Hacker News Reader

Comment by blackbagboys | Hacker News Reader

Would "75% of enforcement done in neighborhoods of color" still be an issue if 75% of crime occurs in those same neighborhoods?

View on HN

This argument is so seductive, but it bugs me that people assume that "crime rates" are based on how often people are committing crimes. It's how often crimes are reported / how often criminals are caught.

Given the greater prevalence of things like stop and frisk in neighborhoods with large minority populations, is it any surprise more people are caught for things like drug possession? That's going to further skew the stats, leading to more enforcement in those neighborhoods (since they are "high crime").

Drug use is actually higher in young white populations than it is among young black populations, but because of where law enforcement spends their time the incarceration rates differ wildly.

Look at the distribution of violent crime.

Yes, that would be one of many possible corrective measures.

It seems like you're making an argument that crime statistics should not be used as a metric to focus law enforcement efforts.

What other metrics should we use? If you simply assign patrol routes based on population you are going to under-serve areas with higher crime and over-police areas that don't need it.

I'm arguing that it's a more complicated statistical problem than "we see more criminal 'events' in this neighborhood, so we should send more cops there."

Offering criticism without offering an alternative isn't helpful.

Of course it's complicated, and of course crime stats are a simplification. Most stats are. The point is, stats are much better than going by biased "gut feelings".

Yes there are underlying issues with stats such as underreporting, but the solution isn't to get rid of the stats. It's to improve the underlying cause of the bias in the data, such as improving police/public relations.

Your conclusion and the one I implied in my comment are not mutually exclusive.

My solution, implied in my comment, was not to use naive statistical models or ideas (i.e. 75% of "crime").

An "implied" solution isn't a solution because it was never suggested.

And your now stated solution is easier said than done, and certainly not immune to the same kind of biases a naive solution may be subject to.

Possession is a red herring. The CompuStat statistics don't even count drug crimes and gun possession crimes--the incidence of which might be affected by the intensity of policing due to stop-and-frisk: http://www.nyc.gov/html/nypd/downloads/pdf/crime_statistics/....

Here is the NYC crime map: https://maps.nyc.gov/crime. Look at the maps for felony assault or rape. Stop-and-frisk isn't going to change the incidence rate of those crimes.

You obviously have a more sophisticated take on the question. Parent comment referenced "75% of crime," not "75% of violent crime" or "75% of rapes and aggravated assaults recorded in compustat."

The general point holds - there are corrective measures that need to be taken to avoid skewing crime data as a result of increased enforcement, and in the case of some jurisdictions those measures are being taken.

What's not obvious to me is whether police are increasing the severity of the charges based on where they are, even if the charges wouldn't necessarily hold up in court. There's a case to be made that that would be an efficient tactic - public defenders will encourage plea bargains and it gives the DA more leverage to settle the case quickly and efficiently. The opposite may be true when booking a drunk banker who gets in a fistfight, or a privileged college kid who rapes his date behind a dumpster.

Also, the discussion generally is not about CompuStat, it's about "quality of life" improvements prosecuted with the use of secret databases that are not publicly available. So, you know, there's that.

The disparity doesn't go away if you use victimization surveys.

It's important to keep in mind that in a "racist society" machine learning will also tend to be racist (but much more so). After all, the "AI" is learning from human data, so it's certainly not as objective as you would think a "machine" to be.

https://www.propublica.org/article/machine-bias-risk-assessm...

https://enterprisersproject.com/article/2016/9/beware-biases...

http://www.nytimes.com/2016/06/23/us/backlash-in-wisconsin-a...

And similarly, in a society in which biological differences exist between races, an objective, unbiased AI will be tend to be racist.

This isn't nuanced, it doesn't add much to the discussion. It's "I think people of color commit more crimes" wrapped in an acceptable veneer.

And what if "people of color commit more crimes" is statistically true?

Since pigmentation doesn't cause crime, we should find the root cause.

Is it acceptable to assign more police officers to neighborhoods of root cause?

Get out of here with your common sense that doesn't fit the narrative.

It is a question of a crime taking place vs one which results in an arrest, convictions and statistics. I worked on Wall St for many years -- there are a lot of drugs on Wall St, more than i've seen anywhere else. Yet the rules are rarely enforced as they are in other neighborhoods. So of course, Wall St wont make it into the statistics. It is a vicious cycle.

But how can one figure that out?

Isn't enforcement of a law (through an arrest for example) the first indication that a crime might have occurred?

If so, how to figure out if the 75% attribution is fair or just a result of selection/confirmation bias?

There are numerous reasonable proxies; locations of murders, reported assaults or robberies, calls to the police department reporting a possible crime, etc.

Is that what is actually being used, though?

I suspect there is no way for the public to know.

Reply on news.ycombinator.com