We next compared the fit the Cholesky decomposition to the two uni-directional (PGD→CU & CU→PGD) and reciprocal causation (PGD↔CU) models. As shown in Table 2, the PGD→CU model could be rejected at all ages. The causal CU→PGD and reciprocal PGD↔CU models both provided a good fit to the data as judged by the non-significant changes in log likelihood and lowest sample size adjusted Bayesian Information Criterion (BIC) [which has been shown to outperform the more traditionally used Akaike Information Criterion (66)]. In the reciprocal interaction model, the standardized causal pathways (β1) from PGD→CU at each time period were negative and therefore more likely to be artifactual than substantive. They were also small (−0.14, −0.27 and −0.17) which meant that PGD explained very little variance in CU, so the results more closely resemble the CU→PGD model. Therefore, the more parsimonious CU→PGD causal model was chosen as the best fitting.
