areas.\nHere it is the Koomot activity.\naround Schönbuch (Bebenhausen area) This hike explores the really nice Schönbuch wooded area around Bebenhausen (a village in Tübingen district), with a good opportunity to spot some wild animals.\nHere it is the Koomot activity.\naround Hagelloch to Bebenhausen Before arriving in Bebenhausen again, this hike crosses the area surrounding Hagelloch.\nHere it is the Koomot activity.\nfrom WHO to Bebenhausen Probably the easiest hike of all, it is almost mandatory if you are living in WHO. Once you reach Bebenhausen, you\u0026rsquo;ll be able to visit the Bebenhausen Abbey. Even if coming back means going uphill, this remains a quite easy and short hike.\nHere it is the Koomot activity.\nCredits I really want to say thank you to my friend Armin, who has been my trusted hiking guide for months, and his amazing group!\n","permalink":"https://lucabontempi.com/blog/tuebingen_hiking_guide/","summary":"\u003cp\u003eDuring my stay in Tübingen (Germany), hiking was one of my favourite activities and I have to admit the surrounding area is very suitable to hike.\u003c/p\u003e\n\u003cp\u003eTherefore, I created a collection of all the hikes I did and I also selected some of the best ones, which will be immediately presented as Top Hikes.\u003c/p\u003e\n\u003cp\u003eEach hike will be presented through the \u003ca href=\"https://www.strava.com/athletes/4403103\"\u003eStrava activity\u003c/a\u003e, which often also include some pictures, and I\u0026rsquo;ll also include a link to the \u003ca href=\"https://www.komoot.it/user/2218098976141\"\u003eKomoot activity\u003c/a\u003e, since it allows you to download the GPX file, which can be used to replicate the hike.\u003c/p\u003e","title":"A Tübingen Hiking Guide"},{"content":"I created a collection of 36 photos that, in my opinion, best represent my experience in Tübingen (Germany). These pictures cover a period from October 2021 to September 2022 and were mainly taken in Tübingen and in the surrounding area, with just few exceptions coming from some other attractive German cities.\nThe collection is entitled \u0026ldquo;One year in Tü. A Chapter.\u0026rdquo; and it is fully available on my Flickr profile. All the photos are published under the CC BY-NC-SA license.\nHere you can view every photo with a little bit of commentary!\n1/ This is one of my favourite glimpses of Marktplatz.\n2/ The Stocherkahnrennen is probably one of the highlights of the summer semester in Tübingen. It is held (almost) every year since 1956 and the date to save is the Corpus Christi day (on a Thursday, usually in June).\n3/ A view of the Stiftskirche from the Schloss Hohentübingen (more precisely from here).\n4/ When I firstly arrived in Tübingen, in October 2021, Germany was close to the biggest change of the last years. Angela Merkel then retired and Germany started to be a little bit more confused.\n5/ This elevetor (at least twice as old as me) was literally the very first thing I saw in Tübingen.\n6/ WHO and its \u0026ldquo;iconic\u0026rdquo; buildings. I feel like I spent too much time there.\n7/ Maybe I should have posted the Universitätsbibliothek cafeteria, probably I spent more time there than studying\u0026hellip;\n8/ A vintage road bike in front of Neue Aula.\n9/ The inside of Neue Aula, built in 1846.\n10/ Tübingen is a bike friendly city, kind of a small Amsterdam. At least, they have bike lanes\u0026hellip;\n11/ The tunnel crossing the Schloss hill.\n12/ Just some Italians chatting in b\u0026amp;w.\n13/ I didn\u0026rsquo;t expect a car museum to be so good. Mercedes-Benz Museum in Stuttgart.\n14/ In 1477, Eberhard I of Württemberg, whose motto was attempto (\u0026ldquo;I dare\u0026rdquo;), founded the University of Tübingen. — Wikipedia\nThis is his statue in Stuttgart.\n15/ Vfb has been, from that day on, my favourite German team!\n16/ Rothenburg ob der Tauber was probably my favourite city amongst those I visited in Germany.\n17/ The attractions at the Stuttgarter Frühlingsfest.\n18/ I enjoyed some high level tennis at the ATP 250 in Stuttgart. Two-time Olympic gold medalist Andy Murray warming up in the picture.\n19/ Hiking is one of my favourite activities. This picture was taken during a hike, close to the Schopflocher Torfmoor.\n20/ Burg Hohenzollern is one of the most scenic castles in the area. This photo was taken from the Zoller Horn, during a hike.\n21/ Another hiking destination is the Schönbuchturm in Herrenberg. If you are crazy enough you could go there by foot, and come back.\n22/ Herrenberg is a small, but cute, town not far away from Tübingen.\n23/ People enjoying Neckar and the sun.\n24/ A view of Holzmarkt from the bell tower of the Stiftskirche.\n25/ Getting ready for a thursday night in Clubhaus.\n26/ Tübingen and its decorated street lamps.\n27/ This night of the Tübingen Stadtfest was probably the funniest one in town.\n28/ Another fairytale castle in Lichtenstein.\n29/ A detail of the Sigmaringen Castle, built on the Danube. A hike could be a good idea to visit it.\n30/ I swear that most buses are more modern (but less stylish).\n31/ The 2022 European Championships were a multisport event held in Munich this summer. Fabio Jakobsen and Arnaud Démare (in the picture) were first and second in the cycling men\u0026rsquo;s road race.\n32/ I\u0026rsquo;m a great fan of theme parks. Europa-Park is the German response to that big French park (but Gardaland rules).\n33/ This summer, at the European Championships in Köln, the German national team retired the jersey of Dirk Nowitzki, the European basketball GOAT, as of now.\n34/ The most famous view in Tübingen, from Neckarbrücke.\n35/ My empty room, in Fichtenweg, the day I left Tübingen.\n36/36 The Italian border on the Splügen Pass in October, coming back to Italy.\nAll these pictures were taken using a Nikon D5600 or an Oppo Find X2 Neo.\n","permalink":"https://lucabontempi.com/blog/tuebingen_collection/","summary":"\u003cp\u003eI created a collection of 36 photos that, in my opinion, best represent my experience in Tübingen (Germany). These pictures cover a period from October 2021 to September 2022 and were mainly taken in Tübingen and in the surrounding area, with just few exceptions coming from some other attractive German cities.\u003c/p\u003e\n\u003cp\u003eThe collection is entitled \u0026ldquo;\u003ca href=\"https://www.flickr.com/photos/77058054@N07/albums/72177720328778423/\"\u003eOne year in Tü. A Chapter.\u003c/a\u003e\u0026rdquo; and it is fully available on my \u003ca href=\"https://www.flickr.com/photos/77058054@N07/\"\u003eFlickr profile\u003c/a\u003e. All the photos are published under the \u003ca href=\"https://creativecommons.org/licenses/by-nc-sa/4.0/\"\u003eCC BY-NC-SA license\u003c/a\u003e.\u003c/p\u003e","title":"One year in Tübingen. A Chapter. A Photo Collection"},{"content":"Hey there, I\u0026rsquo;m Luca!\nI hold a Master’s in International Business and Entrepreneurship (M.Sc.) from the University of Pavia, where I focused on the digital side of business, and another Master’s in European Management (M.Sc.) from the University of Tübingen. For my thesis, I explored what makes people download certain apps from app stores.\nBefore that, I earned my Bachelor’s in Business Administration (B.Sc.) at the University of Brescia, where, in my thesis, I researched the IT industry and the shift from software as a product to software as a service (in Italian).\nToday, I’m part of IBM, helping partners get the most out of our technology and uncover new opportunities to grow.\nOutside of work, I’m always up for a conversation about tech, video games, movies, or even politics. To recharge, I like running and cycling—nothing beats a long ride or run to clear the mind.\nFor more details, feel free to take a look at my concise CV and get in touch with me through the Contacts section.\n","permalink":"https://lucabontempi.com/about/","summary":"\u003cp\u003eHey there, I\u0026rsquo;m Luca!\u003c/p\u003e\n\u003cp\u003eI hold a Master’s in \u003ca href=\"https://mibe.cdl.unipv.it/en\"\u003eInternational Business and Entrepreneurship\u003c/a\u003e (M.Sc.) from the University of Pavia, where I focused on the digital side of business, and another Master’s in \u003ca href=\"https://uni-tuebingen.de/en/fakultaeten/wirtschafts-und-sozialwissenschaftliche-fakultaet/faecher/fachbereich-wirtschaftswissenschaft/wirtschaftswissenschaft/studium/studiengaenge/master/msc-european-management/\"\u003eEuropean Management\u003c/a\u003e (M.Sc.) from the University of Tübingen. For my thesis, I explored \u003ca href=\"https://lucabontempi.com/blog/determinants_of_download_on_mobile_app_stores/\" title=\"Determinants of Download on Mobile App Stores - An Empirical Analysis\"\u003ewhat makes people download certain apps from app stores\u003c/a\u003e.\u003c/p\u003e\n\u003cp\u003eBefore that, I earned my Bachelor’s in Business Administration (B.Sc.) at the University of Brescia, where, in my thesis, I researched the \u003ca href=\"https://lucabontempi.com/il_mercato_informatico/\" title=\"The Information Technology Industry: Software as a Product and a Service\"\u003eIT industry and the shift from software as a product to software as a service\u003c/a\u003e (in Italian).\u003c/p\u003e","title":"About"},{"content":"You can contact me by email:\nluca@lucabontempi.comEmail You can also find me on: LinkedInProfessional profile StravaSport and activities KomootRoutes and outdoor adventures FlickrPhotography ThreadsPosts and conversations XPosts and conversations ","permalink":"https://lucabontempi.com/contacts/","summary":"\u003cp\u003eYou can contact me by email:\u003c/p\u003e\n\u003ca class=\"contact-link\" href=\"mailto:luca@lucabontempi.com\"\u003e\n    \u003cspan class=\"contact-icon\" aria-hidden=\"true\"\u003e\u003csvg xmlns=\"http://www.w3.org/2000/svg\" viewBox=\"0 0 24 21\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"\n    stroke-linecap=\"round\" stroke-linejoin=\"round\"\u003e\n    \u003cpath d=\"M4 4h16c1.1 0 2 .9 2 2v12c0 1.1-.9 2-2 2H4c-1.1 0-2-.9-2-2V6c0-1.1.9-2 2-2z\"\u003e\u003c/path\u003e\n    \u003cpolyline points=\"22,6 12,13 2,6\"\u003e\u003c/polyline\u003e\n\u003c/svg\u003e\u003c/span\u003e\n    \u003cspan class=\"contact-text\"\u003e\u003cspan class=\"contact-label\"\u003eluca@lucabontempi.com\u003c/span\u003e\u003cspan class=\"contact-description\"\u003eEmail\u003c/span\u003e\u003c/span\u003e\n  \u003c/a\u003e\n\n\u003cp\u003eYou can also find me on:\n\u003cdiv class=\"contacts-grid\"\u003e\n\u003ca class=\"contact-link\" href=\"https://www.linkedin.com/in/lucabontempi1/\" target=\"_blank\" rel=\"noopener noreferrer\"\u003e\n    \u003cspan class=\"contact-icon\" aria-hidden=\"true\"\u003e\u003csvg xmlns=\"http://www.w3.org/2000/svg\" viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"\n    stroke-linecap=\"round\" stroke-linejoin=\"round\"\u003e\n    \u003cpath d=\"M16 8a6 6 0 0 1 6 6v7h-4v-7a2 2 0 0 0-2-2 2 2 0 0 0-2 2v7h-4v-7a6 6 0 0 1 6-6z\"\u003e\u003c/path\u003e\n    \u003crect x=\"2\" y=\"9\" width=\"4\" height=\"12\"\u003e\u003c/rect\u003e\n    \u003ccircle cx=\"4\" cy=\"4\" r=\"2\"\u003e\u003c/circle\u003e\n\u003c/svg\u003e\u003c/span\u003e\n    \u003cspan class=\"contact-text\"\u003e\u003cspan class=\"contact-label\"\u003eLinkedIn\u003c/span\u003e\u003cspan class=\"contact-description\"\u003eProfessional profile\u003c/span\u003e\u003c/span\u003e\n  \u003c/a\u003e\n\n\u003ca class=\"contact-link\" href=\"https://www.strava.com/athletes/4403103\" target=\"_blank\" rel=\"noopener noreferrer\"\u003e\n    \u003cspan class=\"contact-icon\" aria-hidden=\"true\"\u003e\u003csvg xmlns=\"http://www.w3.org/2000/svg\" viewBox=\"0 0 24 24\" fill=\"currentColor\" stroke-width=\"2\"\u003e\n    \u003cpath\n        d=\"M15.387 17.944l-2.089-4.116h-3.065L15.387 24l5.15-10.172h-3.066m-7.008-5.599l2.836 5.598h4.172L10.463 0l-7 13.828h4.169\" /\u003e\n\u003c/svg\u003e\u003c/span\u003e\n    \u003cspan class=\"contact-text\"\u003e\u003cspan class=\"contact-label\"\u003eStrava\u003c/span\u003e\u003cspan class=\"contact-description\"\u003eSport and activities\u003c/span\u003e\u003c/span\u003e\n  \u003c/a\u003e\n\n\u003ca class=\"contact-link\" href=\"https://www.komoot.it/user/2218098976141\" target=\"_blank\" rel=\"noopener noreferrer\"\u003e\n    \u003cspan class=\"contact-icon\" aria-hidden=\"true\"\u003e\u003csvg xmlns=\"http://www.w3.org/2000/svg\" viewBox=\"0 0 48 48\" fill=\"none\" stroke=\"currentColor\" stroke-linecap=\"round\"\n    stroke-linejoin=\"round\" stroke-width=\"3\"\u003e\n    \u003cpath\n        d=\"M24 2.5A21.5 21.5 0 0 0 7.17 37.38l10.22-10.22a7.32 7.32 0 0 1 11.79-8.34h0a7.32 7.32 0 0 1 1.43 8.34l10.22 10.22A21.5 21.5 0 0 0 24 2.5Zm0 25L10.64 40.82a21.41 21.41 0 0 0 26.72 0Z\" /\u003e\n\u003c/svg\u003e\u003c/span\u003e\n    \u003cspan class=\"contact-text\"\u003e\u003cspan class=\"contact-label\"\u003eKomoot\u003c/span\u003e\u003cspan class=\"contact-description\"\u003eRoutes and outdoor adventures\u003c/span\u003e\u003c/span\u003e\n  \u003c/a\u003e\n\n\u003ca class=\"contact-link\" href=\"https://www.flickr.com/photos/77058054@N07/\" target=\"_blank\" rel=\"noopener noreferrer\"\u003e\n    \u003cspan class=\"contact-icon\" aria-hidden=\"true\"\u003e\u003csvg xmlns=\"http://www.w3.org/2000/svg\" viewBox=\"0 0 24 24\" fill=\"currentColor\"\u003e\n    \u003cpath\n        d=\"M5.334 6.666C2.3884 6.666 0 9.055 0 12c0 2.9456 2.3884 5.334 5.334 5.334 2.9456 0 5.332-2.3884 5.332-5.334 0-2.945-2.3864-5.334-5.332-5.334zm13.332 0c-2.9456 0-5.332 2.389-5.332 5.334 0 2.9456 2.3864 5.334 5.332 5.334C21.6116 17.334 24 14.9456 24 12c0-2.945-2.3884-5.334-5.334-5.334Z\" /\u003e\n\u003c/svg\u003e\u003c/span\u003e\n    \u003cspan class=\"contact-text\"\u003e\u003cspan class=\"contact-label\"\u003eFlickr\u003c/span\u003e\u003cspan class=\"contact-description\"\u003ePhotography\u003c/span\u003e\u003c/span\u003e\n  \u003c/a\u003e\n\n\u003ca class=\"contact-link\" href=\"https://www.threads.com/@bonte.lu\" target=\"_blank\" rel=\"noopener noreferrer\"\u003e\n    \u003cspan class=\"contact-icon\" aria-hidden=\"true\"\u003e\u003csvg xmlns=\"http://www.w3.org/2000/svg\" viewBox=\"0 0 192 192\" fill=\"currentColor\" stroke=\"none\"\u003e\n    \u003cpath\n        d=\"M141.537 88.9883C140.71 88.5919 139.87 88.2104 139.019 87.8451C137.537 60.5382 122.616 44.905 97.5619 44.745C97.4484 44.7443 97.3355 44.7443 97.222 44.7443C82.2364 44.7443 69.7731 51.1409 62.102 62.7807L75.881 72.2328C81.6116 63.5383 90.6052 61.6848 97.2286 61.6848C97.3051 61.6848 97.3819 61.6848 97.4576 61.6855C105.707 61.7381 111.932 64.1366 115.961 68.814C118.893 72.2193 120.854 76.925 121.825 82.8638C114.511 81.6207 106.601 81.2385 98.145 81.7233C74.3247 83.0954 59.0111 96.9879 60.0396 116.292C60.5615 126.084 65.4397 134.508 73.775 140.011C80.8224 144.663 89.899 146.938 99.3323 146.423C111.79 145.74 121.563 140.987 128.381 132.296C133.559 125.696 136.834 117.143 138.28 106.366C144.217 109.949 148.617 114.664 151.047 120.332C155.179 129.967 155.42 145.8 142.501 158.708C131.182 170.016 117.576 174.908 97.0135 175.059C74.2042 174.89 56.9538 167.575 45.7381 153.317C35.2355 139.966 29.8077 120.682 29.6052 96C29.8077 71.3178 35.2355 52.0336 45.7381 38.6827C56.9538 24.4249 74.2039 17.11 97.0132 16.9405C119.988 17.1113 137.539 24.4614 149.184 38.788C154.894 45.8136 159.199 54.6488 162.037 64.9503L178.184 60.6422C174.744 47.9622 169.331 37.0357 161.965 27.974C147.036 9.60668 125.202 0.195148 97.0695 0H96.9569C68.8816 0.19447 47.2921 9.6418 32.7883 28.0793C19.8819 44.4864 13.2244 67.3157 13.0007 95.9325L13 96L13.0007 96.0675C13.2244 124.684 19.8819 147.514 32.7883 163.921C47.2921 182.358 68.8816 191.806 96.9569 192H97.0695C122.03 191.827 139.624 185.292 154.118 170.811C173.081 151.866 172.51 128.119 166.26 113.541C161.776 103.087 153.227 94.5962 141.537 88.9883ZM98.4405 129.507C88.0005 130.095 77.1544 125.409 76.6196 115.372C76.2232 107.93 81.9158 99.626 99.0812 98.6368C101.047 98.5234 102.976 98.468 104.871 98.468C111.106 98.468 116.939 99.0737 122.242 100.233C120.264 124.935 108.662 128.946 98.4405 129.507Z\"\u003e\n    \u003c/path\u003e\n\u003c/svg\u003e\u003c/span\u003e\n    \u003cspan class=\"contact-text\"\u003e\u003cspan class=\"contact-label\"\u003eThreads\u003c/span\u003e\u003cspan class=\"contact-description\"\u003ePosts and conversations\u003c/span\u003e\u003c/span\u003e\n  \u003c/a\u003e\n\n\u003ca class=\"contact-link\" href=\"https://twitter.com/luca_bontempi1\" target=\"_blank\" rel=\"noopener noreferrer\"\u003e\n    \u003cspan class=\"contact-icon\" aria-hidden=\"true\"\u003e\u003csvg xmlns=\"http://www.w3.org/2000/svg\" viewBox=\"0 0 24 24\" fill=\"currentColor\"\u003e\n    \u003cpath\n        d=\"M18.244 2.25h3.308l-7.227 8.26 8.502 11.24H16.17l-5.214-6.817L4.99 21.75H1.68l7.73-8.835L1.254 2.25H8.08l4.713 6.231zm-1.161 17.52h1.833L7.084 4.126H5.117z\"\u003e\n    \u003c/path\u003e\n\u003c/svg\u003e\u003c/span\u003e\n    \u003cspan class=\"contact-text\"\u003e\u003cspan class=\"contact-label\"\u003eX\u003c/span\u003e\u003cspan class=\"contact-description\"\u003ePosts and conversations\u003c/span\u003e\u003c/span\u003e\n  \u003c/a\u003e\n\n\u003c/div\u003e\n\u003c/p\u003e","title":"Contacts"},{"content":"Luca Bontempi 📍 Milan Area, Italy | E-mail: luca@lucabontempi.com | LinkedIn\nSummary I am a Partner Technical Specialist at IBM, operating at the intersection of enterprise technology and strategic business growth. My work focuses on empowering the Global System Integrators (GSIs) ecosystem, enabling partners to demonstrate, architect, and deploy solutions across the full IBM technology portfolio at scale. By translating complex technical architectures into clear commercial value, I help partners expand their autonomous delivery capabilities and drive joint client engagements.\nMy practical approach to IT is grounded in a strong academic foundation in innovation economics. Holding two Master of Science degrees in European Management (Eberhard Karls University of Tübingen) and International Business and Entrepreneurship (University of Pavia), I focused my graduate research on platform economics, specifically analyzing the market determinants of mobile app adoption and the impact of digital platforms on business models.\nI combine this macro-level understanding of digital ecosystems with hands-on technical enablement. Whether orchestrating tailored technical workshops, supporting partner teams through Proofs of Concept (PoCs), or aligning cross-functional stakeholders, my goal is to turn emerging IT trends into high-quality, actionable market solutions.\nExperience Partner Technical Specialist · Jul 2026 – Present Associate Partner Technical Specialist · Jan 2026 – Jun 2026\nAssociate Partner Technical Specialist - Partner · Jun 2023 – Dec 2025\nIBMSegrate, Italy · Full-time\nAccount Scope: Progressed from enabling 13 partners on a single portfolio area to owning the technical relationship with three global and Italian System Integrators (GSI) across the full IBM technology portfolio. Partner Enablement: Drive technical enablement and solution activation for strategic GSIs, expanding their autonomous delivery capabilities across the IBM portfolio. Solution Architecture: Work with partner architects to translate client requirements into integrated IBM solutions, and facilitate joint client engagements through to technical validation. Technical Advocacy: Deliver tailored technical workshops and training sessions, actively supporting partner teams during the execution of Proofs of Concept (PoCs). Technical Sales Intern · Mar 2023 – May 2023 IBMSegrate, Italy · Internship\nInternal Tooling: Co-developed an internal analytics tool to track, assess, and optimize partner ecosystem skills, improving reporting efficiency across the unit. Pre-Sales Support: Assisted the Ecosystem team in orchestrating technical enablement initiatives and preparing pre-sales assets for key partners. Administration Intern · Oct 2019 – Dec 2019 New Master SrlMagnago, Italy · Internship\nIntern in a transportation company serving the sport industry. Education Master\u0026rsquo;s Degree in European Management · Sep 2021 – Feb 2023 Eberhard Karls University of TübingenTübingen, Germany · Double Degree Programme\nScore: Gut (1.57/1.0). Master\u0026rsquo;s Thesis: Conducted empirical research on the market determinants of mobile app downloads and the impact of digital platforms on business models (Determinants of Download on Mobile App Stores - An Empirical Analysis). Attended courses: Marketing-Mix Management, eBusiness, Innovation and Technological Change, Advanced Empirical Public Economics. Master\u0026rsquo;s Degree in International Business and Entrepreneurship · Sep 2020 – Jul 2023 University of PaviaPavia, Italy · Double Degree Programme\nScore: 110/110. Curriculum: Digital Management. Attended courses: Information Systems for Managers, Exponential Coding with AI, Business Analytics with Excel, Applied Project Management, Financial Innovation \u0026amp; Fintech, Capital Markets and EU Company Law. Bachelor’s Degree in Business Administration · Sep 2016 – Mar 2020 University of BresciaBrescia, Italy\nScore: 102/110. Bachelor\u0026rsquo;s Thesis: \u0026ldquo;The Information Technology Industry: Software as a Product and a Service\u0026rdquo; (in Italian). Scientific High School Diploma · Sep 2011 – Jun 2016 Liceo Scientifico AntoniettiIseo, Italy\nLanguages Italian: Native English: Full Professional Proficiency (C1) Skills Domain Expertise Technical Sales, Partner Enablement, Solution Architecture, B2B Enterprise Sales\nTech \u0026amp; Data Python, SQL, Enterprise Software Architecture concepts, Digital Platform Models\nSoft Skills Cross-functional collaboration, Technical training delivery, Stakeholder management\nProfessional Certifications IT Architecture: Architectural Thinking (IBM) Sales: Global Sales School - Sales Activation (IBM) Project Management: Introductory Certificate in Project Management (IPMA Italy) Leadership Sports Management: Sport Director and Coach of a Youth Cycling Team (since 2016) ","permalink":"https://lucabontempi.com/cv/","summary":"\u003ch2 id=\"luca-bontempi\"\u003eLuca Bontempi\u003c/h2\u003e\n\u003cp\u003e📍 Milan Area, Italy | E-mail: \u003ca href=\"mailto:luca@lucabontempi.com\"\u003eluca@lucabontempi.com\u003c/a\u003e | \u003ca href=\"https://www.linkedin.com/in/lucabontempi1/\"\u003eLinkedIn\u003c/a\u003e\u003c/p\u003e\n\u003ch2 id=\"summary\"\u003eSummary\u003c/h2\u003e\n\u003cp\u003eI am a Partner Technical Specialist at IBM, operating at the intersection of enterprise technology and strategic business growth. My work focuses on empowering the Global System Integrators (GSIs) ecosystem, enabling partners to demonstrate, architect, and deploy solutions across the full IBM technology portfolio at scale. By translating complex technical architectures into clear commercial value, I help partners expand their autonomous delivery capabilities and drive joint client engagements.\u003c/p\u003e","title":"Curriculum Vitae"},{"content":"Below you can find the full document version of my Master Thesis, written at the Eberhard Karls University of Tübingen and at the University of Pavia and submitted for the first time during February 2023.\nThe research question investigates which characteristic out of reputation, popularity and developer\u0026rsquo;s brand, reflected in some elements in the presentation page on app stores, is the most effective in predicting user\u0026rsquo;s choice to download an app.\nThis same thesis is also available as a blog post.\n","permalink":"https://lucabontempi.com/determinants_of_download_on_mobile_app_stores_doc/","summary":"\u003cp\u003eBelow you can find the full document version of my Master Thesis, written at the \u003cstrong\u003eEberhard Karls University of Tübingen\u003c/strong\u003e and at the \u003cstrong\u003eUniversity of Pavia\u003c/strong\u003e and submitted for the first time during \u003cstrong\u003eFebruary 2023\u003c/strong\u003e.\u003c/p\u003e\n\u003cp\u003eThe research question investigates which characteristic out of \u003cstrong\u003ereputation\u003c/strong\u003e, \u003cstrong\u003epopularity\u003c/strong\u003e and \u003cstrong\u003edeveloper\u0026rsquo;s brand\u003c/strong\u003e, reflected in some elements in the \u003cstrong\u003epresentation page\u003c/strong\u003e on \u003cstrong\u003eapp stores\u003c/strong\u003e, is the most effective in predicting \u003cstrong\u003euser\u0026rsquo;s choice to download an app\u003c/strong\u003e.\u003c/p\u003e","title":"Determinants of Download on Mobile App Stores - An Empirical Analysis"},{"content":"","permalink":"https://lucabontempi.com/explore/","summary":"","title":"Explore"},{"content":"Seconda di quattro puntate. Nella prima ho raccontato di chi parlano questi numeri, cioè di un settimo dei tesserati, quelli arrivati almeno una volta nei primi cinque. Qui rispondo alla domanda da cui è nato tutto lo studio: da che età il risultato in gara dice qualcosa sul futuro?\nQuando ho cominciato ero convinto che a tredici anni non ci fosse niente da vedere.\nLo lasciava pensare la letteratura, perché alle età più basse è difficile separare la prestazione dalla maturazione. Lo diceva il buon senso: un ragazzino di terza media che vince una gara Esordienti (così si chiamano in Italia i tredici e quattordici anni, gli Under 15 della letteratura) sta battendo dei coetanei con magari un anno di pubertà in meno. E lo diceva l\u0026rsquo;esperienza. Avevo in testa una lunga fila di campioncini di quell\u0026rsquo;età che poi non si sono più visti.\nMi aspettavo di trovare zero, e di poterlo dire con precisione. È andata diversamente.\nGuardare, prima di modellare Per la risposta più semplice la statistica non serve. Prendi i ragazzi che poi sono diventati professionisti, guardi dove stavano in classifica a tredici anni e lo confronti con dove stavano tutti gli altri.\nIl piazzamento è in percentile: una scala in cui 100 è il primo della classifica, 50 è a metà e 0 è l\u0026rsquo;ultimo. Serve a rendere confrontabili stagioni e categorie con un numero diverso di partecipanti.