On the Freakonomics blog, Steven Pinker had this to say:
There are many statistical predictors of violence that we choose not to use in our decision-making for moral and political reasons, because the ideal of fairness trumps the ideal of cost-effectiveness. A rational decision-maker using Bayes’ theorem would say, for example, that one should convict a black defendant with less evidence than one needs with a white defendant, because these days the base rates for violence among blacks is higher. Thankfully, this rational policy would be seen as a moral abomination.
I’ve seen a common theme on LW that is more or less “if the consequences are awful, the reasoning probably wasn’t rational”. Where do you think Pinker’s analysis went wrong, if it did go wrong?
One possibility is that the utility function to be optimized in Pinker’s example amounts to “convict the guilty and acquit the innocent”, whereas we probably want to give weight to another consideration as well, such as “promote the kind of society I’d wish to live in”.
A rational decision-maker using Bayes’ theorem would say, for example, that one should convict a black defendant with less evidence than one needs with a white defendant, because these days the base rates for violence among blacks is higher.
One would compare black defendants with guilty black defendants and white defendants with guilty white defendants. It’s far from obvious that (guilty black defendants/black defendants) > (guilty white defendants/white defendants). Differing arrest rates, plea bargaining etc. would be factors.
Where do you think Pinker’s analysis went wrong, if it did go wrong?
He began a sentence by characterizing what a member of a group “would say”.
One would compare black defendants with guilty black defendants and white defendants with guilty white defendants. It’s far from obvious that (guilty black defendants/black defendants) > (guilty white defendants/white defendants). Differing arrest rates, plea bargaining etc. would be factors.
60% of convicts who have been exonerated through DNA testing are black. Whereas blacks make up 40% of inmates convicted of violent crimes. Obviously this is affected by the fact that “crimes where DNA evidence is available” does not equal “violent crimes”. But the proportion of inmates incarcerated for rape/sexual assault who are black is even smaller: ~33%. There are other confounding factors like which convicts received DNA testing for their crime. But it looks like a reasonable case can be made that the criminal justice system’s false positive rate is higher for blacks than whites. Of course, the false negative rate could be higher too. If cross-racial eyewitness identification is to blame for wrongful convictions then uncertain cross-racial eyewitnesses might cause wrongful acquittals.
Yes. It’s important to remember that guilty defendants aren’t the same thing as convicted defendants. A rational decision-maker using Bayes’ theorem wouldn’t necessarily put all that much weight on the decisions of past juries, knowing as we do that they’re not using Bayes’ theorem at all. And, of course, a Bayesian would need exactly the same amount of evidence to convict a black defendant as they did a white defendant. That question is whether skin colour counts as evidence.
The conviction rate for black defendents is sometimes much higher than the conviction rate for whites, so the solution you’ve suggested here would intensify the racial disparity.
If you instituted a policy to require less evidence to convict black defendants, you would convict more black defendants, which would make the measured “base rates for violence among blacks” go up, which would mean that you could need even less evidence to convict, which...
No, you’d just need to keep track of how often demographic considerations influenced the outcome, so that any measure of “base rates for violence among blacks” you use for such decisions is independent of the policy.
(That’s not to say that such a policy would be a good idea of course)
It’s not clear that “the base rates for violence among blacks is higher” is meant to be measured by convictions. I interpreted it to be based on sociological data, for example, and in that case there would be no feedback loop. Pinker didn’t cite a source, unfortunately. A very quick stroll past Google Scholar 12 shows that a common source used is arrest data in the FBI’s Uniform Crime Report. Plainly there are also important ways in which arrest data may be biased against blacks, but I’m hesitant to simply dismiss a finding based on that difficulty as I’m willing to bet that researchers in the field would have attempted to account for the difficulty.
The right kind of data would come from things like the National Crime Victimization Survey which collects data outside the criminal justice system. The base rate for the offender in a violent crime being black, is, according to that survey, lower than the probability that a given person arrested for a violent crime is black. So it looks to me that the evidence is long screened out by the time the case gets to the court room.
Right. I’m comparing particular categories of crime—reports from robbery victims to arrests for robbery, reports of rape to arrests for rape etc. I’m definitely not comparing total arrests for violent crimes to reports of violent crimes minus homicide.
