History is not a statistical science. It is our best attempt at assembling some idea of what happened in the past. That idea is often wrong, and any good historian will tell you that if you base your stuff only on secondhand information, it's bad history. Further, any good historian will tell you that even with extensive primary sources, we can still be led astray if all of those sources are biased for the same reason.
Finally, while the ravages of time do act as a selection process, that selection process is not nearly as biased as people volunteering information by self-selection. It IS biased, but the bias is (usually) much more random, which means the data we derive from it can be more meaningful, even if not absolutely so.
Archaeology helps us do the best history.