Variance constraints strongly influenced model performance in growth mixture modeling: a simulation and empirical study

Table 4 Model fit Indices for growth mixture models of aggressive behavior in the TRacking Adolescent Individuals’ Lives Survey (TRAILS), The Netherlands, 2001–2017: text bolded for the best fitting model for each number of classes

Model	# Classes	BIC	aBIC	AIC	Entropy
0: Unconstrained	1^a	− 2326	− 2367	− 2400	–
	2	− 4393	− 4475	− 4541	0.69
	3	− 4885	− 5009	−5107	0.67
1A: Residual variance time constrained	1^b	− 2300	− 2325	− 2345	–
	2	− 4090	− 4141	− 4182	0.7
	3	− 4553	− 4630	− 4690	0.65
	4	− 4738	− 4840	− 4921	0.66
	5	− 4929	− 5056	− 5157	0.68
1B: Residual variance classes constrained	2	− 3421	− 3506	− 3575	0.56
	3	− 3149	− 3235	− 3303	0.54
	4	− 3261	− 3369	− 3455	0.54
1C: Random effects constrained	2	− 4143	− 4219	− 4280	0.67
1C: Random effects constrained	3	− 4791	− 4902	− 4991	0.66
2A: Residual variance constrained time and classes	2	− 2873	− 2921	− 2959	0.46
	3	− 3029	− 3099	− 3154	0.58
	4	− 3186	− 3278	− 3352	0.55
	5	− 3360	− 3474	− 3565	0.59
2B: Random effects and residual variance time constrained	2	− 3843	− 3888	− 3923	0.66
	3	− 4459	− 4522	− 4573	0.7
	4	− 4685	− 4768	− 4833	0.71
	5	− 4889	−4991	− 5072	0.71
2C: Random effects and residual variance classes constrained	2	− 2705	− 2763	− 2808	0.79
	3	− 2916	− 2989	− 3047	0.78
	4	− 3032	− 3121	− 3192	0.76
	5	− 3155	− 3260	− 3344	0.73

For some models, solutions are not shown for all number of classes because in those instances no solutions could be computed: In Model 0 and 1C for 4 and 5 classes, and in Model 1B for 5 classes
BIC Bayesian Information Criterion, aBIC Adjusted BIC, AIC Aikake Information Criterion
^aThis model is the same as the 1-class models 1B, 1C and 2C,
^bThis model is the same as the 1-class models 2A and 2B

ISSN: 1471-2288