\na tredici anni (Esordienti, primo anno) posizione tipica chi non diventerà professionista 49° percentile chi diventerà professionista 81° percentile «posizione tipica» è la mediana, cioè il valore che lascia metà del gruppo sopra e metà sotto; professionista vuol dire aver corso in una squadra di primo o secondo livello entro i venticinque anni\nA diciotto anni la distanza si allarga ancora, da 47 a 94.\nQuanto si separano due gruppi si può riassumere in modo elegante. Prendi tutte le coppie possibili formate da un futuro professionista e da un futuro non professionista, e conti quante volte il professionista sta davanti. A tredici anni succede nel 74% dei casi, a diciotto nell'89%1. Sulle scale convenzionali la separazione a tredici anni sta appena sopra il confine fra «media» e «grande». Un segnale enorme non è, ma è molto lontano dal niente che mi aspettavo.\nLe due curve sono i percentili mediani dei due gruppi, e il gruppo si conosce solo guardando indietro: a tredici anni nessuno sapeva chi fosse chi.\nSu quel 74% servono due precisazioni, perché è facile leggerlo male.\nLa prima: il confronto è fra chi era in classifica quell\u0026rsquo;anno, non fra tutti i ragazzi. Chi a tredici anni non ha mai fatto un punto non entra né fra i professionisti né fra gli altri. La cifra dice dunque quanto la classifica separa al suo interno; su tutti gli altri ragazzi non dice niente.\nLa seconda: i professionisti del confronto sono i futuri professionisti, cioè ragazzi di cui oggi conosciamo l\u0026rsquo;esito e che a tredici anni non lo avevano scritto in fronte. È un confronto costruito guardando indietro.\nInoltre il 74% e l'89% non si confrontano direttamente, perché non riguardano le stesse persone: a tredici anni il conto gira su tutti quelli che erano in classifica allora, a diciotto solo su quelli che ci sono ancora, molti meno e già selezionati. Il paragone pulito, sulle stesse persone a ogni età, arriva più avanti.\nNon è un confronto fra due gruppi soltanto C\u0026rsquo;è un secondo modo di guardare gli stessi dati: invece di dividere i ragazzi in arrivati e non arrivati, li divido in quattro secondo quanto lontano sono andati: chi non è diventato professionista, chi lo è diventato senza mai entrare nei primi cinquecento del mondo, chi in quei cinquecento c\u0026rsquo;è entrato, e chi è arrivato nei primi cento del PCS Ranking di procyclingstats.com.\nA tredici anni le mediane dei primi tre gruppi sono 49, 69 e 891. Viene fuori una scala ordinata: più uno è andato lontano, più stava avanti già a tredici anni. Un rumore casuale non produce una scala del genere, ed è la ragione principale per cui non credo che il 74% sia un artefatto.\nL\u0026rsquo;ultimo gradino però la rompe. La mediana di chi è arrivato nei primi cento del mondo, a tredici anni, è 84: sotto quella del gruppo precedente. Sono sei atleti, e con sei atleti (un campione molto ridotto) una mediana può fare qualunque cosa.\nPer approfondire, nel documento tecnico: Il gradiente per livello raggiunto.\nÈ rendimento, o è la data di nascita? È il dubbio più ragionevole che si possa avere, e va chiuso.\nFra ragazzi della stessa annata, chi è nato a gennaio ha fino a dodici mesi di sviluppo in più di chi è nato a dicembre. Nei dati si vede benissimo: nella classifica Esordienti i nati nel primo trimestre sono 2,13 volte quelli dell\u0026rsquo;ultimo, tenuto conto di come sono distribuite le nascite in Italia2. Se il piazzamento a tredici anni misurasse in buona parte quanto presto uno è cresciuto, il 74%, riferito ai casi in cui il professionista sta davanti al non professionista da Esordiente, racconterebbe soprattutto quello.\nPer verificarlo rifaccio il modello con l\u0026rsquo;età relativa dentro, contata in giorni fra la nascita e la fine dell\u0026rsquo;anno, e guardo che fine fa il peso del piazzamento. Non gli succede niente. Il vantaggio di dieci punti di percentile è 1,40 prima e 1,40 dopo, e la capacità di distinguere passa da 0,736 a 0,738: si muove nella terza cifra decimale. L\u0026rsquo;età relativa da sola, usata come unico predittore, arriva a 0,513. Praticamente una monetina.\nNascere a gennaio serve, dunque? Serve moltissimo, ma prima, per entrare in classifica: il rapporto di due a uno fra primo e ultimo trimestre dice proprio quello. Fra i ragazzi che in classifica ci sono già, invece, la data di nascita non distingue più chi arriverà. Infatti fra i professionisti quel rapporto scende a 1,5 contro l'1,7 di tutti i classificati, uno scarto che su 77 atleti non si distingue dal caso. È un vantaggio che fa entrare in classifica da giovani, e che con il talento non c\u0026rsquo;entra.\nIn Italia si nasce di più fra maggio e settembre, quindi l\u0026rsquo;atteso si scosta dal 25% per trimestre: la figura tiene conto della stagionalità reale delle nascite.\nC\u0026rsquo;è poi un modo indipendente di mettere alla prova quella spiegazione, e viene dal ciclismo femminile. È una delle poche domande della serie a cui posso rispondere per le ragazze esattamente come per i ragazzi. Tutte quelle sulla previsione hanno bisogno di sapere chi è arrivato, e un esito femminile confrontabile con quello maschile non c\u0026rsquo;è, per ragioni che racconto nella quarta puntata; qui basta la data di nascita. Il ragionamento è questo. Le ragazze maturano prima, e a tredici anni molte hanno già attraversato la pubertà, mentre fra i coetanei maschi la differenza di sviluppo fra chi è nato a gennaio e chi a dicembre è al suo massimo. Se il vantaggio viene dalla maturazione, fra le atlete deve essere più debole.\nLo è: a tredici anni, sulle stesse annate di nascita e con lo stesso atteso demografico, i nati nel primo trimestre sono 1,98 volte quelli dell\u0026rsquo;ultimo fra i maschi e 1,51 volte fra le femmine, e la differenza fra i due sessi regge a un test (p = 0,007)2. Il numero maschile è diverso dal 2,13 di prima perché qui le annate sono soltanto quelle in cui anche le ragazze sono osservate: 5 544 atleti e 874 atlete.\nDopo i quattordici anni, però, le due linee smettono di somigliarsi. È la parte che non torna. Fra i maschi lo squilibrio cala di categoria in categoria. Fra le femmine scende più in fretta, tanto che in Allieve non si distingue più dall\u0026rsquo;atteso demografico (1,20 con p = 0,14), e poi risale in Juniores (1,42), dove torna a distinguersene. Su 681 e 370 atlete quelle due cifre hanno intervalli larghi, e non ci costruirei sopra niente. Con ragionevole sicurezza si può parlare solo dei tredici anni, dove le atlete sono 874 e la distanza dai coetanei è netta.\nA tredici anni lo squilibrio fra le atlete c\u0026rsquo;è, ma è più contenuto di quello fra i coetanei. Dopo, i valori femminili poggiano su poche centinaia di atlete e non seguono una linea.\nÈ una conferma indiretta, perché i due movimenti differiscono in molte altre cose oltre all\u0026rsquo;età della pubertà. Tuttavia il femminile fa da controllo a un risultato maschile: è una differenza di maturazione a dire perché nascere a gennaio conta, e perché poi smette.\nQuanto conta la classifica, esattamente Un modello permette di mettere un numero sul vantaggio: quanto conta salire di dieci posizioni percentuali3?\ncategoria età quanto moltiplica le odds Esordienti, primo anno 13 ×1,40 (fra 1,25 e 1,57) Esordienti, secondo anno 14 ×1,56 Allievi, primo anno 15 ×1,70 Allievi, secondo anno 16 ×1,98 Juniores, primo anno 17 ×1,64 Juniores, secondo anno 18 ×2,28 (fra 1,91 e 2,80) Under 23, primo anno 19 ×1,32 Ogni riga è un modello a sé, con l\u0026rsquo;intervallo al 95% e la scala logaritmica, perché un odds ratio si legge in rapporti e non in differenze.\nCome si legge questa tabella? Il numero moltiplica le odds, non la probabilità. Fra due Esordienti che differiscono di dieci posizioni percentuali, quello davanti ha il 40% di odds in più di arrivare al professionismo: il rapporto fra la sua probabilità di farcela e quella di non farcela è 1,40 volte quello dell\u0026rsquo;altro. Con un esito raro come questo anche la probabilità cresce di quasi il 40%. Non si può però leggere quel 40% come quaranta punti percentuali: nella realtà, si passa da circa il 2,7% a circa il 3,7%, un punto in più.\nIl peso cresce con l\u0026rsquo;età, da 1,40 a 2,28, ma non in modo regolare. Le due righe che scendono non vanno lette come cali del segnale: a ogni passaggio di categoria cambia la popolazione. In Juniores primo anno i professionisti sono già il 9,7% della lista contro il 4,5% della cella precedente, e in Under 23 sono il 37%. Chi confronta quei numeri come se misurassero la stessa cosa ha già sbagliato la lettura.\nPer approfondire, nel documento tecnico: Perché la colonna «% pro» non va letta come un segnale.\nC\u0026rsquo;è poi un dettaglio che sembra un risultato e non lo è. Nelle categorie a lista unica la percentuale di futuri professionisti è più alta al primo anno che al secondo: sembrerebbe che il primo anno selezioni meglio. In realtà, come ho raccontato nella prima puntata, al primo anno i posti a punti sono pochi, e i classificati sono in media 181 contro 320. Chi c\u0026rsquo;è è già più selezionato, e un gruppo più selezionato contiene per forza una quota maggiore di futuri professionisti. Il primo anno predice meglio, allora? Si risponde solo confrontando le stesse persone, e la risposta è no: il secondo anno discrimina meglio in tutte le categorie, 81% contro 70% in Esordienti, 85% contro 81% in Allievi, 87% contro 78% in Juniores.\nQuanto aggiungono i risultati più recenti Il piazzamento negli Allievi aiuta a prevedere chi diventerà professionista. Ma quanto aggiunge rispetto a ciò che sapevamo già dai risultati negli Esordienti?\nPer scoprirlo, prendiamo i 102 ragazzi presenti nelle classifiche di tutte le categorie. È un gruppo piccolo e molto selezionato, ma permette di confrontare le previsioni sugli stessi atleti. Partiamo dal solo anno di nascita, poi aggiungiamo i risultati negli Esordienti, negli Allievi, negli Juniores e infine negli Under 23: a ogni passaggio misuriamo quanto migliora la previsione4.\nil modello conosce… quanto ci prende entro che margine solo l\u0026rsquo;anno di nascita 51% (cioè: nulla) 40-62% più gli Esordienti 58% 46-69% più gli Allievi 66% 55-77% più gli Juniores 81% 73-89% più l\u0026rsquo;Under 23 82% 74-90% Il confronto regge solo perché i 102 atleti sono sempre gli stessi: cambiando gruppo a ogni passo si misurerebbe chi è rimasto, non l\u0026rsquo;informazione aggiunta.\nLa terza colonna della tabella, che a volte si potrebbe anche omettere, qui serve: gli intervalli sono larghi fra i sedici e i ventitré punti, e i primi tre si sovrappongono abbondantemente. Di gradini che si staccano davvero ce n\u0026rsquo;è uno solo, quello degli Juniores. Gli Juniores da soli aggiungono quanto tutte le categorie precedenti messe insieme, e sono anche l\u0026rsquo;unico passaggio in cui il guadagno si distingue dal caso.\nAttenzione però a non trasformare i passi non significativi in zeri. Esordienti e Allievi migliorano l\u0026rsquo;adattamento del modello in modo distinguibile dal caso; semplicemente non cambiano l\u0026rsquo;ordine in cui i ragazzi vengono messi in fila. Sono due domande diverse: una chiede quanto il modello sia in accordo con i dati, l\u0026rsquo;altra se, dati due ragazzi, ci azzecchi su chi mettere davanti.\nPoteva essere una stranezza di quel sottocampione, quindi ho fatto la stessa domanda in altri due modi. Una foresta casuale, cioè un algoritmo che si arrangia da solo a trovare le combinazioni utili, ha ricevuto diciassette variabili invece delle due del modello semplice e ha guadagnato 2,0 punti percentuali di capacità predittiva. In cima alla sua classifica di importanza c\u0026rsquo;è proprio il piazzamento in Juniores. Una regressione penalizzata, che mette tutte le categorie in un modello solo e butta via quelle che non si guadagnano il posto, ne ha tenute 2 su 8: Juniores secondo anno e Under 23.\nTre strade diverse, la stessa conclusione: quasi tutta l\u0026rsquo;informazione utile sta nell\u0026rsquo;ultima misura che hai. Le stagioni precedenti non si sommano all\u0026rsquo;ultima: sono in gran parte la stessa cosa vista da più lontano.\nPer approfondire, nel documento tecnico: Quanto si somigliano le categorie, Un modello più complicato farebbe meglio? e E se si usassero tutte le categorie insieme?.\nE se contassi anche chi in classifica non c\u0026rsquo;era Questo studio conta solo chi è in classifica. L\u0026rsquo;alternativa è trattare l\u0026rsquo;assenza come un rendimento peggiore di qualunque presenza, e ho provato anche quella.\nLa risposta dipende dall\u0026rsquo;età in un modo che non mi aspettavo. A diciotto anni mettere gli assenti in fondo alza la capacità di distinguere, dall'89% al 94%: a quell\u0026rsquo;età non essere in classifica è già un segnale forte, e in effetti dei 77 futuri professionisti soltanto 3 non c\u0026rsquo;erano. A tredici anni invece la abbassa, dal 74% al 69%, perché in classifica a quell\u0026rsquo;età ci sono 59 dei 77 futuri professionisti, e gli altri diciotto finirebbero in fondo pur essendo destinati ad arrivare5.\nNon esserci è un\u0026rsquo;informazione, ma lo diventa tardi. Ed è la misura più precisa di una cosa detta nella prima puntata: un nome che manca dalla classifica non è stato bocciato da nessuno, e quasi un terzo di chi sparisce ricompare.\nPer approfondire, nel documento tecnico: Le conclusioni dipendono dalle scelte di disegno?.\nCosa resta Il risultato a tredici anni un\u0026rsquo;informazione la porta, e chi lo nega per prudenza ti sta dicendo una cosa gentile e falsa. Se guardi i piazzamenti dei tuoi Esordienti stai guardando qualcosa che in media ha a che fare con il futuro, dentro la popolazione osservata.\nNon lo leggerei però come «la letteratura si sbagliava». Gli studi precedenti facevano pensare a un segnale debole alle età più basse, e le due cose stanno insieme: qui il campione è italiano, ampio e non limitato a chi era già competitivo a livello internazionale, ed è proprio il tipo di popolazione in cui un segnale precoce ha modo di vedersi.\nE la data di nascita non c\u0026rsquo;entra. Nascere a gennaio aiuta a entrare in classifica e non a diventare professionista: chi seleziona a tredici anni sta in parte selezionando la data di nascita, e lo sta facendo a vuoto.\nCosa significa, in pratica? Per prevedere chi diventerà professionista, i risultati più recenti contano molto più di quelli ottenuti tre anni prima. Ma i risultati passati possono servire a rispondere a un’altra domanda: il ragazzo sta migliorando o peggiorando? È il tema della prossima puntata.\nCome lo sappiamo\nIl predittore è il percentile dentro la cella stagione × categoria × anno di categoria, che rende confrontabili classifiche di lunghezza diversa. L\u0026rsquo;esito è essere arrivati a correre in una squadra professionistica di primo o secondo livello entro i venticinque anni.\nI modelli sono regressioni logistiche con la correzione di Firth, necessaria perché l\u0026rsquo;esito è raro, meno del 3%, e senza di essa le stime sarebbero distorte verso l\u0026rsquo;alto. Sono aggiustati per anno di nascita, dato che le coorti recenti hanno avuto meno tempo per arrivare; l\u0026rsquo;aggiustamento sposta i coefficienti di meno di 0,01.\nLe percentuali di coppie sono l\u0026rsquo;area sotto la curva ROC detta in italiano: la quota di coppie, formate da un futuro professionista e da un futuro non professionista, in cui il modello mette davanti quello giusto.\nCalcolo in report/moduli/punteggi.py.\u0026#160;\u0026#x21a9;\u0026#xfe0e;\u0026#160;\u0026#x21a9;\u0026#xfe0e;\nCalcolo in report/moduli/rae.py. Le nascite attese vengono da Eurostat, tavola demo_fmonth.\u0026#160;\u0026#x21a9;\u0026#xfe0e;\u0026#160;\u0026#x21a9;\u0026#xfe0e;\nModelli in R/16_univariati.R, resi da report/moduli/univariati.py.\u0026#160;\u0026#x21a9;\u0026#xfe0e;\nModelli in R/18_annidati.R. Foresta casuale in R/27_confronto_ml.R, regressione penalizzata in R/17_penalizzato.R.\u0026#160;\u0026#x21a9;\u0026#xfe0e;\nAnalisi di sensibilità in scripts/10_sensibilita.py.\u0026#160;\u0026#x21a9;\u0026#xfe0e;\n","permalink":"https://lucabontempi.com/blog/ciclismo-giovanile-a-che-eta-si-riconosce-un-futuro-professionista/","summary":"\u003cp\u003e\u003cem\u003eSeconda di quattro puntate. Nella \u003ca href=\"/blog/ciclismo-giovanile-di-chi-stiamo-parlando/\"\u003eprima\u003c/a\u003e ho raccontato di chi parlano questi numeri, cioè di un settimo dei tesserati, quelli arrivati almeno una volta nei primi cinque. Qui rispondo alla domanda da cui è nato tutto lo studio: da che età il risultato in gara dice qualcosa sul futuro?\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eQuando ho cominciato ero convinto che a tredici anni non ci fosse niente da vedere.\u003c/p\u003e\n\u003cp\u003eLo lasciava pensare la letteratura, perché alle età più basse è difficile separare la prestazione dalla maturazione. Lo diceva il buon senso: un ragazzino di terza media che vince una gara Esordienti (così si chiamano in Italia i tredici e quattordici anni, gli Under 15 della letteratura) sta battendo dei coetanei con magari un anno di pubertà in meno. E lo diceva l\u0026rsquo;esperienza. Avevo in testa una lunga fila di campioncini di quell\u0026rsquo;età che poi non si sono più visti.\u003c/p\u003e","title":"A che età si riconosce un futuro professionista nel ciclismo?"},{"content":"Questa serie parla dei ranking giovanili italiani e del passaggio al professionismo. Prima dei miei numeri racconto cosa la ricerca sa già e, soprattutto, di chi parlano i numeri delle prossime puntate. È lì che si annidano quasi tutti gli errori.\nVent\u0026rsquo;anni di ricerca, in poche righe Al Mondiale juniores su strada ogni nazione porta al massimo sei corridori, e a portarne sei sono solo le nazioni in cima al ranking. Gli Juniores (o Under 19) sono la categoria dei diciassette e diciotto anni. Chi ci ha corso ricorda benissimo quanto sia difficile strappare una convocazione. Chi arriva su quella linea di partenza ha superato ogni selezione che il ciclismo giovanile riesca a mettere in piedi. Più predestinato di così non si può.\nEppure nemmeno vincerla, quella gara, garantisce qualcosa. Da quando esiste il Mondiale juniores su strada, solo tre corridori hanno vinto sia quello sia il Mondiale élite: Greg LeMond (juniores nel 1979, élite nel 1983 e nel 1989), Mathieu van der Poel (juniores nel 2013, élite nel 2023) e Remco Evenepoel. È lui il corridore sul podio nella foto di copertina, scattata ai Mondiali juniores di Innsbruck del 2018, l\u0026rsquo;anno in cui li vinse. Il Mondiale élite lo avrebbe vinto quattro anni dopo, nel 2022.\nNel 2006 un gruppo di ricercatori tedeschi1 ha raccolto i risultati delle grandi gare di venticinque anni di ciclismo: 27 454 risultati di 8 004 atleti di 108 paesi, dal 1980 al 2004, e non solo su strada. Poi ha guardato la stessa storia dai due lati. In avanti: dei ragazzi passati dal Mondiale juniores, quanti hanno poi corso una grande gara fra i grandi, cioè una delle corse che lo studio conta come tali, dai Mondiali ai grandi giri? Trentaquattro su cento. All\u0026rsquo;indietro: fra gli atleti con risultati nelle grandi gare élite, quanti erano passati dal Mondiale juniores? 29,4 su cento.\nLe due percentuali rispondono a domande opposte, e nessuna delle due va molto oltre un terzo: la maggioranza resta fuori da qualunque lato si guardi.\nLe due cifre rispondono a domande opposte e non misurano la stessa cosa. Messe insieme, però, dicono qualcosa di scomodo: andare forte da ragazzi non è indispensabile per arrivare in alto, e non basta neanche lontanamente a garantirlo. Due predestinati su tre non correranno mai una grande gara da grandi, e sette su dieci di quelli che a quelle gare ci arrivano non erano passati da quel Mondiale. Da allora la domanda è stata ripresa in Norvegia, in Belgio, in Spagna, in Olanda e due volte in Italia, su campioni molto diversi fra loro, e le conclusioni si assomigliano parecchio2. Sono cinque.\nI futuri professionisti andavano già meglio degli altri da ragazzi. È il risultato più replicato del campo: lo hanno trovato gruppi diversi con metodi diversi.\nIl segnale cresce man mano che ci si avvicina al traguardo. Lo studio italiano più citato, su 1 345 corridori delle classifiche federali, indica come momento più informativo il primo anno da Under 23, cioè i diciannove anni.\nSotto i quindici anni, invece, il sospetto è che ci sia poco da vedere. Uno studio belga sui tredici e quattordicenni ha misurato maturazione, coordinazione e capacità motorie: tutto insieme spiegava poco più di un quinto delle differenze nel rendimento in gara due o tre anni dopo. La lettura che se ne è ricavata è che a quell\u0026rsquo;età si misuri soprattutto chi si è sviluppato prima.\nPoi ci sono i test di laboratorio, e qui suona strano: soglia e watt per chilo descrivono bene il livello di un ragazzo adesso, ma aggiungono poco su dove arriverà. Fa eccezione uno studio spagnolo su 65 Under 23, dove una prova a cronometro di otto minuti separa bene chi diventerà professionista. Quanto bene? Il 54 per cento di quelli che la prova indicava come futuri professionisti ce la fa davvero. Poco più della metà, e si parla di ventenni già selezionatissimi.\nInfine il mese di nascita: nelle popolazioni giovanili i nati a inizio anno sono più di quanti la demografia ne giustifichi, tanto più quanto più l\u0026rsquo;età è bassa, mentre fra chi arriva l\u0026rsquo;effetto non si trova. Uno studio su quasi tremila corridori delle squadre Continental dice che l\u0026rsquo;effetto si vede soprattutto fra chi non ce l\u0026rsquo;ha fatta. Nella seconda puntata rifaccio quel conto sui miei dati, perché è il dubbio più ragionevole che si possa avere su un risultato a tredici anni.\nC\u0026rsquo;è poi un dato che torna ovunque: la stragrande maggioranza di chi va forte da giovane non arriva comunque. I numeri vanno letti con il loro denominatore attaccato, perché metterli in fila come se misurassero la stessa cosa porta fuori strada. Due su tre dei juniores del Mondiale non correranno una grande gara élite. Il 15 per cento dei primi dieci della classifica Allievi italiana, cioè i quindici e sedici anni, ha poi fatto punti nel World Tour, il circuito di primo livello. Cinque su quarantotto di un\u0026rsquo;intera annata olandese sono arrivati a una squadra Continental, il il terzo livello delle squadre UCI, o superiore2. I punti di partenza sono diversi (i convocati, i primi dieci, tutti) e lo sono anche i traguardi. La scala cambia ogni volta, la direzione resta la stessa.\nRestano fuori due cose, e sono il buco che questa serie prova a riempire. Nessuno ha mai guardato i tredici e quattordici anni partendo dai risultati di gara su una popolazione ampia e non preselezionata: gli studi italiani cominciano dagli Allievi, quello belga include gli Under 15 ma solo per chi aveva già un piazzamento nei primi dieci. E nessuno studio basato sui risultati di gara ha mai scritto quanti dei ragazzi segnalati poi arrivino. Eppure è esattamente il numero che useresti per decidere.\nCi sono poi due cose che gli studi non dichiarano mai. Non le ho trovate in nessuno dei ventidue che ho riassunto2, e me ne sono accorto provando a rifare i conti.\nLa prima è da quanti si parte. I convocati al Mondiale, i primi dieci, i ragazzi a ranking sono già il risultato di una selezione, e quanti fossero prima non lo scrive nessuno. Senza quel numero due percentuali di due studi diversi non si possono mettere in fila: sotto hanno popolazioni diverse.\nLa seconda è che spesso un ragazzo che smette di comparire nelle classifiche viene contato come se avesse smesso di correre. Gli imbuti di questa letteratura si costruiscono così, e chi non compare più nella categoria successiva finisce fra i persi. Nessuno ha verificato quanta parte di quelle sparizioni sia invece gente che continua a correre senza più andare a punti, o che rientra l\u0026rsquo;anno dopo. Quanti rientrano lo misuro più avanti in questa puntata. Non sono pochi.\nNon posso farne una critica dall\u0026rsquo;alto, perché valgono in pieno anche per questo studio. Anche i miei numeri partono da un gruppo già scremato, e anch\u0026rsquo;io vedo sparire dei ragazzi dalla classifica senza sapere se abbiano smesso di correre o se vadano soltanto più piano. L\u0026rsquo;unica differenza che posso rivendicare è di averle misurate invece di lasciarle implicite. Il resto della puntata parla di questo.\nLa classifica da cui parte tutto Lo studio parte dalla classifica che ciclismo.info pubblica ogni anno per ogni categoria giovanile. Nel periodo considerato, dal 2007 al 2025, raccoglie 28 037 piazzamenti stagionali di 11 094 ragazzi3. Ragazzi e non ragazze, e la ragione sono i numeri: le atlete in classifica sono fra un settimo e un decimo degli atleti, le divisioni professionistiche femminili esistono solo dal 2020 e la classifica Under 23 femminile, che nel maschile è quella che dice di più, non esiste. Dove la domanda non ha bisogno di sapere chi ce l\u0026rsquo;ha fatta, invece, le ragazze ci sono.\nciclismo.info è l\u0026rsquo;archivio più completo che esista sul ciclismo giovanile italiano. Una precisazione, però, va fatta subito: l\u0026rsquo;archivio ufficiale della federazione è un\u0026rsquo;altra cosa, questo è un portale che raccoglie e ordina i risultati per conto proprio. E dentro non ci sono tutti i giovani ciclisti italiani. Per entrarci devi aver fatto almeno un punto, e i punti li prendono solo i primi cinque di ogni gara: cinque alla vittoria e a scendere fino a uno al quinto posto4.\nQuanto pesa questo filtro? Si può misurare, perché la federazione pubblica ogni anno quanti sono i tesserati per categoria5. I numeri che seguono sono medie fra alcune stagioni.\ncategoria tesserati per stagione in classifica quota Esordienti 3 214 497 15,5% Allievi 2 673 371 13,9% Juniores 1 687 262 15,5% Under 23 1 060 162 15,3% La barra è la media delle stagioni, la linea va dalla più bassa alla più alta. Nel conteggio dei tesserati c\u0026rsquo;è anche chi corre altre specialità e non può comparire in una classifica su strada, quindi uno su sette è un limite inferiore.\nIn classifica finisce circa un tesserato su sette. Più del valore colpisce quanto stia fermo: quattro categorie, da cinque a sette stagioni ciascuna, sempre lo stesso ordine di grandezza, in una forbice fra il 12,2% e il 17,2%. Gli altri sei su sette corrono senza mai andare a punti, oppure corrono in una specialità diversa dalla strada. La tessera infatti è per categoria e non per disciplina: un Esordiente che fa solo mountain bike è dentro quel conteggio, pur non potendo comparire in nessuna classifica su strada.