Anonymization of job applicant resumes is a commonly recommended strategy to increase diversity in organizations. However, real-world, large-scale tests of anonymization have shown mixed results for increasing diversity in interview call-back rates and job offers. We consider decision-makers’ social dominance orientation (SDO), a measure of anti-egalitarianism/endorsement of group-based hierarchy, to illustrate one reason why anonymization has not been universally successful in the field. Across three pre-registered studies (N=3,000), we show that SDO predicts the likelihood of adopting anonymized hiring processes and moderates the effects of anonymization: lower SDO individuals are both more likely to opt into using anonymized hiring processes, and less likely to hire individuals from underrepresented groups when anonymization is used. Thus, ironically, opt-in anonymization policies can reduce the diversity of who is selected. More broadly, we suggest that policy evaluations need to account for the possibility that heterogeneous treatment effects and selection effects regarding who is most likely to adopt policies may inadvertently lead to outcomes that are contrary to the stated policy goals.