It’s a smaller sample than all cars I ever see, sure. But it’s still fairly big. If I look at a car I’ll usually try to look at the drivers face too, just because it’s useful information to know where they are directing their attention. I don’t think that’s particularly idiosyncratic and so it seems like a somewhat cheap argument to me to imply that people are basically just willfully deluding themselves.

The issue is cognitive biases make the total sample size irrelevant, which is something to be aware of. Unfortunately, said biases are not willful which makes them far more insidious a problem to notice and deal with.

Anyway, as a simple test try and count how many cars you can see next time you go for a drive, how many people’s gender you can infer, and how many you can positively identify based on their looks. It’s significantly safer to try this as a passenger.

> Anyway, as a simple test try and count how many cars you can see next time you go for a drive, how many people’s gender you can infer, and how many you can positively identify based on their looks. It’s significantly safer to try this as a passenger.

Might be interesting to try, but I think in a situation where I notice a bad driver I'm much more likely to identify their gender than in a random passing car.

But I live in a comparatively small Austrian city riddled with traffic lights, weird intersections, crosswalks, and way too much traffic in general, so I get ample opportunity to look at slow moving cars. If you're American your culture around cars is probably very different and that might explain why we seem to have a different intuition about this.