(Yet) Another Empirical Analysis of File Sharing

SEPTEMBER 24, 2006

Just came across another paper evaluating the effect of filesharing published earlier this year. Authored by Norbert J. Michel (now of Nicholls State University) and is entitled The Impact of Digital File Sharing on the Music Industry: An Empirical Analysis (Berkley Press’ Topics in Economic Analysis & Policy: Vol. 6: No. 1, Article 18) is available at: http://www.bepress.com/bejeap/topics/vol6/iss1/art18 under a ‘quasi-open-acess’ policy (which so far has at least had the effect of preventing me accessing it).


The first file-sharing software, Napster, was shut down in 2001, but the copying technology’s impact on the music industry is still passionately debated. This paper uses micro-level data from the Consumer Expenditure Survey to examine the impact of Internet file sharing on music sales. Music industry representatives argue that the practice decreases CD sales, while supporters of file-sharing allege the practice could actually increase sales. Using household-level data from the Consumer Expenditure Survey, we find support for the claim that file-sharing has decreased sales.

May have some commonality with the same author’s earlier Digital File Sharing and the Music Industry: Was There A Substitution Effect?, Review of Economic Research on Copyright Issues, 2005 Issue, vol. 2(2), pp. 20-32.


Exact same approach as Hong’s previous work (see summary in http://www.thefactz.org/economics/p2p_summary.html) but less detailed:

  • Hong uses data 1996-2001 (p.31) while Norbert uses data from 1995-2003 (p.4).

  • Identification strategy: Compare CD purchases (available at micro-level from the CEX) between ‘computer owner’ and non-‘computer owners’ (CEX variable again) and attribute differences to file-sharing.

  • Both use a Difference-in-Differences (DiD) approach

  • Hong does quite a bit of extra such as

    • estimating a demand system for entertainment goods (17 ff.) in an effort to account for the affect of the changing prices of other entertainment goods (videos declined in price over the sample period)
    • kernel matching (12 ff.) to deal with possibility of other underlying differences between treatment (owners) and control group (non-owners)
  • Conclusions: Norbert estimates a 13% decline while Hong settled on an 18% decline (or 33% taking the less conservative figure on p.28)

Just as with Hong (and Zentner’s Broadband variable) the major concern is regarding identification strategy: it is difficult to be confident of estimates that depend on assuming that differences in purchases between computer owners and non-owners can be attribued entirely to file-sharing, particularly when computer-ownership may be associated with so many other activities and characteristics.