Comparison of logistic, log binomial and modified poisson regression models in identifying factors associated with early marital union in Uganda

Abstract

Introduction: Early marital union, defined as marriage under the age of 18, affects 33,000 lives globally each day, with 43% of Ugandan women aged 25-49 experiencing it. This study aimed to compare how well ordinary logistic, log-binomial, and modified poison regression models fit the same data in identifying factors associated with early marital unions in Uganda. Objectives: To assess the performance of three binary statistical models and use the best model to identify the factors associated with early marital unions in Uganda. Methodology: Using UDHS-2016 data on a weighted sample of 13,768 ever-married women, the study assessed model performance based on metrics such as AIC, AUC, negative log-likelihood, variance, and Wald chi-square in R-software. The outcome variable was binary, signifying whether a respondent got married before turning eighteen years (coded as 1) or not (coded as 0). Results: The logistic model exhibited the lowest AIC, highest AUC, and lowest Log likelihood, while modified Poisson had the smallest variance. Multivariable multilevel logistic regression revealed that women with secondary or higher education had a 63% decreased odds of early marital unions (AOR=0.37, 95% CI: 0.3, 0.4) as compared to those with no formal education. Other associated factors included age at first sex (AOR= 0.05, 95% CI: 0.08, 0.1), internet use (AOR=0.61, 95% CI: 0.5, 0.8), region (AOR= 0.87, 95% CI: 0.8, 1.0), source of drinking water (AOR= 1.22, 95% CI: 1.1, 1.4 & AOR =1.21, 95% CI: 1.0, 1.5), and recent sexual activity (AOR= 1.14, 95% CI: 1.05, 1.24). Conclusion: Although logistic regression model was a superior fit, Modified Poisson model provided precise estimates. Early marital unions were significantly reduced among educated women. Recommendations include prioritizing education beyond primary school for girls, promoting responsible internet use, and implementing region-specific interventions addressing socio-cultural and economic factors contributing to early marriages in Uganda.