MerQuR+Statistics Glossary

Statistics Glossary → Robust ML estimators (MLM, MLR; Satorra-Bentler)

Robust ML estimators (MLM, MLR; Satorra-Bentler)concept

Estimators for structural equation and confirmatory factor models that keep the ML estimates but correct the standard errors and the chi-square test when the data are not normally distributed.

Standard maximum likelihood (ML) estimation assumes multivariate normal data. With skewed or heavy-tailed data the parameter estimates usually remain consistent, but the standard errors come out too small and the model chi-square is inflated, so the model is rejected unnecessarily. MLM uses the same ML estimates but gives the Satorra-Bentler scaled chi-square and robust standard errors (it requires complete data). MLR uses Huber-White sandwich standard errors and the Yuan-Bentler scaled chi-square, and can also work with missing data (FIML).

A scaling correction factor (c) clearly above 1 indicates a substantial departure from normality. Scaled chi-square values cannot simply be subtracted when comparing two nested models; a scaled difference test needs a special correction. These estimators are meant for continuous but non-normal data; for ordinal items with few categories, methods based on polychoric correlations such as WLSMV are more suitable.

An everyday example

On a survey item, most answers pile up at 'strongly agree' and the distribution is heavily skewed. The model fitted with standard ML has a chi-square of 180, while the Satorra-Bentler scaled chi-square is 125 (a correction factor of about 1.44). The robust method takes into account that part of the misfit comes not from the model but from the non-normality of the data.

In MerQur

It appears as the 'Estimator (ML, MLM, MLR)' option in Confirmatory Factor Analysis, 'Estimator (ML, MLR)' in Structural Equation Modeling and 'Estimator' (ML, MLM Satorra-Bentler, MLR Huber-White, GLS, WLS ADF) in Path Analysis.

Analyses where this concept appears

Confirmatory Factor Analysis (CFA)Structural Equation Modeling (SEM)Path Analysis

Related concepts

Maximum likelihood (ML) estimationMardia's test of multivariate normality