NUMBER OF RELEVANT PREMISES

The means and standard deviations for the two groups on the performance measures for number of relevant premises on the Integrative Reasoning problems are shown in Table 5.4.

5.4.2.1 Decision Accuracy

To explore the effect of number of relevant premises, performance one-relevant problems (2-1 and 3-1) was compared with problems where all premises were relevant (2-2 and 3- 3). The 1-1 problems were not included in this analysis since they do not fall clearly into either category. The results are shown in Figure 5.3. Several o f the variables had skewness and kurtosis that were unacceptable, and an arcsine transformation failed to reduce these to acceptable levels. Since no appropriate nonparametric test was available, repeated measures ANOVA was carried out in addition to nonparametric tests. ANOVA with two within-subject factors (condition; number of relevant premises) found that the three-way interaction was not statistically significant (F=0.24, df=l,46, p=0.63); nor were the two-way interactions of group x condition (F=1.05, df=l,46, p=0.31), or group x number of relevant premises (F=1.65, df=l,46, p=0.21) significant. There was no significant main effect of group (F=0.17, df=l,46, p=0.68).

There was no significant interaction between condition and number o f relevant premises (F=0.22, df=l,46, p=0.64), but there was a significant main effect o f number of relevant premises (F=38.71, df=l,46, p<0.001), and the means revealed that in both conditions Ss got more one-relevant than all-relevant problems correct (see Figure 5.3). There was a significant effect o f condition (F=98.54, df=l,46, p<0.001), with both groups getting more problems correct in the memory-aid relative to the standard condition (see Figure 5.3).

Nonparametric tests were also carried out. Mann-Whitney U-tests were used to test group differences on each problem type in the two conditions. The groups were not significantly different in the number of correct answers achieved on one-relevant problems in either the standard (U=218.0, p=0.13) or the memory-aid (U=276.0, p=0.56) conditions; nor were there significant group differences on all-relevant problems in the standard (U=281.5, p=0.89) or memory-aid (U=247.0, p=0.36) conditions. The effect o f the number of relevant premises was assessed using Wilcoxon tests to compare performance across groups on one-relevant versus all-relevant problems. The difference was found to be

significant in both the standard (Z=-4.15, pO.OOOl) and memory-aid (Z=-3.89, pO.OOOl) conditions, with better performance on one-relevant problems. Finally, the effect of condition was assessed using Wilcoxon tests to compare performance across groups in the standard versus memory-aid conditions. The difference was found to be significant for both the one-relevant (Z=-4.77, p<0.0001) and all-relevant (Z=-4.48, p<0.0001) problems, with more correct solutions in the memory-aid condition in both cases. This pattern of findings was consistent with those of the parametric analysis reported above in finding no group differences, but that Ss achieved more correct solutions on one-relevant than all- relevant problems, and in the memory-aid relative to the standard condition.

5.4.2.2 Response Time

The effect of number of relevant premises on decision time was explored by comparing decision time for one-relevant (2-1 and 3-1) and all-relevant (2-2 and 3-3) problems, and the results are shown in Figure 5.4. Some of the variables showed unacceptable skewness and kurtosis, but a log 10 transformation reduced this to acceptable levels, and parametric tests were therefore appropriate. As described in section 5.4.2.1, an overall effect of condition on decision time would be difficult to interpret because of the different measures used in the two experimental conditions, but both conditions were included in the analysis so as reveal any interactions with condition. Repeated measures ANOVA with two within-subject factors (condition; number of relevant premises) showed that the three-way group x condition x number of relevant premises interaction was not significant (F=0.04, df=l,46, p=0.83); nor were the two-way interactions of group with condition (F=0.14, df=l,46, p=0.71) or group with number o f relevant premises (F=0.70, df=l,46, p=0.40) significant. There was no significant main effect of group (F=1.60, df=l,46, p=0.21).

There was a significant condition x number of relevant premises interaction (F=39.85, df=l,46, p<0.001), and paired t-tests were carried out with a post-hoc significance level o f 0.0125. There were significant differences in response time between one- and all- relevant problems in both the standard (t=7.53, df=47, p<0.001) and memory-aid (t=16.48, df=47, p<0.001) conditions. The means revealed that these differences were in different directions. On one-relevant problems, Ss had faster response times in the memory-aid condition relative to the standard condition, whereas for the all-relevant problems the reverse was true (see Figure 5.4). There were significant effects of condition (F=10.39, df=l,46, p=0.002), and number of relevant premises (F=210.73, df=l,46, p<0.001).

