Skip to main content
. Author manuscript; available in PMC: 2021 Nov 1.
Published in final edited form as: Soc Sci Med. 2020 Jul 29;265:113182. doi: 10.1016/j.socscimed.2020.113182

Table A1.

Comparison of fit statistics for measurement invariance by gender

G-squared Degrees of Freedom
Wave 1
# classes Model 1 Model 2 Difference Model 1 Model 2 Difference
2 40523 41784 1260 1769381 1769425 44
3 39808 40757 950 1769335 1769401 66
4 39272 40196 924 1769289 1769377 88
5 38853 39736 884 1769243 1769353 110
6 38494 39312 818 1769197 1769329 132
7 38194 39033 839 1769151 1769305 154
Wave 3
# classes Model 1 Model 2 Difference Model 1 Model 2 Difference
2 27045 28274 1229 331697 331735 38
3 26365 27456 1091 331657 331714 57
4 25957 26937 980 331617 331693 76
5 25669 26499 829 331577 331672 95
6 25415 26200 785 331537 331651 114
7 25210 26002 793 331497 331630 133
Wave 4
# classes Model 1 Model 2 Difference Model 1 Model 2 Difference
2 22479 23017 537 331697 331735 38
3 21643 22282 639 331657 331714 57
4 21315 21892 577 331617 331693 76
5 21053 21679 626 331577 331672 95
6 20841 21500 658 331537 331651 114
7 20626 21329 703 331497 331630 133

Notes: Models 1 and 2 are identical, except Model 2 includes measurement invariance for gender when Model 1 does not. Clustering is not used in either of these models, and gender is a grouping variable for both models. For all classes examined at all waves, interpretation of the G-squared statistic (as a chi-square value given the degrees of freedom) indicates a significant difference (p<.01), demonstrating improved model fit for measurement invariance compared to just using gender as a group.