Reusing SEM estimates

[Richard Rogers]

I feel like I'm missing something basic here. In order to reuse the results of a polychoric correlation in a SEM model, I have to clear the original data set from which the correlation was run. Here's the code, which some of you no doubt recognize, and which works:

polychoric `thevars';
matrix polychR=r(R);
matrix dir;
matlist R;

forvalues i=1/`: word count `thevars' ' { ;
forvalues j=1/`i' { ;
local setcor `setcor' `=polychR[`i',`j']' ;
} ;
if `i' < `: word count `thevars' ' local setcor `setcor' \ ;
} ;

local N = _N ;
clear ; /* here is where I lose the original observations */
ssd init `thevars' ;
ssd set obs `N' ;
ssd set cor `setcor' ;

sem (Y -> `thevars');

I want to do further analysis of the SEM results, which means I have to apply the estimates back on the original observations. I saved the estimates as follows:

estimates save temp, replace;

Then I do the following:

clear;
use original.dta;
estimates use temp;

If I use -sem- without the polychoric correlations, at this point I would use the -predict- command. Here is where I am breaking down. I also could not rerun the SEM model using the estimates as a seed. I admit feeling stupid at this point--this is no doubt simple. What am I missing?

Thanks.

[Joseph Coveney]

I'm not sure whether it will work, but perhaps you can merge or append the dataset to the one where the estimates are saved.

But I thought that the basic supposition behind using ssd in the first place is that you don't have any original dataset, that all you have is a set of summaries.

Also, the reason for feeding a polychoric correlation matrix to factormat or sem is that the endogenous manifest variables are (ordered) categorical. It's not clear to me what you're trying to do with predict after sem, having acknowledged that the nature of the variables that would be used in predictions is not appropriate.