Three-way repeated-measures ANOVA of reward expectancy ratings
| Factor | Mean square | F | Significance |
|---|---|---|---|
| Stage | 0.233 | F(1,17) = 0.320 | 0.579 |
| Phase | 13.498 | F(1,17) = 18.622 | 0.000469 |
| Stimulus | 4.021 | F(1,17) = 1.59 | 0.224 |
| Stage × phase | 0.274 | F(1,17) = 0.437 | 0.518 |
| Stage × stimulus | 99.418 | F(1,17) = 42.526 | 0.000005 |
| Phase × Stimulus | 0.341 | F(1,17) = 0.627 | 0.439 |
| Stage × phase × stimulus | 11.724 | F(1,17) = 8.951 | 0.008 |
There were no outliers in the data and the ratings were normally distributed, as assessed by inspection of a boxplot and Shapiro–Wilk’s test of normality (all p > 0.05). In this analysis, the assumption of sphericity for all three main factors (stage, phase, and stimulus) and their two- and three-way interactions was automatically met, because all these factors had only two levels. As shown above, there was a statistically significant three-way interaction between stage, phase, and stimulus (F(1,17) = 8,961, p = 0.008). Bonferroni corrected post hoc tests comparing the difference in reward expectancy ratings between CS+ and CS− at each stage showed a significantly higher reward expectancy rating of color A compared to color B during late acquisition (p = 0.007). In the reversal stage, a significantly higher differential reward expectancy rating of the new CS+ versus the new CS− was observed (p = 0.017).