Table 5.4 Means (and standard deviations) for num ber of relevant premises in the standard and memory-aid conditions on the Integrative Reasoning problems Dysphoric Standard Memory-aid Controls Standard Memory-aid % Correct solutions One-relevant premise 88.54 99.48 82.81 98.96 (12.18) (2.55) (13.70) (3.53) Arcsine 2.58 3.05 2.37 3.02 (0.46) (0.14) (0.44) (0.19) All-relevant premises 67.19 87.50 67.71 89.06 (20.13) (14.28) (22.70) (18.55) Arcsine 1.99 2.58 2.02 2.37 (0.54) (0.51) (0.61) (0.44)

Mean response time secs.

One-relevant premise 4.28 4.26 4.86 4.53 (1.36) (1.71) (2.10) (1.60) Log 10 0.61 0.60 0.65 0.63 (0.12) (0.16) (0.16) (0.16) All-relevant premises 7.40 10.98 9.04 12.29 (4.67) (5.92) (5.62) (5.56) Log 10 0.80 0.99 0.88 1.05 (0.24) (0.20) (0.25) (0.19)

5.4.2.3 Correlation of Performance with Severity of Depressive Symptomatology

Pearson correlation coefficients were calculated between each of the performance measures and scores on the Beck Depression Inventory (BDI1 and BDI2). This was carried out within each group because the selection criteria for the two groups (dysphoric >10; nondysphoric <6) ensured that, taken together, the scores on the BDI could not have a normal distribution. There were no significant correlations with either BDI1 or BDI2. Table 5.5 shows the results for BDI2. This is the measure which was taken during the actual test session, and might therefore be expected to show a higher correlation with task performance than BDI1 which was completed as part o f the initial screening process (see section 4.3.3.2).

Table 5.5 Correlation of BDI2+ with num ber of relevant premises in the standard and memory-aid conditions on the Integrative Reasoning problems Dysphoric Standard Memory-aid Controls Standard Memory-aid % Correct solutions One-relevant premise .16 .01 .24 -.40 Arcsine -.22 .01 .25 -.40 All-relevant premises .09 .28 -.21 .20 Arcsine .09 -.18 -.20 .20

Mean response time secs.

One-relevant premise -.34 .10 .06 .40 Log 10 -.33 -.05 .08 .46 All-relevant premises -.22 -.10 .19 .19 Log 10 -.28 -.07 .22 .20 5.5 SUMMARY OF RESULTS

1. There was no significant effect of group or interaction with group on any of the measures.

2. Ss in both groups achieved fewer correct solutions on problems with more premises, but this effect was reduced in the memory-aid relative to the standard condition.

3. Ss in both groups took longer to respond to problems with more premises. In the memory-aid condition, Ss were significantly faster in responding to 1-1 problems, but significantly slower on 3-3 problems, relative to the standard condition.

4. Ss in both groups achieved fewer correct solutions on all-relevant relative to one­ relevant problems, and in the standard compared with the memory-aid condition.

5. Ss in both groups were faster on one-relevant premises and slower on the all- relevant problems in the memory-aid condition compared with the standard condition.

6. There were no significant correlations between the performance measures and scores on the Beck Depression Inventory (BDI1 and BDI2).

5.6 DISCUSSION

This experiment was designed to examine the effects of storage and processing demands on an Integrative Reasoning task which varied the number of premises, the number of relevant premises, and the need to store information. There were no significant differences between the dysphoric and control Ss on any task measures, and so the general prediction of dysphoric impairment was not supported. The failure to find any significant group differences on a reasoning task similar to other tasks, such as the Discrimination Learning task reported in chapter IV, found to be sensitive to depressive deficits was the most interesting aspect of this experiment, and one which needs to be examined in more detail. Several specific predictions were made regarding the performance of dysphoric Ss on the Integrative Reasoning task, and these will be considered in turn.

