Lost on the Web: Does Web Distribution Stimulate or Depress Television Viewing?

Transcription

1 Lost on the Web: Does Web Dstrbuton Stmulate or Depress Televson Vewng? Joel Waldfogel The Wharton School Unversty of Pennsylvana August 10, 2007 Prelmnary comments welcome Abstract In the past few years, YouTube and other stes for sharng vdeo fles over the Internet have vaulted from obscurty to places of centralty n the meda landscape. The fles avalable at YouTube nclude a mx of user-generated vdeo and clps from network televson shows. Networks fear that avalablty of ther clps on YouTube wll depress televson vewng. But unauthorzed clps are also free advertsng for televson shows. As YouTube has grown quckly, major networks have responded by makng ther content avalable at ther own stes. Ths paper examnes the effects of authorzed and unauthorzed web dstrbuton on televson vewng between 2005 and 2007 usng a survey of Penn students on ther tendences to watch televson seres on televson as well as on the web. The results provde a glmpse of the way young, Internet-connected people use YouTube and related stes. Whle I fnd some evdence of substtuton of web vewng for conventonal televson vewng, tme spent vewng programmng on the web 4 hours per week far exceeds the reducton n weekly tradtonal televson vewng of about 25 mnutes. Overall tme spent on network-controlled vewng (televson plus network webstes) ncreased by 1.5 hours per week. I am grateful to the Mack Center at Wharton for fnancal support. I thank Ben Shller and Davd Rothschld for admnsterng surveys, Arjun Shah for dlgent data nput, and Hannah and Sarah Waldfogel for educatng me about YouTube. All errors are mne.

2 In the past few years, YouTube and other stes for sharng vdeo fles over the Internet have vaulted from obscurty to places of centralty n the meda landscape. YouTube allows users to post vdeo fles up to 10 mnutes n length or 100 megabytes n sze avalable to anyone at any tme. Founded n February 2005, YouTube was named Tme Magazne s Inventon of the Year for 2006 and s now among the top 10 stes on the Internet. The fles avalable at YouTube nclude a mx of user-generated vdeo and clps from network or studo-created fare. Networks fear that users wll vew clps on YouTube nstead of watchng through conventonal channels, depressng televson vewng. But YouTube mght nstead help them. Unauthorzed clps are, n one sense, free advertsng for televson shows. Gven the seral nature of televson programmng, wth epsodes as complements for one another, the avalablty of easly accessble clps onlne could stmulate conventonal televson vewng. Another possble effect of unauthorzed web dstrbuton s also bengn for networks: those vewng materal on YouTube mght otherwse not have watched televson. Whle ther vewershp of unauthorzed fles mght not smulate conventonal vewershp, t may represent a gan to consumers wthout offsettng losses to the content creators. As YouTube has grown quckly, major networks have responded by makng ther content avalable at authorzed onlne stes. As of July 10, 2007, abc.com offered full epsodes of about 20 seres. Four recently-ared epsodes of Grey s Anatomy are avalable as are four epsodes of Lost. CBS, NBC, Fox, and Comedy Central also offer ther shows onlne. Network stes dffer from unauthorzed stes n that they offer full 1

3 epsodes rather than the excerpts avalable at, say, YouTube. The network web offerngs also nclude advertsng. The possblty that content avalable on the web may ether stmulate or depress televson vewng s by now famlar from the debates over the effect of fle sharng n musc and moves. 1 Here, as n the examples of musc and moves, the queston of whether authorzed or unauthorzed web dstrbuton stmulates or cannbalzes conventonal televson consumpton s an emprcal one. The seral nature of televson programmng and the attendant complementarty across epsodes of a seres gve the suggeston that web dstrbuton stmulates televson vewng a veneer of plausblty. To measure such effects, I have undertaken a survey, askng students on a college campus about ther habts n vewng televson as well as vdeo on the web. The results of the survey provde a glmpse of the way young, Internet-connected people use YouTube and related stes. I am able to provde answers to the followng questons: 1) What unauthorzed and authorzed stes do people use for vdeo? 2) How much tme they spend watchng varous forms of web vdeo and tradtonal televson? 3) Whch shows are most commonly vewed on the web as opposed to on televson? 4) Do authorzed and unauthorzed web vdeo use dsplace conventonal televson use? The queston of whether YouTube stmulates or cannbalzes nterest n conventonal televson vewng ganed addtonal promnence n March of 2007 when 1 See Lebowtz (2006), Oberholzer-Gee and Strumpf (2007), Rob and Waldfogel (2006, forthcomng), Zentner (2006). 2

