1

Digitization, Unbundling, and Piracy: Consumer Adoption amidst Disruptive Innovations in the Music Industry

Byungwan Koh Haskayne School of Business

University of Calgary Calgary, AB T2N 1N4

bkoh@ucalgary.ca

Il-Horn Hann Robert H. Smith School of Business

University of Maryland College Park, MD 20742 ihann@rhsmith.umd.edu

Srinivasan Raghunathan

Naveen Jindal School of Management University of Texas at Dallas

Richardson, TX 75083 sraghu@utdallas.edu

Abstract

In the music industry, the Internet and digital music formats have fundamentally altered the way music is packaged, distributed and consumed. The Internet and digital music formats have enabled music to be purchased as an individual song (digital single) or as an album (digital album), or to be enjoyed without paying for it (unlicensed music). Building on multi-generation diffusion models, we identify, and quantify the demand migration patterns among these various consumption options: (i) from CD to digital album (generation substitution), (ii) from CD to digital single (unbundling), (iii) attrition from CD to unlicensed music, (iv) attrition from digital album to unlicensed music, and (v) attrition from digital single to unlicensed music. We find that the introduction of purchased digital music – digital album and digital single – option has weakened the attrition effect of online music piracy (unlicensed music) on the demand for CD. Since digital single and digital album were introduced, about 7 billion units of the demand for CD that would have migrated to unlicensed music has migrated to either digital single or digital album. Cannibalization due to unbundling, rather than attrition due to piracy, is now the dominant factor that leads the migration of demand for CD to digital music formats and the decline in industry revenue. However, attrition is emerging as a challenge for digital single and digital album. Especially, in recent years, while the rate of increase of attrition effect of online music piracy on digital single is decreasing, that on digital album is increasing. Although we focus on the demand migration patterns in the music industry, our model can be easily extended and used to shed some light on the complicated migration of demand in the industries where ‘tipping across markets’ or ‘platform envelopment’ phenomenon occurs. Keywords: Disruptive innovation, technology substitution, generation substitution, attrition of demand, unbundling, multi-generation diffusion, music sales, digital music.

 

2

1. Introduction

The phenomenon of new technologies replacing old ones is ubiquitous in many industries. Products based

on new technologies often have higher quality or lower cost than those based on the old technologies,

which induces consumers to migrate from the earlier generation products to the successive generation

ones. Sometimes, the new technology can be disruptive in the sense that it can fundamentally redefine the

characteristics of the industry or the market (Christensen 1997).

One often-cited example of such a disruptive force is the Internet along with digital music

formats in the music industry (Cellan-Jones 2013). It has fundamentally changed the way music is

packaged, distributed to, and consumed by consumers. Before the introduction of the Internet, music was

distributed primarily using physical media such as LP (Long Play), cassette, and later CD (Compact

Disc). The Internet, assisted by the development of digital sound recording formats (e.g., MP3), has made

digital music that is distributed without a physical medium the consumers’ preferred choice in recent

years.1 While the digital formats and the Internet as a distribution may seem innocuous, it had a drastic

impact on the music industry.2 First, digital music has transformed the industry from one that

predominantly distributed and sold music as a bundle of songs in a single medium (i.e., an album) to one

that distributed music also as an individual song (i.e., single). On the supply side, the negligible marginal

cost of distribution in the case of digital music as compared to significant marginal cost of distribution for

earlier formats such as LP, tape, and CD, has made such unbundling of music viable. On the demand side,

with the consumers rapidly adopting distribution channels such as peer-to-peer (P2P) networks, the

labels’ move to unbundling of music has allowed consumers to acquire and consume specific songs they

prefer thereby foregoing seller-determined bundles which may contain songs they do not prefer. This has

                                                            1 In this paper, by digital music, we are referring to music in digital format that can be downloaded from the Internet. On the other hand, the CD music, though it is stored digitally, is physical in the sense that it is available for purchase as a physical disc. 2 The Internet has disrupted other industries also in a similar way. Newspapers, magazines, movies, television, and software are some other examples where a similar phenomenon is observed.

 

3

significantly altered the music industry’s market structure.3 Second, the Internet facilitates sharing of

digital contents at minimal cost and provides consumers with various options to consume music without

paying for it though many of those are illegal.4 Consumers download music from P2P and Torrent

networks and rip music from various Internet radio channels. Altogether, the Internet has substantially

changed the way music is consumed by enabling digital music to be purchased as an individual song

(digital single) or as an album (digital album), or to be enjoyed without paying for it (unlicensed music).5

A key difference between purchased digital music – digital single and digital album – and unlicensed

music is that the former is a direct revenue source for the music industry but the latter is not.

It is imperative to understand the dynamics of consumption of music as consumers migrate from

the CD format to digital music formats. The dynamics is especially important because unlike the typical

migration of demand, where a product of a successive generation replaces a product of the previous

generation, digital music formats have given rise to a phenomenon where several new products replace

the old one (i.e., CD). For instance, after the introduction of digital music formats, the demand for music

has migrated (i) from CD to digital album, (ii) from CD to digital single, and (iii) from CD to unlicensed

music. Furthermore, demand also has migrated (iv) from purchased digital music (digital album and

digital single) to unlicensed music. For the music industry, the implications of the various migration

patterns are rather different. While the migration from CD to purchased digital music (digital album and

digital single) has implications in terms of product design, the migration to unlicensed music has

implications pertaining to piracy mitigation and conversion of non-paying consumers to paying

consumers.

                                                            3 Industry experts discussed “single vs. albums” issue at New Music Seminar 2013. Some of their views are summarized at: http://news.cnet.com/8301-13645_3-57589371-47/will-the-single-kill-the-album/. 4 We note that while consumers do not pay for unlicensed music, they may still incur costs to consume such music. For example, when music is shared through a P2P network, consumers face the risk of being caught for piracy, and downloading a virus. Please see Danaher at al. (2010, page 1141) for a discussion of different categories of costs that a pirate may incur. 5 The Internet also enables streaming of digital contents and introduces streaming music channels such as YouTube, Pandora and Spotify; hence, streaming digital music is clearly another available music consumption option. However, in our study, we model the demand migration patterns from CD to digital music formats, and hence, for consistency with the CD format, we focus on the digital music formats that consumers can possess. In our model, streaming music channels indirectly affect the music industry through ripped music files from these channels.

 

4

There is considerable research on the impact of the Internet on the music industry. Most studies

examined the effect of online file sharing on music sales (e.g., Zentner 2005, 2006; Michel 2006; Rob and

Waldfogel 2006; Hong 2007, 2011; Oberholzer-Gee and Strumpf 2007; Liebowitz 2008; Andersen and

Frenz 2010). These studies mainly focused on the association between CD sales and online file sharing.

Chellappa and Chen (2009) and Dewan and Ramaprasad (2014) included digital music sales in their

models, but like previous studies, they also focused on the association between (aggregate) music sales

and online file sharing. Koh et al. (2013) and Waldfogel (2010) explored how the availability of

purchased digital music option affected the association between music sales and online file sharing. In

summary, the prior literature in this stream of research provided insights into the aggregate association

between online file sharing and music sales. Closest to our paper (in terms of the overall research question

addressed) would be Danaher et al. (2010) that investigate the impact of legitimate digital media

distribution channel on physical sales and digital piracy channels in the DVD market. They used the

removal of NBC content from Apple’s iTunes store in December 2007 and its subsequent restoration in

September 2008 as natural experiments to estimate the short-term effect of the availability of legitimate

digital distribution channel on piracy. In contrast to prior literature, we examine the demand migration

patterns among various music consumption options (i.e., CD, digital album, digital single, and unlicensed

music) available to a consumer, and quantify various migrations at a micro level not found in the existing

literature. We do this by taking a product diffusion point of view and by developing a multi-generation

diffusion model that is able to capture and quantify the multiple migration dynamics. We discuss why a

multi-generation diffusion model is an appropriate model for our context in section 2.

