Single missing data imputation in PLS-SEM

An important source of bias in structural equation modeling (SEM) employing the partial least squares method (PLS) is missing data. Deletion methods, such as listwise and pairwise deletion, have traditionally been used to deal with missing data. These methods are perceived as leading to selective loss of data and significant related biases. Missing data imputation methods, on the other hand, do not resort to deletion. We discuss five single missing data imputation methods in the context of PLS-SEM employing the PLS Mode A algorithm. Among these five methods, two hierarchical methods are new. The results of a Monte Carlo experiment suggest that Multiple Regression Imputation yielded the least biased mean path coefficient estimates, followed by Arithmetic Mean Imputation. With respect to mean loading estimates, Arithmetic Mean Imputation yielded the least biased results, followed by Stochastic Hierarchical Regression Imputation and Hierarchical Regression Imputation. Our study suggests that single missing data imputation methods perform better with PLS-SEM than expected based on past research on their performance with other multivariate analysis techniques such as multiple regression and covariance-based SEM. The methods are implemented as part of the software WarpPLS, starting in version 5.0.