It can be completely biased. This is why my definition is probably different than yours in the first place.

On the other hand, your sample may be also biased, so it goes both ways.