This paper also contributes to the literature on diffusion models. Diffusion models of single

product formulates the bell-shaped growth of product sales using innovation (which represents the

communication of product through mass media) and imitation (which represents the word-of-mouth)

parameters (e.g., Fourt and Woodlock 1960; Mansfield 1961; Floyd 1962; Rogers 1962; Chow 1967;

Bass 1969). Studies have extended the single product diffusion models to multiple products by including

complements (Peterson and Mahajan 1978; Bucklin and Sengupta 1993; Jun and Park 1999; Dewan et al.

 

5

2010; Niculescu and Whang 2011), generation substitutions (Norton and Bass 1987; Mahajan and Muller

1996; Islam and Meade 1997; Jun and Park 1999; Kim et al. 2000; Danaher et al. 2001; Chu and Pan

2008; Michalakelis et al. 2010; Jiang and Jain 2012), competition (Givon et al. 1995; Libai et al. 2010),

and network effects (Goldenberg et al. 2010; Hann et al. 2013). Unlike other multi-generational diffusion

models, in which the demand from a previous generation product migrates to a single new generation

product, our model allows the demand from a previous generation to migrate to multiple new generation

products (e,g., from CD to digital album, digital single, and unlicensed music). In addition, our model

allows individual diffusion characteristics and the migration patterns to vary across these new generation

products, whereas prior studies in this stream make restrictive assumptions about individual diffusion

parameters.

One of the main advantages of using the diffusion model approach (over other empirical

techniques) is that the model works well with relatively small number of data points (Moutinho and

Hutcheson 2011) and without decision variables such as price and advertising (Bass et al. 1994). It has

been noted that the Bass model (Bass 1969) can be applied even when as few as four data points are

available.6 Furthermore, if the percentage changes of decision variables (e.g., price) over time are

approximately constant, then the estimation results without decision variables are observationally

identical to that with decision variables.7

The remainder of the paper is organized as follows. In the next section, we develop our model

that allows us to analyze and estimate the generation substitution, unbundling, and attrition effects. In

Section 3, we estimate our model using two estimation approaches; nonlinear seemingly unrelated

regression (NLSUR) and nonlinear system GMM (generalized method of moments). We discuss and

compare the various migration effects and the impact they had on the music industry in Section 4.

                                                            6 However, Heeler and Hustad (1980) show that the estimation quality is improved when the data includes the peak. 7 The trends of prices of music formats over time have been relatively stable. The average suggested list price of CD went down significantly immediately after CD was introduced in 1983 and steadily decreased afterwards (Recording Industry Association of America 2007). CD prices were surprisingly unresponsive to the diffusion of unlicensed music (Mortimer et al. 2012). Furthermore, most songs at iTunes have been sold consistently at $.99 per song since iTunes was launched in 2003.

 

6

Section 5 summarizes our findings and discusses the implication for the music industry. We conclude in

Section 6 with limitations and future research opportunities.

2. Model Development

In this section, we develop a model for the migration of demand for music – from CD to digital music

formats – and attrition of demand from purchased music (CD, digital album and digital single) to

unlicensed music by extending the standard multi-generation diffusion model (Norton and Bass 1987).

We first briefly review the Norton and Bass model. Then, we describe the music industry context we

study in this paper, and illustrate why the standard multi-generation model is inadequate to model this

context. Finally, we describe the model we use in this paper.

2.1. Review of the Norton and Bass Multi-generation Diffusion Model

Let there be two successive generations, G1 and G2, of a product which are introduced in the market at

time 0 and 02 respectively. Let tFi be the cumulative adoption probability of generation i by time

t . In the Norton and Bass model, the sales (in terms of number of units) of generation i in time period t ,

tSi , is given by the following:

ttFmtS 111 ,

ttFmtS 2222 ,

where

2211 tFtFmt ,

otherwise. 0

,0 if 1

1t

ep

qe

tF tqp

i

i

tqp

iii

ii

    (1)

In equation (1), the expression for iF t provided on the right hand side models the S-shaped

curve of cumulative sales over the life time observed for many products. In this expression ip is referred

 

7

to as the coefficient of innovation and iq is referred to as the coefficient of imitation of generation i . 1m

is the market potential of G1 and 2m is the increase in (or unique) market potential for G2 over and above

that of G1. That is, the market potential for G2 is 21 mm . t is the decrease in demand for G1 and

increase in demand for G2 in time period t due to the generation substitution (i.e., migration from G1 to

G2 in time period t ). As G2 diffuses, more of the demand for G1 migrates to G2, and the demand for G1

becomes zero when the demand for G2 is saturated (i.e., 122 tF ).

2.2. Our Context: Music Industry

Since its introduction in 1983, the CD had been the dominant music recording and distribution format.

Starting late 90s, when residential broadband Internet access started to become mainstream, digital music

began to spread on the Internet. However, the music industry did not start selling digital music until 2003.

Please note that while the proliferation of the (broadband) Internet8 and various unlicensed music options

(i.e., online music piracy channels such as Napster) began in late 90s, digital music for purchase (e.g.,

iTunes) became available only in 2003. During the period 1997 – 2003, there was no option to “purchase”

digital music. When the music industry started selling digital music, it offered digital music in two forms:

digital album, which is a bundle of songs similar to CD, and digital single, which is an individual song.

As a consequence, the demand for music has migrated in different ways, as illustrated in Figure 1.

                                                            8 In this paper, we refer to the broadband Internet simply as the Internet. The broadband Internet enables various online music piracy channels (unlicensed music options) by providing high-speed data transactions and interactive delivery services. The broadband Internet connection became available in 1996 and began to grow in 2000.

 

8

Note: a. Generation substitution; b. Unbundling; c. Attrition of demand from CD to unlicensed music; d. Attrition of demand from purchased digital music to unlicensed music.

Figure 1. Migration patterns among different formats of music

In Figure 1, the arrows labeled “a”, “b”, and “c” represent migrations of demand for music from

CD to digital music formats. However, each of these migrations is different in character.

a. Although initial sound quality of digital music was inferior to that of CD, there is now little

difference between the two (Atkinson 2008). Further, digital music provides several benefits over

CD. A consumer can instantly get (download) digital music from online stores, can store more

songs in a smaller device, and play them on a number of different devices. Therefore, a part of the

demand for CD has migrated to digital album. This migration can be characterized as the

generation substitution (or technology substitution), which is similar to that defined in the multi-

generation diffusion literature (e.g., Norton and Bass 1987; Mahajan and Muller 1996; Islam and

Meade 1997; Jun and Park 1999; Kim et al. 2000; Danaher et al. 2001; Chu and Pan 2008;

Michalakelis et al. 2010; Jiang and Jain 2012).

b. The digital music format allows consumers to purchase an individual song, as opposed to an

album, which is a bundle of several songs. It is commonly known that a significant portion of

 

9

consumers in the music market prefer to consume only a subset of the songs that are bundled as

part of an album (Elberse 2010). Therefore, some of these consumers may hesitate to buy an

entire album for the sake of consuming only one or two songs they like. According to a survey

conducted by Amberg and Schröder (2007), 69.4 percent of interviewed persons indeed preferred

to buy an individual song over buying an entire album. Therefore, another part of demand for CD

has migrated to digital single due to this unbundling on top of benefits of digital music that we

discussed above. We call this migration from CD to digital single as the unbundling.

c. The remaining part of demand for CD has migrated to unlicensed music. Although, as stated in

footnote 4, there may be some cognitive/annoyance costs associated with consuming unlicensed

music, some consumers may prefer to consume it rather than buy CD. In general, a direct

monetary payment is not required to consume unlicensed music. Therefore, unlicensed music is

not a direct revenue source for the music industry. We call the migration from CD to unlicensed

music as the attrition of demand from CD to unlicensed music.

d. Unlicensed music has affected not only the demand for CD but also the demand for purchased

digital music – digital album and digital single. That is, some demand for purchased digital music

has migrated to unlicensed music.9 Therefore, similar to the relationship between CD and

unlicensed music, we call the migration from purchased digital music to unlicensed music as the

attrition of demand from purchased digital music to unlicensed music.