\nQuell\u0026rsquo;uno su sette sottostima quanti dei ciclisti su strada finiscono in classifica, perché nel denominatore ci sono ragazzi che nel numeratore non possono entrare. Resta che la classifica mostra solo la punta del ciclismo giovanile, ed è di quella punta che parlano tutti i numeri delle prossime tre puntate.\nLe coorti, e il numero che userò sempre Le coorti principali, cioè i gruppi di nati negli stessi anni che seguo nel tempo, sono i nati fra il 1996 e il 2000. Hanno avuto tutti il tempo di arrivare o di non arrivare. Sono 2 813 ragazzi, 77 professionisti. Professionista, in tutta la serie, vuol dire aver corso in una squadra di primo livello (WorldTeam) o di secondo (ProTeam) entro i venticinque anni. La finestra d\u0026rsquo;età serve a rendere confrontabili annate diverse; chi arriva più tardi qui non risulta.\nDi quei 2 813, però, soltanto 2 183 erano in classifica già da Esordienti, la categoria dei tredici e quattordici anni, l\u0026rsquo;Under 15 della letteratura. L\u0026rsquo;Under 23, invece, copre quattro annate, dai diciannove ai ventidue. È dagli Esordienti che parte l\u0026rsquo;imbuto6.\ncategoria atleti in classifica di cui partiti dagli Esordienti quota dei 2 183 Esordienti 2 183 2 183 100% Allievi 1 741 1 303 59,7% Juniores 1 051 745 34,1% Under 23 342 224 10,3% professionisti 77 66 3,0% le due colonne di mezzo non coincidono perché in ogni categoria entrano anche ragazzi che negli Esordienti non c\u0026rsquo;erano: la quota è calcolata sulla terza colonna, non sulla seconda\nSul primo scalino c\u0026rsquo;è chi a tredici anni era già arrivato almeno una volta nei primi cinque, e i giovani ciclisti italiani sono molti di più. Da lì all\u0026rsquo;ultimo scalino si scende di trentatré volte, molto meno di tre ordini di grandezza.\nGuardiamo l\u0026rsquo;ultima riga. I professionisti delle coorti sono 77, ma solo 66 erano in classifica già da Esordienti; gli altri 11 sono entrati più tardi. Mettere 77 sopra 2 183 vorrebbe dire contare al numeratore gente che nel denominatore non c\u0026rsquo;è. La quota è calcolata sui sessantasei che da lì venivano davvero: tre su cento.\nLetto al contrario, è il numero più utile di questa puntata: uno su sette dei futuri professionisti, a tredici anni, in classifica non c\u0026rsquo;era. Se guardi una lista di Esordienti sapendo che i migliori d\u0026rsquo;Italia sono lì dentro, ricordati che uno su sette di quelli che arriveranno, in quel momento, non c\u0026rsquo;è ancora.\nTra quegli undici ci sono anche Christian Scaroni, vincitore di una tappa al Giro d’Italia 2025, e Jonathan Milan, due volte maglia ciclamino al Giro e maglia verde al Tour 2025: entrambi compaiono nelle nostre classifiche soltanto da Allievi di secondo anno.\nSembra un imbuto, e in parte lo è. Ha però una proprietà che rovina le letture semplici: non è lo stesso gruppo che si assottiglia. Fra un quarto e un terzo degli atleti di ogni categoria non è mai comparso in Esordienti, e in Under 23 sono più di uno su tre. Chi legge l\u0026rsquo;imbuto come «di quei duemila ne sono rimasti trecento» conta fra i sopravvissuti gente che non era mai partita.\nIl numero che userò per tutta la serie porta allora la sua etichetta attaccata: ventisette su mille fra chi è mai comparso in classifica, cioè 77 su 2 813. Il tre su cento della tabella è un\u0026rsquo;altra frazione: 66 su 2 183, i soli partiti dagli Esordienti.\nTre cose da sapere prima di leggere una classifica La prima: sparire da una classifica non significa aver smesso. Il 30,6% degli atleti salta almeno una stagione e poi ricompare; il 6,7% del totale torna dopo un’assenza di almeno due stagioni.6. Un rientro dimostra che quell\u0026rsquo;assenza non era un abbandono. Inoltre, accanto ai 77 professionisti delle coorti ci sono 108 atleti ancora a punti dopo i ventidue anni che professionisti non sono diventati. Sono più dei professionisti stessi. Chi non diventa professionista non ha per forza smesso. Tornerà utile nella seconda puntata, quando l\u0026rsquo;assenza da una classifica diventerà un\u0026rsquo;informazione da maneggiare con cura.\nPer approfondire, nel documento tecnico: Uscire dalla classifica non è smettere, Quando si smette e Il passaggio di categoria è una rottura?.\nLa seconda: come sono fatte le categorie. Negli Esordienti la fonte pubblica due classifiche separate, una per annata: primo e secondo anno corrono gare loro e non si fanno concorrenza. Dagli Allievi in su la classifica è una sola e le due annate ci convivono, correndo le stesse gare. La conseguenza si vede nei posti a punti, cioè i cinque piazzamenti che in ogni gara danno punti7.\ncategoria quota dei posti presa dal primo anno Esordienti, classifiche separate 49,7% Allievi, classifica unica 26,7% Juniores, classifica unica 33,3% Under 23, classifica unica 11,8% il 33,3% degli Juniores rompe la discesa, e non so dire perché: in quella categoria quelli del primo anno sono una fetta più grossa del gruppo che in Allievi\nIl 49,7% degli Esordienti è quello che ci si aspetta da due classifiche separate. Il numero da guardare è quanto resta al primo anno dove la lista è una sola.\nChe con due classifiche separate le annate si dividano i posti a metà è ovvio, e vale anche fra le ragazze: ciascuna corre contro le proprie coetanee. La domanda interessante è un\u0026rsquo;altra. Quanto riescono a raccogliere i più piccoli dove la classifica è una sola e le gare sono le stesse per tutti? Poco più di un quarto in Allievi, un terzo in Juniores, un ottavo in Under 23. Il metro è la parità: con due annate sarebbe metà; in Under 23, che di annate ne ha quattro, un quarto. Rispetto a una divisione uniforme fra annate, il primo anno raccoglie circa metà dei posti che gli spetterebbero negli Allievi e Under 23, e due terzi negli Juniores. I posti nella top 5 ovviamente non spariscono: li vincono i più grandi. Chi legge la classifica di un quindicenne al primo anno di Allievi sta guardando un ragazzo che corre contro avversari con un anno di sviluppo in più.\nEppure qualcuno riesce a emergere già contro i più grandi: Filippo Ganna, nel 2011, da Allievo di primo anno raccolse 65 punti e sei vittorie, chiudendo settimo nella classifica delle due annate. Sarebbe poi diventato due volte campione del mondo a cronometro.\nChe il meccanismo sia davvero questo lo dice anche il ciclismo femminile, dove la fonte ha cambiato struttura in corsa. Fino al 2021 le Esordienti avevano una classifica sola, e in quel periodo il primo anno raccoglieva il 28,5% dei posti: lo stesso ordine di grandezza del 26,7% maschile, in un\u0026rsquo;altra categoria e in un altro movimento. Dal 2022 le classifiche sono due e la quota sale a metà8, come c\u0026rsquo;era da aspettarsi.\nStessa categoria e stessa età, prima e dopo il cambio di regolamento: il numero che porta l\u0026rsquo;informazione è il 28,5% di sinistra, cioè quanto raccoglie il primo anno quando la lista è una sola.\nPer approfondire, nel documento tecnico: Ma il primo anno non sparisce per mancanza di posti, Quanto sono concentrati i punti e Un cambio di regolamento che vale un esperimento.\nLa terza: il calendario si sta accorciando. In una stagione la classifica registra in media 320 gare in Esordienti e 146 in Under 23. Soprattutto, fra il 2009 e il 2025 le gare registrate calano del 40% negli Esordienti e del 58% in Under 23. Colpa della pandemia? No: il calo era cominciato prima, e dopo il 2020 non si è tornati ai valori di prima. Nella finestra in cui il confronto è possibile i tesserati Esordienti calano di circa un decimo e le gare di quasi un quinto. Il movimento si restringe, e il calendario si restringe più in fretta.\nQuattro linee per categoria dal 2007 al 2025: gli Esordienti scendono da 433 gare nel 2009 a 261, gli Allievi da 642 a 309, gli Juniores da 430 a 199, l\u0026rsquo;Under 23 da 222 a 88, con un crollo comune nel 2020.\nMeno gare vuol dire meno posti a punti, quindi una classifica che si assottiglia da sola: due annate lontane non si possono confrontare come se niente fosse. Il calo, tuttavia, riguarda il movimento maschile e non il ciclismo giovanile in generale. Dove il confronto regge, cioè in Allievi e Juniores, fra il 2011 e il 2025 le gare maschili calano di oltre un terzo mentre quelle femminili tengono, e il divario si stringe: da una gara femminile ogni sette maschili a una ogni cinque in Allievi, da una ogni dieci a una ogni cinque in Juniores8.\nPer approfondire, nel documento tecnico: Quanti posti c\u0026rsquo;erano prima e L\u0026rsquo;attrito della classifica e quello vero.\nCosa misura quel punteggio Un\u0026rsquo;ultima cosa sullo strumento. Nel 2025 due ricercatori olandesi hanno pubblicato la critica più efficace che si potesse muovere a uno studio di questo tipo: un ranking federale che non pesa le gare per livello mette sullo stesso piano chi corre gare internazionali e chi corre gare sotto casa.\nSulla classifica italiana quella critica è disinnescata a metà, perché il regolamento le gare le pesa. Lo si vede indirettamente dividendo i punti per il numero di piazzamenti: 2,67 punti per piazzamento in Esordienti, 2,73 in Allievi, 3,02 in Juniores e 4,17 in Under 239. Il salto cade esattamente dove entrano in funzione i moltiplicatori del regolamento, che fanno valere di più le gare di livello alto: dagli Juniores in su. In Esordienti e Allievi la scala è piatta, e torna con il fatto che a quelle età un calendario internazionale non esiste.\nL\u0026rsquo;altra metà della critica resta in piedi. Un ranking premia comunque chi va forte nelle gare più frequenti in calendario: uno scalatore, in un calendario fatto di percorsi veloci, ha meno occasioni di andare a punti. Per correggerlo servirebbe sapere com\u0026rsquo;era fatta ogni gara, e quel dato non esiste.\nPer approfondire, nel documento tecnico: Lo stesso punteggio è lo stesso risultato? e Ma i pari merito sono moltissimi.\nCosa resta Ogni percentuale che leggerai nelle prossime puntate ha come denominatore un settimo dei ragazzi tesserati, un gruppo già selezionato, e non la loro totalità e quindi non si possono estendere a tutti i tesserati. Di quanto, esattamente, non lo so.\nL\u0026rsquo;imbuto non è lo stesso gruppo che si assottiglia: in ogni categoria c\u0026rsquo;è una quota consistente di ragazzi che negli Esordienti non c\u0026rsquo;erano, ed entrare tardi è normalissimo. Un futuro professionista su sette, a tredici anni, non era in classifica, ed è un settimo diverso da quello dei tesserati qui sopra, che è la quota di chi in classifica ci entra.\nAdesso sai di chi parliamo. A che età quel piazzamento comincia a dire qualcosa sul futuro? È il tema della prossima puntata.\nCome lo sappiamo\nLe classifiche vengono da ciclismo.info, stagioni 2007-2025. I tesserati vengono dai dati statistici pubblicati dalla Federazione Ciclistica Italiana e coprono le stagioni 2018-2025. Le due serie non sono perfettamente allineate, perché il tesseramento è per categoria e non per specialità: la copertura calcolata è quindi un limite inferiore.\nLa letteratura è riassunta in una rassegna di ventidue studi pubblicati fra il 2006 e il 2025, con riferimento completo, DOI e limiti di ciascuno. Cinque li ho letti per intero; per gli altri ho lavorato su abstract e citazioni incrociate.\nSchumacher Y.O. et al. (2006), Success in elite cycling: a prospective and retrospective analysis of race results, Journal of Sports Sciences 24(11), 1149-1156. DOI 10.1080/02640410500457299.\u0026#160;\u0026#x21a9;\u0026#xfe0e;\nTutti gli studi citati in questa sezione, con riferimenti e limiti, stanno in docs/literature_review.md.\u0026#160;\u0026#x21a9;\u0026#xfe0e;\u0026#160;\u0026#x21a9;\u0026#xfe0e;\u0026#160;\u0026#x21a9;\u0026#xfe0e;\nCalcolo in report/moduli/provenienza.py.\u0026#160;\u0026#x21a9;\u0026#xfe0e;\nFederazione Ciclistica Italiana, Norme Attuative 2027, Cat. Esordienti / Allievi / Donne Esordienti / Donne Allieve, art. 4.4.21 per la scala dei punti e artt. 4.2.1 e 4.2.5 per le due annate che corrono separate. Sono le norme della federazione, cosa diversa dal regolamento con cui la fonte costruisce la propria classifica, che pesa le gare per livello.\u0026#160;\u0026#x21a9;\u0026#xfe0e;\nCalcolo in report/moduli/copertura.py, dai tesserati raccolti in riferimenti/tesserati_fci.csv.\u0026#160;\u0026#x21a9;\u0026#xfe0e;\nCalcolo in report/moduli/attrito.py.\u0026#160;\u0026#x21a9;\u0026#xfe0e;\u0026#160;\u0026#x21a9;\u0026#xfe0e;\nCalcolo in report/moduli/posti.py. I posti sono stimati dai piazzamenti nei primi cinque, cinque per gara.\u0026#160;\u0026#x21a9;\u0026#xfe0e;\nCalcolo in report/moduli/ragazze.py.\u0026#160;\u0026#x21a9;\u0026#xfe0e;\u0026#160;\u0026#x21a9;\u0026#xfe0e;\nCalcolo in R/30_misura.R, reso da report/moduli/misura.py.\u0026#160;\u0026#x21a9;\u0026#xfe0e;\n","permalink":"https://lucabontempi.com/blog/ciclismo-giovanile-di-chi-stiamo-parlando/","summary":"\u003cp\u003e\u003cem\u003eQuesta serie parla dei ranking giovanili italiani e del passaggio al professionismo. Prima dei miei numeri racconto cosa la ricerca sa già e, soprattutto, di chi parlano i numeri delle prossime puntate. È lì che si annidano quasi tutti gli errori.\u003c/em\u003e\u003c/p\u003e\n\u003ch2 id=\"ventanni-di-ricerca-in-poche-righe\"\u003eVent\u0026rsquo;anni di ricerca, in poche righe\u003c/h2\u003e\n\u003cp\u003eAl Mondiale juniores su strada ogni nazione porta al massimo \u003cstrong\u003esei\u003c/strong\u003e corridori, e a portarne sei sono solo le nazioni in cima al ranking. Gli Juniores (o Under 19) sono la categoria dei diciassette e diciotto anni. Chi ci ha corso ricorda benissimo quanto sia difficile strappare una convocazione. Chi arriva su quella linea di partenza ha superato ogni selezione che il ciclismo giovanile riesca a mettere in piedi. Più predestinato di così non si può.\u003c/p\u003e","title":"Ciclismo giovanile e professionismo: il panorama attuale"},{"content":"Questo progetto è nato da una domanda che non ha nulla a che fare con gli app store. Quando ho scritto la tesi, usare l\u0026rsquo;AI generativa era sconsigliato, il che in pratica significava non usarla. ChatGPT era pubblico da circa due mesi e nessuno contestava la regola.\nTre anni dopo, quella stessa regola sembra quasi impossibile da far rispettare e ancora più difficile da giustificare. Farei fatica a trovare uno studente che non si faccia aiutare dall\u0026rsquo;AI per la tesi, e ancora più fatica a sostenere che non dovrebbe farlo. È questo ribaltamento, non la letteratura sugli app store, che mi ha spinto a riaprire il mio lavoro quest\u0026rsquo;anno. Non volevo riscriverlo. Volevo vedere cosa sarebbe emerso da una seconda lettura, più scettica, di qualcosa a cui avevo dato fiducia per tre anni senza più toccarlo.\nOgni grafico e ogni numero qui sotto provengono da uno script nel repository della revisione.\nNel febbraio 2023 ho consegnato una tesi su una domanda circoscritta: quando una persona arriva sulla pagina di un\u0026rsquo;app nello store, cosa la convince a installarla? Ho studiato tre elementi: il voto medio delle recensioni, il numero di download e il brand dello sviluppatore, espresso dal nome dell\u0026rsquo;app, dall\u0026rsquo;icona e dal nome dello sviluppatore. Ho modificato gli screenshot di una vera pagina del Play Store per ottenere una versione forte e una debole di ciascun elemento. Ogni partecipante vedeva tre pagine, una per elemento, in ordine casuale, ciascuna con una delle due versioni estratta a caso. Nessuno vedeva entrambe le versioni dello stesso elemento. Poi misuravo quella che la letteratura chiama intenzione di download.\nLa conclusione pubblicata era questa: il brand conta di più, la reputazione viene seconda e la popolarità non fa praticamente nulla, almeno in una categoria con scarsi effetti di rete.\nQuest\u0026rsquo;anno ho riaperto i file. I dati hanno retto: la randomizzazione funzionava, non ci sono valori mancanti e ogni regressione della tesi si riproduce fino alla quarta cifra decimale. L\u0026rsquo;analisi, invece, non ha retto. Avevo commesso tre errori, nessuno nei dati. Tutti e tre riguardavano quello che ne avevo fatto.\nPrimo errore: tre regressioni non bastano per un confronto La conclusione principale nasceva dall\u0026rsquo;accostamento di tre coefficienti: 1,53 per il brand, 0,50 per la reputazione, 0,36 per la popolarità. Il brand è chiaramente il più grande. Fine.\nPeccato che quei numeri provenissero da tre regressioni diverse, su tre gruppi diversi di partecipanti. La tesi non verifica mai se 1,53 differisca da 0,50. Ho guardato i due numeri e ho tratto la conclusione.\nHo quindi eseguito il test che avrei dovuto fare dall\u0026rsquo;inizio: un unico modello su tutte le 1.473 risposte, tenendo conto del fatto che ogni persona ne forniva tre.\nElemento Effetto sull\u0026rsquo;intenzione di download (scala 1–7) IC al 95% Brand dello sviluppatore +1,05 da 0,81 a 1,29 Reputazione (voto) +0,85 da 0,61 a 1,09 Popolarità (download) +0,22 da −0,02 a 0,46 Nella media dei tre giudizi espressi da ciascuno, brand e reputazione non sono distinguibili (p = 0,24). Entrambi superano chiaramente la popolarità. La mia gerarchia a tre livelli è in realtà a due, almeno in quella media, e la precisazione conta.\nLa parte interessante è quanto mi sia costata la frammentazione. Dividere un campione in nove sottocampioni ha disperso gran parte della potenza statistica che avevo raccolto. L\u0026rsquo;effetto della reputazione è aumentato di due terzi, da 0,50 a 0,85. Circa metà dell\u0026rsquo;aumento dipende dalla precisione: sulla stessa domanda relativa alla prima app, un unico modello su tutti i dati stima 0,68, non 0,50. L\u0026rsquo;altra metà deriva da una domanda diversa, perché la reputazione acquista forza nei giudizi successivi.\nGli stessi tre effetti, stimati in tre modi. Punti vuoti: la tesi, una regressione per elemento sul proprio sottocampione. Triangoli: la stessa domanda, sulla prima app, con un modello unico. Punti pieni: la media dei tre giudizi, che risponde a una domanda diversa.\nNon volevo fidarmi ciecamente del mio nuovo modello preferito, quindi ho ripetuto il confronto in tutti i modi difendibili: 24 combinazioni di campione, tipo di modello e controlli. Una non convergeva e l\u0026rsquo;ho esclusa. Il brand risulta superiore in 4 delle 23 rimaste. Non sono quattro risultati casuali: sono esattamente quelli che usano soltanto la prima app vista da ogni partecipante.\nNon è una coincidenza. Nel 2022 avevo scelto la prima app di proposito, perché la prima schermata è l\u0026rsquo;unica che si giudica senza avere ancora nulla con cui confrontarla. Su quella quantità, quella per cui era stato costruito il disegno sperimentale, il brand supera effettivamente la reputazione: +0,76 punti, p = 0,03, e +1,02 (p = 0,007) nel campione effettivamente usato dalla tesi.\nLa conclusione principale, quindi, non era un artefatto. La tesi, però, faceva due affermazioni: il brand vince sulla prima app e perde il vantaggio dopo che l\u0026rsquo;utente ne ha viste altre. Non ne verificava nessuna. Ora il test c\u0026rsquo;è. La prima regge (+0,76, p = 0,03). La seconda non raggiunge la significatività (p = 0,10): i dati mostrano una tendenza, non un cambiamento dimostrato.\nL\u0026rsquo;errore è più circoscritto di «la classifica crolla», ma più difficile da liquidare. Nove regressioni su nove sottocampioni non condividono parametri. Questo significa che non era possibile verificare alcun confronto tra i tre elementi, e infatti non ne è stato verificato nessuno. Un unico modello su tutte le 1.473 risposte restituisce la stessa stima per la prima app, più il test che mancava nella tesi, più ciò che succede dopo aver visto le alternative.\nOgni specificazione difendibile dello stesso confronto. Il brand vince chiaramente solo dove mirava il disegno: sulla prima app vista. Negli altri casi brand e reputazione non sono distinguibili.\nSecondo errore: il mio «manipulation check» non lo era Il risultato che gli altri mi ripetevano era quello controintuitivo: il numero di download non conta. In un mercato ossessionato dalle installazioni, è una frase che colpisce. Non è però esattamente ciò che sostenevo. Avevo usato app scanner perché hanno scarsi effetti di rete: uno scanner non diventa più utile perché lo usano milioni di altre persone. Il numero di download può quindi segnalare ciò che pensano gli altri, ma non promettere un valore aggiuntivo. La mia affermazione era che la popolarità fosse inefficace in quel contesto.\nNon penso più che i dati sostengano nemmeno questo. Per due ragioni. Una avrei dovuto vederla nel 2022. L\u0026rsquo;altra l\u0026rsquo;avevo vista, ma non ne avevo tratto le conseguenze.\nLa prima è imbarazzante. Sono tornato alle immagini usate come stimolo. La versione a bassa popolarità mostra 10.000+ download accanto a 84.000 recensioni. Più recensioni che download. È impossibile. Il numero di recensioni era uguale nelle due condizioni: il difetto riguarda quindi soltanto la versione debole, proprio quella su cui si reggeva il confronto.\nLa seconda è più sottile e riguarda l\u0026rsquo;etichetta, più che l\u0026rsquo;analisi. La tesi riporta, modello per modello, un manipulation check: la quota di partecipanti che confermava di aver notato ogni elemento. In realtà avevo chiesto, una sola volta e alla fine, quali fattori hai preso in considerazione?, con nove caselle da selezionare. Non verifica che la manipolazione sia stata percepita: raccoglie il racconto a posteriori di ciò che le persone pensano di aver fatto.\nLo sapevo già allora. La tesi definisce la domanda «una versione di compromesso, e non ideale». La scelta successiva era deliberata: misurarla, riportarla, non condizionarvi alcuna analisi. Era la scelta giusta e la rifarei. Quello che non ho fatto è esplicitarne la conseguenza. Uno studio in cui l\u0026rsquo;unica misura della percezione è un\u0026rsquo;autodichiarazione successiva al trattamento non verifica che qualcuno abbia visto la manipolazione. La parola check, ripetuta in nove tabelle, suggerisce silenziosamente il contrario.\nA pagarne il prezzo è soprattutto la popolarità, perché è lì che una verifica della percezione sarebbe servita di più. Tra il 60% che ha selezionato «numero di download», l\u0026rsquo;effetto della popolarità più che raddoppia, fino a +0,49, e diventa significativo. Lo propongo come esempio, non come correzione, perché il confronto è quasi circolare: chi è influenzato dai download è anche più propenso a dire di averli guardati. Resta il fatto che, tra uno stimolo difettoso e una misura dell\u0026rsquo;attenzione non verificabile, il risultato nullo aveva meno sostegno di quanto gli avessi attribuito.\nCosa resta, allora, della popolarità? Prima di guardare i risultati ho fissato il più piccolo effetto rilevante nella pratica. Ogni modo ragionevole di ricavarlo porta tra 0,26 e 0,33 punti della scala. Rispetto a quella soglia:\nsulla prima app vista, senza alternative già osservate, la popolarità funziona: +0,45, e l\u0026rsquo;effetto è effettivo; sulla seconda è −0,22 e sulla terza +0,40. Entrambi i risultati sono inconcludenti. Gli intervalli oltrepassano la soglia in entrambe le direzioni: non posso distinguere «nessun effetto» da «troppo piccolo per contare», né da un effetto piuttosto grande. All\u0026rsquo;inizio avevo riassunto il secondo punto con un unico numero, +0,07 dopo il confronto, usando un modello che tratta seconda e terza app come un\u0026rsquo;unica condizione. È lo stesso genere di errore che avevo appena criticato nella tesi: comprimere −0,22 e +0,40 in un numero descrive una situazione che nessuno ha vissuto. Inoltre assecondava una storia ordinata che mi piaceva, perché la terza app non è distinguibile dalla prima: differenza di 0,05, p = 0,86. Il calo riguarda soltanto la seconda e nessuna differenza tra posizioni supera la correzione per averle esaminate tutte e tre.\nQuali risultati nulli sono davvero nulli. La fascia indica l\u0026rsquo;irrilevanza pratica, fissata prima dei test. Un intervallo molto più ampio della fascia significa che lo studio non riusciva a vedere, non che non ci fosse nulla da vedere.\n«Il numero di download non conta» era troppo forte. Quello che posso difendere è più limitato e meno memorabile: dove gli effetti di rete sono bassi, i download influenzano l\u0026rsquo;intenzione sulla prima app vista; dopo, questo studio non sa dirlo. È comunque abbastanza per ritirare la conclusione che avevo pubblicato.\nTerzo errore: il silenzio non è una prova La tesi conclude che non ci sia interazione tra i tre elementi. Per ogni schermata ne veniva modificato uno solo, quindi qui «interazione» può significare soltanto qualcosa di più ristretto: l\u0026rsquo;effetto del voto di questa app dipende dal brand di un\u0026rsquo;app vista prima? La sezione metodologica riconosce questo limite. L\u0026rsquo;abstract no.\nIl problema più profondo è che, comunque, la risposta non dice nulla. Il disegno avrebbe potuto rilevare soltanto interazioni di 1,3–1,5 punti sulla scala a sette punti: tanto quanto gli stessi effetti principali, compresi tra 0,2 e 1,0, mentre le interazioni sono solitamente molto più piccole. Non trovarle non dimostra che non esistano. Lo strumento era troppo grossolano per vederle.\nI due risultati di moderazione su cui avevo costruito consigli manageriali sono il caso opposto. Qui i dati possono dire più di «non significativo»: entrambi gli effetti rientrano nel margine che avevo fissato in anticipo come troppo piccolo per contare. Non sono semplicemente indimostrati: sono praticamente assenti.\nCosa regge e una cosa nuova L\u0026rsquo;effetto del brand supera ogni controllo. Degli 81 coefficienti stimati in nove regressioni, esattamente due sopravvivono alla correzione per il numero di test, ed entrambi sono l\u0026rsquo;effetto del brand.