I presumed as much, but that problem may still be noticeable: every rape that’s also a murder won’t get counted. I don’t know how frequently the various violent crimes are paired with murder, though, or how large the difference between reported victimization and arrests are.
Pinker didn’t address evidence screening off other evidence. Race would be rendered zero evidence in many cases, in particular in criminal cases for which there is approximately enough evidence to convict. I’m not exactly sure how often, I don’t know how much e.g. poverty, crime, and race coincide.
It is perhaps counterintuitive to think that Bayesian evidence can apparently be ignored, but of course it isn’t really being ignored, just carefully not double counted.
The problem isn’t using it as evidence. The problem is that it is extremely likely that humans will use such evidence in much greater proportion than is actually statistically justified. If juries were perfect Bayesians this wouldn’t be a problem.
Yes indeed. And the same goes, by the way, for other kinds of rational evidence that aren’t acceptable as legal evidence (hearsay, flawed forensics, coerced confessions, etc).
The more interesting question isn’t for the jury—for whom the race of a defendant has long been swamped by other evidence—but for a police officer deciding whether or not a person’s suspicious behavior is sufficient reason to stop and question them. Not only does including race in that assessment seem rational but it is something police officers almost certainly do (if not consciously) which makes it rather more interesting as a policy question.
Do you care to elaborate? The interpretation of your response that comes to my mind is that you dissent from the moral viewpoint that Pinker expresses.
I do not consider it laudable that, when someone makes a rational suggestion, it is seen as a moral abomination. If it’s a bad idea, there are rational ways to declare it a bad idea, and “moral abomination” is lazy. If it is a good idea, then “moral abomination” goes from laziness to villainy.
If his argument is “this causes a self-fulfilling prophecy, because we will convict blacks and not convict Asians because blacks are convicted more and Asians convicted less, suggesting that we will over-bias ourselves,” then he’s right that this policy is problematic. If his argument is “we can’t admit that blacks are more likely to commit crimes because that would make us terrible people,” then I don’t want any part of it. Since he labeled it a moral abomination, that suggests the latter rather than the former.
The latter is probably not his intended meaning given that he states “these days the base rates for violence among blacks is higher.”
I think calling something a “moral abomination” means it directly conflicts with your values, rather than only being a “bad idea.” For example, lying may be a bad idea but probably not a moral abomination to a consequentialist whereas killing the healthiest humans to reduce overpopulation would not only be a bad idea because it would be killing off the workforce, it directly conflicts with our value against killing people.
The laziness in calling something a “moral abomination” is failing to specify what value it is conflicting with. Of course, having such a complex, context-dependent, and poorly objectively defined value as “non-discrimination” might be unfashionable to some.
The latter is probably not his intended meaning given that he states “these days the base rates for violence among blacks is higher.”
Those are the words he puts in the mouth of “a rational decision-maker using Bayes Theorem,” whose conclusion he identifies as a moral abomination. It is ambiguous whether or not he thinks that belief should pay rent.
I think calling something a “moral abomination” means it directly conflicts with your values, rather than only being a “bad idea.”
The purpose of indignation is not to make calculations easier, but to avoid calculations.
I think he’s just saying that not all rational evidence should be legal evidence. I don’t think that he should be read according to LW conventions when he calls lower evidence standards for blacks a “rational policy”. He doesn’t mean to say that it would be rational to institute this policy (and yet somehow also morally abominable). He means that institutionalizing Bayesian epistemology in this way would be morally abominable (and hence not rational, as folks around here use the term).
I think he’s just saying that not all rational evidence should be legal evidence.
Sure; in which case calling it a moral abomination is laziness. (The justification for holding legal evidence to a higher standard is very close to the self-fulfilling prophecy argument.)
It’s already been pointed out that being a member of a group is evidence, so the evidence standards are identical. This is important because some evidence screens off other evidence.
The problem with our conversation is that Pinker’s argument is so wrong, with so many errors sufficient to invalidate it, that we are having trouble inferring which sub-components of it he was right about. I encourage moving on from what he meant to what the right way to think is.
One (more) reason to be uncomfortable with such an argument: “black” doesn’t carve nature at its joints.
(Whereas, relevantly for such arguments, “poor” does—though I believe that arguing that way leads down the path that has been called “reference class tennis”.)
