TheEffectiveness and Targeting of TelevisionAdvertising

RON SHACHAR Tel-AvivUniversity Tel-Aviv,Israel, 69978 [email protected]

BHARAT N. ANAND HarvardBusiness School ,MA 02163 [email protected]

Televisionnetworks spendabout 16 % oftheir revenues on tune-ins, which arepreviews or advertisements for their own shows.In thispaper, we examinetwo questions. First, what isthe informational content inadvertis - ing?Second, is this level of expenditures consistent withprofit maximiza - tion?To answer these questions, we use anew andunique micro-levelpanel dataseton thetelevision viewing decisions of a largesample of individuals, matchedwith data on showtune-in advertisements.The difference in effectivenessof advertisements between ‘‘regular’’ shows(about whichview - ers areassumed to havesubstantial informationa priori) and ‘ ‘specials’’ (about whichthey have very little ) reveals the value of information in advertisementsand the different roles that informationcan play.The number ofexposures for each individual is likely to be correlatedwith their preferences,since networks target theiraudiences. We addressthis endogene - ityproblem by controlling for observed, and integrating the unobserved, characteristics ofindividuals, and find that theestimated effects oftune-ins arestill large. Finally, we findthat actual expenditureson tune-ins closely match thepredicted optimal levels of spending.

1. Introduction Advertisingexpenditures by televisionnetworks are substantial . In 1995,for example, the three majornetworks spent approximately

We would like to thankGreg Kasparian and David Poltrack of CBSfortheir help in obtainingthe datafor this study,John W . Emersonfor comments andassistance in programming,and Barry Nalebuff, Subrata Sen, Idit Shachar,and many of our colleagues forhelpful discussions .

Q 1998M assachusettsInstitute of Technology . Journalof Economics &ManagementStrategy, Volume 7, Number 3, Fall 1998, 363 ] 396 364 Journal ofEconomics & Management Strategy

16% oftheir revenues ontune-in advertisements, 1 morethan twice as much asfirms in otherindustries . The networksalso rank among the topten firmsin the economyin termsof dollars of advertising expenditures .2 Thisraises an obvious question: are networks spend- ing toomuch onadvertising? 3 Indeed, one maysuspect that net- worksunderestimate their advertisingcosts, since these aremostly opportunitycosts; consequently, advertisingexpenditures mightex- ceed the optimalamount . Answering thisquestion requires anunder- standingof the effects oftune-ins onindividuals’ viewing choices, since advertisingrevenues area function of showratings, which are aggregationsof individualviewing decisions . In thisstudy, we esti- matethese effects using anew andunique micro-level panel dataset onthe televisionviewing decisionsof a largesample of individuals, matchedwith data on showtune-ins . The secondquestion that we addressin thispaper isof more general interest:what is the informationalvalue ofadvertising? A spiriteddebate onwhether advertisingis ‘ ‘informative’’ or‘ ‘persua- sive’’ hasgone onfor some time .4 Understandingthe relativeimpor- tanceof informing andpersuading hasboth positive and normative implications . However,distinguishing these effects empirically is difficult. We baseour solution to this identification problem onthe followinglogic: if advertisinghas information content, then the ef- fectsof tune-ins onviewing decisionsshould differ acrossshows accordingto individuals’ prior information about each show . For example, individualsmay possess very littleinformation about the timingand attributes of shows that are aired once— ( specials) com- pared toshows that are aired frequently ( regulars).5 We estimatethe differential effects of tune-ins onviewing decisionsfor these two

1. ‘‘Tune-in’’ usually refers to anadvertisement for atelevision show . Advertise- ments forTV shows represent acost forthe networks, andother advertisements supply their revenues . Theads we focus onin this paper areof the first type . 2. Thenetworks usually air12 minutes of commercials duringeach hour of programming . In1995, they used about2 of these 12minutes ontune-ins fortheir shows. Since advertising revenues represent almost all of the networks’revenues, and tune-ins represent most of their advertisement effort, we proxythe shareof revenues spent onadvertisements as16 % . We then estimate their spendingon advertising in dollars,using these numbersand data on networks’ revenues . 3. Roberts andSamuelson (1988 ) use adifferent approachthan ours to answera similar question in the context of the tobacco industry . 4. Earlywork onthis canbe traced to Galbraith(1967 ) and Solow ( 1967); Tirole (1989, pp. 289] 290) labels these asthe ‘‘partial’’ view versus the ‘‘adverse’’ view of advertising . Formodels dealingexplicitly with the informationaleffects ofadvertising, see Butters( 1977) and Grossman and Shapiro (1984 ). 5. This is similar to the oft-cited distinction between ‘‘search’’ goodsand ‘ ‘experi- ence’’ goods,although not identical, since regularshows havefeatures commonto both types of goods . TheEffectiveness and Targeting of Advertising 365 kindsof shows . If there were noinformational content in advertising, then the effects oftune-ins shouldnot differ acrosssuch shows with different preexisting ‘‘informationstocks .’’Thus,the productvaria- tionin the dataprovides us with a clearway to identify the informa- tionalvalue ofadvertising .6 Ourestimates of the effects of tune-ins maybe biased,since networksgenerally target their tune-ins foreach show towards cer- taingroups of individuals . Forexample, tune-ins forcomedies are moreoften airedduring othercomedies, thus targeting individuals mostlikely towatch comedies in the firstplace . If these unobserved differences in individualpreferences arenot controlled for, the effects of tune-ins onviewership will be upwardbiased . We dealexplicitly withthis endogeneity problem in the estimationbelow . To our knowledge, noother micro-level studyaddresses this issue in any detail.7 Thisis the firststudy focusing explicitly onthe effects oftune-in advertisementson individual behavior .8 There aretwo main advan- tagesof examining the effects ofadvertisingin the TVindustry . First, advertisingis undertaken by firmsusing, in general, variousinstru- ments(price discounts,promotions, etc .) andvarious media (televi- sion,billboards, magazines, etc .). These different formsof advertising maybe substitutesor complementsin consumption 9:forexample, the marginaleffectiveness ofmagazine ads may depend onthe amount ofadvertising generated by othermeans, concurrently or in the past . In orderto accuratelyestimate the marginaleffectiveness ofaparticu- larform of advertising or assess the relativeefficacy ofalternative formsof advertising, one wouldrequire dataon the amountof advertisingvia other forms; this is usually difficult toobtain . This issueis of less concern in the TVindustryfor two reasons: first, almostall advertising by networksis in the formof tune-ins; second, the price of TVservicesis zero from an individual’ s standpoint . Hence, we donot need toconsider issues arising from the existence ofmultiple advertisinginputs .

6. Afew previous studies attempt to distinguish informative effects frompersua- sive effects in advertising, either bysubjective measurementof the informationcontent in ads(Resnik andStern, 1978 ) or by econometric identification techniques similar to ours( Ackerberg,1995 ). All these studies focus onsingle-product returns of advertising . 7. Thetargeting of advertising maybe important in contexts of bothrepeat and nonrepeatpurchase, and hence is likely to bean issue of generalconcern . 8. See,for example, Berndt(1991 ) or Tellis andWeiss ( 1995) for surveys of the literature onthe effects of advertising . 9. Strictly, the advertising itself is not consumed;rather, ads may convey informa- tion thatis of value in makingdecisions . 366 Journal ofEconomics & Management Strategy

Second, previousstudies suggest that the returnsto advertising maydiffer acrossproducts .10 These differences areuseful in revealing the different rolesof advertisingin influencing individual choice . However,it is difficult tocompare results across studies, given nonuniformityin datasets,methodologies, or products under exami- nation. Since ourdataset contains information for multiple TVshows that,in principle, differ substantiallyin their attributes,we can examine suchcross-product differences withina commonsetting . Ourfocus here isonly oncharacterizing the variationbetween regu- larsand specials; we plan tofurther exploitthe productvariation in the datasetin future researchon this issue . The restof the paper isorganized as follows . We firstdescribe the datasetsused, since thisassists in understandingthe structureof the model. In Section 2,we describe the constructionof the dataset andpresent descriptivestatistics on the samplepopulation, viewer- ship patterns,show characteristics, and characteristics of tune-ins . In Section 3,we present the modelused toestimate the effects of tune-ins, anddiscuss identification issues . We discussthe resultsin Section 4,andtheir implicationsfor network strategies in Section 5 . In Section 6,we discussissues for further study .

