I wonder if they get this by doing an analysis of the text. There are a couple ways of doing this, one involves looking at grammatical and syntax features (a la Gender Genie) and the other is more qualitative, looking at certain subject keywords and emotional keywords as well as comment length. It wouldn't surprise me if comments on this site, analyzed using either method, tended to be more free of gender tells than the writing of the general population. That being said, it's difficult to get more than 65% accuracy no matter how you do it.
Like my girlfriends and me, we feel so betrayed that Sun bought MySQL. My best friend Sarah said that she felt like she had a real relational database situation, commits and all. Sometimes I just want to cuddle with my PHP manual, especially when I have a headache from looking at my screen so long. Maybe after I finish refactoring this program, I settle down with a nice cup of green tea, then go shopping for a USB cable and some new shoes. After a long bubble bath, I'll get back to the rest of my code.
If you ever needed an example of what is wrong w/ all the panel-based demographic and analytics info sites - this is it.
This kind of methodology works fine if you are analyzing a very very large site that many people use (facebook, myspace, etc). It's old thinking to think that a panel of people can accurately measure niche content. It might work for nielsen on TV, but it obviously is completely broken for web sites.
I got about the same from YC News recently. I'd guess there are about between 5,000-7,000 uniques. Why?
1) My story (mike's too) got on the front page and had some decent discussion. I think 15-20% of uniques reading a featured story is a reasonable estimate.
2) The site is about news and community, so it doesn't make much sense to check infrequently. I'd guess that the uniques/week isn't much below the uniques/month. So it wouldn't matter that the story didn't stay up for a long time.
3) The leaderboard goes up to the top 50, with karma dropping by about 100 per 5 spots towards the end. If you take the rule of thumb that 90% just read, 9% contribute a little, and 1% contribute a lot, then if the top 1% is 50 people, you get 5,000. If it's 70 people, then you get 7,000. I recognize most of the names on the top 50, and I bet I'd recognize half of the top 51-100. So again, 5,000-7,000.
the contribution percentages may be different here b/c we have a lot of the kind of people who take initiative, aren't afraid to put their ideas forward, know how to create user accounts, and have stuff to say.
According to the site, they use "panel-based information," which is culled from voluntarily shared anonymous user statistics. Maybe the correct assumption here is that the women who visit ycombinator are more likely to share usage statistics than the men?
Uh, The boys and I think your algorithm needs a little tweaking.