Bestseller Lists and Product Variety Author(S): Alan T

Home , Bestseller

Bestseller Lists and Product Variety Author(s): Alan T. Sorensen Source: The Journal of Industrial Economics, Vol. 55, No. 4 (Dec., 2007), pp. 715-738 Published by: Wiley Stable URL: http://www.jstor.org/stable/4622404 . Accessed: 06/03/2014 16:11

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp

. JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].

Wiley is collaborating with JSTOR to digitize, preserve and extend access to The Journal of Industrial Economics.

http://www.jstor.org

This content downloaded from 130.212.18.200 on Thu, 6 Mar 2014 16:11:23 PM All use subject to JSTOR Terms and Conditions THE JOURNALOF INDUSTRIALECONOMICS 0022-1821 VolumeLV December2007 No. 4

BESTSELLERLISTS AND PRODUCT VARIETY*

ALAN T. SORENSEN+

Thispaper uses detailedweekly data on salesof hardcoverfiction books to evaluatethe impactof the New YorkTimes bestseller list on salesand product variety. In order to circumvent the obvious problem of simultaneityof sales and bestsellerstatus, the analysis exploits time lags and accidental omissions in the construction of the list. The empiricalresults indicate that appearingon the list leads to a modest increase in sales for the average book, and that the effect is more dramaticfor bestsellersby debut authors.The paperdiscusses how the additionalconcentration of demandon top-sellingbooks couldlead to a reduction in the privately optimal number of books to publish. However,the data suggest the opposite is true: the marketexpansion effect of bestsellerlists appearsto dominateany businessstealing from non-bestsellingtitles.

I. INTRODUCTION

THEPERCEIVED IMPORTANCE OFBESTSELLER LISTS is a salient feature of multimedia industries.Weekly sales rankingsfor books of various genres are published in at least 40 different newspapers across the U.S., and making the list seems to be a benchmark of success for authors. In the movie industry, box office rankings are watched closely by movie studios and widely reportedin television and print media. Sales of music CDs are tracked and ranked by Billboard Magazine, whose weekly charts are prominently displayed in most retail music stores. Ostensibly, the purpose of bestseller lists is to simply report consumers' purchases. However, there are a number of reasons why the conspicuous publication of sales rankings may directly influence consumer behavior (in addition to merely reflecting it). Bestseller status may serve as a signal of quality: for example, bookstore patrons who are unfamiliar with a particular author may nevertheless buy her current bestseller, thinking that its popularity reflects other buyers' (favorable) information about the book's quality.1 Publicized sales rankings would also directly affect consumer

*Iam thankfulto the HooverInstitution, where much of thisresearch was conducted, and to Nielsen BookScanfor providingthe book salesdata. The researchhas benefittedfrom helpful conversationswith Jim King, Phillip Leslie, and Joel Sobel, among many others. Scott Rasmussenprovided excellent research assistance. Any errorsare mine.

tAuthor's affiliations:Graduate School of Business,Stanford University and NBER, 518 MemorialWay, Stanford,California, 94305-5015, U.S.A. e-mail: [email protected].

Thereis an extensiveliterature on quality-signalingwhen products' qualities are uncertain. See, for example,(Milgrom and Roberts[1986]). ? 2007 The Author. Journal compilation C 2007 Blackwell Publishing Ltd. and the Editorial Board of The Journal of Industrial Economics. Published by Blackwell Publishing Ltd, 9600 Garsington Road, Oxford OX4 2DQ, UK, and 350 Main Street, Maiden, MA 02148, USA. 715

This content downloaded from 130.212.18.200 on Thu, 6 Mar 2014 16:11:23 PM All use subject to JSTOR Terms and Conditions 716 ALANT. SORENSEN behaviorin the presenceof social effects, with bestsellerlists servingas a form of coordinatingmechanism.2 For example, teenagerswho want to listento musicthat is 'hot'can look to the Billboardcharts to find out whatis popular, and people may favor movies at the top of the box-office charts because they want to be conversantin popularculture. In the specificcase of books, bestsellerstatus also triggersadditional promotional activity by retailers. For the same reasons that bestsellerlists may directlyaffect consumers' purchasedecisions, they may also causesales to be morehighly concentrated on the few bestselling products. This, in turn, could influence product variety:if the additionalsales accruingto bestsellersas a directconsequence of the publicationof the list come at the expenseof non-bestsellingproducts, the optimal number of products to offer may decrease (relative to what would have been optimalin the absenceof a bestsellerlist). For example,if publicizedbox officerankings cause ticket sales to be more concentratedon blockbusters,they may also make it unprofitableto incurthe fixedcosts of producinga film whose popularityis expectedto be only marginal.3 This paper examinesthese issues in the context of the book publishing industry,looking specificallyat the impactof the New YorkTimes bestseller list on sales of hardcoverfiction titles. The empiricalanalysis addresses two questions. First, does being listed as a New YorkTimes bestseller cause an increasein sales?Second, does the influenceof the bestsellerlist also affect the number of books that are published, and if so, in which direction? Obvious simultaneity problems make answering the first question a nontrivialempirical exercise; however, subtletiesin the constructionand timingof the New YorkTimes list can be exploitedto identifyits impact.The resultssuggest a modestincrease in salesfor the typicalbestseller when it first appearson the list, and the increaseappears to reflectan informationaleffect rather than a promotional effect: the effects are concentratedin the first week a book appears on the list, and they are most substantialfor new authors. Regardingthe second question, the impact of bestsellerlists on product variety is theoreticallyambiguous, since it depends on whether marketexpansion or business-stealingeffects dominate. Although the data are less than ideal for addressingthis question,I presentindirect evidence suggestingthe business-stealingeffects of bestsellerlists are unimportant: if anything,bestseller lists appearto increasesales for both bestsellersand non-bestsellersin similargenres. Although this paperis the firstto explorethe impact of bestsellerlists on product variety, similar questions have been previously addressedin a numberof contexts.Early theoretical work emphasized the tradeoffbetween

2 See, e.g., (Beckerand Murphy[2000], Banerjee [1992], and Vettas[1997]). 3 This line of reasoningwill be clarifiedin sectionII.

? 2007 The Author. Journal compilation ? 2007 Blackwell Publishing Ltd. and the Editorial Board of The Journal of Industrial Economics.

This content downloaded from 130.212.18.200 on Thu, 6 Mar 2014 16:11:23 PM All use subject to JSTOR Terms and Conditions BESTSELLERLISTS AND PRODUCT VARIETY 717 quantityand diversityin the presenceof scale economies(Dixit and Stiglitz [1977])and the effects of market structureon product variety (Lancaster [1975]).Empirical studies of productvariety have been undertakenfor the radio broadcastingindustry (Berry and Waldfogel [2001], and Sweeting [2005]),the music industry(Alexander [1997]), and for retail eyeglasssales (Watson[2003]), to name a few examples. In the following section, I outline a basic theoretical framework for understandinghow bestsellerlists can influencethe numberof books that get publishedin equilibrium.Section 3 providesa briefdescription of the book industryand the dataset.The empiricalanalysis (Section 4) proceedsin two parts:first, the data areused to identifyand quantifythe directimpact of the bestsellerlist on sales;second, substitutionpatterns in the data are analyzed to determinethe likelydirection of the list'simpact on productvariety. Some broad implicationsof the results(as well as alternativeinterpretations) are discussedin the concludingsection.

