Technical Appendix

Survey Data

The findings presented in this report come from data collected by researchers from a stratified random sample of commercial banks¹ in the United States. In 2016, researchers developed, piloted, and conducted a 57-question survey to uncover variation in the costs of entry-level checking accounts. In addition to the costs and fees of entry-level checking accounts, survey questions covered topics such as banks’ strategies for serving consumers (e.g., whether branches operated extended hours during evenings and weekends, offered non-English language services, used ATMs, and offered online and/or mobile banking) and transaction processing (e.g., whether transactions were processed in chronological order). The survey was piloted and data were collected between March and December 2016 from a stratified random sample of retail banks identified from the FDIC’s list of 6,186 active banks. The FDIC’s procedures from the FDIC (2016c) Small Business Lending Survey² were implemented to select a stratified random sample, including stratifying by banks’ asset amounts and metropolitan and non-metropolitan areas.³ Contact information for each bank’s main branch was used for survey data collection and subsequent analyses.⁴ The sample included 1,976 banks and 1,625 banks completed the survey. The final analytic sample included 1,344 banks with complete data on the outcome variables.

Racial Makeup and Demographic Characteristics

To analyze relationships between geographic variation in demographic characteristics and checking account costs and fees, geocoded⁵ survey responses were combined with data from the 2011-2015 5-year sample of the American Community Survey (ACS) (Minnesota Population Center 2011). ACS data were gathered for neighborhoods (i.e. census tracts) as well as cities and towns (i.e. census places and county subdivisions) on the following: racial makeup (i.e. percent non-Hispanic/Latinx white, non-Hispanic/Latinx black, non-Hispanic/Latinx Asian, and Hispanic/Latinx) and demographic characteristics, including percent foreign born, educational attainment (i.e. percent of adults with a college degree and percent with less than high school), poverty rate, homeownership rate, and median age.

Banking Characteristics

Controls were included to measure banks’ asset holdings and location in rural areas (i.e. variables for bank size⁶, asset class, and rural location of the branch), as well as the job title or role of the bank employee that responded to the survey (i.e. variables for teller, customer service, or other job [such as retail or sales representative, branch manager or bank vice president]⁷) and whether they held a supervisory role. The financial services environment was also measured, including the presence of other commercial banks and alternative financial services (AFS). Data from 2014 Federal Deposit Insurance Corporation (FDIC) summary of deposits, 2014 National Credit Union Administration (NCUA) call reports, and 2015 InfoGroup proprietary business listings were used to calculate the geographic density of both traditional and alternative financial services (i.e. the number of each per square mile in the census tract).

Methods

A series of regression models were estimated for each outcome of interest, and the results described in this report are based on models with full controls: racial makeup, demographic characteristics, and bank characteristics. Interaction terms were used to test for evidence of racialized discretion, such as interacting indicator variables for the survey respondent’s job title or role (i.e. teller, customer service, or other role) with racial makeup for predicting checking account costs and fees. Regression coefficients were also used to estimate how checking account costs and fees vary across the typical neighborhood-level (or place-level) racial makeup experienced by white, black, Asian, and Latinx Americans. The 2011-15 ACS data were used to calculate the white-black, white-Asian, and white-Latinx exposure indices on the census tract-level for the entire United States excluding Puerto Rico, which measure the average tract-level percent black, Asian, and Latinx among white Americans.⁸ These values were then multiplied by the coefficients for percent black, Asian, and Latinx in fully-controlled models and the sum was added to the product of the coefficient for each covariate and that covariate’s mean.⁹