I edited my original comment to expound. The "evidence" he cites is incredibly thin, in my professional opinion.

Because correlation does not imply causation.

But if one factor correlates more with crime than all the others, it sure looks like that one is the cause.

Can I use this comment (anonymized of course) as example in the next course I teach on regression stats? Because it's the exact opposite of how you're supposed to do regression inference. Selecting feature sets by beta alone is a recipe for disaster.

Consider the following example:

A + B => Y (one A and one B causes one Y) A => 2Z (two A causes one Z)

If you have a data set featuring B, Z, and Y then a regression analysis will yield that Y = 2Z + B, and you would conclude that Z is "a root cause" of Y over B. However, by construction Z has no causal effect on Y, only B does. This is literally the point of "correlation does not imply causation".

Moreover, the white-paper the op-ed author is citing [0] doesn't even have "fatherlessness" as the most correlated variable: it's poverty across the board (and by a lot). Even if your argument was sound it wouldn't be correct.

[0] https://ifstudies.org/ifs-admin/resources/reports/ifs-strong... Tables 2 and 3

Thank you, I should have read the white paper better, you are right, the correlation with single parent households is actually one of the weakest.

By the way, do I read correctly from the regression analysis stats, that "single young men age 15-29" is -negatively- correlated with all categories of crime?

Yes, that’s correct. Mixed significance pattern and wide-ish SEs so a little challenging to make claims across crime types IMO.

This demo tends to commit the most recorded crimes, which means that they’re getting arrested or murdered. Presumably in an area where lots of crimes are occurring the corresponding increase in arrests and murders leads to lower population counts on the US Census (which I believe was their source of truth on demo stuff).