movie piracy on peer-to-peer networks—the case of kazaa

14
Movie piracy on peer-to-peer networks—the case of KaZaA q Marc Fetscherin * Institute of Information Systems, Engehaldenstrasse 8, 3012 Bern, Switzerland Received 23 June 2004; accepted 23 June 2004 Abstract Content providers from the movie industry argue that peer-to-peer (P2P) networks such as KaZaA, Morpheus, or Audiogalaxy are an enormous threat to their business. They blame these networks for their recent decline in sales figures. However, this argument would only apply if consumers can access high quality copies easily and quickly on these networks. This paper presents a simple model outlining the consumerÕs tradeoff between downloading the movie legally or acquiring it illegally through copying from such networks. The model shows that there are mainly two factors affecting consumer demand for copies: the probability of get- ting high quality copies and the risk associated with copying. The paper goes on to empirically 0736-5853/$ - see front matter Ó 2004 Elsevier Ltd. All rights reserved. doi:10.1016/j.tele.2004.06.005 q Legal notice: We do not encourage individuals to share copyrighted material on peer-to-peer (P2P) networks neither do we support the illegal sharing of copyrighted material on such file-sharing systems. In addition, we respect and support international copyright agreements and regulations (e.g., Berne Convention, TRIPS) as well as international (e.g., EC Directives) and local laws (e.g., DMCA, the Swiss URG). The goal of this study was to evaluate the quality and quantity of illegal movie files available on P2P networks and to what extent this poses an economical threat for the movie industry. In that respect, we had to download copyrighted material in order to reach our research goals. However, we emphasize that we did not at any time either (re) distribute these files nor did we provide them for uploading during the entire study. Finally, after having completed this study, all downloaded files have been deleted permanently from our systems. In this study we do not show how to bypass protection technologies or to do any copyright infringement, neither do we provide any circumvention technology, system, or solution, nor how to ‘‘trade’’ with illegal copies. The purpose was purely academic. * Corresponding author. Tel.: +41 31 631 47 83; fax: +41 31 631 46 82. E-mail address: [email protected]. URL: http//:www.ie.iwi.unibe.ch. Telematics and Informatics 22 (2005) 57–70 www.elsevier.com/locate/tele

Upload: marc-fetscherin

Post on 26-Jun-2016

213 views

Category:

Documents


1 download

TRANSCRIPT

Telematics and Informatics 22 (2005) 57–70

www.elsevier.com/locate/tele

Movie piracy on peer-to-peer networks—thecase of KaZaAq

Marc Fetscherin *

Institute of Information Systems, Engehaldenstrasse 8, 3012 Bern, Switzerland

Received 23 June 2004; accepted 23 June 2004

Abstract

Content providers from the movie industry argue that peer-to-peer (P2P) networks such as

KaZaA, Morpheus, or Audiogalaxy are an enormous threat to their business. They blame

these networks for their recent decline in sales figures. However, this argument would only

apply if consumers can access high quality copies easily and quickly on these networks. This

paper presents a simple model outlining the consumer�s tradeoff between downloading the

movie legally or acquiring it illegally through copying from such networks. The model shows

that there are mainly two factors affecting consumer demand for copies: the probability of get-

ting high quality copies and the risk associated with copying. The paper goes on to empirically

0736-5853/$ - see front matter � 2004 Elsevier Ltd. All rights reserved.

doi:10.1016/j.tele.2004.06.005

q Legal notice: We do not encourage individuals to share copyrighted material on peer-to-peer (P2P)

networks neither do we support the illegal sharing of copyrighted material on such file-sharing systems. In

addition, we respect and support international copyright agreements and regulations (e.g., Berne

Convention, TRIPS) as well as international (e.g., EC Directives) and local laws (e.g., DMCA, the Swiss

URG). The goal of this study was to evaluate the quality and quantity of illegal movie files available on

P2P networks and to what extent this poses an economical threat for the movie industry. In that respect,

we had to download copyrighted material in order to reach our research goals. However, we emphasize

that we did not at any time either (re) distribute these files nor did we provide them for uploading during

the entire study. Finally, after having completed this study, all downloaded files have been deleted

permanently from our systems. In this study we do not show how to bypass protection technologies or to

do any copyright infringement, neither do we provide any circumvention technology, system, or solution,

nor how to ‘‘trade’’ with illegal copies. The purpose was purely academic.* Corresponding author. Tel.: +41 31 631 47 83; fax: +41 31 631 46 82.

E-mail address: [email protected].