II. A SIMPLE THEORETICAL FRAMEWORK Considera highly simplifiedmodel in which a singlepublisher chooses how many manuscriptsto publish.4There are K manuscriptsunder considera- tion. Printing and marketing a manuscript requires a fixed cost, F, in additionto the (constant)marginal printing cost c. Priorto publication,the manuscriptscan be rankedin orderof expectedpopularity, with r being the index of the rth-bestbook among the K alternatives.The marketprice of a publishedbook, p, is taken as given,and the post-publicationprice does not adjustto reflecta book's relativepopularity.5 The expecteddemand for the rth best book is given by D(K) . 8(r; K), where D(K) can be interpreted as the level of aggregatedemand for books (whichmay dependon how many are offered), and 0(r; K) is a function determininghow aggregatedemand is allocated among books depending on their relative popularity (e.g., we could have O(r;K) - 1). Only books with positive expected profits will be published,-_ so the numberof books will be the maximumK* such that (p - c)D(K*)O(K*;K*) > F, as illustrated in Figure 1. Now considerthe potential impact of publishinga bestsellerlist, so that consumersobserve the sales ranks of the top L books. Suppose that the

4 The modelcould also applyto movie studios'decisions about how manyfilms to produce, or to recordlabels' decisions about how manyartists to 5 sign. This assumptionwould be absurdin mostcontexts, but in thiscase it is at leastdescriptive of a curious practicein multimediamarkets. Prices of books, movies, and CD's almost never reflectthe popularityof the individualproducts. (Clerides [2002]) provides direct evidence and a thoroughdiscussion of pricingissues in the book industry.)Typically, 'price points' for books and CDs are determinedbefore they aremarketed, and subsequentadjustments are extremely infrequent.Movie ticketprices are even more rigid: in the summerof 2002, for example,it cost the sameamount to see 'Chicago'(which won the Oscarfor best pictureand was a box-office success)as 'BoatTrip' (which flopped at the box officeand was universallyridiculed by critics).

? 2007 The Author. Journal compilation ? 2007 Blackwell Publishing Ltd. and the Editorial Board of The Journal of Industrial Economics.

This content downloaded from 130.212.18.200 on Thu, 6 Mar 2014 16:11:23 PM All use subject to JSTOR Terms and Conditions 718 ALAN T. SORENSEN

F/(p-c)

L * K

L K**K*

Figure 1 The publish/no-publish margin aggregate consumer response to a bestseller list leads to an increased concentrationof sales on bestsellers:letting O(r;L,K) denote the expected marketshare of the rth-rankedbook among K alternativeswhen a bestseller list of lengthL is published,and O(r;0,K) denote the marketshare when no list is published, we assume that O(r;L,K) > ( < )O(r;0,K) if r < ( > )L. If bestseller lists have any direct impact on consumer behavior, the effect would almost certainlyhave this feature.The most obvious mechanismfor this effect is informational: if consumers are uncertain about books' qualities,and they believethat at least some past purchasershad meaningful informationabout the books they purchased,then bestsellerstatus would be a signal of quality.Alternatively, social effects may lead to higherdemand for bestsellers:consumers may want to readwhat everyoneelse is readingin the interestof keepingup with what is popular-e.g., they don't want to be left out of the conversationwhen they go to the cocktailparty. In the market for hardcoverfiction books, an additionalmechanism pushes sales toward bestsellers: retailers routinely discount bestsellers and position them prominentlyin their stores. If the overall level of demand D(K) is independentof any bestseller-list effects,then the list unambiguouslyreduces the numberof books that can be

? 2007 The Author. Journal compilation ? 2007 Blackwell Publishing Ltd. and the Editorial Board of The Journal of Industrial Economics.

This content downloaded from 130.212.18.200 on Thu, 6 Mar 2014 16:11:23 PM All use subject to JSTOR Terms and Conditions BESTSELLERLISTS AND PRODUCT VARIETY 719 D

I I

F/(p-c)K

L K* K**

Figure 2 List may increase overall level of demand profitablypublished. This is illustratedin Figure1: the increasedsales for the top L books come at the expense of the non-listed books, shifting the publish/no-publishmargin to the left (from K* to K**).More realistically, however,the publicationof a bestsellerlist could increaseoverall demand in addition to changing the allocation of demand across titles-i.e., the additional promotion and information about bestsellers could attract consumersthat otherwisewould not have purchasedany book at all. In this case, the impact of bestseller lists on the publish/no-publishmargin is ambiguous. (Figure 2 illustratesa case in which more books would get publishedin the presenceof a list, even though the list leads to a relatively higherconcentration of sales among bestsellers.) This framework,while obviously oversimplified,illustrates the principal ideas underlying the empirical analyses to follow. Ideally, we want to examinesales data to see if indeed O(r;L,K) > O(r;0,K) for r < L-that is, to see if bestsellerlists causean increasein demandfor bestsellersrelative to non-bestsellers.In orderto say anythingabout whetherbestseller lists affect the number of books that get published,we must then ask a much more subtle question of the data: how are sales of relatively unpopular (non- bestselling)books affectedby the publicationof bestsellerlists? That is, how ? 2007 The Author. Journal compilation ? 2007 Blackwell Publishing Ltd. and the Editorial Board of The Journal of Industrial Economics.

This content downloaded from 130.212.18.200 on Thu, 6 Mar 2014 16:11:23 PM All use subject to JSTOR Terms and Conditions 720 ALANT. SORENSEN does to for r close to O(r;L,K) compare 0(r;0,K) very K.6 Unfortunately(but not surprisingly),the available data are inadequate for answeringthis questiondirectly. Instead, we will look for indirectevidence of substitution between bestselling and non-bestsellingtitles, which would suggest the potentialfor (and the likely directionof) productvariety effects.7

III. BACKGROUNDAND DATA III (i). TheBook Industry In the U.S., the vast majorityof books are producedby a small numberof largepublishing houses like RandomHouse and HarperCollins.8 The odds againsta manuscript'sbeing acceptedby one of thesepublishing houses are long, especiallyin the case of fiction. Thirty per cent or fewer of available manuscriptsin any given year are in print, and althoughninety per cent of published books are nonfiction, seventy per cent of the manuscripts submitted to traditional publishers are fiction (Suzanne [1996]).9Most successfulmanuscripts are brokeredto publishersby literaryagents. These agentsare typically reluctant to take on first-timeauthors, and theirfees tend to be steep (around 15 per cent of authors' royalties).However, using an agent greatlyincreases the author'schances of success:unsolicited manuscripts (manuscriptsreceived 'over the transom') are estimated to have fifteen thousand to one odds against acceptance (Greco [1997]). Manu- scriptsare sometimessold to publishersby auction, but this method is the exceptionrather than the rule and is used primarilyby established,brand- name authors. The decisionto extenda contractto an authoris made only afterreview of the manuscript'squality and salability by several stages of editors. If a manuscriptsurvives the review process, the publisheroffers the author a contract granting royalty payments in exchange for exclusive marketing rights as long as the publisherkeeps the book in print. Royalties average seven percentof the wholesale price on hardcoverbooks by new authors,

6 Note that the illustrationin Figure1 makesit appearthat salesof the K'h-rankedbook and the L + ls'-rankedbook areequally affected, which need not be thecase. For instance,it is quite plausiblethat the impactis a decliningfunction of a book's rank.Moreover, only the effecton books at the marginis relevantfor determiningthe numberof books that get published. 7 Becausethis model does not explicitlyspecify consumers' preferences, it saysnothing about the welfareeffects of a reductionin productvariety. This paper will be deliberatelyagnostic on this point, sincethe welfareeffects could plausibly go eitherway: on the one hand,fewer books could mean foregone surplusfrom titles that would have appealedto readerswith diverse tastes; on the other hand, fewer books could mean that less time is wasted reading bad literature. 8 For thefirst quarter of 2003,the top six publishingconglomerates accounted for over80 per cent of unit sales in adult fiction. 9These numbers imply that 10 per cent of fictionmanuscripts get published.

? 2007 The Author. Journal compilation ? 2007 Blackwell Publishing Ltd. and the Editorial Board of The Journal of Industrial Economics.

