Ars Technica has spent the past few months crunching numbers from millions of Steam pages in order to create an accurate sample size of the whole. The first article has more information on how the data is gathered if you are interested.
Source: http://arstechnica.com/gaming/2014/04/steam-gauge-addressing-your-questions-and-concerns/2/
How accurate are your numbers?
Since publishing, we've had a few more developers reach out either privately or publicly to offer their own Steam sales data for comparison to our estimates. With over a dozen "real world" spot checks in hand now, we have yet to see an instance where the error in our numbers is more than 10 percent off from the actual numbers developers have access to. Sometimes our error is much less than that, of course, and the error can go in either direction (though so far our numbers seem to over-estimate slightly more often than under-estimating).
While 10 percent isn't a small functional margin of error, it's also much better than a simple shot in the dark guess. If we're reporting sales of two million units for a game, you can be pretty confident the actual sales number is somewhere between 1.8 and 2.2 million.
Source: http://arstechnica.com/gaming/2014/04/steam-gauge-addressing-your-questions-and-concerns/2/