URL: http//:www.ie.iwi.unibe.ch.

58 M. Fetscherin / Telematics and Informatics 22 (2005) 57–70

estimate the probability of getting such high quality copies. To date, our results show that

there is a very low probability of obtaining high quality movies. We further tested our model

by conducting simulation analyses to better understand current and future consumer behavior.

The first simulation shows that, with the current low probability of getting high quality copies,

the majority of consumers would prefer to download movies legally. The results of the second

simulation focusing on the future behavior of consumers showing that the most important fac-

tors are: the risk of being caught, the perceived value of the original, the availability (proba-

bility) of high quality copies and the price of the original. Our research findings not only offer

insight into the consumer behavior, but also to what extent the movie industry faces piracy.

These results may serve as a new impulse for copyright holders and policy makers, allowing

them to make appropriate decisions and take targeted actions to fight piracy.

� 2004 Elsevier Ltd. All rights reserved.

Keywords: Peer-to-Peer; Piracy; Digital Music; Downloading

1. Introduction

Piracy of digital content, especially Internet piracy, has increased significantly in

the last few years. Technology advances in hardware and software enable individuals

to virtually capture, store, copy, and modify any kind of digital content. By using

highly popular file sharing systems such as KaZaA, Morpheus, or Audiogalaxy, itis possible for anyone to share any type of digital content such as games, software,

music files or movies on a mass scale. This illegal file-sharing of copyrighted content

poses an enormous threat to content providers who are the only ones authorized to

distribute and sell their products.

A number of papers have explored various economic mechanisms and effects in-

volved in piracy. From social and economic perspectives, piracy may increase or de-

crease social welfare (Holm, 2000). Those arguing piracy has a negative impact are

content providers, who claim that piracy costs them several billion USD a year(SIIA, 2001; IFPI, 2002). For brevity, we will omit the term illegal or pirate and refer

simply to copy in this paper. Takeyama (1994) argues that expectations of a future

illegal market for copies may also create expectations of future price cuts on the orig-

inal. She illustrates that such expectations may cause immediate reductions in de-

mand and profit for the seller of the original. Hui and Png (2002) argue that the

demand for music CDs decreases with piracy, suggesting that ‘‘theft’’ outweighed

the ‘‘positive’’ effects of piracy. Piracy may also undermine the supply side of orig-

inals since content providers may be left unpaid (Holm, 2000).On the other hand, there are those who argue that piracy has a positive impact by

stimulating the demand side. Holm (2000) and Alvisi et al. (2002) argue copying re-

duces monopoly prices and increases the quantity supplied. The main idea is that the

introduction of lower quality products may be a way through which the monopolist

manages to capture some consumers that would otherwise prefer to copy the prod-

uct. Orsorio (2002) shows that in developing markets, companies stand to make

more money if they view each new illegal user as one more mouthpiece for the soft-

M. Fetscherin / Telematics and Informatics 22 (2005) 57–70 59

ware and one less customer for the competition. Liebowitz (2002) presents two pos-

sible means by which peer-to-peer copying would not harm the revenues of copyright

holders—indirect appropriability and exposure effects.

Although there have been theoretical efforts to understand the economic conse-

quences of copying, it is difficult to find studies that empirically estimates the param-eters judged to be important in markets exposed to piracy. This paper presents a

simple model of consumers� tradeoff between buying and copying. The paper goes

on to measure the quantity and quality of movie copies available on such networks

by taking the most known and used file-sharing system KaZaA as a proxy.

This paper is structured as follows. Section 2 presents a simple model which takes

into account the consumer purchasing behavior of digital content with the presence

of copies. The research method is described in Section 3 whereas Section 4 presents

estimates for the variables dicussed in earlier. Section 5 conjoins on one hand theestablished model and on the other hand the empirical results by undertaking simu-

lation analyses. Two simulations are presented, one for the current consumer behav-

ior, the second for the future consumer behavior. Finally, this paper provides a

discussion and conclusion in Section 6.

2. The model

For simplicity, we will compare the cases where a consumer can either legally

download a movie or he can get a copy from peer-to-peer networks. Thus, individ-

uals have two possibilities to acquire digital content, buy it legally or copy it. The

following model outlines the tradeoff an individual faces. This model excludes the

possibility that an individual does not want the content.

