Evidence against mood-congruent attentional bias in ...

Psychiatry Research 230 (2015) 496–505

Contents lists available at ScienceDirect

Psychiatry Research journal homepage: www.elsevier.com/locate/psychres

Evidence against mood-congruent attentional bias in Major Depressive Disorder Philip Cheng n, Stephanie D. Preston, John Jonides, Alicia Hofelich Mohr, Kirti Thummala, Melynda Casement, Courtney Hsing, Patricia J. Deldin Department of Psychology, University of Michigan, 530 Church Street, Ann Arbor, MI 48109, USA

art ic l e i nf o

a b s t r a c t

Article history: Received 4 March 2015 Received in revised form 22 September 2015 Accepted 27 September 2015 Available online 9 October 2015

Depression is consistently associated with biased retrieval and interpretation of affective stimuli, but evidence for depressive bias in earlier cognitive processing, such as attention, is mixed. In five separate experiments, individuals with depression (three experiments with clinically diagnosed major depression, two experiments with dysphoria measured via the Beck Depression Inventory) completed three tasks designed to elicit depressive biases in attention, including selective attention, attentional switching, and attentional inhibition. Selective attention was measured using a modified emotional Stroop task, while attentional switching and inhibition was examined via an emotional task-switching paradigm and an emotional counter task. Results across five different experiments indicate that individuals with depression perform comparably with healthy controls, providing corroboration that depression is not characterized by biases in attentional processes. & 2015 Elsevier Ireland Ltd. All rights reserved.

Keywords: Depression Attention Executive Function Cognitive science

1. Introduction Biased attentional processing has been a popular consideration as a causal mechanism in Major Depressive Disorder (MDD) in the last few decades, as reflected both in the theoretical framework of the etiology and phenomenology of depression (Beck et al., 1987; Nolen-Hoeksema, 1991; Clark and Beck, 1999; Alloy et al., 2000), and in research pursuits (for overview, see Koster et al., 2009). Specifically, it has been purported that individuals with MDD might demonstrate biased deployment of attention towards negatively valenced information in the environment for additional cognitive processing. In turn, the cascade of cognitive events lead to better retrieval and recognition of negative information in memory storage (Bradley et al., 1995), constituting a cognitive vulnerability to the onset and maintenance of MDD. The influence of biased attentional processing as a contributing mechanism in MDD is evident in current interventions. For example, Cognitive Behavioral Therapy identifies “mental filtering” as a cognitive distortion, commonly illustrated with the example of selectively attending to negative criticisms presented in an otherwise positive performance evaluation. Subsequently, interventions such as Attentional Bias Modification have been enacted to correct mental filtering via the practice of attending to more n

Corresponding author. Fax: þ1 734 615 0573. E-mail address: [email protected] (P. Cheng).

http://dx.doi.org/10.1016/j.psychres.2015.09.043 0165-1781/& 2015 Elsevier Ireland Ltd. All rights reserved.

benign or positive stimuli in the environment (Papageorgiou and Wells, 2000). However, the specific cognitive processes that contribute to a depressive “mental filtering” is still unclear. Theoretically, disruptions to multiple components during information processing can lead to the same observed phenomenon. In the example above, possible cognitive biases include (1) enhanced detection of negative criticism during the performance review (i.e. selective attention), (2) enhanced rehearsal of the negative feedback following the review (i.e. rumination), or (3) discriminatory retrieval of the negative criticisms when recounting the review (i.e. recall bias). Delineation of the specific vulnerabilities in the information processing system is important because it enhances the precision of intervention targets, which could potentially increase response to cognitive therapy. While rumination and biased recall of episodic memory in depression has been consistently established (Dalgleish and Werner-Seidler, 2014), the empirical evidence for attentional bias in MDD has been less consistent. While some studies have shown attentional bias in MDD, many of these have been conducted in non-clinical samples (e.g., Whitmer and Banich, 2007; Brailean et al., 2014; Cooper et al., 2014). Furthermore, a plethora of studies are unable to replicate any depression-related biased processing (Mogg et al., 1995; Mathews et al., 1996; Williams et al., 1997; Gotlib et al., 2004a, 2004b). Additionally, many studies showing attentional biases in depression have relied on the dot-probe task (e.g., McCabe and Gotlib, 1995; Gotlib et al., 2004a, 2004b; Leyman

P. Cheng et al. / Psychiatry Research 230 (2015) 496–505

et al., 2007). The dot-probe paradigm arguably requires participants to engage in two different tasks serially: the first is to attend to the emotional stimuli, and the second to identify the location of the dot following the emotional stimuli. While the first half of the task may measure selective attention for mood-congruent stimuli, the second half of the task adds a component of attention switching. However, since the task only measures latency to identifying the dot, it disallows disentanglement of different components of attention switching, such as the processes of orienting towards and disengaging from emotional stimuli (Posner and Petersen, 1989). In order to distinguish these disparate cognitive processes, tasks specific to attention orienting and attentional flexibility may be used. The goal of this study was to systematically delineate the role of attentional biases in MDD, particularly in examining three different components in attention. Three attentional tasks were employed in five experiments; two in dysphoric samples, and three in samples of individuals diagnosed with MDD. The first is an emotional Stroop task that measured selective attention to emotional stimuli, and was completed in three experiments; two were conducted using a larger dysphoric sample in order to maximize power, and one in a sample of individuals diagnosed with MDD. The second task was conducted in a sample of participants diagnosed with MDD, and employed a set-switching/inhibition task that examined attentional flexibility in engaging and disengaging from emotional stimuli. Finally, the third task was also conducted in a sample diagnosed with MDD, and examined attentional flexibility under a higher working memory load using a modified Garavan counting task. All three paradigms in this study required participants to navigate goal-oriented tasks while simultaneously processing emotional faces, thereby enabling an index of how task-performance is influenced by competing emotional stimuli. The first three experiments examined selective attention via an emotional Stroop task (Preston and Stansfield, 2008). The emotional Stroop task examines whether task performance is influenced by deficits in inhibition of task-irrelevant mood-congruent stimuli. The fourth experiment uses a task-switching paradigm to examine if attentional flexibility (setswitching and set-inhibition) is impaired by mood-congruent stimuli. Finally, the fifth experiment further examines set-switching under a higher cognitive load, using a counter task that also requires an additional updating of working memory.

