Abstract:
|
In many applied settings of mental health research, the questionnaire items in measurement instruments do not approximate continuous, normally distributed variables but instead are ordinal. Properties of these instruments are often most accurately evaluated using structural equation models (SEM) for ordinal data. Most empirical studies have missing data. To overcome this issue, multiple imputation can be performed followed by analyses of the imputed datasets. However, no published article addresses the mechanism of pooling the test of model fit across multiply imputed datasets for models with ordinal variables. To address this gap in the literature, through simulations we evaluated whether the D2 procedure, which was originally developed for pooling m Wald tests of the joint significance of a set of k parameters across m multiply imputed datasets with continuous variables, can be extended to combining model fit statistics of SEM models across multiply imputed datasets with ordinal data. Our findings suggest that the D2 procedure may be a reasonable procedure to use, especially when the analysis model includes auxiliary variables that can compensate for the loss of information.
|