Thread by @JohnFPfaff

I think conditional outcomes is a huge thing we don't think about enough in general. Like when talking about crime during Covid.

Did, for example, crime in the subway in NYC go up or down? It's... actually really hard to say, bc it all depends on what we condition on.

https://twitter.com/JohnFPfaff/status/1480589936323661840

The raw data--victimizations per 1M riders--suggested an increase, in no small part bc total ridership plunged.

But... the plunge wasn't random. What if those who started WFH were always less likely to be victimized in the first place?

That was almost surely the case.

Park Slopers who already took the subway at peak rush hour from a safe Brooklyn station to a safe Manhattan station stopped taking the train.

Late night workers, early morning shifts? They were now "essential workers" and still on the train.

So if we condition not on raw ridership but some more complicated and nuanced measure of "ridership at risk," we may get a completely diff sense of what happened during the pandemic.

Hard to measure? Sure. (Tho less so in the smartphone era.) But the better way to think abt it.

Someone #onhere, the last time I talked about this (can't find it how), posted a really interesting paper arguing that this is why school bus stops seem like crime "hotspots": they have a lot of ppl. Condition on the number of ppl, and they're unremarkable.

I see this with prison populations too.

Did the incarceration rate rise in the 1970s or fall?

Seems like a crazy question! But....

Depends on the denominator--on what we condition on.

There's never one right answer. It always depends on what we are trying to think about.

But I feel, sometimes, like we just instinctive use some conditionals/denominators without really thinking carefully about it at all.

And that's ... really bad.