2. The Datasets We use three datasetsin thisstudy . The firstincludes dataon the viewing choicesof about17,000 individuals during one week in November 1995 . The seconddescribes the attributesof the shows offered tothese individualsby the four leading networksduring this week. The number oftune-ins foreach show and the timeof their airingconstitute the thirddata set . Nielsen Media Researchmaintains a sampleof over5000 house- holdsnationwide by installinga Nielsen People Meter( NPM) for eachtelevision set in the household . Using 1990Census data, the sampleis designed toreflect the demographiccomposition of viewers nationwide. The sampleis revised regularly, ensuring, in particular, thatno single householdremains in the samplefor more than two years. The NPMusesa specialremote control to record arrivals and departuresof individual viewers, as well asthe channel being watchedon each television set . Although the NPMiscalibrated for

10. Forexample, Batraet al . (1995) summarize the variation in estimated returns acrossproducts examinedin prior studies . Forconflicting results regardingthe exis- tence of diminishing returns to advertising, see Simonand Arndt ( 1980) and Rao and Miller (1975). TheEffectiveness and Targeting of Television Advertising 367 measurementseach minute, the dataset available to Nielsen clients providesa recordof whether ornot each viewer tuned intoeach of the alternativesduring eachquarter hour . The rawdataset records whether ornot an individual was watchingtelevision in eachquarter hour, and if so,her choiceof network. An individualwho was watching television, but did not chooseany network, is coded as watching a nonnetworkchannel . Thismight be acable channel,PBS, ora localindependent channel . Eachquarter hour is defined asa timeslot . We use the datafor ,8:00to 11:00 P.M.,forthe five weekdays startingMonday, November 6,1995 . Thus,we observe viewers’choices in 60time slots . Table Ipresentsthe programmingof the four majornetworks (ABC, CBS, NBC,andFOX ) overthis time period. Thisstudy confines itself toeast-coast viewers, to avoid problems arisingfrom ABC’ sMondaynight programming . ABC featuresMonday Night Football,broadcast live acrossthe country; depending onlocal starting and ending timesof the footballgame, ABCaffiliatesacross the countryfill their Mondaynight schedule witha varietyof othershows . Adjusting forthese programming differences by region wouldunnecessarily complicatethis study . Finally, viewerswho never watchedtelevision during weeknight prime timeand those younger thansix years of age areeliminated fromthe sample . Fromthis group, werandomlyselect 1675 individu- als. Their viewing choicesprovide anadequate and rich set of informationthat is used in the estimation . Mostindividuals chose not to watch TV (55 .5% ) in any given timeslot, and many preferred towatcha nonnetworkchannel ( 18 .4% ). NBCwasthe mostwatched network, with 8 .9% ofthe individuals tuned in,on average, in agiven timeslot, followed by ABC(7 .7% ), CBS (6.1% ),andFOX adistantfourth (3 .6% ). The highest-ratedshow during the week under considerationwas ER (amedicaldrama on NBC),witha 21 % share,11 while the CBSnewsmagazine 48 Hours hadthe lowestrating with a 2 .9% share. An interesting aspectof viewing behaviorwas the persistence in choicesof individuals . Two outof any three viewersin anygiven timeslot watched the same channel in the previoustime slot . The degree ofpersistence waseven higher acrosstime slots spanned by asingle show,and varied accordingto demographic characteristics .12

11. ‘‘Share’’ is defined here to bethe fraction of all individuals( including those who donot watch TV) who aretuned in to the show in the given time slot . 12. Shacharand Emerson (1996 ) examine the phenomenonof persistence in more detail, usingthe samedataset . 368 Journal ofEconomics & Management Strategy

DayNetwork 8:00 8:30 9:00 9:30 10:00 10:30

Mon. ABC The Marshal Pro Football: Philadelphia at Dallas

Can’t Hurry CBSThe Nanny Murphy Brown High Society Chicago Hope Love

NBC Fresh Prince In the House Movie: She Fought Alone of Bel-Air

FOX Melrose Place Beverly Hills 90210Affiliate Programming: News

Home Tue.ABC Hudson Street Coach NYPD Blue Improvement

CBS The Client Movie: Nothing Lasts Forever

NBC Wings News RadioFrasier Pursuit of Dateline NBC Happiness

FOX Movie: Bram Stoker’s Dracula Affiliate Programming: News

Wed.ABC Ellen The Drew Grace Under The Naked Prime Time Live Carey Show Fire Truth

CBS Bless this Dave’s World Central Park West Courthouse House

NBC Seaquest 2032 Dateline NBC Law & Order

FOX Beverly Hills 90210 Party of Five Affiliate Programming: News

Thu. ABC Movie: : It’s All in the Game Murder One

CBS Murder, She Wrote New York News 48 Hours

NBC The Single Caroline in E.R. Guy the City

FOX Living SingleThe Crew New York Undercover Affiliate Programming: News

Fri. ABC Family Boy Meets Step by Step Hangin’ With 20/20 Matters World Mr. Cooper

CBSHere Comes the Bride Ice Wars: USA vs The World

NBC Unsolved Mysteries Dateline NBC Homicide: Life on the Street

FOX Strange Luck X-Files Affiliate Programming: News

The purpose ofthis study is to examine the effects oftune-ins onviewing decisions . In Section 3,we present the viewing-choice modelthat we use toestimate these effects . The structureof this modelis similar to that of Shacharand Emerson (1996 ), whichfound thatdemographic characteristics of a show’s castand of individuals, TheEffectiveness and Targeting of Television Advertising 369

table II. SummaryStatistics: Individual DemographicCharacteristics

Variablea Mean Std. Dev

Kids 0.0794 0.2704 Teens 0.0627 0.2425 Gen-X 0.2400 0.4272 Boom 0.2764 0.4474 Old 0.4191 0.4936 Female 0.5319 0.4991 Male 0.4681 0.4991 Family 0.4304 0.4953 Adult 0.8579 0.3492 Income 0.8333 0.2259 Educat 0.7421 0.2216 Urban 0.4149 0.4929 Basic 0.3642 0.4813 Premium 0.3588 0.4798 a Definitions(dum myvariables,u nlessotherwi sespecifie d):‘‘Kids’’ :individual isbetwee ntheages of 7 and 11. ‘‘Teens’’: individualisbetwee ntheages of 12 and 17 . ‘‘Gen-X’’ :individual isbetw eenthe ages of 18 and 34. ‘‘Boom’’:individual isbetwee ntheages of 35 and 49 . ‘‘Old’’ :individual isage 50 orolder . ‘‘Female’’ :individualisa female . ‘‘Male’’: individual isa male . ‘‘Family’’ :individual livesin ahousehold with afemaleolder than 18 and herkids . ‘‘Adult’’ :individual isolder than 18 . ‘‘Income’’:thereare six levelsof incom eonthe u nit interval . ‘‘Educat’’ :thereare five categoriesof educationon the unit interval . ‘‘Urban’’ :individual livesin anurbanarea . ‘‘Basic’’ :individual hasbasic cable . ‘‘Premium’’:individual has premiumcable .

aswell asfive showattributes —action,comedy, romance, suspense, andfiction— are importantin explaining viewing decisions . We briefly describe eachof these variablesbelow . Table IIdefines allthe individual demographicvariables and presentstheir summarystatistics . Show demographicsare described in Table III. In Table IV,we present the variousattributes for each show,each of whichis normalized to lie between 0and1 . These attributeswere subjectively measuredby four researchassistants whoboth watched the entire tapesof the week’s shows,and used preexisting knowledge aboutthe showsin constructingthese mea- sures.13

13. Fordetails onthe construction of these attribute measures,see Shacharand Emerson(1996 ). 370 Journal ofEconomics & Management Strategy

table III. SummaryStatistics: Cast Demographics of Shows

Show Cast Mean Std. Dev.

Gen-X 0.3438 0.4787 Boom 0.2813 0.4532 Family 0.1719 0.3802 Male 0.4531 0.5017 Female 0.2500 0.4364 Black 0.0727 0.2603

Definitions(all dummyvariables):‘‘Gen-X’’: them ain charactersin theshow are betwee ntheages of 18 and 34. ‘‘Boom’’:themain characte rsin theshow are betw eenthe ages of 35 and 49 . ‘‘Family’’ :them ain charactersin theshow are m embersof a family . ‘‘Male’’ :them ain charactersin theshow are m ale . ‘‘Female’’ :them ain charactersin theshow are fem ale . ‘‘Black’’ :them ain charactersin theshow are black .

table IV. Show Attributes

Show Network Action ComedyFiction RomanceSuspense FreshPrince of Bel -Air NBC 0.11 1.00 0.67 0.50 0.20 Melrose Place FOX 0.33 0.00 0.67 0.75 0.80 The Marshal ABC 0.56 0.00 0.67 0.38 0.60 The Nanny CBS 0.00 1.00 0.67 0.63 0.20 Can’t Hurry Love CBS 0.00 1.00 0.67 0.88 0.20 In The House NBC 0.22 1.00 0.67 0.25 0.20 Pro Football ABC 1.00 0.14 0.00 0.00 1.00 BeverlyHills , 90210 FOX 0.22 0.43 0.67 1.00 0.60 SheFought Alone NBC 0.67 0.29 0.33 0.75 0.80 MurphyBrown CBS 0.11 1.00 0.67 0.50 0.20 HighSociety CBS 0.00 1.00 0.67 0.38 0.20 ChicagoHope CBS 0.67 0.14 0.33 0.63 0.80 Roseanne ABC 0.00 1.00 0.67 0.75 0.20 Bram Stoker ’s Dracula FOX 0.67 0.00 1.00 0.63 0.80 Wings NBC 0.11 1.00 0.67 0.75 0.20 The Client CBS 0.22 0.14 0.33 0.25 0.60 NewsRadio NBC 0.00 1.00 0.67 0.38 0.20 HudsonStreet ABC 0.00 1.00 0.67 0.88 0.20 NothingLasts Forever CBS 0.33 0.00 0.67 0.63 0.80 Home Improvement ABC 0.22 1.00 0.67 0.75 0.20 NBC 0.00 1.00 0.67 0.63 0.20 Pursuitof Happiness NBC 0.00 1.00 0.67 0.88 0.20 Coach ABC 0.00 1.00 0.67 0.63 0.20 Dateline NBC NBC 0.22 0.00 0.00 0.00 0.40 NYPD Blue ABC 0.56 0.14 0.33 0.63 0.80 Bless ThisHouse CBS 0.00 1.00 0.67 0.50 0.20 Ellen ABC 0.00 1.00 0.67 0.25 0.20 TheEffectiveness and Targeting of Television Advertising 371

table IV. (Continued)

