Experimental Design
Our general experimental design follows those of similar email-correspondence studies (Ahmed and Hammarstedt 2009; Hanson et al. 2016; Giulietti, Tonin, and Vlassopoulos 2019; Schwegman 2019).
Treatment is clustered at the mortgage loan originator (MLO) levels, and we send two emails to each MLO. We paired our two emails to each MLO as follows. In random order, we sent one email from a same-gender couple (randomized to either be female or male), and one email from a different-gender couple (randomized so that the emailer is either female or male). Independently and in random order, one couple was always White (both individuals have White names), and the other couple was African American (both African American names) with 75% probability on average, and White otherwise. We varied the average probability of 75% for having an African American couple by the proportion African American in the metropolitan (or micropolitan) statistical area (MSA) so-as to make our sample of African American couples more population representative.
In addition to race and sexual orientation signals, we randomly vary some demographic and socioeconomic characteristics of prospective borrowers: credit scores, occupation, and fertility. We signal for creditworthiness by randomly assigning both the prospective borrower and their spouse either low credit scores, high credit scores, or no mention of credit scores, each with a probability of one third. We randomly include occupation and tenure of employment, which are both seen as important in the loan decision. Using data from the U.S. Bureau of Labor Statistics (2021), we select low-income and high-income occupations that are common enough to be needed in all communities and provide variation in wages or salaries. Finally, for one third of our sample of MLOs, we add mention of expecting a child.