This site makes extensive use of JavaScript.
Please enable JavaScript in your browser.
Live
PTR
10.2.7
PTR
10.2.6
Beta
Racism
Post Reply
Return to board index
Post by
Squishalot
Mytie, there are many things wrong with racial profiling. Here are a few, in broad strokes:
1. Most of what's termed racial profiling is unconstitutional.
2. It's not an effective way to police. It may be that more murders in NYC are committed by blacks than whites, but that's clearly a correlation, not a causation. There are better indicators, and police watching for gang affiliation or behavior patterns consisent with drug dealing will be more effective than police looking at race.
(Note: The fact that the NYC frisk law has lowered crime rates does not imply that profiling is an effective way to decide whom to frisk. The result of that law is that the police frisk more people with less cause - that's going to affect crime rates even if you choose people to frisk at random.)
3. It induces bias. Introducing racial profiling into enforcement policy leads to more false positives for the race in question and more false negatives elsewhere. It's also worth noting that this bias can be used - the more racial profiling you do, the easier it is for, e.g. gangs to transport drugs with a white courier or terrorists to use a white convert for their attack.
4. It's discriminatory and unfair. It sounds trite but it's true, it would probably feel more true if it was you getting searched by the police in front of your kids once a week just for going to the store for milk.
I posted something in another thread where this was brought up. The article linked at the time showed that approximately 10% of people who were stopped and frisked were then charged with various offences (illegal weapon possession, drug possession / dealing, etc.). Now, unless you're going to suggest that 1 in 10 people on the street are likely to be charged with a criminal offence, then the profiling for potential stop-and-frisk targets, however it's done, has managed to identify people more likely than the general population to offend.
Irrespective of what we think about racial profiling, we do need to bear this in mind. Of course, we can't completely exclude parts of society from being targets, but we want any sort of preventative measure to be generally effective, don't we? To use an extreme example, if 1 in 100 people without criminal tendencies get pulled in for a frisk, I'd like to hope that the proportion of criminals getting pulled in is much higher.
I'm a huge supporter of analytics and the methods used, and if race happens to be a strongly performing variable in the equation, then I would have no qualms using it
in the absence of other information
. In relation to your point about gang affiliation and drug dealing patterns, I have no doubt that the police are using those as well, considering the relatively high hit rate.
Including more data in a profile would be better if the subsequent analysis was done correctly, but in the real world it's not - cops walking the streets are (in general) using more instinct than Bayes' Theorem. Of course overestimating the strength of a correlation would be fine if false positives didn't matter - and if you're not the one being profiled, maybe they don't. But the thesis of the article you linked was that to others, they do.
I'd like to think that the cops walking the streets are generally briefed on what key characteristics of potential criminals to look out for. Perhaps I'm wrong, but I don't think the PDs are *that* blind to the world.
Post by
MyTie
1. Most of what's termed racial profiling is unconstitutional.Where in the constitution is it illegal to take race into account when making a profile of a person.
2. It's not an effective way to police. It may be that more murders in NYC are committed by blacks than whites, but that's clearly a correlation, not a causation. There are better indicators, and police watching for gang affiliation or behavior patterns consisent with drug dealing will be more effective than police looking at race.
(Note: The fact that the NYC frisk law has lowered crime rates does not imply that profiling is an effective way to decide whom to frisk. The result of that law is that the police frisk more people with less cause - that's going to affect crime rates even if you choose people to frisk at random.)It is effective, statistically. If we steralize the argument of emotion, and look at the math, if a certain race is statistically more likely to perpetuate a crime, and that race is targeted by law enforcement more than other races, the law enforcement will be more effective than if it were evenly applied. To say "it isn't effective", I'd like to know how you arrived at that conclusion.
3. It induces bias. Introducing racial profiling into enforcement policy leads to more false positives for the race in question and more false negatives elsewhere. It's also worth noting that this bias can be used - the more racial profiling you do, the easier it is for, e.g. gangs to transport drugs with a white courier or terrorists to use a white convert for their attack.I'm not saying that only blacks should be stopped, but that race should be a factor when making a profile. And your insistance that it is biased is true, but it is an accurate and mathematically effective bias.
4. It's discriminatory and unfair. It sounds trite but it's true, it would probably feel more true if it was you getting searched by the police in front of your kids once a week just for going to the store for milk.It is fair. It is fair that a certain race is crime prone and so therefore gets more attention from law enforcement. That's fair. It would be unfair to apply the same law enforcement to everyone regardless of real indicators, which INCLUDES BUT IS NOT LIMITED TO race.