Show Network Action ComedyFiction RomanceSuspense BeverlyHills , 90210 FOX 0.22 0.43 0.67 1.00 1.00 Seaquest 2030 NBC 1.00 0.00 1.00 0.13 0.80 TheDrew Carey Show ABC 0.11 1.00 0.67 0.00 0.20 Dave’s World CBS 0.00 1.00 0.67 0.50 0.20 Dateline NBC NBC 0.11 0.00 0.00 0.00 0.80 Partyof Five FOX 0.00 0.29 0.67 1.00 0.20 Grace UnderFire ABC 0.00 1.00 0.67 0.38 0.20 Central Park West CBS 0.22 0.14 0.67 0.75 0.80 TheNaked Truth ABC 0.11 1.00 0.67 0.88 0.20 Law andOrder NBC 0.67 0.43 0.33 0.13 1.00 Courthouse CBS 0.33 0.29 0.67 0.88 0.80 Prime Time Live ABC 0.78 0.14 0.00 0.25 0.80 Murder, She Wrote CBS 0.44 0.14 0.67 0.38 0.80 LivingSingle FOX 0.00 1.00 0.67 0.75 0.40 Columbo: It’sAllin the Game ABC 0.44 0.43 0.67 0.75 0.80 Friends NBC 0.00 1.00 0.67 0.88 0.60 The Crew FOX 0.00 1.00 0.67 0.63 0.20 TheSingle Guy NBC 0.00 1.00 0.67 0.88 0.20 Seinfeld NBC 0.11 1.00 0.67 0.63 0.60 New York News CBS 0.56 0.14 0.33 0.50 0.60 New York Undercover FOX 0.89 0.29 0.33 0.50 1.00 Caroline inthe City NBC 0.00 1.00 0.67 1.00 0.20 48 Hours CBS 0.22 0.00 0.00 0.25 0.40 ER NBC 1.00 0.29 0.33 0.50 0.80 Murder One ABC 0.33 0.00 0.33 0.50 0.80 StrangeLuck FOX 0.78 0.29 1.00 0.50 0.80 FamilyMatters ABC 0.22 1.00 0.67 0.50 0.20 UnsolvedMysteries NBC 0.44 0.00 0.33 0.00 1.00 CBSFridayNight Movie CBS 0.33 1.00 0.00 0.50 0.20 BoyMeets World ABC 0.22 1.00 0.67 0.50 0.20 IceWars: USA vsTheWorld CBS 0.78 0.14 0.00 0.00 0.60 X-Files FOX 1.00 0.29 1.00 0.25 1.00 Step by Step ABC 0.00 1.00 0.67 0.75 0.20 Dateline NBC NBC 0.22 0.14 0.00 0.25 0.40 Hangin’ with Mr. Cooper ABC 0.22 1.00 0.67 0.50 0.20 Homicide:Life on theStreet NBC 0.67 0.29 0.33 0.13 1.00 20 r 20 ABC 0.22 0.13 0.00 0.13 0.40

The variableTune Injt indicateswhether there wasa tune-in advertisementfor a show jt (i.e.,ashowon network j in agiven time slot t ) in eachtime slot during the week . Obviously,there areno tune-ins fora showin anytime slot subsequent towhen itis aired . Table Vpresentsthe showsordered by their number oftune-ins . 372 Journal ofEconomics & Management Strategy

table V. Show Tune-Ins

Show Network Day Tune-Ins The Nanny CBS Monday 0 The Marshal ABC Monday 0 Melrose Place FOX Monday 0 FreshPrince of Bel -Air NBC Monday 0 Pro Football ABC Monday 1 Home Improvement ABC Tuesday 2 Roseanne ABC Tuesday 2 Coach ABC Tuesday 2 BeverlyHills , 90210 FOX Monday 2 FamilyMatters ABC Friday 2 Wings NBC Tuesday 2 NYPD Blue ABC Tuesday 2 X-Files FOX Friday 2 MurphyBrown CBS Monday 2 HudsonStreet ABC Tuesday 2 Can’t Hurry Love CBS Monday 2 In The House NBC Monday 2 ChicagoHope CBS Monday 2 Ellen ABC Wednesday 3 Bram Stoker ’s Dracula FOX Tuesday 3 Dateline NBC NBC Wednesday 3 BoyMeets World ABC Friday 3 Seaquest 2030 NBC Wednesday 3 Step by Step ABC Friday 3 UnsolvedMysteries NBC Friday 3 Hangin’ with Mr. Cooper ABC Friday 3 Grace UnderFire ABC Wednesday 3 HighSociety CBS Monday 3 TheNaked Truth ABC Wednesday 3 Dateline NBC NBC Friday 3 Courthouse CBS Wednesday 4 StrangeLuck FOX Friday 4 NewsRadio NBC Tuesday 4 The Client CBS Tuesday 4 Homicide:Life on theStreet NBC Friday 4 Frasier NBC Tuesday 4 New York Undercover FOX Thursday 4 LivingSingle FOX Thursday 4 Law & Order NBC Wednesday 4 SheFought Alone NBC Monday 4 TheDrew Carey Show ABC Wednesday 4 Bless ThisHouse CBS Wednesday 4 NothingLasts Forever CBS Tuesday 5 Columbo: It’sAllin the Game ABC Thursday 5 Pursuitof Happiness NBC Tuesday 5 Dateline NBC NBC Tuesday 5 TheEffectiveness and Targeting of Television Advertising 373

table V. (Continued)

Show Network Day Tune-Ins BeverlyHills , 90210 FOX Wednesday 5 The Crew FOX Thursday 5 Central Park West CBS Wednesday 5 Murder, She Wrote CBS Thursday 6 Dave’s World CBS Wednesday 6 Partyof Five FOX Wednesday 6 Prime Time Live ABC Wednesday 6 CBSFridayNight Movie CBS Friday 6 Seinfeld NBC Thursday 6 Friends NBC Thursday 6 TheSingle Guy NBC Thursday 7 IceWars: USA vsTheWorld CBS Friday 7 Caroline inthe City NBC Thursday 7 Murder One ABC Thursday 8 New York News CBS Thursday 8 20 r 20 ABC Friday 8 ER NBC Thursday 9 48 Hours CBS Thursday 10

Since the datacontain information only ontune-ins thatwere aired onprime timestarting on Monday,it is not surprising thatthe shows onMonday have fewer tune-ins thando shows aired later in the week. Interestingly, the three showswith the largestnumber of tune-ins were airedin the sametime-slot, Thursday at 10:00 P.M. This mayindicate the strategicuse oftune-ins by networks .

The variableCount i jt indicates the number oftimes that an individual i wasexposed toa tune-in fora show jt. Thisvariable was createdby matchingthe Nielsen dataon individuals’ viewing choices

withthe informationin Tune In jt. In principle, the effect of tune-ins canthen be estimatedby examining the effects ofthis variable on individualviewing decisions,after controlling for other characteris- tics. The resulting estimateswill be biasedin general, however . The followingexample illustratesthe sourceof this bias . Eight tune-in advertisementsfor the ABCnewsmagazine 20 r 20 airedduring the followingtime slots and shows: Monday 10:30 ( MondayNight Football );Tuesday9:45 (Coach);Tuesday10:15 ( NYPD Blue);Wednesday 9:15 (Grace UnderFire );Wednesday 10:15 ( Prime Time Live);Thursday9:15 ( Movie: Columbo);Thursday10:15 ( Murder One);andFriday 9:45 ( Hangin’ with Mr. Cooper). In Table VI, we estimatea probitmodel to test the hypothesisthat individuals who 374 Journal ofEconomics & Management Strategy

table VI. An Exampleto Illustrate Targeting a

Variable Coefficient StandardError Lead-In 1.4938 0.1371 MondayNight Football 0.0321 0.1666 Coach 0.3149 0.1544 NYPD Blue 0.1452 0.1561 Grace UnderFire y 0.0920 0.1489 Prime Time Live 0.9942 0.1388 Columbo y 0.0267 0.2048 Murder One 0.7413 0.2089 Constant y 1.7985 0.0663 No. of observations 1675 Loglikelihood y 386.44

a Dependentvariable :thedecision to watch 20 r 20.

were exposed toany of these tune-ins hada higher propensity to watch 20 r 20. The dependent variableis equal toone if aperson watched 20 r 20,andzero otherwise . The explanatoryvariables are: (1) adummy variableset equal toone if the personwas watching ABCin the previoustime slot (i .e.,at9:45 on Friday ), andzero otherwise(this captures the ‘‘lead-in’’ effect) ,and(2 ) foreach tune-in, adummy variableequal toone forindividuals who watched the tune-in andzero for those who did not . The estimationresults reveal that not all tune-ins were effective . Forexample, viewersexposed tothe tune-in while watching NYPD Blue didnot have a higher propensity ex post to watch 20 r 20. On the otherhand, some tune-ins —thoseaired during Coach, Prime Time Live, and Murder One,forexample —were indeed effective . However, these resultsdo not necessarily indicate that the advertisements during these showswere effective, since there isan alternative explanation . The networksprobably donot choose the timingof their tune-ins randomly;rather, they mayair the adfor a showduring othershows with similar characteristics in orderto target a particular audience. One ofthe adsfor 20 r 20 occurredduring the ABCnews magazine PrimeTime Live . AsTable VIshows,being exposed toanad during PrimeTime Live hadthe strongestexplanatory power among the exposure variables,increasing the viewing probability by almost 21% (with a t-value ofabout 7 ). Thismay merely indicate,however, thatsome viewers may like newsmagazines more than others and thushave a higher ex ante propensity towatch both PrimeTime Live and 20 r 20. Since suchpreferences areunobserved, anindividual’ s TheEffectiveness and Targeting of Television Advertising 375 exposure totune-ins for 20 r 20 airedduring PrimeTime Live is endogenous14 ;hence, the estimateof its effect isbiased . Tosummarize, since the networksair their tune-ins during showswith a viewing audience thathas a higher ex ante propensity towatch the promotedshow, the effect ofexposure toan ad on viewing decisionsreflects boththe effectiveness ofthe adand the targetingof the network . We explicitly dealwith this endogeneity problem in the modelbelow .

