You are on page 1of 15

Converting Pirates Without Cannibalizing Purchasers: The Impact of Digital Distribution on

Physical Sales and Internet Piracy


Author(s): Brett Danaher, Samita Dhanasobhon, Michael D. Smith and Rahul Telang
Source: Marketing Science, Vol. 29, No. 6 (November-December 2010), pp. 1138-1151
Published by: INFORMS
Stable URL: http://www.jstor.org/stable/40959556
Accessed: 14-01-2016 12:35 UTC

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/
info/about/policies/terms.jsp
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content
in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship.
For more information about JSTOR, please contact support@jstor.org.

INFORMS is collaborating with JSTOR to digitize, preserve and extend access to Marketing Science.

http://www.jstor.org

This content downloaded from 14.139.237.35 on Thu, 14 Jan 2016 12:35:42 UTC
All use subject to JSTOR Terms and Conditions

Marketing Science

infilili
iM.

Vol.29,No. 6, November-December
2010,pp. 1138-1151
ISSN0732-2399
1eissn1526-548X
1101290611138

doi 10.12877mksc.1100.0600
2010INFORMS

PiratesWithoutCannibalizing
Converting
Purchasers:
The ImpactofDigitalDistribution
on
PhysicalSales and Internet
Piracy
BrettDanaher

Massachusetts
02481,bdanaher@wellesley.edu
Wellesley
College,Wellesley,

SamitaDhanasobhon,MichaelD. Smith,RahulTelane

HeinzCollege,CarnegieMellonUniversity,
15213
Pittsburgh,
Pennsylvania
mds@andrew.cmu.edu,
|sdhanaso@andrew.cmu.edu,
[
rtelang@andrew.cmu.edu

ofdigitalchannelsformediadistribution
has raisedmanyimportant
availability
questionsformarketers,
whetherdigitaldistribution
channelswillcannibalizephysicalsales and whetherlegitimate
digital
notably,
distribution
channelswill dissuade consumersfromusing (illegitimate)
digitalpiracychannels.We address
thesetwo questionsusingtheremovalof NBC contentfromApple's iTunesstorein December2007,and its
and we analyzethe
in September
restoration
2008,as naturalshocksto thesupplyoflegitimate
digitalcontent,
DVD
and
the
store.
this
on
demand
BitTorrent
channels
Amazon.com
of
shock
through
piracy
impact
and Amazon.com,documenting
levelsof
To do thiswe collectedtwo largedata sets fromMininova.com
piracyand DVD sales forbothNBC and othermajornetworks'contentaroundtheseevents.We analyzethese
model and findthatNBC's decisionto removeits contentfromiTunesin
data in a difference-in-difference
Thisis
December2007is causallyassociatedwithan 11.4%increasein thedemandforNBC's piratedcontent.
twiceas
roughlyequivalentto an increaseof48,000downloadsa day forNBC's contentand is approximately
largeas thetotallegalpurchaseson iTunesforthesame contentin theperiodprecedingtheremoval.We also
decreasein piracyforthesame contentwhenit was
findevidenceof a smaller,and statistically
insignificant,
at
we see no changein demandforNBC's DVD content
2008.Finally,
to theiTunesstorein September
restored
channelon iTunes.
Amazon.comassociatedwithNBC's closingor reopeningofitsdigitaldistribution
distribution
channels;cannibalization
Keywords:Internet;
piracy;digitaldistribution;
Publishedonlinein
History:Received:April29,2009;accepted:June16,2010;processedby DuncanSimester.
Articlesin AdvanceOctober 20, 2010.

We can't competewithfree.That'san economicparadigm

- JamesGianopulos,Co-Chairman,
thatdoesn'twork.
Twentieth
CenturyFox (quotedin Thompson2003)
You'll neverstop [piracy].Whatyou have to do is com-

petewithit.- SteveJobs,CEO, Apple Inc. (quotedin


Goodell2003)

1. Introduction

channelshas
ofdigitaldistribution
The development
For
raisedmanyimportant
questionsformarketers.
and moviestudios,twoofthemoreimportelevision
tantquestionsare (1) can paid digitaldistribution
channelsserve as an attractivealternativeto consumptionthrough("free")digitalpiracychannels,
cannibalizeDVD box
and (2) will digitaldistribution
setsales?
Withrespectto thefirstquestion,thequote above
fromJamesGianopulos,co-chairmanof Twentieth
of manyin theindusCenturyFox,is representative
if not impossible,
it
is
that
who
claim
difficult,
try

channels
use paid digitaldistribution
to successfully
to competewitha free(albeitillegal)piracychannel.
SteveJobs,CEO
On theotherside of thisargument,
claimsthatdigitaldistribution
ofAppleIncorporated,
channels,such as Apple's iTunesvideo store,offer
to competewithpiracy
studiosthebest opportunity
the ease and convenienceof
channelsby mimicking
pricepoint.
piratedchannelsat a competitive
Withrespectto thesecondquestion,thereis ample
feel
evidencein thebusinesspressthatDVD retailers
will
channels
thatstudios'distribution
digital
through
cannibalizesales ofDVDs. Forexample,
significantly
in late2006,afterDisneyfinalizeda deal to distribute
itsmoviesthroughiTunes,pressreportsclaimthata
WalmartexecutivevisitedHollywoodStudiosto tell
themthat"it will retaliateagainstthemforselling
movieson Apple'siTunes[store]"(Arango2006).This
reportgoes on to note thatWalmart,whichmakes
up an estimated40% of studios' DVD sales, made
good on this threatby sending"'cases and cases'
of DVDs back to Disney" (Arango2006). Similarly,

1138

This content downloaded from 14.139.237.35 on Thu, 14 Jan 2016 12:35:42 UTC
All use subject to JSTOR Terms and Conditions

PiratesWithoutCannibalizingPurchasers
Danaher et al.: Converting
2010INFORMS
Science29(6),pp. 1138-1151,
Marketing

Target,whichmakes up an estimated15% of DVD


themnotto
sales,senta letterto studiosthreatening
followDisneyintodigitaldistribution
(McBrideand
aftersendingthisletter,
Marr2006) and, reportedly
ofinter"ordereditsstoresto takedown a multitude
nal signssteering
customers
to Disneyproducts"and
displayswith
replacedDisney'sendcappromotional
(Menn2006).
displaysforDisney'scompetitors
However,whereasthese questionshave received
and in thepress,we
muchdiscussionin theindustry
are not aware of any empiricalstudiesthataddress
thedegreeto whichdigitaldistribution
ofmediacontentaffects
demandforphysicalcontentand demand
forInternet
piracy.The goal of thispaper is to anathat
lyzethesequestionsthrougha quasi-experiment
occurredon December1, 2007. During August of
withthe Apple
2007,NBC expresseddissatisfaction
iTunesstore'spricingpolicy.WhereasNBC (and other
in pricmedia companies)wanted more flexibility
was
a
ing, Apple
enforcing one-price-fits-all
policy
acrossnearlyall television
episodesforsale on iTunes.
Whennegotiations
brokedown,NBC announcedthat
it would removeall of its contentfromiTuneson
December1, 2007, a significant
move since NBC
40%
of
all
video
contenton the
reportedly
supplied
iTunes store.1In response,Apple refusedto offer
NBC's fall2007seasonforsale starting
in September
2007,and on December1, 2007,Apple removedall
olderNBC contentfromiTunes.
In our analysis,we use this event as an exogenous shockto the legal digitalsupply of all older
seasons of NBC television.This contentwas generally availableforsale on DVD, on iTunes(priorto
December1), and through
piracy,and thuswe study
a marketwitha physicalsales channel,a digitalsales
channel,and a piracychannel.We accountforthe
of a timetrendby comparingchangesin
possibility
piracyand DVD sales two weeksbeforeversustwo
weeks afterDecember1 forNBC's competitor
networks,ABC, CBS, and Fox (all of whichcontinueto
offertheircontenton iTunesand thus receivedno
shockon December1). We thencontrastthis time
trendwith the change in piracyand Amazon.com
DVD sales forNBC, arguingthatany differences
for
NBC contentafterDecember1, over and above the
difference
forsimilartelevision
was caused
networks,
the
removal
of
NBC
content
from
iTunes.
by
Finally,
we ask whetherour findings
are upheldin a second
on September
9,2008,whenNBC restored
experiment
all ofitscontentto theiTunesstore.
We findthatthe removalof NBC contentfrom
iTunescaused an 11.4% increasein piracyfor its
whichcorresponds
to 27 morepirateddowncontent,
loads perday perepisode,or about48,000totaladditionalpirateddownloads per day. This numberis
1See Msnbc.com
(2007).

1139

nearlytwiceas largeas the daily numberof downloads of theseepisodes on iTunesin the two weeks
priorto December1, implyinga fixedcostassociated
withthe decisionto pirate:once individualsstartto
pirate,theypiratemorecontentthantheywouldhave
originally
purchased.Thismayalso implya spillover
- thatpiracyofcontenton othernetworks
effect
could
haveincreasedas a resultofNBC's decisiontoremove
itsowncontent
fromiTunes.Althoughwe cannotposbecause of the lack
itivelyidentifythis externality
of an appropriate
our resultsare concounterfactual,
sistentwithsuch an effectbecause non-NBCpiracy
increasedby 5.8% over this timeperiod (and thus
thedocumented11.4%increasein piracymayunderstatethe truedisplacementof piracyby digitaldistribution).
Finally,
althoughstudyingNBC's returnto
iTunesin September2008is complicatedby thestart
of a new season of television,
our evidencesuggests
thattherestoration
ofNBC contentto iTunescauseda
which
smaller,
statistically
insignificant
dropinpiracy,
is also consistent
witha fixedcostto piracy.
In contrastto thestrongcorrelation
betweenlegitimate digital distribution
and piracy,we find no
changein theAmazon.comsales rankofNBC television season box sets in the fourweeks surrounding
December1 relativeto the baselinechangein nonNBC box sets,implyingthatwhile customerswho
cannotpurchasedigitallymayturnto piracy,theydo
not considerDVD box sets- at least those sold on
- as a substitute
Amazon.com
to digitaldownloads.

