MANOVA indicated an overall difference between groups when we examined the total number of inconsistent choices made, F(6,58) = 5.18, P < .001 (η2 = .349). Significant differences were present between groups for all 6 conditions: fruit, F(1,64) = 12.80, P < .001, landscapes, F(1,64) = 23.81, P < .001, puppies, F(1,64) = 15.59, P < .001, vegetables, F(1,64) = 15.46, P < .001, graded valence, F(1,64) = 11.05, P < .001, and same valence, F(1,64) = 23.18, P < .001. Thus, results indicate that individuals with SZ made significantly more inconsistent choices than CN (see ).
Average Count of Inconsistencies per Condition.
MANOVA also indicated that SZ subjects showed a higher average magnitude of inconsistent ratings than did CN subjects, F(6,58) = 4.08, P = .002 (η2 = .297). Significant differences were found on 4 out of 6 conditions, fruit, F(1,64) = 18.70, P < .001, landscapes, F(1,64) = 07.37, P = .009, graded valence, F(1,64) = 05.10, P = .027, same valence, F(1,64) = 07.37, P = .009. The analyses for puppies, F(1,64) = 0.35, P = .55, and vegetables, F(1,64) = 01.51, P = .22, were nonsignificant but also showed patients to make larger magnitude inconsistent choices. Thus, individuals with SZ show a general pattern of making larger magnitudes of inconsistent choices than CN (see ) (After entering WASI full-scale IQ as a covariate, MANCOVA was significant for the total number of inconsistencies [F = 5.57, P < .001, η2 = .38] and magnitude of inconsistency [F = 3.83, P < .001, η2 = .30]. The group × emotion interaction also remained significant for the graded valence condition after controlling for IQ, F(3,63) = 5.95, P = .001 [η2 = .09]. Thus, our effects cannot be accounted for on the basis of group differences in IQ. We elected not to add these analyses to the main text due to problems related to the overmatching fallacy, whereby matching diagnostic groups on variables that are not independent of the illness causes variance directly attributable to the variable of interest to be removed due to overmatching on a variable, such as IQ. In studies examining patient groups, such as individuals with SZ, the overmatching fallacy may be particularly relevant due to the neurodevelopmental nature of the disorder).
Average Magnitude of Inconsistencies per Condition.
Fig. 3. Mean Rank Order and SE for Stimulus Categories in the Graded Valence Condition. Lower values = more frequently preferred item (ie, rank of 1 = most frequently preferred item); higher values = less frequently preferred item (ie, rank of 12 = least frequently (more ...)
It is important to note that there were a similarly high number of inconsistent choices and large magnitude of inconsistent choices in the same valence condition as the fruit, landscape, puppies, and vegetables conditions. This suggests that failures of transitivity are not limited to conditions with a constrained semantic context but are also at hand when patients are presented with stimuli that are of similar valence but from a variety of semantic contexts.
We also examined whether SZ and CN made preference judgments in a normatively valenced fashion (ie, prefer highly positive > mildly positive > mildly negative > highly negative images) by examining preference assignments in the graded valence condition. Repeated measures ANOVA indicated a significant group (SZ vs CN) × condition (highly positive, mildly positive, mildly negative, highly negative) interaction, F(3,63) = 4.51, P = .01 (η2 = .07), as well as a significant main effect for Condition, F(3,63) = 79.69, P < .001 (η2 = .56); however, the between-subjects effect of group was nonsignificant, F(1,63) = 0.12, P = .73 (η2 = .00). The significant within-subjects effect indicates that the stimulus manipulation was successful and that stimuli were generally preferred in a normative fashion (ie, highly positive > mildly positive > mildly negative > highly negative). Paired-samples t-tests conducted separately for CN and SZ, indicated that CN preferred highly positive > mildly positive (t = −2.55, P = .02) and mildly negative > highly negative (t = −3.29, P = .003). In contrast, SZ patients did not prefer highly positive > mildly positive (t = −1.67, P = .16) nor did they prefer mildly negative > highly negative (t = −1.64, P = .11). Paired-samples t-tests also indicated that both CN and SZ preferred positive items significantly more than negative items (CN: t = −9.86, P < .001; SZ: t = −6.64, P < .001). Thus, although individuals with SZ show a normative pattern of preferring positive over negative items, they made less robust fine-grained distinctions within each valence category than CN (see figure 3).
Correlations between behavioral measures of frequency of inconsistency, magnitude of inconsistency, and rank-order preference on the graded valence condition (ie, the average preference position from 0 to 11 that items within each normatively defined valence category were ranked within that condition) with Chapman scale anhedonia and MATRICS working memory and global impairment are presented in . Given our unique interest in the same valence and graded valence conditions, we presented correlations with these variables specifically; however, we also present correlations with the average inconsistency and magnitude across all 6 conditions for completeness. Results indicated that higher levels of physical anhedonia were associated with both a greater number of inconsistencies and a greater magnitude of inconsistency. Higher physical anhedonia levels were also associated with reduced rank-order preference for highly positive, mildly positive, and highly negative items in the graded valence condition, suggesting that more severe anhedonia is associated with a tendency to like positive stimuli less strongly and dislike negative stimuli less strongly. Working memory was significantly correlated with a greater magnitude of inconsistency but not greater numbers of inconsistent choices; however, general cognitive impairment was associated with both total number of inconsistencies and magnitude of inconsistency in the graded and same valence conditions. Furthermore, when we calculated separate positive and negative valence gradation scores (average rank order of high emotion – average rank order of mild emotion), these measures were significantly correlated with working memory (positive: r = 0.37, P < .03; negative: r = −0.39, P < .02), suggesting that poorer working memory performance is associated with less fine-grained distinctions between highly positive to mildly positive stimuli and mildly negative to highly negative stimuli. Positive and negative valence gradation scores were not significantly correlated with general cognition, suggesting that the ability to make fine-grained valence discriminations is specifically linked to working memory. Correlations between behavioral variables and the total and subscale global scores from the SANS, as well as BPRS positive, negative, and disorganized symptoms, were nonsignificant.
Correlations between Chapman Anhedonia Scales, Working Memory, and Behavioral Measures in Schizophrenia (SZ) Patients and Controls (CN)