It was predicted that dysphoric Ss would be differentially impaired by the increased storage demands associated with increasing the number of premises. The findings showed that Ss in both groups generally achieved fewer correct solutions and had longer response times as the number of premises increased. This was consistent with the findings reported by Salthouse (1992) and Salthouse et al. (1989; 1990), suggesting that this task manipulation was successful. There were no significant group differences or interactions with group for decision accuracy or response time associated with number of premises. However, for response time, the interaction between group and number of premises did approach significance, and the means indicated that dysphoric Ss tended to be faster than controls as the number of premises increased. This was an unexpected finding, and will be given further consideration later in this discussion. However, there was no evidence that the dysphoric Ss were differentially impaired by increasing storage demands.

Storage demands were also manipulated in the current experiment by the introduction of a memory-aid condition, analogous to the memory-aid condition for the Discrimination Learning problems in chapter IV, in which the premises and the question were presented simultaneously, and left on view until the S made a response. This removed the necessity for Ss to store the premises prior to giving a response. It was predicted that if dysphoric Ss have reduced storage capacity then they should be differentially facilitated by the memory-aid condition. Ss in both groups achieved significantly more correct solutions on 2- and 3-premise problems in the memory-aid relative to the standard condition, while there was no difference between the conditions on the 1-premise problems. This suggests that reducing the need to store the premises in the memory-aid condition facilitated performance, and attenuated the effects of increasing the number of premises for both groups. There were also significant interactions between condition and number of premises and number o f relevant premises in response time, and these will be considered below. However, again there were no significant interactions between group and condition, indicating that dysphoric Ss were not differentially facilitated by the reduced storage demands of the memory-aid condition. This is consistent with the finding that dysphoric Ss were not differentially facilitated by a similar memory-aid manipulation in the Discrimination Learning task described in chapter IV.

It was predicted that dysphoric Ss would be differentially impaired by the increased processing demands when all rather than only one premise was relevant to the answer. Ss in both groups achieved fewer correct solutions and had longer response times on all- relevant relative to one-relevant problems. This differs from the findings reported by Salthouse et al. (1990). They reported that Ss' decision accuracy did not differ significantly on one-relevant compared with all-relevant problems, although their response times were significantly longer for the all-relevant problems. The reason for this discrepancy between the current findings and those of Salthouse et al. (1990) will be considered in more detail below. Again, there were no significant group differences or interactions with group for decision accuracy or response time associated with the number of relevant premises. This suggests the requirement to integrate premises increased task difficulty for Ss in both groups, but dysphoric Ss did not appear to be differentially impaired by the increased demands on processing capacity. There was no interaction of condition with number o f relevant premises for decision accuracy, suggesting that Ss' ability to integrate the premises successfully on all-relevant problems was not facilitated

by the memory-aid condition. This is consistent with the fact that the memory-aid condition reduced storage, but not processing, demands.

There were significant interactions for response times between condition and number of premises, and between condition and number o f relevant premises, and inspection revealed both were cross-over interactions. Ss were faster on one-relevant problems, but slower on all-relevant problems in the memory-aid relative to the standard condition. Similarly, Ss were faster on the 1-premise problems, but slower on the 3-premise problems in the memory-aid relative to the standard condition. The fact that Ss were faster on 1-premise and one-relevant problems in the memory-aid relative to the standard condition is consistent with the predicted facilitation o f performance in the memory-aid condition. However, the fact that Ss were slower on the more demanding 3-premise and all-relevant problems in the memory-aid condition is unexpected, and requires explanation.

The fact that Ss were faster on the more demanding problems in the standard condition suggests that they may have adopted a less demanding strategy for these problems. In both the standard and memory-aid conditions, the solution to each problem was always either 'increases' or 'decreases', and so there was a 50% chance of achieving the correct solution by use of a simple guessing strategy, and therefore guessing could potentially be a relatively successful strategy. An inspection of the mean scores reveals that decision accuracy on the 3-premise problems and the all-relevant problems in the standard condition was above 50%, but much lower than any o f the other problem types. It is therefore possible that some Ss adopted a guessing strategy when the task demands exceeded capacity in the standard condition.

The failure to find any significant correlations between performance measures and scores on the Beck Depression Inventory (BDI1 and BDI2) also needs to be addressed. In section 4.6 it was noted that this could suggest either that there was no relationship between severity of depressive symptomatology (as measured by the BDI) and task performance, or that a relationship did exist, but that the measures used made it difficult to detect. Considering the first alternative, in the current experiment there was no evidence that the dysphoric group was impaired relative to the controls on any measure of performance on the Integrative Reasoning task, and therefore the lack o f a relationship between task performance and severity of depressive symptoms is perhaps unsurprising.

