Another thing to consider as far as selection bias goes is why the sample of LAGS is so much smaller for both your database and Niedams. An obvious answer is that TAG is the book style and most people play by the book.
Another potential answer is thet LAGS tend to go broke fast unless they are really good and you don't have a large sample because the bad ones bust and quit playing. If this is true, your winrate for LAGs is biased upwards because you are not observing the negative winrates for the players who quit because they bust.