Post by
168916
This post was from a user who has deleted their account.
Post by
MyTie
Firstly you are conflating causation with correlationNot at all. I never said anything about correlation, nor what causes what. I said that if you apply a remedy to a statistically more likely situation in which the problem exists, that problem will be statistically more likely to find the remedy. That has nothing to do with correlation or causation. I actually don't know what you are talking about here. Seriously... what., and secondly I explained how I arrived at that conclusion. Adding race into profiling induces unnecessary bias because people are not good at dealing with conditional probability (even people who ought to be, like doctors - studies have shown as much).source. Further, even though conditional probability may not be perfect, it is used effectively in many professions, especially docs. There are medical conditions that affect blacks more than whites, and whites more than blacks, and docs take race into consideration when finding the most likely problem. It isn't their sole source of diagnosis, but to say it is "unnecessarily biased" is... well... could you back up why it is unnecessary to consider race issues when addressing problems that disproportionately affect certain races?
Post by
557473
This post was from a user who has deleted their account.
Post by
MyTie
P.S. MyTie, you profile pic horrifies and distracts me. WTF is it?
It's a bearded toothy rooster, gleefully laughing at my debate opponents. Remember that next time you're arguing with me. That horrifying rooster is watching you.
Post by
168916
This post was from a user who has deleted their account.
Post by
MyTie
You suggested that certain races are "crime prone" and that we can extrapolate from that. You are describing a correlation, but treating it as if there was causation involved.I never said that being black causes a person to commit a crime. What causation are you pointing to? I'm saying there is a correlation, and that correlation provides a statistical advantage to applying a very effective bias.The point of it all is that even if you knew that 90% of all crimes in a neighborhood were committed by blacks, if you then went there and saw a white guy and a black guy and thought you knew which of them was more likely to be a criminal, you'd be incorrect.No, I would be correct. I would be incorrect if I said I knew which of them WAS a criminal. If you go to Arizona, and there is a blonde haired blue eyed German, and a latino spanish speaking Mexican, and you know that one of them is an illegal immigrant, which one is
more likely
to be the illegal? As in, it is more probable. No one is trying to offer causation proof of race based crime.
Post by
168916
This post was from a user who has deleted their account.
Post by
Squishalot
Sure, why not? I don't know any data to that effect, but it'd be a damned stupid program if they aren't managing to do better than a coin flip.
Indeed, which is why I think it's 'successful' in that respect, in the same way that credit scorecards from banks are generally successful in discriminating against high risk applicants.
But I don't know what you mean this to establish. Are you suggesting that the NYC program is indeed profiling by race? They say they aren't, and if that's true does it not argue against your point?
I'm suggesting that it doesn't matter so much how they're profiling, providing that they're a) getting better-than-chance results; b) that they're getting reasonably strong results that are going to be hard to improve on; and c) they're obtaining a wide enough sample to feed more data back into their analysis. If race is a part of that profile, then I see nothing wrong with that, because it's effective. When NYPD say they aren't, I believe what they're saying is "we aren't specifically targeting black people", not "our profiling tool doesn't rank black people high on the suspicion list".
I'd be highly concerned if they were getting 100% black people, for example, no matter how strong their analytics model predicts (black = crime, white = no crime), because they need to get fresh data for their predictive models. However, that's not the case - it's not as if all 600k+ people they've pulled over on the street are non-white folks.
Post by
168916
This post was from a user who has deleted their account.
Post by
Squishalot
But using this program as a background to discuss racial profiling doesn't really make sense, given that the police say they aren't doing it and we can surmise that the program would lower crime rates with or without it.
Happy to move on from it - I'm just discussing it because it came up.
(Incidentally, the support you've offered here is begging the question - you are in essence saying "I see nothing wrong with it so long as it's effective", when whether it's effective is what's disputed.)
But it is effective, as you've suggested you agree with - it's more likely to uncover criminals than blind chance is.
This is another problem with racial profiling I didn't mention - the more you do it, the more your data will appear to justify it.
This statement is generally incorrect. If you only sample people of a particular race, you simply won't have the data to justify it, because there won't be enough race data for it to power your analysis. For race to be a significant variable in a scorecard, you need to have a mix of inputs to help your program determine the type 1 / type 2 errors. If you end up profiling 99% black people and 1% white, race will inevitably disappear from your profiling algorithm because the program won't have a big enough sample of white people to determine if race is significant or not, therefore, the profiling will no longer be justified.