2. Emotional Stroop (experiments 1–3) The emotional Stroop task is a variant of the classic color-naming Stroop task modified to examine emotional biases in selective attention. Previous research using the emotional Stroop task in depression typically replaces the content of the word stimuli from colors (e.g., red, blue, green) to emotional words (e.g., sad, down, tired). Use of this task has produced inconsistent depression related Stroop effects, and it has been suggested that effects are more likely to be detected if depressive schemas are activated prior to the task, or if self-relevant stimuli are used (Mogg and Bradley, 2005). The former appears to suggest that the attentional bias in depression might be state-dependent, and the latter may be confounded by the fact that self-relevant stimuli have likely been rehearsed and elaborated, and therefore may relate to components of information processing other than attention. Alternatively, a different variation of the emotional Stroop task may be more robust in eliciting a depression-related Stroop effect without reliance on priming or self-relevant stimuli. This may be achieved through the use of facial images, as evidenced by the increased semantic accessibility of emotions via facial expressions compared to verbal stimuli (Glaser and Glaser, 1989). Facial

497

expressions are also processed innately and automatically (Izard, 1994), as evinced by automatic facial mimicry and matching selfreported feelings in response to subliminal presentations of facial expressions (Sonnby-Borgström, 2002). This variant of the emotional Stroop task has been previously tested (Preston and Stansfield, 2008; Hofelich and Preston, 2012), with results demonstrating semantic interference at the level specific to the emotion. Together, this suggests that affective biases may be enhanced with the use emotional faces as stimuli. In this variant of the emotional Stroop task, valenced word stimuli are superimposed onto emotional face stimuli, and participants are instructed to selectively attend to the valence of the word. In some trials, the valences of the word and face are congruent, whereas others are incongruent. Congruent trials should yield faster response times than incongruent trials. Furthermore, incongruent trials where the facial emotions (i.e. distracters) are mood-congruent with the participant should yield slower response times. For example, if depression is characterized by attentional bias to mood-congruent stimuli, depressed individuals should exhibit increased response times for incongruent trials with sad faces as distracters. This is because of the additional cognitive effort required to inhibit the negative distracter stimuli, which may be activating task irrelevant and self-preoccupying processes (Holmes, 1974; Dawkins and Furnham, 1989; De Ruiter and Brosschot, 1994). Alternatively, depressed individuals may also show decreased reaction on all trials with mood-congruent stimuli, including both sad words and faces. Three different samples using the emotional Stroop task were examined in the following experiments. Experiment 1 was an archival study in a combined sample of two undergraduate populations (Preston and Stansfield, 2008; Hofelich and Preston, 2012). This non-clinical sample was used in order to maximize sample size, and to examine depression severity (as indexed by the Beck Depression Inventory; Beck et al., 1961) as a continuous variable in a general sample of college students. Data from experiment 2 was from an unpublished archival dataset also completed in a sample of undergraduates, and differs from experiment 1 in the use of varying stimulus presentation times from subliminal to supraliminal. Data from experiments 1 and 2 were collected for the purpose of investigating trait emotionality (empathy, alexithymia) and facial mimicry, and no analyses involving depression were previously published. Finally, experiment 3 was conducted in a sample of individuals diagnosed with MDD using the Structured Clinical Interview for the DSM-IV. In all three experiments, it was hypothesized that if depression is related to mood-congruent attentional biases, results should demonstrate that dysphoric or depressed individuals show increased response times to incongruent trials with sad faces as distracters. Alternatively, depressed or dysphoric participants may also demonstrate a generalized affective bias as indexed by longer reaction times to emotional stimuli (word or face). 2.1. Experiment 1 2.1.1. Methods 2.1.1.1. Participants. One hundred and six undergraduate students from a large midwestern university (54 female; mean age 19.18, range 18–27) participated in the study for course credit or $10. Sample mean on the BDI (mean ¼9.5, SD ¼7.3) fell in the range of normal mood fluctuation, with 10 participants scoring in the moderate to severe range of symptom severity. Participants did not differ by sex or age between the two groups (see Table 1). 2.1.1.2. Materials and procedure. Stimuli in this experiment consisted of emotional adjectives superimposed onto pictures of emotional faces. Stimuli were produced using Adobe Photoshop (Adobe Systems Inc., San Jose, CA), and delivered using E-Prime

498


Table 1 Demographic information for participants in each experiment. Experiments 3, 4, and 5 utilized samples that were diagnosed with Major Depressive Disorder.

Experiment 1 Experiment 2

Sex (M:F)

Age

BDI (SD)

BDI range

54:52 47:32

19.18 (1.47) 18.78 (1.45)

9.46 (7.33) 11.73 (10.71)

0 to 43 0 to 41

HC

Experiment 3 Experiment 4a Experiment 5 a

MDD

Sex (M:F)

Age

Sex (M:F)

Age

9:16 5:18 14:11

23.80 (5.29) 20.13 (1.74) 20.96 (3.20)

9:23 6:24 7:16

24.07 (6.29) 23.23 (5.38) 23.09 (5.27)

Groups differed by age.