3. Model,Estimation, and Identification

In eachtime slot t,individual i makesher viewing choice Cit from amongsix options . She mayeither chooseto watch a particularshow onanyof the four networks,watch nonnetwork (including cable) TV, ornotwatch TV (i .e.,pursue someoutside alternative ). Eachof these alternativesis indexed by j, with j s 1indexing the outsideoption, j s 2, . . . ,5correspondingto the four networksABC, CBS, NBC, andFOX, respectively, and j s 6denoting cable TVandother non- networkchannels . An individual i isassumed to derive utility fromalternative j in time slot t given by Ui jt. Time slotsare defined every 15minutes . The structurewe imposeon Ui jt isgiven by the following:

Ui jt s h t q Zjt( Xi b q a i )

r r 2 ( ) q( t 1 Count i jt q t 2 Count i jt ) 1 y Special jt

s s 2 q( t 1 Count i jt q t 2 Count i jt ) Special jt v 4 q( Yi, net, t D ) I Ci, t y 1 s j q e i jt for j s 2, . . . , 5,

v 4 Out Ui , out, t s h out q Yi, out, tg out q d out I Ci, t y 1 s 1 q a i q e i, out, t , v 4 Non Ui , non, t s h non q Yi, non, tg non q d non I Ci, t y 1 s 6 q a i q e i, non, t , where Xt is an l = l diagonalmatrix of individualdemographics, Zjt is a 1 = l rowvector of showcharacteristics, Special jt isa dummy

14. Basedon the characteristics ofdemand, the networks probablychoose bothhow manytune-ins to airfor each show, andwhere to locate( ortarget ) these tune-ins . Most previous studies ofadvertising ignore the targetingissue, andfocus onthe endogene- ity problem arisingfrom the choice of numberof ads . While this is clearly anissue of concern when usingaggregate-level data,the use of individual-level dataavoids this problem, since the optimal numberof tune-ins bynetworks is not chosen separatelyfor eachindividual . 376 Journal ofEconomics & Management Strategy

variablethat is equal toone if the showon network j at time t is a

special,and I isthe indicatorfunction . Yi, net, t isa vectorof show and individualcharacteristics such as show continuity and gender; Yi, out, t and Yi, non, t arevectors of individual characteristics such as income, location,age, education, and family size . The detailedstructure of utilitiesis provided in the Appendix . Weallowfor switching costs in individual behavior,as captured

by the parameters d out, d non andthe vectorof parameters D . For example, individualswatching cable TVinaparticulartime slot may continue towatch it in the subsequent timeslot due toswitching costs,thus inducing statedependence . We estimatethe switching costsin the decisionto watch network TV, watchcable TV,ornot watchTV atall, and allow the switchingcosts to vary for males (relativeto females ) ,andfor continuation shows (i .e.,showsthat span morethan one timeslot ). 15 The matchbetween individualcharacteristics Xi and show characteristics Zjt maybe animportant component of preferences as well. Forexample, malesmay be morelikely towatch sports shows, teens towatch BeverlyHills 90210 ,andwomen to watch shows with female casts,less action, and more romance . Such differences in viewing behaviorare used toidentify the parameters b . The choiceof whichinteraction variables for show and individual characteristics, 16 Zjt Xi,toinclude isessentially ad hoc . These variablesare defined only forthe variousnetwork alternatives, j s 2, . . . , 5.

Individualsalso may differ in their unobserved preferences, a i, overvarious kinds of shows . Forexample, someindividuals may like towatch comedies, while othersprefer dramas . Moreover,such heterogeneity maybe importanteven aftercontrolling for simple

observeddifferences in preferences acrosspeople, ascaptured by Xi. If tune-ins forcomedies are primarily placed in othercomedies, viewerswho are more likely towatch comedies in the firstplace will be exposed tomore tune-ins . Separating the effects of tune-ins from these unobserved differences in preferences isimportant for obtain- ing unbiased estimatesof tune-in effects .

15. Theseinclude age,education, gender,income, andfamily status here . 16. Amore comprehensive set of interaction variables is included in Shacharand Emerson(1996 ). Our choice of which variables to include here is motivated in partby the results ofthe estimation there . TheEffectiveness and Targeting of Television Advertising 377

We assumea simple discretedistribution for these unobserved preferences, where individualsare any of K types. Thus

a withprobability I 1 p1 a 2 withprobability p2 a s . i í . . a withprobability 1 y p p . J k ls k l v Action ComedyRomance Suspense Here a 1 isa parametervector — a 1 , a 1 , a 1 , a 1 , Fiction a 1 4—thatindicates the preferences ofindividuals of type 1for showscharacterized according to their level of action,comedy, etc; similarly, a 2 represents the preferences oftype-2 individuals,and so on. The parameters a 1, a 2 , . . . , a K areidentifiable if there isany systematicpattern in the viewing decisionsof one group ofindividu- alsrelative to others . Identificationof the probability ofeach of these types in the population, p1, p2 , . . . , pK ,isstraightforward . The focusof this estimation is on the effects oftune-ins on viewing behavior . We measurean individual’ s exposure totune-ins fora given show,Count i jt , simply asthe number ofadsfor that show thatshe isexposed toover the week .17 Aquadraticterm in exposure allowsfor a simple nonlinearstructure for the effects ofcounts on viewing decisions . Differences in the decisionto watch a showamong individualsexposed todifferent countsof tune-ins areused toiden- tify the parameters t 1 and t 2. Notall individuals with preferences ofa given type, a k , will watchexactly the sameshows during the week; hence, there will be variationin the exposure oftune-ins amongindividuals of anygiven type. Forexample, someindividuals of type 1maybe exposed to moretune-ins fora particularcomedy than for others of the same type, simply due toidiosyncratic variation in viewing decisions . Similarly,there will alsobe variationin the exposure totune-ins fora particularshow among individuals of each of the othertypes . This variationallows us to identify the effects oftune-ins, afterallowing

17. Since the tune-in dataare available onlyfor the sameweek asthe shows were aired,there is measurementerror in Count i jt. Forexample, the shows onMonday will havefewer tune-ins thanthose later in the week . Inprinciple, it is possible to estimate adifferent tune-in parameter t forshows airedon different daysof the week . Estimates obtainedduring a preliminary stageof this research—allowing forlinear effects of tune-ins only,and without distinguishing between regularshows and specials—did not reveal significant differences in tune-in effects forshows ondifferent days. Inview of these results, since we areprimarily concerned here with differences between regularshows andspecials, we donot explicitly correct forthis measurement error when constructing the likelihood function . 378 Journal ofEconomics & Management Strategy forindividuals’ preferences tovary in unobserved ways . The results of the estimation,with and without controlling for unobserved het- erogeneity, arepresented in Section 4below . In orderto estimate the variationin the effects oftune-ins accordingto individuals’ prior information about a show,we sepa- rateshows into two categories, specials (new, one-time shows) and regulars. Individualsmay be presumed toknow more about a ’s characteristics—itstime slot, its cast, whether itis a comedy, etc.—thanabout a special . If the effect oftune-ins doesnot depend on their informationalcontent, these effects shouldnot differ between s s r r specialsand regulars, i .e., t 1 and t 2 wouldbe equal to t 1 and t 2. Thus,this provides a sourceof identification of the informational contentin advertisingrelative to other effects, suchas persuasion or signaling( Milgromand Roberts, 1986 ). The informationalcontent of tune-ins maybe oftwokinds . First, tune-ins maysimply increaseawareness about the existence of a show. Second, they mayprovide informationabout the show’s at- tributes.18 Individualsare likely topossess much moreinformation aboutboth the existence andattributes of regular showsthan about thoseof specials . Forregular shows,then, the roleof tune-ins maybe simply toserve asreminders of the show’s existence andits at- tributes. Since mostof this information could easily be conveyed via afew tune-ins, the informationalvalue foran individual would diminishas the number oftune-ins viewed increases . Forspecials, however,tune-ins maybe importantboth in informing people about the existence of sucha showand in conveying informationabout its attributes. Here, asingle tune-in isunlikely tobe nearly aseffective; instead,a seriesof tune-ins maybe necessaryto convey information aboutthe different aspectsof anew show . Forthe samereason, althoughdiminishing marginaleffectiveness oftune-ins shouldapply here aswell, the rateat which these returnsdiminish is likely tobe smallerthan for regular shows . Accordingto thissimple characteriza- tionof the informationalcontent in tune-ins, the marginalvalue of tune-ins shouldbe largerfor regular shows(relative to specials ) at lowlevels oftune-ins, but shoulddecline morerapidly as well . More generally, however,the identificationstrategy is based on the logic thatif tune-ins conveyinformation, their effectiveness islikely to

18. Asimilar distinction between the informationalroles of advertising hasbeen madein the previous literature . Butters( 1977), forexample, focuses onthe role of advertising in conveying informationabout a product’s existence andits price . Gross- manand Shapiro (1984 ) augment this to consider informationabout a product’s other attributes( such aslocation ) aswell . TheEffectiveness and Targeting of Television Advertising 379 differ acrossshows for which viewers possess different preexisting stocksof information . We return toa discussionof these different effects in ourpresentation of the results .