As seen from the above description of the music industry we study in this paper, there are

multiple types of migration (and attrition) that occur simultaneously – not only there is migration from

one generation (i.e., CD format) to the next generation (i.e., digital music formats), but also there is

attrition from products (e.g., digital album or digital single) to another product (i.e., unlicensed music)

                                                            9 As it is shown in Danaher et al. (2010), there might be a sampling effect, and some demand for unlicensed music has migrated to digital music. However, we follow a multi-generation diffusion approach, and cannot measure the attrition effect and sampling effect of unlicensed music separately. In our model, the attrition of demand from purchased digital music to unlicensed music captures the net effect of these two.

 

10

within the same generation. Furthermore, not all migrations are equivalent. For instance, migration from

CD to digital album is different from CD to digital single in the sense that the former is from album to

album formats whereas the latter is from album to single formats. Such distinctions in the products

themselves, which have a significant effect on industry’s revenue, are not considered in existing multi-

generational models. Therefore, there is a need to formulate a diffusion model that captures the migration

patterns in the music industry context.

2.3. Our Diffusion Model

We extend the standard multi-generation diffusion model (Norton and Bass 1987) in order to study the

generation substitution, unbundling, and attrition effects. We use subscripts CD, DA, DS, and FM to

denote, respectively, CD, digital album, digital single, and unlicensed music. CD represents the old

generation and DA, DS, and FM represent the new generation of technologies. We adjust the timeline

such that CD is introduced at time zero. Both digital album and digital single are introduced at time D ,

and unlicensed music is introduced at time DFM . Building on the Norton and Bass model, we write

sales (in number of units) in time period t for music in each of the formats as the following:

ttFmtS CDOCDCDCD , (2)

tttFmtS ADAGSDDADADA , (3)

tttFmtS ADSUBDDSDSDS , (4)

ttttFmtS ADSADAACDFMFMFMFM , (5)

tFi is identical to that defined in equation (1). CDm is the market potential (peak sales) of CD, and DSm ,

DAm , and FMm are the increase in market potential (peak sales) for digital single, digital album, and

unlicensed music. The demand for CD migrates to digital album, digital single, and unlicensed music

(i.e., tttt ACDUBGSCDO ). tCDO is the overall decrease in demand for CD due to the

introduction of digital music formats (indexed by O-CD), and tGS , tUB , and tACD , respectively,

capture the generation substitution (GS), unbundling (UB), and attrition of demand from CD to unlicensed

 

11

music (ACD). tADA and tADS capture the attrition of demand from digital album (ADA) and digital

single (ADS) to unlicensed music respectively. is the parameter that adjusts for the difference in units

between album and single formats; that is, one-unit demand of CD migrates to β-unit demand of

individual songs (i.e., digital single and unlicensed music). Similarly, one-unit attrition of demand from

digital album is equivalent to β-unit of additional demand for unlicensed music. A CD album has on

average of 12 songs.10 Thus, if unbundling does not affect the consumption behavior of consumers, we

would likely have = 12 (i.e., one-unit demand for CD migrates to 12-unit of individual songs), whereas

if it does, we would likely have < 12.

We develop the migration parameters, tGS , tUB , tACD , tADA , and tADS in the

following manner. Until digital music became available for purchase, (i.e., DFM t ), unlicensed

music had been the only digital music consumption option for consumers; hence, when DFM t , we

have: 0 tt UBGS , and we model tCDO  in the same way as Norton and Bass. That is, when

DFM t , FMFMCDCDACDCDO tFtFmtt .

After purchased digital music – digital album and digital single – option became available (i.e.,

Dt ), some demand for CD that would have migrated to unlicensed music in the absence of digital

album and digital single migrated to digital album and digital single instead; hence, we hypothesize (and

later empirically test) that the attrition from CD to unlicensed music, tACD , reduces over time. We use

an exponential function to model the decrease of this attrition over time and have the following to model

for tACD when Dt .

FMFMCDCDt

ACD tFtFmet (6)

                                                            10 Dave Taylor, a self-proclaimed industry veteran, analyzed 428 CD albums and found that on average a music CD has 12.54 songs. Details are available at: http://www.askdavetaylor.com/do_most_music_cds_have_12_tracks.html.

 

12

A portion of the reduction in attrition from CD to unlicensed music becomes tGS and the

remaining becomes tUB . We conjecture (and later empirically test) that, fraction of the demand that

migrates from CD to purchased digital music chooses digital single, and 1 fraction chooses digital

album. Therefore, we derive the following models for tGS and tUB when Dt .

DDACDCDt

GS tFtFmet 11 (7)

DDSCDCDt

UB tFtFmet 1 (8)

where Dtt if Dt ; and 0 otherwise.

The attrition from purchased digital music to unlicensed music is modeled similar to the Norton

and Bass model, and the model we have for the attrition from CD to unlicensed music prior to the

introduction of purchased digital music as the following.

FMFMDDADAADA tFtFmt (9)

FMFMDDSDSADS tFtFmt (10)

A consumer generally does not pay to consume unlicensed music. Hence, it is difficult to track

the transactions of unlicensed music, and a direct measure for demand for unlicensed music, tSFM , is

not readily available. Prior studies, therefore, have used a proxy measure such as the Internet penetration

(Zentner 2005; Liebowitz 2008; Bender and Wang 2009), Internet use or access (Zentner 2006; Hong

2007, 2011), and computer ownership (Michel 2006).11 Following this prior literature, we use the

(broadband) Internet penetration as a proxy measure for the demand for unlicensed music. We assume

tNtS IntFM , where tNInt is the number of Internet subscribers by time t and is the parameter

for the average per subscriber consumption of unlicensed music. Using this assumption and equations (6),

(9), and (10), we rewrite equation (5) as:

                                                            11 Similarly, the Internet penetration is used as a proxy measure for unlicensed movie consumption in Smith and Telang (2010).

 

13

FMFMDDADADDSDSCDCDt

FMInt tFtFmtFmtFmemtN . (11)

Unfortunately, we cannot estimate both FMm and in one model. Hence, we assume that

0FMm and focus on . 0FMm implies that there is no “new” demand for unlicensed music; that is,

the entire demand for unlicensed music is from CD.12 The assumption that 0FMm , therefore, captures

the upper limit of attrition of demand from CD to unlicensed music.

In summary, the diffusion model we use for the music industry is given by equations (1) – (11)

collectively.13 These equations reduce to the standard Norton and Bass model if we consider only CD and

unlicensed music or only CD and digital album.

3. Data and Model Estimation

3.1. Data

As a measure of sales for CD, digital album, and digital single, we collect annual units shipped of these

three for the time period 1982 – 2012 from RIAA (Recording Industry Association of America) year-end

industry shipment and revenue statistics reports.14 As a measure of Internet penetration, which is a proxy

measure of demand for unlicensed music, we collect the number of (broadband) Internet subscribers on a

yearly basis for the time period 1997 – 2012 from Word Bank Databank.15 We show the time series of

units shipped of CD, digital album, and digital single, and Internet penetration in Figure 2.