This content downloaded from 130.212.18.200 on Thu, 6 Mar 2014 16:11:23 PM All use subject to JSTOR Terms and Conditions BESTSELLERLISTS AND PRODUCT VARIETY 721 and may increaseonce the book achievesa certainlevel of sales (Suzanne [1996]). Upon the delivery of a completed manuscripta large portion of expectedroyalties are given to the authorin the form of an advance;many authors never receiveadditional payments because advances often exceed royaltiesearned from actual sales. Publishers retain a largeshare of royalties in escrow accounts to compensate for returnedbooks; booksellers may returnunsold books to the publishersfor full price, and thereforereturn rates exceedingfifty per cent are common (Greco [1997]). In spite of the difficulty that authors seem to face in getting their manuscriptspublished, the book industrygenerates an astonishingflow of new books each year. Across all categories(fiction and nonfiction)and all formats (hardcover,trade paper, and mass-marketpaper), over 100,000 titles were publishedin the year 2000 alone. In adult fiction, the numberof new books published(called 'title output'within the industry)has increased dramaticallyover the past decade.The industry'strade publication, Bowker Annual,reports that title output for hardcoverfiction more than doubled from 1,962 in 1990 to 4,250 in 2000. In contrast, the rate of increasewas muchmore gradual prior to 1990.In fact, Bowkerreports that the numberof fiction titles in 1890was over 1,100, so title output had less than doubledin the 100 years prior to 1990 (Bogart[2001]). The dramatic increases in title output in the 1990's were roughly concomitant with an increase in the concentration of sales among bestsellers.From the mid-1980'sto the mid-1990's,the shareof total book sales representedby the top 30 sellers nearly doubled (Epstein [2001]).In 1994,over 70 percentof total fiction saleswere accountedfor by a merefive authors:John Grisham,Tom Clancy,Danielle Steel, Michael Crichton, and StephenKing (Greco [1997]).1o Publishersand authorsemploy a numberof marketingstrategies in their attemptsto achievethe kind of successenjoyed by these top-sellingauthors. Publishers'marketing budgets may range betweenten and twenty-fiveper cent of net sales(Cole [1999]),and authorsare expectedto appearpublicly in promotion tours. Book reviews are highly sought after but difficult to obtain: tens of thousands of books are publishedeach year in the United States, but the New York Times(for example) reviews only one per day. Publishers also may pay retail stores for shelf space or inclusion in promotionalmaterials. Book marketersconcentrate their efforts on creating a successfullaunch; retail stores may removelow-selling new releasesfrom their shelvesafter as little as one week (Greco [1997]).

10 Rosen [1981] provides a classic explanation for the presence of such 'superstars.' A critical piece of the argument is that the convexity of returns results from the imperfect substitutability of the products offered-i.e., reading several unremarkable books may not be a good substitute for reading a single great one.

? 2007 The Author. Journal compilation ? 2007 Blackwell Publishing Ltd. and the Editorial Board of The Journal of Industrial Economics.

This content downloaded from 130.212.18.200 on Thu, 6 Mar 2014 16:11:23 PM All use subject to JSTOR Terms and Conditions 722 ALANT. SORENSEN In spite of all the resourcesspent on marketing,one of the best kinds of publicity--appearing on a bestsellerlist-cannot be bought." Bestseller lists have long played an important role in the book industry. Regular publicationof bestsellerlists beganin 1895,when a literarymagazine called TheBookman started printing a monthlylist of the top six best-sellingbooks. The New YorkTimes Book Reviewbegan publishingits bestsellerlist as a regularfeature in 1942. Although many other prominentlists now exist,12 the New YorkTimes list is generallyconsidered the most influentialin the industry(Korda [2001]).

III (ii). Data The main dataset to be analyzedconsists of weekly national sales for over 1,200 hardcoverfiction titles that were releasedin 2001 or 2002. The sales data wereprovided by Nielsen BookScan,a marketresearch firm that tracks book sales using scannerdata from an almost-comprehensivepanel of retail booksellers.13Additional information about the individualtitles (such as the publication date, subject, and author information)was obtained from a variety of sources, includingAmazon.com and a volunteerwebsite called Overbooked.org.Table 1 reportssummary information for the books in the data, brokeninto three subsamples. The overallsample represents a relativelylarge fraction of the universeof hardcoverfiction titles releasedin this time period, though it is likely to be somewhatskewed toward popular books.14 Books that neversold morethan 50 copies in a singleweek (nationwide)were dropped from the sample,since their weekly sales numbersappeared to be mostly noise. Also, some books are excludedfrom the empiricalanalyses if theirrelease dates were difficult to determine.'"In suchcases, the numberof books is reducedto 799. As will

11Occasionally an authorhas testedthis propositionby purchasingnumerous copies of his own book, in hopes of pushingit onto a bestsellerlist. None of these attemptshas ever been trulysuccessful; the typicaloutcome has beenconsiderable embarrassment for the perpetrator once the schemewas uncovered. 12Most majorU.S. newspaperspublish their own local list (sometimesin additionto the New YorkTimes list), and a numberof national lists compete with the New YorkTimes list for attention (e.g., Publishers Weekly, Wall Street Journal, USA Today, and BookSense). 13BookScan collects data throughcooperative arrangements with virtuallyall the major bookstore chains, most major discount stores (like Costco), and most of the major online retailers(like Amazon.com).They claim to trackat least 80 per cent of total sales. 14 In constructingthe set of candidatebooks to track,we had to firstlocate a book (and its ISBN number)in orderto considerit. Obscure,slow-selling books are(by definition)harder to locate. 15 Threesources of informationwere used in determiningthe exactweek of release.For most titles,Amazon.com lists the exactday of release.If not, BookScanreports the monthof release, and 'eyeballing'the data usuallyreveals an obvious releasedate. If for any reasonthe release datewas not obvious-e.g., the Amazon.comand BookScanrelease dates didn't match, and/or the releasedate wasn't obvious from looking at the salesdata-the titlewas excludedfrom the samplewhenever the resultsmight be sensitiveto the accuracyof the releasedate.

? 2007 The Author. Journal compilation ? 2007 Blackwell Publishing Ltd. and the Editorial Board of The Journal of Industrial Economics.

This content downloaded from 130.212.18.200 on Thu, 6 Mar 2014 16:11:23 PM All use subject to JSTOR Terms and Conditions BESTSELLERLISTS AND PRODUCT VARIETY 723

TABLEI SUMMARYSTATISTICS All books with 26 + weeks of data (n = 1,217): Min Median Max Mean Std. Dev.

List price 10.95 24.95 68.85 24.28 2.63 Releaseweek* 1 57 83 49.64 24.18 New author 0 0 1 .15 .36 Sales, first 6 months 87 3,960 1,443,345 35,183 111,444 Max. one-week sales 50 486 360,133 6,903 24,453

Books with reliablerelease dates (n = 799): Min Median Max Mean Std. Dev.

List price 10.95 24.95 60 24.44 2.38 Releaseweek* 1 56 83 48.70 24.52 New author 0 0 1 .14 .35 Sales, first 6 months 87 8,063 1,443,345 51,396 134,445 Max. one-weeksales 50 972 360,133 10,227 29,626

New York Times bestsellers(n = 205): Min Median Max Mean Std. Dev.

List price 10.95 25 29.95 25.05 2.25 Releaseweek* 1 40 83 41.33 23.89 New author 0 0 1 .05 .22 Sales, first 6 months 20,963 93,507 1,443,345 174,987 222,530 Max. one-weeksales 2,108 18,438 360,133 36,047 50,058 Week firstappeared 4 4 34 4.5 2.4 Weeks on list 1 4 30 5.8 5.3

*Forrelease week, the firstweek of 2001is week 1, and the firstweek of 2002is week53. be shown in the next section, for nearlyall books the vast majorityof sales occur in the first 4-6 months, and subsequent sales are relatively uninformative.In light of this, only the first 26 weeks of sales are included in the samplefor any given book. Data on bestseller-liststatus come directlyfrom the New YorkTimes. The analysis focuses on the New York Times list because it is a nationally published list (and the sales data are national), and because it is almost universallyregarded as the most influentiallist in the industry.A majorityof retail booksellers (including online bookstores) have special sections devoted to New YorkTimes bestsellers and offer price discounts on these titles, and authorsare sometimesoffered bonuses for everyweek theirbook appears on the New York Times list. To constructits list, the New YorkTimes surveys nearly 4,000 bookstores eachweek, in additionto a numberof book wholesalerswho serveadditional types of booksellers(like supermarkets,newsstands, etc.) The reportedsales figures from these respondents are then extrapolated to a nationally representativeset of salesrankings using statistical weights. Because the New York Times list is constructed using sampling methods, it often makes 'mistakes'--i.e., books that should have made the list in a given week ? 2007 The Author. Journal compilation ? 2007 Blackwell Publishing Ltd. and the Editorial Board of The Journal of Industrial Economics.

