Recently we’ve seen several studies of the impact of filesharing on CD sales. We have enough data now to draw some (very) preliminary conclusions, assuming the studies are correct. Despite the apparent contradictions between the various studies, I think there is a plausible theory that can explain them all – a Grand Unified Theory of Filesharing.

First, let’s review the three main results that have to be explained.

• Survey-based studies, which ask people whether they use the Internet, whether (and how much) they use filesharing, and how many CDs they buy, find that people who fileshare buy fewer CDs.
• The recent econometric study by Oberholzer and Strumpf, based on per-album time-series data on filesharing activity, CD sales, and other factors, found that filesharing has little or no effect on CD sales.
• Eric Boorstin’s study found, controlling for differences in personal income, that there is a strong positive correlation between Internet usage and CD purchasing. This held true for all age groups, except the 15-24 group, for whom Internet usage correlates negatively with CD purchasing.

(It’s undisputed that CD sales have dropped sharply in recent years, but there are several plausible causes for that drop. That’s a topic for another day. Here, I’ll assume only that filesharing is not the only cause of the sales drop, so that we don’t need filesharing to explain the drop.)

The Grand Unified Theory explains the study results by breaking down the users of filesharing into two subpopulations, which I will call Free-riders and Samplers.

Free-riders are generally young. They have few if any moral qualms about filesharing, and they tend to assume that others feel the same way. They use filesharing to accumulate libraries of music, as an alternative to buying CDs.

Samplers are generally older and more risk-averse. They are highly engaged with cultural products of all sorts. They are morally conflicted about filesharing, and use it mostly to download songs that either aren’t for sale, or that they don’t value enough to pay for. They buy music that they really like, and filesharing causes them to find more music they like, so it tends to increase their CD purchases.

Now let’s look at how the theory explains the studies’ results.

In survey-based studies, Free-riders admit to filesharing and to buying fewer CDs because of their filesharing. But Samplers are reluctant to confess their filesharing to a stranger, being more risk-averse and more attuned to the dubious moral status of filesharing (not to mention its illegality). The result is that Free-riders are overcounted in survey-based studies, and Samplers are undercounted, so survey-based studies find that filesharing depresses CD sales.

The Oberholzer and Strumpf study measured the actual impact of both Free-riders and Samplers, and found that the lost sales caused by Free-riders are balanced by the increased sales due to Samplers.

The Boorstin study had different results for different age groups. His 15-24 age group was mostly Free-riders, who buy fewer CDs when they have Internet access, because their filesharing substitutes for purchases. His older age groups were mostly Samplers, who buy more CDs because of filesharing, and who are also, because of their high level of cultural engagement, predisposed to both Internet usage and CD purchasing. Therefore he found that young Internet users buy fewer CDs, while older Internet users buy a lot more.

So there you have it: a theory that explains the study results, and that seems plausible (to me, at least). Of course, there are lots of caveats here. One or more of the studies might be wrong; or the studies might be right but the theory wrong. But bear with me for a bit longer as I explore the possible consequences of the theory.

The theory says that the net effect of filesharing on CD sales is roughly zero, because of a balance between the negative impact of the Free-riders and the positive impact of the Samplers. But what happens in the future? It all depends on what happens to today’s Free-riders.

Perhaps today’s Free-riders will mature into Samplers, to be replaced by a new generation of Free-riders, so that the effects of the two groups continue in a rough balance. Or perhaps today’s Free-riders, never having known anything else, will keep Free-riding as they get older, and the balance will tip toward Free-riders.

It’s also worth noting that the theory does not predict whether (illegal, free) filesharing will reduce online sales of music. Probably the answer depends on what the online alternatives look like, and how convenient they are to use.

So the theory can explain the present situation, but it doesn’t make strong predictions about the future; or, if you prefer, the theory comes in several flavors, which differ in their future predictions. If we had a better handle on what makes one person a Free-rider and another a Sampler, we could make better predictions.

[Thanks to Eric Boorstin and Andrew Appel for helping me develop and refine these ideas.]

Eric Boorstin, a senior at Princeton, just filed his senior thesis, Music Sales in the Age of File Sharing. The thesis includes a clever study of the impact of Internet usage on CD sales. This is a twist on previous studies, which have tried to correlate CD sales to usage of filesharing. The tradeoff here is that although Internet usage is one step removed from filesharing, the data on Internet usage are much more detailed and much more reliable than the data on filesharing usage.

Eric worked from two datasets. The first dataset came from SoundScan, and gave him aggregate sales of CDs, on a week-by-week basis, for many separate metropolitan areas in the U.S. The second dataset came from the U.S. Census Bureau, and contained data on population, income, and Internet usage, broken down by age group and geographic area. The census data came from 1998, 2000, and 2001. Combining these datasets, he ended up with data for CD sales, age group demographics, income, and Internet adoption, at three different points in time, in ninety-nine separate metropolitan areas in the U.S.

Eric took these datasets and did a regression to determine the correlation between Internet adoption rate and CD sales, broken down by age group. He controlled for differences in personal income. (For more methodological details, see the thesis.)

For people in the 15-24 age group, he found a significant negative correlation between Internet adoption and CD sales. For people in all of the age groups older than 25, he found the opposite

Today’s New York Times has a great story by John Schwartz on last week’s filesharing study and the reaction to it. There’s a nice summary of the study itself, and some discussion and criticism of it.

The criticism seems to fall into two categories. One category is the appropriate scholarly caution toward a new result that hasn’t been peer-reviewed yet. Although economists who have seen the study say its methodology looks reasonable, there may be other unknown factors yet to be discovered that will cast doubt on the study. The other category of criticism comes from people who don’t criticize the study’s methdology but just point to other types of studies that give different results.

The article notes that these other studies haven’t been peer-reviewed either, and that some of their sponsors have agendas. Anybody who has been around for a while knows to be very skeptical of certain kinds of studies done by certain kinds of consulting firms.

Eric Rescorla offers two interesting posts on the recently released study on filesharing. First, Eric summarizes the study’s methodology. Then he discusses the implications if the study turns out to be correct that filesharing does not reduce sales.