                                                            12 0FMm implies that some demand for unlicensed music is new; that is, some consumers who did not purchase

any CD when CD was the only option now consume unlicensed music. Hence, if 0FMm , not all demand for

unlicensed music would have been the demand for CD. 13 Note that equation (5) is replaced with equation (11) when we use tNInt as the proxy to estimate tS FM . 14 The first music CD was released in 1983, and the digital option (iTunes) was introduced in the market in 2003. The sales (units shipped) of music CD in 1982, and sales of digital music in the time period 1982 – 2003 are zero. 15 Word Bank Databank reports per 100 people (broadband) Internet subscribers. Since other data (sales of each music format) is not weighted by the population size, for consistency, we collected the population data also from Word Bank Databank, and converted per 100 people subscribers to the total number of subscribers.

 

14

Figure 2. Time series of unit shipped of CD, digital single, digital album, and the number of broadband Internet subscribers

3.2. Model Estimation

The multi-generation diffusion literature (e.g., Norton and Bass 1987) argues that the behavioral process

for the adoption of successive generations is expected to be similar; hence, the literature assumes that

coefficients of innovation and imitation are constant across generations (i.e., ppi and qqi i ).

However, in the context of the diffusion of music formats, the adoption processes for digital single and

unlicensed music are likely to be different from that for CD and digital album because the characteristics

of digital single and unlicensed music are substantially different from that of CD and digital album (e.g.,

an individual song vs. an album). Therefore, we relax the assumption of constant imitation and innovation

factors and assume that OMDSDACD pppp and qqi for all i .16

Following prior studies (e.g., Libai et al. 2009; Dewan et al. 2010), we estimate equations (1) –

(4) and (6) – (11) simultaneously using nonlinear seemingly unrelated (NLSUR) regression. Seemingly

unrelated (SUR) regression is substantially more efficient than ordinary least square (OLS) regression

when estimating multiple time-series equations simultaneously (Creel and Farell 1996). Further, the result

                                                            16 For the cell phone industry, Islam and Meade (1997) and Danaher et al. (2001) showed that the assumption of constant imitation factor can be rejected.

0

2

4

6

8

10

0.0

0.3

0.6

0.9

1.2

1.5

1982

1983

1984

1985

1986

1987

1988

1989

1990

1991

1992

1993

1994

1995

1996

1997

1998

1999

2000

2001

2002

2003

2004

2005

2006

2007

2008

2009

2010

2011

2012

Bro

adb

and

In

tern

et S

ub

scri

ber

s(1

0 m

illi

on p

erso

ns)

Un

it s

hip

ped

(B

illi

on u

nit

s)

CD

Digital single

Digital album

Internet penetration

 

15

of White’s general test for heteroscedasticity rejects the null hypothesis of homosecdasticity, and hence,

we use heteroscedasticity-consistent estimates (White 1980).

Generalized Method of Moments (GMM) estimation is another method for improving efficiency

when heteroscedasticity is present in the model (Wooldridge 2001). Hence, to check the robustness of our

results, we also estimate the model using nonlinear system GMM.17 Following Suarez et al. (2013), we

use lagged dependent variables as well as non-GMM exogenous variables as instruments. For non-GMM

exogenous variables, we collect the units shipped (sales) of other physical music (other than CD such as

LP, cassette and DVD audio) (denoted as tSOP ) and mobile music (denoted as tSM ) in time period t

from RIAA year-end industry shipment and revenue statistics report, and for each dependent variable, we

impose moment conditions instrumenting the sales of (i) different products in the same technology

generation and (ii) products in the different technology generation as illustrated in Table 1. The validity of

instruments is tested using Hansen’s J test statistics (Hansen 1982).

Table 1. Instruments for each dependent variable

Dependent Variable

Instruments

Lagged Dependent Variable

Non-GMM Exogenous Variable (i)

Non-GMM Exogenous Variable (ii)

tSCD 1tSCD tSOP tStStS MDSDA

tSDA 1tSDA tStS MDS tStS OPCD

tSDS 1tSDS tStS MDA tStS OPCD

tSFM 1tSFM - tStStS MDSDA &

tStS OPCD

We report estimation results and model fit from NLSUR along with nonlinear system GMM in

Table 2. The adjusted R-squares from both NLSUR and nonlinear system GMM are noticeably high and

all parameter estimates are significant at one percent level. The insignificant Hansen’s J statistics suggests

that the instruments for nonlinear system GMM estimation are valid; that is, instruments are uncorrelated

                                                            17 By using non-linear system GMM, we control for heteroscedasticity of unknown forms as well as endeogeneity and serial correlation in the error terms.

 

16

with the error terms. The results shown in Table 2 suggest that there is a strong overall statistical support

for our model to capture the demand migration dynamics in the music industry context. Figure 3

illustrates the model fit visually, and the figure also lends support to the strong model fit to the data.

Table 2. Parameter estimates and model fit

NLSUR Nonlinear system GMM

Est.

Std. Err.

Sig. Est. Std. Err.

Sig.

Market potential

CD (mCD) 2.0738 0.5367 *** 2.0631 0.4032 ***

Digital album (mDA) 1.9189 0.5661 ** 1.4709 0.2838 ***

Digital single (mDS) 1.1264 0.2851 *** 1.3300 0.1750 ***

Innovation effect

Album (pCD & pDA) 0.0092 0.0017 *** 0.0080 0.0010 ***

Digital single (pDS) 0.1794 0.0351 *** 0.1769 0.0267 ***

Unlicensed music (pFM) 0.0131 0.0027 *** 0.0127 0.0022 ***

Imitation effect

CD (qCD) 0.1649 0.0282 *** 0.1796 0.0281 ***

Digital album (qDA) 0.3231 0.0462 *** 0.3855 0.0275 ***

Digital single (qDS) 0.6023 0.1233 *** 0.5834 0.1554 ***

Unlicensed music (qFM) 0.4793 0.0313 *** 0.5022 0.0293 ***

Consumption behavior change due to unbundling (β)

1.1067 0.3018 ** 1.0429 0.2037 ***

per Sub. consumption of unlicensed music (δ) 0.2406 0.0360 *** 0.2441 0.0336 ***

Proportion of digital single (α) 0.7739 0.0477 *** 0.7880 0.0325 ***

Control for time effects (γ) 0.2159 0.0345 *** 0.2343 0.0256 ***

Adj. R-Square

CD 0.9766 0.9750

Digital album 0.9669 0.9346

Digital single 0.9989 0.9987

Unlicensed music 0.9997 0.9997

Number of observation 31 30

Number of instruments - 9

Hansen’s J statistic - 12.71

Degree of Freedom - 26

P-value - 0.9864 *** p<0.001; ** p<0.01

 

17

(a) (b)

Figure 3. Model fit to the data of (a) NLSUR and (b) nonlinear system GMM

A detailed examination of the model parameters provided in Table 2 reveals the following

interesting observations.

(i) Our model estimates that if there was no digital music (i.e., digital single, digital album, and

unlicensed music were unavailable), the market potential (peak sales) of CD ( CDm ) could have

reached 2.07 billion (2.06 billion in the nonlinear system GMM estimation) units, approximately

twice the true peak of CD sales, which were 0.94 billion units in 2000. Similarly, if there is no

unlicensed music, the market potentials of digital single ( DSCD mm ) and digital album (

DACD mm 1 ) would reach 2.79 billion (2.98 billion in the nonlinear system GMM estimation)

and 2.33 billion (1.88 billion in the nonlinear system GMM estimation) units respectively.18 The

digital single and digital album sales are apparently still growing and have not reached the peak yet.

                                                            18 Recall that DAm and DSm reported in Table 3 are the increase in the market potential of digital album and digital

single respectively over the market potential of CD.

