The data is sampled from the Heyzap app. So English focused social games will tend to be overestimated.

We did calibrate on quite a few data points. I will think about normalizing for the above bias too