version 1.0 run on an LCD monitor positioned at eye level 30 in. away from the participant. Facial stimuli from the Pictures of Facial Affect (PFA; Ekman and Friesen, 1976) were used, and included happy, sad, and angry faces. Overlaid adjectives were utilized from published prototypes (Shaver et al., 1987). Participants were instructed to identify the emotion of the word (happy, angry, sad) using the keyboard as quickly and accurately as possible. All trials were provided with 500 ms of feedback; errors were indicated by a red border around the screen while correct trials were indicated with a green border. Trials were followed by an inter-stimulus interval of 500 ms. A total of 144 trials were included in the task. For additional details, please refer to Hofelich and Preston (2012), which contains different analyses using the same data. Analyses in the referenced paper were focused on differences in facial mimicry on the emotional Stroop task between high and low trait empathy groups. No analyses were conducted using depression scores. Response times from incorrect trials, as well as trials exceeding four standard deviations of participant mean were excluded from analyses. G*Power (Faul et al., 2007) was used to compute the minimum effect size detectable in this sample with adequate power ( 40.8) for the interactions of interest. Partial eta squared was used as the measure of effect size. A partial eta squared (ηp2 ) value of 0.01, 0.06, and 0.14 corresponds to a small, medium, and large effect size, respectively (Cohen, 1969; Richardson, 2011). Power analyses using an inter-correlation value of 0.70 revealed adequate power (4 0.8) to detect a small effect size (ηp2 = 0.011) for the 2-way Congruence Depression interaction (nonsphericity correction ϵ ¼1.00), and a small effect size (ηp2 = 0.006) for the 3-way Emotion Congruence Depression interaction (nonsphericity correction ϵ ¼0.824). 2.1.2. Results Reaction time was submitted to a general linear model (GLM) with Emotion (Happy, Sad, Angry), Congruence (Congruent, Incongruent), and Depression (BDI scores) as independent variables. Results confirmed a Stroop effect, with longer reaction times to incongruent trials (M¼ 943.14, SE¼22.27) compared to congruent trials (M¼845.08, SE¼ 17.02), main effect of Congruence, F(1,104)¼ 32.442, po0.001 (ηp2 = 0.238). An emotional Stroop effect was also detected across the sample, Congruence Emotion, F(2,208)¼ 20.680, po0.001 (ηp2 = 0.166), with the largest emotional Stroop effect for trials with happy faces. The Stroop effect did not differ by depression severity, Congruence Depression, F(1,104)¼.009, p¼ 0.925 (ηp2 < 0.001), nor did the emotional Stroop effect differ by depression severity, Emotion Congruence Group, F(1,104)¼ 0.780, p¼0.460 (ηp2 = 0.007). Two additional analyses using GLM were also conducted in order

to further examine the specific hypothesis that biased processing for specific emotions (i.e., sadness) in depression would be reflected in general reaction times to emotional stimuli. The first model included Face Emotion (Happy, Sad, Angry) and Depression (BDI scores) as independent variables, and the second model included Word Emotion (Happy, Sad, Angry) and Depression (BDI scores) as independent variables. No significant interaction effects were detected, though participants showed faster reaction times to happy words (M¼816.52, SE¼17.85) compared to sad (M¼958.85, SD¼23.39) and angry words (M¼937.44, SE¼21.78), main effect of Word Emotion, F(1,104)¼19.663, po0.001 (ηp2 = 0.159). 2.2. Experiment 2 2.2.1. Methods 2.2.1.1. Participants. Seventy-nine undergraduate students (47 female, 32 male) from a midwestern university participated in the study for course credit. The average age was 18.77, with a range between 18 and 26. Dysphoria in this sample was higher than that of experiment 1, with the sample mean on the BDI (mean ¼11.7, SD¼ 10.2) falling in the range of mild mood disturbance, with 19 participants scoring in the moderate to severe range of symptom severity. Participants did not differ by sex or age between the two groups (see Table 1). 2.2.1.2. Materials and procedure. Stimuli in this experiment were similar to that of experiment one. In this experiment stimulus presentation times varied from subliminal to supraliminal in 8 discrete time bins (16.667 ms, 33.333 ms, 83.333 ms, 100 ms, 150 ms, 200 ms, 300 ms, and 400 ms), randomized across the task. In order to preserve power, presentation times below 200 ms were designated as subliminal, and presentation times 200 ms and longer were designated as supraliminal. Participants were instructed to respond as quickly and as accurately as possible. Only angry and sad stimuli were included in this experiment in order to keep the number of trials manageable. Participants were instructed to identify the emotion of the word (angry, sad) as quickly and accurately as possible. Response times from incorrect trials, as well as trials exceeding four standard deviations of participant mean were excluded from analyses. G*Power (Faul et al., 2007) was used to compute the minimum effect size detectable in this sample with adequate power (4 0.8) for the interactions of interest. Partial eta squared was used as the measure of effect size. A partial eta squared (ηp2 ) value of 0.01, 0.06, and 0.14 corresponds to a small, medium, and large effect size, respectively (Cohen, 1969; Richardson, 2011). Power analyses using an inter-correlation value of 0.53 revealed adequate power (40.8) to detect a moderately small effect size (ηp2 = 0.022) for the 2-way Congruence Depression interaction (nonsphericity correction ϵ ¼1.00), and a small effect size (ηp2 = 0.012) for the 3-way Emotion Congruence Depression interaction (nonsphericity correction ϵ ¼1.00). 2.2.2. Results Reaction time was submitted to a GLM with Emotion (Sad, Angry), Congruence (Congruent, Incongruent), Presentation Time (subliminal, supraliminal), and Depression (BDI scores) as independent variables. Results confirmed a Stroop effect, with longer reaction times in incongruent trials (M¼783.87, SE¼ 19.83) compared to congruent trials (M¼731.15, SE¼16.03), main effect of Congruence, F(1,77)¼ 15.752, po0.001 (ηp2 = 0.170). An emotional Stroop effect was also detected, with larger effects for angry compared to sad trials, Congruence Emotion, F(1,77)¼ 15.363, po0.001. Both the Stroop and emotional Stroop effect also differed by presentation times, with a larger effect in the supraliminal compared to