4 Vacom sued Google (YouTube s parent) for $1 bllon dollars n damages. Accordng to Vacom s complant, 100,000 of ts clps were avalable at YouTube, and users had vewed these clps 1.5 bllon tmes (see Helft and Fabrkant, 2007). In ts flng Vacom charged that the recent $1.65 bllon acquston prce for YouTube reflects the webste s enormous popularty. YouTube s value, however, s bult largely on the unauthorzed appropraton and explotaton of copyrghted works belongng to others, especally Plantffs. As a result, a large part of YouTube s value s drectly attrbutable to the avalablty of Plantffs copyrghted works on YouTube s webste. 2 The damages that Vacom suggests n ts complant depend on whether clps vewed at YouTube cannbalze or stmulate authorzed vewng of Vacom s propertes. Ths paper proceeds n three sectons. Secton 1 provdes context and theoretcal background. Secton 2 descrbes the underlyng survey and the resultng data used n the study. Secton 3 presents results. A bref concluson follows. I. Context and Theoretcal Background YouTube was founded n February 2005 and has grown very rapdly. Ranked by daly traffc from Alexa.com users, YouTube grew from obscurty a rank of roughly 100,000 th n md 2005 to nearly the top 100 by the start of 2006 (Fgure 1). 3 Between January 2006 and July 2006, monthly unque audence grew from 4.9 mllon to 19.6 mllon vstors, accordng to Nelsen. 4 Tme Magazne named YouTube the Best 2 See Vacom Internatonal Inc. v. YouTube and Google Inc., COMPLAINT FOR DECLARATORY AND INJUNCTIVE RELIEF AND DAMAGES, accessed at July 9, Alexa.com provdes a tme seres on the use of YouTube (and many other stes), compled from users of ther toolbar, and they make these statstcs avalable at Alexa.com. 4 See accessed July 10,

5 Inventon of 2006 n November By late 2006 YouTube was n the top 10, where t has remaned through md Accordng to Alexa.com, YouTube was the 5 th most popular ste on the web on July 10, Apart from flng the lawsut descrbed n the ntroducton, the networks have responded to YouTube s emergence manly by makng ther programmng avalable onlne n varous authorzed forms. In late 2005 ABC and then NBC began sellng epsodes of popular shows on Tunes (Pennngton, 2006). In early 2006 CBS began to offer epsodes of ts popular shows for sale at Google s Vdeo Store (Mlls, 2006). Even as they were tryng to sell epsodes onlne, they were also expermentng wth free authorzed dstrbuton. Late n 2005, networks expermented wth free streamng content at ther webstes. CBS streamed epsodes of Two and a Half Men and How I Met Your Mother on Yahoo! durng a Chrstmas break experment that ncreased vewershp for the two shows on the network, accordng to CBS Entertanment Presdent Nna Tassler (Pennngton, 2006). A few months later the networks changed ther strateges, makng programmng avalable for streamng free onlne (Grand Rapds Press, 2006). In May 2006, ABC became the frst network to offer full-length epsodes onlne (Petrecca, 2006). By the fall of wth the dawn of the televson season all of the major networks were offerng multple shows onlne wthout charge (Zap2It.com, 2006). Many observers 5 See 6 See accessed July 10,

6 vewed ths strategy as a response to the popularty of dgtal recordng devces and pracy ssues that major network broadcasters are facng. 7 The rapd growth of both authorzed and unauthorzed dstrbuton of network s vdeo content on the web between 2005 and 2007 makes ths perod an auspcous one for documentng effects of web dstrbuton on vewershp of network content on televson vewng. The growth n programmng vewed on the web n ths perod reflects newly avalable supply and s exogenous to the users. The changes n the envronment over ths perod subject ntense web users such as on-campus college students to a natural experment: What happens to conventonal televson vewng of people who are now able to vew clps or shows at varous authorzed or unauthorzed webstes? Vdeo obtaned over YouTube dffers from musc or move fle sharng n that YouTube lterally provdes excerpts, whle musc and move fle sharng typcally provde an unauthorzed user wth a complete copy of the fle n queston. Whle a musc downloader mght be moved by hs conscence to purchase a legal copy of a song, he gets very lttle drect beneft from makng ths purchase once he has downloaded a copy. Smlarly, because most move fle sharng s done va coped DVDs, the possessor of a prated copy has lttle to gan, save perhaps a salved conscence, from purchasng a move he has already obtaned wthout payment. Here, though, some of the vdeo avalable for vewng on the web s of poor qualty by desgn. YouTube hosts vdeo, and the clps are no more than ten mnutes or 7 See also accessed July 10,