2. LiteratureReview

As thispaperaddressestheinteraction
betweenlegitimatedigitaland physicaldistribution
channels,as
well as the interaction
betweenlegitimateand illechannels,the paper fits
gitimatedigitaldistribution
into two main literatures:
the marketingliterature
between various distribution
studyinginteractions
channels,and theeconomicsand information
systems
literature
on onlinepiracyofdigitalgoods.
Withrespectto thepiracyliterature,
mostexisting
studiesexaminethe effectof onlinepiracyon physical media sales. Most of thesestudiesanalyze the
impactofpiracyon musicCD sales,witha fewrecent
studiesexaminingtheimpactof piracyon movieor
televisionrevenue.The challengein this literature
is typicallyidentification,
as the correlation
between
sales
and
downloads
of
each
movie
physical
pirated
or song is predominantly
drivenby unobservedhetacrossgoods.
erogeneity
in
this literatureaddress the identificaPapers
tion issue in severaldifferent
ways: throughcrossshocks
to demand,or
variation,
country
exogenous
results.
With
to
variasurvey
respect cross-country
tion,Zentner(2005),Hui and Png (2003),and Peitz

This content downloaded from 14.139.237.35 on Thu, 14 Jan 2016 12:35:42 UTC
All use subject to JSTOR Terms and Conditions

Danaher et al.: Converting


PiratesWithoutCannibalizingPurchasers
Science29(6),pp. 1138-1151,
2010INFORMS
Marketing

1140

and Waelbroeck(2004) use international


panel data
on music sales and pirateddownloads,with each
studyfindingthatpiracydisplacesCD sales to some
extent.
Danaherand Waldfogel
(2008)examSimilarly,
ine theimpactofonlinepiracyofHollywoodmovies
box office
on international
revenueand findevidence
ofdisplacement
ofticketsales by onlinepiracy.
In thecontextofexogenousshocks,one ofthetests
and Strumpf(2007) takes
used by Oberholzer-Gee
in
the
German
school
holidays
systemas exogenous
demandshocksand findslittleor no displacement
associatedwith music piracy.Likewise,Smithand
Telang(2009)use thetelevisionbroadcastof a movie
as an exogenousdemandshockfortheDVD and find
littleorno displacement
ofDVD sales frompiracyfor
moviesbroadcaston television.
Finally,in the contextof surveydata, Rob and
Waldfogel(2006)use surveydata froma population
of collegestudents,askingwhetherindividualswho
piratemusicpurchaseless music,includingcontrols
such as statedvaluationsof the albumsin question
orpersonaltasteformusic.Rob and Waldfogel(2007)
use a similarapproachto studythe effectof movie
of moviessuch as thepiracyon paid consumption
DVD rental,and DVD purchase.In
aterattendance,
bothstudies,the authorsfinddisplacementof paid
consumption
by piracy.
Thus, the majorityof existingempiricalstudies
of
in the literature
findsome degreeof substitution
for
unpaid "pirated"consumption paid consumption,2whichraisesthequestionofhow firmsshould
ofpiracy.Recent
combatthenegativeeffects
optimally
thisquestion
examined
in
literature
have
the
papers
in thecontextof litigation
againstpirates,protection
of mediacontentthroughdigitalrightsmanagement
and purposefully
(DRM) systems,
damagingtheperoffilesharingnetworks.
formance
ofthefirstanti-piracy
Addressingtheeffectiveness
et al. (2008)
and
Blackburn
tool,
(2004)
Bhattacharjee
examinetheimpactoftheRIAA's legalthreats
against
individualfilesharersduringthesummerof 2003as
withbothsetsofauthorsfinding
a quasi-experiment,
is higher,filesharthatwhenthethreatof litigation
is stillsubstancontent
of
but
declines
availability
ing
Vernik(2009)
tial.In thecontextof DRM protection,
and Sinha et al. (2010) argue thatthe presenceof
DRM may increase
piracyby reducingthe usability
ofthepurchasedfiles,causingconsumerswho otherwise would have purchasedto pirateinstead.In the
of file-sharing
contextof degradingtheperformance
2We notethat
manyanalyticpapershave foundthatpiracyneed
Forexample,Jain(2008)
industries.
tothecopyright
notbe harmful
arguesthatpiracyneed notbe harmfulto theextentthatit may
conserveas a pricediscrimination
device,allowingprice-sensitive
andreducing
sumerstopaynothing
amongother
pricecompetition
consumers.

Christinet al. (2005) studythe impactof


networks,
severaldifferent
on fourpop"poisoning"strategies
ular peer-to-peer
and
findthat
networks
file-sharing
theinjectionof a fewreplicateddecoyscan strategically manipulateusers' perceptionof contentavailabilityin thenetwork.
tool thatmedia comHowever,anotherimportant
panies may be able to use to reducethe impactof
piracyis directly
competingwithpiracychannelsby
channels.To the
addinglegitimate
digitaldistribution
bestofourknowledge,
thereareno papersintheliteraturethatexaminetheinteraction
betweenlegitimate
and illegitimate
distribution
channels,
(piracy)digital
and one contribution
of thispaperis to examinethis
question.
Our question is also related to the interaction
distribution
whichhas been
channels,
amongdifferent
in
the
Forexamstudied
literature.
marketing
widely
show
that
coordinaand
ple, Jeuland
Shugan(1983)
channelsleads to higher
tion betweendistribution
Chianget al. (2002),
profits.Extendingthisfinding,
Chu et al. (2007),and Webb (2002) develop stratechangies forfirmsto managemultipledistribution
In the contextof directdistribution
nels effectively.
channels,Balasubramanian
(1998)uses analyticmodels to show thatthe presenceof directdistribution
channels,includingInternetchannels,yieldshigher
returns
whentheproductis welladaptedto thechanalso note
literature
nel.Otherpapersin themarketing
two channelsare, the
that the more differentiated
less likelytheyare to cannibalizeone another(e.g.,
Friedmanand Furey1999,Viswanathan
2005).
fewpapersin thelitHowever,thereare relatively
eraturethat attemptto directlymeasurethe effect
on physicalchannelsales. One
of digitaldistribution
etal. (2002),whouse a samis Deleersnyder
exception
and
Dutch
British
of
85
newspapersthatadded
ple
findthatwhennewsand
channels
distribution
digital
available
content
make
their
online,theonline
papers
contenthas only a small impacton physicalnewsand Naik (2003)
paper sales. Likewise,Biyalogorsky
disfindthatTowerRecords'additionof an Internet
cannibalize
tributionchannel did not significantly
its retailsales. Withrespectto video distribution,
(2009)uses surveydatatoshowthatauthoWaldfogel
rizedYouTubeviewingoftelevisioncontenthas only
viewon over-the-air
effect
a smallnetdisplacement
the
between
achieve
complementarities
ing and may
twochannels.
aresigthechallengesofidentification
In summary,
to the
in thisdomain.Our paper contributes
nificant
literature
by beingthefirstpaper,to our knowledge,
thesupto use a naturalquasi-experiment
impacting
contentto addresstheidentification
ply oflegitimate
to theliterature
problem.Our paper also contributes

This content downloaded from 14.139.237.35 on Thu, 14 Jan 2016 12:35:42 UTC
All use subject to JSTOR Terms and Conditions

PiratesWithoutCannibalizingPurchasers
Danaher et al.: Converting
2010INFORMS
Science29(6),pp. 1138-1151,
Marketing

by proposinga techniqueto collectdata documenting piracylevels over time and by being the first
estimatehow
paper we are aware of to empirically
interactwith
channels
distribution
digital
legitimate
bothphysicaland piracychannels.
demandthrough

3. Theory

3.1. Digital Distributionand Piracy


of a digital
Theorydoes notclearlypredicttheeffect
in a digitalpiratechansaleschannelon consumption
nel.On one hand,iTunespurchases(byfarthedominantlegitimate
digitalchannelforvideo purchases)
and pirateddownloadsare similarin thatbothprovidehigh-quality,
usuallyfastfiledownloadsthatcan
a telebe viewedon a computeror,withsomeeffort,
visionor portablevideo device.3
Giventhesesimilarities,
one mightask whyanyone
a digitaldistribution
wouldpurchasethrough
channel
ifpiracyis free.Forthisto occur,theremustbe some
nonfinancial
costto piracy.Thereare severalpossible
intowhichthatcostcould fit:
categories
(i) Theremay be a cost to learn to use BitTorrent,whichwould be akinto a fixedcost (especially
becauseBitTorrent
is generallyconsideredto be easy
to use oncelearned).
(ii) Individuals may experiencemoral qualms
aboutpirating,
whichcouldhave thecharacteristic
of
a fixedcost(see,forexample,Naginand Paternoster
1991in thecontextoffixedmoralcoststo committing
crimes)or a variablecost.
(iii) Individualsmay fearbeing caughtand punished,a cost whichis also variablewithrespectto
downloads.
(iv) It is possible that pirated downloads are
viewed as less convenient(and lowerquality)com(eitherbecauseofthe
paredwithiTunesconsumption
relativeease ofuse ofiTunesversuspiracysitesor the
in qualitythroughpiracysites).This cost
variability
would also be variablewithrespectto thenumberof
downloads.
Giventhisinformation,
itis straightforward
to anahow
the
removal
of
a
sales
channel
lyze
digital
might
considera standarddownimpactpiracy.Specifically,
ward slopingdemand curve formedia downloads
and two cases: one wherethe decisionto purchase
insteadofpirateis drivensolelybythehighfixedcost
of piracyand one wherethe decisionto purchaseis
drivensolelyby thehighvariablecostsofpiracy.
If thecoststo piracyare largely"fixed"in nature
and this fixedcost is sufficiently
large,then consumerswill forgoinvestingthe fixedcost of piracy
3Piratedfiles,of course,tendto be easierto shareor use on a
ofdevices(inlargepartbecauseofthelackofdigitalrights
variety
restrictions
on playback),
whereasiTunesdownloads
management
tendtohavemoreconsistent
quality.

