Hello all and thanks in advance for reading/responding to this. I am relatively inexperienced with GSEM and interpreting odds ratios and, therefore, find myself very confused.
I have been using the SEMbuilder (version 16) to build a GSEM model. I have 6 predictors (two ordinal, three bernoulli (binary) and one latent predictor measured by five bernoulli) loading on to one latent dependent variable. This latent dependent is measured using 9 bernoulli/binary observed variables. I have been using logit across the board as the link. Ideally I would like to present comparable coefficients so have been using estat eform to expoentiate my coefficients in to odds ratios. However, I am unsure as to how I can interpret this.
Firstly, does using a logit link make sense? Is my latent dependent considered continuous (seeing as it's made up of 9 binary variables like a scale)?
And, if so, does this mean I shouldn't use odds ratios in my interpretation (is there anything else I can do if this is the case?)
Just for context, I also plan to use this same model to run a multi-group GSEM so is there anything here that would mean I can't do this.
Apologies if these are very basic questions. TIA and I have included my synthax below in case it's needed
. gsem (Parent_inv -> visit_school, family(bernoulli) link(logit)) (Parent_inv -> knw_teach_reg, family(be
> rnoulli) link(logit)) (Parent_inv -> know_smc, family(bernoulli) link(logit)) (Parent_inv -> study_help,
> family(bernoulli) link(logit)) (Parent_inv -> check_bk, family(bernoulli) link(logit)) (Parent_inv -> t
> ell_story, family(bernoulli) link(logit)) (Parent_inv -> sampch_percpq8, family(bernoulli) link(logit))
> (Parent_inv -> sampch_percpq9, family(bernoulli) link(logit)) (Parent_inv -> sampch_percpq10, family(ber
> noulli) link(logit)) (Parent_inv -> resp_education, family(ordinal) link(logit)) (Parent_inv -> Role_con
> , ) (Parent_inv -> sampch_percpq7, family(bernoulli) link(logit)) (Parent_inv -> sampch_percpq6, family(
> bernoulli) link(logit)) (Parent_inv -> teacher_parent_meet, family(ordinal) link(logit)) (Parent_inv ->
> sampch_percpq11, family(bernoulli) link(logit)) (Role_con -> paresp_1, family(bernoulli) link(logit)) (R
> ole_con -> paresp_2, family(bernoulli) link(logit)) (Role_con -> paresp_3, family(bernoulli) link(logit)
> ) (Role_con -> paresp_4, family(bernoulli) link(logit)) (Role_con -> paresp_5, family(bernoulli) link(lo
> git)), latent(Parent_inv Role_con ) nocapslatent
I have been using the SEMbuilder (version 16) to build a GSEM model. I have 6 predictors (two ordinal, three bernoulli (binary) and one latent predictor measured by five bernoulli) loading on to one latent dependent variable. This latent dependent is measured using 9 bernoulli/binary observed variables. I have been using logit across the board as the link. Ideally I would like to present comparable coefficients so have been using estat eform to expoentiate my coefficients in to odds ratios. However, I am unsure as to how I can interpret this.
Firstly, does using a logit link make sense? Is my latent dependent considered continuous (seeing as it's made up of 9 binary variables like a scale)?
And, if so, does this mean I shouldn't use odds ratios in my interpretation (is there anything else I can do if this is the case?)
Just for context, I also plan to use this same model to run a multi-group GSEM so is there anything here that would mean I can't do this.
Apologies if these are very basic questions. TIA and I have included my synthax below in case it's needed
. gsem (Parent_inv -> visit_school, family(bernoulli) link(logit)) (Parent_inv -> knw_teach_reg, family(be
> rnoulli) link(logit)) (Parent_inv -> know_smc, family(bernoulli) link(logit)) (Parent_inv -> study_help,
> family(bernoulli) link(logit)) (Parent_inv -> check_bk, family(bernoulli) link(logit)) (Parent_inv -> t
> ell_story, family(bernoulli) link(logit)) (Parent_inv -> sampch_percpq8, family(bernoulli) link(logit))
> (Parent_inv -> sampch_percpq9, family(bernoulli) link(logit)) (Parent_inv -> sampch_percpq10, family(ber
> noulli) link(logit)) (Parent_inv -> resp_education, family(ordinal) link(logit)) (Parent_inv -> Role_con
> , ) (Parent_inv -> sampch_percpq7, family(bernoulli) link(logit)) (Parent_inv -> sampch_percpq6, family(
> bernoulli) link(logit)) (Parent_inv -> teacher_parent_meet, family(ordinal) link(logit)) (Parent_inv ->
> sampch_percpq11, family(bernoulli) link(logit)) (Role_con -> paresp_1, family(bernoulli) link(logit)) (R
> ole_con -> paresp_2, family(bernoulli) link(logit)) (Role_con -> paresp_3, family(bernoulli) link(logit)
> ) (Role_con -> paresp_4, family(bernoulli) link(logit)) (Role_con -> paresp_5, family(bernoulli) link(lo
> git)), latent(Parent_inv Role_con ) nocapslatent
