Table 2.

Longitudinal Measurement Invariance for Ethnic/Racial Identity Exploration, Commitment, and Centrality

		Model Fit Indices				Model Comparisons
Variable	Invariance Levels	χ2	df	CFI	TLI	Δχ2	Δdf	p	ΔCFI	ΔTLI	Invariance Achieved
Exploration	Configural	173.391	129	.979	.969						Yes
	Metric	187.616	141	.978	.970	14.225	12	.287	.001	−.001	Yes
	Strong	217.465	153	.969	.962	29.849	12	.003	.009	.008	Yes
	Strict	247.052	168	.962	.957	29.587	15	.014	.007	.005	No
Commitment	Configural	485.538	294	.954	.941						Yes
	Metric	502.550	312	.954	.945	17.012	18	.522	.000	−.004	Yes
	Strong	531.237	330	.952	.945	28.687	18	.052	.002	.000	Yes
	Strict	563.583	351	.949	.945	32.346	21	.054	.003	.000	No
Centralitya	Configural	654.172	374	.929	.906						Yes
	Metric	681.332	395	.928	.909	27.160	21	.166	.001	−.003	Yes
	Strong	758.340	416	.914	.897	77.008	21	.000	.014	.012	No

Note. Longitudinal factorial invariance was tested across four levels from the least restrictive to the most (configural, metric, strong, and strict; Widaman, Ferrer, & Conger, 2010). Configural invariance is established if the same set of items load well on the latent factor. Metric invariance exists if the factor loading of each item is invariant over time. Strong invariance can be achieved if the intercept of each item (i.e., the mean) is invariant over time. Finally, strict invariance exists if the residual variance of each item shows over-time invariance. Each invariance level was established if its model fit did not differ significantly from that of the previous invariance level and at least two of the following three criteria were met: Δχ² significant at p < .05, ΔCFI ≥ .01 and Δnon-normed fit index ≥ .02 (Schwartz et al., 2011; see Cheung & Rensvold, 2002, for a discussion of criteria).

^aWe did not proceed to test strict invariance for centrality because strong invariance was not achieved.