1141

and purchasea certainquantityof legitimatedigital content.However,once the legitimatechannel


is removed,some consumerswill investthe fixed
cost associatedwithpiracyand, owingto the lower
marginalcosts of piracy,will increasetheiroverall
On theotherhand,ifthecostto
levelofconsumption.
piracyis variablein nature(i.e., each piratedcontentimposesa marginalcost) and the marginalcost
is constantin the numberof piratedepisodes,then
once the legitimatechannelis removed,consumers
willdecrease
ofpiratedcontheirlevelofconsumption
tentrelativeto purchasedcontent.
Thereare, of course,otherpossiblestructures
for
the nonfinancial
cost of piracy,includinga mix of
fixed and variable costs as well as increasingor
decreasingmarginalcost.4However,even in these
one can show thatwhen the digital
environments,
channel
is removed,it is possiblethatan
purchase
individualwould eithernot turnto piracy,beginto
piratea numberofepisodesless thanor equal to the
numberofpurchasesshehad beenmakingon iTunes,
orevenbeginpirating
morecontent
thanshehad ever
we
purchased.In short,because of data limitations,
cannotmodelpiracycostsat an individuallevel;however,we can use thetheoryoutlinedabovetoinferthe
generalnatureofpiracycostsperceivedbyconsumers
by observingchangesin piracyafteriTunesremoval.
an increasein piracyafteriTunesremoval
Specifically,
thatis largerthanthepreremoval
iTunessales would
be consistent
witha fixedcostto piracyamonga substantialnumberof users,eitherthroughlearningor
moralcosts.
On the otherhand, the theorybehindthe substiofiTunespurchasesforDVD box setsis less
tutability
clearthanit is forthepiracy.It is possiblethatdigital goods cannibalizephysicalsales. However,unlike
thecomparison
ofdigitalsales topiracy,
digitalgoods
do not necessarilyhave lower profitmarginsthan
physicalgoods do. Moreover,even if digitalgoods
differenmarginsare lower,theremay be sufficient
tiationbetweendigitaland physicalgoods thatthey
customersegments.Finally,even
appeal to different
ifbox setsand filedownloadsare substitutes
foreach
other,it is possiblethatindividualswho "go digital"
willbe disinclinedto go backto thephysicalproduct
(indeed,thiswould be analogousto a fixedcostassociatedwithbeginningto download/watchtelevision
online).
In summary,
iTunescustomers
have
mayotherwise
been pirates,mayotherwisehave purchasedthebox
nothave consumedthecontent
set,or mayotherwise
4In fact,itis
costis increasing,
possibletoshowthatifthemarginal
thenconsumers
mightmixbehaviors,
piratingsomeepisodesand
others.
purchasing

This content downloaded from 14.139.237.35 on Thu, 14 Jan 2016 12:35:42 UTC
All use subject to JSTOR Terms and Conditions

PiratesWithoutCannibalizingPurchasers
Danaher et al.: Converting
2010INFORMS
Science29(6),pp. 1138-1151,
Marketing

1142

FollowingSmithand Telang(2009),we use thelevel


trackerfiles at
of daily downloads of BitTorrent
for
a
Mininova.orgas proxy piracyactivityon the
programsin our sample. The websiteMininovais
- the filesthat
a searchenginefor torrenttrackers
allow you to linkto othercomputersand download
servesas
BitTorrent
a specificpieceofmediacontent.
a usefulproxyforvideo piracyas it was the most
popular sourceof piratedvideo downloads during
our studyperiod(Smithand Telang2009).Mininova
is a usefulproxyfordownload levels throughthe
BitTorrent
protocolbecause it was the most poputrackersite duringour studyperiod
lar BitTorrent
it posts a largenumberof
to
Alexa.com,6
according
televisiontrackerfiles,and unlikesome othersites,
on the cumulativenumber
it providesinformation
of downloadsforall trackerfilesdownloadedfrom
itssite.
Our piracy data set containsthe daily number
of downloads for5,200 unique televisionepisodes
to roughly75 unique series)starting
(corresponding
November16,2007.The data includetheseriesname,

season number,and episode numberof each televias well as thenumberoftimesthatfile


sionprogram,
was downloadedeach day.We also added indicators
forthenetworkthatowns therightsto theshow,the
genreof theshow,and whethertheshow is a series
thatis stillproducingnew episodes(such as Heroes)
or a "catalog"series(suchas theoriginalStarTrek).
This data set was createdfroma largerdata set
all televisiontrackers
we collectedmonitoring
posted
to Mininova. We collected data at the torrent
level startingin November2007,obtainingroughly
210,000recordsper day and yieldinga data set of
We
for180,000torrents.
over68 millionobservations
this
data
set
extractedthe torrentfilenames from
and interpreted
thefilenamesto code theseries,season,and episodeforour televisiondata. Whena file
containsmultipleepisodes of a televisionshow,we
countedthisas a downloadforeachepisodecontained
inthefile.Becausemultiplefilesfrequently
maptothe
sameepisodeoftelevision(forexample,theremaybe
filesthatcontain,say,season 1,
six different
torrent
we thencollapsedthe
of
4,
Anatomy),
Grey's
episode
data to the episode level by adding the totaldaily
filesmapdownloadsforan episodeacrossall tracker
of the
observations
that
to
episode. Repeated
ping
same contentovera periodof a monthshouldminierrorsin ourdata.
mizeanypotentialmeasurement
Forour analysis,we focuson piracyamongteleviforNBC and itssubsidiaries(USA
sionprogramming
and the Sci-Fi Channel (now Syfy)).We also anafromtheother
programming
lyzepiracyfortelevision
networks
television
ABC,CBS,and Fox- as a
major
control.We removedall contentfromthe 2007-2008
season fromthe data because, as noted above, the
NBC content
beingsoldon iTunespriortoDecember1
onlyincludedepisodespriortothe2007-2008season.7
Thus, our analysiscompareschangesin piracyfor
content.
older"out-of-season"
in ouranalysiswe focuson thetimeperiod
Finally,
two weeksbeforeand two weeks afterDecember1,
2007 (as well as the two weeks beforeand after
September9, 2008) to best isolatethe effectof the
removalof NBC contentfromiTuneson piracy.Our
willbe to comparethechangein piracy
mainstrategy
forNBC contentafterDecember1 to the changein
arguingthatany increpiracyfornon-NBCcontent,
mentalNBC change over and above the non-NBC
to theremovalofNBC content
changeis attributable
fromiTunes.
December1,2007
tonotethatalthough
Itisimportant
was the officialdate of NBC's removalfromiTunes

5As we discussin detailin 4,itis important


to notethatofnecessitywe are askingthesequestionsin relationto older,off-season
formostofouranalysis.
content
6Availablefrom
accessedApril19,2009.
http://www.alexa.com/,

7This is because,
startingin August2008,Apple did not add
any new NBC contentto the iTunesstorein responseto NBC's
thattheywould removeall of theircontentin
announcement
December.

at all. Wehaveshownthatitis eventheoretically


posof contenton iTunescould
siblethattheavailability
displacemorepirateddownloadsthanthenumberof
episodesbeingpurchasedon iTunes.
Followingthisanalysis,we ask the followingtwo
majorempiricalquestions:5
(i) Whathappensto thelevelofpiracyoftelevision
contentwhen thatcontentis removedfromiTunes
(and whenit is returned)?
(ii) Whathappensto DVD sales of televisionseasons sold on the Internetwhen those seasons are
removedfromtheiTunesstore?
The answersto (i) and (ii) also provideevidence
ofiTunespurchasesthatcomefrom
ofthepercentage
otherwisewould-bepirates,the possibilitythatone
network'sdecisionto use (or not use) a digitaldischannelcan influencepiracyof contenton
tribution
cost
and theshapeofthenonfinancial
othernetworks,
curveassociatedwithpiracy.

4. Data

To addressthesequestions,we use paneldata on consumptionofpiratedtelevisioncontentthroughmajor


trackersites,and panel data on sales of
BitTorrent
DVD season box sets at Amazon.com.We describe
thesedata in moredetailbelow.
4.1. Piracy Data

This content downloaded from 14.139.237.35 on Thu, 14 Jan 2016 12:35:42 UTC
All use subject to JSTOR Terms and Conditions

Danaher et al.: Converting


PiratesWithoutCannibalizingPurchasers
Science29(6),pp. 1138-1151,
2010INFORMS
Marketing

and December2 is thefirstentireday on whichthe


iTunesstoreheld no NBC content,Apple actually
begantheremovalprocesson November30 and continuedthroughDecember1. Thus we mightexpect
to see some increasein piracyas earlyas November 30, but we conservatively
removeDecember1
fromthedata and code December2 as thefirst
day of
in
the"postremoval"
data.
the
as
our
data
If,
period
show,piracybeganto increaseas soonas theremoval
began,thenour selectionof thislaterremovaldate
will lead to an underestimate
of the truechangein
caused
the
content's
removal
fromiTunes.
piracy
by
It is also worthnotingthatpast studieson Internetpiracyhave rarelymade use of eventsor "quasibecause theseeventsoftenoccurwith
experiments"
shortnotice,and data collectionon piracycannot
beginsoon enoughto matchthe event.Thus a contribution
of our studyis the methodof data collection,whichallowsus to tracka good indexofpiracy
over timeand analyze thesedata when shocksare
observed.
4.2. DVD Sales Data
To analyzethe effectof the December1 experiment
on DVD sales,we use panel data on sales ranksof
DVD season box sets on Amazon.comforthe same
daterange:November18 through
December15,2008.
We selectedAmazon.comas our reference
pointfor
in
DVD
sales
the
removal
ofa digchanges
following
ital channelbecause Amazon.comhas an estimated
90% shareof onlineDVD sales in the UnitedStates
(Netherby2005), and it seems plausiblethatusers
who are no longerable to purchasetelevisioncontentthroughan Internet
channel(i.e.,iTunes)would
be disproportionately
moreinclinedtopurchaseDVD
box sets fromanotherInternetchannelratherthan
frombrick-and-mortar
retailers.
However,giventhat
Amazon.com
is onlythefourth
largestsellerforDVDs
intheUnitedStates,behindbrick-and-mortar
retailers
and
Best
News
Walmart,
2006),a
Target,
Buy (DVD
conservative
of
our
results
is
that
interpretation
they
apply only to consumptionthroughInternetDVD
channels.
To analyzechangesin DVD sales at Amazon.com,
we collectdaily observations
of the priceand sales
rankforeach DVD Amazonsells.We thenfocuson
DVD box setsoftelevisioncontentforthefourmajor
thetelevisionseries,season,and
networks,
capturing
network
names.
We interpret
the sales rankbased on priorwork8
thathas shown thatthe relationship
betweensales
rankand sales followsa Paretodistribution:
= aRank.
Quantity

