The patients with BD were older (M = 40.8 years, SD = 13.3) than the HC participants (M = 29.6 years, SD = 16.0; p = .02) but did not differ in age from the patients with MDD (see ). The HC participants were more likely (75%) to be students than were the participants with BD (28%) or MDD (30%; p = .003). The patients with BD were more likely than patients with MDD to be taking psychiatric medications (77% vs. 25%; p = .001). No differences were found between the groups in general cognitive ability as assessed by Raven's Progressive Matrices (see ).
There were no differences between the bipolar I (n = 17) and II (n = 19) patients on any of the demographic or symptom variables in . Thus, the bipolar I and II patients were combined into one group (n = 36) for the primary analyses.
Number of Sentences Completed on the SST
First, we examined the overall number of sentences completed under the four experimental conditions using a 2 (no load, load) × 2 (no reward, reward) × 3 (BD, MDD, or HC) factorial analysis of variance (ANOVA). There was a main effect of the load manipulation, F
(1, 73) = 67.25, p
< .001. Post hoc comparisons using the Holm–Bonferroni sequential rejective method (Holm, 1979
) indicated that regardless of diagnostic group, participants completed more sentences under no-load as compared with load conditions (p
< .05). There was also a main effect of reward, F
(1, 73) = 8.37, p
= .005, indicating that participants completed more sentences under reward conditions than under nonreward conditions. Number of sentence completions was treated as the denominator when comparing groups on the proportion of negative or hyperpositive sentence completions.
Negative and Hyperpositive Sentence Completion Scores
For each participant, two proportion variables were calculated: (a) the proportion of negative sentence completions in each SST condition, calculated as the number of negative/hyperpositive and negative/neutral word strings completed in the negative (depressive) direction, divided by the total number of completed sentences in that condition, and (b) the proportion of hyperpositive sentences completed in each condition, calculated as the number of hyperpositive/neutral sentences completed in the hyperpositive (manic) direction, divided by the total number of completed sentences in the relevant condition.
There was no effect of sentence type (negative/neutral vs. negative/hyperpositive) on the proportion of sentences completed in the negative direction in any of the participant groups: Main effect of sentence type, F(1, 73) = 0.84, p = .36; Sentence Type × Group interaction, F(2, 73) = 0.45, p = .64. However, there was a main effect of sentence type (negative/hyperpositive vs. hyperpositive/neutral), F(1, 73) = 79.67, p = .001, and a two-way interaction between sentence type and group, F(1, 73) = 12.06, p = .001, on hyperpositive sentence completions, indicating that when given negative/hyperpositive sentences, HC participants completed more in the hyperpositive direction than did participants with BD (Holm–Bonferroni corrected p < .05). To adjust for these differences, we calculated the percentage of hyperpositive sentences as the proportion of sentences completed in the hyperpositive direction only when participants were given hyperpositive/neutral options (e.g., “Superior most I'm others to equal”). The negative/hyperpositive sentences were included only when calculating the total proportion of negative sentence completions. Neither negative nor hyperpositive completion scores were correlated with participants' age, gender, Raven's Progressive Matrices scores, or student or nonstudent status (for all, ps > .05, N = 76).
Was the Load Manipulation Successful?
Prior to testing the primary hypotheses, we examined whether participants in the three groups differed in their ability to recall the six-digit number after the L/NR or L/R tasks. A 2 (L/NR, L/R) × 3 (BD, MDD, HC) factorial ANOVA showed that the BD patients recalled fewer digits than did the controls in the L/NR condition, F(2, 75) = 3.08, p = .05, but not in the L/R condition, F(2, 75) = 1.84, p > .10. Hence, we conducted secondary analyses comparing the groups on the proportion of negative or hyperpositive sentence completions in the load conditions using only the 54 participants (BD, n = 21/36 [58.3%]; MDD, n = 15/20 [75.0%]; HC, n = 18/20 [90.0%]) who recalled five or more digits.
Do Patients With BD Unscramble More Negative Sentences Under Load and Reward Conditions?
The primary study hypothesis was that patients with BD would complete proportionately more sentences in the negative direction than would HC participants under conditions of cognitive load and more sentences in the negative direction than either MDD or HC participants under conditions of reward. To test this hypothesis, we conducted a 2 × 2 × 3 ANOVA, with no load/load and no reward/reward as within-subject variables and group (BD, MDD, HC) as the between-subjects variable; the percentage of negative sentence completions was the dependent variable.
There was a three-way interaction between load, reward, and group, F
(2, 73) = 4.11, p
= .02, indicating that in the absence of load and reward (the NL/NR condition), the participants with BD produced a greater proportion of negative sentences than did the HC participants (Holm–Bonferroni corrected p
< .05; Cohen's d
statistic = 0.87) but did not differ significantly from the participants with MDD (p
> .05; see ).1
The BD participants also produced a greater proportion of negative statements in the NL/R condition than did the HC (corrected p
< .05, d
= 1.32) and MDD participants (p
< .05, d
= 0.80). The MDD group did not differ from the HC group in either the NL/NR or the NL/R condition (p
s > .05).