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.619

82

1984

1986

1988

1990

1992

1994

1996

1998

2000

2002

2004

2006

2008

2010

2012

Units Shipped (Billion units)

CD

Digital Single

Digital Album

Estimated (NLSUR)

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1982

1984

1986

1988

1990

1992

1994

1996

1998

2000

2002

2004

2006

2008

2010

2012

Units Shipped (Billion units)

CD

Digital Single

Digital Album

Estimated (Nonlinear system GMM)

 

18

(ii) Our estimation results confirm that innovation and imitation factors are indeed not identical across

different music formats.19 All ip and iq are significant at 0.1 percent level, and significantly different

across each different formats. The comparison of innovation and imitation factors between different

music formats provides several interesting results. First, the innovation and imitation effects for all

forms of digital music (i.e., digital album, digital single, and unlicensed music) are estimated to be

stronger than those for CD. This result is plausible because of the following reasons. Digital music

generally has a lower transaction cost, and thereby, has a lower barrier for diffusion compared to CD.

Also, since digital music is distributed through the Internet, the potentially stronger word-of-mouth

effect on the Internet may favor better diffusion of digital music compared to CD. Second, due to a

smaller unit size (i.e., individual song), digital single would incur a lower opportunity cost than

digital album, and hence, has a stronger innovation and imitation effects. Third, given that unlicensed

music is also distributed/shared over the Internet and consumers do not “pay” for it, it is somewhat

surprising to find that the innovation and imitation effects for unlicensed music are estimated to be

weaker than that for digital single.20 We believe that the inconvenience and risks associated with

consuming unlicensed music (e.g., consumers are often asked to register at the website and install

multiple software programs to consume unlicensed music, and unlicensed music listeners face various

risks such as risks of being caught as well as privacy breaches and computer virus) might cause the

weaker innovation and imitation effects.

(iii) We scale the data and use billion as a unit for sales (units shipped) of music and 10 million as a unit

for the Internet subscribers. Hence, = 0.2406 (0.2441 in the nonlinear system GMM estimation)

                                                            19 The literature (e.g., Norton and Bass 1987) argues that the assumption of constant ip and iq can be easily

verified using the model fit. If the model without the assumption fits better than with the assumption, it suggests that the assumption needs to be relaxed. 20 While some industry experts argue that iTunes’ price (99 cents) for an individual song (download digital single) is attractive enough to stop people consuming unlicensed music (see http://online.wsj.com/article/SB10001424052970204002304576629463753783594.html), some others argue that compared to the costs to consume unlicensed music, 99 cents might be still too high (e.g., Sandy Pearlman’s 5-cent solution).

 

19

implies that each Internet subscriber consumes, on average, about 24 units of unlicensed music every

year. This estimate is of the same magnitude as some other estimates made using different

approaches. For instance, Institute for Policy Innovation (IPI) estimated that 2.64 billion songs were

consumed illegally in 2005 (Siwek 2005). The number of (broadband) Internet subscribers in 2005

was 51 million; hence, this is equivalent to 52 units of unlicensed music consumption per Internet

subscriber a year. Although, IPI’s estimate was greater than ours, we would like to note that the

estimation of IPI was based on the assumption that 20 percent of illegal music consumption could

have been converted to music sales if unlicensed music was unavailable. In contrast, IFPI

(International Federation of the Phonographic) noted that only 10 percent of illegal downloads is a

loss to the industry (Lindvall 2011). If we follow the assumption of IFPI, the estimated magnitude (26

units) of unlicensed music consumption per Internet subscriber of IPI is similar with ours.

(iv) Our model estimates that the attrition of demand for music from CD to unlicensed music diminishes

by about 20 percent every year ( = 0.2156 in the NLSUR and 0.2343 in the nonlinear system GMM

estimations) with the introduction of purchased digital music – digital album and digital single. This

suggests that the option of purchasing digital music indeed replaces some of the attrition of demand

from CD; that is, the purchased digital music option converts some non-paying music consumers to

paying music consumers. In fact, the report of NPD group shows that overall illegal consumption of

music dropped by 26 percent in 2012 (Whitney 2013).

(v) The model estimates that about 80 percent ( = 0.7739 in the NLSUR and 0.7880 in the nonlinear

system GMM estimations) of the demand that migrates from CD to purchased digital music chooses

digital single and the remaining 20 percent chooses digital album. That is, the unbundling effect

dominates the generation substitution effect in the music industry. Our estimate is somewhat

consistent with a survey that Amberg and Schröder (2007) report in their paper. They state that 69.4

percent of their interview respondents express a preference for purchasing an individual song over an

entire album.

 

20

(vi) Finally, the model reveals that one unit demand for CD converts to about 1.1 units demand for digital

single ( = 1.1067 in the NLSUR and 1.0429 in the nonlinear system GMM estimations). Again, this

result supports the view that most of the (bundled) songs in a CD album are not preferred by

consumers, and that when they have the option of purchasing individual songs, they are more likely to

buy only their preferred songs. Our estimate is somewhat similar with that in Elberse (2010) that

estimates an overall reduction of one-third of the music sales due to the unbundling.

In summary, the overall model fit, statistical significance of model parameters, and the

remarkable closeness of our estimates of various parameters to others’ estimates derived using different

methodologies offer a significant level of confidence regarding the findings and implications derived

from the model.

4. Quantification of Migration Patterns and Impact on the Music Industry

In this section, we quantify the impact of the Internet and digital music formats on the music industry by

computing how the demand for music migrates (i) from CD to various digital music formats and (ii) from

purchased digital music to unlicensed music over time. Since the estimation results from NLSUR and

nonlinear system GMM are not significantly different,21 for convenience sake, we use the estimation

results from NLSUR to present our analysis.

Using equations (6) – (10), we compute the generation substitution, tGS , unbundling, tUB ,

attrition of demand from CD to unlicensed music, tACD , attrition of demand from digital album to

unlicensed music, tADA , and attrition of demand from digital single to unlicensed music, tADS , in

each time period t . Figure 4 shows these migration patterns over the study period. We make the following

interesting observations from the observed migration patterns.

                                                            21 The result of t-test suggests that the differences in parameter estimates are not significant at 5 percent level.

 

21

(a) (b)

Figure 4. (a) Migration of demand from CD and (b) attrition of demand from purchased digital music

(i) Over the 1997 – 2012 period, an estimated 13.04 billion units of demand for CD has migrated to

digital music – digital album, digital single, and unlicensed music. That is, if digital music was

unavailable, by end of 2012, 13.04 billion more units of CD could have been sold.22 Overall, 50

percent of this “loss in CD sales” is due to the unbundling (i.e., migration to digital single) and 47

percent is due to the attrition to unlicensed music. Only 3 percent of the loss is due to the generation

substitution (i.e., migration to digital album). In our estimates, the unbundling effect and the attrition

effect dominate the generation substitution effect. This implies that the effects of unbundling and

piracy dominate the pure effect of “digital” music. Furthermore, during the period 1997-2012, the

unbundling effect slightly exceeds the attrition effect, suggesting that the “loss in CD sales” is caused

slightly more by the “unbundling” of the album than by the “free” digital music.

(ii) Figure 4(a) shows how the relative migration patterns from CD to different digital music formats have

changed over time. After the introduction of digital album and digital single, the attrition of demand

from CD to unlicensed music has steadily decreased each year. Out of a total of 6.16 billion units of

                                                            22 14.11 billion units of CD were sold in about 30 years, from 1983 to 2012.

0.0

0.3

0.6

0.9

1.2

1.5

1.8

1996

1997

1998

1999

2000

2001

2002

2003

2004

2005

2006

2007

2008

2009

2010

2011

2012

Migration of demand from CD (Billion units)

Attrition of demand

Unbundling effect

Generation Substitution

0

0.4

0.8

1.2

1.6

2

2003 2004 2005 2006 2007 2008 2009 2010 2011 2012

Attrition of demand from purchased digital music (Billion units)

Digital single

Digital album

 

22

attrition from CD to unlicensed music during 1997 – 2012 period, 26 percent (1.60 billion units)

occurred before 2004 when purchased digital music option became available. Migration of demand

from CD to digital album and digital single, on the other hand, has increased steadily since 2004. This

indicates that the music industry’s strategy of providing consumers with an option to purchase music

in digital format was successful in stemming the attrition of demand for music from CD to unlicensed

music. The rate of increase is clearly higher for migration to digital single than digital album. In 2012,

72 percent of migration from CD was to digital single, 16 percent was to unlicensed music, and the

rest was to digital album. Please note that this is an estimate for year 2012. As we discussed above,

overall, 47 percent of “loss in CD sales” was due to online music piracy.