(1)

8See Chevalier and Goolsbee


(2003), Brynjolfssonet al. (2003), and
Ghose et al. (2006)

1143

Thus, following the experimentproposed by


Chevalierand Goolsbee(2003),one can estimatethe
of therelationship
betweenAmazon.com
parameters
sales rankand actualsales of theproduct.However,
forthepurposeofthisstudy,we simplynotethatthis
betweenpriceand sales
impliesthattherelationship
rankis bestmodeledas an elasticity.
Priorresearch9
has dealt with the nonlinearrelationshipbetween
Amazon sales and Amazon sales rankby analyzing
the effectof eventsor explanatory
variableson the
is found,theexperimenlog ofsales rank.Ifan effect
tal approachcan allow us to translatethecoefficient
intotheactualeffect
on sales,butthisturnsoutto be
here
as
findno statistical
we
effect.
unnecessary
As in our analysisof piracy,we will comparethe
changein sales rank of NBC season box sets after
December1 to thechangein sales rankofABC,CBS,
and Fox seasonbox setsto determine
theeffect
ofthe
removalofNBC's digitaldistribution
channel.
4.3. SummaryStatistics
The simplestpossible analysisof this quasi-experimentis a comparisonof meansbeforeand afterthe
removalofNBC contentfromiTuneson December1.
So to start,we list some veryhigh-levelsummary
statistics
and ask whetherif the averagenumberof
daily downloads of an NBC episode is greaterin
the two weeks afterDecember1 than in the two
weeksbefore.We also testthehypothesis
ofwhether
moreuniqueNBC episodesbecomeavailablethrough
piracywhen NBC contentis removedfromiTunes.
Table 1 displaysthesesummarystatistics.
In rows1, 2, and 3, we reportthemeandownload
statistics
forNBC and non-NBCshows. Duringour
studyperiod,theaveragenumberofdailydownloads
forNBC episodesincreasedby 28, whereastheaverTable 1
age fornon-NBCshows decreasedslightly.
a
basic
trend
for
our
results
that
NBC
downprovides
loadshaveincreasedcomparedwiththecontrol
panel.
We use a balancedpanel to construct
thesestatistics
and use thisin our subsequentregressions;
i.e., we
onlyuse episodesthatwereavailablebothbeforeand
aftertheevent.If a new episodeappears,thenit is a
and we ignoreitto avoid thebias in
shift,
supply-side
thedemand-sideshift.This approachis conservative
as ignoringnew episodes made availableon piracy
aftertheremovalofNBC contentfromiTunesshould,
ifanything,
introduce
additionalcompetition
forinterestamongtheotherepisodesin oursample.
In rows 4, 5, and 6, we outlinethe supplysides
of the effectof contentremovalfromiTunes.In the
two weeks afterDecember1, therewere a net 129
moreuniqueNBC episodesavailableforpiracythan
9See Smithand
Telang (2008).

This content downloaded from 14.139.237.35 on Thu, 14 Jan 2016 12:35:42 UTC
All use subject to JSTOR Terms and Conditions

Danaher et al.: Converting


PiratesWithoutCannibalizingPurchasers
2010INFORMS
Science29(6),pp. 1138-1151,
Marketing

1144

Table1

whethernon-NBCpiracyand NBC piracyhave similartimetrendsbeforeNBC content


was removedfrom
iTunes.
Because
the
raw
data
are
verynoisy,we proNBCnetworks Non-NBC
networks
vide a statistical
testand thenprovidea plot of the
237
405
fitted
valuesto outlinethetrendsin piracy.We utilize
a
longertimeframe(fromNovember15 to Decem265
397
ber 24) forthe plot. We firstestimatethe following
difference-in-difference
model:10
28
-8

DataSummary
November
18 Through
Statistics,
Piracy
December
15,2007

1. Meandaily
downloads
12/1(balanced
before
panel)
2. Meandaily
downloads
after
12/1(balanced
panel)
3. Change
4. No.ofunique
episodes
available
before
12/1
5. No.ofunique
episodes
available
after
12/1
6. Change

1,683

3,400

1,812

3,383

129

-17

in the previoustwo weeks.A moredetailedanalysis shows that147 new NBC episodesbecameavailable throughpiratedchannelsin thetwoweeksafter
December1,and 18episodesceasedtobe available.In
fornon-NBCcontent,
we do notobserveany
contrast,
new episodes becomingavailable,and we observe
17 episodes ceasingto be available.The numberof
because we would
new NBC episodes is striking,
less popular
older
content
to
become
simply
expect
a
less
over
time.
as
result,
available)
(and,
Examiningthe newlyavailable episodes,we discoverthatentireseasonsof some less popularNBC
- seasonsthatwerenotavailableon Mininova
content
beforeDecember1 but were available on iTunesbecome available on Mininova afterDecember 1.
These seriesinclude,forexample,a numberof seasons of Saved by theBell and Xena: WarriorPrincess.

None of these seasons had new or updated box


sets releasedor new syndicationdeals duringthe
of
theinference
timeperiodof thestudy,supporting
In
iTunes.
with
removal
from
NBC's
short,it
causality
thattheseentireseasonsofolderNBC teleis striking
visionshowsbecameavailableforpiracyimmediately
aftertheremovalof theiTuneschannel,whereasno
new non-NBCcontentbecameavailable.We conjecturethatincreaseddemandfortheseshows through
largepiracyswarmto
piracyallowed a sufficiently
existafterthe contentwas no longeravailable on
iTunes.
However,thesesimplestatistics
onlytella partial
for
not
control
as
do
episodeimportant
story they
nor do theyreflectan approprilevel heterogeneity
betweendigital
ate model of the actualrelationship
and piracy.Thus,we turnto regression
distribution
modelsforour mainempiricalanalysis.

5. Results
5.1. Impact on NBC Piracy

First,we provideevidencein supportof our idenassumption:thatnon-NBCpiracyis a good


tifying
forNBC piracy.As evidence,we test
counterfactual

= a/ +
log(Downloads/i)

Dec24

tDt

f=Novl5
Dec24

+ e,f/ (2)
r(D,xNBC,.

f=Novl5

whereNBC, is an indicatorvariableequal to one if


episode i is broadcaston NBC, Dt is a vectorof date
fixedeffects
foreachday (fromNovember15,thefirst
date in our data throughDecember24),a{ is a vector
ofepisode-levelfixedeffects,
Downloads,,is thetotal
numberof pirateddownloadsof episode i on day i,
and ytcapturesthedifferences
in NBC piracyoverthe
non-NBCpiracyon any day t. If non-NBCepisodes
are a good controlforNBC episodes,thenwe would
expectytto be zero fordatesbeforetheevent.Thus
we testifyt= 0 forall t fromNovember15 toNovember 30. Second,assumingthatnon-NBCpiracyis an
we can visuallyanalyzewhathapadequatecontrol,
NBC
to
pens
piracyrelativeto non-NBCpiracyin the
after
removal.
period
in Figure1,where
We plottheresulting
coefficients
non-NBCpiracylevelsare givenby jij + t,and NBC
piracylevelsare givenby /I,+ t+ yt,wherejlj and
fii are the averagebaselinepiracyfornon-NBCand
NBC episodes,respectively.
Ifnon-NBCpiracyis a good controlforNBC piracy,
thenwe would expectytto be equal to zero forall
One can see fromthechart
datesbeforethetreatment.
above thatnon-NBCpiracytracksNBC quite well
untilNovember30, whichis thefirstdate thatNBC
began removingits episodesfromiTunes.More formally,a Wald testof the null hypothesisthatall yt
are jointlyequal to zero forall t beforeDecember1
couldnotbe rejectedat the5%, 10%,or even20% levit is quite clearfromthe graphthat
els. Conversely,
NBC piracyincreasesbeyondnon-NBCpiracyimmetheremovalfromiTunesand,forall
diatelyfollowing
but a few dates,remainsat thisincreasedlevel for
the 25 days followingthe removalof NBC content
fornearlyall ytfortimes
fromiTunes.The -statistics
between
afterDecember1 indicatethatthedifference
at
NBC piracyand non-NBCpiracyis significant
the5% level and suggestthatNBC piracyincreased
10Note thatwe do not includea maineffect
in thisequationfor
as it would be
cannotbe identified,
theNBC dummy.Thiseffect
subsumedentirely
by theepisodefixedeffects.

This content downloaded from 14.139.237.35 on Thu, 14 Jan 2016 12:35:42 UTC
All use subject to JSTOR Terms and Conditions

PiratesWithoutCannibalizingPurchasers
Danaher et al.: Converting
Science29(6),pp. 1138-1151,
2010INFORMS
Marketing

1
Figure

1, 2007
December
NBCvs. Non-NBC
Piracy
Surrounding

relativeto non-NBCpiracyafteriTunes
significantly
removal.
AlthoughFigure1 providesstrongevidencethat
the removal of the digital distributionchannel
- and thatthis increasewas mainincreasedpiracy
tainedforat least the timeperiod of our data- it
does notconveniently
allow us to measuretheoverall averageincreasein piracycausedby theevent.To
avoid holidayeffects,
we use data fromNovember
18 to December15,whichgivesus twoweeksbefore
and two weeks afterthe event.Giventhisinformaas a stantion,we nextrunthefollowingregression
dard difference-in-difference
model to estimatethe
of theremovalof NBC contentfrom
aggregateeffect
iTunes:11
= a{ + Dt+ yNBQ xDf + eit. (3)
log(downloads/7)
Thismodelis similartomodel(2),exceptthathere,Dt
is a singleindicator
variableequal to one iftheobservationoccursin thetwoweeksafterDecember1,2007
and equal to zero if it is in the two weeks before
thatdate. Thus y capturesthe averageeffectof the
eventon NBC's piracyrelativeto thecontrolgroup's
piracy.A positivevalueindicatesthatNBC piracyhas
increasedby about 100*y percentin theperiodafter
removalrelativeto thepiracyof non-NBCchannels.
As before,
we includeepisode-level
fixedeffects
in the
formofa and daily-level
fixedeffects
in theformof
withdate fixed
Dr We also estimatea specification
effects
wherewe use a dummyforeach day similar
to (2).
We use a log specification
of our dependentvariable forseveralreasons.First,thedownloaddata are
11Observations
on December1 are removedfromthe
occurring
data as NBC was in the processof removingepisodes on this
date.Including
theseobservations
wouldnotmaterially
impactour
results.