subliminal trials, Congruence Presentation Time, F(1,77)¼15.363, po0.001 (ηp2 = 0.166), and Emotion Congruence Presentation Time, F(1,77) ¼7.002, p¼0.010 (ηp2 = 0.083), largely due to a shorter reaction times in congruent trials for the supraliminal presentation times. Depression symptom severity did not significantly impact the Stroop effect, Congruence Depression, F(1,77)¼ 0.560, p¼0.774 (ηp2 = 0.007), nor did it impact the emotional Stroop effect Congruence Emotion Depression, F(1,77)¼0.967. p¼0.329 (ηp2 = 0.012). The Stroop and emotional Stroop effects also did not vary based on presentation times and depression symptom severity, Congruence Depression Presentation Time, F(1,77)¼ 0.023, p¼0.879 (ηp2 < 0.001), and Congruence Emotion Depression Presentation Time, F(1,77)¼ 0.271, p¼ 0.604 (ηp2 = 0.004). Two additional analyses using GLM were also conducted in order to further examine the hypothesis that biased processing for specific emotions in depression would be reflected in general reaction times to emotional stimuli. The first model included Face Emotion (Sad, Angry) and Depression (BDI scores) as independent variables, and the second model included Word Emotion (Sad, Angry) and Depression (BDI scores) as independent variables. Reaction times did not differ between sad and angry stimuli, which also did not further differ by depression severity, Word Emotion Depression, F(1,77)¼ 0.954, p ¼0.332 (ηp2 = 0.012), and Face Emotion Depression, F(1,77) ¼1.744, p ¼ 0.191 (ηp2 = 0.022). 2.3. Experiment 3 2.3.1. Methods 2.3.1.1. Participants. Participants between the ages of 18 and 45 were recruited through flyers and internet advertisements posted in a small midwestern city. Interested participants completed an initial phone screen, followed by the completion of the Structured Clinical Interview for the DSM-IV (SCID; First et al., 2002) conducted by advanced doctoral students in clinical psychology, or doctoral level clinicians. Eligible participants provided written consent prior to the SCID. Interviews were recorded and randomly selected for inter-rater reliability. Exclusion criteria included head injury resulting in loss of consciousness, seizures, major medical illness, developmental disabilities, and co-morbid Axis I disorders (with the exception of secondary anxiety due to its high comorbidity). All subjects were right-handed and spoke English as their primary language. This sample included 25 healthy controls (HC; 16 female) and 32 participants with MDD (23 female). Participants with MDD had not used any antidepressant medications within the past 6 weeks. Groups showed significant differences on BDI scores, t(55) ¼ 7.846, p o0.001 (MDD ¼19.34, HC ¼ 1.16), and did not differ by sex or age (see Table 1). The experiment was approved by the Institutional Review Board. 2.3.1.2. Materials and procedure. The emotional Stroop task used in this experiment is similar to that of experiment one. Happy, sad, angry, and neutral stimuli were used in this experiment, and adjectives were derived from the Affective Norms for English Words list (ANEW; Bradley and Lang, 1999). Facial stimuli were produced using Adobe Photoshop (Adobe Systems Inc., San Jose, CA), and delivered using E-Prime version 1.0 run on an LCD monitor positioned at eye level 30 in. away from the participant. G*Power (Faul et al., 2007) was used to compute the minimum effect size detectable in this sample with adequate power (40.8) for the interactions of interest. Partial eta squared was used as the measure of effect size. A partial eta squared (ηp2 ) value of 0.01, 0.06, and 0.14 corresponds to a small, medium, and large effect size, respectively (Cohen, 1969; Richardson, 2011). Power analyses using an inter-

499

correlation value of 0.75 revealed adequate power (40.8) to detect a small effect size (ηp2 = 0.016) for the 2-way Congruence Group interaction (nonsphericity correction ϵ ¼1.00), and a small effect size (ηp2 = 0.011) for the 3-way Emotion Congruence Depression interaction (nonsphericity correction¼ 0.829). 2.3.2. Results Reaction times to trials were submitted to a 3-way Emotion (Happy, Sad, Angry, and Neutral) Congruence (Congruent Trials, Incongruent Trials) Group (HC, MDD) repeated-measures analysis of variance. Results confirmed a Stroop effect with longer reaction times to incongruent trials (M ¼1290.16, SE¼49.98) compared to congruent trials (M ¼1255.59, SE ¼52.30), main effect of Congruence, F(1,55) ¼5.909, p ¼0.018 (ηp2 = 0.097). An emotional Stroop effect was also detected, F(2.245,123.476) ¼12.260, po 0.001 (ηp2 = 0.157), with the largest emotional Stroop effect for trials with happy faces. Neither the Stroop nor the emotional Stroop effects differed by group, Congruence Group, F(1,55) ¼ 0.236, p ¼0.629 (ηp2 = 0.004), and Emotion Congruence Group, F (2.245, 123.476) ¼ 0.976, p¼ 0.388 (ηp2 = 0.017) (see Fig. 1). Additional 2-way Face Emotion (Happy, Sad, Angry, Neutral) Group (MDD, HC) and Word Emotion (Happy, Sad, Angry, Neutral) Group (MDD, HC) repeated-measures analysis of variance were also conducted in order to further examine potential general processing bias for specific emotions. Groups did not show differential reaction times to specific facial emotions, Face Emotion Group, F(1.729,95.100) ¼ 0.400, p¼ 0.641 (ηp2 = 0.007), or emotional words, Word Emotion Group, F(2.657,146.153) ¼1.763, p¼ 0.163 (ηp2 = 0.031). A main effect of Word Emotion, F (2.657,146.153) ¼14.770, p o0.001 (ηp2 = 0.212) was detected, with slower reaction times to sad words (M¼1393.87, SE¼61.04) compared with happy words (M ¼1175.31, SE ¼61.64). 2.4. Discussion – Experiments 1–3 Across all three experiments, robust Stroop and emotional Stroop effects were detected, as was a bias to respond more quickly to happy than sad stimuli. However, these effects did not differ by depression severity or SCID-verified MDD. This suggests that neither self-reported dysphoria nor a diagnosis of MDD is associated with a clinically significant depression-related attentional biases. While participants in the first two experiments were not clinically diagnosed with MDD, results were consistent with the third sample of SCID-verified patients with MDD and HCs. Finally, power analyses revealed that the analyses of interest in both experiments had adequate power to detect a small effect size. While larger samples may increase power to detect even smaller effect sizes, the clinical importance and utility of detecting such small effect sizes should be considered.

