These percentages are supposedly Bayesian estimates, so it basically just means that it isn’t easy to tell one way or another but the thing was more inclined to take it as female. If the thing is well calibrated it would be right 63% of the time and wrong 37% of the time with this estimate. But at least for my tests it was right even more often—it seems other people had different experiences.
This is based on all the estimates that people have voted on. So it’s not strange if it’s only getting 63 − 70% correct; it’s giving many estimates which are less certain than this.
Darn—claims my blog is 63% woman. Not sure how to take that!
These percentages are supposedly Bayesian estimates, so it basically just means that it isn’t easy to tell one way or another but the thing was more inclined to take it as female. If the thing is well calibrated it would be right 63% of the time and wrong 37% of the time with this estimate. But at least for my tests it was right even more often—it seems other people had different experiences.
Just clicked through to the following screen after selecting “no—it didn’t get it right” to see the resulting poll:
Yes − 63% No − 32% Don’t know − 5%
This is based on all the estimates that people have voted on. So it’s not strange if it’s only getting 63 − 70% correct; it’s giving many estimates which are less certain than this.