(iii) While Figure 4(a) shows that the attrition effect of unlicensed music on CD has decreased since the

introduction of purchased digital music, Figure 4(b) demonstrates that the attrition effect of

unlicensed music on digital album and digital single has increased. Over the 2004 – 2012 period,

unlicensed music has caused the attrition of 7.13 billion units (an average of 0.79 billion units per

year) of digital single and 2.21 billion units (an average of 0.25 billion units per year) of digital

album.23 These numbers suggest that the average attrition effect of unlicensed music on digital single

(0.79 billion units per year) is stronger than that on CD (0.39 billion units per year) and digital album

(0.25 billion units per year). This result is plausible since unlicensed music would be a more direct

substitute for digital single than CD and digital album.

(iv) Between digital single and digital album, the rate of increase of attrition effect of unlicensed music on

the former has decreased but on the latter has increased (Figure 4(b)). In fact, the attrition effect of

unlicensed music on digital single has remained almost flat during the period 2010 – 12. This

suggests that unlicensed music poses more challenge for digital album in recent years analogous to

the challenge it posed to CD before the introduction of purchased digital music option.

                                                            23 By end of 2012, 7.13 billion units of digital single and 0.53 billion units of digital album were sold.

 

23

In summary, the magnitudes of the different migration effects suggest that while the unlicensed

music (i.e., online music piracy) was a huge factor in the reduction in demand for CD, after the

introduction of purchased digital music, it is the unbundling effect that has contributed the most to the

decline in demand for CD. However, the attrition effect of unlicensed music on the music industry has not

been gone away. Instead of luring demand away from CD, the unlicensed music has been taking more

consumers away from digital single, and especially, digital album, and the rate of increase on the digital

album has increased in recent years.

5. Implications and Discussion

A few implications for the music industry emerge from our estimation results. First, our result confirms

that the attrition effect of unlicensed music on the music industry is substantial. However, we would like

to emphasize that it could have been worse if the industry did not provide consumers with the purchased

digital music – digital single and digital album – option and focused only on stopping people from

consuming unlicensed music. Since the purchased digital music option was introduced, about 7 billion

units of demand for CD that would have migrated to unlicensed music has migrated to either digital single

(6.5 billion units) or digital album (0.5 billion units). Further, digital album and digital single have created

“new” demand for (digital) music. In 2012, 1.4 billion and 117 million units of digital single and digital

album were sold respectively. This is already larger than CD sales in 2000 (943 million unites) when it

reached the peak and apparently digital single and digital album sales are still growing. It is estimated that

the peak sales of purchased digital music would be greater than that of CD by about 3 billion units. These

findings support the view that the industry’s strategy needs to be focused on converting the attrition of the

demand to profitable migration, rather than stopping the attrition, by offering a new product.

Second, it is widely recognized that consumers prefer only a small fraction of songs bundled in an

album. In fact, in our estimation, one unit demand of CD has converted only to 1.1 unit demand of digital

single, and about 80 percent of the demand that migrates from CD to purchased digital music chooses

digital single – the remaining 20 percent chooses digital album. Further, it was online music piracy when

 

24

purchased digital music was unavailable, but in the presence of purchased digital music, unbundling is the

dominant factor that leads the decline in CD sales. In 2012, 72 percent of migration from CD was to

digital single, 16 percent was to unlicensed music, and the rest was to digital album. Due to this

dominance of (digital) single format, although consumers are purchasing more (digital) music than ever,

the music industry’s revenue is still down from when the revenue reached the peak in 2000. In 2012, the

industry generated $1.2 billion of revenue from 117 million units of digital album sales whereas $1.6

billion of revenue from 1.4 billion units of digital single sales.24 The revenue from digital single and

digital album together in 2012 was only about 20 percent of the peak revenue from CD in 2000 ($13.2

billion). Cannibalization due to unbundling, rather than attrition due to online music piracy, is now the

dominant factor that leads the decline in industry revenue. These findings suggest that the music industry

likely benefitted tremendously by implementing the (seller-defined) bundling strategy though this

bundling strategy does not seem to be a good strategy anymore. The music industry may need to exploit

the benefits offered by an album format (i.e., higher revenue) by adopting a new product design strategy.

One possible strategy could be to allow consumers to create their own customized (digital) bundles.

Although Apple dropped it from iTunes in late 2012, iMix might be an example. iMix allowed any users

to create a digital bundle (playlist) and make it available for purchase at iTunes.

Third, while purchased digital music option has reduced the attrition of demand from CD to

unlicensed music, attrition to unlicensed music is emerging as a challenge for purchased digital music. On

average, every year, there are attrition of 0.79 billion units of digital single and 0.25 billion units of digital

album to unlicensed music. Especially, compared to 2011, in 2012, while the attrition effect of unlicensed

music on digital single increased only by 1.4 percent, that on digital album increased by 29 percent. These

findings support the notion that the industry may have to move away from offering seller-defined bundles

and move towards a new product design strategy such as a hybrid of album and single formats.

                                                            24 The average prices of CD, digital album, and digital single, respectively, are $14.41, $10.07, and $1.05.

 

25

6. Conclusion

Many industries have been disrupted by technological innovations and not all disruptive technologies

improve the industry’s (or firm’s) profitability (Adner and Zemsky 2005). In the music industry, the

Internet and digital music formats have fundamentally altered the way music is packaged, distributed and

consumed. The major contribution of our research lies in identifying, isolating, and quantifying the

different components of the disruption caused by the Internet and digital music formats. The different

components have different implications for the music industry.

Despite the strong statistical results that we obtained in our study, it has a few limitations. First,

following prior literature, we used the Internet penetration as a proxy measure for the demand for

unlicensed music by assuming that on average, Internet subscribers consume the same amount of

unlicensed music. Although the direct measure for the demand for unlicensed music is nearly unavailable,

it would be worthwhile to look at a direct measure and verify the results. Second, we assumed that there is

no “new” demand for unlicensed music (i.e., OMm = 0) and measure the upper limit of the impact of

online music piracy. Once a direct measure for the demand for unlicensed music becomes available, we

can relax the assumption and measure more precise impact of unlicensed music. Third, as many diffusion

studies (e.g., Bass 1969, 1995; Mahajan and Peterson 1978; Bucklin and Sengupta 1993; Mahajan and

Muller 1996), we estimated the model using relatively fewer number of data points. However, we would

like to note that our data covers the full lifecycle of each music format. Lastly, despite its advantage, the

diffusion model approach has some limitations. Since the model uses aggregate market-level data, the

model does not consider user heterogeneity (Van den Bulte and Stremersch 2004; Susarla et al. 2012) and

provides limited individual-level implications (Mahajan et al. 1990). In addition, the model does not

capture the direct effects of decision variables such as price and advertising (Chandrasekaran and Tellis

2007). Alternatively, prior studies have used a log-logistic distribution accelerated failure time model

(Xue et al. 2011) and the decomposed diffusion model (Susarla et al. 2012). Nonetheless, the diffusion

model has been successfully applied to estimate the sales in many different industries including retail

 

26

service, telecom, information technology, agriculture, pharmaceutical, and consumer durable goods (see

Mahajan et al. 1990, Bass 2004, and Meade and Islam 2006 for the review of diffusion models). In

particular, Bass (1995), Bewley and Griffiths (2003), Pae and Lehmann (2003), and Boswijk and Franses

(2005) apply the model and estimate the sales of recording formats (e.g., LP, tape, and CD). As it is

shown in Chen et al. (2013), given that “communication” is a significant factor that derives music sales,

the diffusion model might be a good technique to measure the diffusion of and demand migration patterns

among different music consumption options.

