You can weight by frequency of activity—say, approximately "players sampled from matches selected uniformly at random"—and it'd largely still be true. The top f

You can weight by frequency of activity—say, approximately "players sampled from matches selected uniformly at random"—and it'd largely still be true. The top few percentiles do practice more often, but overwhelmingly so. Glicko metrics for two-player games basically have this property, so if you want to try it out in the wild, you can pick up a game with a Glicko-rated ladder and see how hard it is to get your mean rating estimate up +2 SDs.