This content downloaded from 130.212.18.200 on Thu, 6 Mar 2014 16:11:23 PM All use subject to JSTOR Terms and Conditions 724 ALANT. SORENSEN (because their sales exceeded the sales of the book listed at rank 15) are sometimes omitted, and the ordering of listed books sometimes doesn't reflectthe 'true'ranking of sales (as indicatedby the BookScandata). Also, assemblingthe list takes time, so the printedbestseller list reflectsrankings from three weeks prior. Both of these features of the New York Times list-the mistakesand the time lags-are criticalin the empiricalanalysis of its impact on sales.

IV. EMPIRICAL ANALYSIS AND RESULTS IV (i). Skewnessin Book Sales The most striking pattern in the data is that book sales are remarkably skewedin two importantways. First,the distributionof salesacross books is heavilyskewed. Figure 3 plots total salesin the firstsix monthsagainst sales rankfor the top 100books in the sample.Even when looking only at the top decile of books, the skewness of the distributionis striking.Of the 1,217 books for which at least 26 weeks were observed,16the top 12 (1 per cent) accountfor 25 per cent of total six-monthsales, and the top 43 (3.5 percent) accountfor 50 per cent. The 205 books that made it to the New YorkTimes bestsellerlist account for 84 per cent of total sales in the sample.The most popular book in the sample, SkippingChristmas by John Grisham, sold more copies in its first three weeks than did the bottom 368 books in their first six months combined. Second, book sales tend to be skewedwith respectto time for any given title:that is, salestend to be heavilyconcentrated in the firstfew weeksafter a book's release.Of the 1,217books in the samplefor which26 or moreweeks are observed,898 (73.8 percent) hit theirsales peak sometimein the firstfour weeks. The median 'peak week' is week 2. Somewhat surprisingly,this pattern also seems to hold for debut authors, for which one might expect gradualdiffusion of informationand thereforean S-shapedsales path. For new authors, 112 of 182 books (61.5 per cent) peak in the first4 weeks, and the medianpeak week is 4. This second form of skewness is important to keep in mind when interpretingthe models and results in the following sections. The steady decayof a book's salesover time is the dominantpattern: with the exception of seasonaleffects (e.g., Christmas),any otherchanges in a book's salestend to be second-orderrelative to this decay trend.17Essentially, the sales paths

16 Books for which the release date was questionable are not excluded here, since getting the release date right isn't critical for this exercise. 17Decay-like sales patterns are not unique to the book industry; for example, Einav [2003] reports similar patterns for movie box office sales and incorporates exponential decay directly into his empirical model.

? 2007 The Author. Journal compilation ? 2007 Blackwell Publishing Ltd. and the Editorial Board of The Journal of Industrial Economics.

This content downloaded from 130.212.18.200 on Thu, 6 Mar 2014 16:11:23 PM All use subject to JSTOR Terms and Conditions BESTSELLERLISTS AND PRODUCT VARIETY 725

1500

U) c. 0o E c( 1000

C.)

-. c 500. cl)

1 100 Overallsales rank

Figure 3 Skewness of book sales typicallyresemble exponential decay patterns.Eyeballing the time path of sales for all the books in the sample,one rarelyobserves a book's sales 'take off' after it hits the bestsellerlist; if anything, making the list appears to temporarilyslow the pace of decline.

IV (ii). Do Bestseller Lists Directly Affect Sales? Theoryclearly suggests that bestsellerlists may do more than simplyreflect consumerbehavior: the lists may directlyinfluence consumer behavior, so that a book's appearanceon the bestsellerlist has an independenteffect on its sales. However,measuring such an effect is a difficultempirical problem: the set of books that receivethe 'treatment'of being listed as bestsellersis clearlynot random,and a naiveempirical approach would likely confuse the directionof causality.There is obviously a correlationbetween the level of sales and bestsellerstatus (by the very definitionof a bestsellerlist), but we cannot infer from this correlationthat being listed as a bestsellercauses highersales. However,given the availabledata and the subtletiesin the constructionof the New York Times list, there are at least two ways we can attempt to identify the list's direct influence.One strategyis to exploit the so-called 'mistakes'that are sometimesmade in the list. As mentionedpreviously, the

? 2007 The Author. Journal compilation ? 2007 Blackwell Publishing Ltd. and the Editorial Board of The Journal of Industrial Economics.

This content downloaded from 130.212.18.200 on Thu, 6 Mar 2014 16:11:23 PM All use subject to JSTOR Terms and Conditions 726 ALAN T. SORENSEN

TABLE 2 CHARACTERISTICS OF LISTED BOOKS VS. OMITTED BOOKS

Per cent of 'mistakes' Per cent of bestsellers t-stat Genre listing this genre (N = 75) listing this genre* (N = 44) (p-value)

Literature/Fiction 51 64 - 1.39 (0.17) Mystery/Thriller 51 36 1.53 (0.13) Romance 21 11 1.47 (0.14) Science Fiction 12 7 0.96 (0.34) Religion 1 0 1.00 (0.32) Horror 3 2 0.13 (0.89)

Average list price: 24.98 24.54 1.09 (0.28)

*Only books that were ranked 13-15 when they first appeared on the bestseller list are included in the bestsellers group. Numbers are based on books' genres as listed on Amazon.com. Percentages add to more than 100 because books typically list more than one genre. Prices are publishers' list prices, from BookScan. processused to generatethe New YorkTimes list is inexact.Although the list is by and largequite accuratewhen comparedwith the 'true'sales numbers availablefrom BookScan,it is not uncommonfor a bestsellingbook to be missed-i.e., a book may not appear on the list even though its sales exceeded the sales of listed books. In principle,these mistakes provide a means of identifyingthe effect of appearingon the list, by serving as an appropriatecontrol group. (Comparinglisted books to unlistedbooks is a bad experiment,since whether a book is listed is a nearly deterministic function of the dependentvariable; but comparinglisted books to books that shouldhave been listed is, in principle,a valid experiment-as long as the 'mistakes'are randomoccurrences.) During the years 2001-2002, there were 182 instances in which a hardcoverfiction book was not listed as a New YorkTimes bestseller when in fact it should have been,'8 representing109 differentbooks. (In several cases,there were multiple weeks in whicha book shouldhave been on the list but was not.) The majorityof these (roughly70%) werenarrow misses: had the books been listed, they would have been ranked 13-15 on the list. In orderto constructa faircomparison, I focus on two sets of books:those that were listed at rank 13, 14, or 15 when they first appearedon the New York Timeslist (n = 44), and those that shouldhave appeared at 13, 14,or 15when they weremistakenly omitted (n = 75). Table2 summarizessome observable characteristicsof the books in the two groups. If the omissions were not random, but ratheran attempt by the New York Timesat 'editorializing' the list, we might expect to see a differentsubject composition among the omitted books. The distributionof subjects and list prices appearsto be mostly similarbetween the two groups, lending some confidencethat the

18To be precise,I will say a book shouldhave been listed if, for the weekthat was relevantfor generatingthe list, the book'ssales (according to BookScan)exceeded the salesof the book that was listed at #15.

? 2007 The Author. Journal compilation ? 2007 Blackwell Publishing Ltd. and the Editorial Board of The Journal of Industrial Economics.