Although we focus on the demand migration patterns in the music industry, we would like to note

that our model can be easily extended and used to shed some light on the complicated migration of

demand and competition landscape in the industry, where ’tipping across markets’ or ’platform

envelopment’ phenomenon occurs (Eisenmann et al. 2011). In many industries, a company has moved

into an adjacent industry (or market) by bundling new features into its existing product (Gawer and

Cusumano 2008). For instance, Apple has moved into portable PC market by adding new features to its

“tablet” device. This tablet device has complicated the competition landscape in the portable PC market

and fragmented the demand for the portable PC. For the portable PC industry, it would be important to

understand the migration patterns of the demand from an old portable PC (e.g., an older version laptop

PC) to new portable PCs (e.g., a newer version laptop PC and tablet PC) as well as the competition

between the tablet device and laptop PCs to make a good strategic decision for the future. The model

presented in this paper can be used to analyze such decisions.

 

27

References

Adner, R., and P. Zemsky. 2005. Disruptive Technologies and the Emergence of Competition. RAND Journal of Economics. 36(2) 229-254.

Amberg, M., and M. 2007. Schröder. E-Business Models and Consumer Expectations for Digital Audio Distribution. Journal of Enterprise Information Management. 20(3) 291-303.

Andersen, B., and M. Frenz. 2010. Don’t Blame the P2P file-sharers: The Impact of Free Music Downloads on the Purchase of Music CDs in Canada. Journal of Evolutionary Economics. 20(5) 715-740.

Atkinson, J. 2008. MP3 vs AAC vs FLAC vs CD. Stereophile. Available at: http://www.stereophile.com/features/308mp3cd.

Bass, F.M. 1969. A new-product growth model for consumer durables. Management Science. 15(1) 215-227.

Bass, F.M. 1995. Empirical generalizations and marketing science: A personal view. Marketing Science. 3(2) G6-19.

Bass, F.M. 2004. Comments on a new product growth for model consumer durables. Management Science. 50(12) 1833-1840.

Bass, F.M., T.V. Krisnan, and D.C. Jain. 1994. Why the Bass model fits without decision variables. Marketing Science. 13(3) 203-223.

Bender, M.T., and Y. Wang. 2009. The Impact of Digital Piracy on Music Sales: A Cross-Country Analysis. International Social Science Review. 84(3/4) 157-170.

Bewley, R., W.E. Griffiths. 2003. The penetration of CDs in the sound recording market: issues in specification, model selection and forecasting. International Journal of Forecasting. 19(1) 111-121.

Bockstedt, J.C., R.J. Kauffman, and F.J. Riggins. 2006. The Move to Artist-Led On-Line Music Distribution: A Theory-Based Assessment and Prospects for Structural Changes in the Digital Music Market. International Journal of Electronic Commerce. 10(3) 7-38.

Boswijk, H.P., and P.H. Franses. 2005. On the Econometrics of the Bass Diffusion Model. Journal of Business & Economic Statistics. 23(3) 255-268.

Bucklin, L.P., and S. Sengupta. 1993. The co-diffusion of complementary innovations: Supermarket scanners and UPC symbols. Journal of Product Innovation Management. 10(2) 148-160.

Cellan-Jones, R. 2013. Music, TV and Digital Disruption. BBC. Available at: http://www.bbc.co.uk/news/technology-21371870.

Chandrasekaran, D., and G.J. Tellis. 2007. A Critical Review of Marketing Research on Diffusion of New Products. in Naresh K. Malhotra (ed.) Review of Marketing Research, Volume 3, Emerald Group Publishing Limited, 39-80.

Chellappa, R. and C. Chen. 2009. MySpace Killed the Radio Star? The Impact of Online Sampling on Song Sales. Thirtieth International Conference on Information Systems. Phoenix, AZ.

 

28

Chen, H., P. De., and Y. Hu. 2012. IT-Enabled Broadcasting in Social Media: An Empirical Study of Artists’ Activities and Music Sales. Working paper. Available at SSRN: http://ssrn.com/abstract=2201430 or http://dx.doi.org/10.2139/ssrn.2201430.

Chow, G.C. 1967. Technological Change and the Demand for Computers. The American Economic Review. 57(5) 1117-1130.

Christensen, C.M. 1997. The Innovator’s Dilemma: When New Technologies Cause Great Firms to Fail. Harvard Business School Press, Boston.

Chu, C.P., and J.G. Pan. 2008. The Forecasting of the Mobile Internet in Taiwan by Diffusion Model. Technological Forecasting and Social Change. 75(7) 1054-1067.

Creel, M., and M. Farell. 1996. SUR Estimation of Multiple Time-series Models with Heteroscedasticity and Serial Correlation of Unknown Form. Economic Letters. 53(3) 239-245.

Danaher, B., S. Dhanasobhon, M.D. Smith, and R. Telang. 2010. Converting Pirates without Cannibalizing Purchasers: The Impact of Digital Distribution on Physical Sales and Internet Piracy. Marketing Science. 29(6) 1138-1151.

Danaher, P.J., B.G.S. Hardie, and W.P. Putsis Jr. 2001. Marketing-Mix Variables and the Diffusion of Successive Generations of a Technological Innovation. Journal of Marketing Research. 38(4) 501-514.

Day, G.S. 1981. The Product Life Cycle: Analysis and Applications Issues. Journal of Marketing. 45(4). 60-67.

Dewan, S., D. Ganley, and K.L. Kraemer. 2010. Complementarities in the diffusion of personal computers and the internet: implications for the global digital divide. Information Systems Research. 21(4) 925-940

Dewan, S., and Ramaprasad, J. 2014. Social Media, Traditional Media, and Music Sales. MIS Quarterly. 38(1) 101-121.

Eisenmann, T., G. Parker, and M. Van Alstyne. 2011. Platform Envelopment. Strategic Management Journal. 32(2) 1270-1285.

Elberse, A. 2010. Bye-Bye Bundles: The Unbundling of Music in Digital Channels. Journal of Marketing. 74(3) 107-123.

Federal Communication Commission. n.d. Communications History – Making the Connections. Available at: http://transition.fcc.gov/omd/history/internet/making-connections.html.

Floyd, A. 1962. Trend Forecasting: A Methodology for Figure of Merit. In Technological Forecasting for Industry and Government: Methods and Applications. J.R. Bright ed. Prentice-Hall.

Fourt, L.A., and J.W. Woodlock. 1960. Early Prediction of Market Success for New Grocery Products. Journal of Marketing. 25(2) 31-38.

 

29

Gardner, E. 2013. Recorded Music Industry Posts First Revenue Growth in 13 Years. The Holloywood Reporter. Available at: http://www.hollywoodreporter.com/news/recorded-music-industry-revenue-rises-424574.

Gawer, A., and M.A. Cusumano. 2008. How Companies Become Platform Leaders. MIT Sloan Management Review. 49(2) 28-35.

Givon, M., V. Mahajan, and E. Muller. 1995. Software Piracy: Estimation of Lost Sales and the Impact on Software Diffusion. Journal of Marketing. 59(1) 29-37.

Goldenberg, J., B. Libai, and E. Muller. 2010. The Chilling Effects of Network Externalities. International Journal of Research in Marketing. 27(1) 4-15.

Hann, I.H., B. Koh, and M.F. Niculescu. 2013. The Adoption of Multi-Generational Platforms in the Presence of Intergenerational Services. NET Institute Working Paper No. 12-17. Available at SSRN: http://ssrn.com/abstract=2164390 or http://dx.doi.org/10.2139/ssrn.2164390.

Hansen, L.P. 1982. Large Sample Properties of Generalized Method of Moments Estimators. Econometrica. 50(4) 1029–1054.