7 100 megabytes n length. 8 The vdeo of televson content at network webstes s of hgher qualty, but lke the unauthorzed vdeos at YouTube, t s streamed and can only be watched on a computer. Perhaps more mportant than ther poor qualty, epsodes can serve as complements for one another: watchng one epsode or excerpt onlne can elevate a vewer s nterest n seeng other epsodes through authorzed channels (on televson or at a network webste). 9 Hence, t s far more plausble here than n musc or moves to suppose that web dstrbuton use mght stmulate demand for conventonal or authorzed televson vewng. Even f web dstrbuton use does not stmulate conventonal televson vewng, there s reason to thnk - more than n musc or moves that web dstrbuton use would not dsplace conventonal televson vewng. Unauthorzed web dstrbuton allows users to tme-shft ther vdeo vewng even f they lack a dgtal vdeo recorder or VCR. Moreoever, web dstrbuton allows vewng on a computer rather than a televson. Thus t allows vewng by people unavalable when programmng s ared as well as people wthout ready access to a televson set. The queston of whether web dstrbuton stmulates or depresses conventonal televson vewng has two parts, both of whch may be analyzed wth a smple demand curve. Frst, there s the queston of whether the vewng on the web would otherwse have occurred f vewers only had access to conventonal televson. Consder the demand curve n Fgure 1 that shows potental vewers wllngness to pay to vew programmng over conventonal televson. The prce here s largely metaphorc gven that programmng s mostly free, at least at the margn. For our purposes, the prce 8 See accessed July 9, Ths observaton has multple theoretcal antecedents, ncludng Shapro and Varan (1999), Varan (2000), and Bakos, Brynjolfsson, and Lchtman (1999). 6

8 conssts of the wllngness to watch commercals and vewers ablty to schedule ther lves to be avalable when programmng s broadcast. Pror to the avalablty of web dstrbuton, vewers watch q 0 shows, enjoyng some consumer surplus of CS whle networks get revenue proportonal to REV. 10 There s also some deadweght loss (DWL), n that some programmng would be watched f vewers could tme-shft and/or avod commercals. The effect of web avalablty depends on whether users watch programmng they would already have watched (.e. f ther valuatons exceed the prce ). If vewers watch on the web n nstances n whch ther valuatons exceed the prce, web dstrbuton wll cannbalze conventonal vewng. On the other hand, f web vewngs s drawn from the regon of the vewer demand curve where valuatons fall short of the prce p 0, then web dstrbuton wll rase consumpton wthout reducng televson vewng. Because of the seral nature of many programs, watchng an epsode (or an excerpt) on the web can stmulate nterest n watchng other epsodes of the same show on televson. Ths shfts the demand curve out, perhaps rasng the number of nstances n whch people pay for conventonal televson. The hgher valuaton functon rases the number of nstances n whch consumers would be wllng to watch televson, but because they have the opton of vewng on the web nstead, t s not clear whether televson vewng wll rse. The avalablty of programmng on the web has a theoretcally ambguous effect on the use of conventonal televson. 10 The revenue receved by networks s not lterally the regon REVdepcted n Fgure 1. Instead, t s a rectangle that equals the product of q 0 and the prce per vewer for advertsng, whch s generally not equal to the value of the programmng to the vewer. 7

9 II. Data The data for ths study come from a survey admnstered on the campus of the Unversty of Pennsylvana n May The survey was gven to 287 persons on campus, who were asked about ther use of televson and the web for vewng televson programmng. They were asked how often they watch vdeo over the web, whch authorzed (e.g. abc.com) and unauthorzed (e.g. YouTube) vdeo stes they use, how many weekly hours they spent watchng conventonal televson, vdeo at authorzed webstes, and vdeo at unauthorzed webstes durng the televson season. I also asked them to ndcate whch shows they watched. For each seres they ndcated vewng mode (on televson, authorzed web stes, unauthorzed web stes) and frequency (watched through that mode sometmes or frequently). I asked for ths seres-level enumeraton for both the season just endng as well as for the prevous televson season ( ). The resultng data nclude the numbers of shows that respondents frequently or sometmes watch va one of three modes: on televson, on the web va authorzed stes (such as abc.com), or unauthorzed for both the season and the precedng season. Respondents lsted up to sx stes that they use to vew televson programmng. YouTube was by far the most common response: 244 of 283 respondents lsted YouTube frst. Other unauthorzed stes mentoned repeatedly nclude tv-lnks.co.uk, peekvd.com, and bttorrent. Authorzed stes, ncludng abc.com, nbc.com, fox.com, cbs.com, and cnn.com, are also mentoned frequently. 8

