Mediation analysis using GSEM and -ldecomp- w/ categorical outcome??

[Laura Serra]

Hello!

I am looking to identify the direct and indirect effects of three variables (left-right economic values scale; liberal-authoritarian values scale; and having a degree) on vote choice (a categorial variables with several parties).

What I want to test is whether education (thus the degree variable) mediates the effect of left-right and liberal-authoritarian values on vote choice.

I have run a GSEM model because from what I've gathered that's what I need if my outcome is categorical, and to do so I have used the following code:

gsem (lr1 -> degree2, family(bernoulli) link(logit)) (lr1 -> vote, family(multinomial) link(logit)) (libauth -> degree2, family(bernoulli) link(logit)) (libauth -> vote, family(multinomial) link(logit)) (degree2 -> vote, family(multinomial) link(logit))

I have then tried to use the ldecomp command to identify direct and indirect effects through this code:

ldecomp vote, direct(libauth) indirect(degree2) or

but I get the following error message:

an error occurred when bootstrap executed _ldecomp

what am I doing wrong, or what should I do instead...???

Best wishes,

Laura

[Sherine Maui]

Hi Laura,

I've recently posted a similar question (https://www.statalist.org/forums/for...-using-ldecomp) - have you managed to figure out how to use ldecomp when the outcome is categorical (unordered categories)?

[Chris Boudreaux]

This might be of interest to you both

https://stats.idre.ucla.edu/stata/fa...e-group-model/

[Sherine Maui]

Hi Chris, thank you for the link but I think in the example the outcome variable (science) is continuous rather than categorical

[Chris Boudreaux]

Sorry I do not know. Perhaps since your outcome is unordered, you can analyze the categories separately in a logit framework. Then you can be sure your interpretation is correct.