1145

heavilyskewed,and hence log transformation


provides a betterfit.12Second, previouspapers have
foralbum sales
extensivelyused log specifications
and piracy (e.g., Hendricksand Sorensen 2009).
Third,a log specificationaccounts for nonlineareffect.For example,piracyand
ityin the treatment
sales changesmay be particularly
largeforpopular
will
and
episodes,
log specification capturesome of
thisnonlinearity.
However, by taking logarithms,we implicitly
oftheremovalofthedigitaldisassumethattheeffect
In a linearmodel,
tribution
channelis proportional.
a fewlargeepisodescan have a significant
effect
on
our result.Thus,althoughmostof theepisodesmay
show a largeproportional
increasein piracy,ifa few
ones
do
not
increase
large
enough,a linearspecificationestimatewill differfromthelog specification.
Giventhefitofourdata and based on priorliterature,
we use log transformation
as our primaryspecification.However,we also estimatea linearspecification
and findthatour resultsare robustto using a linear specification
(resultsare discussedin moredetail
below and in footnote
16).
One might worry that downloads of episodes
withina seasonorseriesmaybe correlated.
Forexaman
in
1
increase
of
of
the first
ple,
piracy episode
season of Heroesmay be correlatedwithepisode 12
of the second season of Heroes.Therefore,
following
Bertrand
et al. (2004),we clusterour standarderrors
at the serieslevel,whichallows all episodeswithin
our 76 unique seriesto be correlated.
We also estimaterobuststandarderrorstoallowforheterogeneity
acrossseries.
Theresultsofmodel(3) aredisplayedincolumns(i)
and (ii) of Table 2. Column (i) reportsresultsfor
model(3),and column(ii) adds datefixedeffects,
producingverysimilarresults.The variableof interest
in theregression
is y becauseit indicatesthepercent
changeinpirateddownloadsforNBC overand above
anychangefornon-NBCprograms.
The coefficients
fromtheregressions
in columns(i)
and (ii) showthatwhereasnon-NBCpiracyincreased
by 5.8% during this time period, NBC piracy
increasedby an additional11.4%overand above this
level.Thus,theremovalofNBC contentfromiTunes
causedan 11.4%increasein piracyoverand abovethe
changein thenon-NBC"controlgroup."Thisshows
a significant
substitution
betweenlegitimatedigital
distribution
and piracychannels.Based on 95% confidenceintervalscomputedusingtheclusteredstandard errors,the removalof NBC's iTuneschannel
causedan increasein piracybetween3.4%and 19.4%.
12We also rana Box-Coxteston themodelto determine
thebest- thetestproduceda 0 = 0.04,
transformation
ofdownloads
fitting
thatthelog transformation
indicating
quitestrongly
producesthe
bestfitto ourdata.

This content downloaded from 14.139.237.35 on Thu, 14 Jan 2016 12:35:42 UTC
All use subject to JSTOR Terms and Conditions

Danaher et al.: Converting


PiratesWithoutCannibalizingPurchasers
Science29(6),pp. 1138-1151,
2010INFORMS
Marketing

1146

Table2

ofLogofDaily
LeastSquaresRegressions
Ordinary
18 Through
November
December
Downloads,
15,2007
and2008

After
12/1

0.058*
(0.029)
x
After
0.114**
12/1
Removed
(0.041)
4.513*
Constant
(0.011)
effects
No
Datefixed
No.ofepisodes
6,005
Observations
161,784
R2
0.028

LeastSquaresRegressions
ofLogofDaily
Ordinary
18 Through
Downloads
November
byGenre,
December
15,2007
Drama

2008

2007
(i)

Table3

(i)

(iii)

0.337**
(0.029)
0.115** -0.055
(0.041)
(0.058)
4.314*
4.296*
(0.013)
(0.027)
No
Yes
6,005
6,376
161,784
170,556
0.15
0.16

(iv)

-0.052
(0.059)
3.806*
(0.109)
Yes
6,376
170,556
0.37

downloads
before
orafter
Thedependent
variable
is ln(total
Notes.
pirated
include
are listedin parentheses.
Regressions
September
9). f-Statistics
with
attheserieslevel.
standard
errors
clustered
fixed
effects
episode-level
at0.05,Significant
at0.10.
at0.01,**significant
*Significant

x
After
12/1
NBC
Constant
Observations
No.number
ofepisodes
R2

Action

Comedy

Sci-fi

0.011
0.112+
0.223*
0.213*
(0.064)
(0.092)
(0.062)
(0.060)
5.050** 3.704** 4.920** 4.210**
(0.049)

(0.055)

27,378
1,014
0.21

51,830
1,925
0.24

-0.029

40,734
1,510
0.20

(0.076)

11,738
435
0.27

Notes.The dependent
variableis ln(totalpirateddailydownloads).
bothepisodeanddatefixed
arelistedinparentheses.
Includes
f-Statistics
we
withstandard
errors
clustered
effects,
byseries. Forthisanalysis,
thatdid notfitintothesefourmajor
severalshows/episodes
dropped
categories.
at0.10.
at0.10,**significant
at0.05,Significant
*Significant

the December1 event.However,althoughthe eviof a spillover


withthepossibility
denceis consistent
this
effect
this
cannot
precisely
identify
In columns(iii)and (iv) we runthesamemodelsfor effect, study
counterfactual.
of
the
lack
of
an
because
appropriate
thesame timeperiodbutfor2008(i.e.,November18
increase
December15,2008),whentherewas no event In fact,theresultsabove show a significant
through
time
the
same
in
non-NBC
periodin
piracyduring
In theseregressions,
at iTunes.13
ortreatment
y is negthe
2007
event
for
thattheincreasein 2008. Also, a linearspecification
ativeand insignificant,
indicating
because
in
NBC piracyover and above non-NBCpiracyis not shows a small decrease non-NBCpiracy,
popularnon-NBCepisodes experisomehowtypicalduringthis timeof year- further a few extremely
of
in piracy(whilethemajority
decreases
enced
large
betweenthe 2007
evidenceof a causal relationship
an increase).
increaseand theremovalofNBC contentfromiTunes. non-NBCepisodesexperienced
in NBC piracy,
the
increase
the
source
of
To
explore
The increasein non-NBC piracy observed here
our
results
in
we
break
down
Table
3
by type of
could,however,be derivedfromsome otheroutside
four
model
for
and
run
(3)
separate
genre
program
factoror be a generaltimetrendforall piracyduring
fiction
science
and
action,
drama,
comedy,
groups:
thisperiod,in whichcase thedifference-in-difference
(sci-fi).Conventionalwisdomsuggeststhatcomedy,
on NBC
oftheeffect
modelis an accurateestimation
and action programming
sci-fi,
appeals more to a
piracy.However,we have notbeen able to determine
knownthatthis
It
is
well
duringthistimeframethatmight youngerdemographic.14
anyoutsidefactors
also
more
is
likelyto indulge
demographic
resultinan increasein television
piracydemandunre- younger
withthisconConsistent
in
2008).
(Liebowitz
piracy
latedto NBC's iTunesdecision.
in piracy
the
increase
that
show
results
our
and one thatis derived jecture,
An alternateexplanation,
the
for
above
is
and
sci-fi
for
action,
20%;
comedy
chanthedigitaldistribution
from3,is thatremoving
for
increase
and
the
is
about
increase
11%;
piracy
nel could have a spillovereffectif the nonfinanand
is
drama
programming only slightlypositive
cial cost of piracyis largelyfixed.Thus the 5.8%
increasein non-NBCpiracyfoundin column(i) of insignificant.
Returningto our main resultsin model (3), we
Table 2 could be a resultof the December1 NBC
that there are a few ways to interpretthe
note
and not a generaltimetrend.If thiswere
treatment
overall
percentagechange in piracyresultingfrom
thecase, thenour resultswould understatethe disof iTunescontent.The firstis to calcuthe
removal
placementof piracyby the iTuneschannel,because late the
impliedaverageunitincreasein piracyper
the changein non-NBCpiracywould no longerbe
as
to predictwhatshould episode theaveragenumberofpiratedNBC downcounterfactual
an appropriate
have happened to NBC piracyin the absence of loads per episode priorto December1 (237) multiincreasein piracyin ourmodel
pliedbytheestimated
13In the2008
therewereabout370moreepisodesavailregression,
able foranalysisthanin the2007regression
(in partbecauseofthe
additionalNBC episodesthatappearedafterDecember1, 2009).
thesenewepisodesfromthe2008analysisproHowever,
removing
ducesno material
changesto ourresults.

