Just thought I'd say hi. Nothing to add.

You bring up a good point that I implicitly referred to by saying “sexual identity” and for terrorist mining “perfect information.” The statistics don't work even when you have ideal information, which depending on environment and circumstances may or may not be true.

For instance, on the Kinsey scale/behavioral surveys, about as many men are predominantly homosexual that don't identify as such as there are that do. Plus there are about 3x more that have varying degrees of active bisexuality. The quality of identity information as reflective of behavior is extremely varying based on geographic and socioeconomic variables (although increasingly less so).

In the US where religious identity is pretty open, the information is pretty good, while in China it would be poor.

The worse the information, the harder it will be to get good predictors at all, and you'll start getting a lot more false positives and negatives.

Terrorists groups purposefully keep as few network connections as possible to make it harder to infiltrate. Thus in reality, not only do you have thousands of false positives, but the number of false negatives is really high too.

Once people start hiding information (or even outright lying) then these sorts of studies fall apart very quickly in saying anything definitive.

The one thing that does come out of this is that, in general, if the algorithm says you're not gay, it's pretty unlikely that you actually are. But (taking it to your larger point about terrorists) nobody's out there trying to figure out who's NOT a terrorist.