n markets for books music movies and other cultural products individuals often have information - PDF document

differences in academic and professionalachievement as a self-fulfilling prophecy.Since Merton, social scientists have con-sidered the possibility of self-fulfilling pro-phecies in a wide variety of domains, andthese studies can be classified according to theunit of analysis at which the self-fulfillingdynamics are thought to take place: individ-ual, dyadic, and collective (Biggs forthcom-ing). At the level of the individual, for exam-ple, a substantial body of work has exploredthe placebo effect on health outcomes; that is,whether, and under what conditions, individu-als experience improved health outcomes thatare attributable to receiving a specific sub-stance, but that are not due to the inherentpowers of the substance itself (Stewart-Williams and Podd 2004). Moving up to thelevel of the dyad, researchers have exploredthe effects of teacher expectations on studentoutcomes (Rosenthal and Jacobson 1968;Jussim and Harber 2005) and physician prog-nosis on patient health (Christakis 1999).Finally, at the collective level (i.e., situationsinvolving more than two individuals), sociolo-gists have explored the ÒperformativityÓ ofeconomic theory (Callon 1998), includingwhether the predictions of economic modelslead people to change their behavior in such away as to make the original prediction cometrue (MacKenzie and Millo 2003). A separatebody of work by economists, meanwhile, hasexplored self-fulfilling prophecies with regardto financial panics (Calomiris and Mason1997), investment bubbles (Garber 1989), andeven business cycles (Farmer 1999).At face value, these studies suggest thatover a wide range of scales and domains, thebelief in a particular outcome may indeedcause that outcome to be realized, even if thebelief itself was initially unfounded or evenfalse. Skeptics, however, remain unconvincedthat expectations or perceptions should betreated as fundamental elements of the socialenvironment, on par with individual prefer-ences (Rogerson 1995), and much of theempirical evidence that might persuade themis mixed or ambiguous (Biggs forthcoming).One reason that skeptics remain is that withobservational data alone it is virtually impos-sible to prove that any given real-world eventwas caused by the self-fulfillment of falsebeliefs. This difficulty arises because thestrongest support for such a claim wouldrequire comparing outcomes in the presenceor absence of false beliefs, but in almost allcases only one of these outcomes is observed(Holland 1986; Sobel 1996; Winship andMorgan 1999).Given these limits of observational data, itis no surprise that our best understanding ofself-fulfilling prophecies in cultural marketscomes from experimental and quasi-experi-mental methods. For example, by exploitingerrors in the construction of the New YorkTimesbestseller list, Sorensen (2007) foundthat books mistakenly omitted from the listhad fewer subsequent sales than a matched setof books that correctly appeared on the list.Hanson and Putler (1996), moreover, per-formed a field experiment in which theydirectly intervened in a real market by repeat- our site, participants could only download asong after listening to and rating it. However,they could listen to, rate, and download asmany or as few songs as they wished.Upon arrival to the website, subjects wererandomly assigned into one of a number ofexperimental groups. During an initial set-upphase, 2,211 participants were assigned to oneof two ÒworldsÓ4Ñindependent and socialinfluenceÑwhich differed in the availableinformation about the behavior of other par-ticipants that was presented with the songs. Inthe social influence world, the songs weresorted from most to least popular and accom-panied by the number of previous downloadsfor each song. In the independent world, how-ever, the songs were randomly reordered foreach participant and were not accompanied byany measure of popularity. Although the pres-ence or absence of download counts was notemphasized, the choices of participants in thesocial influence condition could clearly beinfluenced by the choices of previous partici-pants, whereas no such influence was possiblein the independent world. although there were two different conditions Unknown CitizensFalling Over1611BeerbongFather to Son1411The FastlaneTil Death do us Part (I donÕt)1412Evan GoldRobert Downey Jr.1312Ember SkyThis Upcoming Winter1313Miss OctoberPink Aggression1313Silent FilmAll I