3. Task-switching (experiment 4) Attention inflexibility has also been included as a characteristic deficit in MDD (Davis and Nolen-Hoeksema, 2000), and is purported to account for perseverative cognitions such as depressive rumination (Whitmer and Banich, 2007). Attention rigidity can be a result of at least two different processes: (1) deficits in switching from the current task to a different task (higher switch-costs), or (2) the inability to deactivate previously relevant thoughts or goals (lower set-inhibition). In MDD, these two processes may cause or contribute to rumination because of a difficulty engaging in new and non-depression related thoughts, or because of a difficulty suppressing depressive thoughts that continue to compete for attentional resources.

500


Fig. 1. (a) Left panel: Incongruent trials by valence by group. (b) Right panel: Congruent trials by valence by group. Error bars represent one standard error.

While it is clear that those with MDD exhibit ruminative cognitions about negative information in memory, it is less clear if this perseveration is influenced by attentional rigidity for negative stimuli. Attentional rigidity can be tested through a task-switching paradigm (Mayr and Keele, 2000), where participants are asked to attend to one specific stimulus out of multiple stimuli presented simultaneously. Each trial differs in the type of stimuli the participant is cued to attend to, and attentional flexibility is measured by comparing response times between different combinations of trials. For example, attentional flexibility may be measured by comparing reaction time to trials preceded by another with the same cue (repeat trials), and trials preceded by another with a different cue (switch trials). Longer response times on switch trials compared to repeat trials indicate a higher switch-cost, or decrease attentional flexibility. Attention inhibition can also be measured by comparing reaction time to trials that return to the same cue after a switch (e.g., for trial types A, B, and C, inhibitory trials would be measured by the response time of the second “A” in an ABA combination) and reaction time to trials that are preceded by two trials with different cues (i.e., RT to “A” of CBA; also known as control trials). Difficulty in set-inhibition is measured by reduced reaction time to inhibitory trials compared to control trials because a faster response represents incomplete inhibition and therefore increased accessibility of the previous task. If a deficit in attention flexibility is a significant factor in MDD, results should demonstrate differential set-switching and set-inhibition to mood-congruent stimuli. Specifically, the MDD group would be expected to show increased difficulty in set-switching to sad relative to happy stimuli, and relative to sad stimuli in HCs. Results in the MDD group should also show increased difficulty in inhibiting (smaller set-inhibition) sad stimuli compared to happy stimuli, as well as compared to sad stimuli in HCs. If no differences are detected, this would indicate that attentional inflexibility may not be a significant contributor to mood-congruent biases in MDD. 3.1. Methods 3.1.1. Participants Recruitment and clinical screening procedures were similar to that of Experiment three. This sample included 52 SCID diagnosed individuals (30 with MDD). Groups did not differ by sex, but the depressed group was slightly older than the HC group

(see Table 1). As such, analyses included age as a covariate. Antidepressant medication use was not an exclusionary criteria in this study, and all depressed participants demonstrated clinically significant symptoms at time of study. The experiment was approved by the University of Michigan's Institutional Review Board, and consent was obtained from all subjects prior to the study. 3.1.2. Materials and procedure Facial images depicting happy, sad, and neutral emotions were obtained from three different validated databases: the Pictures of Facial Affect (Ekman and Friesen, 1976), the FACES database from the Max Planck Institute for Human Development (Ebner et al., 2010), and a previously validated database used by the senior author (Deldin et al., 2000). Experimental stimuli were produced using Adobe Photoshop (Adobe Systems Inc., San Jose, CA), and delivered using E-Prime version 1.0 run on an LCD monitor positioned at eye level 30 in. away from the participant. Stimuli were selected to be 75% Caucasian and 25% Non-Caucasian, with 50% male/female within each racial category. Each image was manipulated on three dimensions: orientation, size, and contrast. Images differing on orientation were rotated at an 8° angle both clockwise and counter clockwise, while images differing on size were both magnified or shrunk by 20%. Images differing on contrast were adjusted using the Levels function in Photoshop. High contrast was set with input levels of 0, 0.2, and 140 for the black point, gray midpoint, and white point respectively. Low contrast was set with input levels of 0, 2.2, 255 with output levels set at 0, 150. The task was adapted from the task-switching paradigm used in Mayr and Keele (2000). In this task, participants were presented with a 2 2 matrix of images, with three of the four images differing on dimensions of orientation, size, and contrast. The fourth image was the original image, and placements of the images were randomized. Prior to stimulus presentation, participants were provided with a cue that directed them to pick the image differing on one of the three dimensions. Responses were made on the number keypad with the same spatial positions as the matrix on the screen (i.e., keys “1,” “2,” “4,” and “5”). Feedback was provided for all incorrect trials and trials with no response (see Fig. 2 for details). Participants completed 12 blocks (4 blocks per valence) of 80 trials, with optional breaks between each trial. Order of valence was counterbalanced between participants. This task was presented under the same conditions as experiment two. Trials followed with the same cues (e.g., two trials both cued with orientation) were categorized as


501

Fig. 2. Task-switching paradigm presentation order and times for Experiment 4.