This content downloaded from 130.212.18.200 on Thu, 6 Mar 2014 16:11:23 PM All use subject to JSTOR Terms and Conditions BESTSELLERLISTS AND PRODUCT VARIETY 727

TABLE3 COMPARISONOF SALES CHANGES: LISTED BOOKS VS. OMITTEDBOOKS Mean Std. Dev. % A sales, firstweek appearingon bestsellerlist (n = 44) - .076 .702 % A sales, week in which mistakenlyomitted (n = 75) - .227 .157

Difference = .151. One-tailed t-test for Ho: difference is zero: t = 1.404, p = .084. omissionsare indeed random mistakes. The only notabledifferences are that the genre'Literature & Fiction'is morelikely (and 'Romance'or 'Mystery& Thrillers'less likely) among books making the list than among omitted books, but thesedifferences are not statisticallysignificant. Moreover, when consideringthe overallcomposition of the list (notjust positions 13-15), the proportionsof Romance and Mysterynovels match almost perfectlywith the proportionsamong the omitted books, so it seems clear the omissions were not the resultof any bias againstparticular genres. Table 3 reportsa comparisonof sales for the two groups. For books that werepublished for the firsttime on the New YorkTimes list, salesdeclined by an averageof 7.6 per cent relativeto the previousweek.19 For mistakenly omittedbooks, sales declinedby an averageof 22.7 per cent. Taken at face value, the differenceimplies that in the firstweek, beinglisted leads to 19per cent more sales than would have otherwise occurred. However, the differenceis statisticallyimprecise: for significancelevels less than .08, we would fail to rejectthe null hypothesisthat the list has no effect using a one- tailed test. The comparisonreported in Table 3 does not control for any covariates,but doing so has very little effect on the estimate.Using simple linear regressions to control for seasonal effects and time-since-release effects yields estimatesof the same approximatemagnitude. Given the availabilityof panel data on book sales, a second strategyfor identifyingthe effect of bestsellerlists is to use all the availabledata (not just 'mistake'books) to measurethe week-by-weekchanges in sales associated with changesin bestsellerstatus. Whereas the resultsin Table 3 are basedon comparisonsacross books (with mistakenlyomitted books serving as the control groupfor the listedbooks), this secondapproach is basedon within- book variationover time (so the control is effectivelythe same book's sales trajectoryprior to its appearanceon the list). Observingsales over several weeks for each book makesit possibleto control for book fixedeffects, thus absorbingthe obviouslyendogenous differences in sales levels for bestseller vs. non-bestsellertitles. Moreover, the time lag involved in the New York Timeslist meansthat we have at least three'pre-treatment' observations on

19 Recall that the dominant pattern in sales over time is a steady decay: even for books appearing on the bestseller list, it is rare for sales to increase from one week to the next.

? 2007 The Author. Journal compilation ? 2007 Blackwell Publishing Ltd. and the Editorial Board of The Journal of Industrial Economics.

This content downloaded from 130.212.18.200 on Thu, 6 Mar 2014 16:11:23 PM All use subject to JSTOR Terms and Conditions 728 ALANT. SORENSEN each bestsellerbefore it hits the list. (Due to the time lag, the soonest a book can appear on the list is at the beginningof its fourth week on bookstore shelves.)In principle,therefore, the datacan be usedto estimatea modelthat controls for book-specificdifferences in the level of sales and book-specific differencesin sales trends observedprior to appearanceon the bestseller list.20 An empirical model of book sales over time must accommodate the dominantdecay trendin sales over time (as describedin section IV(i)) and allow for appearanceson the bestsellerlist to directlyaffect sales. Moreover, goodness of fit is even more important than usual here, since the counterfactualwe want from the model (how many units a book would have sold if it hadn't appearedon the bestsellerlist in a certain week) is essentiallya forecast.Given theseconsiderations, one simpleapproach is to model book sales as an autoregressiveprocess in which the autoregression parameteris a function of covariates:

(1) ?i,sit = + +itsalesit-1 8it,

?it =- Xlt withXi, a set of covariatesfor book i in week t (includingbestseller status) and et, a demand shock that is independent(but not necessarilyidentically distributed)across i and t. The autoregressivestructure is appealingin part becauseit trackssales quite well, so thatthe predictionin anyperiod will never be off by much. Moreover, this specificationfocuses the estimation on changesin sales for a given book ratherthan differencesin the level of sales acrossbooks, and allowsfor book-specificdifferences in the rate of decayin sales (via book-specificconstants in it,). In other words, the model can accommodateunobserved, book-specific heterogeneity in both the level and trajectoryof sales.In orderfor any remainingendogeneity to be importantit musttake a peculiarform. In particular,the timingof a book'sappearance on the bestsellerlist would haveto correspondto weeksof idiosyncraticallyhigh demand in order to bias our estimate of the list's impact. Given that the currentlist reflectssales from three weeks prior, such endogenoustiming seemsvery unlikely. Although publishers are surely strategic about when they releasebooks, it seemssafe to presumethey don't systematically release books exactlythree weeks in advanceof anticipatedpositive demand shocks.21

20 Though implementedsomewhat differently, the empiricalstrategies employed here are similarin spiritto the analysisof Reinsteinand Snyder[2000], who exploitquirks in the timing of movie critics'reviews to identifytheir impact on box office sales. 21 Onepotential caveat is that publishersmay strategicallyavoid releasing books at the same time as major blockbusters,whose release dates are announcedwell in advance. (This is discussedin more detail in section4.3 below.) However,even in this case it is not clearwhy publisherswould time their releases such that the positiveshock comes exactly three weeks after release.

? 2007 The Author. Journal compilation ? 2007 Blackwell Publishing Ltd. and the Editorial Board of The Journal of Industrial Economics.

This content downloaded from 130.212.18.200 on Thu, 6 Mar 2014 16:11:23 PM All use subject to JSTOR Terms and Conditions BESTSELLERLISTS AND PRODUCT VARIETY 729

TABLE4 REGRESSION ESTIMATES: BOOK SALES AND LIST STATUS

I II III IV

NYT-listed .043 .030 (.015) (.019) NYT, week 1 .087 (.025) NYT, week 2+ .011 (.020) NYT, 1s"week off - .002 (.033) NYT-listed x rank .004 (.004) NYT-listed x [1-5] .042 (.016) NYT-listed x [6-10] .040 (.027) NYT-listed x [11-15] .068 (.053) Oprah 6.356 6.286 6.344 6.346 (.111) (.111) (.114) (.112) GMA .108 .100 .118 .109 (.085) (.083) (.086) (.085) Weeks out -.064 -.055 -.066 -.065 (.022) (.021) (.024) (.023)

R2 .977 .978 .977 .977 # books 799 799 799 799 # observations 19,024 19,024 19,024 19,024

Robust standard errors in parentheses. All explanatory variables are interacted with lagged sales, as explained in the text, and each specification includes week fixed effects and book fixed effects. The number of observations does not equal the number of books times 25 because some weeks are missing.

Table 4 reportsestimates of this model for four separatespecifications, usingweeks 2-26 for each book in the sample.All four specificationsinclude a full set of week dummiesto control for seasonal variationin book sales (thereis a large increasein sales in mid to late December,for example),as well as a full set of book dummiesin Column I reports the estimated coefficienton an indicatorthat equals onei/. for everyweek in which the book appearedon the New York Times bestsellerlist. The point estimates are statisticallysignificant at the 5 per cent level, and imply that sales decline about 4 percentagepoints more slowly when a book is listed as a bestseller. Given the many theoreticalreasons why appearanceson a bestsellerlist shouldhave a directimpact on sales,it is not surprisingto find a statistically significanteffect. The moreinteresting question is perhapswhy the list has an impact-i.e., what is the relevant mechanismby which list appearances boost sales?One class of explanationsis based on the idea that bestseller status is informative:for example,bestseller appearances could signal high quality,or signalwhat otherpeople arereading (and therebyfacilitate social coordination). Another possibility is that sales increases result from promotionalactivities undertaken for bestsellers(e.g., prominentplacement on bookstore shelves). A rathermundane explanationwould be that the

? 2007 The Author. Journal compilation ? 2007 Blackwell Publishing Ltd. and the Editorial Board of The Journal of Industrial Economics.