14For
example,see theJune2009 Nielsenreportwithrespectto
forteens(NielsenComin mediaconsumption
genrepreferences
2010
Warner
the
Time
and
reporton mediaadvertising
pany2009)
to youngmales(TimeWarnerCable 2010).
targeting

This content downloaded from 14.139.237.35 on Thu, 14 Jan 2016 12:35:42 UTC
All use subject to JSTOR Terms and Conditions

PiratesWithoutCannibalizingPurchasers
Danaher et al.: Converting
2010INFORMS
Science29(6),pp. 1138-1151,
Marketing

(11.4%)to obtainan averageincreaseof 27 pirated


to NBC's decision
downloadsperepisodeattributable
Because
channel.15
to removetheiTunesdistribution
therewere 1,683NBC episodes availableforpiracy
priorto December1, and the averageepisode experiencedan increaseof 27 pirateddownloads,using
thismethod,we would concludethattheremovalof
NBC contentfromiTunescaused a totalincreaseof
about48,000pirateddownloadsper day ofNBC conintervalaroundthisnumber
tent.The95% confidence
is 13,500to 77,400pirateddownloadsper day.
Anotherway to look at this increaseis to compareitwithiTunespurchasesofNBC episodesbefore
theremovalof thecontent.To do this,we wereable
to obtaindata showingthatNBC sold slightlyover
320,000episodesof itscontentthroughiTunesin the
twoweekspriorto December1,2007,or about23,000
episodesper day.Thus,our resultssuggestthatthe
unitincreasein piracywas about twiceas large as
Thisresult,although
saleson iTunes.16
thepreremoval
in
was
at
surprising first, predicted ourtheorysection
We discuss
ifthefixedcostofpiracyweresignificant.
detailin 6.
thisin further
our model is measuringthe increase
Importantly,
in piracyacrossepisodes in percentageterms.One
increase
practicalquestionis whetherthepercentage
is
driven
have
witnessed
thatwe
being
disproportionately
by the mostor least popularepisodes.To
this,we firstseparatedout the top 20%
investigate
of themostdownloadedepisodes(the"head") from
theremaining
episodes(the"tail").17
Usingthissplit,
to
model examinewhether
we rana tripledifference
theincreasein NBC piracy,overand above non-NBC
forthehead (themostpopular
piracy,was different
20%ofepisodes)thanforthetail(theremaining
80%):

1147

Table4

forHeadand
ofPiracy
Change
Analysis
TailTitles
DVDboxsets

After
12/1
After
12/1x NBC
After
12/1x Head
After
12/1x Headx NBC
Constant
Observations
No.ofepisodes
R2

0.086* (0.034)
0.104**(0.0424)
-0.110* (-0.044)
-0.039 (0.069)
4.518* (-0.010)
161,784
6,005
0.35

variable
is InfiTunes
Notes.Thedependent
downloads),
include
inparentheses.
f-Statistics
arelisted
Regressions
clustered
with
standard
errors
fixed
effects
episode-level
attheserieslevel.
at0.05.
at0.01, ^significant
*Significant

whethertheepisodeis in
dummyvariableindicating
thetop 20% in termsofnumberofdownloads.Thus,
theincreasein piracycaused by NBC's
y represents
removalfromiTunesforthelesspopulartailepisodes,
and y+ v indicatestheincreasein piracyforthemost
popularepisodes.18
The resultsin Table 4 suggestthaton a percentage basis the changein piracyforthemostpopular
no different
than
episodesin thehead was statistically
thatforthe tail. However,we also note thatinterpretingthesepercentageresultsin lightof the unit
downloadsofhead and tailtitleson bothiTunesand
throughpiracysuggeststhatiTunesdownloadsfor
head titleswere farmorelikelyto convertto piracy
thaniTunesdownloadsoftailtitleswere.
Thisfinding
also shedslighton thearguments
outlinedin thetheorysectionregarding
fixedversusvariin the resultsabove,
able cost to piracy.Specifically,
ourprimary
a
for
fixed
costto piracyis the
argument
in
fact
that
the
increase
caused
piracy
by theremoval
logPownloadsJ = f+jSDf+yNBQx Dt+8Dtx Head;
of theiTuneschannelwas largerthanthesize of the
+ vDtx Head,x NBCf+e/t. (4) iTunesmarketpreremoval.
A possibleobjection
tothis
is a storyofvariablecost:whentheiTuneschannelis
In model(4) Dt is again a dummyvariableequal to
removed,some percentageof iTunescustomersturn
oneifthedateis afterDecember1,and Head, is nowa
topiracy.Basedon how theBitTorrent
protocolworks,
thisadditionaldemandbecomessupplyand makes
15
we haveexplainedwhythelog modelis theappropri- download
Although
speedsfaster,
reducingthevariablecostof
atespecification
here,we also testeda linearmodel.Thecoefficient
and
additionalnew pirates
downloading
attracting
on theinteraction
termbetweenNBC and the afterDecember1
not
have
even
been
iTunescustomers).
(who
may
dummyin thelinearmodelis about25 downloadsperday and is
because
of
the
nature
of the number
at
However,
the
95%
level.
Thus
for
our
coefficient
of
a
interest,
significant
linearmodelproducesresultssimilarto thoseimpliedby our log
of connectionsmaintainedin BitTorrent
"swarms,"
model(a unitincreaseof 25 versusan increaseof 27 forthelog
episodes witha small numberof downloaderswill
model).
experiencea muchlargerreductionin variablecosts
16
at the95% confidence
level,we cannotreject (increasein download
Although,
notably,
speeds)fromadditionaldownthenullhypothesis
thattheincreasein pirateddownloadswas less
loaders
than
with
a largenumberof downepisodes
thanthesizeoftheNBC iTunesmarket
beforeitwas removed.This
loaderswill. Therefore,
if decreasingvariablecostis
is primarily
due to thelargestandarderrorsthatresultfromclus-

teringat theserieslevel,as we onlyobserve88 uniquetelevision


seriesin ourdata.
17Ourclassification
followsthewidelyused "80/20"Paretoprinciple,althougha 90/10or 70/30splityieldssimilarresults.

18As before,severalmain effectsand


of
pairwisecombinations
theseeffects
havebeenleftoutofthemodelas theywouldbe subsumedby theepisodeand datefixedeffects.

This content downloaded from 14.139.237.35 on Thu, 14 Jan 2016 12:35:42 UTC
All use subject to JSTOR Terms and Conditions

PiratesWithoutCannibalizingPurchasers
Danaher et al.: Converting
2010INFORMS
Science29(6),pp. 1138-1151,
Marketing

1148

causingthe increasein piracy,we would expectto


observethisphenomenonmoststrongly
amongless
in
towhat
this
is
contrast
However,
popularepisodes.
theincreasein piracyfor
we observethat,ifanything,
themostpopularhead showswas largerthanthesize
oftheiTunesmarketfortheseshows.
To close, we note thatthe main resultsreported
this sectionare robustto a varietyof
throughout
A linearmodel also shows
alternative
specifications.
a similarlysized increasein NBC piracyover and
above non-NBCpiracy.Changesto the timeframe,
the removalof the Sci-Fiand USA networksfrom
theanalysis,and theinclusionofcontrolsfortheage
ofeach episodedo notmaterially
impacttheresults.
Coding November30 as the firstday of the "after
removalfromiTunes"periodproduceseven stronger
results,likelybecause NBC startedremovingconone mightworry
tenton November30. Additionally,
of NBC's removalfrom
thattheearlyannouncement
iTunes (announcedseveral monthsbeforeit happened)mighthave caused people to switchto piracy
beforetheactualremoval.Ifthiswerethecase,piracy
would have spikedbeforeDecember1, causingour
the true effect.Thus, the
resultsto underestimate
model appears to producerobustand, if anything,
results.
somewhatconservative,

sales rankbecause of the Paretodistribupredicting


tionof sales acrossbox sets.Thus,we runa similar
model to the one we ran for
difference-in-difference
follows:
as
piracy,specified
= /;+ aDt + /3NBQx Dt+ 8log(Plt
) + eit, (5)
lnRank/7

is definedas theAmazon.comsalesrank
whereRank/7
of season box set /on day i, Dt is an indicatorvariable equal toone in thetwo-week
periodafterDecember 1, Pitis the priceof box set i on day t, and i,
is a vectorof fixedeffectsforeach season box set.
theAmazonsales rankas well as
We log transform
withpriorliterathe daily Amazon priceconsistent
the
and
on
ture
based
providedin 4. As
explanation
on a balanced
are
conducted
above,our regressions
do
not
observe
anynew
sampleof titles.Indeed,we
time
frame.19
this
DVD box settitlesadded during
One mayworrythatNBC (orAmazon)maychange
pricesof NBC DVDs in responseto the event.Our
data do notshowanyunusualpricechangesforNBC
DVDs. We also had a discussionwithNBC personnel and do notbelieveNBC strategically
changedthe
prices.However,inclusionof pricedata providesa
. Omissionofpricgood controlforourkeyestimate,
variablebias.We
to
omitted
lead
information
may
ing
in
Table6.
this
for
results
regression
present
the
has
that
note
We
Sales
Set
predictedeffect
DVD
Box
price
raising
5.2. Impacton Amazon.com
increasedrank.
as
observed
of
sales,
iTunes
by
the
decreasing
To determinethe degreeto which
digiin sales rank
no
observe
also
We
DVD
change
of
significant
channeldisplacespurchases
tal distribution
sets after
box
to
non-NBC
relative
box
sets
NBC
for
to
tests
we use similar
box setssold on theInternet,
sales
the
of
the
removal
1.
December
Thus,
is
digital
thoseabove. The dependentvariablein thiscase
sets
DVD
box
of
sales
increase
to
not
seem
did
channel
theAmazon.comsales rank,and thusa decreasein a
litthe
Given
Internet.20
the
on
sold
priormarketing
DVD's rankindicatesan increasein salesofthatDVD.
Viswanathan
differentiation
on
channel
erature
(e.g.,
Table5 comparesmeansforsales ranksof NBC and
non-NBCbox setsbeforeand afterDecember1, 2007. 2005),thisfindingcould suggestthatconsumersconsubWe see fromTable 5 thatthe mean rankfornon- siderillegaldigitaldownloadsa muchstronger
than
downloads
for
stitute
legalphysical
legal digital
NBC box setsincreasedby 10%,meaningthatfewer
non-NBCtelevisionseriesbox sets were sold after purchasesare forlegaldigitaldownloads.We discuss
thisresultin moredetailbelow.
December1 thanbefore.Theincreasein rankforNBC
box setswas only6%, whichcould indicatethatthe
iTunes
removalof NBC contentfromiTunescaused some 5.3. NBC's Returnto
thebestexperabove
results
The
represent
presented
additionalpurchasesof DVD box sets.However,as
the
determine
find
to
we
can
iment
relationship
ofmeans
thiscomparison
withouranalysisforpiracy,
and
between
distribution,
physical
digital
piracy,
does notaccountforchangesin pricethatmayoccur
on
occurred
another
sales.
However,
experiment
duringthistimeperiod(especiallywiththeapproachan
when September9, 2008, when, afterreaching agreeingholidays)noris a linearmodelappropriate
mentwithApple,NBC restoredall of its contentto
Table5

of Means),
DVDSales Rankat Amazon.com
(Comparison
December
11Through
November
15,2007

19Discussionswithknowledgeable
induspersonsin thetelevision
are
released
sets
DVD
box
all
almost
that
immediately
trysuggest
season,whichstartsin September.
networks beforethefalltelevision
Non-NBC
NBCnetworks
20We
analyzedDVD box setsat thelevelofbox setsales perday,
toDec.1
35,384
salesrank
24,553
Amazon.com
prior
ofsalesand changesbybox
becausepriceis a significant
predictor
Dec.1
after
38,785
salesrank
26,056
Amazon.com
set and by theday However,ifwe wereto clusterour standard
3,401
1,503
Change
errorsat theserieslevel,it would merelyinflatethem,and thus
10%
6%
% Change
our resultswould remaincloseto zeroand insignificant.

