JSTOR

I. Introduction

Several recent high profile legal cases have created a renewed focus on digitized products, intellectual property, and related copyright and pricing issues in an increasingly Internet‐enabled world (Clark 1999; Bravin 2000; Federal Trade Commission 2000; Crombie 2001; Gomes 2001; Gomes and Mathews 2001). Downloading, sampling, and sharing digital goods by Internet users who do not own these items in other forms has become a major issue. For the music recording and distribution industry, for example, this problem has turned quite acute.1 A Pew (2000) study found that about 14% of Internet users downloaded digitized music files from the Internet for free. Clark (2000) projected that by 2005, illegal online music sharing would result in annual sales losses of $3.1 billion. This indicated a significant disruption in the business model of the music industry.

The technology that facilitates such online sampling of digital audio and other digital goods is improving rapidly. Various software packages make it increasingly easier for consumers to search, download, and subsequently share music files on line with others (Ahlberg 2000). These use variations of peer‐to‐peer network models to share music in compressed formats with comparatively minor losses in sound quality.2 This phenomenon of digital music sharing was predicted by Alexander (1994b). With an increase in Internet connection speed and the availability of better search techniques, search and download times for these digital goods are being cut down significantly.3 Experts forecast that decreasing prices for data storage and faster connection speeds will soon allow consumers to use e‐mail to send entire disks of music (Clark 2000).

Music (like such items as digitized photographs and video clips) is an information good and, specifically, an experience good whose true value is revealed to a consumer only after it has been consumed (Nelson 1970). Music needs relatively little time (as well as little skill) to consume. In addition, it has the characteristics of a quasi‐public good in that once the good is provided to some consumers, it is very difficult to preclude other consumers from consuming it. In this context, the so‐called free‐rider, an individual who consumes a public good without actually paying for it, can undermine market efficiencies (Alexander 2002). Further, music items are also affected by network externalities, such as fan clubs that share information.4 Digital technology undoubtedly makes such information sharing and the sampling of such goods easier (Barua et al. 2001) and less costly (Cunningham, Alexander, and Adilov 2004), as the positive externality created by more samples leads to more sharing and further reduces the cost of information sharing and sampling. However, from a long‐term perspective, it is unclear whether such technology hurts intellectual property and related revenues. Intellectual property potentially has been under threat ever since the development of the printing press (Shapiro and Varian 1999); however, historically, new business models have consistently embraced new technology to generate increased revenues from intellectual property. The central motivation for our study stems from the fact that digital technology and products provide potential new avenues for revenue generation. Our primary focus is on the economic principles for such experience goods that potentially guide consumers and producers into new business models to reap benefits from the new medium.

At a fundamental level, artists create (or produce) music, which consumers pay to listen to (and enjoy). The dissemination of music has used various forms and technologies over time. The effect of online sharing technologies on music sales is inconclusive and is largely based on anecdotal journalistic evidence (Evangelista 2000; King 2000a, 2000b; Mathews 2000b; Mathews and Peers 2000; Peers and Gomes 2000). The proponents of online sharing argue that a lot of consumers download music to sample it and then subsequently purchase a CD if they like the music and that sharing therefore serves a useful marketing function by broadening the market for music (Boston 2000). Some have claimed that there is little evidence that online music sampling has actually decreased overall sales (Mathews and Peers 2000; Peers and Gomes 2000). Such sampling also potentially benefits artists by helping new artists to become “known.” Proponents also argue that digital compression decreases the quality of music in relation to a CD and that consumers with a high value for music potentially would also purchase the higher quality CD (http://www.mp3.com).5 Opponents of online music sharing, in particular many in the recording industry and some artists, argue that such sharing undermines CD sales. Their fundamental concern is this: “How can you build a business when the product you have developed is being cloned and given away on a mass scale for free?” (http://www.ifpi.org). They argue that piracy threatens the future of artists, composers, and record producers.

Many experts question the long‐term effectiveness of focusing on Internet “piracy facilitators” (Jerry 1987; Mason 1990; Garber 1996; Hardie et al. 1999). The common refrain is that the “genie is now out of the bottle” and simply shutting down such services will have limited effect (Hansen, Borland, and Yamamoto 2000; McCarthy 2000). Despite the scope and publicity that music sharing technology has triggered, little research exists on the effect that digital music sampling has on subsequent music sales.6 A substantially larger body of research has examined the related problem of software piracy. While software and music are both “information goods,” in that the marginal cost of production is virtually zero, certain key characteristics differentiate the two: (i) the quality of an original CD song is better than that of its electronically transferable compressed version. Software, however, requires a lossless compression for proper functioning. (ii) Music files are much smaller than a typical software application; hence, they take much less time to transfer and consume. (iii) Consumption of music requires little specific skill as compared to software; hence, the increased consumer base adds significant dynamics to the issue. (iv) Consumers closely relate a performer with a music product, unlike developers with a software product, which creates issues of personalized valuations for that musical product that depend on the performer. (v) The volume of available music is significantly larger than the existing volume of software products. This provides a far greater product sampling base compared to software products, which introduces additional levels of dynamics in music sampling and its analysis.

The focus of this research is on the economic dynamics of online digital music sampling. In particular, we study potential consumer benefits from digital music sampling technologies, the effect of such technologies on consumer purchasing and pirating behavior, and the impact of new sampling technologies on sales of music “superstars.”