repeat trials, whereas trials that were preceded by a different cue (e.g., size followed by orientation) were coded as switch trials. Additionally, set-inhibition was also examined by comparing inhibitory trials to control trials. Inhibitory trials were categorized by two trials of the same cue separated by a trial of a different cue (e.g., Size, Orientation, Size), whereas control trials are defined by three consecutive trials of different cues (e.g., Contrast, Orientation, Size). Only correct trials were included in the analysis. G*Power (Faul et al., 2007) was used to compute the minimum effect size detectable in this sample with adequate power (40.8) for the interactions of interest. Partial eta squared was used as the measure of effect size. A partial eta squared (ηp2 ) value of 0.01, 0.06, and 0.14 corresponds to a small, medium, and large effect size, respectively (Cohen, 1969; Richardson, 2011). Power analyses using an inter-correlation value of 0.81 revealed adequate power (40.8) to detect a small effect size (ηp2 = 0.014) for the 2-way Trial Type Group interaction (nonsphericity correction ϵ ¼1.00), and a small effect size (ηp2 = 0.008) for the 3-way Trial Type Emotion Group interaction (nonsphericity correction ϵ ¼0.829). 3.2. Results Two repeated-measures analysis of variance were conducted with reaction time as the dependent variable. The first analysis included trials assessing for set-switching, Trial Type (Switch, Repeat) Emotion (Happy, Sad, Neutral) Group (HC, MDD). The second included trials assessing for set-inhibition, Trial Type (Inhibitory, Control) Emotion (Happy, Sad, Neutral) Group (HC, MDD). The first analysis confirmed a significant set-switching effect similar to that found by Mayr and Keele (2000), with results revealing significantly longer reaction time to switch trials (M ¼1261.22, SE ¼39.99) compared to repeat trials (M¼1148.57, SE¼ 33.14), main effect of Trial Type, F(1,50)¼ 4.841, p o0.032 (ηp2 = 0.088). However, the set-switching effect did not differ by group alone, Trial Type Group, F(1,50)¼0.275, p ¼0.603 (ηp2 = 0.005), nor did it differ by emotion and group, Trial

Type Emotion Group, F(1.986,99.302) ¼ 1.407, p ¼0.250 (ηp2 = 0.027) (see Fig. 3). The second analysis found a marginal set-inhibition effect, with a longer mean reaction time to inhibitory trials (M ¼1270.09, SE¼39.72) compared to control trials (M ¼1249.36, SE¼ 40.29), main effect of Trial Type, F(1,50) ¼2.991, p ¼0.09 (ηp2 = 0.056). Similarly to the first analysis, no set-inhibition effect was detected by group, Trial Type Group, F(1,50)¼0.275, p ¼0.603 (ηp2 = 0.005), or by emotion and group, Trial Type Emotion Group, F (1.986,99.302) ¼0.469, p ¼0.626 (ηp2 = 0.009) (see Fig. 4). 3.3. Discussion While a set-switching effect and a marginal set-inhibition effect were reproduced, results did not suggest that those with MDD show increased difficulty in switching attention towards happy stimuli, or inhibiting previously seen sad stimuli, as compared to HCs. Effect sizes of group comparisons were also small, suggesting that even if the sample were underpowered to detect the true effect, the size may not necessarily be clinically interesting or impactful. As expected, reaction times to emotional stimuli were longer than that to neutral stimuli, though this effect also did not differ by group. These results are consistent with those using the emotional Stroop paradigm.

4. Counter task (experiment 5) Although the previous four experiments have demonstrated a lack of bias in the attentional phase of the information processing system, it has also been proposed that attentional biases in depression may only be unmasked under conditions of higher cognitive load (Wegner and Wenzlaff, 1996; Wenzlaff and Bates, 1998; Wenzlaff et al., 2001). However, these studies have mostly been conducted in undergraduate populations without clinical validation of diagnosis. In order to test this hypothesis, we employed an emotional variant of a Counter Task used in Garavan et al. (1999) conducted in a clinically validated sample with MDD.

502


Fig. 3. (a) Left panel: MDD compared to HC groups on Repeat Trials. (b) Right panel: MDD compared to HC groups on Switch Trials. Error bars represent one standard error.

In this task, participants were asked to keep a count, without using external aids, of the number of Happy and Sad faces presented one at a time. This task not only requires set-switching each time the stimuli changes from one emotion to another, it also requires additional cognitive resources dedicated to updating and maintaining two different sources of information simultaneously in working memory. If affective biases in MDD manifest during early attention under higher cognitive load, results should show differential reaction times to switch trials relative to repeat trials under the two emotional conditions. Specifically, mood-congruent biases in MDD would result in longer reaction times in switching to happy trials relative to sad trials, as well as compared to happy trials in HCs. 4.1. Methods 4.1.1. Participants Recruitment and clinical screening procedures were identical to that of experiment three. This sample included 25 (11 females) in the HC group and 23 (16 females) in the MDD group (SCID validated). Groups did not differ by age or gender (see Table 1). The experiment was approved by the University of Michigan's

Institutional Review Board, and consent was obtained from all subjects prior to the study. 4.1.2. Materials and procedure A total of 240 happy and sad facial stimuli were obtained from images used in experiment three, and were presented under the same conditions. Participants were presented with a series of happy and sad faces one at a time, and were instructed to keep a running count of happy and sad faces as they are presented. When participants had successfully updated their count, they pressed a button to advance to the next stimulus, and reaction time to button press was recorded. A total of 36 blocks were completed, each block consisting of between 8 and 18 trials. At the end of each block, participants reported the number of happy and sad faces they counted with feedback on the accuracy of their counts. Participants were given the option for breaks in between each block. Reaction times to trials that were followed by the same emotion (e.g., happy face followed by another happy face) were categorized as repeat trials, whereas trials preceded by a different emotion were categorized as switch trials. Order of stimulus presentation was randomized, with equal distribution of repeat and

Fig. 4. (a) Left panel: Inhibitory compared to Control trials for HC group. (b) Right panel: Inhibitory compared to Control trials for MDD group. Error bars represent one standard error.


