Tool · Experimental Research on Race
Power calculator for correspondence audit studies.
How many applications do you need to reliably detect racial discrimination? This tool computes required sample sizes for paired and unpaired audit designs, using exact formulas from Fleiss, Levin & Paik (2003) and Schlesselman (1982).
Use this tool alongside the Validated Name Picker — power first, then names.
Power curve
Required sample size vs. effect size at your chosen α and design. Orange line marks your current effect size.
Sensitivity table
Required sample at your chosen α and design across a range of effect sizes. Highlighted row matches your current settings.
Formulas. Paired (McNemar): Schlesselman, J.J. (1982). Case-Control Studies. Oxford; Fleiss, J.L., Levin, B., & Paik, M.C. (2003). Statistical Methods for Rates and Proportions, 3rd ed. Wiley (eq. 8.14–8.17). Unpaired (two-proportion z-test): Fleiss et al. (2003), eq. 4.19. The ρ parameter generalizes the paired formula to allow for within-pair callback correlation. Setting ρ = 0 assumes independence (the conservative default). With ρ > 0 (the realistic case for hiring audits, where busy firms call back both applicants more), the discordant-pair probability πd shrinks, each discordant pair carries more signal, and fewer job postings are needed. Job-posting savings shown are relative to an equivalent between-subjects design (2 × n per group). See also: Neumark, D. (2012). Detecting discrimination in audit and correspondence studies. Journal of Human Resources, 47(4), 1128–1157.