First, we develop an economic model that incorporates incentive structures for producers and consumers of music items and we derive their implications for consumer surplus and producer profits. We then identify various market scenarios that affect the incentive structures and study the economic outcomes. Important propositions derived from the analyses state that (i) with the advent of new technology, consumers’ incentive for sampling and buying a music item is closely related to the value of the item to an individual consumer; (ii) the producer, via economic and technological instruments, can effectively combat piracy and increase revenues; and (iii) sampling technologies threaten the phenomenon of superstars, and this threat is proportional to the cost of sampling. Extensive empirical investigations, based on surveys and Billboard ranking charts, lend support to the economic model and validate the key results.

The remainder of the article is organized as follows. Section II discusses the related literature. Section III develops an analytical model of the economic factors, and Section IV presents the empirical study. A discussion of the key results and concluding remarks are the subjects of Section V.

II. Related Research

Digital goods are expensive to produce for the first copy (high fixed costs) and are inexpensive to reproduce and distribute for subsequent copies (near‐zero variable costs). Digital products also exhibit the fundamental characteristics of a public good, in that sharing with others does not reduce the consumption utility of the product. These traits of digital products facilitate their extensive and often illegal distribution worldwide. The growing importance of digital piracy has spurred research on the behavioral and economic understandings of piracy activity, especially in the area of software piracy (Conner and Rummelt 1991; Eining and Christensen 1991; Solomon and O’Brien 1991; Givon, Mahajan, and Muller 1995; Glass and Wood 1996; Gopal and Sanders 1997, 1998, 2000). Studies have reported that females, older individuals, and individuals with an ethical predisposition toward legal justice tend to pirate less. However, as mentioned earlier, digital music sampling has several unique characteristics, and little research exists on its effect on future sales.

The literature on software piracy suggests that economic factors play a key role in an individual’s decision to pirate. A number of studies have reported that the price of software has a significant impact on piracy (Cheng, Sims, and Teegen 1997; Gopal and Sanders 1997, 2000). As the price of the product increases, the net value from obtaining an illegal product increases—thus the negative impact of price on piracy. Gopal and Sanders (1998) highlight the income effect on national piracy levels. Their main recommendation is that the price of the product should be indexed with the affordability levels. In a similar result for global audio piracy, Burke (1996) finds that economic development, rather than copyright regulations, differentiates high and low piracy nations. The value of the product also plays a crucial role on an individual’s decision—higher‐valued consumers typically tend to purchase rather than pirate because they realize higher surplus from consumption (Conner and Rummelt 1991; Cheng et al. 1997; Gopal and Sanders 1998).

Technology that facilitates (or hinders) the piracy activity also plays an important role in piracy behavior (Conner and Rummelt 1991; Gopal and Sanders 1997; Bhattacharjee, Gopal, and Sanders 2003). Conventional wisdom suggests that easier access to pirated software increases piracy activity. Software firms have employed piracy protection technologies that raise the cost of pirating software as a weapon to combat piracy. Technologies such as encryption have been used to prevent easy piracy, with the intent that the prohibitive cost would cause would‐be pirates to legally purchase the product. However, Gopal and Sanders (1997) argue that, in the face of increasing preventive controls, individuals who do not legitimately acquire a digital good would simply do without it and that this behavior represents a drain on software publisher profits. They advocate the use of deterrent controls that rely on educational and legislative schemes to thwart piracy and increase seller revenues.

Some studies have argued that software piracy may not necessarily harm software publishers. Conner and Rummelt (1991) argue that, in the presence of network externalities (where the value a user derives depends on the size of the user base), the utility of the software increases with piracy because it increases the number of other individuals using it. The utility of product consumption increases with the total number of individuals using it. Givon et al. (1995), using innovation diffusion models, suggest that piracy provides word‐of‐mouth advertising for the software product and thus represents an efficient form of “sampling” that leads to a future purchase.

The economic argument, as it relates to incomplete information or information asymmetries, is this: “If consumers do not have accurate information about market prices or product quality, the market system will not operate efficiently. This lack of information may give producers an incentive to supply too much of some products and too little of others. In other cases, some consumers may not buy a product, even though they would benefit from doing so, while other consumers may buy products that leave them worse off. For example, consumers may buy pills that guarantee weight loss only to find that the pills have no medical value” (Pindyck and Rubenfield 2005, 612).

In essence, music consumers do not have accurate information on the quality of the music because the music is an experience good. Music publishers, because of the delay in obtaining market information for all of their music, may overinvest in certain music genres and underinvest in others. A typical strategy to overcome the inefficiencies and uncertainties in the market is to focus on the superstars. Music sampling has the potential to reduce this uncertainty, increase market efficiencies, and permit a broader base of talented musicians and singers to be successful.

