Anne Fleur Kortekaas-Rijlaarsdam

103 REINFORCEMENT LEARNING IN ADHD: EFFECTS OF MPH 5 as for all possible new stimulus pair configurations of stimuli A-F (AB, AC, AD, AE, AF, BC, BD, BE, BF, CD, CE, CF, DE, DF and EF), based on feedback provided in the learning phase. During the test phase, no feedback was provided. The dependent variable was generalization of learning, which is the transfer of learning from the learning context to the novel stimulus pairs. This variable was defined by the difference in accuracy from the last block of the learning phase (original stimulus pairs AB, CD, EF) to the test phase (novel stimulus pairs only) (see (Königs et al., 2016)): a smaller difference reflected better generalization of learning. Reversal Learning Phase The test phase was followed by a reversal learning phase. As conditioning was found to be strongest for AB and CD pairs, these pairs were used in the reversal phase. To ensure a strong association of the S-R mappings before reversing this association, the reversal learning phase started with a short repetition of the learning phase for AB and CD pairs. Instructions were identical to those in the learning phase. After five adjacent correct responses for both pairs (with a minimum of 10 trials and a maximum of 40 trials) the reversal trials started without warning. In these trials feedback was reversed: in AB pairs, 100% of the B-responses and 0% of the A-responses were rewarded with positive feedback; 100% of the A-responses resulted in negative feedback. For the CD pair, 85% of the D-responses and 15% of the E-responses were rewarded with positive feedback. The reversal phase was terminated when participants performed above- chance level (see above), with a minimum of 10 trials. All children performed above- chance level after the 10 th trial. The dependent variable was accuracy for the first 10 trials after the feedback reversal, which was defined as the proportion correct responses excluding trials suspected of anticipatory responses (reaction time <200 ms) (0.1% of trials). Analyses All analyses were performed using SPSS version 23.0 (IBM, 2015). Demographics, including estimated IQ of the ADHD and TD group were compared using analysis of variance (ANOVA) or χ 2 tests. Normality of residuals of dependent variables was inspected and when necessary outliers or influential cases were rescaled to the nearest observation +1 unit or data were transformed (Tabachnick & Fidell, 2012). Learning task performance Generalized Linear Mixed Models (GLMMs) with a random intercept were used to test the effects of group (ADHD using placebo versus TD) and treatment (MPH versus placebo), entering stimulus pair and either group or treatment as fixed factors.