SEM: post-estimation commands for determining good/bad items in latent variables

[Andrew Derkin]

Does Stata SEM have any post-estimation commands that allow you to determine the extent to which an item contributes to good model fit?
And similarly if an item contributes to worse model fit?

E.g. If the model contains a latent variable consisting of 10 items, do any post-estimation commands tell us if any of these 10 items contribute to a poor model?

Thank you

[Richard Williams]

The modification indices can give you insights as to which variables/parts of the model are having problems. See

help sem estat mindices

[wbuchanan]

It also depends on your definition of good/bad items. A good item may not make a statistically significant difference to the overall model fit, but the item information function may indicate that it provides more information over a given range of theta that is not covered as well by other items. The IRT commands would give you estimates of item parameters independent of subjects which you could then use to simplify your model and test the goodness of fit.

[Andrew Derkin]

Thanks for the responses!