\nBrand e reputazione hanno la stessa forza soltanto in media. Separando le posizioni nella sequenza, il quadro è più preciso di quanto sostenessi all\u0026rsquo;inizio:\nPrima app vista Seconda Terza Brand dello sviluppatore +1,44 +0,88 +0,92 Reputazione +0,68 +0,63 +1,13 Popolarità +0,45 −0,22 +0,40 Effetti nelle tre posizioni della sequenza, con intervalli al 95%. Il brand è più forte sulla prima app, la reputazione sulla terza, ma nessuna differenza tra posizioni supera la correzione per confronti multipli. È il profilo di un andamento, non un andamento dimostrato.\nIl brand domina quando non c\u0026rsquo;è ancora nulla con cui confrontarsi. È il divario che la tesi aveva trovato, e regge. Dopo aver visto alternative, i due convergono: 0,89 contro 0,92, in un modello che tratta ciò che viene dopo la prima app come un\u0026rsquo;unica condizione anziché due. In termini probabilistici, il brand supera la reputazione con probabilità del 99% prima del confronto e del 48% dopo: un lancio di moneta. I dati non dimostrano però che le due situazioni differiscano tra loro (p = 0,10), e per nessuno dei tre elementi una differenza a coppie tra posizioni supera la correzione. Quel movimento è quindi il profilo di qualcosa che merita un test adeguato. Non è un risultato dimostrato.\nLa grandezza plausibile degli effetti alla luce dei dati. Quando le curve si sovrappongono, i dati non distinguono i due elementi: ecco cosa significa non distinguibili.\nUna lettura pratica c\u0026rsquo;è, ma regge solo se regge quel profilo. Un brand forte vale di più quando l\u0026rsquo;utente non sta confrontando alternative: un collegamento da una ricerca, una pubblicità, un consiglio. I voti diventano preziosi quando si esplora e si confronta. Per un nuovo concorrente è una notizia migliore di quanto sembri: non puoi comprarti il brand di Adobe, ma puoi essere l\u0026rsquo;app che supera il confronto.\nPoi c\u0026rsquo;è un risultato davvero nuovo, che prima di eseguire l\u0026rsquo;analisi avevo previsto per iscritto non sarebbe emerso. Le persone differiscono molto per quanto sono influenzate da questi elementi, e la differenza è sistematica. Chi è in generale più propenso a scaricare app è meno influenzato da ciò che mostra la pagina: correlazione −0,59. I segnali funzionano sugli indecisi. Una media complessiva lo nasconde del tutto.\nUn punto per partecipante: propensione di base al download e sensibilità agli elementi della pagina. I segnali funzionano sugli indecisi. I punti sono previsioni del modello, riportate verso la media quando il partecipante fornisce poche informazioni.\nQuattro controlli che non hanno cambiato nulla Una nuova analisi vale quanto i controlli che avrebbero potuto metterla in difficoltà. Quattro non ci sono riusciti.\nLe persone non trattano una scala da 1 a 7 come un righello. La risposta 6 copre un intervallo dell\u0026rsquo;intenzione sottostante quasi doppio rispetto alla risposta 3: anche chi è piuttosto sicuro si ferma un passo prima del massimo, evitando gli estremi della scala. Ho ristimato tutto senza assumere distanze uguali. Gli effetti sono cambiati di 0,7–3,0 punti percentuali e la graduatoria è rimasta la stessa.\nDove si collocano le soglie delle risposte sulla scala dell\u0026rsquo;intenzione sottostante, rispetto a una scala con distanze uniformi.\nGli effetti non dipendono in modo rilevabile da chi risponde: né età, né genere, né istruzione, né occupazione. Gli studenti rispondono alla reputazione circa il doppio degli occupati. È una storia ordinata, ma non supera il proprio test. Con un campione composto al 69% da studenti, non avrebbe potuto superarlo.\nEscludere i 33 partecipanti che hanno risposto troppo in fretta o dato la stessa risposta a un\u0026rsquo;intera scala fa aumentare leggermente tutti gli effetti, come ci si aspetta eliminando rumore. Non cambia altro.\nInfine, i grafici della tesi non usano le risposte grezze da 1 a 7, ma una versione centrata sulla media, sottraendo una costante per ogni elemento. Non è un problema se il modello sa già quale gruppo sta osservando, e il mio lo sapeva: ristimando sulle risposte grezze, nessun coefficiente cambia di più di 0,00000000005. Per il lettore, però, un costo c\u0026rsquo;è. Le altezze delle tre linee non sono confrontabili, solo le pendenze lo sono, e in quattro pannelli su sei la versione centrata ordina i tre elementi diversamente da quella grezza. Avevo letto una pendenza, quindi la conclusione regge. In un disegno a misure ripetute sarebbe invece stato importante centrare per partecipante, non per elemento. Non basta sottrarre una media: occorre includere il partecipante nel modello, proprio ciò che mancava alla mia ANOVA. Farlo nel modo sbagliato, per sottrazione, avrebbe ridotto ogni effetto di circa un terzo.\nLe stesse medie sulla scala grezza e sulla scala centrata usata nei grafici della tesi. Le pendenze sono identiche. L\u0026rsquo;ordine delle tre linee no.\nCosa mi hanno insegnato tre anni di distanza Tre cose, nessuna sugli app store.\nProteggere la quantità che si vuole stimare e dividere il campione sono cose diverse: ho fatto la seconda pensando di fare la prima. Misurare soltanto la prima app vista era l\u0026rsquo;intuizione giusta: è l\u0026rsquo;unico giudizio espresso prima di avere qualcosa con cui confrontarsi. Stimarlo con nove regressioni scollegate era sbagliato, perché regressioni senza parametri condivisi non consentono quel confronto, ed è esattamente il confronto che ho fatto. Ho enunciato entrambe le metà del risultato senza verificarne nessuna.\n«Non significativo» non è un risultato finché non dici cosa avresti potuto rilevare. Ogni risultato nullo che ho riportato meritava una frase sul più piccolo effetto che il disegno era in grado di vedere. Tre si dissolvono davanti a quella domanda.\nRiguarda gli stimoli. Il problema peggiore del progetto, uno screenshot con più recensioni che download, non era nei modelli né nei dati. Era in un\u0026rsquo;immagine che nel 2022 avevo guardato cento volte senza controllarla davvero.\nUna nota su come ho lavorato. La nuova analisi si è avvalsa dell\u0026rsquo;AI. Era nata da una piccola curiosità: come avrebbe scritto un\u0026rsquo;AI la mia tesi magistrale? La risposta era meno interessante della domanda a cui ha portato: come l\u0026rsquo;avrebbe revisionata? Non ha trovato nulla che non avrei potuto trovare da solo nel 2023. Ha fatto l\u0026rsquo;unica cosa che non avevo fatto: verificare ogni affermazione rispetto alle prove che avevo già, comprese quelle che facevano comodo. È una soglia bassa e la prima volta non l\u0026rsquo;ho superata.\nNel 2023 avevo usato l\u0026rsquo;AI soltanto per controllare la grammatica inglese. Quest\u0026rsquo;anno lo stesso tipo di strumento mi ha aiutato a smontare e ricostruire l\u0026rsquo;analisi in meno di due settimane. Nessuno dei tre errori richiedeva una macchina: una calcolatrice e un po\u0026rsquo; di pazienza li avrebbero individuati già nel 2022. Mancava la decisione di dubitare di qualcosa in cui credevo già. E quella decisione resta mia.\nL\u0026rsquo;AI accelera l\u0026rsquo;esecuzione: ripetere ventiquattro specificazioni richiede pochi minuti. Il giudizio, capire se un risultato nullo sia davvero nullo o se un risultato comodo resista al dubbio, continua alla velocità di sempre. Il rischio è confondere le due velocità. Ho scritto questo articolo con lo stesso tipo di aiuto, quindi merita lo stesso esame critico. Se nasconde un quarto errore, vorrei saperlo.\nRevisione completa, codice, regole decisionali fissate in anticipo e log sono nel repository della revisione; tesi originale, dati e codice sono qui.\n","permalink":"https://lucabontempi.com/it/blog/reopening-my-thesis/","summary":"\u003cp\u003eQuesto progetto è nato da una domanda che non ha nulla a che fare con gli app store. Quando ho scritto la tesi, usare l\u0026rsquo;AI generativa era sconsigliato, il che in pratica significava non usarla. ChatGPT era pubblico da circa due mesi e nessuno contestava la regola.\u003c/p\u003e\n\u003cp\u003eTre anni dopo, quella stessa regola sembra quasi impossibile da far rispettare e ancora più difficile da giustificare. Farei fatica a trovare uno studente che non si faccia aiutare dall\u0026rsquo;AI per la tesi, e ancora più fatica a sostenere che non dovrebbe farlo. È questo ribaltamento, non la letteratura sugli app store, che mi ha spinto a riaprire il mio lavoro quest\u0026rsquo;anno. Non volevo riscriverlo. Volevo vedere cosa sarebbe emerso da una seconda lettura, più scettica, di qualcosa a cui avevo dato fiducia per tre anni senza più toccarlo.\u003c/p\u003e","title":"Ho chiesto a un'AI di rivedere la mia tesi. Ha trovato tre errori che avrei potuto trovare da solo"},{"content":"Nota: pubblico questo articolo nell\u0026rsquo;agosto 2023, ma ho cominciato a scriverlo nel settembre 2022, quando vivevo ancora a Tübingen. Nel frattempo molte cose potrebbero essere cambiate, soprattutto i prezzi, influenzati dall\u0026rsquo;inflazione dell\u0026rsquo;ultimo anno. Ho comunque deciso di completarlo per chiudere il trittico di articoli che mi ero proposto di scrivere.\nDopo un anno da studente a Tübingen, provo a condividere quello che ho imparato sulla città e sui suoi dintorni.\nIl mio anno da studente internazionale a Tübingen sta per finire e sarebbe un peccato disperdere ciò che ho scoperto in questi mesi. Per questo ho deciso di scrivere una guida e raccontarvi le mie esperienze in questa città meravigliosa. Passeremo dai paesaggi e dallo sport alle esperienze da ricordare, fino al cibo e al divertimento. Cominciamo!\nEsperienze e sport Molte delle esperienze del mio anno a Tübingen si ritrovano anche nella raccolta fotografica dedicata alla città.\nSono un grande appassionato di sport dal vivo e di eventi sportivi, ai quali ho dedicato una parte importante del mio budget. Purtroppo non sono riuscito a partecipare ad alcuni appuntamenti, come la Chocolart, sospesa nel 2021 per le restrizioni legate al Covid.\nTübingen Tigers Mi piace molto il basket e sono stato felicissimo di scoprire che Tübingen aveva una squadra, i Tigers. I Tübingen Tigers giocavano nella seconda divisione tedesca e, contrariamente a quanto si potrebbe pensare, l\u0026rsquo;atmosfera nel palazzetto era fantastica!\nGiocano alla Paul Horn-Arena e sono disponibili anche biglietti scontati per studenti, intorno ai 10 euro.\nVfB Stuttgart Per trovare un grande club calcistico bisogna spostarsi a Stoccarda. Il VfB Stuttgart ha vinto cinque titoli di Bundesliga, l\u0026rsquo;ultimo nel 2007. Negli anni precedenti al mio soggiorno, però, aveva faticato a rimanere in prima divisione.\nI biglietti delle partite erano una delle poche cose che costavano meno che in Italia: meno di 30 euro, se ricordo bene. Lo stadio è bello e l\u0026rsquo;esperienza merita.\nATP 250 di Stoccarda Quando ho scoperto che a Stoccarda si teneva un torneo di tennis, ho passato qualche ora a chiedermi se andarci. Alla fine sono stato contento di aver scelto i quarti di finale, anche se il biglietto per quelle quattro partite sfiorava gli 80 euro.\nHo avuto l\u0026rsquo;occasione di vedere dal vivo grandi giocatori come Matteo Berrettini, il campione olimpico e di Wimbledon Andy Murray e Nick Kyrgios. Considerati i prezzi di altri tornei, direi che sono stato fortunato con quella giornata.\nStuttgarter Frühlingsfest Tra la fine di aprile e l\u0026rsquo;inizio di maggio, a Cannstatt, si tiene per una ventina di giorni la Stuttgarter Frühlingsfest.\nLa festa è piena di attrazioni da luna park, cibo tradizionale tedesco e, ovviamente, birra. Un\u0026rsquo;esperienza tedesca da non perdere, secondo me.\nStocherkahnrennen La tradizionale Stocherkahnrennen è probabilmente uno dei momenti più belli del semestre estivo a Tübingen. Si tiene quasi ogni anno dal 1956, nel giorno del Corpus Domini, un giovedì che cade solitamente a giugno.\nÈ l\u0026rsquo;occasione migliore per sentirsi parte della grande comunità di Tübingen!\nStadtfest Tübingen Un\u0026rsquo;altra serata davvero divertente è stata quella dello Stadtfest a luglio: un intero fine settimana dedicato alla città, con tanti eventi musicali, cibo e birra. Se vivessi ancora a Tübingen, non me lo perderei!\nUmbrisch-Provenzalischer Markt A settembre, nel centro di Tübingen, si tiene l\u0026rsquo;Umbrisch-Provenzalischer Markt, una curiosa fiera che unisce la cultura gastronomica francese e quella italiana. Una bella occasione per assaggiare qualcosa di diverso e passare una piacevole serata.\nEuropa-Park Europa-Park è considerato il secondo parco divertimenti più grande d\u0026rsquo;Europa e il primo in Germania. Adoro questi parchi e mi è piaciuta molto anche l\u0026rsquo;ambientazione dedicata ai diversi paesi europei. Ci sono inoltre almeno tre o quattro montagne russe particolarmente adrenaliniche, sopra la media. Raggiungerlo da Tübingen, però, non è semplice: per una visita in giornata l\u0026rsquo;unica opzione era prendere un Flixbus molto presto al mattino.\nEscursioni Ho già raccontato ampiamente le mie escursioni nella zona: trovate tutto nell\u0026rsquo;articolo dedicato agli itinerari nei dintorni di Tübingen.\nLuoghi da vedere Vorrei segnalare anche alcuni luoghi nei dintorni più ampi di Tübingen che ho avuto il piacere di visitare. Non elencherò ogni città e paese, facilmente rintracciabili in una guida turistica, per esempio la Lonely Planet sul Baden-Württemberg: mi concentrerò su castelli e punti panoramici.\nCastelli Nell\u0026rsquo;area intorno a Tübingen ci sono tre castelli principali, più uno. Il primo, anche il più semplice da raggiungere, è l\u0026rsquo;Hohenzollernschloss Sigmaringen. Ho visitato gli interni e ne è valsa la pena.\nIl secondo è il maestoso Burg Hohenzollern, del quale ho apprezzato il cortile molto più degli interni. In questo caso consiglio un\u0026rsquo;escursione, che permette anche di ammirare il castello da prospettive diverse.\nIl terzo è lo Schloss Lichtenstein, quello che più di tutti sembra uscito da una fiaba. Il piccolo cortile interno è molto bello e offre un\u0026rsquo;ottima vista sul castello appoggiato alle rocce.\nA questi tre si aggiunge un quarto edificio, anche se si tratta più di una reggia che di un castello. A nord di Stoccarda si trova il Residenzschloss Ludwigsburg, un bellissimo palazzo con giardini splendidi. Consiglio di visitarlo in primavera o in estate, dedicando tempo alle piante e ai fiori del parco. In estate organizzano anche spettacoli di fuochi d\u0026rsquo;artificio a tema musicale che mi hanno sorpreso in positivo.\nPunti panoramici Vorrei segnalare alcuni punti rialzati da cui ammirare i paesaggi del Württemberg. Il primo, il mio preferito, è lo Schönbuchturm a Herrenberg, a circa 15–20 km da Tübingen. Dalla cima della torre si gode di una bella vista sul parco dello Schönbuch.\nIl secondo è la torre della televisione di Stoccarda, la Fernsehturm, alta più di 216 metri. Dalle terrazze panoramiche si vede la città e si spazia dai boschi e vigneti nei dintorni fino al Giura Svevo e alla Foresta Nera.\nIl terzo e ultimo è la Rottweil Test Tower, anche la più alta, con i suoi 246 metri. A 232 metri si trova la piattaforma panoramica più alta della Germania.\nCucina tedesca Comincio la parte sul cibo con uno dei miei argomenti preferiti. Quando trascorro del tempo in un posto, cerco sempre di conoscerne a fondo la cultura locale. Per alcuni amici italiani adattarsi ai piatti tedeschi è stato difficile; io, invece, li ho apprezzati quasi subito. Ho quindi preparato una breve selezione che comprende sia locali economici sia ristoranti dai prezzi più tipicamente tedeschi, forse fuori dal budget quotidiano di uno studente, ma che consiglio di provare almeno una volta. Il segreto, credo, è trovare un posto buono ed economico in cui tornare spesso e scegliere con cura le occasioni per spendere di più.\nSe con il cibo tipico è stato amore quasi immediato, non posso dire lo stesso delle bevande. Le birre del Baden-Württemberg non mi sono piaciute, mentre ho trovato qualche vino accettabile. Non apprezzare la birra in Germania può mettere a rischio la sopravvivenza.\nI miei consigli sono Gasthaus Bären, Landgasthaus Waldhäuser Hof e Restaurant Mauganeschtle Tübingen. Eviterei invece di trasformare posti come la Gasthausbrauerei Neckarmüller in un\u0026rsquo;abitudine. È una scelta turistica molto comune per la posizione, a due passi dalla Neckarbrücke, ma il servizio è pessimo, caratteristica condivisa con molti altri ristoranti tedeschi, e la qualità è piuttosto incostante. Ci andrei ogni tanto per una birra e un Flammkuchen, evitando il pasto completo.\nTapas: Gasthaus Bären La Gasthaus Bären è un ristorante e bar economico che ho trovato particolarmente adatto agli studenti. La formula ricorda le tapas: piccole porzioni a prezzi ragionevoli, con piatti da meno di 5 euro. Credo che con 12–15 euro si possa mangiare a sufficienza, a seconda di cosa si beve.\nBären è quel posto su cui tutti possono mettersi d\u0026rsquo;accordo quando nessuno ha voglia di cercare qualcosa di nuovo. Lo consiglierei anche per la prima cena fuori a Tübingen. Qui ho mangiato i migliori Käsespätzle di sempre, ma meritano anche le Zucchini Frikadellen.\nL\u0026rsquo;unico inconveniente è lo spazio limitato: d\u0026rsquo;inverno bisogna arrivare ben prima delle 19 per avere la certezza di trovare un tavolo.\nPrezzo di un pasto: 10–15 euro. Qui trovate la posizione su Google Maps e la pagina Tripadvisor.\nL\u0026rsquo;ambientazione: Landgasthaus Waldhäuser Hof Probabilmente non mi sono mai sentito così immerso nella cultura locale come quando ho provato la Landgasthaus Waldhäuser Hof. Il cibo era buono, ma l\u0026rsquo;ambientazione forse ancora di più, in mezzo alla zona delle fattorie di Waldhäuser. È proprio la posizione il punto forte del locale.\nQui i prezzi salgono rispetto alla prima proposta e si avvicinano a quelli della maggior parte dei ristoranti tedeschi della zona: simili a quelli del Neckarmüller, ma con una qualità, secondo me, decisamente superiore. Consiglierei di provare i piatti di carne.\nPrezzo di un pasto: 25–35 euro. Qui trovate la posizione su Google Maps e la pagina Tripadvisor.\nCucina sveva: Restaurant Mauganeschtle Qui i prezzi salgono ancora, almeno per gli standard di uno studente, ma il Restaurant Mauganeschtle è stato il primo posto che gli stessi svevi mi hanno consigliato per assaggiare la loro cucina. Potete provare Maultaschen, Käsespätzle e molti altri piatti, comprese buone proposte di carne. La qualità è alta e si paga: i piatti principali arrivano a 30 euro.\nPrezzo di un pasto: 30–45 euro. Qui trovate la posizione su Google Maps e la pagina Tripadvisor.\nCucina internazionale Pur essendo una città relativamente piccola, Tübingen offre una discreta scelta di cucine internazionali.\nCucina da pub: Saints \u0026amp; Scholars Saints \u0026amp; Scholars è un pub irlandese in una posizione molto comoda. Essendo così vicino alla biblioteca principale, è sempre stato una valida alternativa alla mensa universitaria.\nConsiderate le porzioni abbondanti, i prezzi sono buoni. Ho apprezzato i classici piatti da pub, come hamburger e club sandwich, sempre accompagnati da patatine. Il locale è inoltre abbastanza grande da permettere quasi sempre di trovare posto.\nPrezzo di un pasto: 20–25 euro. Qui trovate la posizione su Google Maps e la pagina Tripadvisor.\nCucina greca: Faros – an ouzo philosophy A Tübingen ci sono diversi ristoranti greci e Faros potrebbe essere il più ricercato. Il locale in Waldhäuser Straße, vicino a WHO, è piacevole e il cibo è buono.\nI prezzi sono un po\u0026rsquo; sopra la media, considerando che probabilmente vorrete ordinare più di uno o due piatti a persona. Condividerli è un buon modo per provare diverse specialità greche. Io, che adoro i formaggi greci, ho apprezzato sia la Feta Psiti sia il Tiri Saganaki.\nSe invece cercate soltanto una pita veloce, subito fuori dal ristorante trovate il Yellow Donkey, un chiosco da asporto con un menu essenziale.\nPrezzo di un pasto: 30–35 euro. Qui trovate la posizione su Google Maps e la pagina Tripadvisor.\nPizza: Da Michele A Tübingen i ristoranti italiani sono probabilmente tanti quanti quelli tedeschi. Anche le pizzerie non mancano, ma il rapporto tra qualità e prezzo raramente convince. Da Michele, a Derendingen, offre probabilmente la migliore pizza della città a un prezzo che, in Germania, ha senso.\nNon è tutto oro quello che luccica: quasi ogni volta che ci sono stato il servizio lasciava a desiderare. Spesso, però, resta una scelta preferibile alle altre pizzerie.\nPrezzo di un pasto: 20–30 euro. Qui trovate la posizione su Google Maps e la pagina Tripadvisor.\nCibo veloce Il centro di Tübingen è pieno di locali di cibo veloce. Come spesso accade nelle città tedesche, molti servono soprattutto kebab, ma non mancano falafel e cucina asiatica.\nFalafel: Kichererbse I falafel sono una buona alternativa vegetariana al kebab. Per consenso quasi unanime, Kichererbse è probabilmente il posto migliore dove mangiarli. A pranzo c\u0026rsquo;è spesso la fila fuori, segno di quanto sia apprezzato da studenti e residenti. A me e ai miei amici piaceva soprattutto il panino con falafel e melanzane…\nPrezzo di un pasto: 8–10 euro. Qui trovate la posizione su Google Maps e la pagina Tripadvisor.\nKebab: Lustnauer Imbiss In centro ci sono tantissimi posti di kebab, o kebap, soprattutto lungo Mühlstraße. Credo di averli provati tutti, ma nessuno mi è sembrato nettamente migliore degli altri.\nL\u0026rsquo;unico che mi ha sorpreso in qualche modo è stato Lustnauer Imbiss. Ho provato sia il kebab sia altre proposte, come pide e lahmacun, e l\u0026rsquo;ho preferito agli altri locali. Non è lontano dal centro sportivo universitario e dall\u0026rsquo;Österberg: potete prendere qualcosa da mangiare e fare un picnic nei dintorni.\nPrezzo di un pasto: 8–10 euro. Qui trovate la posizione su Google Maps.\nCucina cinese: China Restaurant San Bao Il San Bao è un vero ristorante con servizio al tavolo, ma io ci prendevo soprattutto cibo da asporto. Mi piacevano i noodles, una valida alternativa all\u0026rsquo;onnipresente kebab per le cene in biblioteca. Vicino c\u0026rsquo;è anche Hao\u0026rsquo;s Box: per me erano entrambi buoni e quasi intercambiabili…\nPrezzo di un pasto da asporto: 8–12 euro. Qui trovate la posizione su Google Maps e la pagina Tripadvisor.\nPub, bar e locali notturni A Tübingen è facile trovare un buon pub dove bere una birra, ma trovare buoni cocktail è più difficile. Con tutte le riserve del caso, provo a dare qualche indicazione per passare una piacevole serata in città.\nPub: Stadtpost Se chiedeste a dieci persone quale sia il loro pub preferito a Tübingen, probabilmente ricevereste almeno otto risposte diverse.\nPersonalmente ho sempre avuto un debole per lo Stadtpost, senza ragioni del tutto razionali. Mi piacevano l\u0026rsquo;atmosfera e i prezzi bassi: birra e patatine costavano poco. Sono però abbastanza sicuro che i miei amici preferirebbero altri pub.\nQui trovate la posizione su Google Maps.\nCocktail: Schmitz Katze Anche da Schmitz Katze si trova una bella atmosfera, insieme a ottimi cocktail. Era la mia scelta preferita per chiacchierare con un piccolo gruppo di amici.\nQui trovate la posizione su Google Maps.\nCinema: Kino Atelier Il Kino Atelier, come il Kino Arsenal, è un piccolo cinema monosala che proietta soprattutto film indipendenti tedeschi e internazionali. Qui, per esempio, ho visto Everything Everywhere All at Once alla sua uscita, molto prima che vincesse l\u0026rsquo;Oscar al miglior film. La sala piccola crea un\u0026rsquo;atmosfera intima e accogliente, rendendo la serata un\u0026rsquo;esperienza molto piacevole.\nDiscoteche e feste Non nascondo quello che penso: l\u0026rsquo;offerta di locali notturni a Tübingen ha ampi margini di miglioramento.\nLo Schwarzes Schaf era divertente, ma soltanto con il gruppo di amici giusto. In primavera e in estate le feste organizzate dagli studenti al Clubhaus sono l\u0026rsquo;alternativa migliore alle discoteche, mentre il Kuckuck è la scelta più comoda per chi vive a WHO. Negli altri pochi locali in cui sono stato non ho apprezzato abbastanza l\u0026rsquo;atmosfera da consigliarli.\n","permalink":"https://lucabontempi.com/it/blog/tuebingen_experiential_guide/","summary":"\u003cp\u003e\u003cem\u003eNota: pubblico questo articolo nell\u0026rsquo;agosto 2023, ma ho cominciato a scriverlo nel settembre 2022, quando vivevo ancora a Tübingen. Nel frattempo molte cose potrebbero essere cambiate, soprattutto i prezzi, influenzati dall\u0026rsquo;inflazione dell\u0026rsquo;ultimo anno. Ho comunque deciso di completarlo per chiudere il trittico di articoli che mi ero proposto di scrivere.\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eDopo un anno da studente a Tübingen, provo a condividere quello che ho imparato sulla città e sui suoi dintorni.\u003c/p\u003e","title":"Una guida a Tübingen per studenti internazionali, vissuta in prima persona"},{"content":"Premessa Questo articolo contiene il testo integrale della mia tesi magistrale, scritta presso la Eberhard Karls University di Tübingen e l\u0026rsquo;Università di Pavia e presentata per la prima volta nel febbraio 2023.\nNel settembre 2026 ho riaperto questo lavoro e analizzato nuovamente i dati originali. La revisione ha individuato tre errori nell\u0026rsquo;analisi, nessuno nei dati stessi, ed è raccontata in Ho chiesto a un\u0026rsquo;AI di rivedere la mia tesi. Il testo della tesi qui sotto è quello consegnato allora, lasciato invariato.\nIn questa versione è tradotta in italiano soltanto la Premessa. Il titolo e tutto il testo da «Abstract» in poi restano nelle lingue originali, prevalentemente in inglese, senza traduzioni o revisioni.\nLa domanda di ricerca è quale caratteristica tra reputazione, popolarità e brand dello sviluppatore, riflessa negli elementi della pagina di presentazione sugli app store, sia più efficace nel prevedere la scelta dell\u0026rsquo;utente di scaricare un\u0026rsquo;app.\nIl Capitolo 1 offre un ampio inquadramento dell\u0026rsquo;ecosistema mobile, dei suoi partecipanti e delle sue dinamiche. Il Capitolo 2 ripercorre i risultati della letteratura sugli effetti dei tre fattori. Il Capitolo 3 e il Capitolo 4 presentano infine l\u0026rsquo;analisi empirica, i risultati e la loro discussione.\nLa versione completa in formato documento è disponibile qui, mentre qui trovate una presentazione riassuntiva.