While you have, say, a 80/20 split, there will be sufficient data, and the raw stats will tell you whether it's justified or not. They won't be biased one way or another, for the same Bayesian probability reasons you're trying to teach MyTie about.
Post by
168916
This post was from a user who has deleted their account.
Post by
Squishalot
What is, racial profiling or the NYC program? You seem to be referring to the former but what I agreed about was the latter.
I'm talking about the NYC program overall, of which racial profiling may or may not be a part of. But anyway...
Either way, the measure of whether racial profiling is effective isn't whether it beats blind chance, but whether police are more accurate with it or without it. (Presumably they can beat blind chance either way.) Even if we knew that the NYC program used racial profiling, and agreed it was effective, that wouldn't tell us what removing the profiling element would do to the effectiveness.
If we're going to look purely at race as a variable in a criminal profile, then yes. Arguably though, the addition of a variable will only provide more power to an analysis, so the NYC program including race as a variable would (and should) be at least as powerful as one excluding race as a variable.
You're assuming a particular (and statistically valid) analysis which isn't in evidence... Of course if the officers of MPD are carrying around a phone app that correctly applies multivariate analysis (which would be an interesting experiment) it would be a different story
Likewise, you're assuming statistical analysis isn't being done. If we're starting with the assumption that the officers are going to be working on broad brush 'statistics' (such as your 'blacks are responsible for 70% of the crime' comment), I would agree, but I don't think that it's a reasonable assumption.
I'm assuming that the analysis is done by stats and actuarial majors, with the information being fed to the PDs. The PDs will then provide the general characteristics (i.e. the stereotype) to the actual officers who'll be patrolling the streets. The stereotype that gets provided will be statistically accurate. Whether the actual officers patrol that stereotype or not, or whether they work on gut feel, is entirely different.
On a side note - it's not that hard to carry a phone app for it. All you need is a scorecard that's been generated by your head office, and to plug in the characteristics you observe to derive a probability of criminal behaviour. You can't do any sort of analysis on the street with single observations. What I would actually do is ask them to plug the data in and obtain a probability, investigate the guy (in that order, so your variable recording isn't biased by your known result), then record whether it was good/bad, so as to feed more data back into their investigative database.
Post by
168916
This post was from a user who has deleted their account.
Post by
Squishalot
Have you ever read about why they use decision trees for certain kinds of medical diagnoses instead of taking a full history and weighing the various factors? It's true that adding extra data to a model doesn't generally hurt its predictive power but that is emphatically not true for people.
The reason is because 99% of the time, we don't have sufficient histories to be able to come to anything conclusive about their impact on a problem, or that the histories aren't actually significantly correlated, relative to the more important variables. This is why we do take certain parts of a patient's history that are immediately relevant to the issue in question. (That, and if we believe Dr. House, everybody lies.)
I don't think it's because the data over-fits, I believe it's more because there's too much noise in the data, and that the data quality is garbage in the first instance. When I was working in credit scoring, we actually ignored the 'income' variable for our small-business customers, because lenders would often just shove '$1' into the field during the application process, irrespective of what the business owner actually earned.
Of course, this all assumes that race isn't a significant factor in the first instance, in our hypothetical.
It doesn't resemble anything anyone has ever done, so far as I know, or what people mean when they talk about racial profiling taking place. I mean, being a mathlete I'm sympathetic to what you're suggesting, but you'd need to propose that police forces start profiling before you argue that they should start racially profiling.
I'm not convinced that they don't profile. We know what the classic stereotypes are, people have mentioned them in the last page or two. I don't think it's unbelievable for the cops to have better versions of those. Of course, I don't know anyone in the police force well enough to really have a clue about their process, but I can't imagine that they wouldn't have an idea of what a criminal presents like.
And more broadly, it's worth noting that the point we're discussing here isn't really part of "the profiling debate", as it were. In typical cases you have the civil-liberties side claiming that profiling shouldn't occur but has, while the law-enforcement side claims that profiling shouldn't occur and hasn't. That's certainly the case with the NYC program. The point being that both sides start from the assumption that racial profiling shouldn't be done (on purely discrimination/constitutional grounds, AFAIK), whether it's effective or not.
I'm not sure I agree with that interpretation. I think that the law-enforcement side is claiming that profiling
purely on race
shouldn't occur and hasn't. Profiling credit purely on education shouldn't occur either, but that doesn't mean that education shouldn't be a relevant variable in a credit profile. I'm curious to see anyone from a law enforcement agency who would be willing to say that profiling (on anything) shouldn't be done.
