Full example of using SEM estimates as regression coefficients (CFA)

[Victoria Lorvig]

Hello,

I am trying to construct indices for well being, stress and environmental concern. I am using confirmatory factor analysis (CFA) because I designed my survey to have certain questions relate to certain factors I planned to create. My observed variables are likert scale and binary. I have successfully made SEM models in the GUI and used estat mindices to identify additional covariates to add to my model. Despite many of the constants being 1, the pvals are very low.

What I am having trouble with is what to do next. I have found many tutorials on how to construct these SEM models, but none that then show how to save the estimates and use them as coefficients in a regression. In my case, this would be creating a binomial logit on whether a respondent replied yes or no to changing their behaviour, or a multinomial logit on how long in the future they plan to change their behaviour. Both cases would include variables in addition to the factors, such as demographics.

Do you know of any tutorials, video or written, that show this process?

Thank you in advance.

Using Stata 15 on windows 10.

[Phil Bromiley]

You didn't get a quick answer. You'll increase your chances of a useful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output, and sample data using dataex.

Generally, you can do predictions most easily using built in Stata procedures like predict and margins. As for any other Stata procedure, look at SEM Postestimation. If you really want to do this the hard way, you can also access the parameters and use them directly. You can usually access coefficients directly. If after running sem you issue sem, coefl it will tell you how to refer to the coefficients. You can work with these _b[] values directly, save them in locals, or whatever. You also should know about issuing ereturn list after an estimation - it shows what Stata has saved - you can also access the parameters this way.

[Richard Williams]

I just came across this the other day. Type

help sem_predict

There are options for saving the values of latent variables. Based on your diagram, I think you would want the latent option, as all latent variables are exogenous.

[wbuchanan]

Victoria Lorvig
Is there any reason for not using -gsem- and regressing the outcomes of interest on the latent variables in a single model?

[Victoria Lorvig]

Richard Williams Thank you, I've found what I need under predictions very easily!

[Victoria Lorvig]

wbuchanan I'm using SEM because when I googled CFA that's what it led me to. mindices has suggested adding covariance between some of my latent variables so I'm assuming that SEM is best fitting option. I only learned about these methods Monday, however, so I'd more than welcome to any advice!

[wbuchanan]

Victoria Lorvig
In short, there's no need to save the predictions from the latent variables since you can regress the outcomes on those latent variables in the same model as the measurement part.