“Anyone who accurately observes their environment or is familiar with the relevant data will risk forming the forbidden implicit associations.”

Until then, I only suspected that he was just an ideologue attacking good science.

Consider President-Elect Obama’s attribution of high rates of African-American incarceration to bias in the prosecution, conviction, and sentencing processes within the justice system. In fact, criminologists have worked very hard for decades to find evidence of such bias and failed: differences in incarceration are best explained by differences in crimes. Changing what counts as a crime, e.g. ending drug prohibition, or changing the root causes of crime might help, but Obama’s claim was wrong and distracted from the actual problems. Yet broadcast and print media almost completely failed to communicate this fact, and indeed almost never mention the relevant statistics.

This sort of misconception can result in bad policy, e.g. police departments that allocate policing resources and arrests based on levels of criminal activity are accused of racism, and redistribute their efforts in a way that is less efficient at preventing murders. When police stop African-American drivers more often than other groups, but those stops result in more (proportionally) crimes discovered (e.g. drug or illegal weapon possession), they are directed to change their stopping patterns to less effective ones. When ‘white flight’ took the tax base out of urban cores, neglect of the importance of crime in favor of racism-based explanations mislead about what could reverse the flight (and the decline of crime has led to renewal of urban areas).

Thus, even if we would be wise to restrict emotional images of individual criminals, it is important that dry and sober discussion of crime statistics take place. That does not happen on CBS or Fox News, in the New York Times or the Wall Street Journal, which is why the DOJ statistics are so shocking to anyone drawing solely on our national mass media for information:
http://www.ojp.usdoj.gov/bjs/homicide/race.htm

The most reliable crime statistics are homicides, since deaths and bodies have to be reported and most homicides are solved, and racial disparities are indeed huge. A 33:1 disparity between African-American and Asian-American homicide rates, 8:1 between whites (including Hispanic whites, which are aggregated for these statistics, but not most other federal statistics, reducing the apparent disparity) and African-Americans, probably 10:1 between non-Hispanic whites and African-Americans. About half of all homicides are committed by African Americans, and probably less than a third by non-Hispanic whites. Because demographics vary across the country, in regions where African-Americans or Latinos are much more than 10% of the population either (and often do) group can easily form a supermajority of homicide offenders. Race predicts criminal activity much better than SES, education, or any other such variables, and policy analysis that does not take such large effects into account will be systematically misguided.

Similar issues arise elsewhere, e.g. the idea of ‘equal pay for equal work’ is often presented as though income differences primarily reflected discrimination in hiring and irrational discrimination by employers in pay levels (with no explanation of why sex-indifferent companies don’t make enormous profits by hiring women at 80 cents on the dollar instead of 70). Hilary Clinton did this and proposed legislative intervention in wage-setting, a return of ‘comparable worth.’

In fact, the differences in income can be explained by reduced work hours and continuity of employment due to the unequal distribution of childrearing in our society, differences in education focus, physical labor and risk involved in jobs, preferences (for work involving interaction with people and animals versus inanimate objects) and non-cash benefits (short commutes, a clean office environment, etc). There are various things that could be done to address these factors, e.g. subsidized daycare, changes in the tax system, etc, but a legislative intervention setting wages in numerous industry based on factors other than supply and demand is a horribly wasteful and inefficient way to do things, which is why economists oppose it. But if media self-censorship prevents acknowledgment of any causes other than employer discrimination, then we are more likely to get the bad policy.

Here are some problems for the four points in the Fair Measures article:

1. Racial discrimination laws create an effective presumption of guilt in the presence of group disparities in hiring or admissions, which results in firms and institutions creating unofficial quota systems to match employee or student proportions to population proportions. Implicit bias effects are nowhere near large enough for completely eliminating them to come close to eliminating those disparities, and so they cannot justify affirmative action as it is practiced.

2. Your second point, that more accurate measurements of ‘merit’ should be adopted, is irrelevant for affirmative action as we know it, which is applied to prevent the effects of scientifically-informed personnel selection conducted without knowledge of group membership. There are big differences in work performance across groups:

http://psycnet.apa.org/?fa=main.doiLanding&doi=10.1037/0021-9010.91.3.538
http://en.wikipedia.org/wiki/Intelligence:_Knowns_and_Unknowns

An organization allowed to use IQ tests and other predictively valid measures of work performance to select the employees who will do the best jobs will wind up with big racial disparities, end of story. The Supreme Court cases were about deviating from measures (grades, standardized tests) that *accurately predicted or over-predicted* performance (graduation rates, grades, medical malpractice for the Bakke medical school case) for favored minorities. If we wanted to minimize the number of patients killed by medical malpractice, maximize the performance of life-saving firefighters, etc, the data suggest that predictively valid psychometrics and physical tests should be used, measurements that would result in forbidden group disparities.

3. Debiasing techniques are handy if you’re going for accuracy, but with numerical quotas they’re unneccessary unless implicit associations cause mistakes in rank-ordering candidates within groups.

4. There will be a long wait indeed for the elimination of implicit bias in the presence of real differences in ability and performance. Anyone who accurately observes their environment or is familiar with the relevant data will risk forming the forbidden implicit associations.