> massive influx of personal anecdotes

a.k.a. sampling from the distribution of interest. Once you have enough anecdotes, you actually have a representative distribution of the real thing. Selection may be biased but you can weight those data points as you like, it's still good data.