[PeterisP]

Actually, you should generally expect that ANY statistical analysis will give significantly different results than what you'd see of people around you.

Many of such parameters exhibit clustering; so if 50% people are X; then you'd expect most people to say "hey, that's wrong, that's not what I see" - since there would be many people that would see that 80+% of their friends would be X; and there would be many people that would see that 20-% of their friends are X, but virtually noone would have a circle of friends that would have the "true" 50/50 split.