BESTSELLER LISTS AND PRODUCT VARIETY LAN T - PDF document

BESTSELLERLISTSANDPRODUCTVARIETYT.SORENSENThispaperusesdetailedweeklydataonsalesofhardcover“ctionbookstoevaluatetheimpactoftheNewYorkTimesbestsellerlistonsalesand behaviorinthepresenceofsocialeffects,withbestsellerlistsservingasaformofcoordinatingmechanism.Forexample,teenagerswhowanttolistentomusicthatishotcanlooktotheBillboardchartsto“ndoutwhatispopular,andpeoplemayfavormoviesatthetopofthebox-of“cechartsbecausetheywanttobeconversantinpopularculture.Inthespeci“ccaseofbooks,bestsellerstatusalsotriggersadditionalpromotionalactivitybyretailers.Forthesamereasonsthatbestsellerlistsmaydirectlyaffectconsumerspurchasedecisions,theymayalsocausesalestobemorehighlyconcentratedonthefewbestsellingproducts.This,inturn,couldin”uenceproductvariety:iftheadditionalsalesaccruingtobestsellersasadirectconsequenceofthepublicationofthelistcomeattheexpenseofnon-bestsellingproducts,theoptimalnumberofproductstooffermaydecrease(relativetowhatwouldhavebeenoptimalintheabsenceofabestsellerlist).Forexample,ifpublicizedboxof“cerankingscauseticketsalestobemoreconcentratedonblockbusters,theymayalsomakeitunpro“tabletoincurthe“xedcostsofproducinga“lmwhosepopularityisexpectedtobeonlymarginal.Thispaperexaminestheseissuesinthecontextofthebookpublishingindustry,lookingspeci“callyattheimpactoftheNewYorkTimesbestsellerlistonsalesofhardcover“ctiontitles.Theempiricalanalysisaddressestwoquestions.First,doesbeinglistedasaNewYorkTimesbestsellercauseincreaseinsales?Second,doesthein”uenceofthebestsellerlistalsoaffectthenumberofbooksthatarepublished,andifso,inwhichdirection?Obvioussimultaneityproblemsmakeansweringthe“rstquestionanontrivialempiricalexercise;however,subtletiesintheconstructionandtimingoftheNewYorkTimeslistcanbeexploitedtoidentifyitsimpact.Theresultssuggestamodestincreaseinsalesforthetypicalbestsellerwhenit“rstappearsonthelist,andtheincreaseappearstore”ectaninformationaleffectratherthanapromotionaleffect:theeffectsareconcentratedinthe“rstweekabookappearsonthelist,andtheyaremostsubstantialfornewauthors.Regardingthesecondquestion,theimpactofbestsellerlistsonproductvarietyistheoreticallyambiguous,sinceitdependsonwhethermarketexpansionorbusiness-stealingeffectsdominate.Althoughthedataarelessthanidealforaddressingthisquestion,Ipresentindirectevidencesuggestingthebusiness-stealingeffectsofbestsellerlistsareunimportant:ifanything,bestsellerlistsappeartoincreasesalesforbothbestsellersnon-bestsellersinsimilargenres.Althoughthispaperisthe“rsttoexploretheimpactofbestsellerlistsonproductvariety,similarquestionshavebeenpreviouslyaddressedinanumberofcontexts.EarlytheoreticalworkemphasizedthetradeoffbetweenSee,e.g.,(BeckerandMurphy[2000],Banerjee[1992],andVettas[1997]).Thislineofreasoningwillbeclari“edinsectionII.ALANT.SORENSEN2007TheAuthor.Journalcompilation2007BlackwellPublishingLtd.andtheEditorialBoardofTheJournalofIndustrialEconomics. quantityanddiversityinthepresenceofscaleeconomies(DixitandStiglitz[1977])andtheeffectsofmarketstructureonproductvariety(Lancaster[1975]).Empiricalstudiesofproductvarietyhavebeenundertakenfortheradiobroadcastingindustry(BerryandWaldfogel[2001],andSweeting[2005]),themusicindustry(Alexander[1997]),andforretaileyeglasssales(Watson[2003]),tonameafewexamples.Inthefollowingsection,Ioutlineabasictheoreticalframeworkforunderstandinghowbestsellerlistscanin”uencethenumberofbooksthatgetpublishedinequilibrium.Section3providesabriefdescriptionofthebookindustryandthedataset.Theempiricalanalysis(Section4)proceedsintwoparts:“rst,thedataareusedtoidentifyandquantifythedirectimpactofthebestsellerlistonsales;second,substitutionpatternsinthedataareanalyzedtodeterminethelikelydirectionofthelistsimpactonproductvariety.Somebroadimplicationsoftheresults(aswellasalternativeinterpretations)arediscussedintheconcludingsection.II.ASIMPLETHEORETICALFRAMEWORKConsiderahighlysimpli“edmodelinwhichasinglepublisherchooseshowmanymanuscriptstopublish.Therearemanuscriptsunderconsidera-tion.Printingandmarketingamanuscriptrequiresa“xedcost,,inadditiontothe(constant)marginalprintingcost.Priortopublication,themanuscriptscanberankedinorderofexpectedpopularity,withbeingtheindexofthe-bestbookamongthealternatives.Themarketpriceofapublishedbook,,istakenasgiven,andthepost-publicationpricedoesnotadjusttore”ectabooksrelativepopularity.Theexpecteddemandforthebestbookisgivenby),where)canbeinterpretedasthelevelofaggregatedemandforbooks(whichmaydependonhowmanyareoffered),and)isafunctiondetermininghowaggregatedemandisallocatedamongbooksdependingontheirrelativepopularity(e.g.,wecouldhave1).Onlybookswithpositiveexpectedpro“tswillbepublished,sothenumberofbookswillbethemaximumsuchthat,asillustratedinFigure1.Nowconsiderthepotentialimpactofpublishingabestsellerlist,sothatconsumersobservethesalesranksofthetopbooks.SupposethattheThemodelcouldalsoapplytomoviestudiosdecisionsabouthowmany“lmstoproduce,ortorecordlabelsdecisionsabouthowmanyartiststosign.Thisassumptionwouldbeabsurdinmostcontexts,butinthiscaseitisatleastdescriptiveofacuriouspracticeinmultimediamarkets.Pricesofbooks,movies,andCDsalmostneverre”ectthepopularityoftheindividualproducts.(Clerides[2002])providesdirectevidenceandathoroughdiscussionofpricingissuesinthebookindustry.)Typically,pricepointsforbooksandCDsaredeterminedbeforetheyaremarketed,andsubsequentadjustmentsareextremelyinfrequent.Movieticketpricesareevenmorerigid:inthesummerof2002,forexample,itcostthesameamounttoseeChicago(whichwontheOscarforbestpictureandwasabox-of“cesuccess)asBoatTrip(which”oppedattheboxof“ceandwasuniversallyridiculedbycritics).BESTSELLERLISTSANDPRODUCTVARIETY2007TheAuthor.Journalcompilation2007BlackwellPublishingLtd.andtheEditorialBoardofTheJournalofIndustrial aggregateconsumerresponsetoabestsellerlistleadstoanincreasedconcentrationofsalesonbestsellers:letting)denotetheexpectedmarketshareofthe-rankedbookamongalternativeswhenabestsellerlistoflengthispublished,and)denotethemarketsharewhennolistispublished,weassumethat)if.Ifbestsellerlistshaveanydirectimpactonconsumerbehavior,theeffectwouldalmostcertainlyhavethisfeature.Themostobviousmechanismforthiseffectisinformational:ifconsumersareuncertainaboutbooksqualities,andtheybelievethatatleastsomepastpurchasershadmeaningfulinformationaboutthebookstheypurchased,thenbestsellerstatuswouldbeasignalofquality.Alternatively,socialeffectsmayleadtohigherdemandforbestsellers:consumersmaywanttoreadwhateveryoneelseisreadingintheinterestofkeepingupwithwhatispopulare.g.,theydontwanttobeleftoutoftheconversationwhentheygotothecocktailparty.Inthemarketforhardcover“ctionbooks,anadditionalmechanismpushessalestowardbestsellers:retailersroutinelydiscountbestsellersandpositionthemprominentlyintheirstores.Iftheoveralllevelofdemand)isindependentofanybestseller-listeffects,thenthelistunambiguouslyreducesthenumberofbooksthatcanbe Figure1Thepublish/no-publishmarginALANT.SORENSEN2007TheAuthor.Journalcompilation2007BlackwellPublishingLtd.andtheEditorialBoardofTheJournalofIndustrialEconomics. pro“tablypublished.ThisisillustratedinFigure1:theincreasedsalesforthebookscomeattheexpenseofthenon-listedbooks,shiftingthepublish/no-publishmargintotheleft(from).Morerealistically,however,thepublicationofabestsellerlistcouldincreaseoveralldemandinadditiontochangingtheallocationofdemandacrosstitlesi.e.,theadditionalpromotionandinformationaboutbestsellerscouldattractconsumersthatotherwisewouldnothavepurchasedanybookatall.Inthiscase,theimpactofbestsellerlistsonthepublish/no-publishmarginisambiguous.