Addressing missing values in structural equation modeling with application to digital transformation survey data and its role in improving the quality of healthcare in Mosul

Abstract

The presence of missing data is a pivotal methodological challenge across various scientific disciplines, particularly in structural equation modeling (SEM), due to its direct impact on the accuracy of statistical estimation and the quality of model fit. This study aimed to evaluate and compare the performance of three common missing data handling methods: full information (FIML), pairwise deletion (PD), and total deletion (LD), within the framework of SEM using questionnaire data on digital transformation and its role in improving the quality of sustainable healthcare in Mosul.The study employed a descriptive-analytical approach with a sample of 303 individuals. SEM analysis was performed using R software, relying on three estimation methods: maximum likelihood (ML), unweighted least squares corrected by mean and variance (ULSMV), and weighted least squares corrected by mean and variance (WLSMV). The effect of each processing method on estimation accuracy, fit quality, and statistical efficiency was evaluated.To support the applied findings, a systematic simulation study was conducted comparing the performance of pairwise and whole deletion with the ULSMV and WLSMV estimation methods under sample sizes of 200, 400, 600, 800, 1000, and 1500 and loss ratios of 0.01 and 0.1, respectively, on a four-factor model, each factor comprising five observed variables. Standardized fit indices were used for evaluation. The results showed that the FIML method outperformed most evaluation criteria in terms of statistical accuracy, fit quality, and stability. The simulation also demonstrated that pairwise deletion outperformed whole deletion at low and medium loss ratios, with relative improvement as sample size increased. ULSMV and WLSMV showed similar performance at low loss, with ULSMV tending towards better stability in smaller samples. These results underscore the need to adopt advanced methods for handling missing data and avoid exclusive reliance on traditional deletion methods in SEM applications.