switch trials. Only trials with a correct final count were included in the analysis. G*Power (Faul et al., 2007) was used to compute the minimum effect size detectable in this sample with adequate power ( 40.8) for the interactions of interest. Partial eta squared was used as the measure of effect size. A partial eta squared (ηp2 ) value of 0.01, 0.06, and 0.14 corresponds to a small, medium, and large effect size, respectively (Cohen, 1969; Richardson, 2011). Power analyses using an inter-correlation value of 0.87 revealed adequate power (4 0.8) to detect a small effect size (ηp2 = 0.011) for the 2-way Trial Type Group and Emotion Group interaction (nonsphericity correction ϵ ¼ 1.00), and a small effect size (ηp2 = 0.007 for the 3-way Trial Type Emotion Depression interaction (nonsphericity correction ϵ ¼1.00). 4.2. Results Reaction times were submitted to a 3-way Emotion (Happy, Sad) Trial Type (Repeat, Switch) Group (HC, MDD) repeatedmeasures analysis of variance. Results confirmed a set-switching effect, with longer reaction times to switch trials (M¼2181.71, SE¼ 83.00) compared to repeat trials (M¼ 1584.05, SE¼53.48), main effect of Trial Type, F(1,46) ¼187.873, p o0.001 (ηp2 = 0.803). However, this effect did not differ by group, Trial Type Group, F (1,46)¼ 1.719, p ¼0.196 (ηp2 = 0.036) Results also demonstrated a main effect of Emotion, F(1,460)¼ 12.248, p¼0.001 (ηp2 = 0.210), with longer reaction times to sad trials (M¼1936.74, SE¼71.02) compared to happy trials (M¼ 1829.02, SE¼65.01). A marginal emotional set-switching effect was also detected, Trial Type Emotion, F(1,46)¼3.759, p¼0.059 (ηp2 = 0.075), though this effect did not differ by group, Trial Type Emotion Group, F(1,46)¼0.035, p¼ 0.853 (ηp2 = 0.001) (see Fig. 5). Results also revealed increased reaction time in the MDD group (M¼2085.09, SE¼ 95.73) compared to the HC group (M¼ 1608.67, SE¼91.82), main effect of Group, F(1,46)¼9.30, p¼0.004. 4.3. Discussion Results from this task revealed a set-switching effect that is almost twice the size of the set-switching effect in experiment four, confirming the increased cognitive load in this task. However,

Fig. 5. Switch and Repeat trials by valence between the two groups for Experiment 5. Error bars represent one standard error.

503

no group differences were detected for the set-switching effect by valence, despite all analyses being adequately powered to detect a small effect size. This further suggests that even under conditions of increased cognitive load, those with MDD do not show affective biases in early attention specific to depression. These results are consistent with those from the previous four experiments. Results from this task did reveal a generally longer reaction time in the MDD group that was not detected in the previous experiments; this is likely a function of the increased task-difficulty.

5. General discussion In all five experiments using three different tasks, participants with diagnosed MDD or dysphoria did not show differential effects of attention to emotional stimuli compared to HCs. Even after increased cognitive load, a bias in attentional flexibility was not detected despite adequate power to detect a small effect size. These results are remarkably consistent across all experiments in this study, and add precision to our understanding of cognitive deficits associated with depression. Together, results indicate that cognitive biases in depression do not robustly impact earlier components of attention, including selective attention, attentional flexibility (including orienting towards or disengaging from stimuli in perception), or updating of working memory. In the context of other studies examining depression-related cognitive biases, it is likely that later components of information processing are more impacted by depression, such as enhanced retrieval of negative information (Williams and Broadbent, 1986; Taylor and John, 2004), reduced ability for sustained processing of positive stimuli (Shestyuk et al., 2005), and a bias towards negative interpretations of ambiguous information (Mogg et al., 2006). This is consistent with a previous eye-tracking study, where depressed and healthy control participants were presented with positive, neutral, threatening, and dysphoric images simultaneously. Visual gaze was tracked across each 30 s trial in order to examine the time course of selective attention. Results demonstrated no difference in initial fixation on emotional images between groups, but the depressed participants continued to show increased fixation on dysphoric images over 30 s (Kellough et al., 2008). Furthermore, experiments of attentional biases that employ stimulus presentation times of one second or greater are more likely to detect a difference in depressed participants (Gotlib and Joormann, 2010), further indicating that depression is not characterized by biases in early attentional processes. One strength of this study is the use of both verbal and facial stimuli. Evidence indicates that storage of affective information is more directly accessible via emotional images than through verbal stimuli (Glaser and Glaser, 1989), suggesting that affective biases should be enhanced when using emotional faces as stimuli. The lack of differential reaction times to emotional stimuli under these conditions further suggest that biases in early attention are not characteristic of the cognitive deficits in MDD. Another strength of this study is the use of valenced task-relevant stimuli, which may have more ecological validity. This is in contrast to the dot-probe task, where the emotional stimulus precedes the task-relevant stimulus (i.e., the appearance of the dot). Together, this suggests that while depressed individuals may mind-wander towards mood-congruent stimuli, this effect is not significant enough to impact attentional processing during goal-oriented tasks. While the majority of participants (experiments 3–5) were clinically validated for MDD via the standard clinical interview (SCID), analyses did not account for various variables that may be relevant, such as engagement in on-going interventions (e.g., medication or psychotherapy), or specific symptom clusters (e.g., cognitive versus somatic symptoms). Additionally, although the

