Help with lrtest in gsem using imputed data

[Michelle Anderson]

Hi everyone,

I am currently conducting a gender analysis using gsem with imputed data. Both my constrained and unconstrained models run smoothly. However, when I try to perform the lrtest, I get the error message :

lrtest unconstrained constrained, force stats
note: model unconstrained does not contain matrix e(V); rank = 0 assumed.
model unconstrained has missing e(ll)
r(498);


Has anyone encountered this issue before or can provide any insights on how to resolve it? Are there any alternative methods to compare constrained and unconstrained models? Or could the problem be related to the multiple imputation?

Here is my multiple imputation code:

mi set mlong

mi register imputed AN_binary BN_binary BED_binary OSFED_binary MSM SSM COG GAG Gender birthweight mom_age_delivery child_ethnicity mom_marital income mom_edu caesarean_section maternal_BMI mother_EDs autism

mi impute chained (regress) GAG COG birthweight mom_age_delivery maternal_BMI (logit, augment) AN_binary BN_binary BED_binary OSFED_binary MSM SSM Gender child_ethnicity income mom_edu caesarean_section mother_EDs autism (mlogit) mom_marital, add(25) rseed(1234) force


My gender analysis code:

mi estimate, cmdok: gsem (MSM -> AN_binary, ) (MSM -> COG, ) (MSM -> GAG, ) (COG -> AN_binary, ) (GAG -> AN_binary, ) (birthweight -> MSM, ) (mom_age_delivery -> MSM, ) (maternal_BMI -> MSM, ), group(Gender) ginvariant(none) cov( e.COG*e.GAG) nocapslatent

estimates store unconstrained

            constraint 1 [Gender_0]AN_binary:MSM = [Gender_1]AN_binary:MSM
            constraint 2 [Gender_0]COG:MSM = [Gender_1]COG:MSM
            constraint 3 [Gender_0]GAG:MSM = [Gender_1]GAG:MSM
            constraint 4 [Gender_0]AN_binary:COG = [Gender_1]AN_binary:COG          
            constraint 5 [Gender_0]AN_binary:GAG = [Gender_1]AN_binary:GAG
            constraint 6 [Gender_0]MSM:birthweight = [Gender_1]MSM:birthweight            
            constraint 7 [Gender_0]MSM:mom_age_delivery = [Gender_1]MSM:mom_age_delivery        
            constraint 8 [Gender_0]MSM:maternal_BMI = [Gender_1]MSM:maternal_BMI          
            

mi estimate, cmdok: gsem (MSM -> AN_binary, ) (MSM -> COG, ) (MSM -> GAG, ) (COG -> AN_binary, ) (GAG -> AN_binary, ) (birthweight -> MSM, ) (mom_age_delivery -> MSM, ) (maternal_BMI -> MSM, ), group(Gender) ginvariant(none) cov( e.COG*e.GAG) nocapslatent constraints(1 2 3 4 5 6 7 8)

      
estimates store constrained

lrtest unconstrained constrained, force stats


Thank you in advance for your help!

[Rich Goldstein]

the basic problem is that MI estimates are not MLE's; there are ways around this that have been discussed in the forum and a search should find them; I am not sure if you can use -mi xeq- where each estimate (i.e., the estimate for each imputed data set) is followed by its own -lrtest- and then one uses some method to combine (I'm not sure whether Rubin's rules apply here)

[Michelle Anderson]

Thank you very much, Rich. I read your previous post and it was very helpful!