3.1.The Likelihood Function Forthe econometrician,the viewing choiceis probabilistic, since we donot observe e i jt. The randomvariables e i jt areassumed to be independent acrossindividuals i andtime slots t,havingthe general- izedextreme-value distribution

6 1y s 1 r (1 y s ) ( ) e 1 ( e k ) F e 1 , . . . , e 6 s exp y e y p e , ( ( ks 2 ) ) where e j denotesthe vectorof disturbances for choice j (the sub- scriptsfor individuals and timeslots are suppressed here ). As McFad- den( 1978) illustrates, under these conditions,the viewing choice probability is

eU1 ( ) P Cit s 1 s 1y s 1 r (1y s ) U1 6 Uk e q p ks 2( e )

y s 1 r (1 y s ) 1 r (1y s ) Uj 6 Uk ( e ) p ks 2( e ) ( ) P Cit s j s 1y s for j s 2, . . . , 6, 1 r (1y s ) U1 6 Uk e q p ks 2( e )

where Uj s Uj y e j. Thisspecification is commonly referred toas the ‘‘nested multinomiallogit,’ ’ andrelaxes the assumptionof indepen- dence ofirrelevantalternatives (IIA ) imposedby the standardmulti- nomiallogit specification . Forexample, itappears reasonable to assumethat an individual first decides whether ornot to watch TV; conditionalon doing so,she then choosesbetween the variouschan- nels. The nested logitmodel we specify here canbe thoughtof as capturingthis two-level representationof the viewing decision . More- over,since the multinomiallogit specification is nested withinthis model(and obtains when s s 0),one canexplicitly testwhich speci- ficationbetter describesthe data . While the disturbances ( e ) areindependent acrosstime slots, the viewing choicesare not, because ofthe switchingcosts . Thus, the conditionalprobability ofeach viewer’ s historyof choicesfor the entire week, Ci s (Ci1, . . . , CiT ),issimply the productof the condi- 380 Journal ofEconomics & Management Strategy tionalprobabilities of eachof his or her choicesmade at each quarter hour. Thatis, for a type- k person,

T k k fi( Ci< V it ; a , u ) s Õ P( Cit s j< V it , Ci, t y 1; a , u ), (3.1) ts 1

v 4 where V it s Zjt , Xi,Counti jt , Specialjt , Yi, out, t , Yi, non, t , Yi, non, t (all observedindividual and show characteristics as well asthe lagged choices), andthe vector u includes ofallthe parametersin the model otherthan the a ’s. Since we donot observe each viewer’ s type, we integrateout these unobservable preferences . The resulting marginaldistribution is

K ( < ) ( < k ) ( ) f2 Ci V i ; u , a , P s p f1 Ci V i ; u , a ? pk 3.2 ks 1

where P isa vectorof type probabilities, p1, . . . , pK , for K viewer types.

Because the e it j andthe type probabilitiesare independent acrossindividuals, the likelihood function issimply the productof the probabilitiesof each individual’ s historyof viewing choices . The parameters u , a 1, . . . , a K , and P arechosen to maximize the log-like- lihoodfunction given by

N ( ) ( < ) ( ) log L u , a , P s p log f2 Ci V i ; u , a , P , 3.3 is 1 where N denotesthe number ofindividuals . The estimatesof the structuralparameters obtained are dis- cussedbelow .

4. Results Mostof the estimatesin Table VII aremotivated and discussed elsewhere.19 We present thembriefly here andthen turnto a thor-

19. See Shacharand Emerson (1996 ). TheEffectiveness and Targeting of Television Advertising 381

table VII. Effectsof Tune-Ins withoutCorrecting forTargeting Bias a

ParameterCoeff . Std. ErrorParameter Coeff. Std. Error

r t 1 0.4236 0.0238 h ABC y 0.7966 0.0621 r t 2 y 0.0429 0.0047 h CB S y 0.9352 0.0627 s t 1 0.2404 0.0463 h NBC y 0.7288 0.0620 s t 2 y 0.0149 0.0119 h FOX y 1.0097 0.0759 s 0.2759 0.0272 h Sp ecial 0.0234 0.0369 h non9 y 0.1373 0.0322 X Kids b ] 0.2007 0.1148 h non10 y 0.0317 0.0408 X Teens out b ] 0.2448 0.1148 h 8:15 0.6733 0.0914 X Gen X out b ] 0.3703 0.0585 h 8:30 0.6459 0.0862 X Boom out b ] 0.3549 0.0549 h 8:45 0.6849 0.0936 X Old out b ] y 0.0285 0.0473 h 9:00 0.5061 0.0840 B Kids out b ] y 0.3174 0.1295 h 9:15 0.7912 0.0908 B Teens out b ] y 0.2571 0.0952 h 9:30 0.8525 0.0866 B Gen X out b ] 0.1733 0.0541 h 9:45 0.9644 0.0931 B Boom out b ] 0.3569 0.0490 h 10 :00 0.9323 0.0820 B Old out b ] 0.1351 0.0429 h 10 :15 1.2245 0.0965 Fa Fa out b ] 0.3267 0.0526 h 10 :30 1.4854 0.0939 Fa NoFa out b ] 0.0275 0.0466 h 10 :45 1.4315 0.0991 Black Income b ] y 0.2440 0.0424 Fe Fe b ] 0.0939 0.0333 d out 3.7657 0.0563 Fe Ma b ] y 0.0227 0.0371 d non 2.0182 0.0808 M a Fe b ] y 0.1010 0.0338 d net 1.2474 0.0605 M a M a b ] 0.0177 0.0360 d cont 1.8249 0.0744 d sam p y 0.3564 0.0492 g Kids y 0.6079 0.0955 d dram a 0.4200 0.0825 g Teens 0.6241 0.2263 d new s y 0.5545 0.0634 g Ge n X y 1.1669 0.0919 d sport y 0.8934 0.0685 g Boom y 1.2977 0.0929 d MA y 0.1093 0.0172 g Old y 1.5139 0.0863 g Income 0.3087 0.0422 g Education 0.0775 0.0477 g Urban8 y 0.0041 0.0375 g Urban9 y 0.1308 0.0380 g Urban10 y 0.2476 0.0356 non g Basic 0.3949 0.0215 non g Prem ium 0.5182 0.0245 non g Teens 0.6241 0.2263 non g Ma le 0.1499 0.0219 382 Journal ofEconomics & Management Strategy

table VII. (Continued)

ParameterCoeff . Std. ErrorParameter Coeff. Std. Error

A Kids F Old b ] 0.1550 0.2240 b ] y 0.3592 0.1331 A Teens F Female b ] 0.0659 0.2318 b ] 0.1113 0.0656 A Gen X F Education b ] y 0.0736 0.1623 b ] y 0.0984 0.1455 A Boom R Kids b ] 0.0771 0.1613 b ] y 0.2842 0.2102 A Old R Teens b ] 0.0050 0.1445 b ] 0.1819 0.1838 A Fem ale R Gen X b ] y 0.0735 0.0790 b ] 0.2911 0.1444 A Education R Boom b ] 0.3907 0.1726 b ] y 0.0216 0.1436 C Kids R Old b ] 0.5123 0.1108 b ] 0.0351 0.1325 C Teens R Female b ] 0.7446 0.2113 b ] 0.2115 0.0689 C Gen X R Education b ] 0.3239 0.0875 b ] y 0.3061 0.1489 C Boom S Kids b ] 0.3719 0.0894 b ] y 0.3442 0.2191 C Old S Teens b ] 0.1754 0.0827 b ] 0.4525 0.3347 C Female S Gen X b ] y 0.0504 0.0442 b ] 0.2839 0.1550 C Education S Boom b ] 0.3868 0.0840 b ] 0.1793 0.1541 F Kids S Old b ] 0.3692 0.2011 b ] 0.2932 0.1437 F Teens S Fem ale b ] 0.1985 0.2106 b ] y 0.0229 0.0747 F Gen X S Education b ] y 0.4724 0.1462 b ] 0.0938 0.1619 F Boom b ] y 0.2680 0.1479

a Dependentvariable :decision byindividu al i to viewalternat ive j in time slot t.

ough discussionof the effectiveness oftune-in ads . Wefirstreport the resultswithout controlling for differences in unobserved preferences acrossindividuals . First,note that the estimateof s issignificantly different from zero ( s às 0.28, std. error s 0.03). In otherwords, we canreject the simple multinomiallogit model in favorof the nested logitspecifica- tion. The precise structureof the utility andmost of the parameters (h , g , d , and D ) areincluded in the Appendix . Briefly, we find that the utility fromthe outsidealternative is a declining function ofage, andis positive in incomeand education . The utility fromthe nonnet- workalternative is higher forindividuals with basic cable, and still higher forthose who subscribe topremium channels . Men andteens havea higher utilityfrom the nonnetworkalternative as well . We alsofind strongevidence forswitching costs, as evidenced by the parameters d . In particular,the transitionpropensities of viewers appearsto decrease during the finales ofdramas, men appearto switchaway marginally more than women, and older viewers switch slightly lessoften thanyounger viewers . TheEffectiveness and Targeting of Television Advertising 383

The b parameterssuggest that ‘ ‘likes attract .’’In particular, individualsprefer showswhose cast demographics are similar to their own. Forexample, showswith a generation-Xcast are most preferred by generation-Xviewers, and baby boomerslike towatch showswith baby boomersin the cast . Similarly,viewers prefer to watchshows about people of their owngender, andfamilies like showsabout families more than people wholive alone . Finally, low-incomepeople prefer showswith blacks in acentralrole (note thatwe haveused the incomevariable for individuals as a proxyfor their race). Viewers in different age groupsdo not differ much in their preference for‘ ‘action .’’However,younger viewerslike comediesas well asshowswith a high fictionlevel . Generation-Xerstend towatch romanticshows, whereas kids do not . Kidsalso do not like towatch showswith a high element ofsuspense, relativeto other viewers . Finally, womenlike watchingromantic shows more than men; and educatedpeople like actionand comedy, but donot appreciate romance.

