Models were built sequentially, starting with a basic latent growth model comprising the PDA variables over time, modeled as slope and intercept. Different sets of mediators were tested in turn. As per the recommendations of Bollen and Long (1993), models were considered to fit the data if the model chi-square was not significant, the Comparative Fit Index (CFI) was at least .97, and the root mean square error of approximation (RMSEA) was below .10. Models were tested on an iterative basis, and individual paths were added or removed on the basis of modification indices (Jöreskog & Sörbom, 1984).