The economics of music includes studies in the market structure of firms in the industry (Alexander 1994a, 1997) and the effect of chart success of an album on future sales (Strobl and Tucker 2000). The phenomenon of the superstar effect in music, and its related economics, has been well documented. A superstar owes his or her existence to intrinsic elements of talent (Rosen 1981); extrinsic elements of circumstance, or “luck”; and user expectations based on past performance (Adler 1985; MacDonald 1988; Hamlen 1991; Towse 1992). Models of superstardom have been presented by Chung and Cox (1994), Ravid (1999), and Crain and Tollison (2002), among others. At the heart of the superstar phenomenon lies the desire by consumers to minimize their search and sampling costs by choosing the most popular artist (Adler 1985). Music as an experience good requires information (or knowledge) about the particular item before its consumption. The search for information is costly, especially for relatively unknown artists. In such a case, consumers balance their additional search costs for unknown artists or items of music with their existing knowledge of a known “popular” artist. In a statistical sense, consumers correlate past performance with future outcomes (MacDonald 1988) and try to minimize the variability in their expectations of individual performances. There is also evidence that newly released recordings have higher demand elasticities than older or well‐known music items (Mixon and Ressler 2000).

Behavioral models of digital piracy are important for initiating educational and legal campaigns to reduce piracy (Mathews 2000a). These constitute indirect methods to modify consumer ethics and attitudes. They are investigated in detail in (Gopal et al. 2004), which also reports that deterrent strategies—used successfully in antisoftware piracy campaigns—have limited effect on music piracy. Additionally, music publishers are acknowledging that they “have to make buying music easier than stealing music” (http://www.business2.com/magazine/2000/12/23296.htm).

III. Analytical Modeling of Sharing Technologies

In this section, we present an economic model that analyzes consumer incentives to sample, pirate, and purchase music offerings, and the ensuing impact on seller’s revenues. The model also captures the economic implications of sharing technologies on music offerings by relatively unknown and well‐known artists. This is followed by an empirical model that is designed to validate the key results of the economic analysis via both primary and secondary data sources.

Commercially offered music is a quintessential experience good whose true value is revealed to the consumer only after the initial consumption (i.e., experiencing or sampling the music). The volume of music that is commercially available is vast and spans a panoply of genres, languages, artists, and themes.

Consider a music item that a consumer can legally purchase at a price of P.7 Let the function denote consumer i’s probability density function of the value for the music offered, and let . According to this denotation, V_i represents the maximum value attached by consumer i to music in general. Of course, this varies across consumers, and those with larger values of V_i display a higher affinity for music consumption. The function captures a priori expectations of music item K by consumer i, and it is a proper probability density function in that it satisfies the condition .

The uncertainty depicted in f_i^K(v) above arises from the a priori expectation of the value of music item K based on the artist’s reputation and information available about K. Given the large volume of available music, no consumer has all of the information on all of the music items. Further, music is an experience good whose true consumption utility can only be determined through a costly knowledge acquisition process. Hence, the functional form relates to a consumer’s prior exposure to the particular music and can vary across consumers and also across music items under consideration. The variance of v is the consumer’s uncertainty regarding the underlying value of the music item. Traditionally, variance reduction for a music item occurs through coverage in the press, live performances, music‐oriented radio stations, music television outlets, and other mechanisms that advertise the value of the music item to consumers. An important determinant of the variance is the artist(s) who created the music item. This arises from the prevalence of stardom in the music industry. From an economic standpoint, stardom is a market device used by consumers to economize on the learning and information acquisition costs. Instead of diversifying indefinitely across a large number of artists, which may necessitate significantly large costs of searching and learning, consumers may prefer to patronize a limited number of stars (Adler 1985). A superstar is a known entity whose music is widely listened to and enjoyed by many users; hence a consumer would place a lower uncertainty on music released by such superstars. A consumer who decides to directly purchase the music item K realizes the following expected benefit:

A. Consumer Incentives with Sharing Technologies

Internet‐enabled sharing technologies enable consumers to reduce the information uncertainty regarding commercially available music. Much to the chagrin of the recording industry, they also permit users to obtain illegal copies of the music without paying proper remuneration to the recording industry, the artists, and other entities involved in the creation and distribution of the music.

A consumer who considers using online sharing technologies to experience the music without first paying for it incurs a cost that we denote as C_sample. This captures the time and effort expended by the user in searching, downloading, and listening to the illegal copy of the music.8 If a consumer decides to sample, C_sample becomes a sunk cost. The following discussion details the economic benefits to a consumer from sampling.9

A consumer decides to sample if the net expected benefit from sampling is positive and larger than the benefit from direct buying (eq. [1]). After downloading a song, a consumer faces three subsequent choices: buy a legitimate version of the song, keep the illegal copy (this constitutes piracy), or discard the downloaded song.

Let the actual value of music item K to consumer i be denoted as λV_i. Clearly, λ assumes a value between zero and one. This value is revealed to the consumer only after she has experienced the music item. The decisions a consumer makes prior to experiencing a music item are governed by uncertainties regarding the value of λ. We capture this uncertainty via variable x ( ). Thus, prior to downloading the music item, the consumer is unaware of its true value,10 and therefore the sampling decision is driven primarily by the expectations captured in . The expected benefit (ignoring the sunk cost C_sample) of purchasing the item after sampling is Similarly, if the consumer decides to pirate the item, the realized value is

In the above expression, F ( ) captures the reduction in the consumption value of pirated music. The value deterioration can result from such reasons as poorer quality of downloaded music,11 availability of only partial music online for sampling,12 and other services that a legitimate seller might offer.13 Here E denotes the enforcement penalty due to copyright violations of owning an illegal music item, and k is the probability of enforcement. While the current legal strategy employed by the recording industry targets only entities that facilitate illegal sharing of music (e.g., Napster), consumers remain liable for owning pirated music. If the consumer decides to delete the music item after sampling, the realized value is zero. From (2) and (3), it follows that a consumer who samples will subsequently buy if

A consumer would, however, pirate the item if . This implies that

Finally, a consumer whose realized value is lower than kE/F would discard the downloaded music item. The decision to sample music prior to purchasing is driven by the distribution , the only information a consumer has regarding the value of x. Hence, the net expected benefit from sampling is as follows: where and .