Suppose digital content is produced and there are individuals who desire to con-

sume it. Assume that each individual wants either zero or one unit of the content and

that he makes independent decisions between the legal and illegal acquisition. Let�sdenote the set of Individuals by I and the valuation, or the perceived value of the

content, for a given individual (i) as vi where a uniform distribution is assumed (Pod-

dar, 2002). Let�s also denote the perceived value that the individual places on acquir-

ing the original, as opposed to a copy by voi (Holm, 2000; Cheng et al., 1997; Hui

and Png, 2002). When downloading content, he incurs acquisition costs which can

be broken down into three parts. First, the access costs to be paid to a telecommu-

nication company and/or the Internet Service Provider. Second, he incurs media

storage costs (e.g., hard disk) and third, pay a price for the original. It is assumedthat the access and storage costs are equal for both legal and illegal acquisitions,

and therefore are omitted in the model. Thus, the cost an individual has to

pay for the legal download is the price (Fetscherin, 2003) for the original denoted

by p.

Let us analyze each of these two possibilities in detail. If an individual value the

legal product with voi, he will receive a net benefit (Chen and Iwan, 1999) of:

voi � p ð1Þ

60 M. Fetscherin / Telematics and Informatics 22 (2005) 57–70

The alternative to the model above is for the individual to copy the digital content.

We assume that individuals would only substitute the original version with a copy, if,

and only if, he can easily and quickly find a high quality copy without any risk. Thus,

there are two main factors in our model that affect the demand for copies, the avail-

ability and the risk associated when copying.The availability can be estimated by the fraction of high quality copies available

on such networks compared to the total number of copies available. In other words,

what is the probability of getting a high quality copy? In our model, this is denoted

with the parameter a 2 [1, 0] where 1 indicates availability is 100% for example. It

depends on various factors such as the network topology (centralized versus decen-

tralize), the size of the network in terms of number of users and files available, and

specifically the sharing behavior of users on that network. These factors influence

also the efficiency and effectiveness of the network in terms of search duration anddownloading performance. For example, if the network has a centralized index ser-

ver and there are millions of users connected also willing to share files, the searching

and downloading process will be much faster than on decentralized networks with

fewer peers. This is especially true when the client software enables multi-sourcing,

which means that the same file can be downloaded from different peers simultane-

ously. We agree with Holm (2000) who argued that long search and downloading

times could, in the long term, have a negative impact on the success and usability

of such systems as being the source of copies.We assume that each individual incurs copying costs when making a copy of the

content. These costs are comprised in our model of three components: time to copy,

effort to copy, and the risk of being caught. Time and effort to copy depend on the

strength of copy protection technologies such as Digital Rights Management Sys-

tems (DRMS) in which the risk of being caught depends on the degree of law

enforcement. For simplicity, it is assumed that copy protection technologies do

not restrict individuals in their usage of the legal download digital content and that

the copy does not include any protection technology, otherwise there would not besuch copy. Therefore, these two components are not taken further into account in

our model. The second variable presented in the model is, therefore, the risk associ-

ated when copying. Thus, when individuals are copying files from file-sharing sys-

tems they risk being prosecuted for copyright violation. This is especially an

important factor considering that several thousand people have already been accused

and prosecuted for copyright infringement (Borland, 2004). In the case where an

individual provides copies or is copying, there is a certain probability u 1 that he will

be caught (Hui and Png, 2002; Chiang, 2003; Chellappa and Shivendu, 2002) and f alegal penalty assessed when the theft is detected (Chen and Iwan, 1999). Where f rep-

resents the monetary amount being paid for either paying a fine or the ransom for a

jail sentence. Let x represent the expected risk, in terms of probability of being

caught and fine to be paid (thus x = uf ). Note that x is dependent not only on Digital

1 According to a study conducted by Becker and Clement (2004) heavy user fear to a higher degree that

they are doing something illegal compared to others.

M. Fetscherin / Telematics and Informatics 22 (2005) 57–70 61

Rights Management Systems which enables to detect piracy, but also on the legal

framework the consumer is situated in. In our model, the net benefit an individual

(i) places on the copy is presented as follows:

voia� x ð2ÞThus, consumers buy the legitimate product under the following two conditions.

First, buying must provide more net benefits than copying (1) > (2) and second, buy-

ing must provide more net benefits than not using the content at all, thus, voi�p > 0.

voi � pP voia� x ð3Þ

p6 voið1� aÞ þ x ð4ÞA consumer will copy the movie under the following two conditions. First, copying

must provide more net benefits than buying (2) > (1). Second, copying must provide

more net benefits than not using the content at all, voi a�x > 0. The necessary con-

dition for consumers to engage in copying is:

pP voið1� aÞ þ x ð5Þ

aP 1� p � xvoi

� �ð6Þ

The main focus of this paper is to estimate the availability of high quality copies on

peer-to-peer networks, expressed by the parameter a in our model, by using the mo-

vie industry as an example. The following sections present the research methods used

and the underlying results in relation to the above mentioned factors.