10 Whch shows do people watch on whch meda? The shows most frequently vewed on televson nclude Grey s Anatomy, Entourage, and The Daly Show. The shows most frequently vewed va authorzed web stes nclude Grey s Anatomy, Lost, and The Daly Show. The shows most frequently vewed unauthorzed on the web nclude The Daly Show, South Park, and Scrubs. In both seasons, tradtonal televson remans the most popular method of vewng televson content. Respondents report spendng a mean (medan) of 4.5 (3) hours watchng conventonal televson per week, compared wth 3.8 (1) hours of authorzed web vdeo and 3.0 (1) hour of unauthorzed web vdeo, durng the televson season. 11 The numbers of seres watched va each medum mrror the hours breakdowns. Table 1 reports the number of shows vewed ether frequently or sometmes va each of the three modes durng the two seasons. Of 223 persons wth vald data for both seasons, respondents watched an average of 2.62 seres on televson (ether sometmes or frequently) durng the season, compared wth 0.48 shows at authorzed web stes and 0.53 at unauthorzed web stes. Between the two seasons, however, growth n vewng va the web far outstrpped growth n televson vewng (see Table 1). The number of seres (sometmes or frequently) vewed unauthorzed on the web grew by 96 percent, and authorzed web vewng grew even more, by 188 percent. Over the same perod, the number of seres vewed on conventonal televson n the sample grew by only 11 percent. The sharp growth n web vewng between the and seasons among survey respondents reflects the overall growth n YouTube and network-authorzed web dstrbuton over ths perod detaled above. 11 The survey asked about weekly vewng hours only for the most recent ( ) season. 9

11 Durng the season, televson accounted for 72 percent of seres that respondents vewed (sometmes or frequently), and ths share fell to 55 percent n the followng season. In ths sample nearly half of seres vewed are vewed va the web n the second season. Ths change n the use of the web s an mportant source of varaton I wll use to measure the mpact of web dstrbuton on conventonal televson vewng. ABC, FOX, and NBC are the most watched networks n the season among the survey respondents. For every network TV vewng s more prevalent than ether authorzed or unauthorzed web dstrbuton. In the later season ABC, FOX, and NBC reman the most watched on televson among these respondents. Authorzed web vewng rses sharply across the board and most sharply at ABC where respondents watch as many seres on abc.com as on televson. The rapd growth of authorzed web vewng n the sample suggests that the networks have been successful at undermnng the relatve appeal of unauthorzed dstrbuton. Ths should not be too surprsng gven the qualty of the vewng experence at, say, abc.com n contrast to YouTube. Abc.com offers full epsodes wth hgh pcture qualty, albet wth some commercals, whle unauthorzed stes ether stream wth low resoluton (e.g. YouTube) or for stes used less frequently for vewng televson programs requre actual downloadng (e.g. BtTorrent). III. Emprcal Strategy and Results We have two broad strateges for determnng whether the use of new modes of dstrbuton dsplaces tradtonal televson vewng. 12 Frst, we can ask whether those who watch more web vdeo watch less televson, for example usng the measures of 12 These strateges mrror those adopted n Rob and Waldfogel (2006, forthcomng). 10

12 hours spent durng the televson season. An obvous shortcomng wth ths approach s that those who watch a lot of televson seres on one outlet may watch a lot on the other smply because they lke televson seres, rather than because one complements or stmulates the other. Stll, t s worth lookng at smple statstcs: The raw correlaton of weekly hours spent watchng televson and authorzed web vdeo n s 0.07, and the correlaton of televson hours wth hours spend watchng unauthorzed web vdeo s 0.11 (n both cases nsgnfcantly dfferent from zero). The data on seres rather than hours reveal smlarly non-negatve relatonshps among vewng by dfferent modes. Aggregatng the number of seres vewed sometmes and those vewed frequently, TV and authorzed web vewng are sgnfcantly postvely related n both seasons, as are authorzed and unauthorzed web seres vewng n That s, people watchng more seres on televson also watch more at authorzed web stes. In the earler season, people watchng more seres at authorzed web stes also watched more at unauthorzed stes. The postve relatonshps between vewng va dfferent modes are consstent wth unobserved heterogenety (people who lke televson shows enjoy them va multple meda) but could also reflect complementary between vewng among dfferent modes. Dstngushng ntense from casual use seres vewed frequently from those vewed only sometmes va each mode produces somewhat dfferent results. Whle many correlatons reman postve between the number of seres vewed on TV frequently and the number sometmes vewed va authorzed web stes, between sometmes authorzed and frequent televson, between sometmes authorzed and sometmes unauthorzed others are sgnfcantly negatve, notably the relatonshp 11

13 between the number of seres frequently vewed va unauthorzed channels and the number vewed on televson. Gven the loomng concern of unobserved heterogenety n ths context, these negatve correlatons provde nterestng suggestve evdence of substtuton. 1. Cross Sectonal Approach Regressons provde a more systematc way of analyzng these data, va the followng statstcal model: TV = X β + α WF + α WS + ε, where (suppressng the ndvdual s subscrpt) 1 2 TV = number of seres watched (sometmes or frequently) on conventonal televson durng a season, WF = number of seres watched frequently at authorzed or unauthorzed web stes, WS = number of seres watched sometmes at authorzed or unauthorzed web stes, X = characterstcs of the respondent (age, gender, etc.), and ε = unobserved determnant s of the respondents televson vewng. In ths model the unobserved heterogenety problem s the concern s that ε s correlated wth AU and UN, for example because of unobserved tastes for televson vewng. We can decompose TV nto the separate numbers of seres watched frequently (TVF) and sometmes (TVS), where TV = TVF + TVS. We can also dsaggregate frequent and sometme web vewng nto authorzed and unauthorzed components: WF = UNF + AUF, and 12

