Identifying and understanding discrimination is notoriously difficult because it requires disentangling legitimate differences in ability or productivity from differences in outcomes due to discrimination. For example, if one observes that the members of a certain group (e.g., migrants) are paid less than those of another group (e.g., natives), one cannot say that these differences are due to discrimination as they might be due in large part or even entirely to differences in productivity.
In this project, we plan to use detailed administrative records about the bar exam to access the legal profession in Italy to study discrimination along various important dimensions. The exam consists of two parts, one written test, which is anonymous by design, and an oral test, which is non-anonymous by design. Hence, for each candidate, we observe both outcomes in an anonymous test and in a non-anonymous test, combined with the identities of the evaluators. We also have information about various demographics of both the candidates and the evaluators, which allows us to identify potentially discriminated groups. We can access this administrative archive for many years (the bar exam takes place every year), potentially from the 1950s to 2000.
This is a pretty exceptional source of information that is ideal to study discrimination along a variety of dimensions, such as gender and migration status but also family background (i.e., whether the candidate has a relative who is already a licensed lawyer). Moreover, the setting of the analysis, the bar exam, should not be considered as a mere case study. This is a very high-stakes exam whose outcome affects the professional lives of a very large group of individuals. The entry exam into the legal profession is similar to that of several other regulated professions and our results can be informative well beyond the specific case of Italian lawyers. Furthermore, discrimination in the selection process of professionals can importantly affect the quality of crucial services provided to society, such as legal or medical services. Our results can inform the design of entry regulations with potentially very important implications for welfare.
We have the opportunity to observe the same agents repeatedly both in a setting where there cannot be discrimination by design, as the written test is anonymous, and in a setting where there can be discrimination, as the oral test cannot be anonymous. The main idea is that the differences across groups (by gender, migration status, family background, etc.) in the anonymous written test reflect differences in ability/productivity and, conditional on these, differences in the oral test reflect discrimination. The most natural empirical model to analyse this dataset is a factor model, where we can impose reasonable assumptions and relax them easily for robustness purposes. This setting allows us to go beyond simply detecting the presence of discrimination and also investigate the joint distribution of ability and discrimination. For example, we will be able to say whether it is the most able or the least able agents that are discriminated the most.
This is particularly relevant to uncover potentially important policy trade-offs between equality of opportunity and efficiency. If it is the least able agents who are discriminated against the most, it might eventually be the case that discrimination improves efficiency by indirectly producing a more positive, although less fair, selection of active agents. Moreover, by knowing the identities of the evaluators we will be able to say who discriminates the most and against which groups. The results will allow us to substantially improve our understanding of discriminatory practices and will offer practical recommendations to policymakers.
Important links