The decision made by the consumer to buy directly or to sample before buying is based on the higher of the expected value estimations given by (1) and (6). As defined before, λV_i is the actual or true value of the music item revealed to the consumer who decides to sample, where the value of λ lies between zero and one. The postsampling decision made by a consumer (to buy, pirate, or discard the downloaded file) is driven by the surplus derived from the actions and is determined by λ. Figure 1 depicts the consumer’s decision‐making strategy.

(30 KB)

Fig. 1.— Consumer decision making

Open New Window

A number of factors play into the decision by consumers regarding sampling and subsequent purchase or illegal acquisition of music. In the following discussion, we present the impact of the function . We present two cases, representing high and low degrees of uncertainty in terms of prior expectations. The former relates to music offerings by relatively unknown artists, the latter to releases by well‐known superstars in the music industry. In order to make consumer benefit and seller profit calculations determinate, we employ specific functional forms of and assume uniformity of expectations across consumers. The generalizability of the results from this analysis is tested in the empirical methodology.

B. Value Uncertainty

We model the case of consumer uncertainty regarding the actual consumption value by specifying the function to be uniformly distributed; that is, Let , where I is the set of consumers, and let V_i be continuous in the range . Note that we make no specific assumptions regarding the distribution of V_i other than that it is continuous. Thus the results presented in the discussion that follows should apply for any customer demand function.

The expected benefits from direct buy and sampling are computed using equations (1) and (6): For ease of exposition, let

From the above equations, it follows that the consumer makes the following choices:

Among the samplers, the subsequent buy/pirate/discard decision is dictated by equations (4) and (5), as depicted in figure 1. Two cases of interest arise in this context: and , which are illustrated in figure 2. In the first case, some of these consumers sample first prior to making the purchase decision. When , two interesting scenarios arise: and In the first scenario, some of the consumers with will continue to buy the music item even if they sample the music first. In the second scenario, some of these consumers resort to piracy. When , consumers with continue to buy direct and remain unaffected by the sharing technologies. In this case, consumers with V_i in [γ, 2P] represent a “lost sales” segment—they would not buy directly, given their lower valuation for music compared to that of aficionados. They would also not sample a product with a diminished value (δv), since it would not provide them with positive expected surplus.

(47 KB)

Fig. 2.— Potential decisions of consumers with varying V_i

Open New Window

Proposition 1. The amount of piracy is nondecreasing in P and F and nonincreasing in kE.

Proof. Consumers whose maximum value for music is in the interval pirate music. The interval shrinks with lower P and F and higher values of kE. QED

The above proposition illustrates that price, enforcement, and value‐added services provided to legitimate customers are useful instruments to fight piracy. The popular press has frequently criticized the pricing policies adopted by the music industry, and a number of reports have claimed that high prices contribute to piracy in that industry (Federal Trade Commission 2000; Crombie 2001; Hoover 2001). The seller can eliminate piracy by setting the value of P at .

Consider the impact of C_sample. Note that C_sample is inversely related to γ (eq. [9]). When C_sample is substantially large, only a smaller proportion of samplers subsequently make a legitimate purchase (fig. 2). The underlying rationale is that consumers who incur a substantial cost to search and locate a music item will be loath to further spend P to obtain a legitimate copy. As the C_sample decreases, the total cost to sample and buy decreases, and as a result consumers are more likely to purchase after sampling music. These key results are highlighted in the following propositions.

Proposition 2a. The proportion of consumers who sample is inversely related to C_sample.

Proposition 2b. The proportion of samplers who subsequently buy is inversely related to C_sample.

As the market price of the music item increases, consumers are more likely to sample first rather than directly purchase. This follows because only those consumers with and make direct purchases, and the right‐hand sides of both inequalities increase with P. This implies that, as the market price of a music item increases, more consumers will spend a smaller amount of additional resources to experience the music item before they spend a larger amount to buy it. An alternate explanation is that, as the price increases, buyers exhibit a more risk‐averse attitude toward new music since they do not know its actual value (Mixon and Ressler 2000). This rationale is captured in the following proposition.

Proposition 3. The proportion of consumers who buy directly is inversely related to P.

Sampling, in essence, is a truth‐revelation mechanism that allows consumers who sample to make the purchase decision on the actual value of the music item. A lower actual value (λ) results in a smaller proportion of samplers who purchase the music item. This result is captured in the following proposition.

Proposition 4. The proportion of samplers who do not buy is inversely related to λ.

Proof. Among the samplers, those with the maximum value for music lower than do not then purchase the music item. The proportion of these consumers who either pirate or discard the music increases as λ decreases. QED

We now consider the impact of sharing technologies on seller revenues. Note that, absent sharing technologies, consumers with legally purchase the music. As the sampling costs decrease with increasing availability of free music online, more consumers engage in sampling prior to the purchase decision. A significant concern for the music industry is that this customer base will resort to sampling and then to piracy and that this can negatively affect their bottom line. The following proposition highlights the impact of sharing technologies on the revenues of the music industry.

