We are intrigued by the debate our study has stimulated in various blog-postings on the internet.
Before responding to the debate, we would like to emphasise that, when we entered this research project, we had no preconceived views on what effect downloading of free music may have on music pay-markets.
However, Professor Stan Liebowitz who perhaps is the most prolific advocator against our results (his web-page on IP/File-sharing lists his views), writes in a critique in relation to another paper by Oberholzer-Gee and Strumpf (2007):
"O/S are aware that their conclusion-that file-sharing causes no reduction in record sales-contradicts conventional wisdom. [...] Nevertheless, their results leave an extremely large elephant in the room-if not for file-sharing, why have record sales declined so precipitously?" (Liebowitz, 2007:1)
But, what if we wrongly blame downloaders of free music for declining sales? Then management and policy aimed solely at addressing file-sharing can only fail.
We are in the midst of a technological revolution. New ways in which music is created, delivered and consumed, affect both market and industry structures. There are many different stakeholders affected by this change, including music publishers (the big 4 and the independent), artists and consumers. One should not assume that existing industry structures are also the most optimal in the new technological paradigm. The growth of the online music industry is creating new business opportunities. For example, the responses indicate that free downloads are motivated, among other things, by consumers looking for music that is not available in stores, and that they are not necessarily interested in purchasing the album version but prefer a single file or a digital file. This illustrates a possible opportunity for new artists to enter music pay markets and a change in the way music users prefer music to be delivered and consumed.
We believe that everyone can agree that the better and more diverse data we examine, the closer we come to uncover true relationships. Different data and methods have their unique strengths and weaknesses. If we want to understand the performance of industries and product markets, then we need to understand the strategies of firms who make up these industries and the behaviour of the people who are the consumers. Micro-data is best suited to bring us closer towards that understanding.
Macro data can describe a situation or some relationships at the aggregate level, but such data are limited when it comes to shedding light on explanations of the situation. For this we need micro-data, as we need to understand how the micro behaviour underpin the creation of the situation at the aggregate level. For example, an overall high level of unemployment may seem like an economy in crises at the macro level. However, at the micro level we might find that some declining sectors have huge unemployment and some growing sectors are in need of workers, and this could be due to a very vibrant and dynamic economy with fast changing industry structures.
This type of argument related to conclusions being drawn from macro data, is also the long standing fundamental argument led by evolutionary economists when debating the use and misuse of data and statistics, and it was frequently raised in the methodology debate regarding the proper use of patent statistics.
Therefore, we believe that Liebowitz's analysis, which is based on macro-data, has a weak empirical underpinning, and as a result, the following is probably wrong.
"I have written a recent paper (forthcoming in Management Science) that examines record sales and Internet uses in 99 US cities to measure the impact of file-sharing. While I am partial to my own work, I believe this paper provides the strongest analysis to date of these issues. The methodology avoids many empirical difficulties found in other papers. It concludes that file-sharing is responsible for the entire decline in record sales that has occurred, and that except for file-sharing there would have been an increase in sales since 1999 instead of the strong decline." (Liebowitz, 2007)
His core argument is based on city-level data: record sales measured for an entire city and Internet penetration in a city (used as a proxy for file sharing). Liebowitz writes that, unfortunately, there is no direct measure of P2P music file-sharing. He recognises the weaknesses of his proxy, but these are still the data which underpin his strong conclusion:
"... file-sharing [or internet penetration in his case] appears to have caused the entire decline in record sales and appears to have vitiated what otherwise would have been growth in the industry". (Liebowitz, 2007)
Our own study relies on micro-data obtained by asking individuals to what extent they are engaged in downloading and/or purchasing of music, and about their incentives behind such behaviours. While this may not be without its problems, it clearly has its merits as discussed above if we want to understand the behaviour of people on whom music markets depend on.
The survey provides information on a wide range of variables including:
Downloading from free music markets:
Downloading free music from P2P file-sharing networks, like Kazaa, LimeWire, eDonkey, BearShare or Gnutella
Ripping CDs and copying them onto computers
Downloading free music from promotional website
Downloading music from peoples' private Internet websites
Copying MP3 from friends
Number of CDs purchased by respondents
Number of MP3s purchased (i.e. buying music tracks from online pay-sites like iTunes or Archambault)
Percentage of P2P downloads to ‘hear before buying'
Percentage of P2P downloads due to wishing to not to buy ‘whole album'
Percentage of P2P downloads due to ‘album too expensive'
Percentage of P2P downloads due to music ‘not elsewhere available'
There are many other aspects to the survey, such as several demographic factors, MP3 ownership, supplementary entertainment goods and more.
Liebowitz's key criticism is that we examine a sub-sample of file-sharers; i.e. only people who declared that they engaged in P2P file-sharing. He states:
"The single biggest problem with the report, however, is that the authors present two sets of results, one for the entire sample and one just for downloaders, and it is only the former results that should be considered. It makes little or no sense to look only at downloaders." (Liebowitz, 2007)
He continues with an example to illustrate his point:
"When we compare the efficacy of a drug, we compare those who take the drug with those that do not." (Liebowitz, 2007)
However, there is a key difference between data based on drug trial experiments and this data. In drug trials people are randomly allocated to those who receive the drug and those who receive a placebo. This is not the case in our dataset. Here, belonging to the group of file-sharers and belonging to the group of non-file-sharers is not a random event. It is a deliberate choice made by each individual to either engage in P2P file-sharing or not to engage in P2P file-sharing. Being a file-sharer is an endogenous decision, just like taking illegal drugs. We expect illegal drug takers and file-sharers to behave different from the rest of the population, and we therefore model the differences in their behaviour.
We now move onto a point Liebowitz made in his first posting, and which he revised in his second posting several days later. In his first posting Liebowitz claimed that we do not control for ‘music interest' in our study. This is still reflected in Internet blogs when contributors state that our finding (a positive association between the number of P2P downloads and CD album purchases for the sub-sample of downloaders) is trivial and can be explained away by the fact that people with a high interest in music, do both: purchase CD albums and download P2P files.
In Liebowitz's second version of his critique he now reconizes that we are controlling for music interest (albeit imperfectly in his view, which means something very different). Nonetheless, the numerical example given by Liebowitz, and still included in his website, is based on the assumption that no controls for music interest are made and therefore is hugely misleading.
Given the 0.44 estimate has been a subject of debate, there is something we believe needs clarification. On page 27 the paper suggests that for an increase in the average number of P2P downloads of 1, the number of CD purchases per year will increase by 1.212/2.718282 or 0.44. This is a point estimate. We compute marginal effects at the means of the regressors, and, thus, the point estimate refers to those people who claim to download approximately 13 P2P files in an average month and not to downloaders who claim more or less downloads in an average months. This is because the assumed relationship is non-linear (negative binomial distribution plus log transformation of the number of P2P downloads). Many arguments against the paper (including the one of Liebowitz) seem to be based upon the assumption that we work from a linear relationship.
Some blogs (resembling Liebowiz's first critique) discuss if our paper, which suggests a positive association between P2P downloading and CD sales for the sample of P2P downloaders, is not valid because we have seen a decline in CD sales. Hence, they find it a weakness that we cannot identify why CD sales have dropped. We can of course speculate why CD sales have declined (and as most other writing in the blogs we have some personal views on this), but this is not the aim of this study.