Having outlined the main findings of this experiment, there are two important points that have been identified as needing further consideration: 1) the failure to find any significant group differences or interactions with group on the Integrative Reasoning task; 2) the discrepancy with the findings o f Salthouse and his colleagues in finding a significant reduction in decision accuracy associated with increasing the number of relevant premises in the current experiment.

With regard to the failure to find any evidence o f depressive deficits on the Integrative Reasoning task, there are a number of possible explanations to be considered. One possibility is that the dysphoric students were not sufficiently depressed to show impairment on the experimental task. This explanation seems unlikely since the dysphoric Ss in this experiment had similar mean BDI scores to those o f the Ss in the experiment reported in chapter IV, and also to Ss in a number o f other studies where significant group differences have been reported (e.g. Channon & Baker, 1994; see section 2.3.8.1).

Another possibility is that the Integrative Reasoning task is not sensitive to the effects of dysphoric mood on cognitive function. This might result if the task was either too demanding, resulting in floor effects, or not demanding enough, resulting in ceiling effects. An inspection o f the results indicates that the latter may have been the case, with Ss in both groups close to ceiling on the easier problems (those with fewer premises, fewer relevant premises, and those in the memory-aid condition). Decisions about task design were made on the basis of the findings reported by Salthouse and his colleagues (Salthouse, 1992; Salthouse et al., 1989; Salthouse et al., 1990). The current task was very similar to that reported by Salthouse et al. (1990, Study 2). A comparison o f the performance o f the control group (non-dysphoric undergraduates in their 20s) in the current experiment with the control group (undergraduates in their 20s) in the Salthouse experiment (Salthouse et al., 1990, Study 2) reveals comparable decision accuracy and response time in relation to number of premises. Salthouse et al. (1990, Study 2) reported significant group differences, suggesting that having the control group perform at this level should not preclude the possibility of finding group differences. Thus, it must be concluded that while the Integrative Reasoning task might be sensitive to age-related deficits, it was not sensitive to dysphoric deficits in the current experiment.

A third possibility is that the nature o f the task in some way disguised actual differences between the groups in their task performance, and the pattern o f findings does offer some tentative support for this hypothesis. It was predicted that dysphoric Ss would be differentially impaired by increases in either storage or processing demands, or both. Both storage and processing demands would be greatest on the 3-premise and all-relevant problems in the standard condition, and therefore group differences would be most strongly predicted on these. However, as discussed above, there is some evidence to suggest that Ss may have adopted a guessing strategy on these problems. Since there was a 50% chance o f guessing the correct solution, this was a potentially successful strategy, and could obscure group differences on these problems. There is some evidence to support this hypothesis in the finding of a near-significant interaction between group and condition for response times when the number o f premises was considered. There was some evidence that dysphoric Ss were faster than controls as the number of premises increased, suggesting either that they had greater available capacity than the controls, or a greater tendency to guess as the problems exceeded capacity. Since the dysphoric Ss did not show greater decision accuracy than the controls, a finding that would suggest greater capacity, the possibility that both groups used a guessing strategy, but that dysphoric Ss were more likely to guess seems the most probable explanation.

Having considered the possible reasons for the failure to find the expected depressive deficit on the Integrative Reasoning task, the second question which needs to be addressed is why the current experiment differed from those reported by Salthouse and his colleagues (Salthouse, 1992; Salthouse et al., 1989; Salthouse et al., 1990) in finding a significant interaction between decision accuracy and number of relevant premises. Both the current and earlier Salthouse studies reported a significant effect of number o f relevant premises on response time, with Ss slower to respond as the number o f relevant premises increased. This rules out a simple explanation in terms of different speed-accuracy trade­ offs. Therefore, the first step is to establish whether the discrepancy reflects an actual performance difference, or whether it could be attributable to an artifact of data analysis.

In order to address the question o f whether the discrepancy reflects an actual performance difference, a direct comparison was made o f the relevant data in the current experiment with that reported by Salthouse et al. (1990, Study 2) and Salthouse (1992). These two experiments were selected as being the most similar to the current experiment. In the

current experiment, Figure 5.3 shows a clear separation between one-relevant and all- relevant problems, with poorer performance on the latter. The equivalent figures reported