14 WS = UNS + AUS, where the prefxes UN and AU refer to unauthorzed and authorzed vewng. Then we can also estmate a more flexble set of models: TVF F F = X β + λ UNF + λ UNS + λ AUF + λ AUS + ε, and TVS S S = X β + θ1 UNF + θ 2UNS + θ3 AUF + θ 4 AUS + ε. Ths allows frequent and casual vewng over the web to bear dfferent relatonshps to the numbers of seres vewed frequently, and casually, on televson. Moreover, t allows us to dstngush effects of authorzed and unauthorzed web dstrbuton on televson. Table 2 presents regressons of the numbers of televson seres watched, ether sometmes or frequently, on televson vewng on age, gender, and measures of the numbers of seres watched on authorzed and unauthorzed web stes for the season. The dependent varable n the frst column aggregates both seres watched frequently and sometmes on televson n the season. The number of seres watched on televson bears a negatve and sgnfcant relatonshp to the number of seres vewed on the web (-0.34) and a postve and sgnfcant relatonshp to the number vewed sometmes on the web (0.39). The next two columns examne the two consttuent parts of the total number of seres vewed on televson, those vewed frequently (TVF) and those vewed sometmes (TVS). For the season, web vewngs relatonshps wth total seres watched operate largely through relatonshps wth the number of seres watched frequently on televson: the coeffcents n column (2) are large and sgnfcant, whle the coeffcents n column (3) are ndstngushable from zero. The latter half of Table 2 repeats the exercse of the frst three columns wth ndependent varables dsaggregated to allow dfferent coeffcents on authorzed and unauthorzed web vewng. Whle sgnfcance levels declne, we cannot reject the 13

15 hypotheses of coeffcents on authorzed and unauthorzed web vewng, equaton by equaton. Table 3 repeats the exercse of Table 2 for the latter ( ) season. The frst column s frequent web vewng coeffcent n the total seres televson vewng equaton, n the prevous season, s now and nsgnfcant. The sometme web vewng coeffcent n the same equaton rses from 0.39 n the prevous season to In the latter season, frequent and sometme web vewng contnue to have sgnfcant postve and negatve coeffcents, respectvely, n explanng frequent televson vewng (col 2), but both frequent and sometme web vewng have statstcally sgnfcant coeffcents n explanng sometme televson vewng (col 3). Dsaggregatng web vewng nto authorzed and unauthorzed yelds somewhat dfferent results for relatve to the prevous season. Whle authorzed and unauthorzed web dstrbuton had statstcally ndstngushable coeffcents n the earler season, n unauthorzed dstrbuton measures have consstently more negatve or less postve coeffcents n explanng total and frequent televson vewng. The cross sectonal results contan dstnctve postve and negatve coeffcents. Gven the underlyng concern that unobserved heterogenety would nduce postve relatonshps, t s dffcult to nterpret the postve coeffcents as evdence of complementarty. Stll, two results are nterestng. Frst, the negatve coeffcents n Tables 2 and 3 are more convncngly suggestve of substtuton. Second, the dstnctons between apparent effects of authorzed and unauthorzed web dstrbuton n the latter season suggest that unauthorzed dstrbuton exerts a larger substtutng effect. 14

16 2. Longtudnal Approach A second approach s to use the varaton across ndvduals n the change over tme n ther web and televson vdeo use. That s, we can estmate the model: ΔTVs = α 0 + α1δwfs + α 2ΔWSs + ε s (or ts analogues that dsaggregate frequent and casual vewng), where varables are defned as above except that an observaton s corresponds to a season (s). Ths model dfferences out the unobservable person effect n conventonal televson vewng that s potentally correlated wth vewng at authorzed and unauthorzed webstes. The modelng approach asks whether those wth larger growth n vewng of televson seres at ether authorzed or unauthorzed stes experence larger decreases or smaller ncreases n televson use. Fgure 3 plots the nter-season change n the number of seres watched frequently on televson (on the vertcal axs) aganst that change n the number of seres vewed frequently at unauthorzed web stes (on the horzontal axs). A negatve relatonshp s evdent. The correlaton s and statstcally sgnfcantly negatve. Fgure 4 repeats the exercse replacng unauthorzed wth frequent authorzed webste vewng on the horzontal axs. Agan, a negatve relatonshp s evdent. The correlaton s and sgnfcant. Both frequent authorzed and frequent unauthorzed seres vewng on the web bear sgnfcant negatve relatonshps wth the number of seres vewed frequently on conventonal televson. Table 4 reports regressons of the change n seres watched on televson on changes n seres vewed on the web. The table has the same organzaton as Tables 2 and 3, although varables are now n changes rather than levels. The dependent varable n the frst and second columns s the combned number of seres watched sometmes or 15