(Figure2illustratesacaseinwhichmorebookswouldgetpublishedinthepresenceofalist,eventhoughthelistleadstoarelativelyhigherconcentrationofsalesamongbestsellers.)Thisframework,whileobviouslyoversimpli“ed,illustratestheprincipalideasunderlyingtheempiricalanalysestofollow.Ideally,wewanttoexaminesalesdatatoseeifindeed)forthatis,toseeifbestsellerlistsanincreaseindemandforbestsellersrelativetonon-bestsellers.Inordertosayanythingaboutwhetherbestsellerlistsaffectthenumberofbooksthatgetpublished,wemustthenaskamuchmoresubtlequestionofthedata:howaresalesofrelativelyunpopular(non-bestselling)booksaffectedbythepublicationofbestsellerlists?Thatis,howLK* Figure2ListmayincreaseoveralllevelofdemandBESTSELLERLISTSANDPRODUCTVARIETY2007TheAuthor.Journalcompilation2007BlackwellPublishingLtd.andtheEditorialBoardofTheJournalofIndustrial does)compareto)forveryclosetoUnfortunately(butnotsurprisingly),theavailabledataareinadequateforansweringthisquestiondirectly.Instead,wewilllookforindirectevidenceofsubstitutionbetweenbestsellingandnon-bestsellingtitles,whichwouldsuggestthepotentialfor(andthelikelydirectionof)productvarietyeffects.III.BACKGROUNDANDDATAIII(i).TheBookIndustryIntheU.S.,thevastmajorityofbooksareproducedbyasmallnumberoflargepublishinghouseslikeRandomHouseandHarperCollins.Theoddsagainstamanuscriptsbeingacceptedbyoneofthesepublishinghousesarelong,especiallyinthecaseof“ction.Thirtypercentorfewerofavailablemanuscriptsinanygivenyearareinprint,andalthoughninetypercentofpublishedbooksarenon“ction,seventypercentofthemanuscriptssubmittedtotraditionalpublishersare“ction(Suzanne[1996]).successfulmanuscriptsarebrokeredtopublishersbyliteraryagents.Theseagentsaretypicallyreluctanttotakeon“rst-timeauthors,andtheirfeestendtobesteep(around15percentofauthorsroyalties).However,usinganagentgreatlyincreasestheauthorschancesofsuccess:unsolicitedmanu-scripts(manuscriptsreceivedoverthetransom)areestimatedtohave“fteenthousandtooneoddsagainstacceptance(Greco[1997]).Manu-scriptsaresometimessoldtopublishersbyauction,butthismethodistheexceptionratherthantheruleandisusedprimarilybyestablished,brand-nameauthors.Thedecisiontoextendacontracttoanauthorismadeonlyafterreviewofthemanuscriptsqualityandsalabilitybyseveralstagesofeditors.Ifamanuscriptsurvivesthereviewprocess,thepublisherofferstheauthoracontractgrantingroyaltypaymentsinexchangeforexclusivemarketingrightsaslongasthepublisherkeepsthebookinprint.Royaltiesaveragesevenpercentofthewholesalepriceonhardcoverbooksbynewauthors,NotethattheillustrationinFigure1makesitappearthatsalesofthe-rankedbookand-rankedbookareequallyaffected,whichneednotbethecase.Forinstance,itisquiteplausiblethattheimpactisadecliningfunctionofabooksrank.Moreover,onlytheeffectonbooksatthemarginisrelevantfordeterminingthenumberofbooksthatgetpublished.Becausethismodeldoesnotexplicitlyspecifyconsumerspreferences,itsaysnothingaboutthewelfareeffectsofareductioninproductvariety.Thispaperwillbedeliberatelyagnosticonthispoint,sincethewelfareeffectscouldplausiblygoeitherway:ontheonehand,fewerbookscouldmeanforegonesurplusfromtitlesthatwouldhaveappealedtoreaderswithdiversetastes;ontheotherhand,fewerbookscouldmeanthatlesstimeiswastedreadingbadForthe“rstquarterof2003,thetopsixpublishingconglomeratesaccountedforover80percentofunitsalesinadult“ction.Thesenumbersimplythat10percentof“ctionmanuscriptsgetpublished.ALANT.SORENSEN2007TheAuthor.Journalcompilation2007BlackwellPublishingLtd.andtheEditorialBoardofTheJournalofIndustrialEconomics. andmayincreaseoncethebookachievesacertainlevelofsales(Suzanne[1996]).Uponthedeliveryofacompletedmanuscriptalargeportionofexpectedroyaltiesaregiventotheauthorintheformofanadvance;manyauthorsneverreceiveadditionalpaymentsbecauseadvancesoftenexceedroyaltiesearnedfromactualsales.Publishersretainalargeshareofroyaltiesinescrowaccountstocompensateforreturnedbooks;booksellersmayreturnunsoldbookstothepublishersforfullprice,andthereforereturnratesexceeding“ftypercentarecommon(Greco[1997]).Inspiteofthedif“cultythatauthorsseemtofaceingettingtheirmanuscriptspublished,thebookindustrygeneratesanastonishing”owofnewbookseachyear.Acrossallcategories(“ctionandnon“ction)andallformats(hardcover,tradepaper,andmass-marketpaper),over100,000titleswerepublishedintheyear2000alone.Inadult“ction,thenumberofnewbookspublished(calledtitleoutputwithintheindustry)hasincreaseddramaticallyoverthepastdecade.Theindustrystradepublication,Annual,reportsthattitleoutputforhardcover“ctionmorethandoubledfrom1,962in1990to4,250in2000.Incontrast,therateofincreasewasmuchmoregradualpriorto1990.Infact,Bowkerreportsthatthenumberof“ctiontitlesin1890wasover1,100,sotitleoutputhadlessthandoubledinthe100yearspriorto1990(Bogart[2001]).Thedramaticincreasesintitleoutputinthe1990swereroughlyconcomitantwithanincreaseintheconcentrationofsalesamongbestsellers.Fromthemid-1980stothemid-1990s,theshareoftotalbooksalesrepresentedbythetop30sellersnearlydoubled(Epstein[2001]).In1994,over70percentoftotal“ctionsaleswereaccountedforbyamere“veauthors:JohnGrisham,TomClancy,DanielleSteel,MichaelCrichton,andStephenKing(Greco[1997]).Publishersandauthorsemployanumberofmarketingstrategiesintheirattemptstoachievethekindofsuccessenjoyedbythesetop-sellingauthors.Publishersmarketingbudgetsmayrangebetweentenandtwenty-“vepercentofnetsales(Cole[1999]),andauthorsareexpectedtoappearpubliclyinpromotiontours.Bookreviewsarehighlysoughtafterbutdif“culttoobtain:tensofthousandsofbooksarepublishedeachyearintheUnitedStates,buttheNewYorkTimes(forexample)reviewsonlyoneperday.Publishersalsomaypayretailstoresforshelfspaceorinclusioninpromotionalmaterials.Bookmarketersconcentratetheireffortsoncreatingasuccessfullaunch;retailstoresmayremovelow-sellingnewreleasesfromtheirshelvesafteraslittleasoneweek(Greco[1997]).Rosen[1981]providesaclassicexplanationforthepresenceofsuchsuperstars.Acriticalpieceoftheargumentisthattheconvexityofreturnsresultsfromtheimperfectsubstitutabilityoftheproductsofferedi.e.,readingseveralunremarkablebooksmaynotbeagoodsubstituteforreadingasinglegreatone.BESTSELLERLISTSANDPRODUCTVARIETY2007TheAuthor.Journalcompilation2007BlackwellPublishingLtd.andtheEditorialBoardofTheJournalofIndustrial