have to Say1213Stunt MonkeyInside Out1214Far from KnownRoute 91114Moral HazardWaste of my Life1116Nooner at NineWalk Away1017SibrianEye Patch1017Drawn in the SkyTap the Ride1018Art of KanlySeductive Intro, Melodic Breakdown1022Fading ThroughWish me Luck1023Benefit of a DoubtRun Away1026Salute the DawnI am Error1027Cape RenewalBaseball Warlock v11030Go MordecaiIt Does What Its Told1031The Broken PromiseThe End in Friend935SummerswastedA Plan Behind Destruction936SecretaryKeep Your Eyes on the Ballistics939The CalefactionTrapped in an Orange Peel945 that the download decision was indeedviewed by participants as a consequentialone. Moreover, the download decision wasstrongly related to participantsÕrating deci-sion; the more stars a participant gave a song,the more likely they were to download thatsong.At the time each subject participated,every song in his or her ÒworldÓ had a spe-cific download count and market rank (i.e.,the song with the most downloads had a mar-ket rank of 1). Figures 4aÐcplot the proba-bility that participants listened to the song ofa given market rank. These plots show thatparticipants who were aware of the behaviorof others were more likely to listen to songsthat they believed were more popular. Therewas also a slight reversal of this pattern forthe least popular songsÑfor example, a sub-ject in a social influence world was about sixtimes more likely to listen to the most popu-lar song and three times more likely to listento the least popular song, than to listen to asong of middle popularity rank. The tenden-cy for the subjects to favor the least popularsongs, which at first may appear surprising,is consistent with previous work (Salganik etal. 2006) and could simply be an artifact ofour experimental set-upÑjust as the top spoton a list is salient, so too is the bottom. quality signal regardless of their preferencefor listening to the same songs as others. Wedo not know for sure, however, because ourexperiment was not designed to differentiatebetween these two kinds of influence sincethis differentiation was not necessary to and reveals, for example, that the most popu-lar song during the set-up period was project-ed to finish at the same rank in the invertedand unchanged worlds; thus the impact ofinversion for this song was 0.8The secondmost popular song during the set-up period,however, was projected to finish higher in theunchanged world than in the inverted worlds(5th vs 20th and 23rd), so for this song theimpact of the inversion was negative (i.e., itwas hurt by the inversion). Overall, the finalrankings of almost all songs seem to be per-manently affected by the inversion, wheresongs that were promoted by the inversiontended to do better in the long run, and songsthat were initially demoted tended to doworse. Thus, in our experiment, the manipula-tion of market information, combined with aprocess of social influence, seemed to lead tolong-term changes in the popularity of songs.Whereas Figures 6 and 7 show the out-comes experienced by individual songs,Figure 8shows the outcome experienced bythe entire ÒmarketÓÑspecifically, it shows theSpearman rank correlation "(t)between popu-larity at the end of the set-up period and pop-ularity in the three social influence worlds as afunction of time.9During the set-up period (tothe left of the vertical line) the initial socialinfluence world quickly converged to anapproximate steady state, as evidenced by thecontinuing high value of "(t)!1 for theunchanged world (solid line) after the inver-sion (to the right of the vertical line). By con-trast, the two inverted worlds (dashed lines)started, by definition, with "(t)= Ð1, after the inversion to lock in is that the songs them-selves were of different quality, and that thesedifferences were more salient than the percep- 2003; Frith 2004). Fortunately, our experimen-tal design permits us to proceed withoutresolving these conceptual difficulties bymeasuring instead the intrinsic of thesongs to our pool of participants. Because thebehavior of the participants in the independent clusive. For example, unlike in our experimentwhere popularity was manipulated at a singletime and in a highly artificial manner (i.e.,total inversion), distortion in the real worldmay occur repeatedly, and may also exhibitconsiderable subtlety and variety. As anextreme example of manipulation, total inver-sion seems a natural first case to consider; butthere are of course many other possiblemanipulations that one could explore, even11The dilemma faced by