When it comes to US demographics, “black” covers a “natural” cluster of the population / identifiable blob in thingspace. Sure, there are border cases like mixed-race people and recent African immigrants, just like there are edge-cases between bleggs and rubes. “Is person X black or not?” is probably one of the top yes/no questions that would tell you the most about an American (Along with “Did he vote for Obama?”, “Is he richer or poorer than the median?”, or “Does he live north or south of the Mason-Dixon line?”)
Sure, when it comes to world demographics, or Brazilian demographics, it doesn’t cut reality at it joints as well.
For questions of, say, population genetics, I think that is quite right. But for questions of sociology or social policy I don’t see why one wouldn’t include ‘black’ as part of the ontology.
That’s not too important. If I go to my closet and pull out twenty items of clothing at random, and designate those group A, and designate the rest group B, if I know what is in each group I can still make predictions about traits of random members of either group.
Why single out race? There are other demographic factors that could count too: sex, age, social class … and if a policy of “OK, conviction standards for blacks are lower” would be political suicide, a policy of “OK, conviction standards for working class people” would be even worse.
Even so such policies may indeed increase the accuracy of convictions, 1) they don’t match our intuitions about justice, which I suspect would make people less happy, 2) judges and juries already take such factors into account (implicitly and approximately), so there’s a risk of overcorrecting, and 3) energy would be much better spent increasing the accuracy of conviction with less ambiguous things like cameras, DNA tests, etc.
I do believe that such demographic data can be useful to help direct resources for crime prevention, though.
There are other demographic factors that could count too: sex, age, social class … and if a policy of “OK, conviction standards for blacks are lower” would be political suicide
Whatever works, I don’t have any specific policies in mind (I’m far from being an expert in law enforcement).
But to take a specific example, I don’t think information about higher crime rates for blacks is enough to tell whether we need “increased police presence in the Ghetto”—for all I know, police presence could already be 10 times the national average there.
There is a tendency I dislike in political punditry/activism/whatever (not that I’m accusing you of it, you just gave me a pretext to get on my soap box) to say “we need more X” or “we need less X” (regulation, police, taxes, immigrants, education, whatever) without any reference to evidence about what the ideal level of X would be, and about whether we are above or below that level—sometimes the same claims are made in countries with wildly different levels of X.
“Thankfully” part is wrong. We don’t use any explicit probability thresholds to judge people guilty or not, we rely on judge’s gut feeling about the defendant, which is very likely even more biased.
With a serious probability threshold being black would count slightly against you, but it would be very small bias.
Where do you think Pinker’s analysis went wrong, if it did go wrong?
Pinker is not distinguishing between members of the audience and members of the jury.
For the audience, Pinker is basically correct. The prior probability that the defendant committed the acts charged is fairly high, given the evidence available to the audience that the government charged the defendant with a crime.
But the jury is given an instruction by the judge that the defendant is entitled to a presumption of innocence that must be overcome by the government. In Bayesian terms the judge means, “For social reasons not directly related to evidence, set your prior probability that the defendant is guilty as close to zero as is logically possible for you. If the probability of guilt is sufficiently high after hearing all the evidence, then you may vote to convict.”
If a rationalist juror ignores this instruction and acts like a rational audience member, then the juror is doing wrong, as Pinker notes. But ignoring the judge’s instruction is wrong, and Pinker is not presenting a great insight by essentially highlighting this point.
There shouldn’t be any such distinction. The audience (I assume you mean the courtroom audience) should reason the exact same way the jury does.
The prosecution is required to make an explicit presentation of the evidence for guilt, so that the mere fact that charges were brought is screened off. As a consequence, failure to present a convincing explicit case is strong evidence of innocence; prosecutors have no incentive to hide evidence of guilt! Hence any juror or audience member who reasons “the prosecution’s case as presented is weak, but the defendant has a high likelihood of guilt just because they suspect him” is almost certainly committing a Bayesian error. (Indeed, this is how information cascades are formed.)
See Argument Screens Off Authority: once in the courtroom, prosecutors have to present their arguments, which renders their “authority” worthless.
I wrote this long post defending my point, and about halfway through, I realize it was mostly wrong. I think the screening off point is probably a better description of what’s wrong with Pinker’s analysis. Specifically, the higher rate of crime among blacks should be screened off from consideration by the fact that this particular black defendant was charged.