Inspiteofalltheresourcesspentonmarketing,oneofthebestkindsofpublicityappearingonabestsellerlistcannotbebought.Bestsellerlistshavelongplayedanimportantroleinthebookindustry.Regularpublicationofbestsellerlistsbeganin1895,whenaliterarymagazinecalledTheBookmanstartedprintingamonthlylistofthetopsixbest-sellingbooks.NewYorkTimesBookReviewbeganpublishingitsbestsellerlistasaregularfeaturein1942.Althoughmanyotherprominentlistsnowexist,NewYorkTimeslistisgenerallyconsideredthemostin”uentialintheindustry(Korda[2001]).III(ii).DataThemaindatasettobeanalyzedconsistsofweeklynationalsalesforover1,200hardcover“ctiontitlesthatwerereleasedin2001or2002.ThesalesdatawereprovidedbyNielsenBookScan,amarketresearch“rmthattracksbooksalesusingscannerdatafromanalmost-comprehensivepanelofretailbooksellers.Additionalinformationabouttheindividualtitles(suchasthepublicationdate,subject,andauthorinformation)wasobtainedfromavarietyofsources,includingAmazon.comandavolunteerwebsitecalledOverbooked.org.Table1reportssummaryinformationforthebooksinthedata,brokenintothreesubsamples.Theoverallsamplerepresentsarelativelylargefractionoftheuniverseofhardcover“ctiontitlesreleasedinthistimeperiod,thoughitislikelytobesomewhatskewedtowardpopularbooks.Booksthatneversoldmorethan50copiesinasingleweek(nationwide)weredroppedfromthesample,sincetheirweeklysalesnumbersappearedtobemostlynoise.Also,somebooksareexcludedfromtheempiricalanalysesiftheirreleasedatesweredif“culttodetermine.Insuchcases,thenumberofbooksisreducedto799.AswillOccasionallyanauthorhastestedthispropositionbypurchasingnumerouscopiesofhisownbook,inhopesofpushingitontoabestsellerlist.Noneoftheseattemptshaseverbeentrulysuccessful;thetypicaloutcomehasbeenconsiderableembarrassmentfortheperpetratoroncetheschemewasuncovered.MostmajorU.S.newspaperspublishtheirownlocallist(sometimesinadditiontotheYorkTimeslist),andanumberofnationallistscompetewiththeNewYorkTimeslistforattention(e.g.,PublishersWeeklyWallStreetJournalUSAToday,andBookScancollectsdatathroughcooperativearrangementswithvirtuallyallthemajorbookstorechains,mostmajordiscountstores(likeCostco),andmostofthemajoronlineretailers(like).Theyclaimtotrackatleast80percentoftotalsales.Inconstructingthesetofcandidatebookstotrack,wehadto“rstlocateabook(anditsISBNnumber)inordertoconsiderit.Obscure,slow-sellingbooksare(byde“nition)hardertoThreesourcesofinformationwereusedindeterminingtheexactweekofrelease.Formostliststheexactdayofrelease.Ifnot,BookScanreportsthemonthofrelease,andeyeballingthedatausuallyrevealsanobviousreleasedate.Ifforanyreasonthereleasedatewasnotobviouse.g.,theandBookScanreleasedatesdidntmatch,and/orthereleasedatewasntobviousfromlookingatthesalesdatathetitlewasexcludedfromthesamplewhenevertheresultsmightbesensitivetotheaccuracyofthereleasedate.ALANT.SORENSEN2007TheAuthor.Journalcompilation2007BlackwellPublishingLtd.andtheEditorialBoardofTheJournalofIndustrialEconomics. beshowninthenextsection,fornearlyallbooksthevastmajorityofsalesoccurinthe“rst4…6months,andsubsequentsalesarerelativelyuninformative.Inlightofthis,onlythe“rst26weeksofsalesareincludedinthesampleforanygivenbook.Dataonbestseller-liststatuscomedirectlyfromtheNewYorkTimes.TheanalysisfocusesontheNewYorkTimeslistbecauseitisanationallypublishedlist(andthesalesdataarenational),andbecauseitisalmostuniversallyregardedasthemostin”uentiallistintheindustry.Amajorityofretailbooksellers(includingonlinebookstores)havespecialsectionsdevotedtoNewYorkTimesbestsellersandofferpricediscountsonthesetitles,andauthorsaresometimesofferedbonusesforeveryweektheirbookappearsontheNewYorkTimesToconstructitslist,theNewYorkTimessurveysnearly4,000bookstoreseachweek,inadditiontoanumberofbookwholesalerswhoserveadditionaltypesofbooksellers(likesupermarkets,newsstands,etc.)Thereportedsales“guresfromtheserespondentsarethenextrapolatedtoanationallyrepresentativesetofsalesrankingsusingstatisticalweights.BecausetheYorkTimeslistisconstructedusingsamplingmethods,itoftenmakesmistakesi.e.,booksthatshouldhavemadethelistinagivenweek Table1SummarystatisticsAllbookswith26weeksofdata(nMinMedianMaxMeanStd.Dev.Listprice10.9524.9568.8524.282.63Releaseweek1578349.6424.18Newauthor001.15.36Sales,“rst6months873,9601,443,34535,183111,444Max.one-weeksales50486360,1336,90324,453Bookswithreliablereleasedates(nMinMedianMaxMeanStd.Dev.Listprice10.9524.956024.442.38Releaseweek1568348.7024.52Newauthor001.14.35Sales,“rst6months878,0631,443,34551,396134,445Max.one-weeksales50972360,13310,22729,626NewYorkTimesbestsellers(nMinMedianMaxMeanStd.Dev.Listprice10.952529.9525.052.25Releaseweek1408341.3323.89Newauthor001.05.22Sales,“rst6months20,96393,5071,443,345174,987222,530Max.one-weeksales2,10818,438360,13336,04750,058Week“rstappeared44344.52.4Weeksonlist14305.85.3Forreleaseweek,the“rstweekof2001isweek1,andthe“rstweekof2002isweek53.BESTSELLERLISTSANDPRODUCTVARIETY2007TheAuthor.Journalcompilation2007BlackwellPublishingLtd.andtheEditorialBoardofTheJournalofIndustrial (becausetheirsalesexceededthesalesofthebooklistedatrank15)aresometimesomitted,andtheorderingoflistedbookssometimesdoesntre”ectthetruerankingofsales(asindicatedbytheBookScandata).Also,assemblingthelisttakestime,sotheprintedbestsellerlistre”ectsrankingsfromthreeweeksprior.BothofthesefeaturesoftheNewYorkTimeslistthemistakesandthetimelagsarecriticalintheempiricalanalysisofitsimpactonsales.IV.EMPIRICALANALYSISANDRESULTSIV(i).SkewnessinBookSalesThemoststrikingpatterninthedataisthatbooksalesareremarkablyskewedintwoimportantways.First,thedistributionofsalesacrossbooksisheavilyskewed.Figure3plotstotalsalesinthe“rstsixmonthsagainstsalesrankforthetop100booksinthesample.Evenwhenlookingonlyatthetopdecileofbooks,theskewnessofthedistributionisstriking.Ofthe1,217booksforwhichatleast26weekswereobserved,thetop12(1percent)accountfor25percentoftotalsix-monthsales,andthetop43(3.5percent)accountfor50percent.The205booksthatmadeittotheNewYorkTimesbestsellerlistaccountfor84percentoftotalsalesinthesample.Themostpopularbookinthesample,SkippingChristmasbyJohnGrisham,soldmorecopiesinits“rstthreeweeksthandidthebottom368booksintheir“rstsixmonthscombined.Second,booksalestendtobeskewedwithrespecttotimeforanygiventitle:thatis,salestendtobeheavilyconcentratedinthe“rstfewweeksafterabooksrelease.Ofthe1,217booksinthesampleforwhich26ormoreweeksareobserved,898(73.8percent)hittheirsalespeaksometimeinthe“rstfourweeks.Themedianpeakweekisweek2.Somewhatsurprisingly,thispatternalsoseemstoholdfordebutauthors,forwhichonemightexpectgradualdiffusionofinformationandthereforeanS-shapedsalespath.Fornewauthors,112of182books(61.5percent)peakinthe“rst4weeks,andthemedianpeakweekis4.Thissecondformofskewnessisimportanttokeepinmindwheninterpretingthemodelsandresultsinthefollowingsections.Thesteadydecayofabookssalesovertimeisthedominantpattern:withtheexceptionofseasonaleffects(e.g.,Christmas),anyotherchangesinabookssalestendtobesecond-orderrelativetothisdecaytrend.Essentially,thesalespathsBooksforwhichthereleasedatewasquestionableareexcludedhere,sincegettingthereleasedaterightisntcriticalforthisexercise.Decay-likesalespatternsarenotuniquetothebookindustry;forexample,Einav[2003]reportssimilarpatternsformovieboxof“cesalesandincorporatesexponentialdecaydirectlyintohisempiricalmodel.ALANT.SORENSEN2007TheAuthor.Journalcompilation2007BlackwellPublishingLtd.andtheEditorialBoardofTheJournalofIndustrialEconomics. typicallyresembleexponentialdecaypatterns.Eyeballingthetimepathofsalesforallthebooksinthesample,onerarelyobservesabookssalestakeoffafterithitsthebestsellerlist;ifanything,makingthelistappearstotemporarilyslowthepaceofdecline.IV(ii).DoBestsellerListsDirectlyAffectSales?Theoryclearlysuggeststhatbestsellerlistsmaydomorethansimplyre”ectconsumerbehavior:thelistsmaydirectlyconsumerbehavior,sothatabooksappearanceonthebestsellerlisthasanindependenteffectonitssales.However,measuringsuchaneffectisadif“cultempiricalproblem:thesetofbooksthatreceivethetreatmentofbeinglistedasbestsellersisclearlynotrandom,andanaiveempiricalapproachwouldlikelyconfusethedirectionofcausality.Thereisobviouslyacorrelationbetweenthelevelofsalesandbestsellerstatus(bytheveryde“nitionofabestsellerlist),butwecannotinferfromthiscorrelationthatbeinglistedasabestsellercauseshighersales.However,giventheavailabledataandthesubtletiesintheconstructionofNewYorkTimeslist,thereareatleasttwowayswecanattempttoidentifythelistsdirectin”uence.Onestrategyistoexploittheso-calledmistakesthataresometimesmadeinthelist.Asmentionedpreviously,the 500 1000 1500Sales (thousands), first 6 months 1 100Overall sales rank Figure3SkewnessofbooksalesBESTSELLERLISTSANDPRODUCTVARIETY2007TheAuthor.Journalcompilation2007BlackwellPublishingLtd.andtheEditorialBoardofTheJournalofIndustrial processusedtogeneratetheNewYorkTimeslistisinexact.AlthoughthelistisbyandlargequiteaccuratewhencomparedwiththetruesalesnumbersavailablefromBookScan,itisnotuncommonforabestsellingbooktobemissedi.e.,abookmaynotappearonthelisteventhoughitssalesexceededthesalesoflistedbooks.Inprinciple,thesemistakesprovideameansofidentifyingtheeffectofappearingonthelist,byservingasanappropriatecontrolgroup.(Comparinglistedbookstounlistedbooksisabadexperiment,sincewhetherabookislistedisanearlydeterministicfunctionofthedependentvariable;butcomparinglistedbookstobookshavebeenlistedis,inprinciple,avalidexperimentaslongasthemistakesarerandomoccurrences.)Duringtheyears2001…2002,therewere182instancesinwhichahardcover“ctionbookwasnotlistedasaNewYorkTimesbestsellerwheninfactitshouldhavebeen,representing109differentbooks.(Inseveralcases,thereweremultipleweeksinwhichabookshouldhavebeenonthelistbutwasnot.)Themajorityofthese(roughly70%)werenarrowmisses:hadthebooksbeenlisted,theywouldhavebeenranked13…15onthelist.Inordertoconstructafaircomparison,Ifocusontwosetsofbooks:thosethatwerelistedatrank13,14,or15whenthey“rstappearedontheNewYorkTimeslist(44),andthosethatshouldhaveappearedat13,14,or15whentheyweremistakenlyomitted(75).Table2summarizessomeobservablecharacteristicsofthebooksinthetwogroups.Iftheomissionswerenotrandom,butratheranattemptbytheNewYorkTimesateditorializingthelist,wemightexpecttoseeadifferentsubjectcompositionamongtheomittedbooks.Thedistributionofsubjectsandlistpricesappearstobemostlysimilarbetweenthetwogroups,lendingsomecon“dencethatthe Table2Characteristicsoflistedbooksvs.omittedbooksPercentofmistakeslistingthisgenre(PercentofbestsellerslistingthisgenreLiterature/Fiction51641.39(0.17)Mystery/Thriller51361.53(0.13)Romance21111.47(0.14)ScienceFiction1270.96(0.34)Religion101.00(0.32)Horror320.13(0.89)Averagelistprice:24.9824.541.09(0.28)Onlybooksthatwereranked13…15whenthey“rstappearedonthebestsellerlistareincludedinthebestsellersgroup.NumbersarebasedonbooksgenresaslistedonAmazon.com.Percentagesaddtomorethan100becausebookstypicallylistmorethanonegenre.Pricesarepublisherslistprices,fromBookScan.Tobeprecise,Iwillsayabookshouldhavebeenlistedif,fortheweekthatwasrelevantforgeneratingthelist,thebookssales(accordingtoBookScan)exceededthesalesofthebookthatwaslistedat#15.ALANT.SORENSEN2007TheAuthor.Journalcompilation2007BlackwellPublishingLtd.andtheEditorialBoardofTheJournalofIndustrialEconomics. omissionsareindeedrandommistakes.TheonlynotabledifferencesarethatthegenreLiterature&Fictionismorelikely(andRomanceorMystery&Thrillerslesslikely)amongbooksmakingthelistthanamongomittedbooks,butthesedifferencesarenotstatisticallysigni“cant.Moreover,whenconsideringtheoverallcompositionofthelist(notjustpositions13…15),theproportionsofRomanceandMysterynovelsmatchalmostperfectlywiththeproportionsamongtheomittedbooks,soitseemscleartheomissionswerenottheresultofanybiasagainstparticulargenres.Table3reportsacomparisonofsalesforthetwogroups.Forbooksthatwerepublishedforthe“rsttimeontheNewYorkTimeslist,salesdeclinedbyanaverageof7.6percentrelativetothepreviousweek.Formistakenlyomittedbooks,salesdeclinedbyanaverageof22.7percent.Takenatfacevalue,thedifferenceimpliesthatinthe“rstweek,beinglistedleadsto19percentmoresalesthanwouldhaveotherwiseoccurred.However,thedifferenceisstatisticallyimprecise:forsigni“cancelevelslessthan.08,wewouldfailtorejectthenullhypothesisthatthelisthasnoeffectusingaone-tailedtest.ThecomparisonreportedinTable3doesnotcontrolforanycovariates,butdoingsohasverylittleeffectontheestimate.Usingsimplelinearregressionstocontrolforseasonaleffectsandtime-since-releaseeffectsyieldsestimatesofthesameapproximatemagnitude.Giventheavailabilityofpaneldataonbooksales,asecondstrategyforidentifyingtheeffectofbestsellerlistsistousealltheavailabledata(notjustmistakebooks)tomeasuretheweek-by-weekchangesinsalesassociatedwithchangesinbestsellerstatus.WhereastheresultsinTable3arebasedoncomparisonsbooks(withmistakenlyomittedbooksservingasthecontrolgroupforthelistedbooks),thissecondapproachisbasedonwithin-bookvariationovertime(sothecontroliseffectivelythesamebookssalestrajectorytoitsappearanceonthelist).Observingsalesoverseveralweeksforeachbookmakesitpossibletocontrolforbook“xedeffects,thusabsorbingtheobviouslyendogenousdifferencesinsaleslevelsforbestsellervs.non-bestsellertitles.Moreover,thetimelaginvolvedintheNewYorklistmeansthatwehaveatleastthreepre-treatmentobservationson