Felix Oberholzer and Koleman Strumpf, of Harvard and the University of North Carolina, respectively, have published an interesting study on the effect of file sharing on record sales. They looked at album sales, actual download traffic for individual songs, and several other variables. Their main conclusion that file sharing had little or no effect on sales, and they could not reject the statistical hypothesis that filesharing had no effect at all on sales. Though these effects are not statistically significant, the data suggests that file sharing may boost the sale of the most popular albums, and may depress the sales of less popular albums, with a near-zero net effect on sales.
How much should we trust these results? I don’t know. The authors look like respectable academics, but their methodology is complex enough that I am not qualified to evaluate it. Perhaps a more qualified reader will have something to say about the study’s methodology.
Oberholzer and Strumpf report that factors which increase the inconvenience of free music downloads, such as Internet congestion, fail to impel downloaders to go out and buy the music instead. From this they conclude that downloads do not hurt music sales. This conclusion requires a leap of logic that is unsupported by any evidence. Music fans committed to the notion of free downloads will predictably respond to inconveniences by trying again at more convenient times, but this tells us nothing about what they would do if file-sharing didn’t exist. The best evidence we have on the latter score is what happened in pre-Napster times. In those days, music sales were substantially higher than they are in the file-sharing era. This fact does not by itself prove the record industry’s claim that file-sharing was responsible for the sales slump, but the Oberholzer/Strumpf study certainly does nothing to disprove it.
I just skimmed the paper. It looks like a fairly typical piece of econometrics.
So, the basic problem with trying to estimate the effect of downloading on music sales is that the amount of downloading D and the amount of sales S are both related to the popularity of the album P.
In fact, in some perfect world they would be linearly related. As-in:
D = aP
S = bP
So, in order to decouple these two we’ve got to look for situations in which downloading is easier or harder than normal. So, for instance, suppose that Napster goes down for a week. Do sales go up? In general, it’s not so simple so you end up using what are called instrumental variables. For instance, say you have a measure of Internet performance. During periods when performance is bad, downloads look less attractive. If sales go up during this period (or some later period with a known lag) then you might conclude that downloads depressed sales.
The authors use a bunch of such approaches but don’t really find anything. That doesn’t mean there’s nothing to find (though it’s suggestive) just that we don’t know.
If I get some time, I’ll do a more thorough writeup on EG.