This content downloaded from 14.139.237.35 on Thu, 14 Jan 2016 12:35:42 UTC
All use subject to JSTOR Terms and Conditions

PiratesWithoutCannibalizingPurchasers
Danaher et al.: Converting
2010INFORMS
Science29(6),pp. 1138-1151,
Marketing
Table6

Least Squares RegressionsofLnSales Rank


Ordinary
18 Through
ofSeason DVDBoxSets, November
December15, 2007
DVDboxsets
1.727**(0.103)
-0.023 (0.020)
0.000 (0.048)
2.925 (0.338)
9,555
397
0.29

price)
ln(Amazon
After
12/1
After
12/1x NBC
Constant
Observations
No.ofuniqueDVDboxsets
R2

sales rank).f-Statistics
variableis ln(Amazon
Notes.Thedependent
fixedeffects.
includeDVD-level
arelistedinparentheses.
Regressions
does notmaterially
changetheresults.
Addingdatefixedeffects
at 0.05.
at 0.01,^significant
-^Significant

thisdate also cointhe iTunesstore.Unfortunately,


cides withthenew fallseason of televisionin 2008,
forour empiricalanalwhichpresentscomplications
of
a
new
season of television
The
ysis.
premiere
demand
forolder seasons of
increases
undoubtedly
the same show,causinglarge day-to-dayswingsin
piracyof older contentthatcorrespondto the premieresofnew showsand increasing
thenoise in the
data aroundthisperiod (as evidencedin the large
standarderrorsand low explanatory
powerreported
in Table7).
However,it stillmay be valuable to examinethe
changesin piracyaround this time period to see
whetherour earlierfindingsare supported.In 5.1
we notedthatwhen NBC removedits contentfrom
iTunes,pirateddownloadsof NBC shows increased
by morethanthenumberofiTunespurchasesprevithisas evidenceofa
ouslymade,and we interpreted
fixedcostto piracy.If thenonfinancial
costofpiracy
is largelyfixed,thenwe would not expectto see as
return
toiTunesas leftwhentheconmanycustomers
tentbecomesavailableagainthere.
Table7

Least Squares RegressionsofLog ofDaily


Ordinary
Downloads,August26 Through
September22, 2008

After
12/1
After
12/1x NBC
Constant
Datefixedeffects
Observations
No.ofepisodes
R2

(i)

(ii)

0.063
(0.049)
-0.077
(0.096)
4.306*
(0.021)
No
129,720
4,709
0.015

-0.077
(0.096)
4.275*
(0.053)
Yes
129,720
4,709
0.074

Notes.The dependent
variableis ln(totalpirateddownloadsbefore
and afterSeptember
are listedin parentheses.
Stan9). f-Statistics
darderrorsareclustered
at theserieslevel.
at 0.01,significant
at 0.05,Significant
at 0.10.
'Significant

1149

we runmodel (3) again


To explorethispossibility,
forthe fourweeks surrounding
September9, 2008,
7.
in
Our resultssugTable
these
results
reporting
to iTunes,
was
returned
NBC
content
that
when
gest
7.7%
overand
was
reduced
NBC
content
of
by
piracy
above any change fornon-NBCcontent.However,
and considerthisresultis statistically
insignificant,
alone
the
estimate
suggeststhatthe unit
point
ing
decreasein piracywhen NBC returnedto iTunesis
muchsmallerthanthe unitincreasein piracywhen
NBC leftiTunes.

6. Discussion

Our resultsrepresent
thefirsttest,to thebestof our
we
are
aware
ofthatquantifies
theeffect
knowledge,
of a legal digitaldistribution
channelon bothonline
piracyand onlinesales ofphysicalproducts.As such,
decisionmakersat mediafirmssomemuchtheyoffer
neededevidenceregarding
theabilityoflegal digital
distribution
channelsto competewithillegalpiracy
channelswithoutcannibalizing
physicaldistribution
channels.
in this study,we used NBC's deciSpecifically,
sion to removeits contentfromthe iTunesmusic
storeon December1, 2007as a quasi-experiment
and
foundthattheremovalofNBC's primary
sales
digital
channelcaused an 11.4% increasein piracyof that
contentover and above any changeexperiencedby
networksABC, CBS, and Fox over the
competitor
sameperiod.An 11.4%increasein piracycorresponds
to about27 moredownloadsper day per episode,or
48,000additionalpirateddownloadsof all NBC contentper day.To put thisnumberin perspective,
it is
about twiceas largeas the numberof daily iTunes
sales NBC receivedin thetwo weeksbeforeDecember1. Thisestimateis conservative
becauseMininova
piracy,althougharguablya good proxyforoverall
BitTorrent
piracy,represents
onlya portionofoverall
BitTorrent
televisionpiracy.
We note that this large jump in piracy (larger
than the size of the iTunes market)is consistent
with therebeing a significant
fixedcost to piracy.
In otherwords,our resultsare consistentwiththe
possibilitythatiTunespurchasersmay avoid piracy
because the fixedcost in learningto use BitTorrent
(or the fixedmoral/stigmacost of illegalbehavior)
makespiracyless attractive
thanlegitimate
purchases
iTunes.However,whenthedigitalsaleschanthrough
nel is not available,theseindividualsturnto piracy,
pay thefixedcost,and, owingto theseeminglylow
marginalcostsofadditionaldownloads,beginto consumemorecontentthrough
piracythantheyhad prethisphenomenon
seems
Moreover,
viouslypurchased.
mostprevalent
forthemostpopularepisodesoftelevision,whichsuggeststhatitis notdrivenbya decrease

This content downloaded from 14.139.237.35 on Thu, 14 Jan 2016 12:35:42 UTC
All use subject to JSTOR Terms and Conditions

1150

PiratesWithoutCannibalizingPurchasers
Danaher et al.: Converting
2010INFORMS
Science29(6),pp. 1138-1151,
Marketing

a snapshotintimefora singlemediatype.It
in the variablecost of piracyresultingfrommore represent
studiestoanalyzecompetiwould
be
usefulforfuture
in
swarm.
the
BitTorrent
piratesparticipating
and illegitimate
could even have a
tionbetweenlegitimate
effect
We notethata fixed-cost
digitalchana
results.
that
have
in
to
confirm
our
networks
other
other
nels
effect
for
digital
settings
spillover
once thefixedcostis paid for
Anotherinteresting
sales channel/because
potentialdirectionforfuture
results
Our
ABC
as
well.
"moral"cost of piracy.In 3,
to
the
it
is
relates
research
NBC,
likelypaid for,say,
model
shows
the
with
this
topurchaseorpiraterests
that
the
decision
are consistent
we argued
possibility
when
we
in
non-NBC
cost curve
a 5.8% increase
might largelyon the shape of the nonfinancial
piracy
from
have expecteda decreaseresulting
decreasing associatedwithpiracy.Althoughpartofthatcostmay
interest.
However,thisresultshouldbe treatedcau- be relatedto learningor to the (sometimes)diminevi- ished qualityof the piratedcopies,thesecostsmay
tiouslyowingto a lack of strongcounterfactual
dence fornon-NBCpiracy.Eitherway,theseresults approachzero in the futureas piratesbecomemore
because sophisticatedin theirmethods,consumersbecome
shouldsoundan alarmto contentproviders,
once thefixedcostofpiracyis sunk,it maybe diffi- more technologically
savvy,and clientsoftwarefor
cultto getpiratesto returnto legal options.Indeed, piracybecomeseven moreuser-friendly.
We suspect
in thefuture
efforts
althoughthe returnof NBC contentto iTunespre- thata largepartofantipiracy
may
sentedsome analyticalchallengesas a resultof its need to relyon the consumer's"moral"cost associcoincidingwiththestartofthefalltelevisionseason, ated withpiracy.To pricecompetitively,
digitaldiswe observeda smallerand statistically
would benefitfromknowingmore
efforts
insignificant tribution
decreasein piracyforNBC content(comparedwith abouttheshapeofthismoralcost.As such,we believe
to iTunes.
non-NBCcontent)whenit was returned
or in the
thatuserstudies- whetherin thelaboratory
DVD
box
of
on
sales
distribution's
cost
moral
the
nature
of
the
at
aimed
field
impact
Digital
revealing
channelspresentsa differ- of piracyfordifferent
setssold throughInternet
a
fertile
is
consumers
of
types
however.WhenNBC removedits archived area forfutureresearch.
entstory,
seasonsof televisionfromiTunes,we foundno significantchange in the Amazon.comsales rank for
Acknowledgments
NBC's DVD salesrelativeto thetrendthatwe saw for Authorsare listedin alphabeticalorder.The authorsthank
ofthis Mel Stephens,LowellTaylor,JoelWaldfogel,
non-NBCbox sets.One possibleinterpretation
NitinMehta,
are
not
DVDs
and
at
downloads
that
is
and
this
of
the
editors
workshopparticipants
digital
journal,
finding
Techin theshortterm,and thusaddinga digi- the 2009 NBER SummerWorkshopon Information
substitutes
Conference
channeldoes notlead to a short-term nologyand Economics,the 2009 International
tal distribution
the2008 ZEW Web 2.0 WorkIndustrial
on
interA
similar
set
sales.
DVD
box
in
Organization,
displacement
Systemsand
is thatthereis a fixedcostto digitalviewing shop, the 2008 Workshopon Information
pretation
the
State
Arizona
University, Universityof
and oncea consumerhas "gonedigital," Economics,
oftelevision,
Cornell
at
California
Irvine,
CarnegieMellon University,
she is unlikelyto comeback.
of
the
of
the
University
Maryland,
University
to our University,
We note thatthereare severallimitations
of
the University
SouthernCalifornia,
University,
Temple
our findings
represent Texas at Austin,and the Universityof Texas at Dallas
study.First,and mostnotably,
betweenlegitimate
a testofshort-term
digi- forvaluablecommentson thisresearch.The authorsalso
elasticity
and phys- thankGuillermoJose
taldistribution,
Salahaldin
pirateddigitaldistribution,
Aguilar,TemiAwofisayo,
channels.Thelong-term
icaldistribution
forexcellent
presenceofa
Hussein,SuganthPranesh,and BryanSpringer
channellikelyhas a muchstronger researchassistance.The authorsacknowledgetheNational
digitaldistribution
sales thanthoseobserved ScienceFoundationforgenerousfinancial
channel
on physical
effect
supportprovided
to throughCAREERAwardIIS-0118767(to thethirdauthor)
term.
short
in
the
here
However,it is important
note thatthereis littlethatmedia firmscan do to and CAREERAwardCNS-0546009(to the fourthauthor).
ofdigitalchannelsgiventhe R. Telangalso acknowledgesfinancialsupportofAlfredR
thepenetration
forestall
StudiesFellowship.
associatedwith SloanFoundationIndustry
increasedease, speed,and flexibility
Rather,our
obtainingmedia in digitalenvironments.
be best
would
media
that
resultssuggest
companies
with
servedby competing
piracythroughdigitaldis- References
warnsstudiosoverDVD
Wal-Mart
T. 2006.Retail-Iation:
than
rather
tribution
hopingthatthelack of a legiti- Arango,
Post(September
York
New
downloads.
22) 31.
channelwilldriveconsumers
matedigitaldistribution
mall:A strategic
Mail
versus
1998.
S.
analysisot
Balasubramanian,
away fromthe digitalchanneland back to physical
retailand conventional
betweendirectmarketers
competition
we also note
Sci.17(3) 181-195.
ers.Marketing
purchases.In additionto thislimitation,
thatthespilloverresultsmentionedabove shouldbe
ZUU4.now mucnsnouia
M., E. Durlo,b. Mullainatnan.
Bertrand,
estimates?
we trustdifferences-in-differences
QuartJ.Econom.
cautiouslyowingto thelackofan approinterpreted
249-275.
results
our
that
note
we
119(1)
counterfactual.
Further,
priate

