Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

From |
"JVerkuilen (Gmail)" <jvverkuilen@gmail.com> |

To |
statalist@hsphsun2.harvard.edu |

Subject |
Re: st: SEM |

Date |
Thu, 11 Oct 2012 20:36:51 -0400 |

On Thu, Oct 11, 2012 at 8:14 PM, Tucker, Graeme (Health) <Graeme.Tucker@health.sa.gov.au> wrote: > This is a repeat post (original on 3rd October) with some code in the hope I can elicit a response from someone. > > I find I can fit a full orthogonal EFA in Stata using the "sem" command and get very sensible results. The problem is that the model is unidentified in other popular SEM programs (LISREL, AMOS). How does Stata get around the identification problem? > <snippety doo dah> It's unidentified. My SEM professor, the late R. P. McDonald (yes that one), gave us a really useful rule: Standard errors in an identified model should be proportional to 1/sqrt(n). When they are not, even if the log-likelihood appears OK, it's not. If you convert to a standardized solution it's MUCH easier to see this, so even if you intend to use the unstandardized solution its worth generating the standardized solution for inspection. Here's the standardized output for your model (all I changed was adding "stand" as an option), in which it is brutally apparent that things are bad. I suspect it's just a quirk of fate (or, more likely, simulated data) that the log-likelihood was concave and so you didn't get a bunch of error messages. . use http://www.stata-press.com/data/r12/sem_2fmm . sem (L1 -> a1 a2 a3 a4 a5 c1 c2 c3 c4 c5) (L2 -> a1 a2 a3 a4 a5 c1 c2 c3 c4 c5) , covstruct(_lexogenous, diagonal) latent(L1 L2) stand Endogenous variables Measurement: a1 a2 a3 a4 a5 c1 c2 c3 c4 c5 Exogenous variables Latent: L1 L2 Fitting target model: Iteration 0: log likelihood = -10309.339 (not concave) Iteration 1: log likelihood = -10285.537 (not concave) Iteration 2: log likelihood = -10231.81 (not concave) Iteration 3: log likelihood = -10060.861 (not concave) Iteration 4: log likelihood = -9920.2176 (not concave) Iteration 5: log likelihood = -9726.1648 (not concave) Iteration 6: log likelihood = -9588.4151 (not concave) Iteration 7: log likelihood = -9553.7786 (not concave) Iteration 8: log likelihood = -9540.1666 Iteration 9: log likelihood = -9539.3031 Iteration 10: log likelihood = -9534.884 Iteration 11: log likelihood = -9534.7931 Iteration 12: log likelihood = -9534.793 Structural equation model Number of obs = 216 Estimation method = ml Log likelihood = -9534.793 ( 1) [a1]L1 = 1 ( 2) [a2]L2 = 1 ( 3) [cov(L1,L2)]_cons = 0 ------------------------------------------------------------------------------ | OIM Standardized | Coef. Std. Err. z P>|z| [95% Conf. Interval] -------------+---------------------------------------------------------------- Measurement | a1 <- | L1 | .5115907 15.24519 0.03 0.973 -29.36844 30.39162 L2 | .7461977 10.45207 0.07 0.943 -19.73948 21.23187 -----------+---------------------------------------------------------------- a2 <- | L1 | .4947181 15.47373 0.03 0.974 -29.83324 30.82268 L2 | .7573831 10.10735 0.07 0.940 -19.05266 20.56743 -----------+---------------------------------------------------------------- a3 <- | L1 | .5168972 16.02201 0.03 0.974 -30.88566 31.91945 L2 | .7842178 10.5605 0.07 0.941 -19.91397 21.48241 -----------+---------------------------------------------------------------- a4 <- | L1 | .4698141 14.97014 0.03 0.975 -28.87112 29.81075 L2 | .7327321 9.598579 0.08 0.939 -18.08014 19.5456 -----------+---------------------------------------------------------------- a5 <- | L1 | .4656534 16.90655 0.03 0.978 -32.67058 33.60188 L2 | .8275131 9.513554 0.09 0.931 -17.81871 19.47374 -----------+---------------------------------------------------------------- c1 <- | L1 | -.0649107 17.44735 -0.00 0.997 -34.26109 34.13127 L2 | .8539805 1.326338 0.64 0.520 -1.745595 3.453556 -----------+---------------------------------------------------------------- c2 <- | L1 | -.0322915 17.82959 -0.00 0.999 -34.97764 34.91306 L2 | .8726903 .6600055 1.32 0.186 -.4208968 2.166277 -----------+---------------------------------------------------------------- c3 <- | L1 | -.0827462 17.59861 -0.00 0.996 -34.57539 34.4099 L2 | .8613845 1.690683 0.51 0.610 -2.452292 4.175061 -----------+---------------------------------------------------------------- c4 <- | L1 | -.1043651 17.5395 -0.01 0.995 -34.48116 34.27243 L2 | .8584912 2.132346 0.40 0.687 -3.320831 5.037813 -----------+---------------------------------------------------------------- c5 <- | L1 | -.0156541 18.14773 -0.00 0.999 -35.58455 35.55324 L2 | .8882625 .3202817 2.77 0.006 .2605219 1.516003 -------------+---------------------------------------------------------------- Variance | e.a1 | .1814639 .0257661 .1373814 .2396915 e.a2 | .1816248 .0256626 .1376909 .239577 e.a3 | .1178198 .0192061 .0855976 .1621715 e.a4 | .2423784 .0318532 .1873397 .313587 e.a5 | .098389 .0161366 .0713415 .1356909 e.c1 | .2665038 .0365583 .203675 .3487138 e.c2 | .2373689 .0328135 .1810319 .3112381 e.c3 | .2511698 .0358107 .1899358 .3321453 e.c4 | .2521008 .0363222 .1900793 .3343596 e.c5 | .2107447 .0303307 .1589462 .2794235 L1 | 1 . . . L2 | 1 . . . -------------+---------------------------------------------------------------- Covariance | L1 | L2 | 0 (constrained) ------------------------------------------------------------------------------ LR test of model vs. saturated: chi2(25) = 79.69, Prob > chi2 = 0.0000 -- JVVerkuilen, PhD jvverkuilen@gmail.com "Out beyond ideas of wrong-doing and right-doing there is a field. I'll meet you there. When the soul lies down in that grass the world is too full to talk about." ---Rumi * * For searches and help try: * http://www.stata.com/help.cgi?search * http://www.stata.com/support/faqs/resources/statalist-faq/ * http://www.ats.ucla.edu/stat/stata/

**References**:**st: SEM***From:*"Tucker, Graeme (Health)" <Graeme.Tucker@health.sa.gov.au>

- Prev by Date:
**st: SEM** - Next by Date:
**FW: st: SEM** - Previous by thread:
**st: SEM** - Next by thread:
**FW: st: SEM** - Index(es):