It can be difficult to prove the existence of discriminatory attitudes to the full satisfaction of a social scientist. Ideally, one would like a real-world random event or policy intervention to occur, so that you could compare the real-world behavior of those who either experienced the random event, or not. A couple of recent studies have found specific situations where this can be done.
Here's one example, analyzed by Marco Battaglini, Jorgen M. Harris, Eleonora Patacchini in "Professional Interactions and Hiring Decisions: Evidence from the Federal Judiciary" (January 2020, NBER Working Paper 26726). They note that in US appellate courts, the usual pattern is that cases are heard by three judges randomly assigned from the overall group. They also note that judges have pretty much a free hand in deciding who to hire as clerks. An overview of their work reports:
Using data from the Judicial Yellow Book and a database of federal court cases, the researchers find that judges who hear more cases alongside female colleagues are 4 percentage points more likely to hire at least one female law clerk in the following year. They also examine whether exposure differentially affects judges who vary in meaningful ways — by gender, age, experience level, political affiliation, status, quality, and current law clerk staff composition. The effects of exposure to female judicial colleagues are larger for men, for judges whose current roster of law clerks is majority male, and for less-experienced judges. For example, male judges who hear more cases alongside female judges are 4.3 percentage points more likely to hire a female law clerk subsequently, while female judges who hear more cases alongside female judges are 1.6 percentage points more likely to hire a female law clerk subsequently.
Here's another example from a few years ago, this one from the venture capital industry, from research by Paul A. Gompers and Sophie Q. Wang “And the children shall lead: Gender diversity and performance in venture capital” (2017. Harvard Business School Working Paper 17-103). They look at whether senior partners in venture capital firms have sons or daughters, which for this group we can treat as a random event. Then they look at how likely the firm is to hire women.
My guess is that not many judges or venture capitalists had previously been openly vowing not to hire women as clerks or as as part of venture capital firms; indeed, I'm sure many of had previous hired a few women. In addition, I doubt that these changes involved an open and conscious change-of-heart epiphany about hiring more women. I doubt that a lot of judges were muttering in their chamber: "Well, now that I've been on a panel with a female judge, I'll go ahead and hire more female law clerks." I doubt a lot of venture capitalists were muttering to themselves: "Well, now that I have a daughter, all of a sudden women job applicants seem more plausible to me." Instead, it seems to me that these kinds of studies both demonstrate the existence of an unconscious bias and a mechanism for attenuating that bias.
We find that the proportion of female hires increases by 1.93% if you replace a son with a daughter for the existing partners in a firm. Given that about 8.03% of the new hires are female, this suggests a 24% increase in the probability of hiring a senior female investor when a son is replaced with a daughter for the existing partners. ... [W]e also show that improved gender diversity, induced by parenting more daughters, improves deal and fund performances.
The studies also demonstrate that increasing contact can be a meaningful tool for reducing discriminatory attitudes. Kevin Lang and Ariella Kahn-Lang Spitzer make this point in the context of racial discrimination in their essay "Race Discrimination: An Economic Perspective" (Spring 2020, Journal of Economic Perspectives). They point out that racial discrimination potentially affects many areas: residential location, credit access, educational opportunity, hiring, criminal justice, medical treatment, and so on. Pushing back against discrimination in each of these areas is of course worthwhile, but some ways of pushing back may have broader leverage than others across a range of areas. They write:
[P]olicies to increase interracial contact—like limiting residential segregation—may offer a useful point of leverage. Residential and social segregation may lead to prejudice and taste-based discrimination. Pettigrew and Tropp (2006) provide a meta-analysis of 515 studies and conclude that there is strong support for “intergroup contact theory,” which proposes that contact tends to reduce prejudice. Some economists have contributed to our understanding of this topic. Carrell, Hoekstra, and West (2015), for example, found that having an additional black member in an Air Force squadron of roughly 35 people increased the probability of having a black roommate as a sophomore (usually not a freshman squadron member) by about one percentage point, or about 18 percent. Similarly, exposure to more black peers with high admissions scores increased the probability that whites reported that they had become more accepting of African Americans. Dahl, Kotsadam, and Rooth (2018) find similar positive effects on male attitudes towards female recruits from having been assigned to a squad with a woman member during boot camp in Norway. In particular, given the importance of networks in job search, social distance can directly increase racial disparities in employment (Loury 2000). These studies, together with the large literature outside economics, suggest a public interest in greater integration and reducing social distance across groups.