in which case we choose the complex over the simple if the difference is positive. The first part of the BIC difference compares the accuracy of the two models, and the second part compares the complexity as measured by number of parameters. For example, Ahn et al. (2008) compared the four-parameter PVL model described earlier with the another model called the expectancy valence learning model (EVL) ( Busemeyer and Stout, 2002). The EVL model assumed that α=1 (no risk aversion) and it used the delta learning rule. The two models differ by one parameter (α) and they are also non-nested because of the learning rule. The BIC difference favored the PVL model over the simpler EVL model ( Ahn et al., 2008).