\nAbstract After providing an extensive background on the mobile ecosystem, its participants and its dynamics, this research aims to establish which characteristic out of reputation, popularity and developer\u0026rsquo;s brand, reflected in some elements in the presentation page on app stores, is the most effective in predicting user\u0026rsquo;s choice to download an app. Experimental results show how developer\u0026rsquo;s brand is generally the most decisive element, but, when comparison between apps is introduced in the model, it loses its effectiveness and reputation becomes more relevant. Popularity and peer influence, in the context of low network effects, appear to be unexpectedly ineffective in driving users\u0026rsquo; decisions across all models. Moreover, users\u0026rsquo; involvement in the download process reinforces the effect of reputation, while, when customers are more involved in the specific app category and perceive a higher risk, apps from established brands are more likely to benefit. Finally, it was not possible to find clear evidence of any two-way interaction between reputation, popularity and brand.\n(in Italian) Dopo aver fornito una descrizione dell\u0026rsquo;ecosistema mobile, dei suoi partecipanti e delle sue dinamiche, questa ricerca si pone l\u0026rsquo;obiettivo di stabilire quale caratteristica tra la reputazione, la popolarità e il brand dello sviluppatore, riflesse in alcuni elementi nella pagina di presentazione sugli app store, sia la più efficace nel predirre la scelta dell\u0026rsquo;utente di scaricare un\u0026rsquo;applicazione mobile rispetto ad un\u0026rsquo;altra. I risultati dell\u0026rsquo;esperimento mostrano come il brand dello sviluppatore è generalmente il fattore più influente. Tuttavia, quando la possibilità di comparare le applicazioni è introdotta nel modello, il brand perde parte della sua efficacia e la reputazione diventa più rilevante. La popolarità dell\u0026rsquo;applicazione tra gli altri utenti, in un contesto di scarsi effetti di rete, sembra essere sorprendentemente inefficace nell\u0026rsquo;influenzare le scelte degli utenti in tutti i modelli. In aggiunta, il coinvolgimento dell\u0026rsquo;utente nel processo di download delle applicazioni rinforza l\u0026rsquo;effetto della reputazione, mentre, quando l\u0026rsquo;utente è più coinvolto nella specifica categoria di applicazioni e quindi percepisce un rischio maggiore, le applicazioni sviluppate da brand consolidati hanno più probabilità di essere scelte. Infine, non è stato possibile individuare alcuna interazione bidirezionale significante tra reputazione, popolarità e brand dello sviluppatore.\nChapter 1. Introduction The mobile industry has incredibly changed throughout the years it existed. While the developments in hardware and software components have been in the public eye for many years now, the most impressive change was arguably the shift in power and influence among market participants.\nIn fact, the industry\u0026rsquo;s balance of power, which back in the 1990s and early 2000s used to favour mobile device manufacturers and, in particular, network operators, now sees a hardly disputed winner, which are mobile platform providers, like Google or Apple (Basole et. Karla, 2012, p. 35).\nThe most important year in this transition was 2008 and Apple was probably the main actor. As commonly known, in 2007, Apple launched the first version of the iPhone and it could be argued that the real innovation didn\u0026rsquo;t lay in the new hardware features of this phone: Apple\u0026rsquo;s iPhone wasn\u0026rsquo;t the first one to introduce a capacitive touchscreen, for example, and LG did that some time in advance in 2006 (Ihlwan, 2007). The real breakthrough that initiated the \u0026ldquo;app economy\u0026rdquo; was the introduction, in 2008, of Apple\u0026rsquo;s App Store, a mobile application marketplace with originally a few more than 500 apps and which now accounts for more than 1.5 million applications (Basole and Karla, 2012, p. 32; Ceci, 2022b; Friedman, 2013). The first app stores were launched at the end of the 1990s, but it wasn\u0026rsquo;t until almost a decade later, with the Apple Store, that they started to gain traction (Kouris and Piller, 2014, p. 21). Since then, the increasing availability of mobile apps, supported by enhancements in hardware features, drove the growth of the mobile industry. In fact, the number of smartphone users worldwide almost duplicated only considering the last five-year period, reaching 6.2 billion users in 2021 and being expected to grow up to 7.6 billion in 2027 (Ericsson, 2022). Meanwhile, apps became an incredibly profitable business by themselves and those downloaded from the Google Play Store, the largest app store by the number of available apps at the moment (3.5 million apps in 2022), generated in 2021 more than $ 47 billion in gross revenue, three times more than what was generated back in 2016 (Ceci, 2022a, 2022b).\nClearly, being the owner of a successful app marketplace is a position as desirable as difficult to reach for companies operating in the mobile ecosystem and many appear to fail while attempting that. The complex dynamics lying behind the interaction among market participants (e.g. app developers and customers) are shaped by the mechanics of information economics, which may raise entry barriers and uncertainty. While past literature has already provided useful insights on two-sided platforms,1 a category to which app marketplaces belong, some elements of customer behaviour, when dealing with digital products, need to be explored more deeply. However, before investigating users\u0026rsquo; preferences on mobile app stores and focusing on empirical evidence, an overview of the mobile ecosystem is necessary.\n1.1 Mobile Ecosystem The mobile phone industry is now more than forty years old. In 1978, in fact, a Motorola engineer, Martin Cooper, who is today considered the inventor of the cell phone, made the first mobile phone call using a Motorola DynaTac in New York (Zheng and Ni, 2010, p. 32). Six years later, in 1984, a more recent version of the same Motorola phone was commercialized for the first time. Those phones were also defined as \u0026ldquo;brick phones\u0026rdquo;, due to their shape and size, and, in the following ten or twenty years, the innovation efforts of this industry focused on miniaturization, better power efficiency and battery life, better displays and signal. In the meantime, generations of cellular network technologies were succeeding (from 1G to today\u0026rsquo;s 5G), improving voice communication quality, as well as introducing revolutionary features such as text messaging or Internet access. However, the real disruption in this industry came in 2008, with the introduction of the first smartphones and of the app concept, which overturned power relations between market participants.\nBasole and Karla (2012, p. 28) identified four groups of market participants, mobile platform providers, mobile device manufacturers, mobile network operators, and mobile application providers, to which a fifth group, mobile users, can be added to effectively understand the industry\u0026rsquo;s internal dynamics.\nMobile Platform Providers. Mobile platform providers (MPPs) provide operating systems and middleware solutions for smartphones and they are currently driving the mobile industry. In addition, MPPs often coincide with the owners of major mobile app stores. This double role was arguably assumed for the first time by Apple in 2008, when, while launching the iPhone, it leveraged its experience in content distribution, thanks to the previous existence of iTunes, to introduce the App Store, in order to strengthen its ecosystem and drive demand for iOS devices (Basole and Karla, 2012, p. 36; Cortimiglia et al., 2011, p. 3), before app stores became a profitable business by their own. While a couple of years after its introduction, iOS was installed on 37 percent of all smartphones, a market share similar to the one of Symbian OS (Nokia\u0026rsquo;s own operative system), and Android had just entered the market, as shown in Fig. 1.1, now the latter dominates other mobile operative systems (with about 71 percent of the global market), just followed by iOS and other operative systems with very small shares (Laricchia, 2022).\nFigure 1.1: Mobile OS market shares in 2022 and 2009 (Laricchia, 2022). Interestingly, these two systems were both able to succeed with two different approaches: iOS is considered a relatively closed ecosystem (\u0026ldquo;walled garden\u0026rdquo;), while Android is a more open one (\u0026ldquo;open garden\u0026rdquo;). The unique role of Apple as both MPP and device manufacturer enables it to apply some restrictions, as, for example, the fact that apps can only be installed from the App Store, as sole source (Basole and Karla, 2012, p.36; Cortimiglia et al., 2011, p. 3), and this approach allows more strict testing and verification, that, in combination with the existence of only a few different hardware with that operative system (OS), may lead to a better user experience, with consequently improved customer loyalty and a premium brand image. On the other hand, Android is an open-source OS and the ecosystem is more open: this may lead to higher levels of competition both on software and hardware, with lower prices as result (Müller et al., 2011, p. 74). In terms of app distribution, both Apple and Google act as gatekeepers: they decide what gets published on their stores, also covering important roles in terms of quality assurance and pre-screening of dangerous apps (Müller et al., 2011, p. 73), but, while in Google case, these limits can be circumvented thanks to the presence of other app sources, this is (legally, at least) impossible on Apple\u0026rsquo;s devices.\nOverall, MPPs took over many parts of the value chain, from OS development to payment systems, relegating other participants to more subordinate roles.\nMobile Device Manufacturers. Mobile device manufacturers\u0026rsquo; (MDMs) activity mainly consists in producing and distributing the hardware part of a smartphone. In the past, MDMs generally owned proprietary operative systems installed on their products, only a few native apps were offered with the phone and the possibility of installing third-party apps was generally limited. Their role was strictly connected to the one of network operators, on which they often depended for distributing devices, usually sold in combination with mobile subscriptions (Basole and Karla, 2012, p. 29).\nToday, the only MDMs developing a popular OS are the most important MPPs, Apple and Google, while all the other producers (Samsung, Xiaomi, Oppo, Vivo, etc.) redirected themselves towards Android as the chosen OS for their devices in most cases.\nMobile Network Operators. Mobile network operators (MNOs) have been dominant in the mobile service industry for many years. For decades, they were the main sources of innovation and added value, they supplied SIM cards and they could also dictate some pre-settings, like the browser\u0026rsquo;s homepage or types of applications installed (Basole and Karla, 2012, p. 29). Later, with the smartphone introduction and the app economy emerging, MNOs lost their traditionally central role, their margins shrunk and the risk of becoming dumb pipes is real.\nMobile Application Providers. Mobile application providers (MAPs) are those providing content in the current mobile ecosystem. They develop both web-based and native applications for specific technology platforms, which cover many different categories and use cases. Developers traditionally lacked market accessibility and app stores greatly reduced barriers and time-to-market for them, but, on the other side, they now largely depend on MPPs\u0026rsquo; enabling capabilities and developing tools (Basole and Karla, 2012, p. 30). Anyway, MAPs can now reach a large number of potential customers, at the price of leaving (on average) 30 percent of their app download revenue to gatekeepers (Cortimiglia et al., 2011, p. 5), whose role is particularly restricting in walled gardens, due to the lack of alternative channels.\nMobile Users. On the other side of a mobile app store, there are smartphone users. They search, download and pay for apps and they also later rate them. There is a tendency among users to view the mobile device and available apps as a package: they take into consideration the ecosystem as a whole and then decide which hardware and software combination to buy. In this sense, mobile devices can be considered complementary goods for apps and their price a one-time membership fee to join the ecosystem (Kouris and Piller, 2014, p. 36).\nFor customers, downloading an app and the perception of the risk involved depend more on the MPPs than MAPs: users, in fact, place their trust in platform owners as their main source of information. In addition, they use apps to perform highly fragmented everyday tasks and perceive apps not as complex as traditional desktop software. This makes their buying decision limited, habitual and impulsive, corresponding to generally low levels of involvement (Buck et al., 2014, p. 5). During the download process, users often don\u0026rsquo;t search for information actively and reputation systems (e.g. rating score), as well as peer influence (e.g. sale rankings, number of downloads), drive their decision-making (Buck et al., 2014, p. 5; Liu et al., 2014, p. 330).\n1.2 Mobile Apps as Digital Products The term \u0026ldquo;mobile application\u0026rdquo;, \u0026ldquo;mobile app\u0026rdquo; or just \u0026ldquo;app\u0026rdquo; refers to application software, which is \u0026ldquo;a type of software that allows the user to perform a specific task, that can be installed and run on a range of portable devices such as smartphones and tablets\u0026rdquo; (Liu et al., 2014, p. 327). An app usually distinguishes itself by general software because it is optimized for touch-screen-based mobile devices, it is offered free of charge or for sale and the most of the times it is available for download through a centralized online marketplace where users can also rate and review the app and access some ranking lists (Liu et al., 2014, p. 327). Apps already have a longer history than one could expect: for example, Zheng and Ni (2010, pp. 51-56) were able to individuate two generations of mobile apps already back in 2010. The first generation involved very simple apps, such as ringtones, instant messaging and personal information management apps, among which the first one was the most popular for many years. Then, the second generation included more complex apps like mobile commerce, gaming, music streaming and mobile social networking and some of these are still popular today. Later, a third generation emerged, with features like augmented reality, ubiquitous social networks and merging and seamless switching between different functions (Kouris and Piller, 2014, p. 25).\nMobile apps, exactly like all types of software, are digital goods and as such they share all the typical characteristics of this category of products. Compared to standard economic models, in a digital information setting, costs fall substantially and approach zero. In fact, Goldfarb and Tucker (2019, p. 3) were able to identify five types of cost reduction: lower search costs, lower replication costs, lower transportation costs, lower tracking costs and lower verification costs. Even if these cost reductions are intrinsic to all digital (i.e. intangible) goods and services, they are enhanced by the use of mobile app stores as the main distribution channel.\nIn fact, mobile app stores are an example of two-sided platforms. These platforms are intermediaries that enable exchanges between other players and, by doing that, they efficiently organize supply and demand allowing buyers to access the sellers (and vice versa), facilitating matchmaking and lowering transaction costs. Platforms aggregate information and provide access to that data to market participants and also provide ratings and feedbacks that help to build credibility and trust among participants, making some previously unknown seller characteristics observable, rewarding the better-quality users and creating incentives for lower-quality ones to leave the market. In exchange for easier transactions, these businesses charge a share of their value for the service (Lambrecht et al., 2014, p. 10). In addition, by aggregating market participants, intermediary platforms generate indirect externalities (or indirect network effects): in fact, they usually consist of two-sided (or multi-sided) networks with two (or multiple) types of members, where each type is looking for a network with the largest number of users of the other type. In these cases, network effects become so strong that customers tend to pick a network, instead of choosing a service or a product, since the utility on one side of the market highly depends on the number and quality of participants on the other side (Kouris and Piller, 2014, pp. 39-40; Piccoli and Pigni, 2019, p. 239). Indeed, mobile app stores are characterized by a positive feedback loop: the greater the number of users of the store (reflected in the number of downloaded apps or in the installed subscriber base), the more attractive that marketplace becomes for developers, resulting in an increase of the supply of content and applications and, as a consequence, in higher consumer benefits and increased attractiveness for additional users (Basole and Karla, 2012, p. 35).\nLower Search Costs. It is a lot easier to find and compare information about potential transactions online than it is offline. As a consequence of the decrease in search costs, prices have fallen and product variety increased (Goldfarb and Tucker, 2019, pp. 6-7). While prices online appear to be constantly lower than offline, also for tangible products (e.g. Brynjolfsson and Smith, 2000, p. 579), in purely digital contexts, such as app marketplaces, prices actually reached zero in many cases. Minimal search costs, together with other reasons that will be later discussed, had an important role in spreading and making successful the freemium strategy and increasing the offer of free apps.\nHowever, even if prices are lower, an important level of price dispersion persists online. The ease of changing prices and running price promotions on mobile apps, which also have no marginal costs of production, makes this a common characteristic of mobile app stores too. \u0026ldquo;Free App of the Day\u0026rdquo; promotions are an example of an initiative that could rise price dispersion on app marketplaces. This kind of initiative involves spotlighting in the homepage and zero-pricing an app for 24 hours and it had been used in the past, for some periods, by the Google Play Store, the Apple App Store and the Amazon Appstore, the three most popular app stores today (Chen et al., 2022, pp. 1-2). The objective of this practice is combining digital free sampling, which proved to be effective in reaching new customers, with sponsorship practices, like featuring the app in a noticeable position, consequently increasing visibility and enhancing customer awareness of the app. According to Chen et al. (2022, pp. 6-7), such a promotion may lead to adverse effects in the post-promotion phase: firstly, the removal of zero-price hurts sales, in particular, due to customers\u0026rsquo; expectations of the app being free and their perception of a loss by observing the restored (and increased) price; secondly, promotion-induced reviews, generated by the increased visibility of the app, are less effective than organic ones.\nLow search costs also mean that it is way easier to find rare and niche products online (Goldfarb and Tucker, 2019, p. 8), a phenomenon defined by Anderson (2006, p. 22) as \u0026ldquo;the long tail\u0026rdquo;. The effects of the long tail are particularly visible on app stores: apps generally require just relatively little storage space, making welcoming as many apps as possible quite inexpensive for marketplace owners, with positive benefits overall. Regardless of the number of sales, unsuccessful apps provide additional choices and options for customers, reinforcing the two-sided positive feedback loop (Basole and Karla, 2012, p. 35). In addition, the growing user-friendliness of developing tools allowed many people besides professional developers to publish their apps, further increasing the range of available options. A broader choice of apps, from a negative perspective, leads to higher choice complexity, which could cause delays in the users\u0026rsquo; app download process. This negative effect of the long tail is usually mitigated by recommendation systems, app rankings and improved research, made possible by the available data on customers\u0026rsquo; previous behaviour.\nFinally, an online marketplace can sometimes generate the superstar effect, with homogeneous customers all agreeing on which product is the best and buying it (Goldfarb and Tucker, 2019, p.8). In the case of apps, this phenomenon is extremely strengthened when software benefits from direct network effects and its utility largely depends on the size of the installed base (e.g. instant messaging apps). In these cases, the probability of being in presence of a tipping (or \u0026ldquo;winner-takes-all\u0026rdquo;) market is high.\nLower Replication Costs. In addition to being subject to marginal costs which are close to zero, software is a non-rival, meaning that it can be consumed by one person without reducing the amount of quality available to others, and, in absence of a legal or technological limit, non-excludable good, so it can be reproduced and used by anyone at approximately no cost. These characteristics often generate opportunities, like the possibility of implementing bundling strategies at a very low marginal cost, very common in the software industry, but also some threats.\nLow replication costs and non-excludability are piracy\u0026rsquo;s main reasons for existence. Piracy has been a long-discussed issue in the software industry and illegally distributed copies actually offered in some ways opportunities for free sampling and strengthening network effects in the past, leading researchers and companies to think of piracy as an advantage under certain conditions. Raising the barriers of protection causes, in fact, some users to buy the product and others to live without it or look for alternatives, consequently reducing the installed base and its value also for those users that paid for the program (Reavis Conner and Rumelt, 1991, pp. 136-137). In addition, illegal copies contribute to raising switching costs for pirates, with a higher probability that they would later buy the software. Today, the incentive not to get in the way of illegal distribution has been partially revised, also considering the effectiveness of freemium software in covering the same functions previously fulfilled by piracy, while remaining in control of intellectual property and generating some earnings from advertising, for example (Halmenschlager and Waelbroeck, 2014, pp. 30-31; Nan et al., 2018, p. 283).\nIn some way, compatibility can be seen as a limit to software duplication: in fact, software needs to be compatible with other elements of the ecosystem (the operative system, in particular) and, therefore, in some cases, it cannot be run without some efforts by developers to make this possible. Sometimes, making software compatible with an additional OS can be almost as expensive as it was creating it at first and it needs some incentives from the platform owner side, such as, for example, a large installed base, a high-quality development toolkit or low monetary and technical entry barriers: developers that choose to target two or more ecosystems try to balance between a greater potential market and the cost of porting the product for different platforms (Hyrynsalmi et al., 2012, p. 60; Koch and Kerschbaum, 2014, pp. 1432-1434; Rochet and Tirole, 2003, pp. 1015-1017). Examining the strategy of the most popular apps, multi-homing might be assumed sometimes to be a viable way to increase the circulation of an app. However, the effect could be bi-directional and popularity in a single ecosystem could actually trigger the porting of the app in a second one. In fact, on mobile app stores, the large majority of apps (about 87 to 95 percent) are developed for a single technology platform and apparently, a multi-homing strategy isn\u0026rsquo;t very effective in making an app more popular than others (Ali et al., 2017, p. 5; Hyrynsalmi et al., 2012, p. 71). While at least professional developers would be expected to develop cross-platform apps, one out of five top-rated apps are not present on both Android and iOS, for reasons which reportedly are lack of resources, platform restrictions, low expected revenue in the other platform, user fragmentation within a platform and competition being already present, and, even if developed for both platforms, in any case, they are usually not consistently released in terms of new versions and updates (Ali et al., 2017, pp. 6-7). Obviously, if multi-homing can be hard for companies with skilled developers, it gets even rarer in an ecosystem that contains a good number of app publishers doing it just for fun or lacking a clear monetisation plan (Hyrynsalmi et al., 2012, p. 68). In addition to that, platforms in some industries tend not to promote multi-homing and even to obstacle it when it reduces possibilities of differentiation. While setting conditions that raise developer attractiveness usually increases the share of cross-platform software available, it also reduces user attractiveness when compared to other platforms (Hyrynsalmi et al., 2012, p. 70).\nLower Transportation Costs. The cost of transporting information goods over the Internet is quite near zero. Software has been distributed online and in physical formats for many years now and, while, back in the days, offline distribution was certainly more widespread, the Internet started to pick up as an important distribution channel soon.