"Gaydar" Experiment

This experiment to test the accuracy of "gaydar" appears to me to use a flawed methodology. (Or perhaps the brief article doesn't accurately convey the method and results of the experiment.)

Ambady and colleague Nicholas Rule, both at Tufts University in Medford, Massachusetts, wondered about sexual orientation. They showed men and women photos of 90 faces belonging to homosexual men and heterosexual men for intervals ranging from 33 milliseconds to 10 seconds. When given 100 milliseconds or more to view a face, participants correctly identified sexual orientation nearly 70% of the time. ...

Psychologist David Kenny of the University of Connecticut, Storrs, says the finding demonstrates the brain's remarkable ability to make fast yet accurate appraisals. Still, he notes that with some of the images, accuracy regularly fell below 50%. It's possible that some faces are just hard to read.

Considering that far less than 50% of the population is gay, an experiment based on an equal number of gay and straight videos won't really capture the essence of the "gaydar" phenomenon. It's also important to break out the mistakes into false-positives and false-negatives.

For example, in real life a person "A" who assumes that every person they meet is straight will be right more than 90% of the time. That doesn't mean we all have highly-attuned "gaydar", it's just a reflection of the rarity of homosexuality. Person A will have a false-positive rate of 0% and a false-negative rate of 10% -- the highest possible in a real setting.

A useful measurement of "gaydar" should weight the false-negative error rate higher than the false-positive error rate, because otherwise it's very easy to "game" the measurement system by simply picking "straight" every time. For example, false-negatives could be weighted fives times as heavily as false-positives, which would give Person A a modified error rate of 50% instead of 90%.

(HT: GeekPress and SciTechDaily.)