J.D. hit the mark with his post.
If I survey 7,000 Harvard graduates and find that one group who does a particular thing lives longer than the group that does not, what have I shown? What I have shown is the activity is ASSOCIATED with longevity. However I have not proven causality. This type of research is suggestive, but nothing more.
And when researchers do nothing but troll data for patterns, they also put themselves at risk for finding patterns that are odd random occurances. An important lesson that is taught to UVa medical students is to have them all take a battery of medical tests. Presumably they are all healthy. Then when everyone gets their results, they are all told what the "normal" ranges are. Then test by test, the professor asks the class to raise hands for all who fall outside the "normal" range. At least one hand comes up for each and every test - and they are almost all NORMAL kids. There's supposed to be important lessonS there. Unfortunately I know for a fact that they escape many of these students.
But let's assume that the ASSOCIATION is real and repeatable. Where does a researcher go from there to prove causality? Well he has to do three things: 1) He must subject the "treatment" to the gold standard of research - the randomized trial, 2) he must explain HOW the effect occurs, and 3) he must demonstrate the effect at the fundamental level. When all this is done, then you know FOR SURE that eating chocolate may make you live longer.
In the mean time, you still have hope if YOU ALREADY eat a lot of chocolate. Perhaps then you might be part of that population whose behavior is associated with an extra year of life. If so, then great - buy a bottle of Dom Perignon and celebrate!