Table3Comparisonofsaleschanges:listedbooksvs.omittedbooksMeanStd.Dev.sales,“rstweekappearingonbestsellerlist(n.076.702sales,weekinwhichmistakenlyomitted(n.227.157.151.One-tailedt-testfor:differenceiszero:1.404,Recallthatthedominantpatterninsalesovertimeisasteadydecay:evenforbooksappearingonthebestsellerlist,itisrareforsalestoincreasefromoneweektothenext.BESTSELLERLISTSANDPRODUCTVARIETY2007TheAuthor.Journalcompilation2007BlackwellPublishingLtd.andtheEditorialBoardofTheJournalofIndustrial eachbestsellerbeforeithitsthelist.(Duetothetimelag,thesoonestabookcanappearonthelistisatthebeginningofitsfourthweekonbookstoreshelves.)Inprinciple,therefore,thedatacanbeusedtoestimateamodelthatcontrolsforbook-speci“cdifferencesinthelevelofsalesbook-speci“cdifferencesinsalestrendsobservedpriortoappearanceonthebestsellerlist.Anempiricalmodelofbooksalesovertimemustaccommodatethedominantdecaytrendinsalesovertime(asdescribedinsectionIV(i))andallowforappearancesonthebestsellerlisttodirectlyaffectsales.Moreover,goodnessof“tisevenmoreimportantthanusualhere,sincethecounterfactualwewantfromthemodel(howmanyunitsabookwouldhavesoldifithadntappearedonthebestsellerlistinacertainweek)isessentiallyaforecast.Giventheseconsiderations,onesimpleapproachistomodelbooksalesasanautoregressiveprocessinwhichtheautoregressionparameterisafunctionofcovariates:withasetofcovariatesforbookinweek(includingbestsellerstatus)andademandshockthatisindependent(butnotnecessarilyidenticallydistributed)acrossand.Theautoregressivestructureisappealinginpartbecauseittrackssalesquitewell,sothatthepredictioninanyperiodwillneverbeoffbymuch.Moreover,thisspeci“cationfocusestheestimationonchangesinsalesforagivenbookratherthandifferencesinthelevelofsalesacrossbooks,andallowsforbook-speci“cdifferencesintherateofdecayinsales(viabook-speci“cconstantsin).Inotherwords,themodelcanaccommodateunobserved,book-speci“cheterogeneityinboththelevelandtrajectoryofsales.Inorderforanyremainingendogeneitytobeimportantitmusttakeapeculiarform.Inparticular,thetimingofabooksappearanceonthebestsellerlistwouldhavetocorrespondtoweeksofidiosyncraticallyhighdemandinordertobiasourestimateofthelistsimpact.Giventhatthecurrentlistre”ectssalesfromthreeweeksprior,suchendogenoustimingseemsveryunlikely.Althoughpublishersaresurelystrategicaboutwhentheyreleasebooks,itseemssafetopresumetheydontsystematicallyreleasebooksexactlythreeweeksinadvanceofanticipatedpositivedemandshocks.Thoughimplementedsomewhatdifferently,theempiricalstrategiesemployedherearesimilarinspirittotheanalysisofReinsteinandSnyder[2000],whoexploitquirksinthetimingofmoviecriticsreviewstoidentifytheirimpactonboxof“cesales.Onepotentialcaveatisthatpublishersmaystrategicallyavoidreleasingbooksatthesametimeasmajorblockbusters,whosereleasedatesareannouncedwellinadvance.(Thisisdiscussedinmoredetailinsection4.3below.)However,eveninthiscaseitisnotclearwhypublisherswouldtimetheirreleasessuchthatthepositiveshockcomesexactlythreeweeksafterALANT.SORENSEN2007TheAuthor.Journalcompilation2007BlackwellPublishingLtd.andtheEditorialBoardofTheJournalofIndustrialEconomics. Table4reportsestimatesofthismodelforfourseparatespeci“cations,usingweeks2…26foreachbookinthesample.Allfourspeci“cationsincludeafullsetofweekdummiestocontrolforseasonalvariationinbooksales(thereisalargeincreaseinsalesinmidtolateDecember,forexample),aswellasafullsetofbookdummiesin.ColumnIreportstheestimatedcoef“cientonanindicatorthatequalsoneforeveryweekinwhichthebookappearedontheNewYorkTimesbestsellerlist.Thepointestimatesarestatisticallysigni“cantatthe5percentlevel,andimplythatsalesdeclineabout4percentagepointsmoreslowlywhenabookislistedasabestseller.Giventhemanytheoreticalreasonswhyappearancesonabestsellerlistshouldhaveadirectimpactonsales,itisnotsurprisingto“ndastatisticallysigni“canteffect.Themoreinterestingquestionisperhapsthelisthasani.e.,whatistherelevantmechanismbywhichlistappearancesboostsales?Oneclassofexplanationsisbasedontheideathatbestsellerstatusisinformative:forexample,bestsellerappearancescouldsignalhighquality,orsignalwhatotherpeoplearereading(andtherebyfacilitatesocialcoordination).Anotherpossibilityisthatsalesincreasesresultfrompromotionalactivitiesundertakenforbestsellers(e.g.,prominentplacementonbookstoreshelves).Arathermundaneexplanationwouldbethatthe Table4Regressionestimates:booksalesandliststatusIIIIIIIVNYT-listed.043.030(.015)(.019)NYT,week1.087NYT,week2NYT,1weekoffrank.004.004.042(.016)NYT-listed[6…10].040(.027)NYT-listed[11…15].068Oprah6.3566.2866.3446.346(.111)(.111)(.114)(.112)GMA.108.100.118.109(.085)(.083)(.086)(.085)Weeksout(.022)(.021)(.024)(.023).977.978.977.977#books799799799799#observations19,02419,02419,02419,024Robuststandarderrorsinparentheses.Allexplanatoryvariablesareinteractedwithlaggedsales,asexplainedinthetext,andeachspeci“cationincludesweek“xedeffectsandbook“xedeffects.Thenumberofobservationsdoesnotequalthenumberofbookstimes25becausesomeweeksaremissing.BESTSELLERLISTSANDPRODUCTVARIETY2007TheAuthor.Journalcompilation2007BlackwellPublishingLtd.andtheEditorialBoardofTheJournalofIndustrial