This content downloaded from 14.139.237.35 on Thu, 14 Jan 2016 12:35:42 UTC
All use subject to JSTOR Terms and Conditions

PiratesWithoutCannibalizingPurchasers
Danaher et al.: Converting
2010INFORMS
Science29(6),pp. 1138-1151,
Marketing
Bhattacharjee,S., R. D. Gopal, K. Lertwachara, J. R. Marsden,
R. Telang. 2008. The effectof digital sharing technologieson
music markets: A survival analysis of albums on ranking
charts.ManagementSci. 53(9) 1359-1374.
E., P. Naik. 2003. Clicks and mortar:The effectof onBiyalogorsky,
line activitieson off-linesales. MarketingLett.14(1) 21-32.
Blackburn, D. 2004. On-line piracy and recorded music sales.
Unpublished manuscript,Harvard University,Boston.
Brynjolfsson,E., Y. Hu, M. D. Smith. 2003. Consumer surplus in
the digitaleconomy:Estimatingthe value of increasedproduct
variety.ManagementSci. 49(11) 1580-1596.
Chevalier,J.,A. Goolsbee. 2003. Measuring prices and price competitiononline: Amazon.com and BarnesandNoble.com. Quant.
MarketingEconom.1(2) 203-222.
Chiang, W. Y, D. Chajed, J.D. Hess. 2002. Direct marketing,indirectprofits:A strategicanalysis of dual-channel supply-chain
design. ManagementSci. 49(1) 1-20.
Christin,N., A. S. Weigend,J. Chuang. 2005. Content availability,
pollution,and poisoning in filesharingpeer-to-peernetworks.
Proc. 6th ACM Conf. ElectronicCommerce,ACM, New York,
68-77.
Chu, J.,P. K. Chintagunta,N. J.Vilcassim.2007. Assessing the economic value of distributionchannels:An application to the PC
industry./. MarketingRes. 44(1) 29-41.
Danaher, B., J. Waldfogel. 2008. Reel piracy: The effectof online
movie piracy on filmbox officesales. Workingpaper, Universityof Pennsylvania,Philadelphia.
Deleersnyder,B., I. Geyskens, K. Gielens, M. G. Dekimpe. 2002.
How cannibalisticis theInternetchannel?A studyof thenewspaper industryin the United Kingdom and The Netherlands.
Internat.]. Res. Marketing19(4) 337-348.
DVD News. 2006. Top DVD retailers:Market shares. DVD News
Magazine10(41) 30.
Friedman,L. G., T. R. Furey 1999. The ChannelAdvantage:Going
to MarketwithMultipleSales Channelsto ReachMore Customers,
Sell More Products,Make More Profit.Butterworth-Heinemann,
Oxford,UK.
Ghose, A., M. D. Smith,R. Telang. 2006. Internetexchanges for
used books: An empirical analysis of product cannibalization
and welfareimpact.Inform.SystemsRes. 17(1) 3-19.
Goodell, J. 2003. Steve Jobs: The RollingStone interview.Rolling
Stone (December 3). Retrieved March 15, 2009, http://www
/story/5939600/steve_jobs_the_rolling
.rollingstone.com/news
_stone_interview.
Hendricks,K., A. Sorensen.2009. Informationand the skewness of
music sales. /. PoliticalEconom.117(2) 324-369.
Hui, K., I. Png. 2003. Piracyand the legitimatedemand forrecorded
music. B.E. J.Econom.Anal. Policy2(1), Article11. http://www
.bepress.com/beieap/contributions/vot/issl/artll.
Jain,S. 2008. Digital piracy: A competitiveanalysis. MarketingSci.
27(4) 610-626.
Jeuland,A. P., S. M. Shugan. 1983. Managing channel profits.MarketingSci. 2(3) 239-272.
Liebowitz,S. J.2008. Testingfilesharing's impact on music album
sales in cities.ManagementSci. 54(4) 852-859.
McBride,S., M. Marr.2006. Target,a big DVD seller,warns studios
over download pricing.Wall StreetJournal(October 9) Al.

1151
Menn, J. 2006. Target, Disney in DVD truce. Los Angeles Times
(November 18), http://articles.latimes.com/2006/nov/18/
8.
business/fi-targetl
Msnbc.com. 2007. iTunes to drop NBC shows beforecontractends.
(August 31), http://www.msnbc.msn.com/id/20531069.
Nagin, D. S., R. Paternoster.1991. On the relationship of past
to future participation in delinquency. Criminology29(2)
163-189.
Netherby,J. 2005. Amazon.com dominates in online Dvu sales.
Gale Group,
Report(December 19), Reed Business Information,
FarminetonHills, MI.
Nielsen Company.2009. How teens use media: A Nielsen reporton
the mythsand realitiesof teen media trends.Report,Nielsen
Company, New York, http://blog.nielsen.com/nielsenwire/
reports/nielsen_howteensusemedia_iune09.pdf.
Oberholzer-Gee,F., K. Strumpf.2007. The effectof file sharingon
record sales: An empirical analysis. /. PoliticalEconom.115(1)
1-42.
Peitz,M., P. Waelbroeck.2004. The effectof Internetpiracyon music
Issues
sales: Cross-sectionevidence. Rev.Econom.Res. Copyright
1(2) 71-79.
Rob, R., J. Waldfogel. 2006. Piracy on the high C's: Music downloading, sales displacement,and social welfarein a sample of
college students./. Law Econom.49(1) 29-62.
Rob, R., J. Waldfogel. 2007. Piracy on the silver screen. /. Indust.
Econom.55(3) 379-395.
Sinha, R. K., F. S. Machado, C. Sellman. 2010. Don't thinktwice,it's
all right:Music piracyand pricingin a DRM-freeenvironment.
/. Marketing74(2) 40-54.
Smith,M. D., R. Telang. 2008. Internetexchanges for used digital
goods. Workingpaper,Carnegie Mellon University,
Pittsburgh.
Smith,M. D., R. Telang. 2009. Competing with free:The impact of
movie broadcasts on DVD sales and Internetpiracy.ManagementInform.SystemsQuart. 33(2) 321-338.
Thompson,A. 2003. Tinseltownfollies.New YorkMagazine(May 5),
http://nymag.com/nymetro/movies/columns/hollywood /n
_8677.
Time Warner Cable. 2010. Time Warner Cable media sales:
Target demographics. Accessed May 3, 2010, http://www
.cablemediasales.com/pages/nets/?CP=nets&SP=demo&demo=
ml8-49.
Vernik,D. 2009. Digital music set free:The flipside of DRM. Working paper, Rice University,Houston.
Viswana than, S. 2005. Competing across technology-differentiated
channels: The impact of network externalitiesand switching
costs. ManagementSci. 51(3) 483-496.
Waldfogel,J.2009. Lost on the Web: Does Web distributionstimulate or depress televisionviewing? Inform.Econom.Policy21(2)
158-168.
Webb, K. L. 2002. Managing channels of distributionin the age
of electronic commerce. Indust. MarketingManagement31(2)
95-102.
Zentner, A. 2005. File sharing and internationalsales of copyrighted music: An empirical analysis with a panel of countries.B.E. }. Econom.Anal. Policy5(1), Article21. http://www
.bepress.com/bejeap/topics/vol5/issl/art21.

This content downloaded from 14.139.237.35 on Thu, 14 Jan 2016 12:35:42 UTC
All use subject to JSTOR Terms and Conditions