Looking for Invariance in CFA models

Doug Hemken

October 2015

Comparing Measurement Models

Using the MPlus example data:

* Data for a two factor measurement model
infile x1-x6 using "Z:\PUBLIC_WEB\MPlus\Basics\CFA and MIMIC\ex5.1.dat"

// The congeneric model, where everything varies
sem (L1 -> x1-x3)
estimates store congeneric
(500 observations read)


Endogenous variables

Measurement:  x1 x2 x3

Exogenous variables

Latent:       L1

Fitting target model:

Iteration 0:   log likelihood = -2450.3813  
Iteration 1:   log likelihood = -2450.3813  

Structural equation model                       Number of obs     =        500
Estimation method  = ml
Log likelihood     = -2450.3813

 ( 1)  [x1]L1 = 1
------------------------------------------------------------------------------
             |                 OIM
             |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
Measurement  |
  x1 <-      |
          L1 |          1  (constrained)
       _cons |  -.0222186    .062788    -0.35   0.723    -.1452808    .1008436
  -----------+----------------------------------------------------------------
  x2 <-      |
          L1 |   1.125066   .0988703    11.38   0.000      .931284    1.318848
       _cons |   .0255831   .0624333     0.41   0.682    -.0967839    .1479501
  -----------+----------------------------------------------------------------
  x3 <-      |
          L1 |   1.019625   .0888617    11.47   0.000     .8454592    1.193791
       _cons |   .0346609   .0624935     0.55   0.579    -.0878241    .1571459
-------------+----------------------------------------------------------------
    var(e.x1)|   1.063175   .0956754                      .8912609    1.268249
    var(e.x2)|   .7996467   .0998985                      .6259779    1.021498
    var(e.x3)|   1.008739   .0952739                      .8382702    1.213874
      var(L1)|   .9079904    .125112                      .6930973    1.189511
------------------------------------------------------------------------------
LR test of model vs. saturated: chi2(0)   =      0.00, Prob > chi2 =      .

And some example output:

// Tau-equivalent, all loadings equal
//  a unit change on one measure is equal to that on another
sem (L1@a -> x1-x3)
// also written (L1 -> x1@a x2@a x3@a), or as three separate paths
estimates store tauequivalent
lrtest tauequivalent congeneric
Endogenous variables

Measurement:  x1 x2 x3

Exogenous variables

Latent:       L1

Fitting target model:

Iteration 0:   log likelihood = -2452.1656  
Iteration 1:   log likelihood = -2451.4314  
Iteration 2:   log likelihood = -2451.4237  
Iteration 3:   log likelihood = -2451.4237  

Structural equation model                       Number of obs     =        500
Estimation method  = ml
Log likelihood     = -2451.4237

 ( 1)  [x1]L1 - [x3]L1 = 0
 ( 2)  [x2]L1 - [x3]L1 = 0
------------------------------------------------------------------------------
             |                 OIM
             |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
Measurement  |
  x1 <-      |
          L1 |   1.049357   598.7102     0.00   0.999    -1172.401      1174.5
       _cons |  -.0222186    .063565    -0.35   0.727    -.1468037    .1023665
  -----------+----------------------------------------------------------------
  x2 <-      |
          L1 |   1.049357   598.7102     0.00   0.999    -1172.401      1174.5
       _cons |   .0255831   .0613638     0.42   0.677    -.0946877    .1458539
  -----------+----------------------------------------------------------------
  x3 <-      |
          L1 |   1.049357   598.7102     0.00   0.999    -1172.401      1174.5
       _cons |   .0346609   .0629133     0.55   0.582    -.0886469    .1579687
-------------+----------------------------------------------------------------
    var(e.x1)|   1.019347   .0857791                      .8643556     1.20213
    var(e.x2)|   .8818478   .0783523                      .7409079    1.049598
    var(e.x3)|   .9781344   .0834668                      .8274909    1.156202
      var(L1)|   .9089662   1037.221                             .           .
------------------------------------------------------------------------------
LR test of model vs. saturated: chi2(1)   =      2.08, Prob > chi2 = 0.1488



Likelihood-ratio test                                 LR chi2(1)  =      2.08
(Assumption: tauequivalent nested in congeneric)      Prob > chi2 =    0.1488