A Method to Explore the Best Mixed-Effects Model in a Data-Driven Manner with Multiprocessing: Applications in Public Health Research

• Main Author: Hyemin Han (Author)
• Format: Book
• Published: MDPI AG, 2024-05-01T00:00:00Z.
• Subjects: article
• Online Access: Connect to this object online.
• Notes: 10.3390/ejihpe14050088; 2254-9625; 2174-8144
• Keywords: data science; data-driven analysis; model exploration; mixed-effects model; public health; Public aspects of medicine; RA1-1270; Psychology; BF1-990
• Source: European Journal of Investigation in Health, Psychology and Education, Vol 14, Iss 5, Pp 1338-1350 (2024)

In the present study, I developed and tested an R module to explore the best models within the context of multilevel modeling in research in public health. The module that I developed, <i>explore.models</i>, compares all possible candidate models generated from a set of candidate predictors with information criteria, Akaike information criterion (AIC), and Bayesian information criterion (BIC), with multiprocessing. For testing, I ran <i>explore.models</i> with datasets analyzed in three previous studies in public health, which assumed candidate models with different degrees of model complexity. These three studies examined the predictors of psychological well-being, compliance with preventive measures, and vaccine intent during the COVID-19 pandemic. After conducting model exploration with <i>explore.models</i>, I cross-validated the nomination results with calculated model Bayes Factors to examine whether the model exploration was performed accurately. The results suggest that <i>explore.models</i> using AIC and BIC can nominate best candidate models and such nomination outcomes are supported by the calculated model Bayes Factors. In particular, all the identified models are superior to the full models in terms of model Bayes Factors. Also, by employing AIC and BIC with multiprocessing, <i>explore.models</i> requires a shorter processing time than model Bayes Factor calculations. These results indicate that <i>explore.models</i> is a reliable, valid, and feasible tool to conduct data-driven model exploration with datasets collected from multiple groups in research on health psychology and education.