Newsom USP 655 SEM Winter 2012

Multigroup SEM Example 1

All Parameters Free Across Groups

data: file=C:\Jason\mplus\semclass\stack1.dat; listwise=on; variable:

names = widow panas1 panas2 panas3 panas4 panas5 panas6 panas7 panas8 panas9 panas10 ; grouping is widow (0=notwidow,1=widow); missing = blank;

analysis: type=general; iterations = 200; model=nomeanstructure; information=expected; model: posaff by panas1-panas5*; posaff@1; negaff by panas6-panas10* ; negaff@1; ! ! Note: by default in Mplus, measurement errors and factor correlations are not constrained to be equal across groups;

Model notwidow: posaff by panas1-panas5*; posaff@1; negaff by panas6-panas10* ; negaff@1; Model widow: posaff by panas1-panas5*; posaff@1; negaff by panas6-panas10* ; negaff@1; ! I've freed all factor loadings but set variances to 1, alternatively, some authors ! have suggested using the highest loading item as a reference loading and freeing ! factor variances; output: stdyx ;

Multigroup structural model example--All parameters free ; SUMMARY OF ANALYSIS Number of groups Number of observations Group NOTWIDOW Group WIDOW Number of dependent variables Number of independent variables Number of continuous latent variables 2 159 40 10 0 2

Grouping variable

WIDOW ML EXPECTED 200 0.500D-04 20

Estimator Information matrix Maximum number of iterations Convergence criterion Maximum number of steepest descent iterations Input data file(s)
1

Note: I only illustrate some of the tests that may be needed (e.g., all parameters equal vs. all parameters free, factor variance comparisons, or other specific paths). For brevity sake, I do not illustrate comparison of variances, covariances, or mean and intercept comparisons. See the handout Multigroup SEM for an overview. By default in Mplus Version 6, analyses with meanstructures set the intercepts to zero in the first group and allow them to be freely estimated in the second group (weak factoral invariance).

Newsom USP 655 SEM Winter 2012

C:\Jason\mplus\semclass\stack1.dat THE MODEL ESTIMATION TERMINATED NORMALLY TESTS OF MODEL FIT Chi-Square Test of Model Fit Value Degrees of Freedom P-Value Chi-Square Contributions From Each Group NOTWIDOW WIDOW 72.686 40.260 112.947 68 0.0005

Chi-Square Test of Model Fit for the Baseline Model Value Degrees of Freedom P-Value CFI/TLI CFI TLI Loglikelihood H0 Value H1 Value Information Criteria Number of Free Parameters Akaike (AIC) Bayesian (BIC) Sample-Size Adjusted BIC (n* = (n + 2) / 24) 42 4000.880 4139.198 4006.140 -1958.440 -1901.967 0.971 0.962 1643.772 90 0.0000

RMSEA (Root Mean Square Error Of Approximation) Estimate 90 Percent C.I. 0.082 0.054

0.108

SRMR (Standardized Root Mean Square Residual) Value MODEL RESULTS Estimate Group NOTWIDOW POSAFF BY PANAS1 PANAS2 PANAS3 PANAS4 PANAS5 NEGAFF BY PANAS6 PANAS7 PANAS8 PANAS9 PANAS10 NEGAFF WITH POSAFF Variances POSAFF NEGAFF S.E. Est./S.E. Two-Tailed P-Value 0.046

0.745 0.865 0.791 0.855 0.830

0.054 0.071 0.055 0.057 0.062

13.734 12.145 14.401 14.881 13.367

0.000 0.000 0.000 0.000 0.000

0.798 0.597 0.748 0.380 0.813

0.054 0.065 0.058 0.054 0.083

14.910 9.234 12.804 7.049 9.777

0.000 0.000 0.000 0.000 0.000

-0.674 1.000 1.000

0.050 0.000 0.000

-13.557 999.000 999.000

0.000 999.000 999.000

Newsom USP 655 SEM Winter 2012

Group WIDOW POSAFF BY PANAS1 PANAS2 PANAS3 PANAS4 PANAS5 NEGAFF BY PANAS6 PANAS7 PANAS8 PANAS9 PANAS10 NEGAFF WITH POSAFF Variances POSAFF NEGAFF

1.092 1.252 1.037 1.054 0.914

0.127 0.160 0.132 0.135 0.136

8.599 7.841 7.825 7.788 6.719

0.000 0.000 0.000 0.000 0.000

1.099 0.952 0.843 0.405 0.676

0.136 0.142 0.139 0.127 0.188

8.075 6.710 6.047 3.188 3.601

0.000 0.000 0.000 0.001 0.000

-0.472

0.129

-3.661

0.000

1.000 1.000

0.000 0.000

999.000 999.000

999.000 999.000

Loadings Only Constrained Equal Across Groups

data: file=C:\Jason\mplus\semclass\stack1.dat; listwise=on; variable:

names = widow panas1 panas2 panas3 panas4 panas5 panas6 panas7 panas8 panas9 panas10 ; grouping is widow (0=notwidow,1=widow); missing = blank;

analysis: type=general; iterations = 200; model=nomeanstructure; information=expected; model: posaff by panas1* panas2-panas5; negaff by panas6* panas7-panas10*;

! !