Proposition 5. Sharing technologies have a positive impact on revenues when

Proof. When consumers with will purchase the music item. Thus, the same set of customers who would buy in the absence of sharing technologies will continue to do so. Additionally, sharing technologies provide an incentive to new customers whose value is in the range when as well as those whose value is in the range when to purchase the music after sampling. QED

An important result from the above proposition is that the revenue impact of sharing technologies depends on the actual value of the music to consumers. Music that is highly valued by the consumers will stand to gain from the availability of sharing technologies. Lower‐valued music will see decreased revenues as customers increasingly attempt to decipher the true value of the music via sampling. The proposition also suggests that efforts by the producer to increase kE (enforcement) and lower F (increasing the value‐added for legitimate consumers) constitute a more effective strategy for the music industry than the attempts to make C_sample high for consumers. A high value of C_sample might shore up the existing consumers by dissuading them from downloading and sampling music. However, this strategy will not attract new customers who can potentially enhance the revenues of the music industry, as it has the potential of dissuading customers from engaging in any type of music transaction.

Overall, the propositions provide important welfare implications of online music sharing technologies. Clearly, consumer surplus is always increased by sharing technologies only when it adds to the surplus. Pirates gain as they obtain music without paying for it. For samplers, the surplus is enhanced as their purchasing behavior is based on informed choices. It follows immediately from proposition 5 that, when overall welfare is higher with the presence of online sharing networks than without. Such networks provide useful instruments for the music business to sustain and grow in the presence of new sharing technologies.

C. Stardom and Sharing Technologies

The analysis presented thus far applies to unknown artists, since consumers are assumed to be uncertain as to the true value of the music offerings from these artists. A relatively large amount of information available about superstar artists significantly reduces the variability in consumer expectations of their music. We now analyze the impact of sharing technologies and stardom on sales. In what follows, superscript s denotes a superstar and n denotes an unknown artist. We consider a comparative analysis where the true value of a music item offered by a superstar is the same as that from an unknown artist, that is, .

Proposition 6. When and , a superstar reaps higher sales than the unknown artist.

Proof. When , a consumer would not choose to spend extra to sample and determine the true value of a music item but would aim to maximize surplus by choosing the item with the higher expected value and lower purchasing cost (which may include the sampling cost of the unknown artist’s music). The rationale is that, given music items from a superstar and an unknown artist (with equal true values, i.e., ), the consumer would directly choose the superstar’s music, because it has a higher expected value and lower variability, rather than sample and subsequently buy the unknown artist’s music. This leads to higher sales for the superstar. QED

Proposition 7. When , as (superstar revenue − unknown artist revenue) → 0.

Proof. As the negligible sampling cost allows a consumer to sample and determine the true value of a music item from an unknown artist and a superstar before purchase. Since consumer surplus is maximized by either music item, ceteris paribus, and hence the difference in sales between a superstar’s music and that of an unknown artist is negligible. QED

An immediate observation is that current price premiums on superstar music (Strobl and Tucker 2000) will begin to decrease as sampling costs decrease. As an illustration of propositions 6 and 7, we consider the extreme case of a “perfect” superstar, where the consumer variance in expectations is zero.14 Such a superstar gets wide exposure in the press and significant playtime on radio and television. In essence, consumers do not have to expend additional individual effort to learn and determine the value offered by the star’s music. By employing the methodology detailed before, the consumer decision‐making strategy is as follows:

This is illustrated in figure 3. Note that a consumer will never sample and subsequently buy a perfect superstar’s music item, since she is aware of the true value a priori. Thus, downloading the music of superstars is more likely to result in piracy than the downloading of music from relatively unknown artists. This offers a possible explanation for the high concern expressed by some popular artists (e.g., rock group Metallica and rapper Dr. Dre) regarding the online sharing of music via mechanisms such as Napster (Jenkins 2000).

(15 KB)

Fig. 3.— Consumer decisions based on star status (a priori expectation of music value).

Open New Window

The above discussion suggests that low sampling costs present a significant disadvantage for current superstars and aid the discovery of equally valued music items by less well known artists. This has potentially far‐reaching consequences in music publishing and advertising and should be accounted for in designing new selling strategies. The next section of this article provides empirical validation of the key analytical results discussed thus far.

IV.  Empirical Methodology

In this section, we discuss the empirical evidence used to examine the key propositions enumerated in the analytical model. The data used to investigate the propositions were drawn from two sources. The first data set consisted of primary data collected via a survey and was used to validate consumer choices under various technological and economic parameter settings. A second data set was developed using the Billboard ranking charts and was used to evaluate the propositions related to the superstar phenomenon and sharing technologies.

A. Intention Survey

The survey instrument asks about consumer attitudes toward online music. The primary data collection was completed in two major phases. An initial pilot study was conducted with a sample of 76 graduate students. Subjects were assured complete anonymity. The measurement scales derived from the analytical model were tested for validity, clarity, and consistency. Based on the feedback, a revised survey questionnaire was administered to 200 respondents.

