sem: Very different fit index values in Stata and Mplus

[Chris Martin]

I’m trying to run a simple single-factor model with four continuous indicators. Even though the four indicators have very higher (>.9) intercorrelations, the fit indices in Stata are not good and also it's showing df = 6 rather than 2.

Specifically, in Stata
Chi-square = 1261.533
RMSEA=.768
CFI=.787
TLI=.646
df=6

The sample size is 356.

If I run the same analysis in MPlus
Chi-square = 4.956
RMSEA = .064
CFI = .999
TLI=.997
df=2

For some reason MPlus shows a sample size of 357, so I think it’s including one value with some missing data. Stata is excluding that.

I’m using Stata SE version 16.1 and I have installed all the updates. Here’s the code I’m using . Note that I’m running MPlus from Stata using the runmplus command.

sem (PPR -> PPR1 PPR2 PPR3 PPR4), method(ml) nocapslatent
estat gof, stats(all)
 
runmplus PPR1 PPR2 PPR3 PPR4, model(PPR by PPR1-PPR4;)

Has anyone had similar problems with Stata?

Chris

[Richard Williams]

mPlus defaults to using full information maximum likelihood. In Stata you can say method(mlmv), Or, you CAN tell mplus to use listwise deletion, but check the exact syntax, Or, maybe this would work:

runmplus PPR1 PPR2 PPR3 PPR4 if e(sample), model(PPR by PPR1-PPR4;)

I don't know if one case can cause such huge discrepancies, but you might as well make the samples identical first and see if discrepancies persist.

Also, given that you specify a range of variables for mplus, are they actually in the right order? Maybe try

 
 runmplus PPR1 PPR2 PPR3 PPR4, model(PPR by PPR1 PPR2 PPR3 PPR4;)

[Joseph Coveney]

I’m trying to run a simple single-factor model with four continuous indicators. Even though the four indicators have very higher (>.9) intercorrelations, the fit indices in Stata are not good and also it's showing df = 6 rather than 2.

Has anyone had similar problems with Stata?

No.

A dataset that matches your description gives results that look a whole lot like what MPlus gave you. (See below.) You might want to take closer look.

.ÿ
.ÿversionÿ16.1

.ÿ
.ÿclearÿ*

.ÿ
.ÿsetÿseedÿ`=strreverse("1555797")'

.ÿ
.ÿtempnameÿCorr

.ÿmatrixÿdefineÿ`Corr'ÿ=ÿJ(4,ÿ4,ÿ0.9)ÿ+ÿI(4)ÿ*ÿ0.1

.ÿquietlyÿdrawnormÿppr1ÿppr2ÿppr3ÿppr4,ÿdoubleÿcorr(`Corr')ÿn(357)

.ÿquietlyÿreplaceÿppr4ÿ=ÿ.ÿinÿl

.ÿ
.ÿsemÿ(ppr?ÿ<-ÿPpr),ÿmethod(ml)ÿnocnsreportÿnodescribeÿnofootnoteÿnolog
(1ÿobservationsÿwithÿmissingÿvaluesÿexcluded)

StructuralÿequationÿmodelÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿNumberÿofÿobsÿÿÿÿÿ=ÿÿÿÿÿÿÿÿ356
Estimationÿmethodÿÿ=ÿml
Logÿlikelihoodÿÿÿÿÿ=ÿ-1013.5894

