they only sampled a few thousand leaked responses out of over a million. The margin of error is 2.5% on conclusions because they didn't use enough data, not even close.