Trying to sustain prior assumptions in the face of hard evidence

Interesting but almost self-defeating. From Affirmative Action, Major Choice, and Long-Run Impacts by Zahary Bleemer. From the Abstract:

Estimation of the impact of race-based affirmative action (AA) on the medium- and long-run outcomes of underrepresented minority (URM) university applicants has been frustrated by limited data availability. This study presents a highly-detailed novel database of University of California (UC) applications in the years before and after the end of its AA admissions policy, linked to national educational records and a California employment database. Using a difference-in-difference design to compare URM and non-URM freshman applicants' outcomes two years before and after UC's affirmative action policies ended in 1998, I identify substantial and persistent educational and labor market deterioration after 1998 among URM applicants: each of UC's 10,000-per-year URM freshman applicants' likelihood of earning a Bachelor's degree within six years declined by 1.3 percentage points, their likelihood of earning any graduate degree declined 1.4 p.p., and their likelihood of earning at least $100,000 annual between ages 30 and 37 declined by about 1 p.p. per year. These results suggest that affirmative action's end decreased the number of age 30-to-34 URM Californians earning over $100,000 by at least 2.5 percent. Turning to targeted students' major choice, I link the application records to five universities' detailed course transcript data and find no evidence – despite considerable statistical power – that more-selective university enrollment under AA lowered URM students' performance or persistence in core physical, biological, or mathematical science courses. These findings suggest that state prohibitions on university affirmative action policies have modestly exacerbated American socioeconomic inequities.

OK. Modest effect sizes is an unpleasant acknowledgement that the policy may not be worthwhile in the first place. You want definitive and positive effect sizes.

More importantly, this is a typical example of ideological priors. If we are to justify institutionalized racism, which is what affirmative action is, then we justify it on three grounds. The first is that it rights a prior wrong suffered by the targeted population. The second is that it is an effective policy in raising minority beneficial outcomes. The third is that the benefits outweigh the costs.

In this instance, the first justification is open. Whether you think institutionalized racism today is warranted to address wrongs committed by other parties against other groups of people in the distant past is a philosophical and moral issue without an easy resolution. Virtually all positions have good arguments for them.

Is it an effective policy? Well, if cancelling the policy only has modest impact, i.e. modest effect sizes, then I think it is a fair question whether it is worth pursuing in the first place. Especially when so many confounds (effect based on occurrence in the economic cycle, status of complimentary public policy such as economic set asides, etc.) are not addressed.

Is it a worthwhile policy? Here is where Bleemer seems to fall down. It is not clear from the Abstract whether Bleemer is taking into account all costs and all benefits. Is he netting out the costs of those "beneficiaries" of affirmative action who incur the student loan costs of matriculating without graduating? These can be very large and it is unclear whether they enter into the equation. Does he examine real results, i.e. in the private sector, or does he aggregate private and public sector? If affirmative action graduates disproportionately enter public sector work, then there is a hidden and continuing subsidy which is not being taken into account. Finally, does Bleemer take into account the likely productivity of those candidates displaced by affirmative action? If a displaced candidate earns on average $100,000 pa and the affirmative action candidate earns $40,000, there are very real tax and societal economic costs that need to be taken into account.

Bleemer, perhaps captive to the ideological objective that all people are identical and should have the same outcomes, does not seem to address the whole picture. He only addresses whether affirmative action increases bean counting based on race. Even with that inappropriately narrow focus, he finds the effect to be modest.

Seems to be failed policy if we cannot agree on the original justification, if we cannot determine that it is effective, and if we can be pretty certain that it is more costly than acknowledged.