increasesaremerelypriceeffects,resultingfromtheautomaticdiscountthatmanystoresofferoncurrentbestsellers.Thespeci“cationsreportedincolumnsII…IVofTable4attempttoshedlightonthesourceofthebestsellerlistsimpact,andspeci“callytrytodistinguishwhetherinformationorpromotionisdrivingtheeffects.ColumnIIseparatestheeffectofalistappearancebetweenthe“rstweekandsubsequentweeksonthelist.Ifthesalesincreasesaredrivenbyinformation,weshouldexpectthemtobeconcentratedinweekone,whentheinformationisnewtoconsumers.Ontheotherhand,ifsalesincreasesmostlyre”ectpromotionalactivities,thentheeffectsshouldbeapparentaslongasthebookremainsonthelist.Theestimatesaremoreconsistentwiththeinformation-shockexplanation:theyindicatean8.7percentage-pointchangeintheweekthebook“rstappears(statisticallysigni“cantatthe5percentlevel),withanyeffectsinsubsequentweeks(combined)beingstatisticallyindistinguishablefromzero.Moreover,abooksremovalfromthelist(which,accordingtothebookstoremanagerswithwhomIspoke,abruptlyterminatesanyspecialin-storepromotionalactivityforthebook)haslittleimpactonitssales:thecoef“cientonthedummyfor“rstweekoffisneitherstatisticallynoreconomicallysigni“cant.Giventhisresult,itisveryunlikelythatthemeasuredimpactofbestsellerstatusrepresentsapurepriceeffect.Justasinformationaleffectsshouldbeconcentratedintheweekofabooks“rstappearanceonthelist,theymayalsobeconcentratedamongbooksatthetopofthelist.Thereisobviouslyabigdifferenceinthebuzzgeneratedbyanumber-onebestsellerthanbyabookbarelybreakingthroughtothebottomofthelist,andsocialeffectsinconsumptionarepresumablystrongestforthebooksattheverytop.Bycontrast,theeffectsofin-storepromotionsshouldbeequallyimportantforallbooksonthelist,becauseallofthemareputonaspecialrack,and/orallofthemarediscounted.ColumnsIIIandIVofTable4reportestimatesfromregressionsinwhichtheimpactofalistappearanceisallowedtodependonthebookspositiononthelist.Basedontheseestimates,thereisnoevidencethatthelistsimpactislargerforbooksatthetopofthelist:theinteractionofthelistdummywithlistrankisstatisticallyindistinguishablefromzeroinspeci“cationIII,andthecoef“cientsonlistdummiesforseparategroups(books1…5onthelist,books6…10,andbooks11…15)arestatisticallyindistinguishableinspeci“cationIV(0.13,-value0.88).Ifanything,thepointestimatessuggestlargereffectsforbooksattheofthebestsellerlist.Ininterpretingtheseresults,itisimportanttorememberthattheinformationcontentofanappearanceonthelistisquitelowforsomeauthorseveryoneknowsthatGrishamsnewestnovelwillmakethelist,sothereisnoinformationshockwhenit“rstappearsandtheestimatesinTable4reporttheeffectoverallbooksthatmadethelist.Indeed,ALANT.SORENSEN2007TheAuthor.Journalcompilation2007BlackwellPublishingLtd.andtheEditorialBoardofTheJournalofIndustrialEconomics. itcouldbethatformostbooks,appearingonthelisthasvirtuallynoeffectonsales;onlyappearancesthataresurprisesmakeadifference.TheresultsreportedincolumnsIIIandIVofthetable,suggestingthatappearancesatthetopofthelisthavenomoreimpactonsalesthanappearancesatthebottomofthelist,maysimplyre”ectthatmosttop-rankedbestsellersarebooksthatwereexpectedtobethere.Inordertoexaminethisissuemorecarefully,themodelwasre-estimatedallowingforbook-speci“cheterogeneityintheeffectofbeinglisted.Thatis,themodelwasestimatedwithisanindicatorequaltooneifbookappearedonthebestsellerlistforthe“rsttimeinweek.Themoststrikingfeatureofthesishowtheyrelatetotheauthorshistories.Figure4showskerneldensitiesfortheestimatedsseparatelyforwell-establishedauthorsvs.newerauthors.Forestablishedauthors,thedistributioniscenteredat0.02andisapproximatelysymmetric.Forauthorswhowerelesswell-established,thedistributionhasapositivemodeandalongrighttail,whichisconsistentwiththeideathatbestsellerstatushasanappreciableimpactonlyforsomebooks.Perhapsthemostinterestingresultrelatestothesmallsampleofelevendebutauthorswhomadethelist.Fortheseauthors,theestimatedsareconsistentlylargeandpositive:oftheeleven,nineareinthetopquartile,andsevenareinthetopdecile.Theaveragevalueofamongthesenewauthorsis0.35(comparedwith0.05fortheremainderoftheauthors).Althoughthesamplesizeisobviouslysmallhere,thepatternsareclearlyconsistentwiththehypothesisthatappearingonthebestsellerlistonlyhasanimpactwhentheappearanceisinformative.TheheterogeneityevidentinFigure4isimportanttokeepinmindwheninterpretingthequantitativeresultsfromTable4.Takenatfacevalue,theestimatedcoef“cientsimplyamodesteffectoflistappearancesonsales,eventhoughthespeci“edautoregressivemodelallowstheeffecttopersist.Forexample,comparingabookthatappearedonlyonceonthebestsellerlist(initsfourthweekfromrelease)toabookthatstartedwiththesameinitialsalesbutneverappearedonthelist,andassumingaconstantTwooutlierswereomittedintheestimationofthedensity.One,anextremenegativevalue,wasforabookthatwasannouncedasatelevisionbookclubpickonthesamedayit“rstappearedonthebestsellerlist,sothedatahavedif“cultydistinguishingthetwoeffects.Theother,anextremepositivevalue,wasforabookwithamysteryauthor.TheDiaryofEllenwasnominallywrittenbyanew(but“ctitious)authornamedJoyceReardon,anditwasrumoredthatthebookwasactuallywrittenbyStevenKing.Salesspikedconsiderablywhenthebook“rsthitthebestsellerlist,soitsestimatedisverylarge.Thegroupsofauthorsweresplitbasedonthenumberofpastadult“ctiontitleseachauthorhadpublished.Thetopthird(havingpublished35ormoretitles)weredesignatedtheestablishedauthors,andtheremainingtwothirdswerecalledthenewerauthors.BESTSELLERLISTSANDPRODUCTVARIETY2007TheAuthor.Journalcompilation2007BlackwellPublishingLtd.andtheEditorialBoardofTheJournalofIndustrial equalto0.7,the8percentage-pointincreaseinweek4translatestoa13.7percentdifferenceinexpectedsalesoverthe“rst52weeks.However,theaverageeffectsseemtobemaskingtherealstory:forestablishedbestsellingauthors,thelisthasnodiscernibleimpactonsales;butfornewauthors,appearingonthelisthasarelativelydramaticimpact.(Aone-timeincreaseof0.35inweek4wouldleadtoa57percentincreaseinsalesoverthefollowing52weeks.)Notethatevenattheirlargest,theeffectsofbestsellerlistappearancesaresmallwhencomparedtotheeffectsoftheannouncementvariablesincludedintheregressionsreportedinTable4.TheOprahindicatorisequaltooneifthebookwasannouncedasaselectionforOprahWinfreysbookclubinthatweek,andtheGMAindicatorisequaltooneifthebookwasannouncedasthepickfortheGoodMorningAmericashowsbookclub.Theestimatedeffectsoftheseannouncementsareverylargetheydwarfanyeffectofthebestsellerlist.Thein”uenceoftheannouncementscouldderivefromvarious 0density .4 0 .8Beta established authorsnewer authors Figure4Heterogeneityintheimpactofbestsellerlistonsalesisinitialsales,thentotalsalesoverweeksis.Ifincreasesbyanamountinweek(butthenrevertstoitspreviousvalueandisotherwiseconstant),thenthepercentageincreaseinsalesisequaltoALANT.SORENSEN2007TheAuthor.Journalcompilation2007BlackwellPublishingLtd.andtheEditorialBoardofTheJournalofIndustrialEconomics. sources:theannouncementscouldsimplyalertarelativelylargenumberofconsumerstotheexistenceofthebook;theycouldserveasaqualitysignal;ortheycouldactasacoordinationmechanismbywhichalargenumberofconsumersagreetoreadthesamebook.