17 frequently on televson. Column (1) presents a regresson of the change n the number of seres watched on televson on the change n the numbers of seres vewed frequently and sometmes on the web. Nether frequent nor sometme web vewng has a sgnfcant coeffcent, although the sometme coeffcent (0.17) s nearly twce ts standard error. In column (2) explanng the change n seres watched frequently on televson, frequent web vewng gets a negatve (-0.31) and sgnfcant coeffcent that s smlar to ts cross secton analogues (-0.3 to -0.4), whle sometme web vewng gets an nsgnfcant coeffcent that contrasts wth ts postve and sgnfcant cross sectons analogues (0.3). In column (3) explanng the change n seres watched sometmes on televson both frequent and sometme web vewng measures have postve and sgnfcant coeffcents smlar to ther cross sectonal analogues, partcularly those for The latter half of the table allows authorzed and unauthorzed web vewng to have dfferent coeffcents. For the most part, authorzed and unauthorzed web vewng coeffcents are statstcally ndstngushable, but whle unauthorzed web seres vewng bears no relatonshp to sometme televson vewng, authorzed web vewng has a postve and sgnfcant coeffcent of An addtonal seres vewed frequently va unauthorzed and authorzed web channels reduces the number of seres vewed frequently on televson but ncreases the number of shows vewed sometmes on televson. Evaluatng equatons (2) and (3) from Table 4 at the mean values of frequent and sometmes web vewng n the latter season, reduces the number of seres vewed frequently on televson (TVF) by 0.36, whle they ncrease the number of seres vewed sometmes on televson (TVS) by

18 Determnng the overall effect of the web vewng on conventonal televson requres a way of comparng the amount of vewng assocated wth seres vewed frequently and sometmes. Because we have data on the number of hours respondents spend n all three modes of vewng n the second season, we can run a regresson of, say, weekly hours of conventonal televson vewng on the numbers of seres watched frequently and sometmes, respectvely. That s, we can run the regresson: TVHours = π + π TVF + π TVS + ε The coeffcents π 1 and π 2 from ths regresson map sometme and frequent televson vewng of seres nto weekly hours. They have the nterpretaton that each addtonal seres vewed frequently adds π 1 hours to weekly televson vewng whle each addtonal seres vewed sometmes adds π 2 hours to weekly vewng. Table 5 presents results. The frst two columns report a lnear regresson and a medan regresson, respectvely, for televson. The next 2 columns repeat the exercse for authorzed web vewng, and the fnal two columns for unauthorzed web vewng. The frst column ndcates that an addtonal sometmes-watched seres adds about half an hour to weekly televson vewng, whle an addtonal frequently-watched seres adds over twce as much. The latter number s dffcult to take lterally, snce few programs are broadcast for over an hour per week. Stll, t s comfortng that the frequently coeffcent exceeds the sometmes coeffcent. The overall effect on televson hours can then be estmated as Δ 1 π 2 TVHours = π ΔTVF + ΔTVS. 17

19 Usng the mean estmates of π 1 and π 2 from column 1, the answer s a quarter of an hour (-0.24), whch s 5 percent of the mean value of TV hours, 4.53 hours. Thus, web dstrbuton has on balance, reduced televson vewng hours n ths sample by 5 percent. Separatng the effects of authorzed and unauthorzed web dstrbuton, unauthorzed accounts for 69 percent of the reducton. Whle televson hours declne slghtly, total hours vewng network vdeo content rse overall because of the seres vewng at authorzed and unauthorzed web stes. We can use the hours regressons n the remander of Table 5 to calculate the addtonal hours of assocated wth the authorzed and unauthorzed web vewng n the season. Addtonal hours of weekly vewng of authorzed web vdeo are 1.78, 13 whle the change n unauthorzed web vewng hours s 2.26 hours per week. Thus, overall web vewng rses by 4.04 hours per week, far offsettng the 0.24 hour reducton n weekly televson vewng n the sample. Hours of network-controlled vewng rse by 1.54 per week (a 1.78 hour ncrease n network authorzed web vewng n conjuncton wth a 0.24 hours reducton n tradtonal televson vewng). Much of these respondents vewng would not have occurred absent web dstrbuton. Usng the demand framework ntroduced earler, the addtonal consumpton represents some combnaton of reduced deadweght loss of tradtonal dstrbuton and demand stmulaton effected by web dstrbuton. Concluson 13 AU AU Ths s calculated as π AUF 06 / 07 + π 06 / 07 = 2.14(0.57)+0.68(0.82). 1 2 AUS 18

