Identifying Gender by Handwriting: You’re Probably Not as Good as You Think

> "We were surprised that respondents weren’t better at identifying the gender behind the writings, and that so many were wrong on several of the questions. The participants who submitted handwriting samples were blind to the purpose of the survey, and the samples therefore weren’t intentionally misleading. Perhaps the average person simply isn’t as good a judge as expected."

Perhaps being spammed with a survey wall blocking the content that the user seeks, until they complete a survey (which is what survata does), causes many people to not try very hard. And perhaps the more technically sophisticated users who view survey walls as a signal for low quality, spammy content, simply hit the back button.

The first and last female samples are identical (except for size, they appear to be exactly the same image). Hard to believe that one was identified correctly almost 80% of the time, and the other only 30%.

If the test was given as a list, then I suspect a large number of the results are skewed just by the nature of people's desire to think in terms of pattern-breaking (e.g. Would the results be the same if it was the five male samples followed by the five female samples, or if only one sample was provided per person). Of course, talking about people's "nature" is itself conjecture and probably warrants it's own study.

So, men always guess it's a man, and women always guess it's a woman?

I wonder how a deep-belief network would perform at this, or even just an SVM.