The survey was targeted toward college students as online music copying and sharing is rampant among students (Fine 2000; Jay 2000). Furthermore, this demographic constitutes a significant portion of the music fan base (Holbrook and Schindler 1989); consequently the analysis on this segment is directly relevant to the music industry. The ages of the respondents ranged from 19 to 54 years with an average of 23; 61% were males, 15% worked full time, and 54% worked part time. Of this sample, 52% reported a very high level of interest in music, while another 37% listened to music regularly. The sample group is sufficiently diverse in terms of demographic, economic, and social aspects.

1.Measures To evaluate the impact of economic and technological factors that influence sampling and piracy, the respondents were asked to reveal their online music experiences, to provide demographic information, and to specify preferences for decisions relating to online music activities under varied settings.

Price and sampling cost.—The price variable indicates the retail price of a music CD and was set at $5 and $15. The sampling cost was measured via Internet connection speeds. A high sampling cost scenario involves respondents searching and downloading music via a 56 kbps modem, which on average can take up to an hour to download a typical song. A low sampling cost scenario involves a fast Internet connection such as a cable modem or DSL, which provide significantly lower download times. Price and sampling cost are coded as categorical variables. The pilot study suggested that other technological factors did not have a significant influence.

Decision choice.—For a given sampling cost and price setting, the respondents were asked to take one of the six actions as described in figure 4. If actions A, B, C, or D are selected, it constitutes sampling. Among these, action D constitutes piracy, actions A or C constitute sample and buy, and action B constitutes sample and discard. The traditional direct buy option is captured from the action E.

(49 KB)

Fig. 4.— Illustration of intention survey choices

Open New Window

Value uncertainty regarding the music offering is classifed as follows. If a respondent has listened to a particular piece of music before, its value is assumed to be “known” to her; otherwise, it is “unknown.” A user is assumed to be uncertain about the true value of “unknown” music. The respondents were presented with a total of five types of music choices as depicted in table 1.

Table 1 Choices of Music

Open New Window

2.Hypotheses The propositions developed in the analytical model are tested using the empirical evidence. Table 2 summarizes the hypotheses tested, and the corresponding propositions. The results of the testing of the individual hypotheses validate the comparable propositions, which in turn validate consumer choices under different technological and economic environments.

Table 2 Hypotheses

Open New Window

3.Data Analysis To test hypotheses 1, 2a, 2b, and 3, we estimated a multinomial logit model where Decision = f(price, sampling cost, age, gender, income) for all unknown music choices.15 To test hypothesis 4, we estimated a multinomial logit model, Decision = f(music value, age, gender, income), for the two types of known music choices.

In these models, price, sampling cost, gender, and music value are categorical variables, with the low values coded as 0 and the high values coded as 1. For gender, males are coded as 0. Part A of table 3 presents the results for the unknown music choices, and part B of table 3 presents those for known music choices.

Table 3 Multinomial Logit Model of Unknown Music Choices

Open New Window

For both models, income has little or no significant effect. The lack of income effect is perhaps due to the low cost of a typical music item. In contrast, software, which is much more expensive, exhibits significant income effect (Gopal and Sanders 2000). Females generally have a negative correlation with overall sampling decisions (decisions 1–4), especially with piracy behavior (decision 4), but show a positive effect on sample and buy decisions for unknown music (pt. A of table 3). This suggests that females primarily sample unknown music with an intention to buy.

The results from unknown music choices (table 3, pt. A) show that price has a statistically significant negative correlation with music sampling and buying (decisions 1 and 3). This validates hypothesis 1. It also suggests that sampling costs are negatively correlated with the buying intention of unknown music. For unknown music, sampling cost has a significant negative effect on sampling behavior (decisions 1–4), which validates hypothesis 2a. Sampling costs also have a negative correlation with the decision to sample and buy (decisions 1 and 3), signifying that a decrease in sampling costs leads to increased sampling and buying of unknown music. Hence, hypothesis 2b is valid. Further, market price has a significant negative correlation to a consumer’s direct buy decision (decision 5) for unknown music. This validates hypothesis 3 and also suggests that sampling costs positively affect direct buying.

Results for known music (table 3, pt. B) show that the decision to sample and buy (decisions 1 and 3) is positively related to the music value. This validates hypothesis 4.

B. Superstar Analysis

To test actual consumer buying behavior based on information on a particular music item, linear regression analysis was used to empirically evaluate the Billboard rankings charts archival data. This data set consists of weekly rankings of the published Billboard Top 200 album charts. The rankings are calculated from the total retail sales in the United States for the previous week and are usually published on Saturdays. Hence, these rankings directly reflect actual sales of albums. Billboard rankings are widely used as a marketing tool by the music industry to denote the success of an album—a high ranking suggests a high value of a particular music item to potential customers.

In an earlier study of similar data, Strobl and Tucker (2000) used a sampling technique where they included only those data where artists’ first names began with the letters A–D. Their analysis included a simple count of the number of times an artist appeared in the Top 30 rankings and did not employ a weighted ranking of albums to account for the impact of actual rankings on future sales.16

Sample data for our analysis were collected for the top 200 album rankings for each week for the most recent 10‐year period (1991–2000), resulting in a total of 104,400 observation points. During this period, a total of 2,174 unique artists and 5,355 albums appeared on the charts.17

An artist’s current ranking on the charts is seen to be highly correlated with past rankings. Table 4 presents correlation coefficients of an artist’s chart rankings between pairs of consecutive months, for three sample years.18 All other years (for which data were available) show similar patterns.