Post by
168916
This post was from a user who has deleted their account.
Post by
gamerunknown
There are medical conditions that affect blacks more than whites, and whites more than blacks, and docs take race into consideration when finding the most likely problem.
Using "
race
" as a general term rather than doing an adequate history of a patient is much more prone to error though, because certain population clusters that "pass" as white may be more susceptible, etc.
As for airport profiling, relying heavily on predictors will be exploited for the purpose of criminality. Once feedback enters the public conscious of a particular profile airport security are looking for, terrorists will recruit individuals outside of that profile.
Post by
Squishalot
The moral of which is, in asking whether racial profiling is effective your question should not be whether race has any predictive power but whether it is among the most predictive features available. I've never heard any argument that it is, personally.
The irony of the medical argument, I realised over dinner, is that race is one of the first things that doctors look at when determining the possibility of certain prognoses.
Well, fitting to noise is essentially what overfitting means. But I don't really understand your example - if the "income" feature was untrustworthy, shouldn't your model have automatically assigned it a low weight? I'm assuming you're talking about some kind of regression analysis. Or is assigning a low weight what you mean by "we ignored it"?
I use a different definition of overfitting to you then. When I think about overfitting, I think about something that was well correlated in the past, but due to population change / different circumstances / other, is no longer well correlated. Noisy data generally won't have been well correlated in the past, and therefore won't have been fitted in the first instance.
I meant that nobody is profiling the way you described, with guys back at HQ doing multivariate regression and the people in the field trusting the model over their instincts (which is what statistical analysis demands, after all).
I don't think that people in the field are trusting 'the model' over their instincts - I do think that if they're given a profile of a person, they'll look for someone matching that profile. They're not going to be told that a carjacker was a white female, then hit the streets and pull over all the black guys they see, for example. (Extreme case, but the point stands.)
Sorry, that whole last paragraph should have referred specifically to "the racial profiling debate". Whether or how other kinds of profiling occur probably warrants debate but I can't recall having seen the matter discussed.
My statement still stands - I think the law enforcement agencies are saying that we shouldn't profile purely on race, rather than saying that we shouldn't include race in profiles. I agree with the former - we shouldn't profile purely on race - but I'm still going to use race in my scorecards if it's appropriate to.
As for airport profiling, relying heavily on predictors will be exploited for the purpose of criminality. Once feedback enters the public conscious of a particular profile airport security are looking for, terrorists will recruit individuals outside of that profile.
You mean like that study showing that the most predictive sign of a suicide bomber is a lack of life insurance?
Having worked in analytics, I can safely say that people tend to have a very bad idea of how profiles are calculated, and so it's hard to 'game' the system. If you can explain to me why a male widow is a better credit risk than a single male person, I'll be highly impressed. In the case of insurance, for example, companies generate automatic scripts and spam websites with every combination of inputs to try to reverse-engineer how the companies profile. Without that level of immediate feedback, I don't think we'll ever know how airports truly profile.(##RESPBREAK##)8##DELIM##Squishalot##DELIM##
Post by
ElhonnaDS
This may have been addressed, but since correlation doesn't prove causation, why does the fact that a certain percentage of people who are frisked under this happen to be a minority mean that there is racial profiling?
This may be a little devil's advocate, since I agree that racial profiling is not a good idea in general, but here's a possibility. If we know, based on arrest and conviction rates, murder rates, drug saturation, census info about income, etc. that many low income neighborhoods have high crime rates, and are inhabited by a much higher than average percentage of people of minority backgrounds, then those are hard numbers.
If the police were targeting high crime areas, and were looking for specific colors of clothing associated with gangs, people hanging out late at night or in business locations after the business has closed, then those would be acceptable ways to look for people engaged in criminal activity. If a high percentage of those people happen to be minority, does that then automatically mean that cops were looking for people of a specific race, or could it mean that the neighborhoods where there are higher crime rates happen to have a larger minority community.
I'm just saying that the causation vs. correlation works both ways. I agree that just because a larger percentage of certain ethnicities, per capita, commit a crime, is no reason to suspect particular people just because they happen to be of that ethnicity, and certainly not an indicator that that ethnicity has any inherent inclinations towards that. What I am asking is why similar logic is summarily discarded in the case where a high percentage of people who are frisked are of that ethnicity, but those numbers could ALSO be explained by cops looking in high crime areas, for specific gang colors (many gangs DO recruit within a single or certain ethnicities), etc?
Post Reply
You are not logged in. Please
log in
to post a reply or
register
if you don't already have an account.