\nAnyway, software distribution over the Internet used to present some drawbacks, in particular in the past. In fact, anyone could post software to the Internet, with no accountability and making malicious software downloads quite common, creating the need for a solution for trusted software distribution (Rubin, 1995, p.7). From a certain perspective, app stores answered to that problem, creating a trusted and secure channel even for less technologically advanced customers. In addition, they also provided a more efficient and faster system for automated software updates, making them almost unnoticeable to unaware customers, reducing security threats and improving user experience.\nToday, when getting information on which app to download, users place their trust in app store owners, creating a way closer relationship with intermediaries than third-party software developers (Buck et al., 2014, p. 13).\nLower Tracking Costs. In a digital domain, user activity can be easily recorded and stored (Goldfarb and Tucker, 2019, p. 19), enabling relevant opportunities to make software selling more efficient. A major example of this is personalized recommendations, present on almost every online marketplace as well as on mobile app stores.\nThe availability of information on user behaviour also opened up new price discrimination opportunities: in particular, freemium, a specific kind of versioning, became predominant. Freemium, a combination of the words \u0026ldquo;free\u0026rdquo; and \u0026ldquo;premium\u0026rdquo;, is a business model by which a service or a product is offered free of charge, but a premium version is charged for advanced features, functionality, related products or services (Liu et al., 2014, pp. 327-328). Therefore, the free basic version of the product contains limited features, less privacy, more advertising or limited time usage (Boudreau et al., 2022, p. 1378). Freemium insert itself in a context of sampling and versioning strategies, that proved, for many years, to be efficient in creating a positive brand attitude and awareness when uncertainty about the product is high (Arora et al., 2017, p. 63). In addition to that, mobile applications are largely experience goods and it can be difficult for customers to quantify the true value of the product before using it, while a free version of the software allows customers to get to know the product, expecting that a considerable share will purchase the premium version (Boudreau et al., 2022, p. 1379). The freemium strategy in the app economy is also used to exploit advantages ascribed to positive network effects, product complementarity and an accelerated diffusion process. For companies leading the market, a free version of an app may provide a method for exploiting their advantage and expanding their lead over their competitors; for new entrants, a freemium strategy may provide a way to eat a part of the leader\u0026rsquo;s advantage (Boudreau et al., 2022, pp. 1375-1376). However, unlike traditional software, apps are generally characterized by minimal learning and switching costs and a low degree of interdependence among apps from the same developer (Liu et al., 2014, p. 330), raising fragmentation and competition. Implementing this strategy remains anyway quite complex and a trade-off emerges: on one side, consumers will judge the app on the basis of the quality of the free version (Liu et al., 2014, p. 330); on the other side, if the free app is sufficient for users, it could cannibalize the paid version.\nNevertheless, collecting users\u0026rsquo; information isn\u0026rsquo;t always positive and privacy regulation puts a cost on tracking information flows (Goldfarb and Tucker, 2019, p. 23): on one hand, there are legal compliance costs and, on the other, there is a huge reputation risk. Many users report being very concerned about their privacy, but they still frequently download free apps that monetize their personal data, bringing to light a piracy paradox. Clearly, consumers on the Internet articulate their need for privacy, but they act in an opposite direction (Buck et al., 2014, p. 4).\nLower Verification Costs. As already discussed, mobile app stores, as two-sided platforms, contribute to increasing trust among participants through the rating system. In an offline setting, the company\u0026rsquo;s reputation rests on having an uninterrupted series of transactions with a positive outcome; online and on app stores in particular, the purchase is typically a one-time event, meaning that the reputation system becomes fundamental in providing insights on the app quality and making it public for users, in a way to discourage opportunistic behaviours of app developers, even if the transaction is often a unique event (Müller et al., 2011, p. 67).\nAnyway, mobile apps are continuously evolving at a rapid pace with updates being released and replacing older versions quite frequently. As a result of app updates, user satisfaction may change frequently and traditional rating systems appear sometimes unable to capture these dynamics since they are usually resilient to average rating fluctuations when the app has already a substantial number of previous reviews. This system disadvantages new emerging developers, considering that initial low quality of the app and consequently low ratings may cause the impossibility to come back to a good average score later and it could encourage developers to release a new version of a badly reviewed app as a completely new one, trying to reset its reputation (Ruiz et al., 2015, p. 11).\nMoreover, another problem of review systems on online marketplaces is \u0026ldquo;reputation inflation\u0026rdquo;. In fact, average public feedback scores on these platforms have greatly increased and it looks like customers are reluctant to give a low rating (in particular when the feedback is public) because they fear retaliation or they don\u0026rsquo;t want to harm the seller. In addition, the perceived cost of leaving a negative review increases higher the average rating, pressuring consumers to leave above-average scores (Filippas et al., 2018, p. 30; Horton and Golden, 2015, p. 2). In the long term, this phenomenon makes the reputation system less informative and also creates nonlinear rating expectations for buyers, raising the threshold above which buyers consider the average score a good one.\nAll in all, the advantages of mobile app stores as platforms and distribution channels are overwhelming, even if some aspects still need to be perfectly tuned. Their introduction contributed to making transaction costs, already decreased by digital economics, even lower and helped in democratizing software distribution.\nChapter 2. Literature Review Many authors previously covered customer behaviour on online marketplaces, comparing it with the offline domain (e.g. Brynjolfsson and Smith, 2000; Brynjolfsson et al., 2003). Recently, literature also focused on user behaviour on mobile app stores: for example, Burghers et al. (2016, p. 330) argued how apps are generally low-priced or free and this causes low effort and regret when they are unsatisfied and decide to uninstall them. This means that the perceived risk in the buying decision process is low and leads, in many cases, to low involvement. In addition, some consumers appear to be unconscious and they do not seem to search for relevant information that could reduce the information asymmetries in the app market (Buck et al., 2014, p. 12). There is evidence that decision-making when dealing with app downloading often relies on heuristics (e.g. \u0026ldquo;take the first\u0026rdquo;), which allows users to quickly navigate across offerings, but may also cause customers to overlook some important information (Dogruel et al., 2015, p. 139).\nIn this context, it can be assumed that some elements in the presentation page of an app on a store may be effective in convincing a user to download it. In particular, three main references and their influence on user\u0026rsquo;s intention to download will be investigated below: app reputation, its popularity (and the consequent peer influence) and the developer\u0026rsquo;s brand.\nProduct reputation impact on purchase decisions online has already been covered by many authors. Chevalier and Mayzlin (2006, pp. 353-354), for example, proved that an improvement in online reviews is positively correlated with an increase in sales and the impact of ratings isn\u0026rsquo;t linear, with one-star reviews being more influential than five-star reviews. Moe and Trusov (2011, pp. 455-456) testified how ratings can improve sales, but customer rating behaviour is significantly socially influenced by previously posted ratings. Previous literature investigated eWOM (electronic word-of-mouth) valence, usually reflected by the average rating score, as well as eWOM volume, which is usually measured by the number of reviews. Out of the papers that covered eWOM valence on mobile app stores, seven of them were able to find a positive impact on app success (Böhm and Schreiber, 2014, p. 13; Burgers et al., 2016, p. 340; Chen et al., 2022, pp. 6-7; Finkelstein et al., 2017, p. 38; Gokgoz et al., 2021, pp. 434-435; Jung et al., 2012, p. 936; Lee and Raghu, 2014, p. 161), one found a negative correlation (Liang et al., 2015, pp. 250-251) and four found no significant impact (Roma and Ragaglia, 2016, pp. 184-185; Sällberg et al., 2022, pp. 7-10; Timmerman and Shepherd, 2016, pp. 14-15; Wang et al., 2015, p. 5). On the other side, four authors discovered a positive relation between app success and the number of reviews (Burgers et al., 2016, p. 340; Lee and Raghu, 2014, pp. 149-151; Sällberg et al., 2022, p. 13; Wang et al., 2015, pp. 7-10), while just one found out it was negative (Liang et al., 2015, pp. 250-251). The majority, in both cases, used the number of downloads in a selected time period as a performance metric, while some others preferred app survival in the top-apps lists. Overall, reputation in the mobile ecosystem has already been studied enough to take some preliminary conclusions and the majority of past literature contributions agree that reputation signals quality and it has a generally positive effect on sales.\nThe popularity of an app is reflected in multiple elements of an app store: firstly, the number of downloads is usually shown in the app page, then the top-list presence and app sales rank also signal it is popular among users (Sällberg et al., 2022, p. 3). Popularity translates into more interest among potential customers because it reflects the previous interest of other consumers and this peer influence has been shown to be one of the primary drivers of purchase decisions in online shopping (Liu et al., 2014, p. 330). Salganik et al. (2006, p. 855) have already shown how popularity information and social influence affect other users\u0026rsquo; choices on music platforms, making popular songs even more popular. Also in the software industry, there is evidence that app popularity, reflected, for example, in the presence in top-apps ranking, has a positive impact on app success (e.g. Finkelstein et al., 2017, p. 2679). In addition, his effect may be even stronger in the case of software that generates strong network effects, since its utility increases with popularity and users are expected to consider that. Finally, the consequences of reputation aren\u0026rsquo;t independent of social influence: interestingly, less popular products in the market are more significantly influenced by online reviews (Liu et al., 2014, p. 331).\nThe third and last considered anchor is the brand of the company developing the app. Even if online marketplaces lower prices and reduce information asymmetries, brands remain sources of competitive advantage (e.g. Brynjolfsson and Smith, 2000, p. 580). In the mobile ecosystem, Roma and Ragaglia (2016, p. 184), for example, have already shown how a company\u0026rsquo;s fame positively moderates app success. On mobile app stores, the brand may be reflected in multiple elements of the presentation page: apart from the developer or publisher name, which is usually present, the company\u0026rsquo;s name is sometimes included also in the app\u0026rsquo;s name and in the icon. While, if taken singularly, each one of these elements has a relative importance on purchase decision which is low, at least if compared to average review rating and number of reviews (Böhm and Schreiber, 2014, p. 12), they could combine and consistently influence users. Furthermore, app icons are graphic elements and they are easy to understand for users (Böhm and Schreiber, 2014, p. 8) and Pol (2015, p. 51) showed how consumers prefer apps with a brand logo, well-known brands also gain an additional advantage and the brand attitude has a positive effect on the perceived app quality.\nChapter 3. Empirical Analysis The experimental approach used in this research aims to understand how users\u0026rsquo; intention to download (ITD) varies when three main variables (app reputation, popularity and developer fame) and their values change, taking also into consideration their interaction effects, the effect of comparison and of involvement in mobile apps, in the app purchase process and in the app category selected for the purpose of the experiment. Differently from previous research on the topic, this study focuses on direct user participation as its central characteristic: in fact, most of previous literature investigating the effect of the same variables used the number of downloads or an app survival in app store rankings as performance measures. However, while scraping mobile app stores as a method for data collection allows for larger samples, the algorithms behind app rankings on these stores are not public and their functioning is unknown, meaning they could not directly reflect users\u0026rsquo; preferences. For example, Google\u0026rsquo;s Play Store top apps rank is probably based on a complicated algorithm, that takes into consideration multiple factors, such as app title, description, localization, retention rate, number of reviews, app updating, number of downloads and review ratings, meaning that a high rank does not always indicate a high user satisfaction (Liu et al., 2014, p. 348).\n3.1 Method and Measurement This research relies on an online questionnaire as its data source and the experimental design consists of parallel binary manipulations of the three main variables. For each one of these variables, one out of two manipulated screenshot images of the presentation page of some apps (presented in Appendix A) was shown to participants and their ITD was measured with a 7-point scale. Since intention can be defined as \u0026ldquo;the person’s motivation in the sense of his/her conscious plan to exert effort to carry out a behaviour\u0026rdquo; (Eagly and Chaiken, 1993, p. 168) or as \u0026ldquo;an individual’s conscious plan to make an effort to purchase a brand\u0026rdquo; (Spears and Singh, 2004, p. 56), intention to download is the person’s motivation in the consciously planning to make an effort to download an app (Pol, 2015, p. 18). The screenshots images were taken from Google\u0026rsquo;s Play Store, considering it is the most common app store on the most widespread mobile OS, which is Android, as shown in Fig. 1.1, and supposing that the majority of respondents would have been familiar with it.\nBefore being exposed to the manipulated variables, survey respondents were introduced to the questionnaire and asked some questions aimed to measure their involvement in apps as products (the online survey is integrally available in Appendix B). Firstly, their involvement in mobile apps in general and in the app download process were measured using Laurent and Kapferer’s scale, as presented in Mittal (1995, p. 670). Then, the involvement in the specific app category chosen for the experimental purpose was quantified through a reduced version, presented in Celuch and Taylor (1999, p. 120), of Zaichkowsky (1994, p. 70)\u0026rsquo;s involvement scale, which was already used for mobile apps in Kang et al. (2015, p. 213).\nThe app typology used in the experiment was mobile scanner apps, which allow to digitize physical documents through the smartphone camera and usually to export them in a digital format (e.g. PDF) in just a few seconds, and it was chosen for different reasons. Firstly, this category is quite fragmented, with many apps with small market shares, and the majority of the available options share similar basic features, with just a few exceptions (e.g. better OCR): in order to empirically prove this hypothesis, a preliminary analysis of the scanner apps segment on the Play Store (reported in Appendix C) was conducted and a sample of 24 apps was taken in consideration. Clearly, these apps are relatively easy to develop and there are low entry barriers for companies with the desire to enter this segment. Secondly, mobile scanner apps are subject to really low network effects, allowing to rule out the influence of network externalities and to measure the popularity variable effect by itself. Another important element was the presence in the category of an app developed by a company, Adobe, which is known worldwide for its software. Then, mobile scanner apps are non-hedonic apps and in this case review ratings become more important than ranks (Liu et al., 2014, p. 345). Furthermore, the fact that these apps act as still, at the moment, imperfect substitutes of traditional scanner machines may lead to think they are used by consumers with low involvement in the digitizing process, in line with the quick and impulsive buying setting described by Buck et al. (2014, p. 5), even if, since they belong to the productivity apps category, there is also evidence in the opposite direction (e.g. Sällberg et al., 2022, p. 10). Finally, the vast majority of these apps are offered with the freemium model, meaning that essential features are for free and, consequently, price is not a variable.\nThe order in which the three survey slides in which the app presentation pages of a mobile scanner app were included to measure users\u0026rsquo; ITD was random and also which one out of the two screenshots corresponding to values of a variable was then randomly defined.\nAfter these three slides, a manipulation check was also included, in a compromise (and not ideal) version designed to test the effectiveness of manipulations in the experimental design, but also not to give away the purpose of the data collection. Finally, the last slide included some measures of respondents\u0026rsquo; demographics.\nIn conclusion, by conducting an analysis of variance and using three different regression models for each one of the three focal variables, the experiment aims to investigate their relation, together with the moderation of three involvement measures, with ITD. In addition, by comparing results from models across different variables, it is possible to determine which one is the most and the least effective variable. Finally, by introducing a variable that approximates comparison, the consequences of an app being shown as first or after other manipulated apps are investigated.\n3.1.1 Variables As anticipated, this research aims to determine how three variables, reflected in some elements of the app presentation page on mobile app stores, impact users\u0026rsquo; choice to download an app. In this experimental design, app reputation, popularity and developer\u0026rsquo;s brand are included in form of three binary variables. In order to make these variables manipulation effective, it is necessary to define correctly the two values for each one of the binary variables.\nApart from the three main variables, also involvement measures are included in the analysis models: definitions for involvement variables are available in the variables summary in Table 3.1, as well as for the variable reflecting comparison of an app with others. Since ITD was measured after the exposure to each one of the three main independent variables, three corresponding dependent variables (also defined in Table 3.1) resulted from the data collection and they are separately used in different analysis models.\nTable 3.1: Summary of all variables with definitions and descriptive statistics. Note: this table is also available as an image.\nVariable Definition Descriptive Statistics Mean S.D. Min Max Dependent Variables $$y_{rep}$$ Measured ITD of the apps with manipulated reputation 4.0468 1.7077 1 7 $$y_{pop}$$ Measured ITD of apps with manipulated popularity 4.5255 1.5675 1 7 $$y_{brand}$$ Measured ITD of apps with manipulated developer\u0026rsquo;s brand 4.9674 1.5863 1 7 Independent Variables $$x_{rep}$$ Binary variable, app reputation in terms of average review ratings: high (=1) vs. low (=0) 0.4847 0.4998 0 1 $$x_{pop}$$ Binary variable, app popularity in terms of number of downloads: high (=1) vs. low (=0) 0.5031 0.5000 0 1 $$x_{brand}$$ Binary variable, developer\u0026rsquo;s brand strength and fame: strong (=1) vs. weak (=0) 0.5214 0.4995 0 1 $$inv\\_app$$ Mean-centered involvement measure inmobile apps (as product category) 0 1.0523 -3.6619 1.6715 $$inv\\_dl$$ Mean-centered involvement measure in the app download process 0 1.2974 -3.6933 2.0367 $$inv\\_cat$$ Mean-centered involvement measure in the experimental app category (mobile scanner apps) 0 1.0629 -3.4802 2.5198 $$comp_i$$ Dummy variable, position in which i=rep, pop, brand was shown to users: first (=0) vs. second or third (=1) 0.6667 0.4714 0 1 App Reputation. In this research, app reputation (rep) is defined in terms of the average review rating in the five-star scale currently used on the Play Store. Taking into consideration the dynamics that affect online review ratings, such as, for example, reputation inflation and the not linear impact of good and bad reviews, the threshold above which a review rating is considered to be high cannot be set at the middle of the scale. Indeed, similarly to Roma and Ragaglia (2016, p. 180), this limit was set at the four-star level. This decision was also supported by the preliminary analysis of mobile scanner apps (reported in Appendix C), which shows how the large majority of the most popular scanner apps on the Play Store have average ratings above 4.5 stars. As a consequence, an average rating of 4.9 stars was defined as a high app reputation, while a value of 3.4 was defined as low. In both cases, the number of reviews was set to the same medium-high level, in order to give credibility to the presented rating.\nApp Popularity. As already discussed, app popularity (pop) can be reflected in many elements on an app store. In this case, since the aim was to directly investigate the specific moment in which users take the download decision, the number of downloads was chosen as a metric, also considering it is presented in a spotlighted position in the app page. The number of downloads is presented on the Play Store in form of the last downloads threshold that the app achieved and overcame (e.g. 1M+, where M stands for million), consequently the issue consists in defining above which downloads bracket an app can be considered popular. Since the analysis in Appendix C already took into consideration only popular apps, a biased sample that drives upwards the average downloads number, and the mode out that group was 10M+, the decision was to set this level as a high app popularity. In contrast, the low app popularity level was set to 10K+ (where K means thousand).\nDeveloper\u0026rsquo;s Brand. The developer\u0026rsquo;s brand (brand) affects multiple elements in the app presentation page (app icon, app name and developer\u0026rsquo;s name). In this experiment, a brand was defined as strong when it enjoyed worldwide fame, also outside the mobile ecosystem, and Adobe, which is the eighth largest software company in the world by revenue, with more than 16 billion in 2022 (Murphy and Contreras, 2022), perfectly fits the definition. On the other side, brands were considered weak when they included generic references to their products, without a strong brand identity (e.g. \u0026ldquo;Office Tools\u0026rdquo; or \u0026ldquo;Tools \u0026amp; Utilities Apps\u0026rdquo;).\n3.1.2 Sample and Data Collection In total 545 participants have taken part in the research questionnaire (available in Appendix B), while 491 of them completed it, with a completion rate of 90.09 percent, and, consequently, were included in the sample. The majority of respondents in the sample were between the 18 and 24 years old (59.27%), while the rest was between 25 and 34 (28.11%), between 35 and 44 (5.91%), between 45 and 54 (2.85%), over 55 (2.04%) or under 18 years old (1.83%). More often respondents were females (63.34%) and less frequently males (34.83%), non-binary people (0.81%) or preferred not to specify (1.02%). Regarding their occupation, students were the most common category (69.45%), followed by full-time (19.55%) and part-time employees (5.70%), self-employed workers (2.65%), unemployed (0.81%), pensioners (0.41%) and participants with other occupations (1.43%). Finally, in terms of education level, the majority in the sample had achieved a bachelor\u0026rsquo;s degree (38.70%), while others had a high school diploma (35.85%), a master\u0026rsquo;s degree (18.94%) or a doctorate (1.83%). Some participants didn\u0026rsquo;t conclude high school (2.85%) or undertook other educational paths (1.83%).\nOverall, ITD was measured three times for each one of the 491 participants, for a total of 1473 combined values of yrep, ypop and ybrand.\n3.1.3 Data Analysis Since this data collection corresponds to a repeated measures experimental model, the first step consisted in running an analysis of variance (ANOVA) among the three levels of different groups of respondents from which we are trying to achieve some results. In favour of better describing the ANOVA and also the following models, it is necessary to define i, j, k = rep, pop, brand with i ≠ j ≠ k. Then, it is useful to redefine xi as xin and yi as yin, where n = 1, 2, 3 is the random position (first, second or third) in which users were exposed to the variable i. An ANOVA investigating variance across groups defined by different values on three levels (i, xi and n) that are supposed to affect each measure of yin may help in understanding if some effects on the dependent variable exist and where they come from.