Hello,
I am a new user of Mplus. I am testing CFA and SEM model via WLSMV estimator. I am wondering whetherthe WLSMV estimator is robust for non-normally distributed data for CFA and SEM?
WLSMV is for binary & ordinal data. Binary/ordinal data, by definition, is not normally distributed. So yes, WLSMV takes the inherent non-normality of binary & ordinal data into account.
Yes, the Weighted Least Squares Mean and Variance adjusted (WLSMV) estimator is generally considered robust for non-normally distributed data in Confirmatory Factor Analysis (CFA) and Structural Equation Modeling (SEM). The WLSMV estimator does not assume normality, making it a good choice when dealing with ordinal data, categorical variables, or data that deviate from normal distribution.
WLSMV operates by estimating model parameters based on the polychoric correlation matrix, which is specifically suitable for categorical data. It adjusts the chi-square test statistic to account for non-normality, thus providing more reliable estimates and standard errors under these conditions. However, keep in mind that WLSMV performs best with large sample sizes, as small samples may still affect the accuracy and stability of estimates, even with the adjustments.
Using WLSMV with non-normally distributed data is recommended for models involving ordinal or categorical variables where data normality cannot be assumed, although large sample sizes are ideal to maximize robustness and stability.
Yusuf Adeniyi Jamiu
Thank you so much for the answer! Is there any source supporting this argument "the Weighted Least Squares Mean and Variance adjusted (WLSMV) estimator is generally considered robust for non-normally distributed data in Confirmatory Factor Analysis (CFA) and Structural Equation Modeling (SEM)"?
- Badges
- Science topic