SEM/GSEM issue: Latent variables constrained & GSEM not concave

[Dana Rose Garfin]

Hi - two questions involving an SEM model with weighted panel data.

1) I am building an SEM with two latent constructs predicting a measured outcome. I am interested in how both latent constructs (each have three items) predict this outcome. However, I can not seem to get STATA to show me the coefficient for both latent constructs. It keeps constraining the second latent variable to 1. I type "nocnsreport" but am getting the same issue. Any suggestions?

2) In the same model, I was hoping to run GSEM since my outcome is a count variable. However, it is approaching a normal distribution and may not fit the Poisson distribution very well (N=1,479, Mean=3.651116, SD=2.285445, range:0 12). Runs fine in SEM, but won't converge in GSEM. Is this an error with GSEM? Or can I run the model in SEM even though technically this is a count variable?

Thanks to everyone for their time!

[Richard Williams]

Without seeing any code or output it is difficult to say. But, on #1, my guess is that the model is being identified by fixing that path at 1. If so, fix some other path at 1 to identify.

I would solve the first problem, and then see if the 2nd problem persists,

showing code and output may help you get a better answer.

Put another way, my first guess is that you are specifying the model incorrectly, but without seeing what you”ve done we can”t really tell.

[Dana Rose Garfin]

Here is the code & output. Thank you so much for responding! Here is the code & output. Of note, the covariances selected were based off of the "estat gof, stats(all) estat mindices" command but the results don't really change at all either way. Thank you!

sem (L1 -> Q34a_1w11, ) (L1 -> Q34a_2w11, ) (L1 -> Q34a_3w11, ) (hispanicw11 -> L1, ) (ppagew11 -> L1, ) (prehurexpALLw10 -> L1,) (L1 -> climateactionw11, ) (L2 -> Q34a_4w11, ) (L2 -> Q34a_5w11, ) (L2 -> Q34a_6w11, ) (hispanicw11 -> L2, ) (ppagew11 -> L2, ) (educat4 -> L2, ) (prehurexpALLw10 -> L2,) (L2 -> climateactionw11, ) (gender2 -> climateactionw11, ) (ppagew11 -> climateactionw11, ) (prehurexpALLw10 -> climateactionw11, ) [pweight = weightw11], nocnsreport latent(L1 L2) cov( e.L1*e.L2 e.Q34a_1w11*e.Q34a_3w11 e.Q34a_1w11*e.climateactionw11 e.Q34a_> 3w11*e.L2 e.Q34a_1w11*e.Q34a_2w11 e.Q34a_2w11*e.L2) nocapslatent
(33 observations with missing values excluded)

Endogenous variables
Observed: climateactionw11
Measurement: Q34a_1w11 Q34a_2w11 Q34a_3w11 Q34a_4w11 Q34a_5w11 Q34a_6w11
Latent: L1 L2

Exogenous variables
Observed: hispanicw11 ppagew11 prehurexpALLw10 educat4 gender2

Fitting target model:
Iteration 0: log pseudolikelihood = -23013.506 (not concave)
Iteration 1: log pseudolikelihood = -22145.466 (not concave)
Iteration 2: log pseudolikelihood = -21816.103 (not concave)
Iteration 3: log pseudolikelihood = -21711.39
Iteration 4: log pseudolikelihood = -21674.066
Iteration 5: log pseudolikelihood = -21659.742
Iteration 6: log pseudolikelihood = -21659.132
Iteration 7: log pseudolikelihood = -21657.982
Iteration 8: log pseudolikelihood = -21657.978
Iteration 9: log pseudolikelihood = -21657.978