\nThereafter, in order to derive more precise results, three different linear regression models were used for each one of the focal variables, to cover all possibilities and for a total of nine regressions. Now, the simplest regression model (Model 1) can be defined as:\nwhere u is the error term. The data set used for Model 1 (M1) for i includes only observations in which respondents were exposed to i before j and k (n = 1), in a way that the resulting ITD cannot be biased by comparison with other apps. In addition, it allows to verify the moderating effects of various measures of involvement on the relationship between the focal predictor and ITD. This model is the main source of evidence for the effects of the focal independent variable xi and the role of involvement as moderator.\nThen, Model 2 (M2) allows for two-way interactions also between different focal independent variables and takes into account only observations with n = 3, in the following way:\nwith m = 1, 2. The main objective of Model 2 is to explore whether there is any interaction between each couple of focal variables, reflecting comparison effects which depend on their alternatively high or low values.\nFinally, Model 3 (M3) can be defined considering all observations in the sample, independently from any specific value of n and introducing the compi variable and its interaction:\nIn the case of equation (3.3), differently from (3.1) and (3.2), n can take values which are not only 1 or 3, meaning that, while it allows to use a larger sample, it needs to include the compi variable (presented in Table 3.1), in order to take in account the possibility that results could be affected by comparison with xj and xk, which are not included in Model 3.\nConcluding, after analyzing the preliminary results of the ANOVA, the comparison of coefficients resulting from models with the same value of i, together with the coefficient of compi from Model 3, helps to reach a further objective of this research, which consists in understanding how an approximation of comparison impacts the customer choice, while differences in coefficients from models depending concurrently on i, j or k could be explained by diverse effectiveness of app reputation, app popularity and brand.\n3.2 Results From the ANOVA emerges how variance is statistically significant across groups defined by different focal variables i (F(2, 4) = 42.187, p \u0026lt; 0.01), by different levels of xi (F(1, 4) = 60.841, p \u0026lt; 0.01) and by different presentation positions n (F(2, 4) = 2.369, p \u0026lt; 0.1), meaning that these three dimensions are likely to predict differences in the dependent variables yin. In addition, also the results regarding the three-way interaction i × xi × n are statistically significant (F(4) = 2.681, p \u0026lt; 0.05) and provide further foundations for subsequent analyses.\nThen, the following step consisted of three different models and nine linear regressions. All information from these regressions, including R2 values and manipulation check ratios, are available in Table 3.2. Model 1 included in turn a focal variable (corresponding to different values of i) and involvement measures, together with their interaction terms. From the first model emerges how, when the app was shown as first (n = 1), the developer\u0026rsquo;s brand was the most effective in predicting ITD (β = 1.5289, p \u0026lt; 0.01), followed by reputation (β = 0.4986, p \u0026lt; 0.1) and popularity (β = 0.3611, p \u0026lt; 0.1). Regarding their interactions with involvement, a positive moderating effect of involvement in the download process on reputation was found (β = 0.3861, p \u0026lt; 0.1), a similar effect to the positive one operated by involvement in the specific app category on the developer\u0026rsquo;s brand (β = 0.4001, p \u0026lt; 0.1).\nThe ITD data included in Model 2 was measured after respondents were exposed to the third app (n = 3), which contained a manipulated value of a random one out of reputation, popularity and brand. The main aims of this model were to determine whether the selected focal variable could be moderated by the values of the other two to which users were previously exposed and also to verify if the effectiveness on ITD remained unchanged if compared to Model 1. In the results from Model 2, there was a reversal between the developer\u0026rsquo;s brand (β = 1.0347, p \u0026lt; 0.05) and reputation (β = 1.3351, p \u0026lt; 0.01), which became the most important element of the three, while the effect of popularity was statistically insignificant. No significant evidence of any moderation effect between one focal independent variable on the others surfaced, while there was evidence, confirmed also by Model 3, of a positive moderation of involvement in apps as generic products on all three focal variables xrep, xpop and xbrand.\nFinally, regressions based on Model 3 included all observations, independently from the position in which apps were shown. The main objective was to investigate the moderating effect of the variable compi, while also confirming the levels of effectiveness of x_i, as derived from Model 1 and 2. Out of Model 3 emerged that, when accounting for all the observations (independently from n values), the developer\u0026rsquo;s brand was the best predictor of ITD (β = 1.5028, p \u0026lt; 0.01), followed by reputation (β = 0.4576, p \u0026lt; 0.1) and popularity (β = 0.3677, p \u0026lt; 0.1). Furthermore, there is evidence of a statistically relevant and negative moderating effect of comparison on brand effectiveness (β = -0.7454, p \u0026lt; 0.05) in predicting ITD.\nTable 3.2: Resulting coefficients and standard errors of all regressions. Note: this table is also available as an image.\nModel (i) M1 (rep) M1 (pop) M1 (brand) M2 (rep) M2 (pop) M2 (brand) M3 (rep) M3 (pop) M3 (brand) (Intercept) 3.86914*** (0.18683) 4.63458*** (0.15161) 4.13417*** (0.17373) 3.69158*** (0.26951) 4.14837*** (0.30527) 4.341602*** (0.333232) 3.906095*** (0.190949) 4.64269*** (0.15454) 4.118051*** (0.178167) $$x_{rep}$$ 0.49862* (0.26259) 1.33507*** (0.41115) 0.25362 (0.37742) -0.285980 (0.360243) 0.457599* (0.268971) $$x_{pop}$$ 0.36110* (0.21478) -0.21677 (0.31337) 0.53397 (0.46800) 0.406845 (0.363070) 0.36773* (0.21838) $$x_{brand}$$ 1.52889*** (0.23554) -0.30136 (0.31366) -0.18598 (0.37590) 1.034683** (0.432264) 1.502819*** (0.240948) $$inv\\_app$$ 0.09502 (0.17658) 0.3581** (0.16063) 0.23275 (0.15521) -0.01990 (0.15111) -0.18401 (0.20203) -0.188448 (0.183269) -0.028741 (0.099168) 0.10192 (0.09446) 0.083030 (0.093585) $$inv\\_dl$$ -0.21222 (0.15674) -0.2292* (0.1235) -0.09778 (0.13349) 0.04958 (0.11528) -0.2527* (0.1505) -0.034777 (0.139169) 0.008458 (0.080334) -0.16296** (0.07627) -0.004562 (0.077751) $$inv\\_cat$$ 0.35115* (0.20204) 0.22163 (0.14454) -0.04189 (0.15636) 0.12016 (0.14256) 0.51847*** (0.19119) 0.287315 (0.173767) 0.106364 (0.099436) 0.29048*** (0.09247) 0.067097 (0.089394) $$comp_i$$ -0.353854 (0.227152) -0.28259 (0.19705) 0.473583** (0.211797) $$x_{i}×x_{rep}$$ -0.74373 (0.53923) 0.251013 (0.507205) $$x_{i}×x_{pop}$$ -0.37078 (0.46066) -0.676553 (0.511097) $$x_{i}× x_{brand}$$ 0.16603 (0.46738) 0.39159 (0.54648) $$x_{i}×inv\\_app$$ 0.2172 (0.26755) -0.0579 (0.23277) 0.24104 (0.21184) 0.28639 (0.22414) 0.57307** (0.26890) 0.540929** (0.260608) 0.278238* (0.146581) 0.29718** (0.13356) 0.214755 (0.131852) $$x_{i}× inv\\_dl$$ 0.38614* (0.23255) 0.10327 (0.16633) -0.03326 (0.17832) -0.14902 (0.17801) 0.08568 (0.22437) -0.071512 (0.193776) -0.004344 (0.116336) 0.00806 (0.10656) -0.174237 (0.105738) $$x_{i}× inv\\_cat$$ -0.10643 (0.27486) 0.06545 (0.20051) 0.40013* (0.23065) 0.18534 (0.21138) -0.42987 (0.27914) 0.006635 (0.253132) 0.200011 (0.142551) -0.11139 (0.13035) 0.158655 (0.128399) $$x_{i}× comp_i$$ 0.510013 (0.322484) -0.39837 (0.27846) -0.745391** (0.289854) R2 0.09523 0.139 0.2885 0.1874 0.1417 0.1533 0.098 0.1056 0.1555 Adj. R2 0.05031 0.1057 0.2542 0.1388 0.06958 0.08682 0.08112 0.08885 0.1397 Observations (N) 149 189 153 196 143 152 491 491 491 Manipulation Check (%) 75.84 62.43 68.63 79.08 55.94 63.16 77.19 60.49 67.41 ***p\u0026lt;0.01, **p\u0026lt;0.05, *p\u0026lt;0.1 3.3 Discussion and Managerial Implications The results indicate that the developer\u0026rsquo;s brand is generally the most effective factor in determining dl decision if compared to reputation (review ratings) and popularity (number of dls), but effectiveness varies on presentation order and information available to users (M1, M3). An unexpected outcome is the scarce (and almost zero) effectiveness of popularity, even if this setting had very low network effects. A hypothesis on the reason why popularity is not frequently taken into consideration by users may rest in the necessity for platform owners (as Google, in this case) to keep the ecosystem competitive and attractive for new developers, avoiding single-firm dominated markets and preserving information transparency on the platform (Boudreau et al., 2022, p. 1379), in particular when network effects are present: in fact, the format in which quantity of downloads is presented (e.g. 10M+ vs. 10K+) may be intentionally not very powerful in communicating the magnitude of popularity differences across apps and it may go unnoticed when users are not reviewing information carefully.\nThe high effectiveness of brands in shaping users\u0026rsquo; choices, even in an online market where a wide range of information is available, opens up relevant opportunities for established software companies. In fact, developing different apps, covering many different segments, already proved to be a successful strategy, since Lee and Raghu (2012, p. 161) pointed out how diversifying app portfolios across selling categories is a key determinant of sales performance.\nWhen users are exposed to information about multiple apps and they can consequently compare them, reputation becomes the most effective element (M2), as also shown by the parallel in Fig. 3.1 between ITD for apps shown as first (n = 1) and those shown as third (n = 3). Users may rely more on eWOM when they have the possibility to consider a wider range of options and they just may need some anchors to better understand apps\u0026rsquo; rating levels. An important cue for making an informed decision can be often found in past experience, but, when users have no previous experience with the specific product, they often rely on social experiences, as expressed in reviews (Burgers et al., 2016, p. 329). As a consequence, it is important, in particular for new entrants in an app category that are not likely to be shown in top search results, to publish an app when it is mature enough to avoid the common phenomenon of reputation backlash, in terms of initial low ratings. On the other side, brands become less predictive of ITD when comparison is introduced in the model (as evidenced by the coefficient of compi in M3): a popular brand may combine well with the \u0026ldquo;take the first\u0026rdquo; heuristic, which has already proven to be effective in the case of mobile apps (Dogruel et al., 2015, p. 139), and be penalised when users are actively collecting information.\nFigure 3.1: Mean-centered ITD (yi) depending on high vs. low values (x) of developer\u0026rsquo;s brand, reputation and popularity, with n = 1 and n = 3. Some other interesting evidence regards the moderation effects of involvement on the three focal variables. First, there is predictable and generalized evidence that higher involvement in apps as products corresponds to a higher effectiveness of all three reputation, popularity and brand levels in driving users\u0026rsquo; ITD (M2, M3). Second, when users are highly involved in the app download process, they tend to rely more on reputation, probably because they scrutinise information more closely. Last, when the involvement in the specific app category (mobile scanner apps, in this case) is high and users consequently perceive a higher risk in the app choice, also considering that scanner apps are productivity apps, frequently used for work (Sällberg et al., 2022, p. 10), they appear to preferably choose a strong and known brand (M1), which is a commonly used as a cue when attempting to mitigate it (Matzler et al., 2008, p. 158).\n3.4 Limitations This experimental approach presents some limitations. The first and major limitation lies in the experimental design: in fact, while investigating the effect of three factors on ITD, the selected one was not a traditional three-way factorial design and just one variable was manipulated at each measurement of ITD, while, at the same time, the others were set to intermediate values. As a consequence, estimates of interactions among factors were not ideal, but, on the other side, it allowed to effectively collect results on comparison effects. Consequently, despite the repeated measure design, it was not possible to include all observations of yi in each sample used for each regression, since every measure of ITD was strictly related to a manipulated value of reputation, popularity or developer\u0026rsquo;s brand (i) happening in the specific case.\nSome other limitations lie in the data collection. For example, the screenshot images used in the questionnaire were all collected from Google Play Store, the most used app store, meaning that users of other platforms, for example, Apple iOS, could be used to a different interface and to look for elements in other positions, affecting, therefore, the resulting ITD. In addition, some people, such as those with high levels of involvement, for example, may find limiting the experimental setting, in which many important elements were not considered or even included.\nFinally, the sample may represent a limitation. In fact, the large majority of respondents were students and they were aged between 18 and 24 and results could be biased by this unequal balance of composition. Nowadays, smartphones are widespread across all customer age groups, but this doesn\u0026rsquo;t mean that these groups all share the same levels of involvement and rely on the same factors when making decisions that concern downloading an app or not. Additionally, while this is not certain, participants can be assumed to be mainly from Germany and Italy and there is some evidence that behaviour on app stores and factors relevant to the purchase decision vary across countries (Kübler et al., 2018, p. 37), meaning that results may vary depending on sample geographical origin and cultural background.\nChapter 4. Conclusion 4.1 Summary As a first step, in Chapter 1, an extensive background on the mobile ecosystem, its participants and its dynamics was provided, before reviewing, in Chapter 2, previous literature on user behaviour and determinants of downloads on mobile app stores. Finally, in Chapter 3, the empirical approach of this research was discussed, together with its results. The outcome of this experiment shows how the developer\u0026rsquo;s brand is generally the most decisive element, but, when a comparison between apps is introduced in the model, it loses its effectiveness and reputation becomes more relevant. Popularity and peer influence, in the context of low network effects, appear to be unexpectedly ineffective in driving users\u0026rsquo; decisions across all models. Moreover, users\u0026rsquo; involvement in the download process reinforces the effect of reputation, while, when customers are more involved in the specific app category and perceive a higher risk, apps from established brands are more likely to benefit. Finally, it was not possible to find clear evidence of any two-way interaction between reputation, popularity and brand.\n4.2 Future Directions This research results add up to the substantial literature that already covered customer behaviour on online marketplaces, but some elements remain unexplored. For example, a better experimental design aimed to explore factorial interactions should be developed. Furthermore, peer influence effectiveness and its dynamics on app stores, both in the presence and absence of network effects, should be better investigated.\nBibliography Ali, M., Joorabchi, M. E., \u0026amp; Mesbah, A. (2017). Same app, different app stores: A comparative study. In 2017 IEEE/ACM 4th International Conference on Mobile Software Engineering and Systems (MOBILESoft) (pp. 79-90), IEEE.\nAnderson, C. (2006). The long tail: Why the future of business is selling less of more. Hachette UK.\nArora, S., Ter Hofstede, F., \u0026amp; Mahajan, V. (2017). The implications of offering free versions for the performance of paid mobile apps. Journal of Marketing, 81(6), 62-78.\nBasole, R. C., \u0026amp; Karla, J. (2012). Value transformation in the mobile service ecosystem: A study of app store emergence and growth. Service Science, 4(1), 24-41.\nBöhm, S., \u0026amp; Schreiber, S. (2014). Mobile App Marketing: A Conjoint-based Analysis on the Importance of App Store Elements. In CENTRIC 2014: The Seventh International Conference on Advances in Human-oriented and Personalized Mechanisms, Technologies, and Services.\nBoudreau, K. J., Jeppesen, L. B., \u0026amp; Miric, M. (2022). Competing on freemium: Digital competition with network effects. Strategic Management Journal, 43(7), 1374-1401.\nBrynjolfsson, E., Hu, Y., \u0026amp; Smith, M. D. (2003). Consumer surplus in the digital economy: Estimating the value of increased product variety at online booksellers. Management science, 49(11), 1580-1596.\nBrynjolfsson, E., \u0026amp; Smith, M. D. (2000). Frictionless commerce? A comparison of Internet and conventional retailers. Management science, 46(4), 563-585.\nBuck, C., Horbel, C., Germelmann, C. C., \u0026amp; Eymann, T. (2014). The Unconscious App Consumer: Discovering and Comparing the Information-seeking Patterns among Mobile Application Consumers. In ECIS.\nBurgers, C., Eden, A., de Jong, R., \u0026amp; Buningh, S. (2016). Rousing reviews and instigative images: The impact of online reviews and visual design characteristics on app downloads. Mobile Media \u0026amp; Communication, 4(3), 327-346.\nCeci, L. (2022). Google Play: annual consumer spending on mobile apps 2016-2021. Statista, https://www.statista.com/statistics/444476/google-play-annual-revenue/ (Accessed in January 2023).\nCeci, L. (2022). Number of apps available in leading app stores Q3 2022. Statista, https://www.statista.com/statistics/276623/number-of-apps-available-in-leading-app-stores/ (Accessed in January 2023).\nCeluch, K. G., \u0026amp; Taylor, S. A. (1999). Involvement with services: An empirical replication and extension of Zaichkowsky\u0026rsquo;s personal involvement inventory. The Journal of Consumer Satisfaction, Dissatisfaction and Complaining Behavior, 12, 109-122.\nChen, H., Lachaud, K., \u0026amp; Zhou, W. (2022). The sales effect of “Free App of the Day” on Amazon Appstore: An empirical study. Digital Business, 2(2), 100020.\nChevalier, J. A., \u0026amp; Mayzlin, D. (2006). The effect of word of mouth on sales: Online book reviews. Journal of marketing research, 43(3), 345-354.\nCortimiglia, M. N., Ghezzi, A., \u0026amp; Renga, F. (2011). Mobile applications and their delivery platforms. IT Professional, 13(5), 51-56.\nDogruel, L., Joeckel, S., \u0026amp; Bowman, N. D. (2015). Choosing the right app: An exploratory perspective on heuristic decision processes for smartphone app selection. Mobile Media \u0026amp; Communication, 3(1), 125-144.\nEagly, A. H., \u0026amp; Chaiken, S. (1993). The psychology of attitudes. Harcourt brace Jovanovich college publishers.\nEricsson (2022). Ericsson Mobility Report, https://www.ericsson.com/en/reports-and-papers/mobility-report/reports/november-2022 (Accessed in January 2023).\nFilippas, A., Horton, J. J., \u0026amp; Golden, J. (2018). Reputation inflation. In Proceedings of the 2018 ACM Conference on Economics and Computation (pp. 483-484).\nFinkelstein, A., Harman, M., Jia, Y., Martin, W., Sarro, F., \u0026amp; Zhang, Y. (2017). Investigating the relationship between price, rating, and popularity in the Blackberry World App Store. Information and Software Technology, 87, 119-139.\nFriedman L. (2013). The App Store turns five: A look back and forward. Macworld, https://www.macworld.com/article/221393/the-app-store-turns-five-a-look-back-and-forward.html (Accessed in January 2023).\nGokgoz, Z. A., Ataman, M. B., \u0026amp; van Bruggen, G. H. (2021). There’s an app for that! Understanding the drivers of mobile application downloads. Journal of Business Research, 123, 423-437.\nGoldfarb, A., \u0026amp; Tucker, C. (2019). Digital economics. Journal of Economic Literature, 57(1), 3-43.\nHalmenschlager, C., \u0026amp; Waelbroeck, P. (2014). Fighting free with free: Freemium vs. piracy. Piracy (November 29, 2014).\nHorton, J., \u0026amp; Golden, J. (2015). Reputation inflation in an online marketplace. New York I, 1.\nHyrynsalmi, S., Mäkilä, T., Järvi, A., Suominen, A., Seppänen, M., \u0026amp; Knuutila, T. (2012). App store, marketplace, play! an analysis of multi-homing in mobile software ecosystems. Jansen, Slinger, 59-72.\nIhlwan, M. (2007). LG ties up with PRADA for new phone. Bloomberg, https://www.bloomberg.com/news/articles/2007-01-17/lg-ties-up-with-prada-for-new-phone (Accessed in January 2023).\nJung, E. Y., Baek, C., \u0026amp; Lee, J. D. (2012). Product survival analysis for the App Store. Marketing Letters, 23(4), 929-941.\nKang, J. Y. M., Mun, J. M., \u0026amp; Johnson, K. K. (2015). In-store mobile usage: Downloading and usage intention toward mobile location-based retail apps. Computers in Human Behavior, 46, 210-217.\nKoch, S., \u0026amp; Kerschbaum, M. (2014). Joining a smartphone ecosystem: Application developers’ motivations and decision criteria. Information and Software Technology, 56(11), 1423-1435.\nKouris, I., \u0026amp; Piller, F. T. (2014). App platforms as two-sided markets: analysis and modeling of application distribution platforms for mobile devices. Lehrstuhl für Betriebswirtschaftslehre, insbesondere Technologie-und Innovationsmanagement.\nKübler, R., Pauwels, K., Yildirim, G., \u0026amp; Fandrich, T. (2018). App popularity: Where in the world are consumers most sensitive to price and user ratings?. Journal of Marketing, 82(5), 20-44.\nLambrecht, A., Goldfarb, A., Bonatti, A., Ghose, A., Goldstein, D. G., Lewis, R., Rao, A. and Sahni, N. \u0026amp; Yao, S. (2014). How do firms make money selling digital goods online?. Marketing Letters, 25(3), 331-341.\nLaricchia, F. (2022). Market share of mobile operating systems worldwide 2009-2022. Statista, https://www.statista.com/statistics/272698/global-market-share-held-by-mobile-operating-systems-since-2009/ (Accessed in January 2023).\nLee, G., \u0026amp; Raghu, T. S. (2014). Determinants of mobile apps\u0026rsquo; success: Evidence from the app store market. Journal of Management Information Systems, 31(2), 133-170.\nLiang, T. P., Li, X., Yang, C. T., \u0026amp; Wang, M. (2015). What in consumer reviews affects the sales of mobile apps: A multifacet sentiment analysis approach. International Journal of Electronic Commerce, 20(2), 236-260.\nLiu, C. Z., Au, Y. A., \u0026amp; Choi, H. S. (2014). Effects of freemium strategy in the mobile app market: An empirical study of google play. Journal of management information systems, 31(3), 326-354.\nMatzler, K., Grabner‐Kräuter, S., \u0026amp; Bidmon, S. (2008). Risk aversion and brand loyalty: the mediating role of brand trust and brand affect. Journal of product \u0026amp; brand management, 17(3), 154-162.\nMittal, B. (1995). A comparative analysis of four scales of consumer involvement. Psychology \u0026amp; marketing, 12(7), 663-682.\nMoe, W. W., \u0026amp; Trusov, M. (2011). The value of social dynamics in online product ratings forums. Journal of Marketing Research, 48(3), 444-456.\nMüller, R. M., Kijl, B., \u0026amp; Martens, J. K. (2011). A comparison of inter-organizational business models of mobile app stores: There is more than open vs. closed. Journal of theoretical and applied electronic commerce research, 6(2), 63-76.\nMurphy A., \u0026amp; Contreras I. (2022). The Global 2000. Forbes, https://www.forbes.com/lists/global2000/ (Accessed in January 2023).\nNan, G., Wu, D., Li, M., \u0026amp; Tan, Y. (2018). Optimal Freemium Strategy for Information Goods in the Presence of Piracy. Journal of the Association for Information Systems, 19(4), 266-305.\nPiccoli, G., \u0026amp; Pigni, F. (2019). Information Systems for Managers With Cases, Edition 4.0. Prospect Press, Inc..\nPol, M. (2015). App icon preferences: The influence of app icon design and involvement on quality and intention to download. Master\u0026rsquo;s thesis, University of Twente.\nReavis Conner, K., \u0026amp; Rumelt, R. P. (1991). Software piracy: An analysis of protection strategies. Management science, 37(2), 125-139\nRochet, J. C., \u0026amp; Tirole, J. (2003). Platform competition in two-sided markets. Journal of the european economic association, 1(4), 990-1029.\nRoma, P., \u0026amp; Ragaglia, D. (2016). Revenue models, in-app purchase, and the app performance: Evidence from Apple’s App Store and Google Play. Electronic commerce research and applications, 17, 173-190.\nRubin, A. D. (1995). Trusted distribution of software over the Internet. In Proceedings of the Symposium on Network and Distributed System Security (pp. 47-53), IEEE.\nRuiz, I. J. M., Nagappan, M., Adams, B., Berger, T., Dienst, S., \u0026amp; Hassan, A. E. (2015). Examining the rating system used in mobile-app stores. Ieee Software, 33(6), 86-92.\nSalganik, M. J., Dodds, P. S., \u0026amp; Watts, D. J. (2006). Experimental study of inequality and unpredictability in an artificial cultural market. science, 311(5762), 854-856.\nSällberg, H., Wang, S., \u0026amp; Numminen, E. (2022). The combinatory role of online ratings and reviews in mobile app downloads: an empirical investigation of gaming and productivity apps from their initial app store launch. Journal of Marketing Analytics, 1-17.