CFA and SEM Builder

[Sumita Sarma]

Dear all,

I have 2 questions:

1. While using the sem builder to do a multi-factor CFA, I get only the unstandardized coefficients. When I add the 'stand' options in the stata command line, though I get the standardized coefficients, these do not get reflected in the diagram. Is there a way to get them in the diagram?

2. I am trying to do a CFA with 8 latent constructs (each with 4-6 indicator variables). For some reason, I do not get the MI when I tried 'estata mindices' with only 6 latent constructs. Please advise if there is a maximum number of factors that can be tested in a CFA, or is it due to lower correlation between couple of constructs.

Thanks in advance for your help and directions.
Regards,
Sumita

[Roman Mostazir]

1. Tools > estimation > estimates > Reporting > Display standardized coefficients and values (check it on) [as it is in Stata version 13.1]

2. It will be more helpful to know what you get from Stata rather knowing only "i do not get the MI". If possible, please provide some output using the code delimiters.

[Sumita Sarma]

Hi Roman,

Thanks for your response. Here is the code (using sem builder) after which when I coded 'estat mindices', it ran forever and did not display the MI (only displayed the DFs):
Also am curious: in the sem builder we need to draw the covariance allow to allow the latent constructs to correlate, which then show up in the code with the 'cov' thing, while writing the cov in code means allowing no covriance; am confused whether I am doing it right....

sem (money_costs -> Q14S1, ) (money_costs -> Q14S2, ) (money_costs -> Q14S4, ) (fin_risk -> Q17S1, ) (fin_risk -> Q17S2, ) (fin_risk -> Q17
> S3, ) (fin_risk -> Q17S4, ) (fin_risk -> Q17S5, ) (perf_risk -> Q16S2, ) (perf_risk -> Q16S4, ) (perf_risk -> Q16S5, ) (perf_risk -> Q16S6,
> ) (social_risk -> Q19S1, ) (social_risk -> Q19S2, ) (social_risk -> Q19S3, ) (social_risk -> Q19S4, ) (social_risk -> Q19S5, ) (psy_emo ->
> Q18S2, ) (psy_emo -> Q18S8, ) (psy_emo -> Q18S9, ) (psy_emo -> Q18S10, ) (psy_emo -> Q18S12, ) (psy_risk -> psy_emo, ) (psy_risk -> psy_se
> lf, ) (information -> importance, ) (information -> collection, ) (information -> comparison, ) (importance -> Q11S1, ) (importance -> Q11S
> 2, ) (importance -> Q11S4, ) (importance -> Q11S5, ) (psy_self -> Q22S1, ) (psy_self -> Q22S2, ) (psy_self -> Q22S3, ) (psy_self -> Q22S4,
> ) (psy_self -> Q22S5, ) (collection -> Q12S2, ) (collection -> Q12S3, ) (collection -> Q12S4, ) (comparison -> Q13S2, ) (comparison -> Q13S
> 3, ) (comparison -> Q13S5, ) (comparison -> Q13S6, ) (comparison -> Q13S9, ), covstruct(_lexogenous, diagonal) latent(money_costs fin_risk
> perf_risk social_risk psy_emo psy_risk information importance psy_self collection comparison ) cov( money_costs*perf_risk fin_risk*money_co
> sts fin_risk*perf_risk social_risk*money_costs social_risk*fin_risk social_risk*perf_risk psy_risk*money_costs psy_risk*fin_risk psy_risk*p
> erf_risk psy_risk*social_risk information*money_costs information*fin_risk information*perf_risk information*social_risk information*psy_ri
> sk) nocapslatent nolog stand

[wbuchanan]

What would you expect the modification indices to yield? It looks like you are freely estimating all of the path coefficients, so which paths could be freed to increase model fit? More importantly, what paths do you believe should be specified differently and why not specify the model in that way when fitting your measurement model? Also, have you considered creating item bundles to reduce the number of parameters being estimated?