Ok, so I took the test that I believe you took and because I hesitated when forced to associate gay people with bad terms and straight people with good terms (not knowing that they would reverse this pairing later in the test), my result showed me to have a moderate preference for gay people. I was faster at the straight people + bad terms pairing because I had completed the other pairing and realized the nature of the test. In the explanation, they say that the order of pairing is a real effect and that they normalize for this, as well as giving half the people one pairing first and the other half the opposite pairing first. I'm not sure that they really have that normalization correct. Although I am straight, I really have no preference for either gay or straight people... I don't associate who people choose to love with good or bad terms. If they are nice people, that is really what matters, not anything else. The result showing a preference for gay people was my reluctance to be forced to pair people with good or bad terms when grouped in that way.
So... do I really have an implicit bias? Maybe . But I would hate to have my job depend on such a test! I wonder what would happen if I took it again from a different IP address...
This test is measuring reaction times and whether you follow instructions even if you don't like the answer you have to choose.
When the sun shines, bask.
Classical Guitar forever!