20 The emprcal lterature on fle sharng n musc and moves has, for the most part, found depressng effects of web dstrbuton on legal sales. We too fnd sgnfcant depressng effects of web dstrbuton on conventonal televson vewng, but we also document largely offsettng postve relatonshps, reflectng complementarty. Overall, conventonal televson vewng s reduced slghtly n ths sample, whle overall vewng of network programmng rses substantally. Hours spent vewng televson programmng overall nearly double wth web dstrbuton. Whle conventonal televson vewng falls by about 5 percent, ths s more than offset by ncreases n tme spent vewng network-authorzed web programmng. The networks own web dstrbuton has smaller but smlar effects as the unauthorzed dstrbuton on conventonal televson vewng. Of course, n the network efforts, the network broadcasts advertsng, so the loss the tradtonal vewng s at least partally offset by onlne ad revenue. Ths study has examned the relatonshp between web vewng and conventonal televson vewng for a small group of meda users on a college campus. Ths s a good populaton for study, gven ts ntense computer use. Relatonshps documented n ths sample may offer a glmpse of relatonshps that wll hold for a more general populaton as broadband contnues to spread. But t bears repeatng that ths sample s not representatve of the US populaton generally. Wder-scale samplng would be very useful for determnng typcal effects of web dstrbuton of televson programmng on televson vewng. 19

21 References Bakos, Yanns, Erk Brynjolfsson, and Douglas Lchtman, 1999, Shared Informaton Goods, Journal of Law and Economcs, 42, pp Davd Blackburn, 2004, On-lne Pracy and Recorded Musc Sales, avalable at Davd Boune, Marc Bourreau, Patrck Waelbroeck, 2005, Prates or Explorers? Analyss of Musc Consumpton n French Graduate Schools, at Helft, Mguel and Geraldne Fabrkant, WhoseTube?; Vacom Sues Google Over Vdeo Clps on Its Sharng Web Ste. New York Tmes, March 14, Hong, Seung-Hyun, 2005, The Effect of Dgtal Technology on the Sales of Copyrghted Goods: Evdence from Napster, at Hu, Ka-Leung and Ivan Png, 2003, Pracy and the Legtmate Demand for Recorded Musc, Contrbutons to Economc Analyss, 2, artcle 11. Lebowtz, Stanley J., 1985, Copyng and Indrect Approprablty, Journal of Poltcal Economy 93, pp Lebowtz, Stanley J., 2006, Fle Sharng: Creatve Destructon or just plan Destructon, Journal of Law and Economcs, 49, pp Oberholzer-Gee, Felx and Koleman Strumpf, forthcomng, The Effect of Fle Sharng on Record Sales: An Emprcal Analyss, Journal of Poltcal Economy. Rob, Rafael and Joel Waldfogel, 2006, Pracy on the hgh C s: Musc Downloadng, Sales Dsplacement, and Socal Welfare n a Sample of College Students, Journal of Law and Economcs 49, pp , forthcomng, Pracy on the Slver Screen, Journal of Industral Economcs. Shapro, Carl and Hal Varan, 1999, Informaton Rules. (Harvard Busness School Press, Cambrdge, MA). Varan, Hal, 2000, Buyng, Sharng, and Rentng Informaton Goods, Journal of Industral Economcs, 48, pp Zentner, Alejandro, 2006, Measurng the Effect of Onlne Pracy on Musc Sales, Journal of Law and Economcs, 49, pp

22 Table 1: Seres Vewed, by Network, Mode and Season web web TV authorzed unauthorzed ABC CBS Comedy Central FOX NBC Other total web web TV authorzed unauthorzed ABC CBS Comedy Central FOX NBC Other total web web % change TV authorzed unauthorzed ABC 24% 221% 50% CBS 0% Comedy Central 8% 100% 89% FOX -14% 300% 83% NBC 17% 333% 280% Other 14% 129% 77% total 11% 188% 96% Note: number of seres vewed here ncludes both those that respondents watch frequently and those that respondents watch sometmes. 21

23 Table 2: Web Vewng and Televson Vewng n the Season (1) (2) (3) (4) (5) (6) TV Seres TV Seres TV Seres TV Seres Watched '05- Watched Watched Watched '05- '06 Freq'ly '05- Sometmes '06 TV Seres Watched Freq'ly '05- '06 TV Seres Watched Sometmes '05-'06 Seres Watched '06 '05-'06 Freq'ly on Web (0.1485)* (0.1459)** (0.1030) Sometmes on Web (0.0985)** (0.0968)** (0.0683) Freq'ly Unauth'd (0.1778) (0.1744)* (0.1234) Freq'ly Auth'd (0.2572) (0.2522) (0.1784) Sometmes Unauth'd (0.1782) (0.1748) (0.1236) Sometmes Auth'd (0.1517)** (0.1488)** (0.1052) male (0.2391) (0.2349) (0.1658) (0.2438) (0.2391) (0.1691) age (0.0305) (0.0299) (0.0211) (0.0307) (0.0301) (0.0213) Constant (0.6526)** (0.6411)** (0.4526) (0.6559)** (0.6433)** (0.4551) Observatons R-squared H 0 : Equal Frequent Auth & Unauth Coeffs (p-val) H 0 : Equal Sometmes Auth & Unauth Coeffs (p-val) H 0 : Both sets equal (p-val) Notes: Standard errors n parentheses. * sgnfcant at 5%; ** sgnfcant at 1%. 22