Figure 2 Percentage of negative sentence completions (number of sentences completed in the negative direction divided by total number of sentences completed) by groups and task conditions. NL/NR = no load/no reward; NL/R = no load/reward; L/NR = load/no reward; (more ...)
Consistent with the hypotheses, in the L/NR condition, both the BD and the MDD groups produced higher proportions of negative sentences than did the HC group (BD vs. HC, d = 1.49; MDD vs. HC, d = 0.83; corrected ps < .05). The BD and MDD groups did not differ significantly (p > .05). Group differences were also observed in the L/R condition, in which the BD and MDD groups produced higher proportions of negative sentences than did the HC group (BD vs. HC, p < .05, d = 1.33; MDD vs. HC, p < .05, d = 0.97) but did not differ from each other (p > .05).
Were Differences in Negative Sentence Completions Due to Baseline Levels of Negative Bias?
Responses to the NL/NR condition were conceptualized as each participant's baseline level of negative or hyperpositive bias during the SST task. To rule out the possibility that significant differences between groups in the load and reward conditions were due to initial differences in baseline NL/NR scores, we calculated an impairment index
, a percentage difference score from the NL/NR condition, for each participant in each of the remaining three conditions (NL/R, L/NR, and L/R). For example, to determine the effect of reward on negative sentence completions in the NL/R condition, a reward impairment index was calculated as the percentage of negative completions in the NL/R condition subtracted from the percentage of negative completions in the NL/NR condition. The use of difference scores to control baseline differences is recommended over analyses of covariance when participants are not randomly assigned to groups (Dallal, 2005
Similar to the results of the factorial ANOVAs, there was a significant difference between the groups in degree of impairment due to the load manipulation, F
(2, 73) = 3.68, p
= .03 (see ). The omnibus difference remained significant after controlling for age (p
< .05) and Raven's Progressive Matrices scores (p
< .05). Holm–Bonferroni corrected post hoc comparisons indicated that the BD group showed a more depressive bias due to the load manipulation than did the HC group (p
< .05; Cohen's d
= 0.71), but the MDD and HC groups did not differ (p
> .05). Participants with BD and MDD did not differ in load impairment scores (p
Figure 3 Impairment indices by groups. Impairment indices were calculated as a percentage difference score between the load/no reward (L/NR), no load/reward (NL/R), and load/reward (L/R) conditions and the no load/no reward (NL/NR) condition. Load = impairment (more ...)
There was also a significant difference between the groups in impairment due to the reward manipulation, F
(2, 73) = 4.02, p
= .02 (see ). This omnibus result remained robust after controlling for age (p
= .03) and Raven's Progressive Matrices scores (p
= .003). The group difference was attributable to the higher reward impairment index scores in the BD than the HC group (corrected p
< .05; d
= 0.79); the remaining pairwise group comparisons did not reach significance (p
> .05). Finally, the group comparison in impairment scores in the L/R condition did not reach significance, F
(2, 73) = 2.70, p
Effects of Current Mood
The HRSD and YMRS scores were square root transformed to adjust for positive skew. A comparison of the three groups revealed an omnibus difference in transformed HRSD scores, F(2, 73) = 20.74, p < .0001. Pairwise group contrasts with Holm–Bonferroni adjustments indicated that the BD (untransformed M = 6.2, SD = 5.4) and MDD (M = 4.0, SD = 2.9) groups each had higher (p < .05) HRSD scores than the HC group (M = 0.65, SD = 1.1) but did not differ from each other (p > .05). As expected, the BD group had higher mean YMRS scores (untransformed M = 4.3, SD = 4.4) than did the MDD (M = 0.95, SD = 1.4) and HC groups (M = 0.40, SD = 0.75), F(2, 75) = 11.7, p < .001.
An analysis of covariance (ANCOVA) revealed that diagnosis was no longer significantly associated with negative sentence completion scores in the baseline NL/NR condition, F(2, 73) = 1.61, p = .21, once HRSD scores were covaried; HRSD scores were independently associated with negative completion scores, F(1, 72) = 11.18, p < .002. The group differences in the load impairment index, F(2, 72) = 2.36, p = .10, and the reward impairment index, F(2, 72) = 2.52, p = .09, were weakened by inclusion of HRSD scores in the ANOVA models, although the effect sizes for these pairwise comparisons were similar to ANOVA comparisons that did not include HRSD scores as covariates (load impairment = .70 vs. .71; reward impairment = .77 vs. .79).
