The economist Jordi McKenzie of the University of Sydney published the first study on the impact of music file sharing on music sales (physical and digital) in Australia. His article in the Australian Economic Papers entitled “Illegal Music Downloading and Its Impact on Legitimate Sales: Australian Empirical Evidence” is based on a working paper from August 2009 and was published in December 2009.
With a similar methodological approach to Oberholzer-Gee and Strumpf (2007) he came to the conclusion that “(…) the evidence suggests no discernible impact of dowloading activity on legitimate sales“. More details on his approach and his findings are given here:
Despite repeated enquiries by the author, the Australian Recording Industry Association (ARIA) was unwilling or unable to provide weekly statistics on physical and digital music sales. Thus, McKenzie had to use weekly Top 40 (digital) and Top 50 (physical) singles chart rankings by ARIA as a proxy for music sales. The data set covers a 15-week period from 5 November 2007 to 11 Feburary 2008, including the Christmas/New Year period where retail sales as well as P2P filesharing activity are at their highest. In addition, information on the ranking of the previous week, the number of times a song has entered the charts, and the highest ranking attained to date were included in the data set. The full sample, therefore, consists of 1 350 observations (600 digital and 750 physical) with 91 unique songs in the digital sample and 104 in the physical one. The data was sampled meticulously each Sunday when ARIA released the weekly sales charts. On the following Monday, between 7-10 pm when Internet usage is highest, each song was cross referenced on the file sharing network Limewire, which represents 60% of all music file sharing activity in Australia. This network was chosen to give a proxy for the number of downloads, which cannot directly be measured, occuring for a song. In addition, information relating to the file size, download speed, and bit-rate was collected.
In the regression model the weekly chart rank of a unique song is estimated as the dependent variable on the number of times a song had appeared in the charts (measured in weeks) and the number of file sharers who made the song available for download a week before. Rank is, therefore, transformed into log form to capture the fact that a drop from rank 40 to 50 is not as significant than from rank 1 to 10. The variable ‘number of times in the charts’ is also included in quadratic form in order to account for the ‘aging process’ (life cycle) of a song in the charts. Also a fixed effect, which captures song heterogenity as well as 14 weekly dummy variables to account for weekly fixed effects across the sampling period are included in the model. Finally, an unobserved idiosyncratic error is captured by a dummy variable.
McKenzie ran a two-step regression model. In the first step no proxy and dummy variables were considered, but just the single relationship between chart position and the number of file sharers offering a song for download was measured. The authors found, as expected, a statistically negative impact of file sharing on digital music sales, but suprisingly not on the physical sales. McKenzie, therefore, concluded that there is some degree of market segmentation between the digital and physical music markets. The segmentation derives, as the author suggests “(…) from the new (most younger) computer savy generation, as compared to the more traditional consumer type” (p. 306).
Since the simple regression model does not account for an unobserveable effect on both, music sales and downloading (e.g. popluar songs tend to be purchased as well as ‘illegally’ downloaded at the same time), the author considered an instrumental variable modification and thus he included the additional instruments file-size and bit-rate. In addition, several dummy variables as described above were also integrated in the regression model. As a final result “(…) there is only weak siginificance in the regressions at the ten per cent level for the number of weeks previously spent in the charts. The evidence suggests that – at least using this data – there is no siginificant impact on sales in either market because of P2P activity.” (p. 303). This leads McKenzie to the final conclusion that “[t]his finding may be cautiously interpreted as (illegal) downloading activity having no effect on sales in either the physical or digital market.” (p. 306).