I find myself head-thumpingly frustrated by the fact that all parties in the Copyright Wars continue to get their statistics wrong
. This time, ReadWrite trumpets the idea that participation in social media by musicians drives music sales
I realize that's an appealing and perhaps even common-sense notion and they even quote a graph with the word "Causality" on it, but that is not in fact what is going on and it's not even what the original graph ought to be claiming.
What happened, near as I can make out, is that an outfit called Next Big Media did some data analysis. They looked at some public numbers, such as hits on an artist's Wikipedia page, publicly released iTunes sales, and so on. Then, to their credit, they did some actual statistical analysis. In particular, they did what's called a Granger causality test, which attempts to show that one variable has enough predictive value in its time series to be assigned causative agency in another variable.
Causative agency is much stronger than the usual notion of prediction and it's a tricky thing to pin down. You can, for example, see that in certain months there's a large rise in the number of people wearing overcoats. The calendar date is therefore a good predictor of overcoat use, but it's not a causative factor.
Using a Granger test is good in that it avoids the most simplistic "correlation = causation" failure. However, as Wikipedia and other sources will tell you, Granger Causality is not necessarily true causality. For one thing, it's a test that works only when you have two variables, not three (or more). For another, it's known to fail when there's a (so-called hidden) variable that also follows the same time series. In this case, we can call that variable "popularity". What this study is telling you is that if you can tell when someone is getting popular then you can predict they're going to sell more music.
This, ladies and gentlemen, is not particularly enlightening. We know this, and we further know that public resources such as Facebook pages, Google searches, and Wikipedia article activity are reasonable measures of popularity, particularly when you measure what's popular within the limited subset of the population that is online and connected. Unsurprisingly, this is also the subset of the population that is most likely to buy from iTunes rather than Wal*Mart or other physical music retailer.
There are other methodological flaws in the study - for example, they seem not to be taking into account things like "has just released a new album" or "has appeared on The Simpsons" or "is touring my country" or any of a zillion other factors that may cause jumps in social media popularity, and likewise jumps in sales. I could go on, but you get the gist.
I realize that news outlets have to fill a certain number of (even virtual) column inches, but really when the best thing you can conclude is "artists should make sure their Wikipedia pages are updated and maybe get on Twitter too" - that's pretty lame.