the bands appears to be moresimilar to common-pool resource situations than publicgoods situations because the benefit that a band receivesmay be related to their proportion of the total contribu-tion, not just to the total contribution (Apesteguia andMaier-Rigaud 2006). However, this statement is hard tomake precisely because the payoff functions for the bands important effects on outcomes as has beendemonstrated by numerous scholars workingin the Òproduction of cultureÓ school (Hirsch1972; Peterson 1976; Frith 1978; DiMaggio2000; Peterson and Anand 2004; Dowd 2004).Precisely how our results would changeunder more realistic conditions is difficult topredict. We suspect, for example, that ourfinding that the highest appeal songs tend tosucceed regardless of interference may derivefrom the relatively small number of songs,which prevented the ÒbestÓ songs from escap-ing notice even in the inverted worlds. Thusthis finding may not generalize to more realis-tic scenarios in which the number of songs ismuch greater. Moreover, because we only per-formed one type of manipulation on one set ofsongs, it is unclear how our findings would beaffected either by less severe distortions or byusing a set of songs that are more (or less)similar in terms of appeal. Nor is it obvioushow the results would have differed had oursubjects been exposed to a stronger (or weak-er) form of social influence.In spite of these ambiguities, which wehope will be addressed with additional exper-iments or simulations, we believe that ourfindings are likely to have applicabilitybeyond the specific scope of the experimentitself, and thereby add to our general under-standing of self-fulfilling prophecies in cultur-al markets. We also believe this experimentmay have implications for experimental soci-ology and social psychology more generallyby showing the potential for web-based exper- a number which, to place in thecontext of traditional psychology experiments,is larger than the total enrollment of many uni-versities. Even larger experiments are practi-cal today, and likely to become increasingly soas web-related technology continues to devel-op. Although there are a number of importantissues to consider when conducting web-basedexperimentsÑsome of which are shared withlaboratory experiments, and some of whichare novelÑwe suspect that the ability to run led to some intriguing and even counterintu-itive insights, they have also been confoundedby the difficulty of reconciling models eitherwith micro-level or macro-level empiricaldata. At the micro-level, empirical difficultiesarise because social influence experiments arenot generally designed to differentiatebetween the different ÒrulesÓ governing indi-vidual behavior that are assumed, sometimesimplicitly, in various models (Lopez-Pintadoand Watts 2008). And at the macro-level,empirical verification is plagued by ambigui-ties of cause and effect; that is, very different Dowd, Timothy J. 2004. ÒProduction Perspectives inthe Sociology of Music.Ó Poetics32:235Ð46.Farmer, Robert E. A. 1999. Macroeconomics of Self-fulfilling Prophecies.Cambridge, MA: MITPress.Frith, Simon. 1978. The Sociology of Rock.London,UK: Constable.ÑÑÑ. 2004. What is bad music?Ó Pp 15Ð36 in BadMusic,edited by Christopher J. Washburne andMaiken Derno. New York: Routledge.Gans, Herbert. 1974. Popular Culture and High Influence.Glencoe, IL: Free Press.Kendall, Maurice and Jean Dickinson Gibbons. 1990.Rank Correlation Methods.London, UK:Edward Arnold.Kollock, Peter. 1998. The Economics ofSuperstars.Ó American Economic Review71(5):845Ð58.Rosenthal, Robert and Lenore Jacobson. 1968.Pygmalion in the Classroom: TeacherExpectations and PupilsÕIntellectualDevelopment.New York: Holt, Rinehart andWinston.Salganik, Matthew J. 2007. Success and Failure inCultural Markets, PhD thesis, Department ofSociology, Columbia University, New York.Salganik, Matthew J., Peter Sheridan Dodds, andDuncan J. Watts. 2006. ÒExperimental Study ofInequality and Unpredictability in an ArtificialCultural Market.Ó Science311:854Ð6.Schaller, Mark and Christian S. Crandall, eds. 2003.The Psychological Foundations of Culture.Mahwah, NJ: Lawrence Erlbaum.Schelling, Thomas C. 1978. Micromotives andMacrobehavior.New York: W. W. Norton.Senecal, Sylvain and Jacques Nantel. 2004. ÒTheInfluence of Online Product Recommendationson ConsumersÕOnline Choices.Ó Journal ofRetailing80:159Ð69.Sherif, Muzafer. 1937. ÒAn Experimental Approach tothe Study of Attitudes.Ó Sociometry1(1):90Ð8.Simmel, Georg. 1957. ÒFashion.