4.1.Effects of Tune-Ins Wenowturn to the effect ofthe tune-in variables,which are the focus ofthis study . We find thatthe utility froma showis a positive, concavefunction ofthe number oftimes the individual wasexposed toits ads, indicating that while tune-ins areeffective, they have diminishing returns . The firstexposure toan ad for a regular show increasesthe probability ofwatching the showsby morethan 41 %; the secondexposure increasesthis probability by anadditional 29 %, andthe thirdby about17 %. We interpret the strongresponse to the firstfew adsfor regular showsas an awareness effect: viewersalready knowhow much they like agiven show,and the mainpurpose ofthe adis to remind themof the timingof the show,for example . Note thatone exposure toan ad may not be enough toachieve thiseffect, since viewers,while exposed,may often ignore the televisionduring commercialbreaks . Furthermore,some people mayneed morethan one reminder in ordernot to forget .20 Thus,the secondand the third exposureshave a positive(but smaller) effect onthe individual’s probability ofwatchingthe promotedshow . Figure 1presentsthe effect ofthe Countvariable on the utility fromregular shows(the solidline ) andfrom specials (the dashed line). The difference between these curvesis striking . While the first

20. See alsoOlney, et al . (1991). 384 Journal ofEconomics & Management Strategy

FIGURE 1.EFFECTS OFPROMOTIONS ONVIEWING DECISIONS

adfor a specialis quite effective—itincreases the probability of watchingthe showsby about23 % —itis not as effective asthe first adfor a regular show . Since viewersare not familiar with specials, the firstcouple ofads obviously do not serve as reminders; rather, they arelikely toinform viewers about the show’s attributes,such as itsdegree ofcomedy or action . Therefore, their effect isnot likely to be asstrong as for the regular shows . Forthe samereason, however, their effectiveness maynot diminish as quickly asfor regular shows, given thatthey serve assources of information . Thisinterpretation is stronglysupported by the data,as demonstrated by Figure 1 . The sixthtune-in fora regular showand those following it have anegative effect onthe viewing probability . Thiseffect, previously recognizedin the advertisingliterature, is referred toas wearout. The reasonfor wearout may be either thatafter a while viewersstop attendingto the advertising(Calder and Sternthal, 1976 ), orthat excessiveexposure generates irritation .21

21. Theeffectiveness of exposures mayvary according to their timing aswell; for example, recent exposures mayhave a greatereffect onan individual’ s viewing decisions thanthose further in the past . We havetested this hypothesis andfound no evidence of this formof wearout in the data . TheEffectiveness and Targeting of Television Advertising 385

4.2.Controlling for the Targeting Bias The estimatesof tune-in effectiveness maybe biased,as discussed above,since the variablecapturing an individual’ s exposure totune- insis probably endogenous . In orderto correct for this, we estimate the effect ofads,while allowingfor K-types ofviewers(as outlined in Section 3),eachwith potentially different preferences overthe shows’ attributes(such as comedy and action ). Wereportthe resultsof the estimationwith six types ofviewers in Table VIII .22 Asexpected, the effectiveness oftune-ins decreases when we controlfor unobserved heterogeneity ofpreferences over the showtypes, as demonstratedin Figure 2 . Thisconfirms the biasin ourprevious estimates of the effectiveness of ads,induced by the networks’targeting strategies . However,notice that (1 ) tune-ins are stillstrongly effective, and(2 ) the difference between the effects of tune-ins forspecials and regular showsis similar to that estimated earlier. The sixtypes of viewersdiffer substantiallyin their viewing patterns. These differences canbe illustratedby examining the types’ distinctpreferences andtheir viewing choices . Toinfer these choices, we distributeeach individual in oursample to one ofthe types . The priordistribution for each individual overthe varioustypes isde-

fined by the estimatedprobabilities p1, . . . , pk. Basedon their view- ing historyand using Bayes’s rule, we then estimatethe posterior type probability foreach individual, and assign her tothe group for whichher posteriortype probability isthe highest . The types areordered in Table VIII accordingto their size .23 The largesttype ( 31 .2% ) rarelywatch television . Only 6% ofthem watch the highest-ratedshow for this type, the footballgame . The second largesttype ( 22 .3% ) prefer towatch shows with a high level of suspense andromance, but dislike comedies,relative to other types . Thus,it is not surprising thattheir topfive showsare ER (watched by 36% of them), the CBSTuesdayNight Movie , BeverlyHills 90210 (whichhad, relatively, high levels ofaction and suspense during this week’s episodes), the footballgame, and NYPD Blue. Moreover,each ofthese showsis on a different network,indicating that these viewers arenot loyal to any particular network . Further, although Seinfeld is

22. We did not test fora seventh type, since addingthe fifth andsixth types barely affected the estimated tune-in effects . 23. Group sizes arebased on the estimated probabilities ofan individual belonging to eachtype, p1, . . . , p6 . Theseare calculated directly fromthe estimates m 1, . . . , m 6 in m k K m k TableVIII, where pk s e r(1 q p ks 1 e ) for k s 1, . . . , 6 and m 2 is normalizedto 0 . 386 Journal ofEconomics & Management Strategy

table VIII. Effectsof Tune-Ins after Correcting forTargeting Bias a

ParameterCoeff . Std. ErrorParameter Coeff. Std. Error

r t 1 0.2817 0.0218 h ABC y 0.3568 0.0866 r t 2 y 0.0271 0.0047 h CB S y 0.4686 0.0855 s t 1 0.1515 0.0447 h NBC y 0.2725 0.0866 s t 2 y 0.0017 0.0117 h FOX y 0.5467 0.0943 s 0.3437 0.0267 h Sp ecial y 0.0014 0.0349 h non9 y 0.0766 0.0308 X Kids b ] 0.1874 0.1078 h non10 0.0531 0.0401 X Teens out b ] 0.1981 0.1081 h 8:15 0.6580 0.0920 X Gen X out b ] 0.3529 0.0552 h 8:30 0.6400 0.0881 X Boom out b ] 0.3161 0.0512 h 8:45 0.6649 0.0943 X Old out b ] y 0.0595 0.0448 h 9:00 0.5166 0.0856 B Kids out b ] y 0.3474 0.1209 h 9:15 0.7920 0.0925 B Teens out b ] y 0.2865 0.0885 h 9:30 0.8797 0.0887 B Gen X out b ] 0.1499 0.0510 h 9:45 0.9714 0.0942 B Boom out b ] 0.3187 0.0459 h 10 :00 0.9646 0.0824 B Old out b ] 0.1251 0.0410 h 10 :15 1.2548 0.0962 Fa Fa out b ] 0.2937 0.0518 h 10 :30 1.5415 0.0938 Fa NoFa out b ] 0.0264 0.0457 h 10 :45 1.4911 0.0992 Black Income b ] y 0.2367 0.0392 Fe Fe b ] 0.0670 0.0314 d out 3.8251 0.0538 Fe Ma b ] y 0.0478 0.0362 d non 1.6594 0.0730 M a Fe b ] y 0.1026 0.0317 d net 1.1345 0.0578 M a M a b ] y 0.0008 0.0338 d cont 1.6245 0.0734 d sam p y 0.3111 0.0475 g Kids y 0.9996 0.1416 d dram a 0.3953 0.0790 g Teens y 0.5950 0.2496 d new s y 0.4910 0.0614 g Ge n X y 1.7083 0.1282 d sport y 0.7673 0.0656 g Boom y 1.8195 0.1266 d MA y 0.1056 0.0175 g Old y 2.1334 0.1181 g Income 0.3235 0.0766 g Education 0.1185 0.0856 g Urban8 y 0.0235 0.0475 g Urban9 y 0.1476 0.0484 g Urban10 y 0.2704 0.0470 non g Basic 0.4101 0.0361 non g Prem ium 0.5602 0.0397 non g Teens 0.8986 0.2104 non g Ma le 0.1769 0.0360 TheEffectiveness and Targeting of Television Advertising 387

table VIII. (Continued)

ParameterCoeff . Std. ErrorParameter Coeff. Std. Error

A Kids A ction b ] 0.1937 0.2236 a 1 0.1401 0.1400 A Teens Com edy b ] 0.1650 0.2305 a 1 0.4598 0.1302 A Gen X Fiction b ] y 0.0656 0.1651 a 1 0.2722 0.1275 A Boom Rom ance b ] 0.0630 0.1672 a 1 y 0.3854 0.1356 A Old Su spense b ] 0.0342 0.1516 a 1 0.1979 0.1982 A Fem ale Out b ] y 0.0764 0.0744 a 1 1.7193 0.1494 A Education Non b ] 0.3736 0.1680 a 1 0.9639 0.1517 C Kids b ] 0.1314 0.1492 m 1 0.3378 0.1291 C Teens A ction b ] 0.4152 0.2287 a 3 0.1932 0.1395 C Gen X Com edy b ] y 0.0868 0.1225 a 3 1.1469 0.1266 C Boom Fiction b ] y 0.0483 0.1229 a 3 0.3612 0.1233 C Old Rom ance b ] y 0.3540 0.1139 a 3 y 0.6130 0.1302 C Female Su spense b ] y 0.0645 0.0515 a 3 0.1126 0.1890 C Education Out b ] 0.2572 0.1051 a 3 0.6302 0.1449 F Kids Non b ] 0.2804 0.2003 a 3 y 0.1078 0.1367 F Teens b ] 0.1170 0.1979 m 3 y 0.4366 0.1928 F Gen X A ction b ] y 0.5761 0.1480 a 4 y 0.3724 0.2090 F Boom Com edy b ] y 0.3761 0.1485 a 4 0.5388 0.1713 F Old Fiction b ] y 0.5217 0.1396 a 4 0.6029 0.1707 F Female Rom ance b ] 0.0759 0.0630 a 4 y 0.4624 0.1625 F Education Su spense b ] y 0.1878 0.1401 a 4 0.0708 0.2587 R Kids Out b ] y 0.1527 0.2193 a 4 1.4438 0.1862 R Teens Non b ] 0.4126 0.1959 a 4 y 0.5828 0.2042 R Gen X b ] 0.4142 0.1493 m 4 y 0.6657 0.1739 R Boom A ction b ] 0.1525 0.1524 a 5 y 0.4131 0.1893 R Old Com edy b ] 0.2640 0.1432 a 5 0.9339 0.1451 R Female Fiction b ] 0.2514 0.0677 a 5 0.0462 0.1521 R Education Rom ance b ] y 0.0160 0.1484 a 5 y 0.7290 0.1552 S Kids Su spense b ] y 0.2444 0.2275 a 5 y 0.6439 0.2273 S Teens Out b ] 0.6766 0.3426 a 5 y 0.1076 0.1447 S Gen X Non b ] 0.3690 0.1741 a 5 y 0.1996 0.1359 S Boom b ] 0.3096 0.1777 m 5 y 0.7673 0.2102 S Old A ction b ] 0.3824 0.1688 a 6 0.3193 0.1832 S Female Com edy b ] y 0.0301 0.0749 a 6 y 0.1016 0.1595 S Education Fiction b ] 0.0194 0.1606 a 6 0.0430 0.1499 Rom ance a 6 y 0.2205 0.1850 Su spense a 6 y 0.0609 0.2472 Out a 6 0.3020 0.1872 Non a 6 0.9792 0.1749 m 6 y 0.8578 0.1619 a Dependentvariable :decision byindividu al i to viewalternat ive j in time slot t. 388 Journal ofEconomics & Management Strategy

