I wrote a post about this a while back with some back of the envelope calculations:
The basic upshot of it is - it depends on what you’re looking at. If you’re looking at statistics that are not strongly correlated with private profiles (such as relative pick rates between heroes) then you can be confident in its reliability as long as you use it carefully.
For instance, if you want to look at GM data, you need to look at averages over several weeks before you have reasonably reliable numbers. If you’re looking at all ranks, you don’t need averages over large time periods. The post I linked has some numerical guidelines for these time periods.
On the other hand, if you’re looking at statistics that are probably correlated to public profiles (such as on fire percentage) then the reliability decreases.