Both the title and the y-axis says that the goal is to estimate prevalence from self-ascribed Twitter bios. I'd say author is claiming to do more than simply report on the Twitter bios, they are saying this is an estimator of prevalences. I'm simply providing some evidence to other folks that this not a good estimate of prevalence in the general population.

For your point. There still isn't evidence to say that the self reporting aspect is at fault here. We know the end result is off, but you'll need more evidence to claim whether it's because Twitter users are a good representation of the population, and the bios are over-reporting, or whether the bios reporting is representative, but the Twitter population is not (...or both).