3. Research methods

Three phases were necessary to collect the appropriate data. Phase one involved

the selection of blockbuster movies, phase two entailed the search and the third

phase the evaluation of the movies.

3.1. Phase one—the selection

Although there are no official sales figures for the rental and sale of movies (e.g.,

cinema entries, VHS/DVD sales) available from content providers, we base our selec-

tion of the chosen movies on information provided by Exhibitor Relations Co, 2

which is one of the oldest and most widely quoted film industry statistical research

firm. Our study is based on four blockbuster movies. The key reason we have se-

lected blockbuster movies is the legal demand for these movies is very high; therefore

we also expect the amount of copies to be very high. The four selected movies were

Lord of the Rings: The Two Towers, Analyze That, Minority Report and XXX (Tri-

2 http://www.exhibitorelations.com/.

62 M. Fetscherin / Telematics and Informatics 22 (2005) 57–70

ple X). We conducted search requests several times a week at approximately the

same time on the selected peer-to-peer network for the above mentioned movies.

The main hardware used for the search was an Intel Pentium 4 CPU personal com-

puter with 1.7 GHz processor, 256 MB RAM. Our maximum downloading speed

was 100 MBytes. We worked with the operating system Windows 2000 Professionaland the Norton anti-virus software for Windows 2000. We used the file-sharing pro-

gram KaZaA, specifically KaZaA Lite ++ instead of KaZaA Media Desktop. Both

use the Fasttrack network, a hybrid network, but KaZaA Lite ++ provides addi-

tional features which provide more powerful search requests. Our approach of select-

ing, searching, and downloading movies should be identical to the approach of an

‘‘average’’ Internet user. Only in that respect viable conclusions can be made.

3.2. Phase two—the search

In order to search for one of the four movies mentioned above, various steps were

undertaken. First, we entered the movie title into the ‘‘Search for’’ field and selected

the ‘‘video’’-option. This allows us to focus our search on video files which are labe-

led with the requested movie title. An example of the search mask and the search re-

sults is provided in Fig. 1. Second, in the ‘‘more search options’’ menu we specified

searching for files with at least 100 Mbytes. We picked a minimum of 100 Mbytes, as

smaller files were either not likely to be a complete movie or parts of a movie withreduced audio/video quality to attract users. By pressing the ‘‘search now’’ button,

Fig. 1. Screenshot search results.

M. Fetscherin / Telematics and Informatics 22 (2005) 57–70 63

the search process was executed. For each search request various information was

collected.

3.3. Phase three—the evaluation

After the selection and search phase, we commenced the evaluation phase. Three

additional steps were required to evaluate the quantity and quality of movies shared

on peer-to-peer networks. First we filtered out all obvious fakes. Under the term

‘‘obvious’’ we understand that even the ‘‘average’’ Internet user would not consider

these files for downloading. Therefore we did not either. The selection criteria in-

cluded instances where the filename and title name did not match (excluding typos).

Examples included trailers of a movie that carried the same filename but the word

‘‘trailer’’ was added to the title or vice-versa. Often, a movie was split into two ormore parts (i.e., split files) for easier downloading for individuals without permanent

connections to the Internet and/or smaller bandwidths. For our analysis, we were

forced to download all parts (i.e., split files) in order to make a judgment about

the quantity and quality of full copy version shared on peer-to-peer networks. We

define a version as an entire movie, consisting of either one or several split files.

Since we faced download and storage limitations, we could not download each

newly available file for all four blockbuster movies per search request. However, if

we were not able to download the file or a substantial part of it within 48 hoursof connecting to the network, we aborted the downloading process and classified

the file as ‘‘download not possible’’.