To elaborate on my earlier point, the presumption of innocence also serves to remind the juror that the propensity of the population to commit crimes is screened off by the fact that this particular person went through the arrest and prosecution screening processes in order to arrive in the position of defendant. In other words, a rationalist should not use less evidence to convict black defendants than required for white defendants because this is double counting the crime rate of blacks.
On the Freakonomics blog, Steven Pinker had this to say:
I’ve seen a common theme on LW that is more or less “if the consequences are awful, the reasoning probably wasn’t rational”. Where do you think Pinker’s analysis went wrong, if it did go wrong?
One possibility is that the utility function to be optimized in Pinker’s example amounts to “convict the guilty and acquit the innocent”, whereas we probably want to give weight to another consideration as well, such as “promote the kind of society I’d wish to live in”.
One would compare black defendants with guilty black defendants and white defendants with guilty white defendants. It’s far from obvious that (guilty black defendants/black defendants) > (guilty white defendants/white defendants). Differing arrest rates, plea bargaining etc. would be factors.
He began a sentence by characterizing what a member of a group “would say”.
60% of convicts who have been exonerated through DNA testing are black. Whereas blacks make up 40% of inmates convicted of violent crimes. Obviously this is affected by the fact that “crimes where DNA evidence is available” does not equal “violent crimes”. But the proportion of inmates incarcerated for rape/sexual assault who are black is even smaller: ~33%. There are other confounding factors like which convicts received DNA testing for their crime. But it looks like a reasonable case can be made that the criminal justice system’s false positive rate is higher for blacks than whites. Of course, the false negative rate could be higher too. If cross-racial eyewitness identification is to blame for wrongful convictions then uncertain cross-racial eyewitnesses might cause wrongful acquittals.
Yes. It’s important to remember that guilty defendants aren’t the same thing as convicted defendants. A rational decision-maker using Bayes’ theorem wouldn’t necessarily put all that much weight on the decisions of past juries, knowing as we do that they’re not using Bayes’ theorem at all. And, of course, a Bayesian would need exactly the same amount of evidence to convict a black defendant as they did a white defendant. That question is whether skin colour counts as evidence.
The conviction rate for black defendents is sometimes much higher than the conviction rate for whites, so the solution you’ve suggested here would intensify the racial disparity.
If I suggest reconciliatory solutions rather than try to just delineate mere unfair reality, then let my car’s brake pedal fail half the time, and let my car’s gas pedal work perfectly!
If you instituted a policy to require less evidence to convict black defendants, you would convict more black defendants, which would make the measured “base rates for violence among blacks” go up, which would mean that you could need even less evidence to convict, which...
No, you’d just need to keep track of how often demographic considerations influenced the outcome, so that any measure of “base rates for violence among blacks” you use for such decisions is independent of the policy.
(That’s not to say that such a policy would be a good idea of course)
It’s not clear that “the base rates for violence among blacks is higher” is meant to be measured by convictions. I interpreted it to be based on sociological data, for example, and in that case there would be no feedback loop. Pinker didn’t cite a source, unfortunately. A very quick stroll past Google Scholar 1 2 shows that a common source used is arrest data in the FBI’s Uniform Crime Report. Plainly there are also important ways in which arrest data may be biased against blacks, but I’m hesitant to simply dismiss a finding based on that difficulty as I’m willing to bet that researchers in the field would have attempted to account for the difficulty.
The right kind of data would come from things like the National Crime Victimization Survey which collects data outside the criminal justice system. The base rate for the offender in a violent crime being black, is, according to that survey, lower than the probability that a given person arrested for a violent crime is black. So it looks to me that the evidence is long screened out by the time the case gets to the court room.
Note that the NCVS requires that the victims survive, and does not collect data on crimes like murder, which may cause systematic differences.
Right. I’m comparing particular categories of crime—reports from robbery victims to arrests for robbery, reports of rape to arrests for rape etc. I’m definitely not comparing total arrests for violent crimes to reports of violent crimes minus homicide.
I presumed as much, but that problem may still be noticeable: every rape that’s also a murder won’t get counted. I don’t know how frequently the various violent crimes are paired with murder, though, or how large the difference between reported victimization and arrests are.
Good point.
Pinker didn’t address evidence screening off other evidence. Race would be rendered zero evidence in many cases, in particular in criminal cases for which there is approximately enough evidence to convict. I’m not exactly sure how often, I don’t know how much e.g. poverty, crime, and race coincide.