Structural equation model Number of obs = 1,446
Estimation method: ml

Log pseudolikelihood = -21657.978

----------------------------------------------------------------------------------------------------
| Robust
| Coefficient std. err. z P>|z| [95% conf. interval]
-----------------------------------+----------------------------------------------------------------
Structural |
climateactionw11 |
L1 | -.9167698 .4093453 -2.24 0.025 -1.719072 -.1144678
L2 | 1 (constrained)
ppagew11 | .0216435 .0057948 3.73 0.000 .0102859 .0330011
prehurexpALLw10 | .4493384 .1460343 3.08 0.002 .1631164 .7355604
gender2 | .4094415 .1761863 2.32 0.020 .0641228 .7547602
_cons | 1.64713 .3682647 4.47 0.000 .9253444 2.368915
---------------------------------+----------------------------------------------------------------
L1 |
hispanicw11 | .2234043 .0852968 2.62 0.009 .0562256 .3905831
ppagew11 | -.0051013 .001753 -2.91 0.004 -.0085371 -.0016656
prehurexpALLw10 | .1315016 .0368729 3.57 0.000 .0592321 .2037711
---------------------------------+----------------------------------------------------------------
L2 |
hispanicw11 | .2666044 .1158999 2.30 0.021 .0394448 .493764
ppagew11 | -.0065029 .0026582 -2.45 0.014 -.0117129 -.001293
prehurexpALLw10 | .165791 .0495061 3.35 0.001 .0687609 .2628211
educat4 | .2628 .0964698 2.72 0.006 .0737227 .4518774
-----------------------------------+----------------------------------------------------------------
Measurement |
Q34a_1w11 |
L1 | 1 (constrained)
_cons | 1.533683 .1211403 12.66 0.000 1.296252 1.771114
---------------------------------+----------------------------------------------------------------
Q34a_2w11 |
L1 | 1.061395 .1392396 7.62 0.000 .7884902 1.334299
_cons | 1.439356 .1437193 10.02 0.000 1.157671 1.721041
---------------------------------+----------------------------------------------------------------
Q34a_3w11 |
L1 | .9318193 .1089448 8.55 0.000 .7182913 1.145347
_cons | 1.307562 .113607 11.51 0.000 1.084896 1.530228
---------------------------------+----------------------------------------------------------------
Q34a_4w11 |
L2 | .8870135 .2219959 4.00 0.000 .4519096 1.322117
_cons | 1.743394 .1430707 12.19 0.000 1.46298 2.023807
---------------------------------+----------------------------------------------------------------
Q34a_5w11 |
L2 | .9702443 .2415568 4.02 0.000 .4968017 1.443687
_cons | 1.667628 .1519286 10.98 0.000 1.369854 1.965403
---------------------------------+----------------------------------------------------------------
Q34a_6w11 |
L2 | 1.01283 .2474552 4.09 0.000 .5278267 1.497833
_cons | 1.831151 .1598418 11.46 0.000 1.517866 2.144435
-----------------------------------+----------------------------------------------------------------
var(e.Q34a_1w11)| .2489309 .0676796 .146101 .4241353
var(e.Q34a_2w11)| .1245587 .033947 .0730112 .2124999
var(e.Q34a_3w11)| .0392985 .0278365 .0098048 .1575114
var(e.climateactionw11)| 4.359769 .2562559 3.885368 4.892096
var(e.Q34a_4w11)| .2755542 .0741774 .1625807 .4670302
var(e.Q34a_5w11)| .1693925 .0290305 .1210641 .2370136
var(e.Q34a_6w11)| .2573517 .0333456 .1996343 .3317561
var(e.L1)| .2643785 .0720535 .1549673 .4510369
var(e.L2)| .668053 .3060612 .2721713 1.639757
-----------------------------------+----------------------------------------------------------------
cov(e.Q34a_1w11,e.Q34a_2w11)| .0866102 .0369848 2.34 0.019 .0141213 .1590992
cov(e.Q34a_1w11,e.Q34a_3w11)| -.0009754 .0350034 -0.03 0.978 -.0695808 .06763
cov(e.Q34a_1w11,e.climateactionw11)| .1024735 .0543039 1.89 0.059 -.0039601 .2089072
cov(e.Q34a_2w11,e.L2)| -.0216165 .0429665 -0.50 0.615 -.1058293 .0625963
cov(e.Q34a_3w11,e.L2)| -.0475474 .0434546 -1.09 0.274 -.1327169 .037622
cov(e.L1,e.L2)| .2603415 .0946965 2.75 0.006 .0747398 .4459432
----------------------------------------------------------------------------------------------------



.

[Richard Williams]

Hi Dana. I will try to decider this later, but your output would be far easier to read if you used code tags. See the Statalist FAQ.

[Dana Rose Garfin]

Sorry - I could not see how indecipherable it looked before I posted. I will work on trying to figure out how to use the code tags.

[Dana Rose Garfin]

Here it is as a PDF. I deleted the covariances for ease of interpretation. Thanks a million!!!

[Richard Williams]

sem doesn't know what you consider structural and what you consider measurement. It therefore isn't using the constraints you want to identify the latent variables.

For each latent variable you should explicitly set one of the three paths from latent variable to indicator at 1, e.g. you could do

sem (L1@1 -> Q34a_1w11, )  (L2@1 -> Q34a_4w11, ) etc. ...

You could pick other paths to fix at 1.

Or, fix the Latent Variances at 1:

sem ..., var(L1@1) var(L2@1)

I can't test the above but hopefully this will work and give you more sensible estimates.

If interested, my brief overview of sem is at

https://www3.nd.edu/~rwilliam/stats2/l95.pdf

[Dana Rose Garfin]

Thank you! That worked great! I appreciate the link as well.