## Support Vector Machines and Gender Differences

Earlier I wrote that gender is like a maximally separating hyperplane, including vast within-gender differences? I was trying to use maths to end the war of the sexes.

Some who read it didn’t know what a support vector machine was. You can find out by playing around with this SVM tool. Sprinkle around some pink, blue, and gold dots like I had.

• There should be lots of distance among blues and lots of distance among pinks.
• There should be some clustering of pinks and some clustering of blues.
• I don’t have any data or presuppositions about the golds [trans, queer, hermaphrodite, etc] so I just put them wherever.

Set the kernel to linear (`-t 0`) or radial (`-t 2`), hit `Run` and you’ll see the maximally separating hyperplane — just a line in 2-D.

Thanks to James Grahn for sharing the link to the SVM tool.

Playing around with this, you’ll realise that my earlier claims were less than precise (as some readers pointed out at the time). That’s OK. The point of bringing up gender differences was to suggest a different way of thinking, not to claim I have all the answers.

Just like it’s good to think of men’s and women’s 100m dash times as distributions, the SVM-in-a-high-dimensional-space metaphor is better for thinking about Gender Differences than the most obvious ways we conceive of them (“boys scored 3 points lower than girls on their verbal SAT? Oh no! Our boys can’t communicate!”).

To restate what I said before: In real life, the blue & pink dots are distributed in high dimensional space with vast within-gender differences and lots of overlap among the genders. Too, there are meaningful differences between men and women and an SVM is one way of showing that those exist. In fact, considering the totality of human differences, the SVM should classify better for male-vs-female than dividing along racial lines, economic lines, etc.

That’s my quantitative interpretation of the statement “There is a great difference between the sexes.”