Response accuracy was analyzed with an ANOVA design, involving the factors Age group (younger, older), Learners (positive, negative), Block phase (learning, reversal), Learning condition (positive, negative), and Block half (first, second). The analysis revealed a significant main effect of age group, F(1, 46) = 41.78, p < .0001, #! = .47, which reflects the fact that older adults performed overall worse than younger adults. Furthermore, a significant main effect of block phase, F(1, 46) = 32.24, p < .0001, #! = .39, and a significant interaction between age group and block phase, F(1, 46) = 4.32, p < .04, #! = .05 was obtained. Separate analyses for the two age groups revealed a significant effect of block phase in younger (p’s < .0001, #! = .57) and older adults (p’s < .02, #! = .21). As can be seen in Figure 28 both age groups performed better in the learning compared to the reversal phase and this effect was more pronounced for younger than for older adults.
Figure 28: Mean accuracy for younger and older adults in the learning and the reversal phase, aggregated across learners, learning condition and block half.
Moreover, the analysis showed a significant main effect of learning condition, F(1, 46) = 7.72, p < .008, #! = .06 and an interaction between block phase and learning condition, F(1, 46) = 17.86, p < .0001, #! = .27. Separate analyses for the factor block phase showed a significant effect of learning condition only for the reversal phase (p < .0001, #! = .18), but not for the learning phase (p = .99). This finding reflects the fact that in the reversal phase participants performed better in the negative learning condition (M = .62, SD = .11) than in the positive learning condition (M = .57, SD = .12). In contrast, no effect of learning condition was obtained in the learning phase (negative learning: M = .64, SD = .13, positive learning: M = .64, SD = .14). This finding suggests that in the reversal phase, in which the previously learned mappings produce interference, participants tend to adopt a more conservative response strategy and decide for responses that lead to neutral outcomes in order to avoid negative feedback.
Learning effects: The analysis also revealed significant age differences in the
learning effects that is, it showed a significant effect of block half, F(1, 46) = 137.31, p < .0001, #! = .64, and a significant interaction between age group and block half, F(1, 46) = 30.29, p < .0001, #! = .14. Separate analyses for the two age groups revealed significant effects of block half for younger adults (p < .0001, #! = .86), as well as older adults (p < .0003, #! = .46). As can be observed in Figure 29, a larger increase of accuracy from the first to the second block half was obtained for younger compared to older adults.
Analysis of learning biases: Of most interest in the third experiment were the
learning biases in younger and older adults. As could be expected, according to the definition of positive and negative learners in the this experiment, the analysis revealed a significant interaction between learners, and learning condition, F(1, 46) = 66.28, p < .0001 #! = .52, which reflects the fact that positive learners showed a higher accuracy for the positive compared to the negative learning condition, and vice versa for negative learners. Moreover, a significant interaction between age group, learners and learning condition, F(1, 46) = 7.41, p < .009, #! = .06 was obtained. Post-hoc tests for the two age
groups and the two learner groups showed significant learning biases in overall accuracy in the expected direction for negative and positive learners in younger adults (p’s < .02, #!’s > .39), as well as older adults (p’s < .003, #!’s > .56). As can be seen in Figure 29, the learning biases in overall accuracy were more pronounced for older compared to younger adults, suggesting that the asymmetry in overall performance between learner groups was more pronounced for the elderly. The fact that this pattern of results was obtained for the learning (p < .0001, #! = .57), as well as for reversal phase (p < .0001, #! = .31), suggests that participants responded according to their biases irrespective of interference in the reversal phase. This result validates the distinction between positive and negative learners.
Moreover, the analysis not only revealed significant learning biases in overall accuracy but also with respect to the learning effects, as reflected in a significant interaction between learners, learning condition, and block half, F(1, 46) = 15.56, p < .0003, #! = .23. Separate analyses for the two learner groups showed significant interactions between learning condition and block half for positive learners (p < .0001, #! = .28), but not for negative learners (p < .13).
Figure 29: Mean accuracy in the positive (black) and the negative (grey) learning condition, displayed separately for younger and older adults and for positive and negative learners. Mean accuracy is averaged across block phase and block half.
However, as can be also observed in Figure 30 the absence of an interaction between valence and block half in negative learners is due to the fact that the older negative learners showed a similar increase of accuracy with learning for the positive and the negative learning condition. Post-hoc tests for the two age groups and the two learner groups showed significant interactions between learning condition and block half for younger positive learners (p < .002, #! = .56) and younger negative learners (p < .04, #! = .32). In contrast, for the elderly such an interaction was only obtained for positive learners (p < .009, #! = .49), but not for negative learners (p = .92).
These findings suggest that younger positive learners are biased towards learning better from positive outcomes, whereas younger negative learners are biased towards learning to avoid negative outcomes. For older adults, this pattern of results was only obtained for positive learners, whereas older negative learners were not biased towards better learning in either of the learning conditions.
Figure 30: Mean accuracy learning effects in the positive and the negative learning condition, displayed separately for younger and older adults and positive and negative learners. Learning effects reflect the difference between block half two and block half one and are averaged across the factor block phase.