FIGURE 2.EFFECTS OFPROMOTIONS ONVIEWING DECISIONS their sixthmost popular show,they clearlyfavor dramas —Seinfeld is the only sitcomin their top-ten list . The thirdlargest type ( 14 .4% ) like . Except for ER, all the top15 shows watched by thisgroup aresitcoms . Although this type alsohave a relativelyhigh level of utility fromthe outside alternative,they watchnetwork television frequently —forexample, the tenthhighest ratedshow for this week iswatched by 18 .7% of them. Thisis probably because networktelevision offers exactlythe product—sitcoms —thatthey arelooking for . The fourth(11 .5% ) and the fifth (10% ) types like comedyas well . The fourthtype tend to watchsignificantly less TV thanthe third,whereas the fifth type are lesstolerant toward nonsitcoms than the othertypes —there areno dramasamong their topten shows . The lasttype ( 9 .4% ) like action anddislike comedy . Itis thus not surprising thattheir favoriteshow isthe footballgame (with a 17 % share). Moreover,their high nonnet- workutility is probably due tothe vastoffering of live sportevents on cable TV. Tosummarize, individuals of the firstand last types spend less timewatching network television than the others . Individualsof types 3,4, and 5 almostonly watchsitcoms, and type-2 individuals almostonly watchdramas . These resultssuggest that viewers do not tend toprefer avarietyof shows,and are consistent with the findings TheEffectiveness and Targeting of Television Advertising 389 ofGoettler and Shachar (1996 ). Moreover,since the networksfre- quently aira tune-in fora dramaduring otherdramas and for a sitcomduring othersitcoms, the estimateof the tune-in effectiveness wouldbe upward-biasedwithout controlling for the differences be- tween these types . The differences between the othertypes areof significance aswell . Forexample, type-1 individualsare less likely to watch any televisionshow and thus are less exposed totune-ins . Withoutcontrolling for differences in preferences between these indi- vidualsand others, the relationshipbetween the number ofexposures totune-ins andthe propensity towatch television would also influ- ence ourestimate of t ’s.

5. Are NetworkStrategies Optimal? In thissection, we return tothe questionthat we presented atthe outset:do networks’ expenditures onadvertising exceed the profit- maximizinglevel? Toanswer this, we firstpresent asimple model, whichwe solvefor the optimal(profit-maximizing ) number of tune- insfor each show; then, we comparethis with the actualnumber of tune-ins chosenby the network . Network j shouldchoose the number oftune-ins forshow k thatmaximizes its expected profitfunction . If the networkairs Ck tune-ins—eachof length Lk (in seconds), onaverage— for show k, it losesadvertising fees thatdepend onthe ratingsof the showsduring whichthese tune-ins aired . Here, we donot solve for the network’s decision of when toair each tune-in; thus,we proxythese advertisingfees foreach tune-in by

P ? Rating j ? Lk where P isthe fee thatadvertisers pay foran expected exposure of 24 one viewer forone second . Thus, for Ck tune-ins, the network’s lost fees are P ? Rating j ? Lk ? Ck . We proxyfor Rating j as the average ratingof network j (overall time slots ).

The network’s expected revenues fromairing Ck tune-ins for show k will depend onthe effect ofthese tune-ins onthe ratingsfor that show. Let E(Rating k

P ? ADk ? E(Rating k

24. Note thatan increase in the amountof tune-in time fora show—which depends on Ck —reduces the supply of advertising time, which mayresult in adecline in P as well. We ignore this effect here . 390 Journal ofEconomics & Management Strategy

where ADk isthe length( in seconds) of commercial advertising time availableduring show k.25 Thus,the network’s profitfunction is

p jk s P ? ADk ? E(Rating k

We substitutefor Lk using actualfigures during thisweek foreach show,and base E(Rating k

dE(Rating k < Ck ) r dCk Lk s (5.1) Rating k ADk

Consider,for example, thisoptimization for a . Formost such 1 shows, Lk r ADk s 30 . If the averagenetwork rating (over all its shows), Ratingj, isclose enough tothe expected ratingfor a show,

E(Ratingk )— whichis indeed the casefor most of these shows—then the expressionon the left-hand side in equation(5 .1) gives the percentage effect ofthe marginaltune-in onthe expected ratingof the show. Fora regular show,this was estimated to be 10 % ofthe fourth tune-in, andabout 0 % forthe fifth tune-in; thissuggests that the fourthtune-in wouldbe profitable foralmost all sitcoms, but the fifth would not. Asimilarargument holds for nonsitcom regular showsas well. Withoutfurther distinguishing regular showsaccording to their informationstock (some shows are better knownthan others ) and accordingto the effectiveness oftune-ins acrosssuch shows, we cannotexplain the sourceof actual variation in Ck . (It should be

25. Thisis equalto 6minutes foreach 30-minute segment . 26. Inreality networks donot advertise their shows onthe other networks . This is a rule ofthe game,and we takeit assuch . Examiningthe logic ofthis rule is beyondthe scope of this study . 27. Theoptimal numberof tune-ins forspecials is 8 . However, since the variance of s the estimate of t 2 is high,the variance of the estimated optimal numberof tune-ins is high as well. Notice alsothat while we estimate the effectiveness of tune-ins for specials to beaconcave function, we cannotreject the hypothesis thatthe effectiveness of tune-ins forspecials is alinear ora convex function( i .e.,we cannotreject the s hypothesis t 2 G 0). Thus,we would not like to maketoo much of this result . TheEffectiveness and Targeting of Television Advertising 391 noted,however, that this variation is small .) Ourmodel nevertheless hastwo important predictions . First,regular showsshould have a smallernumber oftune-ins thanspecials . Second, regular shows shouldhave about 4 tune-ins, onaverage . Asit turns out, network strategiesare closely consistent with these implications:the average number oftune-ins forspecials is 5(witha standarddeviation of 1 .4), andfor regular showsis 3 .8(witha standarddeviation of 2 .25). It is notsurprising thatobserved strategies are consistent with our first prediction. Butthe consistencywith our second prediction is reveal- ing, because itindicates that network executives appearto be on targetin assessingthe effect oftune-ins onviewers’choices . The modelwe havepresented here isstylized .28 Nevertheless, these resultsdo suggest that network expenditures onadvertisingare similarto the levels thatmaximize their respective profits .

6. Conclusion While the expenditures ontune-in advertisementsby TVnetworks mayappear excessive, we find thatthey aresimilar to what is predicted by asimple modelof profitmaximization by the networks . Exposureto a maximumof four tune-ins fora showhas a dramatic effect onan individual’ s decisionto watch that show . We also find thatthe effectiveness ofadvertising differs between specialsand regular shows . Since the maindifference between these kinds of showsis in the priorinformation that individuals possess about each, thisresult is indicative of informational content in advertising . While we haveestablished here the value of informationin advertising,it would be interesting toexamine the natureof this informationin moredetail . The differential effectiveness of advertis- ing between specialsand regular showssuggests that advertising conveysat leasttwo distinct types ofinformation—information about ashow’s existence andinformation about its attributes . In order to identify eachof these effects, itwould be useful firstto construct a

28. One issue thatwe havenot explicitly modeled, forexample, is the role of advertisements as signals (see Milgrom andRoberts, 1986 ); accordingto this, shows with manyads are inferred to beof higherquality, which in turn increases the propensity to watch the show . We arenot concerned with the signalinghypothesis, however, forvarious reasons . First, it predicts thatthe effectiveness of adsshould increase with the numberof exposures, contraryto the wearout effect observed in the data. Second,our results rest onthe difference in the effectiveness of specials versus regularshows, not onthe distinction between the persuasive effect andthe signaling effect, which maybe viewed asone of interpretation . Finally,clarifying whether ads aresignals or simply persuasive, thoughlikely to affect the socially efficient level of advertisements, should not changethe implications forprofit maximization . 392 Journal ofEconomics & Management Strategy modelthat explicitly incorporatesthe uncertaintyabout the alterna- tivesin anindividual’ s choiceset as well asthe utilityderived from eachalternative . The datasetwe use mayalso be appropriatein structurallyestimating such a model . The resulting estimatescon- cerning the relativeimportance of each type of informationmay be useful in determining afirm’s strategy,as well asin normative analysis. Finally, ourestimates of the effects oftune-ins allowfor the possibilityof networks targeting particular audiences in scheduling their tune-ins . Since suchstrategies imply thatan individual’ s expo- sure totune-ins maybe correlatedwith her unobserved preferences, estimatesthat do not correct for this endogeneity will be upward biased. Indeed, we find the magnitudeof this bias to be large in the contextof TV tune-ins . Similarbiases are likely toexist in the estimatesobtained from previous studies that analyze the effects of advertisingon the demandfor yogurt, beer, tobacco,and other goods . The methodologywe present in thispaper maybe extended tothese othercontexts .