24 Table 3: Web Vewng and Televson Vewng n the Season (1) (2) (3) (4) (5) (6) TV Seres TV Seres TV Seres TV Seres Watched '06'- Watched Watched Watched '06'- '07 Freq'ly '06- Sometmes '07 TV Seres Watched Freq'ly '06- '07 TV Seres Watched Sometmes '06-'07 Seres Watched '07 '06-'07 Freq'ly on Web (0.0907) (0.0826)** (0.0617)** Sometmes on Web (0.0829)** (0.0755)** (0.0564)** Freq'ly Unauth'd '06-' (0.1360) (0.1237)** (0.0936) Freq'ly Auth'd '06-' (0.1206) (0.1097) (0.0830)* Sometmes Unauth'd '06-' (0.1422) (0.1293) (0.0979)* Sometmes Auth'd '06-' (0.1238)** (0.1126)** (0.0852)* male (0.2581) (0.2352) (0.1756) (0.2728) (0.2482)* (0.1878) age (0.0348) (0.0317) (0.0237) (0.0348) (0.0317) (0.0240) Constant (0.7623) (0.6945) (0.5188) (0.7579) (0.6894) (0.5217) Observatons R-squared H 0 : Equal Frequent Auth & Unauth Coeffs (p-val) H 0 : Equal Sometmes Auth & Unauth Coeffs (p-val) H 0 : Both sets equal (p-val) Notes: Standard errors n parentheses. * sgnfcant at 5%; ** sgnfcant at 1%. 23

25 Table 4: Changes n Web Vewng and Televson Vewng between Seasons (1) (2) (3) (4) (5) (6) Chg # of Seres Watched on Chg # of Seres Watched Chg # of Seres Watched Chg # of Seres Watched on Chg # of Seres Watched Change n Number of Seres Vewed TV Frequently on TV Sometmes on TV TV Frequently on TV Freq'ly on Web (0.0889) (0.0792)** (0.0721)** Sometmes on Web (0.0861) (0.0767) (0.0699)** Freq'ly on Web at Unauthorzed Stes Freq'ly on Web at Authorzed Stes Chg # of Seres Watched Sometmes on TV (0.1378) (0.1229)** (0.1114)** (0.1101) (0.0982)** (0.0890)** Sometmes at Unauthorzed Stes (0.1537) (0.1371) (0.1243) Sometmes at Authorzed Stes (0.1240)* (0.1107) (0.1003)** Constant (0.1433) (0.1276) (0.1163) (0.1439) (0.1284) (0.1163) Observatons R-squared H 0 : Equal Frequent Auth & Unauth Coeffs (p-val) H 0 : Equal Sometmes Auth & Unauth Coeffs (p-val) H 0 : Both sets equal (p-val) Notes: Standard errors n parentheses. * sgnfcant at 5%; ** sgnfcant at 1%. 24

26 Table 5: Translatng Vewng Frequency Into Weekly Hours (1) (2) (3) (4) (5) (6) Weekly Hrs TV Weekly Hrs TV TV Seres Watched Freq'ly '06-' (0.1249)** (0.2660)** TV Seres Watched Sometmes '06-'07 (0.1679)** (0.2360)* Seres Watched Freq'ly Auth'd '06- '07 Seres Watched Sometmes Auth'd '06-'07 Seres Watched Freq'ly Unauth'd '06-'07 Seres Watched Sometmes Unauth'd '06-'07 Weekly Hrs Auth Web Weekly Hrs Auth Web (0.6690)** (0.2517)** (0.6757) (0.1368) Weekly Hrs Unauth Web Weekly Hrs Unauth Web (0.5686)** (0.1388)** (0.6082)** (0.1844)** Constant (0.3901)** (0.4380) (0.9846)* (0.2886)** (0.7543) (0.0000) Observatons R-squared Notes: Odd-numbered columns are lnear regressons. Even numbered regressons are medan regressons wth bootstrapped standard errors. Standard errors n parentheses. * sgnfcant at 5%; ** sgnfcant at 1%. 25

27 Fgure 1: YouTube Traffc Rank from Alexa.com Source: accessed July 10,

28 Fgure 2 27

29 Fgure 3: Change n Frequent Vewng on Web (Unauthorzed) and Televson Chg # of Freq TV Shows Chg # of Freq Unauth Web Shows 28

30 Fgure 4: Change n Frequent Vewng on Web (Authorzed) and Televson Chg # of Freq TV Shows Chg # of Freq Auth Web Shows 29