This content downloaded from 130.212.18.200 on Thu, 6 Mar 2014 16:11:23 PM All use subject to JSTOR Terms and Conditions 730 ALANT. SORENSEN increasesare merely price effects, resulting from the automaticdiscount that many stores offer on currentbestsellers. The specificationsreported in columnsII-IV of Table 4 attemptto shed light on the source of the bestsellerlist's impact, and specificallytry to distinguishwhether information or promotionis drivingthe effects.Column II separates the effect of a list appearancebetween the first week and subsequentweeks on the list. If the salesincreases are drivenby information, we shouldexpect them to be concentratedin weekone, whenthe information is new to consumers.On the other hand, if sales increasesmostly reflect promotional activities,then the effects should be apparentas long as the book remains on the list. The estimates are more consistent with the information-shock explanation: they indicate an 8.7 percentage-point change in the week the book first appears(statistically significant at the 5 per cent level), with any effects in subsequentweeks (combined) being statisticallyindistinguishable from zero. Moreover,a book's removalfrom the list (which, accordingto the bookstore managerswith whom I spoke, abruptlyterminates any specialin-store promotional activity for the book) has little impacton its sales:the coefficienton the dummyfor 'firstweek off is neither statisticallynor economicallysignificant. Given this result, it is very unlikelythat the measuredimpact of bestsellerstatus represents a pure price effect. Just as informationaleffects should be concentratedin the week of a book's first appearanceon the list, they may also be concentratedamong books at the top of the list. Thereis obviously a big differencein the buzz generated by a number-one bestseller than by a book barely breaking through to the bottom of the list, and social effects in consumption are presumablystrongest for the books at the verytop. By contrast,the effectsof in-store promotions should be equally importantfor all books on the list, because all of them are put on a special rack, and/or all of them are discounted.Columns III and IV of Table4 reportestimates from regressions in which the impactof a list appearanceis allowedto dependon the book's position on the list. Based on these estimates,there is no evidencethat the list's impactis largerfor books at the top of the list:the interactionof the list dummy with list rank is statistically indistinguishablefrom zero in specificationIII, and the coefficientson list dummiesfor separategroups (books 1-5 on the list, books 6-10, and books 11-15) are statistically indistinguishable in specification IV (F2,18114= 0.13, p-value = 0.88). If anything,the point estimatessuggest larger effects for books at the bottom of the bestsellerlist. In interpreting these results, it is important to remember that the information content of an appearanceon the list is quite low for some authors-everyone knowsthat Grisham'snewest novel will makethe list, so there is no information shock when it first appears-and the estimates in Table 4 report the average effect over all books that made the list. Indeed,

? 2007 The Author. Journal compilation ? 2007 Blackwell Publishing Ltd. and the Editorial Board of The Journal of Industrial Economics.

This content downloaded from 130.212.18.200 on Thu, 6 Mar 2014 16:11:23 PM All use subject to JSTOR Terms and Conditions BESTSELLERLISTS AND PRODUCTVARIETY 731 it could be that for most books, appearingon the list has virtuallyno effect on sales;only appearancesthat are surprisesmake a difference.The results reportedin columns III and IV of the table, suggestingthat appearancesat the top of the list have no more impact on sales than appearancesat the bottom of the list, may simply reflectthat most top-rankedbestsellers are books that were expectedto be there. In orderto examinethis issue more carefully,the model was re-estimated allowingfor book-specificheterogeneity in the effect of beinglisted. That is, the model was estimatedwith

(2) = + X~ /it fiNEWBSit t, where NEWBSit is an indicator equal to one if book i appeared on the bestsellerlist for the first time in week t. The most strikingfeature of the estimated#P's is how they relate to the authors' histories. Figure 4 shows kernel densities22for the estimated fi's separately for well-established authors vs. newer authors.23For establishedauthors, the distributionis centeredat - 0.02 and is approximatelysymmetric. For authorswho were less well-established,the distributionhas a positive mode and a long right tail, whichis consistentwith the idea that bestsellerstatus has an appreciable impactonly for somebooks. Perhapsthe most interestingresult relates to the small sampleof eleven debutauthors who made the list. For these authors, the estimated r;'sare consistently large and positive:of the eleven,nine are in the top quartile, and seven are in the top decile. The average value of I3iamongthese new authorsis 0.35 (comparedwith 0.05 for the remainderof the authors).Although the samplesize is obviouslysmall here, the patterns areclearly consistent with the hypothesisthat appearingon the bestsellerlist only has an impactwhen the appearanceis informative. The heterogeneityevident in Figure4 is importantto keep in mind when interpretingthe quantitativeresults from Table 4. Taken at face value, the estimatedcoefficients imply a modest averageeffect of list appearanceson sales, even though the specifiedautoregressive model allows the effect to persist. For example, comparinga book that appearedonly once on the bestsellerlist (in its fourthweek from release)to a book that startedwith the same initial sales but neverappeared on the list, and assuminga constant

22 Two outlierswere omitted in the estimationof the density.One, an extremenegative value, was for a book that was announcedas a televisionbook club pick on the same day it first appearedon the bestsellerlist, so the data have difficultydistinguishing the two effects. The other, an extremepositive value, was for a book with a 'mysteryauthor.' TheDiary of Ellen Rimbauerwas nominallywritten by a 'new'(but fictitious) author named Joyce Reardon, and it was rumoredthat the book was actuallywritten by Steven King. Sales spiked considerably when the book firsthit the bestsellerlist, so its estimatedfli is very large. 23 The groups of authorswere split based on the numberof past adult fiction titles each authorhad published.The top third(having published 35 or more titles)were designated the 'established'authors, and the remainingtwo thirdswere called the 'newer'authors.

? 2007 The Author. Journal compilation ? 2007 Blackwell Publishing Ltd. and the Editorial Board of The Journal of Industrial Economics.

This content downloaded from 130.212.18.200 on Thu, 6 Mar 2014 16:11:23 PM All use subject to JSTOR Terms and Conditions 732 ALAN T. SORENSEN

/i\

! \

c/ \ / / \ kBeta /\

0= .\ -.4 Bet.8

established authors newer authors -- -

Figure 4 Heterogeneity in the impact of bestseller list on sales equalto 0.7, the 8 percentage-pointincrease in week4 translatesto a 13.7per cent differencein expected sales over the first 52 weeks.24However, the averageeffects seem to be maskingthe real story:for establishedbestselling authors, the list has no discernibleimpact on sales; but for new authors, appearingon the list has a relativelydramatic impact. (A one-timeincrease in ) of 0.35 in week 4 would lead to a 57 per cent increasein sales over the following 52 weeks.) Note that even at theirlargest, the effectsof bestsellerlist appearancesare smallwhen comparedto the effectsof the announcementvariables included in the regressionsreported in Table4. The 'Oprah'indicator is equalto one if the book was announcedas a selectionfor OprahWinfrey's book clubin that week, and the 'GMA' indicatoris equalto one if the book was announcedas the pick for the 'Good MorningAmerica' show's book club. The estimated effectsof these announcementsare very large-they dwarfany effect of the bestsellerlist. The influenceof the announcementscould derivefrom various

24 If So is initial sales, then total sales over Tweeks is So ECT'•1'. If increases by an amount A in week r (but then reverts to its previous value and is otherwise constant), then the percentage increasein sales is equal to A ;3t-

? 2007 The Author. Journal compilationET-"_1 ? 2007 Blackwell Publishing Ltd. and the Editorial Board of The Journal of Industrial Economics.