Table 4 Bivariate Correlation Coefficients of Artist Rankings on Billboard Top 200 Album Chart

Open New Window

The number of unique artists on the charts (fig. 5) has shown some changes with the introduction of new technologies such as the graphical Web browser (1993), widely available MP3 playback software (1997), and peer‐to‐peer (P2P) file sharing software (1999). These technologies represent watershed events, since the browser made Internet surfing easier for all, created online fan clubs, and lowered sampling costs; MP3 players spurred the conversion of digital music files into smaller MP3 format files; and P2P software blossomed by enabling the sharing of such files, further lowering sampling costs. The effect of these technologies on the consumer buying behavior of music is investigated next.

(33 KB)

Fig. 5.— New technologies and unique artists on Billboard charts (per year and per quarter).

Open New Window

1.Measures To evaluate the impact of an artist’s past reputation on the rankings of their new album, we obtained some information from an artist’s ranking on the charts.

Past reputation.—An artist’s past popularity and reputation is captured by the total amount of time spent by each artist on the charts in the years preceding the release of a new album that is being considered (Strobl and Tucker 2000). This total is weighted to control for the actual chart position that the artist occupied, which also minimizes the multicollinearity effect of an artist’s chart rankings on subsequent calculations.

Debut rank.—This is the rank that a new album by an artist (who has already been on the charts before) achieves on its first appearance on the charts. It measures the initial acceptance of an album by consumers. This ranking is affected by the related artist’s past reputation as well as by limited information through promotional efforts.19

Highest rank.—This is the top rank that a new album attains on the charts (over all subsequent years). This measures the best rank the album achieves over its lifetime on the charts, after consumers have had a chance to gain adequate information on it through various methods.

Chart hit.—This is a binary measure that captures if an artist (of a new album on the charts) has appeared previously on the charts, irrespective of rank. Chart hit is measured as “4‐year reputation” (denoted as 1 if artist appears on charts within 4 years before current album debut date), and “maximum reputation” (denoted as 1 if artist ever appears on charts before current album debut date).

Unique artists per year.—The number of unique artists appearing on the rankings each year is calculated from the charts data. The objective here is to measure whether more artists get ranked on the charts in each subsequent year.

Unique albums per year.—This is the number of unique albums appearing in the rankings.

Internet users per year.—This consists of users online (via the Internet) and is closely related to the amount of useful information online (including about music items and samples). As more information about music items becomes available online, more fans utilize it, and hence sampling costs decrease. The rationale is deduced directly from the network effect and suggests that, as the number of Internet users increase, creating positive network externalities, information availability increases and sampling cost for online music decreases. The rankings data reflect album sales in the United States; hence, we use data on the number of Internet users in the United States per year for the same years (1991–2000).

2.Hypotheses Table 5 presents the hypotheses developed and tested, with the corresponding propositions. Testing the hypotheses provides support for the comparable positions.

Table 5 Hypotheses

Open New Window

3.Data Analysis To minimize seasonal effects with the available data set, we chose a 6‐month period, from July to December 1995, to study the effect of past reputation on the rankings of an artist’s new music album (hypotheses 6a and 6b). During this period, 314 albums made their first appearance on the Billboard Top 200 album charts. Of the artists performing in those albums, 165 of them had appeared on the charts in the preceding years (1991–94). It is not useful to study the difference in ranking between an album’s first appearance on the chart and its best rank, since obviously a ranking difference (of five, say) in the top 20 has a significantly different meaning from the same ranking difference in the bottom 20 of the charts. Hence, the rankings themselves are used to study the effect of artist reputation. The regression models tested are and For hypotheses 7a and 7b, the yearly data are used to test the following models: and To test hypothesis 7c, the data are split into 1995–96 and 1998–2000 (independent) samples to account for the cusp year 1997 when new music compression technologies (MP3) were introduced. The following models are tested to identify any difference before and after the introduction of the new technologies: and Given that a significant majority of the albums do not stay on the charts for over 3 years, these models are satisfactorily estimated from the available data. Table 6 presents the results from the various models designed to test the hypotheses.

Table 6 Regression Results for Hypotheses 6a, 6b, 7a, and 7b

Open New Window

The results (table 6) show that an artist’s past reputation has a very significant positive effect on both the debut rank of a new album by that artist and the highest rank achieved by that album. This validates hypotheses 6a and 6b.

The number of unique artists per year (for the years studied) ranged from 463 to 655, while unique albums ranged from 618 to 921. During the same period, the number of Internet users increased from 3 million to 116.7 million.20 We find strong evidence (table 6) that, over the last decade, the number of unique artists and albums that have appeared on the Billboard Top 200 album charts is statistically related to the number of Internet users. The dependent variable variance is also adequately explained (the adjusted R² is very high). This validates hypotheses 7a and 7b. The implication is that, with the lowering of information sampling costs, consumers become aware of more new albums that they like, leading to more artists and albums being ranked on the charts.

The results (table 7) show a significant difference between the reputation coefficient of 1995–96 and 1998–2000, both for debut rank and highest rank. The impact of reputation, derived from appearance on the Billboard charts, decreased from 1995–96 to 1998–2000, based on lower coefficient value. Using methodology similar to testing difference between means, for a two‐tailed significance of rejects the hypothesis that difference between the coefficients from 1995–96 and 1998–2000 is insignificant. The result holds equally well for the 4‐year reputation as well as maximum reputation coefficients, among all ranks overall as well as specific rank “levels” of the Billboard charts. This validates hypothesis 7c, and it clearly shows an appreciable difference in the chart rankings with the advent of music sharing technologies. This implies that online sharing technologies, which have decreased sampling costs further, have considerably eroded the impact of stardom on sales of music offerings.

