This paper examines the role racial discrimination played in the programming decisions of radio stations in the post-war Jim Crow era. Song constructs a novel, comprehensive dataset of all commercial radio stations across the U.S., including station-level financial information and Black-oriented programming hours, at the start of minority-oriented programming in broadcast media. Using a theoretical framework under the free entry assumption, the researcher develops a test for firm-owner discrimination in the spirit of Becker’s outcome test. Applying this test, Song provides the first set of evidence of firm-owner discrimination in programming decisions in the media market: she shows that Black-oriented stations were significantly more profitable than other stations in the same market. In addition, the profitability of Black-oriented radio stations is correlated with Dixiecrat vote share, a market-level measure of racism in the South. Similar pattern is not observed for the niche market of foreign-language stations.
Antman and Cortes present the first quantitative analysis of the impact of ending de jure segregation of Mexican-American school children in the United States by examining the effects of the 1947 Mendez v. Westminster court decision on long-run educational attainment for Hispanics and non-Hispanic whites in California. Their identification strategy relies on comparing individuals across California counties that vary in their likelihood of segregating and across birth cohorts that vary in their exposure to the Mendez court ruling based on school start age. Results point to a significant increase in educational attainment for Hispanics who were fully exposed to school desegregation.
How and why do racial stereotypes arise, and how do negative stereotypes harm the stereotyped group? McGee studies a model of stereotypes as motivated reasoning when social groups interact. He shows that there exist asymmetric equilibria with biased fiirst-order and higher-order beliefs, where one group chooses to denigrate the other to achieve a competitive advantage, while believing that this other group 'sees us as we see ourselves.' Even though group differences are intrinsically meaningless, agents develop 'racial' beliefs where they think these differences mark differences in ability. Furthermore, stereotypes arise despite the absence of 'inherent' animosity between groups.
The racial wealth gap is the largest of the economic disparities between Black and white Americans, with a white-to-Black per capita wealth ratio of 6 to 1. It is also among the most persistent. In this paper, Derenoncourt, Kim, Kuhn, and Schularick provide a new long-run series on white-to-Black per capita wealth ratios from 1860 to 2020, combining data from the US Decennial Census, historical state tax records, and a newly harmonized version of the Survey of Consumer Finances (1949-2019). Using these data the researchers show that wealth convergence was rapid in the 50 years after Emancipation, but slowed to a halt by 1950. A simple model of wealth accumulation by racial group reveals that even under equal conditions for wealth accumulation, convergence is a distant scenario given vastly different starting conditions under slavery. Accounting for post-Emancipation differences in wealth accumulating opportunities indicates that the racial wealth gap is on track to arrive at a “steady state,” close to today’s levels. Their findings shed light on the importance of policies such as reparations, which address the historical origins of today’s persistent gap, as well as policies that reduce inequality and thereby improve the relative wealth position of Black Americans.
There is a growing concern that algorithmic decision-making can lead to discrimination against legally protected groups, but measuring and reducing such algorithmic discrimination is often challenging because an individual's latent qualification for treatment is often only selectively observed. Arnold, Dobbie, and Hull develop new quasi-experimental tools to overcome this fundamental selection challenge and both measure and reduce algorithmic discrimination in the setting of pretrial bail decisions. They first show that algorithmic discrimination can be measured and reduced using a small number of moments involving individual qualification. The researchers then show how these moments can be estimated by extrapolating quasi-experimental variation across as-good-as-randomly assigned decision-makers. Estimates from New York City show that both a sophisticated machine learning algorithm and a simple regression-based algorithm discriminate against Black defendants, even though defendant race and ethnicity are not included in the training data. Preliminary analyses suggest algorithmic discrimination can be dramatically reduced or eliminated using their methods, with little loss of predictive accuracy.