Confidence Interval for Standardized Beta Coefficients in Linear regression with dummy variables

[Rodrigo Badilla]

Hi all,

I am trying to get Confidence Interval for Standardized Beta Coefficients in Linear regression with dummy variables as independent variables.

I did try with "sem" comand but did show me error:

The following observed variable names will be treated as latent
also I tried with gsem command but did not can use for standarized coefficient.

Do you Know another method to get Confidence Interval for Standardized Beta Coefficients in Linear regression with dummy variables??

Thnks in advance

[Clyde Schechter]

I am interpreting your question to mean that you are interested in a standardized linear regression, and that you are using -sem- to do that, not that you really want to do a structural equations model in the full sense.

If that is the case, then what Stata is telling you is that, by default, -sem- treats any variable whose name begins with an upper case letter as a latent variable. If you are really just using -sem- to do linear regression, then there are no latent variables in your model. So you have two choices. You can specify the -nocapslatent- option in your -sem- command, or you can -rename- all your variables to lower case. Either way, you should then be able to emulate a standardized linear regression with -sem, standardized-.

[Rodrigo Badilla]

Hi Clyde,

Thanks for you reply.

You are right!, Im interesting in a standardized linear regression, and I am using -sem- only to get Confidence Interval for Standardized Beta Coefficients in Linear regression.

Just one little question: How use factor variables using -sem-??

I tried:

sysuse auto, clear
sem(price <- i.rep78 mpg trunk), standardized

but I got error

[Clyde Schechter]

Just one little question: How use factor variables using -sem-??

You can't. -sem- does not support factor variable notation. I don't know why--you'd have to ask StataCorp's developers. But this simply can't be done. You can still get your results from -sem- by using the otherwise largely obsolete -xi- prefix.. -xi:- will create real (not virtual) indicator variables for you and plug those into the -sem- command. It will also handle interactions if you have any (though the necessary syntax is different: see -help xi-). What it will not do is enable you to use -margins- postestimation.

[Rodrigo Badilla]

Thanks Clyde for you reply, I will use xi: to creat dummy variables!
Regards
Rodrigo

[Rodrigo Badilla]

Sorry, but just a little question, I am using as dependent variable a standarized variable (Z-score): Weight_Birth.

When i am use-sem- only to get Confidence Interval for Standardized Beta Coefficients, should use the original variable (non standarized)?? to avoid errors or bias?

[Clyde Schechter]

Yes, use the original, unstandardized, variable. -sem- will then do the standardization for you.

[Rodrigo Badilla]

Super! Thanks Clyde for your time and information!
Regards
Ro