Do women benefit more from gender-homophilous (all-women) or gender-heterophilous (mixed-gender) groups in remote training? Existing theories offer no clear answer, as remote training environments disrupt the social architecture that enables peer effects—weakening not only the collaborative structures that encourage cross-gender exchange but also the interpersonal cues that foster same-gender bonding. We argue that remote training environments reveal a key mechanism through which all-women peer groups confer distinctive benefits: identity-based trust. In all-women peer groups, shared gender identity can help participants transcend the relational barriers of remote interaction, fostering trust-based ties that facilitate mutual support. We test this argument in an 18-month randomized field experiment on a leading online career training platform, which randomly assigned over 2,700 unemployed women to all-women or mixed-gender virtual peer groups. We find that women in all-women groups were significantly more likely to complete their training on time, earn professional certification within a year, and secure in-field employment. Analyses of text communication data reveal three key patterns underlying identity-based trust in all-women groups: (1) multiple shared identities (i.e., marriage, motherhood, career), (2) affective expression, and (3) reciprocal exchanges of support. By contrast, interaction in mixed-gender groups was inhibited, preventing supportive dynamics—even among women—from forming. This study provides the first causal evidence on how peer gender composition shapes women’s career outcomes in remote training and illuminates the microgroup mechanisms through which gender-homophilous groups foster success even in digitally mediated environments. Funding: This work was supported by the Tuck School of Business, the Stanford Institute for Research in the Social Sciences, the Clayman Institute for Gender Research, and the Stanford VMware Women’s Leadership Innovation Lab. Supplemental Material: The online appendix is available at https://doi.org/10.1287/orsc.2025.20602 .
Melin et al. (2026) studied this question.