It is perhaps counterintuitive to think that Bayesian evidence can apparently be ignored, but of course it isn’t really being ignored, just carefully not double counted.
The percentage arrestees who are black is higher than the percentage of offenders who are black as reported by victims in surveys.
(Suggesting the base rate is screened out by the time the matter gets to the court room)
The problem isn’t using it as evidence. The problem is that it is extremely likely that humans will use such evidence in much greater proportion than is actually statistically justified. If juries were perfect Bayesians this wouldn’t be a problem.
Yes indeed. And the same goes, by the way, for other kinds of rational evidence that aren’t acceptable as legal evidence (hearsay, flawed forensics, coerced confessions, etc).
The more interesting question isn’t for the jury—for whom the race of a defendant has long been swamped by other evidence—but for a police officer deciding whether or not a person’s suspicious behavior is sufficient reason to stop and question them. Not only does including race in that assessment seem rational but it is something police officers almost certainly do (if not consciously) which makes it rather more interesting as a policy question.
The word “thankfully.”
Do you care to elaborate? The interpretation of your response that comes to my mind is that you dissent from the moral viewpoint that Pinker expresses.
I do not consider it laudable that, when someone makes a rational suggestion, it is seen as a moral abomination. If it’s a bad idea, there are rational ways to declare it a bad idea, and “moral abomination” is lazy. If it is a good idea, then “moral abomination” goes from laziness to villainy.
If his argument is “this causes a self-fulfilling prophecy, because we will convict blacks and not convict Asians because blacks are convicted more and Asians convicted less, suggesting that we will over-bias ourselves,” then he’s right that this policy is problematic. If his argument is “we can’t admit that blacks are more likely to commit crimes because that would make us terrible people,” then I don’t want any part of it. Since he labeled it a moral abomination, that suggests the latter rather than the former.
The latter is probably not his intended meaning given that he states “these days the base rates for violence among blacks is higher.”
I think calling something a “moral abomination” means it directly conflicts with your values, rather than only being a “bad idea.” For example, lying may be a bad idea but probably not a moral abomination to a consequentialist whereas killing the healthiest humans to reduce overpopulation would not only be a bad idea because it would be killing off the workforce, it directly conflicts with our value against killing people.
The laziness in calling something a “moral abomination” is failing to specify what value it is conflicting with. Of course, having such a complex, context-dependent, and poorly objectively defined value as “non-discrimination” might be unfashionable to some.
Those are the words he puts in the mouth of “a rational decision-maker using Bayes Theorem,” whose conclusion he identifies as a moral abomination. It is ambiguous whether or not he thinks that belief should pay rent.
The purpose of indignation is not to make calculations easier, but to avoid calculations.
I think he’s just saying that not all rational evidence should be legal evidence. I don’t think that he should be read according to LW conventions when he calls lower evidence standards for blacks a “rational policy”. He doesn’t mean to say that it would be rational to institute this policy (and yet somehow also morally abominable). He means that institutionalizing Bayesian epistemology in this way would be morally abominable (and hence not rational, as folks around here use the term).
Sure; in which case calling it a moral abomination is laziness. (The justification for holding legal evidence to a higher standard is very close to the self-fulfilling prophecy argument.)
It’s already been pointed out that being a member of a group is evidence, so the evidence standards are identical. This is important because some evidence screens off other evidence.
The problem with our conversation is that Pinker’s argument is so wrong, with so many errors sufficient to invalidate it, that we are having trouble inferring which sub-components of it he was right about. I encourage moving on from what he meant to what the right way to think is.
One (more) reason to be uncomfortable with such an argument: “black” doesn’t carve nature at its joints.
(Whereas, relevantly for such arguments, “poor” does—though I believe that arguing that way leads down the path that has been called “reference class tennis”.)
Doesn’t it?
When it comes to US demographics, “black” covers a “natural” cluster of the population / identifiable blob in thingspace. Sure, there are border cases like mixed-race people and recent African immigrants, just like there are edge-cases between bleggs and rubes. “Is person X black or not?” is probably one of the top yes/no questions that would tell you the most about an American (Along with “Did he vote for Obama?”, “Is he richer or poorer than the median?”, or “Does he live north or south of the Mason-Dixon line?”)
Sure, when it comes to world demographics, or Brazilian demographics, it doesn’t cut reality at it joints as well.