Heeler, R.M., and T.P. Hustad. 1980. Problems in predicting new product growth for consumer durables. Management Science. 26(10) 1007-1020.

Hong, S. 2007. The Recent Growth of the Internet and Changes in Household-level Demand for Entertainment. Information Economics and Policy. 19(3-4) 304-318.

Hong, S. 2011. Measuring the Effect of Napster on Recorded Music Sales: Difference-in-differences Estimates under Compositional Changes. Journal of Applied Econometrics. 28(2) 297-324.

Islam, T. and N. Meade. 1997. The Diffusion of Successive Generations of a Technology: A More General Model. Technological Forecasting and Social Change. 56(1) 49-60.

Jeong, B.K., M. Khouja, and K. Zhao. 2012. The Impacts of Piracy and Supply Chain Contracts on Digital Music Channel Performance. Decision Support Systems. 53(3) 590-603.

Jiang, Z., and D.C. Jain. 2012. A Generalized Norton-Bass Model for Multigeneration Diffusion. Management Science. 58(10) 1887-1897.

Jun, D.B., and Y.S. Park. 1999. A Choice-Based Diffusion Model for Multiple Generations of Products. Technological Forecasting and Social change. 61(1) 45-58.

Kim, N., D.R. Chang, and A.D. Shocker. 2000. Modeling intercategory and generational dynamics for a growing information technology industries. Management Science. 46(4) 496-512.

Koh, B., B.P.S. Murthi, and S. Raghunathan. 2013. Shifting Demand: Online Music Piracy, Physical Music Sales, and Digital Music Sales. Journal of Organizational Computing and Electronic Commerce. Journal of Organizational Computing and Electronic Commerce. Forthcoming.

 

30

Libai, B., E. Muller, and R. Peres. 2009. The diffusion of services. Journal of Marketing Research. 46(2) 163-175.

Liebowitz, S.J. 2008. Testing File Sharing’s Impact on Music Album Sales in Cities. Management Science. 54(4) 852-859.

Lindvall, H. 2011. IFPI Music Report Dispels the Myths Surrounding Piracy. The Guardian. Available at: http://www.theguardian.com/technology/pda/2011/jan/20/ifpi-report-music-piracy.

Mahajan, V., R.A. Peterson. 1978. Innovation Diffusion in a Dynamic Potential Adopter Population. Management Science. 24(15) 1588-1597.

Mahajan, V., E. Muller, and F.M. Bass. 1990. New Product Diffusion Models in Marketing: A Review and Directions for Research. Journal of Marketing. 54(1) 1-26.

Mahajan, V., and E. Muller. 1996. Timing, Diffusion, and Substitution of Successive Generations of Technological Innovations: The IBM Mainframe Case. Technological Forecasting and Social Change. 51(2) 109-132.

Mansfield, E. 1961. Technical Change and the Rate of Imitation. Econometrica. 29(4) 741-766.

Meade, N., and T. Islam. 2006. Modeling and Forecasting the Diffusion of Innovation – A 25-year Review. International Journal of Forecasting. 22(3) 519-545.

Michalakelis, C., D. Varoutas, and T. Sphicopoulos. 2010. Innovation Diffusion with Generation Substitution Effects. Technological Forecasting & Social Change. 77(4) 541-547.

Michel, N.J. 2006. The Impact of Digital File Sharing on the Music Industry: An Empirical Analysis. Topics in Economic Analysis & Policy. 6(1) 1-22.

Mortimer, J.H., C. Nosko, and A. Sorensen. 2012. Supply Responses to Digital Distribution: Recorded Music and Live Performances. Information Economics and Policy. 24(1) 3-14.

Moutinho, L., and G.D. Hutcheson. 2011. The SAGE dictionary of quantitative management research. Sage Publications Ltd.

Niculescu M.F., and S. Whang. 2011. Codiffusion of wireless voice and data services: An empirical analysis of the Japanese mobile telecommunications market. Information Systems Research. 23(1) 260-279.

Norton, J.A., and F.M. Bass. 1987. A diffusion theory model of adoption and substitution for successive generations of high-technology products. Management Science. 33(9) 1069-1086.

Oberholzer-Gee, F., and Strumpf, K. 2007. The Effect of File Sharing on Record Sales: An Empirical Analysis. Journal of Political Economy. 115(1) 1-42.

Pae, J.H., and D.R. Lehmann. 2003. Multigeneration Innovation Diffusion: The Impact of Intergeneration Time. Journal of the Academy of Marketing Science. 31(1) 36-45.

 

31

Parks Associates. 2013. Research and Markets: Evolution of Digital Music on Connected Devices. Available at: http://www.researchandmarkets.com/research/6r6nx4/evolution_of.

Peterson, R.A., and V. Mahajan. 1978. Multi-product growth models. In Research in Marketing. J. Sheth ed. Greenwich: JAI Press.

Pohlmann K.C. 1989. Compact Disc: A Handbook of Theory and Use. A-R Editions.

Recording Industry Association of America. 2007. The CD: A Better Value Than Ever. An RIAA Report Prepared By the Communications and Strategic Analysis Department of the Recording Industry Association of America. Available at: http://76.74.24.142/F3A24BF9-9711-7F8A-F1D3-1100C49D8418.pdf.

Rob, R., and Waldfogel, J. 2006. Piracy on the High C’s: Music Downloading, Sales Displacement, and Social Welfare in a Sample of College Students. Journal of Law and Economics. 49(1) 29-62.

Rogers E.M. 1962. Diffusion of Innovations. 1st ed. New York: Free Press.

Siwek, S.E. 2007. The True Cost of Sound Recording Piracy to the U.S. Economy. Institute for Policy Innovation, IPI Center for Technology Freedom, Policy Report 188.

Smith, M.D., and R. Telang. 2010. Piracy or Promotion? The Impact of Broadband Internet Penetration on DVD sales. Information Economics and Policy. 22(4) 289-298. Suarez, F.F., M.A. Cusumano, and S.J. Kahl. 2013. Services and the Business Models of Product Firms: An Empirical Analysis of the Software Industry. Management Science. 59(2) 420-435.

Susarla, A., J.H. Oh, and Y. Tan. Social Networks and the Diffusion of User-Generated Content: Evidence from YouTube. Information Systems Research. 23(1) 23-41.

Van den Bulte, C., and S. Stremersch. 2004. Social Contagion and Income Heterogeneity in New Product Diffusion: A Meta-Analytic Test. Marketing Science. 23(4) 530-544.

Waldfogel, J. 2010. Music File Sharing and Sales Displacement in the iTunes Era. Information Economics and Policy. 22(4) 306-314.

Werker, C. 2003. Innovation, Market Performance, and Competition: Lessons from a Product Life Cycle Model. Technovation. 23. 281-290.

Whitney L. 2013. Illegal Music Downloads Dropped in 2012, Says Report. CNET. Available at: http://news.cnet.com/8301-1023_3-57571318-93/illegal-music-downloads-dropped-in-2012-says-report/.

Wooldridge, J.M. 2001. Applications of Generalized Method of Moments Estimation. Journal of Economic Perspectives. 15(4) 87-100.

White, H. 1980. A Heteroskedasticity-Consisetent Covariance Matrix Estimator and a Direct Test for Heteroskedasticity. Econometrica. 48(4) 817-838.

Xue, M., L. Hitt, and P. Chen. 2011. The Determinants and Outcomes of Internet Banking Adoption. Management Science. 57(2) 291-307.

 

32

Zentner, A. 2005. File Sharing and International Sales of Copyrighted Music: An Empirical Analysis with a Panel of Countries. Topics in Economic Analysis & Policy. 5(1) 1-15.

Zentner, A. 2006. Measuring the Effect of File Sharing on Music Purchases. Journal of Law and Economics. 49(1) 63-90.