------------------------------------------------------------------------------
ÿÿÿÿÿÿÿÿÿÿÿÿÿ|ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿOIM
ÿÿÿÿÿÿÿÿÿÿÿÿÿ|ÿÿÿÿÿÿCoef.ÿÿÿStd.ÿErr.ÿÿÿÿÿÿzÿÿÿÿP>|z|ÿÿÿÿÿ[95%ÿConf.ÿInterval]
-------------+----------------------------------------------------------------
Measurementÿÿ|
ÿÿppr1ÿÿÿÿÿÿÿ|
ÿÿÿÿÿÿÿÿÿPprÿ|ÿÿÿÿÿÿÿÿÿÿ1ÿÿ(constrained)
ÿÿÿÿÿÿÿ_consÿ|ÿÿÿ.1036677ÿÿÿÿ.053362ÿÿÿÿÿ1.94ÿÿÿ0.052ÿÿÿÿ-.0009199ÿÿÿÿ.2082553
ÿÿ-----------+----------------------------------------------------------------
ÿÿppr2ÿÿÿÿÿÿÿ|
ÿÿÿÿÿÿÿÿÿPprÿ|ÿÿÿ.9780977ÿÿÿ.0254047ÿÿÿÿ38.50ÿÿÿ0.000ÿÿÿÿÿ.9283054ÿÿÿÿÿ1.02789
ÿÿÿÿÿÿÿ_consÿ|ÿÿÿ.1153587ÿÿÿ.0517012ÿÿÿÿÿ2.23ÿÿÿ0.026ÿÿÿÿÿ.0140262ÿÿÿÿ.2166911
ÿÿ-----------+----------------------------------------------------------------
ÿÿppr3ÿÿÿÿÿÿÿ|
ÿÿÿÿÿÿÿÿÿPprÿ|ÿÿÿ.9766735ÿÿÿ.0267324ÿÿÿÿ36.54ÿÿÿ0.000ÿÿÿÿÿÿ.924279ÿÿÿÿ1.029068
ÿÿÿÿÿÿÿ_consÿ|ÿÿÿ.0804812ÿÿÿ.0522193ÿÿÿÿÿ1.54ÿÿÿ0.123ÿÿÿÿ-.0218668ÿÿÿÿ.1828292
ÿÿ-----------+----------------------------------------------------------------
ÿÿppr4ÿÿÿÿÿÿÿ|
ÿÿÿÿÿÿÿÿÿPprÿ|ÿÿÿ1.013662ÿÿÿÿ.025588ÿÿÿÿ39.61ÿÿÿ0.000ÿÿÿÿÿ.9635104ÿÿÿÿ1.063814
ÿÿÿÿÿÿÿ_consÿ|ÿÿÿ.1128209ÿÿÿ.0532564ÿÿÿÿÿ2.12ÿÿÿ0.034ÿÿÿÿÿ.0084403ÿÿÿÿ.2172015
-------------+----------------------------------------------------------------
ÿÿvar(e.ppr1)|ÿÿÿ.1123873ÿÿÿ.0109267ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ.0928881ÿÿÿÿ.1359796
ÿÿvar(e.ppr2)|ÿÿÿ.0893187ÿÿÿ.0092992ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ.0728319ÿÿÿÿ.1095376
ÿÿvar(e.ppr3)|ÿÿÿ.1109965ÿÿÿ.0106814ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ.0919172ÿÿÿÿ.1340362
ÿÿvar(e.ppr4)|ÿÿÿ.0835828ÿÿÿ.0092242ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ.0673253ÿÿÿÿ.1037662
ÿÿÿÿÿvar(Ppr)|ÿÿÿ.9013236ÿÿÿ.0758325ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ.7643025ÿÿÿÿ1.062909
------------------------------------------------------------------------------

.ÿ
.ÿestatÿgof,ÿstats(chi2ÿrmseaÿindices)

----------------------------------------------------------------------------
Fitÿstatisticÿÿÿÿÿÿÿÿ|ÿÿÿÿÿÿValueÿÿÿDescription
---------------------+------------------------------------------------------
Likelihoodÿratioÿÿÿÿÿ|
ÿÿÿÿÿÿÿÿÿÿchi2_ms(2)ÿ|ÿÿÿÿÿÿ0.750ÿÿÿmodelÿvs.ÿsaturated
ÿÿÿÿÿÿÿÿÿÿÿÿpÿ>ÿchi2ÿ|ÿÿÿÿÿÿ0.687
ÿÿÿÿÿÿÿÿÿÿchi2_bs(6)ÿ|ÿÿÿ1994.765ÿÿÿbaselineÿvs.ÿsaturated
ÿÿÿÿÿÿÿÿÿÿÿÿpÿ>ÿchi2ÿ|ÿÿÿÿÿÿ0.000
---------------------+------------------------------------------------------
Populationÿerrorÿÿÿÿÿ|
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿRMSEAÿ|ÿÿÿÿÿÿ0.000ÿÿÿRootÿmeanÿsquaredÿerrorÿofÿapproximation
ÿ90%ÿCI,ÿlowerÿboundÿ|ÿÿÿÿÿÿ0.000
ÿÿÿÿÿÿÿÿÿupperÿboundÿ|ÿÿÿÿÿÿ0.079
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿpcloseÿ|ÿÿÿÿÿÿ0.850ÿÿÿProbabilityÿRMSEAÿ<=ÿ0.05
---------------------+------------------------------------------------------
Baselineÿcomparisonÿÿ|
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿCFIÿ|ÿÿÿÿÿÿ1.000ÿÿÿComparativeÿfitÿindex
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿTLIÿ|ÿÿÿÿÿÿ1.002ÿÿÿTucker-Lewisÿindex
----------------------------------------------------------------------------

.ÿ
.ÿexit

endÿofÿdo-file


.

P.S.

. rename *, upper

. sem (PPR -> PPR1 PPR2 PPR3 PPR4), method(ml) nocapslatent
variable PPR not found;
Perhaps you meant 'PPR' to specify a latent variable.
For 'PPR' to be a valid latent variable specification, 'PPR' must appear in the latent() option.
r(111);

.

[Joseph Coveney]

. . . but this postscript gives you something closer to what you got . . .

rename *, upper
generate double PPR = runiform()
sem (PPR -> PPR1 PPR2 PPR3 PPR4), method(ml) nocapslatent
estat gof, stats(chi2 rmsea indices)

You might want to take closer look at your dataset.

I recommend sticking to Stata-ish conventions about variable name case, which will allow you to use Stata's defaults for naming latent factors.

[Chris Martin]

Thanks, everyone. I realized that I should have specified "latent(PPR)" as well. Once I added that, it worked fine.