It’s Mason-Dixon, after the two surveyors.
Whoops, thanks!
For questions of, say, population genetics, I think that is quite right. But for questions of sociology or social policy I don’t see why one wouldn’t include ‘black’ as part of the ontology.
That’s not too important. If I go to my closet and pull out twenty items of clothing at random, and designate those group A, and designate the rest group B, if I know what is in each group I can still make predictions about traits of random members of either group.
Why single out race? There are other demographic factors that could count too: sex, age, social class … and if a policy of “OK, conviction standards for blacks are lower” would be political suicide, a policy of “OK, conviction standards for working class people” would be even worse.
Even so such policies may indeed increase the accuracy of convictions, 1) they don’t match our intuitions about justice, which I suspect would make people less happy, 2) judges and juries already take such factors into account (implicitly and approximately), so there’s a risk of overcorrecting, and 3) energy would be much better spent increasing the accuracy of conviction with less ambiguous things like cameras, DNA tests, etc.
I do believe that such demographic data can be useful to help direct resources for crime prevention, though.
Isn’t that the status quo?
Not explicitly, as far as I know (implicitly, possibly, hence what I said about over-correcting).
One would hope not.
Leaning more towards “increased police presence in the ghetto” or “only frisk the Muslims”?
Whatever works, I don’t have any specific policies in mind (I’m far from being an expert in law enforcement).
But to take a specific example, I don’t think information about higher crime rates for blacks is enough to tell whether we need “increased police presence in the Ghetto”—for all I know, police presence could already be 10 times the national average there.
There is a tendency I dislike in political punditry/activism/whatever (not that I’m accusing you of it, you just gave me a pretext to get on my soap box) to say “we need more X” or “we need less X” (regulation, police, taxes, immigrants, education, whatever) without any reference to evidence about what the ideal level of X would be, and about whether we are above or below that level—sometimes the same claims are made in countries with wildly different levels of X.
Single-valued logic
“Thankfully” part is wrong. We don’t use any explicit probability thresholds to judge people guilty or not, we rely on judge’s gut feeling about the defendant, which is very likely even more biased.
With a serious probability threshold being black would count slightly against you, but it would be very small bias.
Pinker is not distinguishing between members of the audience and members of the jury.
For the audience, Pinker is basically correct. The prior probability that the defendant committed the acts charged is fairly high, given the evidence available to the audience that the government charged the defendant with a crime.
But the jury is given an instruction by the judge that the defendant is entitled to a presumption of innocence that must be overcome by the government. In Bayesian terms the judge means, “For social reasons not directly related to evidence, set your prior probability that the defendant is guilty as close to zero as is logically possible for you. If the probability of guilt is sufficiently high after hearing all the evidence, then you may vote to convict.”
If a rationalist juror ignores this instruction and acts like a rational audience member, then the juror is doing wrong, as Pinker notes. But ignoring the judge’s instruction is wrong, and Pinker is not presenting a great insight by essentially highlighting this point.
There shouldn’t be any such distinction. The audience (I assume you mean the courtroom audience) should reason the exact same way the jury does.
The prosecution is required to make an explicit presentation of the evidence for guilt, so that the mere fact that charges were brought is screened off. As a consequence, failure to present a convincing explicit case is strong evidence of innocence; prosecutors have no incentive to hide evidence of guilt! Hence any juror or audience member who reasons “the prosecution’s case as presented is weak, but the defendant has a high likelihood of guilt just because they suspect him” is almost certainly committing a Bayesian error. (Indeed, this is how information cascades are formed.)
See Argument Screens Off Authority: once in the courtroom, prosecutors have to present their arguments, which renders their “authority” worthless.
I wrote this long post defending my point, and about halfway through, I realize it was mostly wrong. I think the screening off point is probably a better description of what’s wrong with Pinker’s analysis. Specifically, the higher rate of crime among blacks should be screened off from consideration by the fact that this particular black defendant was charged.
To elaborate on my earlier point, the presumption of innocence also serves to remind the juror that the propensity of the population to commit crimes is screened off by the fact that this particular person went through the arrest and prosecution screening processes in order to arrive in the position of defendant. In other words, a rationalist should not use less evidence to convict black defendants than required for white defendants because this is double counting the crime rate of blacks.
Could you elaborate? Consequences of what (related to reasoning how)?