Table 2.
Model fit statistics for the frequentist model comparison process.
| Model | RMSEA (p < 0.05) | CFI | AIC | BIC |
|---|---|---|---|---|
| 1. D-AR1-CL1 | 0.082 (0.00) | 0.872 | 14313.79 | 15027.62 |
| 2. D-AR2-CL2 | 0.052 (0.356) | 0.970 | 13909.15 | 14901.55 |
| 3. D-AR2-CL1 | 0.047 (0.796) | 0.963 | 13892.92 | 14676.39 |
| 4. D-RI-AR1-CL1 | 0.029 (1.00) | 0.985 | 13782.79 | 14540.15 |
| 5. D-RI-AR1-CL1 time noninvariant | 0.029 (1.00) | 0.979 | 13749.09 | 14227.87 |
| 6. D-RI-AR1-CL1 GCLM | 0.025 (1.00) | 0.986 | 13733.52 | 14299.36 |
| 7. D-RI-AR2-CL1 | 0.029 (1.00) | 0.987 | 13790.91 | 14617.91 |
| 8. D-RI-AR2-CL2 | 0.029 (1.00) | 0.992 | 13813.73 | 14849.65 |
| +9. D-RI-AR1-CL1-MA1 | 0.026 (1.00) | 0.984 | 13728.54 | 14224.74 |
| 10. D-RI-AR1-CL1-MA1 GCLM* | 0.023 (1.00) | 0.988 | 13712.87 | 14226.48 |
| 11. D-RI-AR1-CL1-MA1-CLMA1 | 0.025 (1.00) | 0.986 | 13729.02 | 14277.45 |
| 12. D-RI-AR1-CL1-MA1-CLMA1 GCLM** | 0.020 (1.00) | 0.991 | 13709.60 | 14275.44 |
| 13. RI-CLPM | 0.025 (1.00) | 0.989 | 13763.11 | 14520.47 |
| 14. RI-CLPM time noninvariant | 0.027 (1.00) | 0.982 | 13735.16 | 14213.95 |
N = 574. RMSEA, root mean square error of approximation; p, probability that the RMSEA is below 0.05, RMSEA < 0.06 = good fit (Schreiber et al., 2006), RMSEA < 0.08 = acceptable fit (Brown, 2006), CFI, comparative fit index, CFI > 0.95 = good fit (Schreiber et al., 2006), CFI > 0.90 = acceptable fit (Brown, 2006); AIC, Akaike information criterion; BIC, Bayesian Information Criterion; AR1, autoregressive effect with 1-unit time-lag; AR2, autoregressive effect with 2-unit time-lag; CL1, cross-lagged effect with 1-unit time-lag; CL2, cross-lagged effect with 2-unit time-lag; GCLM, general cross-lagged panel model with indicators for the unit effect freely estimated, except for those set to 1 to obtain identification (Zyphur et al., 2020a); MA1, moving average effect with 1-unit time-lag; CLMA1, cross-lagged moving average effect with 1-unit time-lag. + With six time points, the introduction of moving averages is best modeled as time invariant (Asparouhov and Muthén, 2023); * “mean stability assumption by restricting the factor loadings to equality over time (after the t = 1 occasion …)” (Zyphur et al., 2021, p. 8); ** first loadings free for identification (Martin and Zyphur, 2022).