YMRS scores bore no relationship to the negative sentence completion variables (for all, p > .10). The greater proportion of negative sentence completions in the BD than the HC group within the NL/NR condition was not affected by covarying YMRS scores, F(2, 72) = 5.23, p < .01, d = 0.94. However, the group differences in load impairment scores, F(2, 72) = 2.38, p = .10, d = 0.50, and reward impairment scores, F(2, 75) = 2.13, p = .13, d = 0.65, became nonsignificant once YMRS scores were covaried.
Do Patients With BD Unscramble More Hyperpositive Sentences Under Load and Reward Conditions?
Next, we tested the hypothesis that, when given hyperpositive/neutral sentence options, the BD patients would react to cognitive load and reward by completing more sentences in the hyperpositive direction. A 2 × 2 × 3 ANOVA was conducted with no load/load and no reward/reward as within-subject variables and group (BD, MDD, and HC) as the between-subjects variable. The results show a main effect of reward, F(1, 73) = 12.48, p = .001. Under reward conditions, all participants produced more hyperpositive statements than they did under nonreward conditions. However, no group differences or Group × Condition interactions were observed in this analysis (see ). Inclusion of HRSD or YMRS scores in ANCOVA models did not affect the results (all ps > .05).
Percentage of Hyperpositive Completions by Groups Across Conditions
Do Patients With BD Report Higher Levels of Thought Suppression Than Do Comparison Groups?
There were significant differences between the groups in the self-reported frequency of negative thought suppression, based on a 1–7 Likert-type rating (see ), F(2, 73) = 4.87, p = .01. The BD group reported more frequent use of negative thought suppression than did the HC group (corrected p < .05) but not the MDD group (p > .05). There was also a significant difference between groups in self-reported success in suppressing negative thoughts, F(2, 72) = 7.52, p = .001, with both the BD group and the MDD group (p < .05) reporting less success than the HC group. However, these differences could be largely attributed to current mood states: When depressive symptoms (HRSD) were covaried, the group differences in frequency ratings, F(2, 72) = .25, p > .10, and success ratings, F(2, 72) = 2.13, p > .10, were no longer significant.
Group Comparisons on Self-Ratings of Thought Suppression
There was also a significant difference between groups in the self-rated frequency of positive thought suppression, F(2, 73) = 3.63, p = .03. The patients with BD reported a significantly higher frequency than did the patients with MDD (p < .05) but not the controls (p > .05). The difference between BD and MDD patients in the frequency of self-reported positive thought suppression was no longer significant when YMRS scores were covaried, F(2, 72) = 2.18, p > .10.
Does Self-Reported Negative Thought Suppression Predict Performance on the SST?
We predicted that (a) frequent use of negative thought suppression would be associated with a greater tendency to complete negatively valenced sentences during the SST, and (b) these associations would be strongest in the BD and MDD groups, especially under load and reward conditions. We examined this hypothesis in multiple regression models, in which the independent variables were diagnosis and self-reported negative thought suppression scores (centered) and their interaction. The dependent variables were the proportion of negative sentences produced in the four conditions.
Indeed, greater self-reported frequency of negative thought suppression was related to a greater proportion of negative sentence completions across conditions, independent of diagnostic group, F(1, 70) = 12.62, p < .001. In contrast, self-reported success in suppressing negative thoughts was related to fewer negative sentence completions across conditions and diagnoses, F(1, 69) = 6.05, p = .016. There were no significant interactions between diagnosis and self-reported frequency or success of negative thought suppression in predicting negative sentence completions in any condition (all ps > .10).
Next, we sought to determine whether the self-reported frequency of thought suppression affected the observed group differences in the proportion of negative sentence completions. We used ANCOVAs to compare the groups on the load and reward impairment indices while including the frequency and, separately, the success of negative thought suppression as covariates. The group differences in the baseline (NL/NR) condition—in which BD participants completed proportionately more negative sentences than did the controls—remained significant when the frequency of negative thought suppression was covaried, F(2, 72) = 3.33, p = .04, d = 0.58, but dropped to a level that approached significance when success scores for negative thought suppression were covaried, F(2, 71) = 3.07, p = .05, d = 0.54.
The proportional difference between the BD and HC groups in impairment due to the load manipulation was no longer significant when the self-reported frequency of negative thought suppression was covaried, F(2, 72) = 2.29, p = .11, d = 0.57, but remained significant when success ratings were covaried, F(3, 74) = 3.40, p = .04, d = 0.77. Finally, the finding of greater reward impairment in the BD relative to the MDD or HC groups was weakened by covarying the frequency of negative thought suppression, F(3, 75) = 3.02, p < .06, d = 0.72. This group difference remained statistically reliable when success ratings were covaried, F(2, 71) = 4.41, p = .016, d = 0.89.