small sample sizes.

poor sampling methods.

high margins of error (or no way to calculate one at all).

an obvious bradley effect.

wait before you react.