This content downloaded from 130.212.18.200 on Thu, 6 Mar 2014 16:11:23 PM All use subject to JSTOR Terms and Conditions BESTSELLERLISTS AND PRODUCT VARIETY 733 sources:the announcementscould simplyalert a relativelylarge numberof consumersto the existenceof the book; they could serveas a qualitysignal; or they could act as a coordinationmechanism by which a large number of consumersagree to read the same book. (The latter mechanismis the apparentobjective of the televisionbook clubs.) Finally,although the specificationsreported in Table4 shouldadequately control for unobservedheterogeneity related to books' varying levels of sales, it is neverthelessuseful to construct a reality check for the key estimated coefficients. For example, instead of regressing sales on an interactionof laggedsales and an indicatorfor 'firstweek on the list,' we can interact lagged sales with an indicator for 'first week almost on the list,' definedas any unlistedbook with sales greaterthan 90 per cent of the 15th- rankedbestseller. If the coefficientsreported in Table4 aremerely an artifact of latentheterogeneity in salesdynamics that is correlatedwith differences in the level of sales, then the coefficienton this variablewould be positiveand significant.In fact, runningthis regressionyields a coefficientestimate of 0.028with a standarderror of .018,lending some additionalcredibility to the resultsreported in the table.

IV (iii). Do Bestseller Lists Affect Product Variety? The resultsof the previoussection indicatethat some books appearingon the New YorkTimes bestseller list enjoya consequentincrease in sales.But to what extent are those extra sales 'stolen'from non-bestsellingtitles? As was explained in section II, whether (and in which direction) a bestsellerlist influencesproduct variety depends on whetherthe list has any impact on books nearthe publish/no-publishmargin. If the consumerwho buys a book becausehe saw it on the bestsellerlist would otherwisehave bought a book near the marginof profitability,then one can arguethat the list reducesthe privately optimal number of books by furtherconcentrating demand on bestsellers. However, it is also possible that the consumer would have otherwisebought no book at all-in that case, sales are more concentrated on bestsellers,but this comes at no expenseto non-bestsellingtitles, and the numberof books publishedis unaffected.Moreover, it is also possible that bestsellerlists increasedemand for all books, for exampleby simplybringing moreconsumers into the bookstore.Bestsellers and non-bestsellerscould, in principle,be complementarygoods if consumersbuy multiplebooks when they visit the bookstore. The data availableare clearlyinsufficient to providea conclusiveanswer to this question.The ideal experimentmight be one in which a large set of consumersmakes purchasesin the presenceof a bestsellerlist, and another set of consumers makes purchases without having any exposure to the bestseller list (either through the media or at the retail outlet itself).

O 2007 The Author. Journal compilation ? 2007 Blackwell Publishing Ltd. and the Editorial Board of The Journal of Industrial Economics.

This content downloaded from 130.212.18.200 on Thu, 6 Mar 2014 16:11:23 PM All use subject to JSTOR Terms and Conditions 734 ALANT. SORENSEN

TABLE5 SUMMARYOF LISTED GENRES Per cent of books Per cent of bestsellers Genre listing this genre listing this genre Literature/Fiction 68 44 Mystery/Thriller 39 56 Romance 11 19 ScienceFiction 6 9 Religion 2 3 Horror 2 3

Basedon books'genres as listedon Amazon.com.Percentages add to morethan 100 because books typically list morethan one genre.

The ubiquityof the New YorkTimes list makesit virtuallyimpossible to find any group of book purchasersthat resemblessuch a control group. What the available data can potentially reveal is indirect evidence of substitutionbetween bestsellersand non-bestsellers.Even this is difficult, however, since the data contain no price variation that would enable estimation of cross-priceelasticities. The only useable variation is time variation in the subject composition of the bestsellerlist. If substitution betweennon-bestsellers and bestsellersis important,then-presuming that books in the same genre are closer substitutes than books in different genres-sales of non-bestsellingbooks should decline when the bestseller list is comprisedof books in similargenres. For example, sales of a non- bestsellingdetective novel would decreasein a week when three detective novels simultaneouslyhit the bestsellerlist. In order to capture these kinds of substitution patterns, a variable summarizingeach book's similarityto the current set of bestsellerswas constructedby comparingthe book's genre(s)(as listedby Amazon.com)to the genresof all books on the currentbestseller list. Specifically,the pairwise similaritybetween books A and B is definedas 2 x (# of genresshared by booksA andB) (3) sim(A,B) = (numberof genreslisted for A) + (numberof genreslisted for B)" Thismeasure is equalto one if the two books' genresare identical, and zero if there is no overlap at all.25Book A's 'average subject similarity'to the current bestseller list is then computed as 1/15 ~E15sim(A, r), where r indexesthe currentbestsellers by rank.Table 5 showsthe relativefrequencies of the genresin the samplefor all books and for bestsellersonly. As is clear from the table, mysteriesand thrillersare the most commonbestsellers, and

25 Amazon.comoften lists multiplegenres for the same book: e.g., 'Contemporaryfiction' and 'Romance.'So sim(A,B)will be less than one unlessbooks A and B list exactlythe same genres.

? 2007 The Author. Journal compilation ? 2007 Blackwell Publishing Ltd. and the Editorial Board of The Journal of Industrial Economics.

This content downloaded from 130.212.18.200 on Thu, 6 Mar 2014 16:11:23 PM All use subject to JSTOR Terms and Conditions BESTSELLERLISTS AND PRODUCTVARIETY 735 week==53 week==54 week==55 week==56 week==57 week==58 O Literature& Fiction Q Mystery&Thrillers • O Romance week==59 week==60 week==61 week==62 week==63 week==64 O ScienceFiction

week==71 week==72 week==73 week==74 week==75 week==76

week==77 week==78 week==79 week==80 week==81 week==82

week==83 week==84 week==85 week==86 week==87 week==88

Figure5 Variationin bestsellerlist compositionover time

romance novels are represented more heavily among bestsellers than among books overall. Importantly, there is substantial variation in the composition of the bestseller list over time. Figure 5 shows a series of star graphs to illustrate variation in list composition for a sample 36-week period. To the extent that there is meaningful substitution between non-bestsellers and bestsellers, we should observe that sales respond to changes in the average similarity variable induced by variation in the list composition. Table 6 reports estimates of a model analogous to the autoregressive model of the previous section, but with similarity measures included in the 1,i.26 Instead of a business-stealing effect, the estimated coefficients seem to suggest a complementarity between bestsellers and non-bestsellers. Weeks in which the genre-composition of the bestseller list is close to a non-bestselling

26 Becausethe purposeis to evaluatethe potentialresponse of non-bestsellersto changesin their similaritywith bestsellers,only books that nevermade the bestsellerlist are used in the estimation.The table reports specifications with and without subject fixed effects; without these fixedeffects, the coefficientson the similaritymeasures could partlyreflect average differences in the prevalanceof certaingenres on the bestsellerlist, whichwould obviouslybe correlated with sales in that genre.

? 2007 The Author. Journal compilation ? 2007 Blackwell Publishing Ltd. and the Editorial Board of The Journal of Industrial Economics.

