Error when using gsem

[Max Coleman]

I'm doing a mediation test with generalized structural equation modeling (gsem). My code looks like this:

local cov "age46 female i.region urbanicity"
    xi: svy: gsem (foreign_born hhinc_continuous `cov' -> DSM_every12, logit) ///
                 (foreign_born `cov' -> hhinc_continuous) ///
                 (2: Black_CaribbeanOrigin -> hhinc_continuous) ///
                 (2: Black_CaribbeanOrigin -> DSM_every12, logit), ///
                  nocaps group(nethrace) ginvariant(none) ///
                  var(e.DSM_every12@v1) var(e.hhinc_continuous@v2)

I get the following error:

invalid covariance specification;
e.DSM_every12 does not identify an error for an uncensored gaussian response with the identity link
an error occurred when svy executed gsem

What am I doing wrong?

UPDATE: I see something in the Stata documentation:

(In gsem, you may not use the cov() option with observed exogenous variables. You also may not use cov() with error terms associated with family (Gaussian, link log.)

Does that mean I simply can't do what I'm trying to do in Stata?

[Clyde Schechter]

Well, you cannot do what you are trying to do, but for a slightly different reason than what you cited in your post.

The problem is that with var(e.DSM_every12@v1) you are trying to constrain the variance of the error term for DSM_every12. But as the outcome variable of a logistic regression, the error variance is automatically (pi^2)/3. You cannot modify that; it's baked in to the use of logistic regression: the "residual" variance is always (pi^2)/3.

[Max Coleman]

Ah, thank you for your help! That's reassuring. Given that the error variance is consistent, it sounds like the solution is to simply delete the term var(e.DSM_every12@v1).