Then the downloaded files were identified and registered with their filename, file

size, resolution, length, and file format. Thirdly we checked the playability, the integ-

rity (e.g., is it the movie it is supposed to be), and the quality of the copy. In regards

to playability, we checked whether the copy was playable by conventional media

players (e.g., Windows MediaPlayer version 7, RealPlayer 8) and if not, what tech-

nical problems we encountered. We defined two categories in this case: either theCODEC was broken or the file was blank (e.g., a dummy file without readable con-

tent). We then checked the integrity of the copy. In other words, we checked whether

it was the movie it was supposed to be according to the title of the file. If it was play-

able and was the ‘‘correct’’ movie, we finally evaluated the quality. We classified cop-

ies into high and low quality categories. The criteria used to allocate the copy into

these categories were the quality of audio and video (bit stream). If there was no

audio or the audio was very poor, the movie was classified as a ‘‘poor’’ quality mo-

vie. In addition, if the video outcome was a screener (e.g., film was gathered with adigital camera in the movie theaters), it was also classified as a ‘‘poor’’ quality movie.

In all other cases, the movie was classified as a ‘‘high’’ quality movie.

4. Empirical results

Empirical data was collected from December 2002 to April 2003. We were able to

conduct 120 search requests for the four selected movies.

64 M. Fetscherin / Telematics and Informatics 22 (2005) 57–70

4.1. Quantity of copies

This section presents the empirical results related to the factors affecting the avail-

ability of copies on peer-to-peer networks, specifically the size of the network in

terms of number of users being online and the number of copies available as wellas the sharing behavior of users. In that respect, we tested the hypothesis that on Ka-

ZaA, there are millions of users online making billions of files available. To test this

hypothesis, we needed to undertake an intraday analysis of the corresponding net-

work. On March 2003, we connected to the network KaZaA 35 times in one day

and analyzed the number of users online, the number of files accessible, and the cor-

responding data volume. Our analysis shows that there are between 3 and 4.5 million

users constantly connected to KaZaA capable of accessing between 650 and 900 mil-

lion files, depending on the time of day.We did the appropriate hypothesis testing, where our hypothesis H0 was l P 2

millions; H1,l < 2 millions. The parameter l represents the average number of users

online on KaZaA at a specific time. The t-score corresponding to the significance le-

vel 0.05 is 1.645; for the data analyzed we obtained a t-value of 18.673. Thus we ac-

cept the hypothesis.

For the second hypothesis, we tested: H0,l P 2 billions; H1,l < 2 billions. The

parameter l represents the average number of files accessible on KaZaA at a specific

moment. The t-score corresponding to the significance level 0.05 is 1.645; for the dataanalyzed we obtained a t-value of �86.682. Thus we reject this hypothesis.

In order to measure the sharing behavior of users, we expressed the null hypoth-

esis as: H0, r 6 0.5; H1, r > 0.5. The parameter r represents the correlation between

the number of users online and the number of files accessible. The t-score corre-

sponding to the significance level 0.05 is 1.645. We arrive at a correlation coefficient

r = 0.995 between the number of users online (USERS_ON) and the number of files

accessible (FILES_ON) as shown in Fig. 2. This value is statistically significant with

3200000 3600000 4000000 4400000USERS_ON

600000000

700000000

800000000

900000000

FILES_ON

files_on = 197665604.30 + 148.77 * users_onR-Square = 0.99

Fig. 2. Relationship between number of users and files found.

M. Fetscherin / Telematics and Informatics 22 (2005) 57–70 65

a = 0.05. Thus we reject the null hypothesis and conclude that users seem to actively

participate in file sharing. By conducting a linear regression analysis, we observed

that with the exogenous variable USERS_ON we arrive to explain 99% of the var-

iance of the endogenous variable files accessible (FILES_ON). We therefore con-

clude that the more people are online, the more files are accessible. However, itshould be noted that not every file means a new movie is available on the network.

Our analysis has shown that there are on average 75 files found per search request,

corresponding on average to 24 different versions of the same movie available at a

specific time on the file-sharing system. Thus, when new peers enter the network,

they either provide a new copy, an additional copy of an ‘‘existing’’ version already

on the file-sharing system. If it were a new copy, this would increase the probability

of getting a movie. On the contrary, if the peer provides an existing version, this

additional copy may not increase the probability of getting a movie but might pos-itively influence the search and downloading process if multisourcing is possible.

4.2. Quality of copies

Three steps were required to calculate the probability of getting high quality mo-

vie copies. The first step of the evaluation process was to filter out the obvious fakes.

