An empirical comparison of some missing data treatments in PLS-SEM

• Published in: PloS one Vol. 19; no. 1; p. e0297037
• Main Authors: Amusa, Lateef Babatunde, Hossana, Twinomurinzi
• Format: Journal Article
• Language: English
• Published: United States Public Library of Science 19-01-2024; Public Library of Science (PLoS)
• Subjects: Algorithms; Analysis; Computer Simulation; Customer satisfaction; Data mining; Datasets; Deletion; Empirical analysis; Evaluation; Expected values; Gene deletion; Missing data; Monte Carlo method; Monte Carlo simulation; Parameter estimation; Performance evaluation; Regression; Social sciences; Structural equation modeling; Validity; Variables; South Africa
• ISSN: 1932-6203; 1932-6203
• DOI: 10.1371/journal.pone.0297037

PLS-SEM is frequently used in applied studies as an excellent tool for examining causal-predictive associations of models for theory development and testing. Missing data are a common problem in empirical analysis, and PLS-SEM is no exception. A comprehensive review of the PLS-SEM literature reveals a high preference for the listwise deletion and mean imputation methods in dealing with missing values. PLS-SEM researchers often disregard strategies for addressing missing data, such as regression imputation and imputation based on the Expectation Maximization (EM) algorithm. In this study, we investigate the utility of these underutilized techniques for dealing with missing values in PLS-SEM and compare them with mean imputation and listwise deletion. Monte Carlo simulations were conducted based on two prominent social science models: the European Customer Satisfaction Index (ECSI) and the Unified Theory of Acceptance and Use of Technology (UTAUT). Our simulation experiments reveal the outperformance of the regression imputation against the other alternatives in the recovery of model parameters and precision of parameter estimates. Hence, regression imputation merit more widespread adoption for treating missing values when analyzing PLS-SEM studies.