There's no way they're not accounting for selection bias. This isn't on online poll where you score among x% of people who took the test. They give the test to a (hopefully) representative sample and score you against that.

I don't think tests like this one are calibrated by those who happen to take it online. I believe they calibrate these tests by looking at a known test taker's results and comparing them to other test results. The other tests would be the high quality ones used in psychology that have a much larger, less biased selection of test takers.