When estimating a model in AMOS, I usually request the Standardized RMR (SRMR, Standardized Root Mean Square Residual) from the plugin menu. When the data is full, i.e. has no missing values, the value of the SRMR appears in the SRMR box after the model is estimated. When the SRMR is requested in an AMOS run on data with missing values (as in Example 17 in the AMOS User's Guide),
there is a warning printed in the SRMR box that the "standardized RMR is not defined when some data values are missing."
I know that some fit estimates are not printed if means and intercepts are estimated, as explained in Technote 1476192 . However, I have also noticed that SRMR is printed when means and intercepts are estimated but the data is full. Why is the SRMR unavailable when there is missing data? Is there a work-around to print the SRMR in this situation?
A. The sample and residual moments are also not printed when there is missing data, as noted in the FAQ (Frequently Asked Questions) page on the amosdevelopment web site:
The SRMR is calculated from sample and residual moments, so it cannot be calculated if those matrices are unavailable. This unavailability of sample and residual moments is the presence of missing data is also discussed in Technote 1570992. A work-around is suggested there that could also be applied to obtain the SRMR in the presence of missing data.
1. Run the saturated model with the incomplete data set as input. (Allow covariances between all pairs of observed variables to be estimated.)
AMOS will use FIML (Full Information Maximum Likelihood) to estimate parameters of the model.
Click the check-boxes for "Implied moments" and "All implied moments" in the Output tab of the
View->Analysis Properties dialog before running the model.
The Implied covariance matrix will be the FIML estimate of the sample covariance matrix.
2. Use the FIML covariance matrix for the saturated model as input data for your theoretical model. Steps for entering a covariance matrix and means as input are described in Technote 1476687 and at
Request the SRMR from the Plugins menu, as described in Technote 1477535, before running the model.