Note: by default in Mplus, measurement errors and factor correlations are not constrained to be equal across groups;

Model notwidow: posaff by panas1* (1) panas2* (2) panas3 (3) panas4 (4) panas5 (5); posaff@1; negaff by panas6* (6) panas7* (7) panas8* (8) panas9* (9) panas10* (10); negaff@1; Model widow: posaff by panas1* (1) panas2* (2) panas3 (3) panas4 (4) panas5 (5); posaff@1; negaff by panas6* (6) panas7* (7) panas8* (8) panas9* (9) panas10* (10); negaff@1; output: stdyx ;

Newsom USP 655 SEM Winter 2012

INPUT READING TERMINATED NORMALLY Multigroup structural model example--Loadings Only Equal ; SUMMARY OF ANALYSIS Number of groups Number of observations Group NOTWIDOW Group WIDOW Number of dependent variables Number of independent variables Number of continuous latent variables 2 159 40 10 0 2

Variables with special functions Grouping variable WIDOW ML EXPECTED 200 0.500D-04 20

Estimator Information matrix Maximum number of iterations Convergence criterion Maximum number of steepest descent iterations

THE MODEL ESTIMATION TERMINATED NORMALLY

TESTS OF MODEL FIT Chi-Square Test of Model Fit Value Degrees of Freedom P-Value Chi-Square Contributions From Each Group NOTWIDOW WIDOW 80.162 56.407

136.569 78 0.0000

Chi-Square Test of Model Fit for the Baseline Model Value Degrees of Freedom P-Value CFI/TLI CFI TLI Loglikelihood H0 Value H1 Value Information Criteria Number of Free Parameters Akaike (AIC) Bayesian (BIC) Sample-Size Adjusted BIC (n* = (n + 2) / 24) 32 4004.502 4109.888 4008.510 -1970.251 -1901.967 0.962 0.957 1643.772 90 0.0000

RMSEA (Root Mean Square Error Of Approximation) Estimate 0.087 90 Percent C.I. 0.062 SRMR (Standardized Root Mean Square Residual) Value 0.143

0.111

Newsom USP 655 SEM Winter 2012

MODEL RESULTS Estimate Group NOTWIDOW POSAFF BY PANAS1 PANAS2 PANAS3 PANAS4 PANAS5 NEGAFF BY PANAS6 PANAS7 PANAS8 PANAS9 PANAS10 NEGAFF WITH POSAFF Variances POSAFF NEGAFF Group WIDOW POSAFF BY PANAS1 PANAS2 PANAS3 PANAS4 PANAS5 NEGAFF BY PANAS6 PANAS7 PANAS8 PANAS9 PANAS10 NEGAFF WITH POSAFF Variances POSAFF NEGAFF S.E. Est./S.E. Two-Tailed P-Value

0.858 0.979 0.854 0.906 0.856 0.867 0.694 0.786 0.396 0.816 -0.711 1.000 1.000

0.050 0.065 0.051 0.054 0.057 0.051 0.059 0.054 0.050 0.076 0.043 0.000 0.000

17.035 15.071 16.632 16.915 15.016 17.098 11.671 14.484 7.973 10.688 -16.386 999.000 999.000

0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 999.000 999.000

0.858 0.979 0.854 0.906 0.856 0.867 0.694 0.786 0.396 0.816 -0.405 1.000 1.000

0.050 0.065 0.051 0.054 0.057 0.051 0.059 0.054 0.050 0.076 0.139 0.000 0.000

17.035 15.071 16.632 16.915 15.016 17.098 11.671 14.484 7.973 10.688 -2.922 999.000 999.000

0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.003 999.000 999.000

Chi-square comparisons
2
All parameters free Loadings only equal
a

112.947 136.569

df 68 78

Comparison to all-free model 2 df 23.6222

10

p < .10, * p < .05, ** p < .01, *** p < .001

Sample Write-up A multigroup structural equation modeling approach was used to compare men and women on the factor loadings of the positive and negative affect scale. To test for partial measurement invariance (weak factorial invariance; Meredith, 1993) across groups, the chi-square from a model with all parameters allowed to be unequal across groups was compared to the chi-square from a model with only the loadings constrained to be equal across groups. No means or intercepts were estimated in these models. The model with all parameters freely estimated in the two groups, fit the data well (CFI = .971, SRMR = .042), according to fit criteria suggested by Hu and Bentler (1999), although the overall chi-square was significant, 2(68) = 112.947 p < .001. The partial invariance model with loadings constrained to be equal across groups had fit that was significantly poorer, 2(78) = 136.569, p < .001, 2(10) = 23.622, p < .01. The Comparative Fit Index for this model indicated good fit (CFI = .962), but the Standardized Root Mean Square Residual (SRMR = .143) suggested the fit could be improved. Further analyses are required to determine which loadings may differ across groups. The findings suggest that the measurement of the two-factor positive and negative affect scale differs across groups, and, thus, caution may be warranted in comparing these groups. 2

In an actual analysis, the researcher should not stop here. Identifying the source of the difference is important for possible modifications.