Table 7 Regression Results for Hypothesis 7C

Open New Window

A closer examination of the number of unique albums that appear on the charts in each year shows some interesting insights (fig. 6). For any given year, there are more unique albums in the lower ranking “levels” than in the higher ones.21 This points to a lower variability among the higher ranked albums compared to low‐ranked ones. This suggests a lower perceived variability in the value of higher ranked albums by consumers compared to lower ranked ones. This implies a higher dominance of the “superstar effect” on the higher ranked albums than on lower ranks, where consumer valuations vary less across a superstar’s album. Moreover, the number of albums on each “level” of ranking shows an increasing trend over time. Similar effects are also distinctly noticeable on the rankings of the number of unique artists per year (fig. 7). The implication is that, as sampling becomes less expensive, the superstar effect is eroded overall, and more users purchase music items based on their actual, not perceived, valuations.

(27 KB)

Fig. 6.— Number of albums in Billboard Top 200 charts, categorized by rank

Open New Window

(25 KB)

Fig. 7.— Number of artists in Billboard Top 200 charts, categorized by rank

Open New Window

V. Discussion and Conclusions

The focus of this research was on the economic dynamics of online digital music sampling and information sharing and its implications for consumers and sellers of such goods. An economic model that incorporates the incentive structure for consumers and sellers has been developed and tested.

The economic analysis provides several interesting insights. From a consumer perspective, it provides an indication of the factors that affect a consumer’s decision to sample music online. We show that decreasing sampling costs not only lead more potential consumers to sample unknown music items but also lead more consumers to buy the music items that they have sampled. This directly follows from the fact that lower sampling costs have a positive effect on the consumer surplus of samplers, which, in turn, has a positive effect on their purchasing intentions. This has major implications for the music industry, in that the industry can potentially reverse the effects of online audio piracy by providing more legal and efficient sampling techniques that consumers could use. This is contrary to the anecdotal belief that online availability of digital music leads only to a drain on profitability. This efficient sampling may be offered in the form of easily searchable indexes of music items, fast download access to music items in different secure formats, provisions of posting consumer reviews on items, creation of fan club sites within the search portal, and so forth. Some of these items are now being made available on online music portals—however, there is little evidence of an integrated offering of such strategies.

The effect on sales depends on the true intrinsic value of the music item, ceteris paribus. The impact of music availability online has a differential impact based on the realized value of the music to the consumers. For higher valued songs, online search and sampling capabilities have a beneficial impact on sales. Lower valued music items are pirated more than higher valued items, ceteris paribus, and consequently sales of those suffer. If a producer is aware of the true value of a song to consumers, he can set the price accordingly to maximize profits. For producers, the model shows that, in the presence of online music sampling, uniform pricing for all music items is a suboptimal strategy. The key challenge is to obtain priors on this realized value, so that differential pricing schemes can be effectively implemented based on music valuations. Techniques to obtain this critical information and to derive appropriate pricing schemes are critically important and viable topics for future research.

Our empirical evaluation provides strong support for the hypotheses that the existing superstar phenomenon in the music business is positively aided by high sampling costs and that this superstar status is threatened by the advent of online music sampling services. Superstars come under increasing threat from two fronts: (a) a greater proportion of sampling of superstar music leads to piracy—users who sample do so with an increased intention to pirate, and (b) decreasing sampling costs lead to an erosion of superstardom. However, there is a greater probability of discovering other high quality music items by lesser known artists with the new technology, which will hurt a superstar’s sales—and, hence, status. Hence, online music sharing technologies tend to threaten some superstars and favor other lesser known artists, ceteris paribus, and it is understandable that some superstars would have reservations regarding it. This has created a schism in the artistic community. There is anecdotal evidence that some superstars, like the heavy metal group Metallica and rapper Dr. Dre, have opposed such sampling technology (Gomes and Mathews 2001). Others, including Don Henley,22 Alanis Morisette, and other lesser known artists and groups, have supported it, including in United States Judiciary subcommittee hearings (Welte 2001).

The economic implications of this analysis are generally applicable to other similar experience goods. A subsequent critical issue for producers and distributors of such goods is to estimate market demand and set differentiated prices to maximize returns. Additional research needs to be conducted to derive enhanced pricing models for such goods that incorporate individual consumer valuations as well as other marketing models that utilize consumer attitudes toward such goods. The enormous level of monetary resources at stake demands further investigation into newer models that maximize the value of digitized intellectual property.

Online music technologies are fundamentally altering the landscape of the music business. While consumers clearly stand to gain from these opportunities, the music industry can also reap significant benefits via effective strategies. Music as an artistic expression transcends economics and the bottom‐line revenues of the music industry. Indeed, as stated by Van Morrison, “Music is spiritual; the music business is not” (http://www.aria‐database.com/cgi‐bin/listgen.pl?quotes). Nevertheless, fundamentally sound business models are critical for enabling the social, artistic, and spiritual dimensions of music to flourish.