Mutually canceling in-group biases can erase all detectable signals of racial discrimination

Evidence

Assume that estimating Equation (3) also revealed that β3 is 2. This means that White judges assign two months longer sentences to minority defendants vis-à-vis White defendants. Put differently, White judges exhibit positive in-group bias by assigning White defendants two months shorter sentences vis-à-vis minority defendants. Note that, in this situation, (β3 + δ3), which signifies how minority defendants are treated vis-à-vis White defendants by minority judges, is equal to -4. In turn, this means that minority defendants receive sentences that are four months shorter when they are matched with a minority judge in comparison to the situation where whey would be matched with a White judge. In summary, both White and minority judges favor their in-group by assigning shorter sentences. … Thus, β1, which can be calculated as [β3+(1- π) δ3], is equal to [2+(1/3)(-6)]=0, and we know that 0 indicates no racial bias in sentencing. … The intuition is straightforward. A researcher who has access to a complete data set with all relevant variables that are needed to investigate racial bias can run Equation (1), and he/she may estimate the coefficient of Minority Defendant as being equal to zero, indicating no racial bias in sentencing. This inference, although correct at the aggregate, could mask more nuanced and layered discriminatory behavior. Specifically, this “no evidence for discrimination” would be an artifact of how White judges discriminate against minorities and minority judges discriminate against Whites, and the magnitudes of these effects and the proportions of White and minority judges in the data are such that they could cancel out at the aggregate. … To put it differently, in a judicial system in which judges favor “their own type” and where there are sufficient number of cases of each type handled by both types of judges, no racial bias effect will be detected in the data (using an analysis in the spirit of Equation 1) if in-group bias effect is similar for both types of judges. This finding would fail to reject the hypothesis of no racial discrimination, and therefore would incorrectly report “no evidence of racial discrimination.” — Joseph Avery, Jim Sidanius, Joel Cooper, Gregory Davis, Amanda N., Bias in the Law, 769

Categorization

• SEED: Article angle: ‘When zero is noise — how aggregate statistics can systematically erase symmetric discrimination’ — a piece on how measurement at the wrong level hides rather than reveals bias.
• DEEPTHINKING: This is a dense logical mechanism worth sitting with: a null result that is itself the product of two active discriminations, with implications far beyond sentencing.
• corroborated: 2 independent passages support this claim — Joseph Avery, Jim Sidanius, Joel Cooper, Gregory Davis, Amanda N., Bias in the Law: A Definitive Look at…, 769
• XREF: connects to: Geographic quotas can function as covert racial discrimination that is nearly impossible to prove — Henry M. Hart, Jr. and Albert M. Sacks, The Legal Process: Basic Problems in…, 1704; Racialized actors can decode subtle discrimination because they recognize its recurring patterns — David Theo Goldberg, John Solomos, A Companion to Racial and Ethnic Studies, 1489