Gender Disparities in Research Return Rates: The Moderating Influence of AI Self-Efficacy and Methodological Design

Abstract

Background: The leaky pipeline of data collection remains a significant challenge as research infrastructures pivot toward Artificial Intelligence (AI) and automated tracking. While women traditionally exhibit higher prosocial participation in research, the introduction of AI-driven methodology may act as a demographic filter, creating a methodological gender gap that threatens data inclusivity.

Objective: This study investigates the moderating influence of AI Self-Efficacy and Gender on research return rates. The goal is to identify if technological invasiveness reverses traditional participation advantages, thereby skewing the data pools used to train and evaluate AI systems.

Methodology: A factorial survey experiment was conducted with a stratified sample of N=450 participants (n=150 Male, n=150 Female, n=150 Non-binary). Participants were randomly assigned to one of three research scenarios: a traditional digital survey (Control), an AI-led laboratory experiment utilizing biometric facial analysis (Treatment A), and a longitudinal quasi-experiment using passive AI tracking (Treatment B). The dependent variable, Probability of Return (P_r), was analyzed using a 3x3 Factorial ANOVA and moderated mediation modeling (Hayes Process Model 1) against a 5-item validated AI Self-Efficacy Scale (α = 0.88).

Results: The analysis revealed a significant interaction effect between gender and research design (F = 18.54, p < .001, n² = .14). In the control group, women reported a higher return rate (72.4%) than men (69.8%). However, a crossover effect occurred in Treatment A, where male return rates (62.1%) significantly outpaced female rates (38.5%). AI self-efficacy was found to be a significant gatekeeper for female participants (beta = .58, p < .001) but had no significant impact on male participation (beta = .12, p = .14). This confirms that while men remain resilient to technological complexity, women’s participation is heavily contingent on their technical confidence.

Conclusion: The transition to AI-mediated research introduces a tech-tax that disproportionately discourages female participation, leading to male-dominated datasets. To ensure the growth of inclusive research, scholars must prioritize AI trust-building and transparent design. This paper provides a vital roadmap for auditing research methodologies to prevent the systemic exclusion of women from the future of knowledge production.