Of the 75 files found per search on average, i.e., 24 movie versions, about 4 versions

on average were obvious fakes. In other words, almost 20% of all copies found persearch request are ‘‘obvious’’ fakes. The remaining files have been re-considered for

further evaluation. We tried to download as many as possible and arrived to down-

load 70 versions or sub-versions (i.e., split files) of the four movies. Although we se-

lected them for downloading, this does not mean that the files could be downloaded

at all. Bandwidth constraints, network routing, or other technical constraints from

the downloading as well as from the uploading party can complicate, hamper or even

disrupt the downloading process. Our analysis shows that in 40% of all cases, a

download was not possible. After selecting and downloading the corresponding files,we analyzed whether the downloaded file was playable at all. In 17% of all cases, they

were not playable. In 33% of these cases, it was a CODEC problem, while in 67% of

the cases, the downloaded file was blank. In the second step, we checked the integrity

of the downloaded file and ensured that the movie in question was the one it was sup-

posed to be. Surprisingly, in 16% of all cases, the downloaded file was another movie.

A possible explanation for this could be the intended introduction of fake versions

(in this case not ‘‘obvious’’ fakes) in order to frustrate users downloading movies.

Companies like Overpeer, Vidius, or Netpd are providing such services to Holly-wood studios. In the third step, we evaluated the quality of the remaining movies.

We classified them into high and low quality movies. In 7% of all cases, the movies

were of high quality, whereas in 20% of all cases, the movies were of poor quality.

Table 1 summarizes these results.

We then analyzed the high quality movies from which additional conclusions can

be drawn. The average file size of high quality movie was 610 Mbytes, which, in most

cases, were split files. In other words, since high quality movies are often bigger in

size, they are split into two or more parts in order to be shared more easily on such

Table 1

Evaluation of downloads

Selected movies to download (N = 70) 100%

Download not possible 40%

File does not play

(Playability)

17%

Not correct movie (Integrity) 16%

Copies 27%

Poor quality movies 20%

High quality movies 7%

66 M. Fetscherin / Telematics and Informatics 22 (2005) 57–70

networks. Thus, in order to get a full copy, one needs to download between 1.2 and

1.8 Gbytes of data. While analyzing poor quality movies, the average file size is

around 450 Mbytes, which is significantly lower than that of high quality movies.

Finally, we tested the hypothesis: H0, p P 50%; H1, p < 50%. The parameter p

represents the probability of getting a high quality movie on KaZaA (measured in

the number of high quality movies). The t-score corresponding to the significance le-

vel 0.05 is 1.645; for the data analyzed we obtain a t-value of �1142.204. We there-

fore reject this hypothesis and conclude that there is not a high probability of gettinghigh quality movies on peer-to-peer networks like KaZaA.

5. Simulations

Using the model presented in Section 2 of this paper and the empirical results ob-

tained in the previous section we conduct two types of simulations. First, we wanted

to test current consumer behavior by taking the estimated a as a constant and var-ying all other variables one by one. Second, we performed a simulation analysis by

simultaneously varying two, three, and finally all four variables (u and f taken to-

gether) in order to better understand future consumer behavior and the impact of

the variables a on the consumer behavior. For both analyses, we vary the variables

as shown in Table 2, where each variable varies by a predefined interval.

For the variable perceived value of the original (VOI), we varied it from 0 (no

willingness to pay) to 6 USD with a 10 step interval. As a proxy for the maximum

Table 2

Descriptive statistics

Minimum Maximum Mean Std. Deviation

VOI 0 6 2.70 1.72

P 3 6 4.50 .94

A 0 100 49.84 31.67

U 0 100 17.69 36.01

F 0 1,00,000 26,567 41,348

DECISION 0 1 .58 .491

M. Fetscherin / Telematics and Informatics 22 (2005) 57–70 67

price, we used the average movie ticket price in the US. For the price of a legal down-

load (P), we used the average price of USD 3 3, which represents the minimal price

charged by legal movie providers such as Movielink or CinemaNow 4. We varied the

price from USD 3 up to USD 6, again in 10 step intervals, in order to make it con-

sistent with the perceived value of the original (VOI). To test the availability of highquality copies on peer-to-peer networks (A), we used the estimated 7% for the first

simulation analysis and varied it from 0% to 100% in ten intervals in the second sim-

ulation. For the ‘‘probability of being caught’’ (U), it ranged from a zero probability

up to one hundred. For the first simulation we run it with values from 0 to 100 with

10 step interval. For the second simulation we also used a 10 step interval. Finally,

for the expected fine, although consumers paid fines in the range of USD 2000–3000

(Borland, 2004), we wanted to make it more consistent with current and future laws

and therefore our range was from zero to 10, 100, 1000, 10,000 and 1,00,000 USD.The first analysis consists of 88 simulations [11(VOI) + 11 (P) + 11(U) · 6(F) =

88]. Our results show that in 80 cases, the consumer chose to legally download the

files and in only 8 cases he would prefer the copy. However, this analysis does not

show the effect of the availability on consumer behavior. Thus, we had to conduct

a simulation analysis in which multiple variables were changed, especially a.

