500 people gives you an error margin of +/- 4.5% (Variance of a binomial is npq=np(1-p), worst case p=1/2, so n/4, std deviation thus sqrt(n)/2, then take 2 std dev to cover 95%, thus sqrt(n), as a percentage of all responses sqrt(n)/n = 100/sqrt(n)%), which is not amazing, but not BS either. Is nobody aware of basic statistics?

When you have a really poor response rate, don't you worry about how it might have biased your sample?

People talked about this a lot with respect to political polling by phone during the last US Presidential election. Very few people answer phone polls these days and figuring out how different they are from the general population is important.

Many of us went through study size calculations when we were first learning how to run clinical trials (well, probalby not many, but everybody in my cohort did). The general conclusion among quantitative scientists is that social science like this is worse than useless because of how much time you have to spend explaining to people that the world is not a simple uniform random sampling.

I'm just here to get angry at a misrepresentative editorialized title!

This is hacker news... our time is too valuable to waste actually reading the article and making an informed opinion.