All well and good, but your sample is usually very small and you as a human being can't handle confounding factors very well. Add on the fact that you are also prone to selection bias...
For example how many of your friends (or friends of friends if your sample is too small) are dead in auto accidents is a lot better predictor for your fate than any global country statistics you can find.
Also, the thing with quite a few statistics is that the risk is say 1/1000, and I honestly have closer to 50 people who I interact with in a month.