In the second analysis, we vary simultaneously two, three or four variables in or-

der to get richer and more diverse data. We conducted combined simulations, such

as VOI and P together [11 · 11 = 121 simulations], and VOI, P and A[121 · 11 = 1331], and a simulation taking all four into account which resulted in

[1331 · (6 · 4 5) = 31944]. Since our dependent variable is the decision whether to

buy or copy it, the most suitable statistical method is the logistic regression. This

is useful for situations in which one wants to be able to predict the presence or ab-

sence of an outcome, based on values of a set of predictor variables (independent).

This is especially true when the dependent variable, in our case the decision to buy or

copy, is dichotomous (nominal). As in OLS regression, a prediction equation for

logistic regression specifies the expected logit as a linear additive function of oneor more independent variables as shown in Eq. (7).

Li ¼ lnpi

1� pi

� �¼ zi ¼ a þ b1X 1 þ b2X 2 þ � � � þ bkX k ð7Þ

Table 3 summarizes the independent variables of the logit regression including their

coefficients, the Wald statistic and their corresponding significance level. The

‘‘Exp(B)’’ column is SPSS�s label for the odds ratio, which can be computed by rais-

ing e to the power of the logit coefficient.

Because the effect of the X�s is nonlinear, the interpretation of parameters is more

difficult than in OLS regression (linear regression). Thus, the effects are nonlinear be-

3 In fact, they are offered most at a price of USD 2.99.4 We limit our study on providers of blockbuster movies.5 Cases where u or f equals 0 is excluded, as this has already been taken into account in the previous

simulations.

Table 3

Variables of Logit Regression

B S.E. Wald df Sig. Exp(B)

Step 1a VOI .22 .011 412.735 1 .000 1.2

P �.29 .020 213.245 1 .000 .74

A �.01 .001 405.474 1 .000 .98

U 2.88 .047 3784 1 .000 17.90

F .00 .000 4070 1 .000 (1.00)b

Constant �.69 .100 48.405 1 .000 .49

a Variable(s) entered on step 1: VOI, P, A, U, F.b As (B) = 0, F will not further take into account in this analysis.

68 M. Fetscherin / Telematics and Informatics 22 (2005) 57–70

tween the independent and dependent variables, but linear in their logs. There are

two ways to interpret the results shown in Table 3. First, we test the individual

parameters of the logit coefficients or the underlying odds ratio (Exp(B)), and sec-

ond, we look at the sign and effect of the independent variable on to the dependent

variable.

All coefficients are significant in the model, with the following having the highest

impact on consumer behavior, in order of importance: the probability being caught

(U), the perceived value of the original (VOI), the availability of high quality copieson peer-to-peer networks (A) and the price of the original (P). Therefore, we arrive at

the following Eq. (8):

Li ¼ lnpi

1� pi

� �¼ zi ¼ �0:69þ 0:22VOI � 0:29P � 0:01Aþ 2:88U ð8Þ

Taking the sign of each logit coefficient, we can conclude that an increase in VOI and

especially U will result in an increase (to some extent) in the willingness to buy the

movie. However, an increase in P and A will reduce (to some extent) the willingness

to pay for a download and consumers tend to copy the movie.

We then tested the goodness of fit of the proposed model, in which we referred to

two tests, the �2 log likelihood value and the Nagelkerke R Square. The �2 log like-lihood was significant and the Nagelkerke R Square was clearly above 0.5, both of

which indicate an overall good fit of the model. In addition, we tested the classifica-

tion performance. We used a confusion matrix or classification table, which provides

information about actual and predicted decisions resulting from a logit regression.

We arrived at a hit rate of 86.6%, which is a good results and again an indicator

of the good fit of the model.

6. Discussion and conclusion

The movie industry argues that they suffer from Internet piracy, especially from

piracy on peer-to-peer networks. In that respect, this paper has presented a simplified

model, taking into account various variables, examining consumer behavior whether

to buy or copy a movie. It has presented two variables which affect the demand for

M. Fetscherin / Telematics and Informatics 22 (2005) 57–70 69

copies: first, the availability of copies in terms of quantity (accessibility) and quality,

secondly, the risk associated when copying with the probability of being caught and

the underlying penalty. This paper then empirically tested the first variable, the avail-

ability of high quality copies on peer-to-peer networks, as it seemed to be the main

argument for the slump of sales by the movie industry. Our results have shown thatthere is a very low probability of getting high quality movie copies. We further tested

our model by conducting two-simulation analyses to better understand current and

future consumer behavior. Our first simulation about the current consumer behavior

shows that the majority of consumers prefer to download movies legally. We there-

fore conclude that file-sharing systems are having a minor effect on the movie indus-

try. By conducting the second simulation about the future consumer behavior, we

show that the most important factor affecting consumer behaviour is the risk of

being caught, followed by the price, the perceived value of the original, the availabil-ity of high quality copies.