(Thelattermechanismistheapparentobjectiveofthetelevisionbookclubs.)Finally,althoughthespeci“cationsreportedinTable4shouldadequatelycontrolforunobservedheterogeneityrelatedtobooksvaryinglevelsofsales,itisneverthelessusefultoconstructarealitycheckforthekeyestimatedcoef“cients.Forexample,insteadofregressingsalesonaninteractionoflaggedsalesandanindicatorfor“rstweekonthelist,wecaninteractlaggedsaleswithanindicatorfor“rstweekalmostonthelist,de“nedasanyunlistedbookwithsalesgreaterthan90percentofthe15rankedbestseller.Ifthecoef“cientsreportedinTable4aremerelyanartifactoflatentheterogeneityinsalesdynamicsthatiscorrelatedwithdifferencesinthelevelofsales,thenthecoef“cientonthisvariablewouldbepositiveandsigni“cant.Infact,runningthisregressionyieldsacoef“cientestimateof0.028withastandarderrorof.018,lendingsomeadditionalcredibilitytotheresultsreportedinthetable.IV(iii).DoBestsellerListsAffectProductVariety?TheresultsoftheprevioussectionindicatethatsomebooksappearingonNewYorkTimesbestsellerlistenjoyaconsequentincreaseinsales.Buttowhatextentarethoseextrasalesstolenfromnon-bestsellingtitles?AswasexplainedinsectionII,whether(andinwhichdirection)abestsellerlistin”uencesproductvarietydependsonwhetherthelisthasanyimpactonbooksnearthepublish/no-publishmargin.Iftheconsumerwhobuysabookbecausehesawitonthebestsellerlistwouldotherwisehaveboughtabooknearthemarginofpro“tability,thenonecanarguethatthelistreducestheprivatelyoptimalnumberofbooksbyfurtherconcentratingdemandonbestsellers.However,itisalsopossiblethattheconsumerwouldhaveotherwiseboughtnobookatallinthatcase,salesaremoreconcentratedonbestsellers,butthiscomesatnoexpensetonon-bestsellingtitles,andthenumberofbookspublishedisunaffected.Moreover,itisalsopossiblethatbestsellerlistsincreasedemandforbooks,forexamplebysimplybringingmoreconsumersintothebookstore.Bestsellersandnon-bestsellerscould,inprinciple,becomplementarygoodsifconsumersbuymultiplebookswhentheyvisitthebookstore.Thedataavailableareclearlyinsuf“cienttoprovideaconclusiveanswertothisquestion.Theidealexperimentmightbeoneinwhichalargesetofconsumersmakespurchasesinthepresenceofabestsellerlist,andanothersetofconsumersmakespurchaseswithouthavinganyexposuretothebestsellerlist(eitherthroughthemediaorattheretailoutletitself).BESTSELLERLISTSANDPRODUCTVARIETY2007TheAuthor.Journalcompilation2007BlackwellPublishingLtd.andtheEditorialBoardofTheJournalofIndustrial TheubiquityoftheNewYorkTimeslistmakesitvirtuallyimpossibleto“ndanygroupofbookpurchasersthatresemblessuchacontrolgroup.Whattheavailabledatapotentiallyrevealisindirectevidenceofsubstitutionbetweenbestsellersandnon-bestsellers.Eventhisisdif“cult,however,sincethedatacontainnopricevariationthatwouldenableestimationofcross-priceelasticities.Theonlyuseablevariationistimevariationinthesubjectcompositionofthebestsellerlist.Ifsubstitutionbetweennon-bestsellersandbestsellersisimportant,thenpresumingthatbooksinthesamegenreareclosersubstitutesthanbooksindifferentgenressalesofnon-bestsellingbooksshoulddeclinewhenthebestsellerlistiscomprisedofbooksinsimilargenres.Forexample,salesofanon-bestsellingdetectivenovelwoulddecreaseinaweekwhenthreedetectivenovelssimultaneouslyhitthebestsellerlist.Inordertocapturethesekindsofsubstitutionpatterns,avariablesummarizingeachbookssimilaritytothecurrentsetofbestsellerswasconstructedbycomparingthebooksgenre(s)(aslistedbyAmazon.com)tothegenresofallbooksonthecurrentbestsellerlist.Speci“cally,thepairwisesimilaritybetweenbooksAandBisde“nedas #ofgenressharedbybooksAandBnumberofgenreslistedforAnumberofgenreslistedforBThismeasureisequaltooneifthetwobooksgenresareidentical,andzeroifthereisnooverlapatall.BookAsaveragesubjectsimilaritytothecurrentbestsellerlististhencomputedas1,whereindexesthecurrentbestsellersbyrank.Table5showstherelativefrequenciesofthegenresinthesampleforallbooksandforbestsellersonly.Asisclearfromthetable,mysteriesandthrillersarethemostcommonbestsellers,and Table5SummaryoflistedgenresPercentofbookslistingthisgenrePercentofbestsellerslistingthisgenreLiterature/Fiction6844Mystery/Thriller3956Romance1119ScienceFiction69Religion23Horror23Basedonbooksgenresaslistedon.Percentagesaddtomorethan100becausebookstypicallylistmorethanonegenre.Amazon.comoftenlistsmultiplegenresforthesamebook:e.g.,Contemporary“ctionandRomance.So)willbelessthanoneunlessbookslistexactlythesameALANT.SORENSEN2007TheAuthor.Journalcompilation2007BlackwellPublishingLtd.andtheEditorialBoardofTheJournalofIndustrialEconomics. romancenovelsarerepresentedmoreheavilyamongbestsellersthanamongbooksoverall.Importantly,thereissubstantialvariationinthecompositionofthebestsellerlistovertime.Figure5showsaseriesofstargraphstoillustratevariationinlistcompositionforasample36-weekperiod.Totheextentthatthereismeaningfulsubstitutionbetweennon-bestsellersandbestsellers,weshouldobservethatsalesrespondtochangesintheaveragesimilarityvariableinducedbyvariationinthelistcomposition.Table6reportsestimatesofamodelanalogoustotheautoregressivemodeloftheprevioussection,butwithsimilaritymeasuresincludedintheInsteadofabusiness-stealingeffect,theestimatedcoef“cientsseemtosuggestacomplementaritybetweenbestsellersandnon-bestsellers.Weeksinwhichthegenre-compositionofthebestsellerlistisclosetoanon-bestselling Figure5VariationinbestsellerlistcompositionovertimeBecausethepurposeistoevaluatethepotentialresponseofnon-bestsellerstochangesintheirsimilaritywithbestsellers,onlybooksthatnevermadethebestsellerlistareusedintheestimation.Thetablereportsspeci“cationswithandwithoutsubject“xedeffects;withoutthese“xedeffects,thecoef“cientsonthesimilaritymeasurescouldpartlyre”ectaveragedifferencesintheprevalanceofcertaingenresonthebestsellerlist,whichwouldobviouslybecorrelatedwithsalesinthatgenre.BESTSELLERLISTSANDPRODUCTVARIETY2007TheAuthor.Journalcompilation2007BlackwellPublishingLtd.andtheEditorialBoardofTheJournalofIndustrial booksgenretendtobeweeksforthenon-bestseller.Consistentwiththe“ndingsinTable4,theeffectseemstobemostpronouncedforbooksappearingonthelistforthe“rsttime:whenthesimilaritymeasureisde“nedonlyforthesubsetofbestsellerswhose“rstappearancewasinthegivenweek,thepositiveeffectofthesimilarityisstronger.Thispatterncouldplausiblyre”ectmultiple-bookpurchases:forexample,weeksinwhichanewromancenovelhitthebestsellerlistdrawmorethantheusualfractionofromanceenthusiastsintobookstores,andtheymaybuyseveralromancenovelswhenvisitingthestore.Althoughtheestimatedcoef“cientsonthesimilaritymeasuresarestatisticallysigni“cant,theyshouldbeinterpretedcarefully.Onecaveattokeepinmindisthatpublisherschoosetheirreleasedatesstrategically.Forexample,thereissomeevidencethatpublishersofnon-bestsellingbookstendtoavoidreleasedatesthatcoincidewiththereleaseofablockbustertitleinthesamegenre.TheregressionsinTable6maybebiasedagainst“ndingbusiness-stealing,becausepublisherschoosetheirbooksreleasedatestoactivelyavoidbusiness-stealing.Thatis,ifbookAwereexpectedtostealbusinessfrombookB,thenbookBsreleasedatewouldbestrategicallydistantfrombookAsrelease,sothatpotentialinstancesofbusiness-stealingarelesslikelytobeobservedinthedata.Anotherthingtokeepinmindwheninterpretingthecoef“cientsisthattheyrepresent(atbest)indirectevidenceofthesubstitutionpatternsbetweenbestsellingandnon-bestsellingtitles,andtheresultsmaynotgeneralizeeasilytoothercategoriesofbooks.Forexample,non-“ctiontitlesonsimilartopicsareperhapslesslikelytoexhibitcomplementarities.Aconsumerlookingtobuyabookoninvestments,forinstance,seemsmorelikelytochoosejustone(instead