Response accuracy improved from the first (M = .91) to the final (M = .95) round in the speeded training phase, F(1, 35) = 22.17, p < .001, η
2 = .39. Reaction times (RTs) also decreased from the first (M = 865 ms) to the final (M = 675 ms) round, F(1, 35) = 174.65, p < .001, η
2 = .61. Although participants responded faster to negative than to neutral pairs in the first round (M
neg = 833 ms, M
neut = 896 ms), F(1, 35) = 9.72, p < .05, η
2 = .15, neither accuracy (M
neg = .96, M
neut = .95) nor RT (M
neg = 664 ms, M
neut = 686 ms) showed significant valence effects in the final round, Fs(1, 35) = 1.63 and 1.19, ps > .20. Since we aimed to examine TNT effects on learned memories in the present study, for the following analyses, we excluded any objects for which participants failed to press the correct keys in the final round of the training phase.
A 3 (condition: think, no-think, control) × 2 (valence: negative, neutral) analysis of variance (ANOVA) was performed on the correct recollection rates from the pair memory test. We found a significant effect of condition, F(2, 70) = 3.96, p < .05, η
2 = .12. Neither the effect of valence, F(1, 35) = 0.20, nor the Valence × Condition interaction, F(2, 70) = 0.37, was significant (ps > .60). In line with previous findings that people can control unwanted recollection for both emotional and nonemotional memories (e.g., Murray et al., 2011), a post-hoc Tukey’s test revealed impaired recall in the no-think condition (M
neg = .58, M
neut = .57) as compared with the think condition (M
neg = .67, M
neut = .63), t = 2.67, p < .05, d = 0.51. Performance did not differ significantly between the control condition (M
neg = .59, M
neut = .59) and the other two conditions, ts = 2.15 and 0.52, ps > .05 (Tukey), which is a typical finding for smaller numbers of TNT repetitions (e.g., Anderson & Green, 2001).
Next, we examined the effects of the TNT manipulations on performance in the response memory test. Since slower responses might reflect a random guess in the response memory test, outlier RTs (two SDs above the median for each condition for each participant) were excluded from the analyses in this section.Footnote 1
A 3 (condition) × 2 (valence) ANOVA was performed on accuracy in the response memory test. The accuracy in the speeded training phase was included as a covariate. This analysis revealed a significant Condition × Valence interaction (Fig. 1a), F(2, 70) = 4.66, p < .05, η
2 = .10. Subsequent analyses found a significant condition effect for neutral pairs, F(1, 70) = 4.53, p < .05, η
2 = .11. When associated pictures were neutral, participants were less accurate in the no-think condition (M = .93) than in the control (M = .98) and think (M = .97) conditions, ts = 2.67 and 2.51, ps < .05 (Tukey), ds = 0.37. In contrast, accuracy for the negative pairs was not influenced by the TNT manipulation, F(1, 70) = 1.08, p > .30.
To examine the TNT effects on RTs in the response memory test, a hierarchical linear model analysis (Raudenbush & Bryk, 2002) was employed, with each item as a level-1 unit and each participant as a level-2 unit. The dependent variable was the RT from each correct response for each participant during the response memory test; incorrect responses were not included in this analysis. Predictor variables were condition, valence, the Condition × Valence interaction, and the RT from the speeded training phase for each item.
We found significant effects of condition and of RT from the training phase, Fs(2, 1427) = 7.76 and 60.71, ps < .01, R
2s = .06 and .02. Furthermore, a significant Condition × Valence interaction emerged (Fig. 1b), F(2, 1427) = 4.58, p < .05, R
2 = .05. For neutral pairs, RTs showed a significant condition effect, F(1, 1427) = 11.72, p < .01, R
2 = .03; participants were slower in the no-think condition (M = 677 ms) than in the control (M = 638 ms) and think (M = 615 ms) conditions, ts = 2.85 and 4.54, ps < .05 (Tukey) , ds = 0.29 and 0.49. In contrast, RTs were not affected when pairs were negative, F(1, 1427) = 0.61, p > .50.
Effects of target recollection on behavioral memories
The results above suggest a possible dissociation between picture memories and behavioral responses; whereas the pair memory test showed similar patterns in the recollection of emotional and nonemotional pictures, the response memory test revealed that retrieval suppression inhibited behavioral responses associated with nonemotional pairs only. This dissociation in outcomes argues against the possibility that the accessibility of behavioral responses depends directly on the accessibility of the picture memories.
To further investigate the relationship between the two measures, trials were categorized on the basis of the pair memory performance: trials in which the participants could successfully remember the associated pictures in the pair memory test (remembered pairs), and trials in which they failed to remember the associated pictures (forgotten pairs). We then examined the TNT effects on the response memory test, while including the pair memory performance as an additional independent variable. This did not alter the significant Valence × Condition interactions in the response memory test for either the RTs, F(1, 1421) = 5.02, p < .01, R
2 = .05, or accuracy, F(2, 70) = 3.82, p < .05, R
2 = .09. Participants were less accurate in the no-think than in the control condition when the associated pictures were neutral, irrespective of pair memory test performance: remembered pairs (M
con = .99 vs. M
NT = .95), F(1, 66) = 4.45, p < .05, R
2 = .02, and forgotten pairs (M
con = .96 vs. M
NT = .91), F(1, 53) = 3.13, p = .08, R
2 = .01. In contrast, when the associated pictures were negative, neither the remembered (M
con = .99 vs. M
NT = .98) nor the forgotten (M
con = .95 vs. M
NT = .98) pairs showed significant differences between the no-think and control conditions, F(1, 66) = 0.26, F(1, 53) = 0.57, ps > .40. Likewise, when pictures were neutral, the RTs were slower in the no-think than in the control condition for both remembered pairs (M
con = 641 ms vs. M
NT = 679 ms), F(1, 547) = 4.98, p < .05, R
2 = .01, and forgotten pairs (M
con = 641 ms vs. M
NT = 686 ms), F(1, 360) = 4.00, p < .05, R
2 = .04. In contrast, when the associated pictures were negative, neither the remembered (M
con = 617 ms vs. M
NT = 642 ms) nor the forgotten (M
con = 631 ms vs. M
NT = 633 ms) pairs showed significant differences, F(1, 547) = 1.96, F(1, 360) = 0.00, ps > .15.