Our results have shown that policy makers are better of focusing on the legal

framework permitting the prosecution of pirates while content providers should fo-

cus on marketing in order to increase the perceived value of the legal download. Like

any research, our study has certain limitations despite the statistical tests. First, the

simplified model presented in this paper can be extended by other variables which

might also affect to a certain extend consumer behavior. These variables might be

economic (income), demographic (age, gender), risk (virus), or cultural factors (edu-cation). Second, the purpose of this paper was to assess the quantity and quality of

movie copies on peer-to-peer networks by taking the most known and spread file-

sharing system KaZAa. Further studies should be conducted with other movies

and file-sharing systems in order to fully understand the scope and scale of movie

piracy on such networks.

References

Alvisi, M., Argentesi, E., Carbonara, E., Piracy and Quality Choice in Monopolistic Markets, URL:

http://www.serci.org/2002/Carbonara.pdf[Accessed: 2003-05-25].

Becker, J., Clement, M., 2004. The Economic Rationale of Offering Media Files in Peer-to-Peer Networks.

Proceedings of the 37th Annual Hawaii International Conference on System Science (HICSS), Hawaii.

Borland, J., RIAA files new round of file-swapping suits, URL: http://news.com.com/2100-1027_3-

5201637.html[Accessed: 200-04-28].

Chellappa, R., Shivendu, S., 2002. Economics of Technology Standards: Implications for Offline Movie

Piracy in a Global Context, 36th Annual Hawaii International Conference on System Sciences

(HICSS�03), Big Island, Hawaii 2002.

Chen, Y., Iwan, P., 1999. Software Pricing and Copyright Enforcement: Private Profit vis-a-vis Social

Welfare. In: Proceedings of the Twentieth International Conference on Information Systems,

Charlotte, North Carolina, USA 1999, pp. 119–123.

Cheng, H.K., Sims, R., Teegen, H., 1997. To Purchase or to Pirate Software: An Empirical Study. Journal

of Management Information Systems 13, 49–60.

Chiang, E., Copyright Protection in U.S. Universities: An Overview of Copyright Piracy, Risk Attitudes

towards Copyright Law, and Willingness-to-Pay, 2003.

Fetscherin, M., 2003. Evaluating consumer acceptance for protected digital content. In: Digital Rights

Management––Technological, Economic, Legal andPolitical Aspects. Springer, Berlin et al., pp. 342–362.

70 M. Fetscherin / Telematics and Informatics 22 (2005) 57–70

Holm, H., The Computer Generation�s Willingness to Pay for Originals when Pirates are Present—A CV

study, School of Economics and Management, Lund University, Lund, 2000.

Hui, K.-L., Png, I., Piracy and the Legitimate Demand for Recorded Music, URL: http://

www.comp.nus.edu.sg/~lung/Piracy.pdf[Accessed: 2002-08-10].

IFPI, IFPI Music Piracy Report, URL: http://www.ifpi.org/site-content/library/piracy2002.pdf[Accessed:

2003-08-15].

Liebowitz, S., Policing Pirates in the Networked Age, Cato Institute 2002.

Orsorio, C., A Contribution to the Understanding of Illegal Copying of Software: Empirical and

analytical evidence against conventional wisdom, URL: http://opensource.mit.edu/papers/oso-

rio.pdf[Accessed: 2003-05-25].

Poddar, S., Economics of Software Piracy and It�s Global Impact, URL: http://www.eco.rug.nl/SOM/

SomSemC/Papers2002/16mei.pdf[Accessed: 2002-10-02].

SIIA, Global Software Piracy Report 2000, URL: http://www.siia.net/piracy/pubs/piracy2000.pdf[Ac-

cessed: 2002-02-10].

Takeyama, L.N., 1994. The Welfare Implications of Unauthorized Reproduction of Intellectual Property

in the Presence of Demand Network Externalities. Journal of Industrial Economics 42, 155–166.