Table6Regressionestimates:booksalesandsimilaritytocurrentbestsellersIIIIIIIVAvg.Similarity:Allbestsellers.044(.033)(.080)Newonlist(.100)(.102)Weeksout.005.005.003.002(.002)(.002)(.002)(.002)Weekdummies?yesyesyesyesSubjectdummies?nonoyesyes.945.945.951.951#books595595595595#observations14,11514,11514,11514,115Robuststandarderrorsinparentheses.Allexplanatoryvariablesareinteractedwithlaggedsales,asexplainedinthetext.Avg.Similaritymeasureshowsimilarabookistothecurrentsetofbestsellers,basedonthebookslistedgenres.Onlybooksthatnevermadethebestsellerlistareincludedintheestimation.Thenumberofobservationsdoesnotequalthenumberofbookstimes25becausesomeweeksaremissing.ALANT.SORENSEN2007TheAuthor.Journalcompilation2007BlackwellPublishingLtd.andtheEditorialBoardofTheJournalofIndustrialEconomics. ofbuyingmany)thanwouldaconsumerlookingforasuspenseorromancenovel.V.DISCUSSION&CONCLUSIONSBasedonthesalesdataanalyzedhere,itseemsclearthatappearingontheNewYorkTimesbestsellerlisthasadirectimpactonabookssales.Estimatesbasedontwoalternativeidenti“cationstrategiesshowthatsalesincreasewhenabookappearsonthelist.However,themagnitudeoftheeffectismodest,andappearstore”ectspikesinthesalesofbooksthatweresurprisesonthelist(e.g.,booksbynewauthors).GiventheamountofattentionpaidtotheNewYorkTimeslist(andthedesperateschemesoccasionallyemployedbyauthorstosecureapositiononthelist),theestimatesofitsimpactmayseemsurprisinglysmall.However,thepresentanalysisignorestwoeffectsthatcouldbequitesigni“canttoauthorsandpublishers.First,whileappearingonthebestsellerlistmayhaveonlyamodestimmediateimpactonthebookssales,itmaydramaticallyincreasethepopularityoffuturebooksbythesameauthor.Second,paperbacksalesmaybein”uencedbywhetherthehardcovereditionwasabestseller.(Indeed,paperbackversionsofbooksthatwerehardcoverbestsellerstypicallyannouncethatfactprominentlyonthefrontcover.)Althoughtheseeffectscantbemeasuredwiththeavailabledata,anecdotalevidencesuggeststheymaybeimportant.Althoughthereareavarietyofreasonswhythebestsellerlistmightdirectlyin”uencesales,thepatternsinthedataaremostclearlyconsistentwithaninformation-basedexplanation.Retail-levelpromotionscannotrationalizetwoofthestudyscentral“ndings:whereasretailerspromotionspersistforthedurationofabookstermonthelist,theestimatessuggestthattheimpactofappearingonthelististransitory,withthebulkoftheeffectrealizedinthe“rstweek.Moreover,theimpactonsalesismostpronouncedamongrelativelyunknownauthors(newauthorsinparticular),apatternthatfavorsinformationoverpromotionasanexplanationfortheeffect.Theimpactofthelistonsales(andtheconsequentincreaseintheconcentrationofdemandonbestsellers)raisestheinterestingcounterfactualquestion:Wouldmorebooksbepublishedifitwerentforbestsellerlists?Whether(andinwhichdirection)thebestsellerlistshiftsthepublish/no-publishmargindependsonthenatureofsubstitutionbetweenbestsellingandnon-bestsellingtitlesi.e.,aretheextrasalesofbestsellersonesthatInunreportedanalysesusingsalesdatafornon“ctiontitles,coef“cientsonsimilaritymeasureslikethoseinTable6indeedsuggestedapatternofsubstitutionratherthancomplementarity.Non“ctiontitleshavetheadvantageofbeingabletocategorizesubjectsmuchmore“nelythanin“ction,butitwasdif“culttoincludenon“ctiontitlesinthebroaderanalysisbecausethecoverageofnon“ctionbestsellersinmysamplewassparse.BESTSELLERLISTSANDPRODUCTVARIETY2007TheAuthor.Journalcompilation2007BlackwellPublishingLtd.andtheEditorialBoardofTheJournalofIndustrial wouldhaveotherwisegonetolesspopularbooks?Theresultshereofferindirectevidencethatforhardcover“ction,bestsellersandnon-bestsellerswithinthesamegenremayinfactbecomplements:weeksinwhichbooksofaparticulargenre“rstappearonthebestsellerlisttendtobestrong-sellingweeksfornon-bestsellersofthesamegenre.Althoughtooindirecttobeconclusive,thisresultsuggeststhatmarketexpansioneffectsdominateanybusiness-stealingassociatedwithbestsellerlists,sothatbestsellerlistsmayinfactincreasethenumberofbookspublishedinequilibrium.REFERENCESAlexander,P.,1997,ProductVarietyandMarketStructure:ANewMeasureandaSimpleTest,JournalofEconomicBehaviorandOrganization,32(2),207…214.Banerjee,A.,1992,ASimpleModelofHerdBehavior,QuarterlyJournalofEconomics107(3),797…817.Becker,G.andMurphy,K.,2000,SocialEconomics:MarketBehaviorinaSocial,(HarvardUniversityPress).Berry,S.andWaldfogel,J.,2001,DoMergersIncreaseProductVariety?EvidencefromRadioBroadcasting,QuarterlyJournalofEconomics,116(3),1009…1025.Bogart,D.(ed.),2001,TheBowkerAnnual,46thedition.Clerides,S.,2002,BookValue:IntertemporalPricingandQualityDiscrimination,InternationalJournalofIndustrialOrganization,20(10),1385…1408.Cole,D.,1999,TheCompleteGuidetoBookMarketing,(AllworthPress,Chicago).Dixit,A.andStiglitz,J.,1977,MonopolisticCompetitionandOptimumProductAmericanEconomicReview,67(3),297…308.Einav,L.,2003,GrossSeasonalityandUnderlyingSeasonality:EvidencefromtheU.S.MotionPictureIndustry,workingpaper,StanfordUniversity.Epstein,J.,2001,BookBusiness:PublishingPast,Present,andFuture,(W.W.Norton).Greco,A.,1997,TheBookPublishingIndustry,(Allyn&Bacon).Korda,M.,2001,MakingtheList:ACulturalHistoryoftheAmericanBestseller,,(Barnes&NobleBooks,NewYork).Lancaster,K.,1975,SociallyOptimalProductDifferentiation,AmericanEconomic,65(4),567…585.Milgrom,P.andRoberts,J.,1986,PricesandAdvertisingSignalsofProductQuality,JournalofPoliticalEconomy,94,796…821.Reinstein,D.andSnyder,C.,2000,TheIn”uenceofExpertReviewsonConsumerDemandforExperienceGoods,workingpaper,GeorgeWashingtonUniversity.Rosen,S.,1981,TheEconomicsofSuperstars,AmericanEconomicReview,71(5),Suzanne,C.,1996,ThisBusinessofBooks,(Wambtac).Sweeting,A.,2005,TooMuchRock&Roll?StationOwnership,ProgrammingandListenershipintheMusicRadioIndustry,workingpaper,NorthwesternUniversity.Vettas,N.,1997,OntheInformationalRoleofQuantities:DurableGoodsandConsumersWordofMouthCommunication,InternationalEconomicReview,38,Watson,R.,2003,ProductVarietyandCompetitionintheRetailMarketforEyeglasses,workingpaper,NorthwesternUniversity.ALANT.SORENSEN2007TheAuthor.Journalcompilation2007BlackwellPublishingLtd.andtheEditorialBoardofTheJournalofIndustrialEconomics.