This content downloaded from 130.212.18.200 on Thu, 6 Mar 2014 16:11:23 PM All use subject to JSTOR Terms and Conditions 736 ALAN T. SORENSEN

TABLE6 REGRESSION ESTIMATES: BOOK SALES AND SIMILARITY TO CURRENT BESTSELLERS

I II III IV

Avg. Similarity: All bestsellers .044 - .228 (.033) (.080) New on list .232 .400 (.100) (.102) Weeks out .005 .005 .003 .002 (.002) (.002) (.002) (.002) Week dummies? yes yes yes yes Subject dummies? no no yes yes R2 .945 .945 .951 .951 # books 595 595 595 595 # observations 14,115 14,115 14,115 14,115

Robust standard errors in parentheses. All explanatory variables are interacted with lagged sales, as explained in the text. Avg. Similarity measures how similar a book is to the current set of bestsellers, based on the books' listed genres. Only books that never made the bestseller list are included in the estimation. The number of observations does not equal the number of books times 25 because some weeks are missing. book's genre tend to be good weeks for the non-bestseller.Consistent with the findingsin Table 4, the effect seems to be most pronouncedfor books appearingon the list for the firsttime: when the similaritymeasure is defined only for the subset of bestsellerswhose first appearancewas in the given week, the positive effect of the similarityis stronger.This pattern could plausibly reflectmultiple-book purchases: for example, weeks in which a new romancenovel hit the bestsellerlist drawmore than the usualfraction of romance enthusiastsinto bookstores, and they may buy several romance novels when visitingthe store. Although the estimated coefficients on the similarity measures are statisticallysignificant, they should be interpretedcarefully. One caveat to keep in mind is that publisherschoose their releasedates strategically.For example, there is some evidence that publishersof non-bestsellingbooks tendto avoid releasedates that coincidewith the releaseof a blockbustertitle in the same genre.The regressionsin Table 6 may be biased againstfinding business-stealing,because publisherschoose their books' release dates to actively avoid business-stealing.That is, if book A were expectedto steal business from book B, then book B's release date would be strategically distant from book A's release, so that potential instances of business- stealingare less likely to be observedin the data. Another thing to keep in mind when interpretingthe coefficients is that they represent (at best) indirectevidence of the substitutionpatterns between bestselling and non- bestsellingtitles, and the resultsmay not generalizeeasily to othercategories of books. For example,non-fiction titles on similartopics are perhapsless likely to exhibit complementarities.A consumer looking to buy a book on investments,for instance,seems more likely to choose just one (instead

? 2007 The Author. Journal compilation ? 2007 Blackwell Publishing Ltd. and the Editorial Board of The Journal of Industrial Economics.

This content downloaded from 130.212.18.200 on Thu, 6 Mar 2014 16:11:23 PM All use subject to JSTOR Terms and Conditions BESTSELLERLISTS AND PRODUCT VARIETY 737 of buying many) than would a consumer looking for a suspense or romancenovel.27

V. DISCUSSION & CONCLUSIONS Based on the sales data analyzedhere, it seems clear that appearingon the New York Times bestseller list has a direct impact on a book's sales. Estimatesbased on two alternativeidentification strategies show that sales increasewhen a book appearson the list. However, the magnitudeof the effect is modest, and appearsto reflectspikes in the sales of books that were surpriseson the list (e.g., books by new authors). Given the amount of attentionpaid to the New YorkTimes list (and the desperateschemes occasionally employed by authorsto securea position on the list), the estimatesof its impactmay seem surprisinglysmall. However, the present analysis ignores two effects that could be quite significantto authorsand publishers.First, while appearing on the bestsellerlist may have only a modest immediateimpact on the book's sales, it may dramatically increase the popularity of future books by the same author. Second, paperbacksales may be influencedby whetherthe hardcoveredition was a bestseller. (Indeed, paperback versions of books that were hardcover bestsellerstypically announce that fact prominentlyon the front cover.) Although these effectscan't be measuredwith the availabledata, anecdotal evidencesuggests they may be important. Although there are a variety of reasons why the bestseller list might directlyinfluence sales, the patternsin the data are most clearlyconsistent with an information-basedexplanation. Retail-level promotions cannot rationalizetwo of the study'scentral findings: whereas retailers' promotions persistfor the durationof a book's termon the list, the estimatessuggest that the impact of appearingon the list is transitory,with the bulk of the effect realizedin the firstweek. Moreover,the impacton salesis most pronounced among relativelyunknown authors (new authors in particular),a pattern that favors informationover promotionas an explanationfor the effect. The impact of the list on sales (and the consequent increase in the concentrationof demandon bestsellers)raises the interestingcounterfactual question:'Would more books be publishedif it weren'tfor bestsellerlists?' Whether(and in which direction)the bestsellerlist shifts the publish/no- publish margin depends on the nature of substitutionbetween bestselling and non-bestsellingtitles-i.e., are the extra sales of bestsellersones that

27 In unreportedanalyses using sales data for nonfictiontitles, coefficientson similarity measures like those in Table 6 indeed suggested a pattern of substitution rather than complementarity.Nonfiction titles have the advantageof being able to categorizesubjects muchmore finelythan in fiction,but it was difficultto includenonfiction titles in the broader analysisbecause the coverageof nonfictionbestsellers in my samplewas sparse.

? 2007 The Author. Journal compilation ? 2007 Blackwell Publishing Ltd. and the Editorial Board of The Journal of Industrial Economics.

This content downloaded from 130.212.18.200 on Thu, 6 Mar 2014 16:11:23 PM All use subject to JSTOR Terms and Conditions 738 ALANT. SORENSEN would have otherwisegone to less popular books? The results here offer indirectevidence that for hardcoverfiction, bestsellersand non-bestsellers withinthe samegenre may in fact be complements:weeks in whichbooks of a particulargenre first appear on the bestsellerlist tend to be strong-selling weeks for non-bestsellersof the same genre. Although too indirect to be conclusive,this resultsuggests that marketexpansion effects dominate any business-stealingassociated with bestsellerlists, so that bestsellerlists may in fact increasethe numberof books publishedin equilibrium.

REFERENCES

Alexander, P., 1997, 'Product Variety and Market Structure:A New Measure and a Simple Test,' Journalof EconomicBehavior and Organization,32(2), 207-214. Banerjee,A., 1992, 'A Simple Model of Herd Behavior,' QuarterlyJournal of Economics, 107(3), 797-817. Becker, G. and Murphy, K., 2000, Social Economics.:Market Behavior in a Social Environment,(Harvard University Press). Berry, S. and Waldfogel, J., 2001, 'Do MergersIncrease Product Variety?Evidence from Radio Broadcasting,' QuarterlyJournal of Economics, 116(3), 1009-1025. Bogart, D. (ed.), 2001, The BowkerAnnual, 46th edition. Clerides, S., 2002, 'Book Value: Intertemporal Pricing and Quality Discrimination,' InternationalJournal of IndustrialOrganization, 20(10), 1385-1408. Cole, D., 1999, The CompleteGuide to Book Marketing,(Allworth Press, Chicago). Dixit, A. and Stiglitz, J., 1977, 'Monopolistic Competition and Optimum Product Diversity,' AmericanEconomic Review, 67(3), 297-308. Einav, L., 2003, 'Gross Seasonality and Underlying Seasonality:Evidence from the U.S. Motion Picture Industry,' working paper, Stanford University. Epstein, J., 2001, Book Business.:Publishing Past, Present, and Future,(W.W. Norton). Greco, A., 1997, The Book PublishingIndustry, (Allyn & Bacon). Korda, M., 2001, Making the List.: A Cultural History of the American Bestseller, 1900-1999, (Barnes & Noble Books, New York). Lancaster, K., 1975, 'Socially Optimal Product Differentiation,' American Economic Review,65(4), 567-585. Milgrom, P. and Roberts, J., 1986, 'Prices and Advertising Signals of Product Quality,' Journalof Political Economy,94, 796-821. Reinstein, D. and Snyder, C., 2000, 'The Influence of Expert Reviews on Consumer Demand for ExperienceGoods,' working paper, George Washington University. Rosen, S., 1981, 'The Economics of Superstars,' American Economic Review, 71(5), 845-858. Suzanne, C., 1996, This Businessof Books, (Wambtac). Sweeting, A., 2005, 'Too Much Rock & Roll? Station Ownership, Programming and Listenershipin the Music Radio Industry,' working paper, Northwestern University. Vettas, N., 1997, 'On the Informational Role of Quantities: Durable Goods and Consumers' Word of Mouth Communication,' InternationalEconomic Review, 38, 915-944. Watson, R., 2003,'Product Varietyand Competition in the Retail Marketfor Eyeglasses,' working paper, Northwestern University.

? 2007 The Author. Journal compilation ? 2007 Blackwell Publishing Ltd. and the Editorial Board of The Journal of Industrial Economics.

This content downloaded from 130.212.18.200 on Thu, 6 Mar 2014 16:11:23 PM All use subject to JSTOR Terms and Conditions