COVID-19 Protective Behaviors: Identifying Key Determinants of Face Mask Use with CIBER Plots and Machine Learning

Objectives: During pandemics, various determinants influence protective behaviors such as face mask use. Current evidence lacks insights into sociodemographic differences and potential room for improvement in these determinants, despite this being essential for effective interventions. We aimed to identify the most promising intervention targets for face mask use during COVID-19, across the whole population and stratified by age and education, using novel analytical approaches.Design: Cross-sectional survey of a representative Finnish adult population sample.Methods: 2272 adults responded to questions about mask wearing and potential determinants (May 2021). We employed two complementary analytical approaches: Confidence-Interval Based Estimation of Relevance (CIBER), which combines bivariate correlations with univariate distributions, and interpretable machine learning (Categorical Boosted Tree regression with SHapley Additive exPlanations).Results: Both analytical approaches identified similar promising intervention targets with some notable differences. Key determinants included injunctive norms (particularly expectations of people present rather than authorities), habit, response efficacy, feeling safe when wearing masks, and perceived disease severity. Age was more influential than education in determining both the relevance of determinants and room for improvement, with older adults showing less room for improvement in key determinants.Conclusions: Promising intervention targets encompass reasoned, habitual, social, and affective elements. Our findings emphasize the importance of socio-contextual determinants like situation-specific injunctive norms, which are often overlooked in behavior change interventions, and highlight the need to consider demographic differences such as age when designing mask-wearing interventions.