Figure 3.
Evidence of distance effects at transfer. Whiskers represent 99% credible intervals for the estimates. Shaded intervals represent 80% credible intervals. Left, Session-by-session of the “distance effect on trial zero” parameter in the logistic regression analysis of performance (βD in Eqs. 1, 2) during all-pairs sessions, averaged across subjects. Because parameters are measured in log-odds units, no distance effect at transfer would correspond to a parameter value of 0.0. Right, Proportion of correct responses for the critical test pair BD on its first presentation. “Trials only” estimates are based only on the first two BD presentations in each phase. “BD model fit” estimates are based on the intercept of a logistic regression of response accuracy that uses only BD trials. “Full model fit” estimates use Equation 1 to predict BD accuracy using all trials and their symbolic distances. Performance above chance indicates that TI has occurred.