> often see people that are above average old, smoking.

it's quite a sample bias, because those who you see smoking are likely quite healthy and thus is _able_ to smoke to old age (or won a genetic lottery to be able to withstand smoking).

What you need to count is how many people smoke, and then the % of them dying earlier than those who don't. Otherwise, sample bias will give you the wrong conclusions, esp. if this bias is affirming your own world view.