Affect Grid scores were analyzed via ANCOVA, with condition as between-subject factor and pre-condition scores as a covariate. Accuracy and reaction times on the RAT were analyzed via generalized linear mixed models, with condition (3 levels: positive affect induction, tea consumption, water consumption) as between-subject factor, difficulty (3 levels: easy, neutral, difficult) as withinsubject factor, and subject as a random effect. Analysis of reaction time for correct responses only was complicated by the absence of correct responses for difficult items for some participants (24 out of 150 participants). The adjustment for these missing data involved the considerable but necessary assumption that reaction times to difficult tasks could be inferred from reaction times to easier tasks. Specifically, participant’s reaction times on difficult tasks were estimated by their performance relative to the average on easy and neutral tasks. The analysis was conducted employing treatment, difficulty level and their interaction as fixed effects, and subject as a random effect. Assessment of the interaction in question was conducted by comparing Least Squared Means as generated by the GLIMMIX procedure in SAS v9. These LSMeans are predicted population margins, in that they estimate the marginal means over a balanced population. Originality of drawings was analyzed via a Kruskal–Wallis test and reaction times on the unsolvable anagram and the workload items were analyzed via one-way ANOVA, condition as between-subject factor. Log transformation was applied if distributions suggested non-normality and a positive skew. There were ten missing observations for the drawings (due to failure to follow instructions) and one for the anagram, which were considered independent of condition and no corrective action was taken. Hypothesis tests employed a 5% level of significance. Where a significant effect or interaction was identified, a Tukey–Kramer adjustment was applied to pairwise tests.