\nSpears, N., \u0026amp; Singh, S. N. (2004). Measuring attitude toward the brand and purchase intentions. Journal of current issues \u0026amp; research in advertising, 26 (2), 53-66.\nTimmerman, J. E., \u0026amp; Shepherd, I. (2016). Does eWOM affect demand for mobile device applications?. Journal of Marketing Development \u0026amp; Competitiveness, 10(3).\nWang, Y., Song, J., \u0026amp; Aguirre-Urreta, M. (2015). An empirical investigation of factors impacting application downloads in mobile app stores. SIGHCI Proceedings, 1-5.\nZaichkowsky, J. L. (1994). The personal involvement inventory: Reduction, revision, and application to advertising. Journal of advertising, 23(4), 59-70.\nZheng, P., \u0026amp; Ni, L. (2010). Smart phone and next generation mobile computing. Elsevier.\nAppendices A App Page Screenshots This online study involves the use of manipulated screenshots images of an app presentation page on Google\u0026rsquo;s Play Store (two for each one of the three main variables). Each couple of screenshots of existing apps was manipulated in a way that the assigned variable could take different values. Only the elements directly related to that variable were admitted to change across each couple of presentation pages. As shown in Fig. A.1, in screenshot images dealing with app reputation (high vs. low) the average rating is the only element manipulated. Similarly, also in Fig. A.2, only the number of downloads is manipulated for app popularity (high vs. low). Finally, Fig. A.3 shows how the elements dealing with the developer\u0026rsquo;s brand (strong vs. weak) are three: the app\u0026rsquo;s icon, the app\u0026rsquo;s name and the developer\u0026rsquo;s name.\nFigure A.1: Screenshot images of the app presentation page with variation in app reputation. Figure A.2: Screenshot images of the app presentation page with variation in app popularity. Figure A.3: Screenshot images of the app presentation page with variation in developer’s brand. B Experimental Questionnaire Note: in this online survey some presentation pages from Google\u0026rsquo;s Play Store were included. These images are available in Appendix A.\nSlide 0. Presentation page and consent form.\nWelcome!\nThis is an online study on apps and app stores.\nThe expected duration time is less than 5 minutes.\nThrough your participation in this survey, you are making an important contribution to my research. I greatly appreciate your support.\nWhy do we collect and use your data\nThe purpose of this online study is to obtain new insights on app preferences and app store customer behavior.\nHow to contact us\nLuca Bontempi - Student at the University of Tübingen and at the University of Pavia. Email: luca.bontempi@student.uni-tuebingen.de\nIf you would like to obtain more information about the processing of your personal data, please click here\nO I agree to the processing of my personal data in accordance with the information provided herein\nO I don\u0026rsquo;t want to participate O Start the survey\nNote: respondents had to agree and select \u0026ldquo;Start the survey\u0026rdquo; in order to proceed to the next slide.\nSlide 1. Involvement in mobile apps and app download process.\nRead and answer the following questions on apps and app stores.\nWhat do you think about mobile apps?\nMobile apps are very important to me.\nStrongly disagree O 1 O 2 O 3 O 4 O 5 O 6 O 7 Strongly agree\nFor me, mobile apps do not matter.\nStrongly disagree O 1 O 2 O 3 O 4 O 5 O 6 O 7 Strongly agree\nI have a strong interest in mobile apps.\nStrongly disagree O 1 O 2 O 3 O 4 O 5 O 6 O 7 Strongly agree\nWhat do you think about downloading apps?\nI choose the mobile apps to download very carefully.\nStrongly disagree O 1 O 2 O 3 O 4 O 5 O 6 O 7 Strongly agree\nWhich mobile apps I download matters to me a lot.\nStrongly disagree O 1 O 2 O 3 O 4 O 5 O 6 O 7 Strongly agree\nChoosing an app is an important decision for me.\nStrongly disagree O 1 O 2 O 3 O 4 O 5 O 6 O 7 Strongly agree\nSlide 2. Involvement in the mobile scanner apps category.\nConsider mobile apps that allow you to digitize documents through the camera (Cam Scanner Apps). Read the following statement.\nTo me digitizing documents using a mobile scanner app is:\nSelect your personal opinion with regard to the following couples of words.\nimportant O 1 O 2 O 3 O 4 O 5 O 6 O 7 unimportant\nrelevant O 1 O 2 O 3 O 4 O 5 O 6 O 7 irrelevant\nexciting O 1 O 2 O 3 O 4 O 5 O 6 O 7 unexciting\nmeans nothing O 1 O 2 O 3 O 4 O 5 O 6 O 7 means a lot to me\nappealing O 1 O 2 O 3 O 4 O 5 O 6 O 7 unappealing\nfascinating O 1 O 2 O 3 O 4 O 5 O 6 O 7 mundane\nworthless O 1 O 2 O 3 O 4 O 5 O 6 O 7 valuable\nnot needed O 1 O 2 O 3 O 4 O 5 O 6 O 7 needed\nSlide 3, 4 and 5. Main experiment.\nImagine you are looking for a scanner app for your smartphone since you need to digitize a document. In the Android App Store (\u0026ldquo;Google Play Store\u0026rdquo;) you come across the following app. Please carefully review and take into consideration only the available information before you proceed.\nA screenshot image of the app presentation page is shown here.\nIt is very likely that I would download this app.\nPlease indicate your agreement with this statement above.\nO 1 - Fully disagree O 2 O 3 O 4 - Neutral O 5 O 6 O 7 - Fully agree\nNote: Slide 3, 4 and 5 were presented to respondents in randomized order. The screenshot image presented was randomly selected for each slide and variable out of the correspondent couple presented in Appendix A.\nSlide 6. Manipulation check.\nPlease indicate which factors you took into consideration when evaluating the apps on the previous slides.\nO Review rating O Number of downloads O Developer brand O App\u0026rsquo;s name O App\u0026rsquo;s icon O Number of reviews O App\u0026rsquo;s screenshots O PEGI rate O \u0026ldquo;About this app\u0026rdquo; section\nNote: answer options were presented in a randomized order.\nSlide 7. Demographics.\nWhat is your age?\nPlease select your age range.\nO Under 18 O 18-24 years old O 25-34 years old O 35-44 years old O 45-54 years old O Over 55\nWhat is your gender?\nPlease select your gender.\nO Female O Male O Non-binary O Prefer not to say\nWhat is the highest degree or level of school you have completed?\nPlease select your degree level.\nO Less than a high school diploma O High school degree or equivalent O Bachelor’s degree (e.g. BA, BS) O Master’s degree (e.g. MA, MS) O Bachelor’s degree Doctorate (e.g. PhD) O Other\nWhat is your current employment status?\nPlease refer to your main occupation.\nO Student O Employed full-time O Employed part-time O Self-employed O Unemployed O Retired O Other\nSlide 8. Conclusion.\nThe survey is over!\nThank you for your contribution!\nC Preliminary Analysis of Experimental App Category A preliminary analysis of a sample of popular mobile scanner apps was conducted in September 2022. Up to 24 free apps with this function were found to be ranked 300 or higher in two categories (Productivity and Business Tools) on the Play Store. All these apps presented similar names: for example, the app\u0026rsquo;s names always included the words \u0026ldquo;scan\u0026rdquo; or \u0026ldquo;scanner\u0026rdquo;, it included the word \u0026ldquo;PDF\u0026rdquo; in 21 cases and the words \u0026ldquo;doc\u0026rdquo; or \u0026ldquo;documents\u0026rdquo; in 9 cases.\nThe average review rating ranged from 3.2 out of 5 up to 4.9, with a mean value of 4.46 (S.D. = 0.42) and median value of 4.6, while the reviews number was 4.06 million at the highest and 2437 at the lowest, with a mean equal to 446850 (S.D. = 992614) and median value of 92305.\nGoogle\u0026rsquo;s app store reports the number of downloads in the form of the last threshold exceeded by that app. In this sample, apps had 100M+ downloads (where M stands for million) at their best and 50K+ downloads (where K stands for thousand) at their worst. The median value was 10M+. Since this analysis includes only apps that were among a pool of popular ones (top 300) in their categories, many other cam scanner apps with fewer downloads and review ratings are available on app stores.\nFinally, one app in this sample was published by Adobe, a developer known worldwide and mainly for its traditional software. Some other apps were developed by companies which can be considered popular in the mobile apps ecosystem, such as, for example, CamSoft, Simple Design or MobiSystems. All the others were mostly developed by unpopular brands.\nSince the word \u0026ldquo;platform\u0026rdquo; may refer to different elements of the software ecosystem, \u0026ldquo;app (or software) store or marketplace\u0026rdquo; will be used from now on referring to platforms from which software can be purchased and downloaded, as, for example, Google Play Store or Apple App Store for apps or Steam for PC videogames. On the other side, \u0026ldquo;technology platform\u0026rdquo; or just \u0026ldquo;platform\u0026rdquo; will refer in a broader sense to all the software and hardware elements that allow an application to run on a device, often including as its fundamental part the operative system (OS). Finally, \u0026ldquo;two-sided platform\u0026rdquo; may be used to generically refer to any two-sided market, beyond those dealing with software applications.\u0026#160;\u0026#x21a9;\u0026#xfe0e;\n","permalink":"https://lucabontempi.com/it/blog/determinants_of_download_on_mobile_app_stores/","summary":"\u003ch1 id=\"premessa\"\u003ePremessa\u003c/h1\u003e\n\u003cp\u003eQuesto articolo contiene il testo integrale della mia tesi magistrale, scritta presso la \u003cstrong\u003eEberhard Karls University di Tübingen\u003c/strong\u003e e l\u0026rsquo;\u003cstrong\u003eUniversità di Pavia\u003c/strong\u003e e presentata per la prima volta nel \u003cstrong\u003efebbraio 2023\u003c/strong\u003e.\u003c/p\u003e\n\u003cp\u003eNel \u003cstrong\u003esettembre 2026\u003c/strong\u003e ho riaperto questo lavoro e analizzato nuovamente i dati originali. La revisione ha individuato \u003cstrong\u003etre errori nell\u0026rsquo;analisi\u003c/strong\u003e, nessuno nei dati stessi, ed è raccontata in \u003ca href=\"/it/blog/reopening-my-thesis/\" title=\"I Asked an AI to Review My Thesis. It Found Three Mistakes I Could Have Found Myself\"\u003eHo chiesto a un\u0026rsquo;AI di rivedere la mia tesi\u003c/a\u003e. Il testo della tesi qui sotto è quello consegnato allora, lasciato invariato.\u003c/p\u003e","title":"Determinants of Download on Mobile App Stores - An Empirical Analysis"},{"content":"Durante il mio soggiorno a Tübingen, in Germania, fare escursioni era una delle mie attività preferite. E devo ammettere che i dintorni si prestano davvero bene.\nHo quindi raccolto tutte le escursioni che ho fatto, selezionando anche alcune delle migliori, che trovate subito all\u0026rsquo;inizio come itinerari consigliati.\nPer ogni escursione trovate l\u0026rsquo;attività su Strava, spesso corredata anche da fotografie, e un collegamento all\u0026rsquo;attività su Komoot, da cui potete scaricare il file GPX per ripetere il percorso.\nBuone camminate!\nTra le migliori: intorno al castello di Hohenzollern Questa escursione parte dalla stazione ferroviaria di Bisingen, sale al castello di Hohenzollern, che si può visitare, e prosegue sullo Zoller Horn, da cui si gode di una vista splendida sul castello, prima di tornare a Bisingen.\nEcco l\u0026rsquo;attività su Komoot.\nTra le migliori: lo Schwarzwald Genießerpfad da Calw Probabilmente è il modo più semplice per avere un primo assaggio della Foresta Nera. Si parte dalla graziosa cittadina di Calw e si segue un percorso ad anello che passa dalla forca cittadina, dove si eseguivano le condanne a morte fino all\u0026rsquo;inizio dell\u0026rsquo;Ottocento, e da altri angoli piacevoli.\nEcco l\u0026rsquo;attività su Komoot.\nTra le migliori: intorno al Teckberg Questa escursione mi ha sorpreso in positivo. Prima si sale sul Teckberg, dove si trova un piccolo castello, poi si visitano alcune rovine e infine si percorre una passerella sopra un\u0026rsquo;area ricca di serpenti e rane. Questa fotografia è stata scattata proprio lì vicino.\nEcco l\u0026rsquo;attività su Komoot.\nTra le migliori: intorno a Bad Urach Bad Urach è una delle mete migliori per le escursioni: qui vi aspettano le cascate e le rovine di Hohenurach.\nEcco l\u0026rsquo;attività su Komoot.\nLungo il Danubio fino a Sigmaringen Se avete visto il Danubio a Budapest, probabilmente vi stupirà quanto sia piccolo a Sigmaringen. Questa escursione segue il fiume ancora giovane, attraversa la Teufelsbrücke, il Ponte del Diavolo, e termina a Sigmaringen. Un\u0026rsquo;ottima occasione per visitare anche il castello.\nEcco l\u0026rsquo;attività su Komoot.\nDa Tübingen WHO alla Wurmlinger Kapelle e ritorno La Wurmlinger Kapelle si trova all\u0026rsquo;estremità opposta dello Schlossberg di Tübingen. È una bella collina, con una buona vista sulle pianure circostanti.\nEcco l\u0026rsquo;attività su Komoot.\nDa Balingen alla cascata di Zillhausen, o quasi Questa escursione porta da Balingen alla cascata di Zillhausen, che purtroppo era asciutta quando ci sono stato. Immagino che sia molto più bella quando la cascata c\u0026rsquo;è davvero.\nEcco l\u0026rsquo;attività su Komoot.\nDa Tübingen allo Schönbuchturm e ritorno Questa è stata probabilmente la più folle: sette ore tra andata e ritorno, attraversando due volte l\u0026rsquo;intero Schönbuch. All\u0026rsquo;andata ho seguito il sentiero panoramico sui paesi vicini; per il ritorno ne ho scelto uno più interno.\nEcco l\u0026rsquo;attività su Komoot.\nDa Tübingen a Herrenberg Un modo più semplice per visitare il parco dello Schönbuch è partire da Tübingen e terminare a Herrenberg. Herrenberg è una graziosa cittadina e nei dintorni si trova lo Schönbuchturm, una torre panoramica da cui si vede tutta la foresta.\nEcco l\u0026rsquo;attività su Komoot.\nDa Tübingen a Entringen Partendo da Tübingen in direzione di Hagelloch, una frazione della città, questa escursione permette di esplorare rapidamente l\u0026rsquo;angolo sudoccidentale del parco dello Schönbuch e termina a Entringen, un piccolo centro confinante con Tübingen.\nEcco l\u0026rsquo;attività su Komoot.\nDa Tübingen a Dettenhausen passando per Bromberg Anche questa escursione attraversa il parco dello Schönbuch, ma invece di dirigersi a ovest come le precedenti va verso nord, fino a Dettenhausen, con qualche saliscendi sulle colline vicine.\nEcco l\u0026rsquo;attività su Komoot.\nIntorno a Rottenburg am Neckar Rottenburg am Neckar è un\u0026rsquo;altra graziosa cittadina vicino a Tübingen. Questa escursione ne esplora i dintorni.\nEcco l\u0026rsquo;attività su Komoot.\nNello Schönbuch, nella zona di Bebenhausen Questa escursione attraversa la bella area boscosa dello Schönbuch intorno a Bebenhausen, una frazione di Tübingen, e offre una buona occasione per avvistare animali selvatici.\nEcco l\u0026rsquo;attività su Komoot.\nDa Hagelloch a Bebenhausen e dintorni Prima di raggiungere di nuovo Bebenhausen, questa escursione attraversa l\u0026rsquo;area intorno a Hagelloch.\nEcco l\u0026rsquo;attività su Komoot.\nDa WHO a Bebenhausen Probabilmente la più semplice di tutte, quasi obbligatoria se vivete a WHO. Una volta a Bebenhausen potete visitare l\u0026rsquo;abbazia. Anche se al ritorno si sale, resta una passeggiata piuttosto breve e facile.\nEcco l\u0026rsquo;attività su Komoot.\nRingraziamenti Un grande grazie al mio amico Armin, che per mesi è stato la mia fidata guida escursionistica, e al suo fantastico gruppo!\n","permalink":"https://lucabontempi.com/it/blog/tuebingen_hiking_guide/","summary":"\u003cp\u003eDurante il mio soggiorno a Tübingen, in Germania, fare escursioni era una delle mie attività preferite. E devo ammettere che i dintorni si prestano davvero bene.\u003c/p\u003e\n\u003cp\u003eHo quindi raccolto tutte le escursioni che ho fatto, selezionando anche alcune delle migliori, che trovate subito all\u0026rsquo;inizio come itinerari consigliati.\u003c/p\u003e\n\u003cp\u003ePer ogni escursione trovate l\u0026rsquo;\u003ca href=\"https://www.strava.com/athletes/4403103\"\u003eattività su Strava\u003c/a\u003e, spesso corredata anche da fotografie, e un collegamento all\u0026rsquo;\u003ca href=\"https://www.komoot.it/user/2218098976141\"\u003eattività su Komoot\u003c/a\u003e, da cui potete scaricare il file GPX per ripetere il percorso.\u003c/p\u003e","title":"Una guida alle escursioni nei dintorni di Tübingen"},{"content":"Ho raccolto le 36 fotografie che, secondo me, rappresentano meglio la mia esperienza a Tübingen, in Germania. Coprono il periodo da ottobre 2021 a settembre 2022 e sono state scattate soprattutto in città e nei dintorni, con qualche incursione in altre belle città tedesche.\nLa raccolta si intitola «One year in Tü. A Chapter.» ed è disponibile integralmente sul mio profilo Flickr. Tutte le fotografie sono pubblicate con licenza CC BY-NC-SA.\nQui trovate ogni fotografia con un piccolo commento!\n1/ Uno dei miei scorci preferiti di Marktplatz.\n2/ La Stocherkahnrennen è probabilmente uno degli appuntamenti più belli del semestre estivo a Tübingen. Si tiene quasi ogni anno dal 1956: la data da segnare è il Corpus Domini, un giovedì che cade solitamente a giugno.\n3/ La Stiftskirche vista dallo Schloss Hohentübingen, più precisamente da qui.\n4/ Quando sono arrivato per la prima volta a Tübingen, nell\u0026rsquo;ottobre 2021, la Germania era alla vigilia del suo cambiamento più importante degli ultimi anni. Poi Angela Merkel si è ritirata e il paese ha cominciato a essere un po\u0026rsquo; più confuso.\n5/ Questo ascensore, che ha almeno il doppio dei miei anni, è stato letteralmente la prima cosa che ho visto a Tübingen.\n6/ WHO e i suoi edifici «iconici». Ho la sensazione di averci passato fin troppo tempo.\n7/ Forse avrei dovuto fotografare la caffetteria della Universitätsbibliothek: probabilmente ci ho passato più tempo che a studiare…\n8/ Una bici da corsa d\u0026rsquo;epoca davanti alla Neue Aula.\n9/ L\u0026rsquo;interno della Neue Aula, costruita nel 1846.\n10/ Tübingen è una città a misura di bicicletta, una piccola Amsterdam. Almeno le piste ciclabili ci sono…\n11/ Il tunnel che attraversa la collina del castello.\n12/ Qualche italiano che chiacchiera, in bianco e nero.\n13/ Non mi aspettavo che un museo dell\u0026rsquo;automobile potesse essere così bello. Il Mercedes-Benz Museum di Stoccarda.\n14/ Nel 1477 Eberardo I di Württemberg, il cui motto era attempto («oso»), fondò l\u0026rsquo;Università di Tübingen. — Wikipedia, traduzione\nQuesta è la sua statua a Stoccarda.\n15/ Da quel giorno il VfB è diventato la mia squadra tedesca preferita!\n16/ Rothenburg ob der Tauber è stata probabilmente la mia preferita tra le città che ho visitato in Germania.\n17/ Le attrazioni della Stuttgarter Frühlingsfest.\n18/ Un po\u0026rsquo; di tennis di alto livello all\u0026rsquo;ATP 250 di Stoccarda. Nella foto Andy Murray, due volte campione olimpico, durante il riscaldamento.\n19/ Fare escursioni è una delle mie attività preferite. Questa fotografia è stata scattata durante un\u0026rsquo;escursione, vicino allo Schopflocher Torfmoor.\n20/ Il castello di Hohenzollern è uno dei più scenografici della zona. Questa fotografia è stata scattata dallo Zoller Horn, durante un\u0026rsquo;escursione.\n21/ Un\u0026rsquo;altra meta per le escursioni è lo Schönbuchturm a Herrenberg. Se siete abbastanza folli, potete andarci a piedi e tornare.\n22/ Herrenberg è una cittadina piccola ma graziosa, non lontano da Tübingen.\n23/ Un po\u0026rsquo; di sole sulle rive del Neckar.\n24/ Holzmarkt vista dal campanile della Stiftskirche.\n25/ Preparativi per un giovedì sera al Clubhaus.\n26/ Tübingen e i suoi lampioni decorati.\n27/ Questa serata dello Stadtfest di Tübingen è stata probabilmente la più divertente che ho passato in città.\n28/ Un altro [castello da fiaba](https://en.wikipedia.org/wiki/Lichtenstein_Castle_(W%C3%BCrttemberg) a Lichtenstein.\n29/ Un dettaglio del castello di Sigmaringen, costruito sul Danubio. Un\u0026rsquo;escursione potrebbe essere una buona occasione per visitarlo.\n30/ Giuro che la maggior parte degli autobus è più moderna, anche se meno elegante.\n31/ Gli European Championships del 2022 sono stati una manifestazione multisport disputata a Monaco quell\u0026rsquo;estate. Fabio Jakobsen e Arnaud Démare, nella foto, sono arrivati primo e secondo nella prova in linea maschile di ciclismo.\n32/ Sono un grande appassionato di parchi divertimento. Europa-Park è la risposta tedesca a quel grande parco francese. Ma Gardaland regna.\n33/ Quell\u0026rsquo;estate, agli Europei di Colonia, la nazionale tedesca ha ritirato la maglia di Dirk Nowitzki, per me il più grande cestista europeo di sempre, almeno fino a quel momento.\n34/ La vista più famosa di Tübingen, dalla Neckarbrücke.\n35/ La mia stanza vuota in Fichtenweg, il giorno in cui ho lasciato Tübingen.\n36/36 Il confine italiano sul passo dello Spluga, a ottobre, tornando in Italia.\nTutte queste fotografie sono state scattate con una Nikon D5600 o un Oppo Find X2 Neo.\n","permalink":"https://lucabontempi.com/it/blog/tuebingen_collection/","summary":"\u003cp\u003eHo raccolto le 36 fotografie che, secondo me, rappresentano meglio la mia esperienza a Tübingen, in Germania. Coprono il periodo da ottobre 2021 a settembre 2022 e sono state scattate soprattutto in città e nei dintorni, con qualche incursione in altre belle città tedesche.\u003c/p\u003e\n\u003cp\u003eLa raccolta si intitola «\u003ca href=\"https://www.flickr.com/photos/77058054@N07/albums/72177720328778423/\"\u003eOne year in Tü. A Chapter.\u003c/a\u003e» ed è disponibile integralmente sul mio \u003ca href=\"https://www.flickr.com/photos/77058054@N07/\"\u003eprofilo Flickr\u003c/a\u003e. Tutte le fotografie sono pubblicate con \u003ca href=\"https://creativecommons.org/licenses/by-nc-sa/4.0/\"\u003elicenza CC BY-NC-SA\u003c/a\u003e.\u003c/p\u003e\n\u003cp\u003eQui trovate ogni fotografia con un piccolo commento!\u003c/p\u003e","title":"Un anno a Tübingen. Un capitolo. Una raccolta fotografica"},{"content":"Ciao, sono Luca!\nHo conseguito un Master in International Business and Entrepreneurship (M.Sc.) presso l’Università di Pavia, con un focus sul digitale, e un altro Master in European Management (M.Sc.) all’Università di Tübingen. Nella mia tesi ho analizzato cosa spinge le persone a scaricare determinate app dagli store digitali.\nIn precedenza, ho ottenuto la Laurea in Economia Aziendale (B.Sc.) all’Università di Brescia, con una tesi sull’industria IT e il passaggio dal software come prodotto al software come servizio.\nOggi faccio parte di IBM, dove aiuto i partner a valorizzare al massimo le tecnologie disponibili e a scoprire nuove opportunità di crescita.\nFuori dal lavoro, mi piace parlare di tecnologia, videogiochi, film, serie TV e, a volte, di politica. Per ricaricarmi adoro correre e andare in bici: non c’è niente di meglio di una lunga corsa o pedalata per liberare la mente.\nPer saperne di più, puoi dare un\u0026rsquo;occhiata al mio curriculum vitae riassuntivo e contattarmi attraverso la sezione Contatti.\n","permalink":"https://lucabontempi.com/it/about/","summary":"\u003cp\u003eCiao, sono Luca!\u003c/p\u003e\n\u003cp\u003eHo conseguito un Master in \u003ca href=\"https://mibe.cdl.unipv.it/en\"\u003eInternational Business and Entrepreneurship\u003c/a\u003e (M.Sc.) presso l’Università di Pavia, con un focus sul digitale, e un altro Master in \u003ca href=\"https://uni-tuebingen.de/de/33648\"\u003eEuropean Management\u003c/a\u003e (M.Sc.) all’Università di Tübingen. Nella mia tesi ho analizzato \u003ca href=\"https://lucabontempi.com/blog/determinants_of_download_on_mobile_app_stores/\" title=\"Determinants of Download on Mobile App Stores - An Empirical Analysis\"\u003ecosa spinge le persone a scaricare determinate app dagli store digitali\u003c/a\u003e.\u003c/p\u003e\n\u003cp\u003eIn precedenza, ho ottenuto la Laurea in Economia Aziendale (B.Sc.) all’Università di Brescia, con una tesi sull’\u003ca href=\"https://lucabontempi.com/il_mercato_informatico/\" title=\"Il mercato informatico: il software come prodotto e servizio\"\u003eindustria IT e il passaggio dal software come prodotto al software come servizio\u003c/a\u003e.\u003c/p\u003e","title":"About"},{"content":"","permalink":"https://lucabontempi.com/it/explore/","summary":"","title":"Esplora"},{"content":"Questa è la relazione finale con cui ho ottenuto la mia laurea triennale in Economia e Gestione Aziendale all\u0026rsquo;Università degli Studi di Brescia.\nCome potete intuire, la relazione tratta uno degli argomenti a cui sono più interessato: lo sviluppo e la commercializzazione dei software. La stesura di questo documento ha rappresentato per me un primo approccio a questi argomenti che non erano coperti nel mio corso di studi ed è stato un buon punto di partenza per la compresione delle dinamiche che muovono quel mercato. In quel momento ero quasi totalmente inconsapevole della letteratura passata sull\u0026rsquo;argomento, risultando superficiale in alcuni passaggi, e sicuramente alcuni argomenti necessitano di un ulteriore approfondimento.\nOltre al documento integrale (disponibile qui sotto), è anche possibile consultare la presentazione riassuntiva qui.\nBuona lettura!\n","permalink":"https://lucabontempi.com/il_mercato_informatico/","summary":"\u003cp\u003eQuesta è la relazione finale con cui ho ottenuto la mia laurea triennale in Economia e Gestione Aziendale all\u0026rsquo;\u003cstrong\u003eUniversità degli Studi di Brescia\u003c/strong\u003e.\u003c/p\u003e\n\u003cp\u003eCome potete intuire, la relazione tratta uno degli argomenti a cui sono più interessato: lo sviluppo e la commercializzazione dei software. La stesura di questo documento ha rappresentato per me un primo approccio a questi argomenti che non erano coperti nel mio corso di studi ed è stato un buon punto di partenza per la compresione delle dinamiche che muovono quel mercato. In quel momento ero quasi totalmente inconsapevole della letteratura passata sull\u0026rsquo;argomento, risultando superficiale in alcuni passaggi, e sicuramente alcuni argomenti necessitano di un ulteriore approfondimento.\u003c/p\u003e","title":"Il mercato informatico: il software come prodotto e servizio"}]