@futurebird @AnnyJoe @bmoe another case where facial recognition and vigilante justice would solve problems but also be problems themselves. Although maybe "extra-juditial justice" would be a better term than "vigilante" for the shopkeeper finding the parents and showing them the video.

@violetmadder

In my experience polling firms skimp on leg work. They hate paying people to go and find and interview their sample because it's expensive.

There are a lot of data out there, but it's very biased in unpredictable ways.

Pulling text from facebook twitter or X has nothing on doing interviews on a sample you have randomized well.

And it's critical to try to find the entire sample to the best of your ability.

@futurebird

Not reliable, no. And I'm starting to think that's one hell of a double-edged sword. Because on the one hand, no they don't REALLY have solid or accurate conclusions on what the resistance is up to, but on the other hand these egomaniacal bastards sure would love to BELIEVE they do which is just as scary in its own very special horrible way as they seem plenty confident enough to use the data to target arrests and bombings.

@violetmadder

The kind of polling work I see tells me they aren't really interested in finding out what people really think or what people REALLY care about.

There are topics they don't even ask about that are massive. The questions are leading and full of assumptions.

@semitones @AnnyJoe @bmoe

A good faith operation wouldn't need to do that. "call your parents they need to come pick you up" is enough of a punishment. (and pay for the item)

But this is very boring and difficult work that requires police who work in the same neighborhood for years (their "visibility" statistical system means I never recognize any of the officers since they are sent all around the city by the computer)

It's a whole different way of looking at the role of police.

@futurebird @violetmadder as someone who has worked in NLP for over a decade, word counts would be a much better way to go. You could probably also cluster them in some way which might help distill the data down.

In fact, this would be a terrible application of an LLM. Anything involving counting or understanding of anything countable, and LLMs fall flat on their face. They generate random words, in essence. Not good when you’re trying to observe something.

@futurebird I feel like polls won't get you the answers you seek.

In the 70s, NYC had a massive crime problem. The subways were dangerous, car break-ins were frequent, petty theft was at an all time high.

Why?

Heroin.

There was a massive (and now well-documented) gang moving tremendous amounts of heroin in the 70s, and lots of people were doing it and having the associated problems. So crime overall was actually high.

@futurebird
Eliminating that heroin supply is what cleaned up New York, not the broken windows policy or stop and frisk.

So there is the perception and the reality, as you say. How would better polling reflect the presence of an organized criminal underground that no longer exists?

@Wyatt_H_Knott

I'm interested in WHY people are so receptive to "crime is out of control" political messaging.

The causes of the crime itself isn't a matter for polling and also is very interesting.

@futurebird @violetmadder
One of my favorite books is “Seeing Like A State” by James C Scott. It’s about how states try to make the world “legible” by creating simplified understandings of the world, and then actually imposing those understandings on to the world. This often works in the short term, but destroys the contextual knowledge and capacities of the people imposed upon in the long run.

This style of polling might better reflect that contextual knowledge