Assessing Predictor Robustness in Healthcare Consumer Research: A Bayesian Model Averaging Approach

Abstract

Healthcare consumers are increasingly acting as active co-producers of their care, especially in the management of chronic diseases, where they assume substantial responsibility throughout the care process. Although research on healthcare consumer behavior is growing, methodological limitations, particularly “researcher degrees of freedom” in variable selection, led to inconsistent conclusions about which service design elements truly influence consumer outcomes. To address this issue, we apply Bayesian Model Averaging (BMA) to evaluate the robustness of key consumer interaction variables across alternative model specifications. Using data from diabetes education programs at six hospitals, we focus on three core service design elements: relational quality (consumer trust in health educators), knowledge types (know-what, know-how, and know-why), and media forms (educational delivery methods such as written material, one-on-one meetings, and group classes). Bayesian model averaging (BMA) reveals that group classes for blood glucose monitoring and practical meal planning know-how emerge as the strongest and most robust predictors across consumer satisfaction, knowledge acquisition, behavior change, and health outcomes. Group-based education formats demonstrated substantially higher robustness compared to individual delivery methods across this intervention hierarchy. Our research highlights the value of BMA in producing more reliable, model-independent insights for consumer research, suggesting that healthcare organizations may achieve greater impact by prioritizing group-based blood glucose monitoring education and practical meal planning knowledge dissemination.