Appendix Here we present the complete structureof the utilityfunction . First, we define allthe variablesthat we use: v Variablesdefining individual characteristics:

Kids i 7] 11years old at November 1995 Teens i 12] 17 years old Gen X i 18] 34 years old Boom i 35] 49 years old Old i 50years old and over Incomei Onunit interval,0 s less than $10,000, 0.20 s between $10,000 and $15,000, 0.40 s between $15,000 and $20,000, 0.60 s between $20,000 and $30,000, 0.80 s between $30,000 and $40,000, 1 s $40,000and over

Education i Onunit interval,0 s 0] 8yearsgrade school, 0.25 s 1] 3yearsof high school,0 .50 s 4 years ofhigh school,0 .75 s 1] 3yearsof college, 1 s 4ormore years college

Urban8PMi Livesin one of25 largest metropolitan areas, and the timeis between 8:00 P.M. and 9:00 P.M.

Urban9PMi Livesin one of25 largest metropolitan areas, and the timeis between 9:00 P.M. and 10:00 P.M. TheEffectiveness and Targeting of Television Advertising 393

Urban10PMi Livesin one of25 largest metropolitan areas, and the timeis between 10:00 P.M. and 11:00 P.M.

Basici Basiccable service Premium i Basicand premium service Malei Male Femalei Female Familyi Liveswith his or her family v Variablesdefining showcharacteristics:

Gen XShowjt Maincharacters are between the agesof 18 and34 BoomShow jt Maincharacters are between the agesof 35 and49 FamilyShow jt Maincharacters are members ofafamily Male Showjt Maincharacters are male Female Showjt Maincharacters are female BlackShow jt Maincharacters are black v Variablesconcerning showcontinuity:

Continuejt Show continuity(middle ofshow ) Sample jt Firstquarter hour of show Drama jt Lastquarter hour of drama Sports jt Sports show News jt News show HourBreak jt The 9:00and 10:00 breaks for the nonnetwork alternative Next,we present the specific structureof the utilities:

Kids Teens Gen X Ui , out, t s h out, t q Kids i ? g q Teens i ? g q Gen X i ? g

Boom Old q Boom i ? g q Old i ? g

Income Education q Incomei ? g q Educationi ? g

q Urban8PMi ? g Urban8 q Urban9PM i ? g Urban9 v 4 q Urban10PMi ? g Urban10 q d out I Ci, t y 1 s 1

Out q a i q e i, out, t , v 4 Ui , non, t s h non 9 ? I 9 P.M. F t - 10 P.M. q h non 10

? Iv 10 P.M. F t - 11 P.M.4

non non q Basici ? g Basic q Premium i ? g Premium non non v 4 q Malei ? g Male q Teens i ? g Teens q d non I Ci, ty 1 s 6

Non q a i q e i, non , t , 394 Journal ofEconomics & Management Strategy

Special Ui , j, t s h j q h ? Special jt

X Kids X Teens q Gen X Showjt ? (Kids i ? b ] q Teens i ? b ]

X Gen X X Boom X Old qGen X i ? b ] q Boom i ? b ] q Old i ? b ] )

B Kids B Teens q Boom Showjt ? (Kids i ? b ] q Teens i ? b ]

B ] Gen X B ] Boom B ] Old qGen X i ? b q Boom i ? b q Old i ? b )

qFamily Showjt

Fa Fa Fa NoFa ? Familyi ? b ] q (1 y Familyi ) ? b ]

Black Income qBlack Showjt ? Incomei ? b ]

Fe Fe Fe Ma qFemale Showjt ? (Femalei ? b ] q Malei ? b ] )

Ma Fe Ma Ma qMale Showjt ? (Femalei ? b ] q Malei ? b ] )

A Kids A Teens qAction jt ? (Kids i ? b ] q Teens i ? b ]

A Gen X A Boom A Old qGen X i ? b ] q Boom i ? b ] q Old i ? b ]

A Female A Education Action qFemalei ? b ] q Educationi ? b ] q a i )

C ] Kids C ] Teens qComedyjt ? (Kids i ? b q Teens i ? b

C ] Gen X C ] Boom C ] Old qGen X i ? b q Boom i ? b q Old i ? b

C ] Female C ] Education Comedy qFemalei ? b q Education i ? b q a i )

F Kids F Teens qFiction jt ? (Kids i ? b ] q Teens i ? b ]

F ] Gen X F ] Boom F ] Old qGen X i ? b q Boom i ? b q Old i ? b

F Female F Education Fiction qFemalei ? b ] q Educationi ? b ] q a i )

R ] Kids R ] Teens qRomancejt ? (Kids i ? b q Teens i ? b

R Gen X R Boom R Old qGen X i ? b ] q Boom i ? b ] q Old i ? b ]

R ] Female R ] Education Romance qFemalei ? b q Education i ? b q a i )

S ] Kids S ] Teens qSuspensejt ? (Kids i ? b q Teens i ? b TheEffectiveness and Targeting of Television Advertising 395

S ] Gen X S ] Boom S ] Old qGen X i ? b q Boom i ? b q Old i ? b

S ] Female S ] Education Suspense qFemalei ? b q Educationi ? b q a i )

r r 2 ( ) q( t 1 ? Count i jt q t 2 ? Count i jt ) ? 1 y Special jt

s s 2 q( t 1 ? Count i jt q t 2 ? Count i jt ) ? Special jt

q d netq ( d contq d sample ? Sample jt q d news ? News jt

qd drama ? Drama jt q d sport ? Sport jt q d MA ? Malei ) ? Continuejt

v 4 ? I Ci, t y 1 s j q e i jt for j s 2, . . . , 5.

References

Ackerberg,D .,1995,‘ ‘Advertising, Learning,and Consumer Choice in Experience GoodMarkets: An Empirical Examination,’’ WorkingPaper, Boston University . Batra, R., D.R. Lehmann, J. Burke, and J. Pae,1995, ‘ ‘WhenDoes Advertising Havean Impact? AStudyof TrackingData,’ ’ Journal of AdvertisingResearch , 35 (5), 19] 32. Berndt, E.,1991,‘ ‘Causalityand Simultaneity between Advertising andSales,’ ’ in The Practiceof Econometrics ,New York:Addison-Wesley, Chapter8 . Butters, G.R.,1977,‘ ‘Equilibrium Distributions of Salesand Advertising Prices,’’ Reviewof EconomicStudies , 44 (3), 465] 491. Calder, B. and B. Sternthal,1976, ‘ ‘Television Commercial Wearout:An Information Processing View,’’ Journal of MarketingResearch , 13, 173] 186. Galbraith,J .K., 1967, TheNew IndustrialState ,Boston:Houghton-Mifflin . Goettler, R. and R. Shachar,1996, ‘ ‘EstimatingShow Characteristics andSpatial Competition in the Network Television Industry,’’ WorkingPaper Series H,No . 5, Yale School of Management,December . Grossman, G. and C. Shapiro,1984, ‘ ‘Informative Advertising with Differentiated Products,’’ Reviewof EconomicStudies , 51 (1), 63] 81. McFadden, D.,1978,‘ ‘Modelling the Choice of Residential Location,’’ in A . Karlquist et al., eds., SpatialInteraction Theory and Residential Location ,North-Holland,Amster- dam. Milgrom, P. and J. Roberts,1986, ‘ ‘Price andAdvertising Signalsof Product Quality,’’ Journal of PoliticalEconomy , 94, 796] 821. Olney, T., M. Holbrook,and R . Batra,1991, ‘ ‘ConsumerResponses to Advertising: The Effects of Ad Content,Emotions, and Attitude toward the Ad onViewing Time,’’ Journal of Consumer Research , 17 (4), 440] 453. Rao, A.G. and P.B. Miller, 1975,‘ ‘Advertising r SalesResponse Functions,’’ Journal of AdvertisingResearch , 15 (2), 7] 15. Resnik, A. and B.L. Stern,1978, ‘ ‘An Analysis of InformationContent in Television Advertising,’’ Journal of Marketing , 41, 50] 53. 396 Journal ofEconomics & Management Strategy

Roberts, M.J. and L. Samuelson,1988, ‘ ‘An Empirical Analysis of Dynamic,Nonprice Competition in anOligopolitistic Industry,’’ RandJournal of Economics , 19 (2), 200] 220. Shachar, R. and J. Emerson,1996, ‘ ‘How Old ShouldSeinfeld Be?’’ WorkingPaper, Yale School of Management . Simon, J. and J. Arndt,1980, ‘ ‘TheShape of the Advertising Response Function,’’ Journal of AdvertisingResearch , 20 (4), 11] 28. Solow, R.,1967,‘ ‘TheNew Industrial Stateor Son of Affluence,’’ PublicInterest , 9, 100] 108. Tellis, G.J. and D.L. Weiss, 1995,‘ ‘Does TVAdvertising Affect Sales? TheRole of Measures,Models, and Data Aggregation,’ ’ Journal of Advertising , 24 (3), 1] 11. Tirole, J., 1989, TheTheory of IndustrialOrganization ,Cambridge,MA: The MIT Press .