504


sample sizes across all experiments were adequately powered to detect a small effect size, it is possible that the effects of interest may be smaller than the experiments were powered to detect. In light of these limitations, future studies may recruit for larger sample sizes in order to achieve adequate power to detect even smaller effect sizes, though the clinical importance and utility should be carefully considered. Alternatively, given the heterogeneity of MDD, use of a larger sample size may be better served in increasing resolution to compare various dimensions in MDD to examine if specific symptom profiles or other clinical variables would be more strongly associated with attentional biases. The clinical implications of these results suggest that cognitive interventions in depression may be more effective when targeting components occurring later in the information processing stream. For example, someone who exhibits “mental filtering” regarding a work performance evaluation may respond more robustly to an intervention that reduces rehearsal and elaboration of criticism (i.e. rumination) and/or increases rehearsal and elaboration of praise (i.e. savoring). In comparison, an intervention that trained attention to be more selective of positive or benign information in the environment may yield a less significant response. Implications of this study are also consistent with studies that have examined the effectiveness of cognitive bias modification (CBM) for depression, which contain components that include both attention training and modification of interpretative bias. In attention training, individuals practice attending to more benign or positive stimuli in the environment (i.e. selective attention), whereas modification of interpretative bias involves the practice of positive interpretations of ambiguous information (Papageorgiou and Wells, 2000; Koster et al., 2009). In a meta-analysis of 45 studies (2591 participants) of CBM for anxiety and depression, Hallion and Ruscio (2011) found that CBM had a remarkably smaller effect for attentional bias than interpretative bias, and that effects were only reliably detected following an acute stressor. Furthermore, when examining anxiety and depression separately, results showed that the intervention was specific to anxiety and not depression. Another recent meta-analysis examining 49 randomized controlled trials of CBM did demonstrate a small effect size for depression (Cristea et al., 2015); however, the study also found that the effect size lost significance after controlling for publication bias. The authors suggest that previous findings in CBM may have been artificially inflated due to factors unrelated to the interventions, such as publication bias and demand characteristics. These data, together with the results of this study, suggest that cognitive interventions for depression may be more effective when targeting cognitive components occurring later in information processing.

6. Conclusion In summary, these series of five experiments provide evidence that impaired early attentional processing is not characteristic of MDD. Rather, affective biases may likely have a larger impact on information that has already become the focus of attention. Results from this study contributes to our understanding of the temporal specificity of depressive affective biases in the information processing stream, and could inform development of interventions targeting specific neurocognitive processes.

Contributors

1. Philip Cheng contributed to the writing of the manuscript, the design and collection of data from experiments 3, 4, and 5, and the analysis of all experiments.

2. Stephanie Preston contributed with the provision of data from experiments 1 and 2, and editing of the manuscript. 3. John Jonides oversaw the design and collection of data from experiments 3, 4, and 5. 4. Alicia Hofelich Mohr provided data from experiments 1 and 2. 5. Kirti Thummala contributed to the design and collection of data from experiment 3. 6. Melynda Casement contributed to the design and collection of data from experiment 3, and editing of the manuscript. 7. Courtney Hsing contributed to the design and collection of data from experiment 2. 8. Patricia Deldin contributed to the design and collection of data from experiments 3, 4, and 5, and editing of the manuscript.

Acknowledgments This work is in part support by in part by NIMH grant MH60655 to JJ, funding from the Rackham Graduate School (Grant no. C114875) to AJH, and from the University of Michigan to SDP and PJD. The authors would like to thank Courtney Behnke and Catherine Cherny for data collection, Brent Stanfield and Katherine Foster for support in analyses, and all research assistants involved in the experiments for their dedication and hard work.

References Alloy, L.B., Abramson, L.Y., Hogan, M.E., Whitehouse, W.G., Rose, D.T., Robinson, M. S., Kim, R.S., Lapkin, J.B., 2000. The Temple–Wisconsin Cognitive Vulnerability to Depression Project: lifetime history of Axis I psychopathology in individuals at high and low cognitive risk for depression. J. Abnorm. Psychol. 109, 403–418. Beck, A.T., Rush, A.J., Shaw, B.F., Emery, G., 1987. Cognitive Therapy of Depression. Guilford Press, New York. Beck, A.T., Ward, C.H., Mendelson, M., Mock, J., Erbaugh, J., 1961. An inventory for measuring depression. Arch. Gen. Psychiatry 4, 561–571. Bradley, B., Mogg, K., Williams, R., 1995. Implicit and explicit memory for emotioncongruent information in clinical depression and anxiety. Behav. Res. Ther. 33, 755–770. Bradley, M.M., Lang, P.J., 1999. Affective Norms for English Words (ANEW): Instruction Manual and Affective Ratings. University of Florida, Gainesville, FL. Brailean, A.M., Koster, E.H.W., Hoorelbeke, K., De Raedt, R., 2014. Attentional modulation by reward and punishment cues in relation to depressive symptoms. J. Behav. Ther. Exp. Psychiatry 45, 351–359. Clark, D.A., Beck, A.T., 1999. Scientific Foundations of Cognitive Theory and Therapy of Depression. John Wiley and Sons, New York. Cohen, J., 1969. Statistical Power Analysis for the Behavioral Sciences. Academic Press, New York. Cooper, J.A., Gorlick, M.A., Denny, T., Worthy, D.A., Beevers, C.G., Maddox, W.T., 2014. Training attention improves decision making in individuals with elevated self-reported depressive symptoms. Cogn., Affect. Behav. Neurosci. 14, 729–741. Cristea, I.A., Kok, R.N., Cuijpers, P., 2015. Efficacy of cognitive bias modification interventions in anxiety and depression: meta-analysis. Br. J. Psychiatry 206, 7–16. Dalgleish, T., Werner-Seidler, A., 2014. Disruptions in autobiographical memory processing in depression and the emergence of memory therapeutics. Trends Cogn. Sci. 18, 596–604. Davis, R.N., Nolen-Hoeksema, S., 2000. Cognitive inflexibility among ruminators and nonruminators. Cogn. Ther. Res. 24, 699–711. Dawkins, K., Furnham, A., 1989. The colour naming of emotional words. Br. J. Psychology 80, 383–389. Deldin, P.J., Keller, J., Gergen, J.A., Miller, G.A., 2000. Right-posterior face processing anomaly in depression. J. Abnorm. Psychol. 109, 116–121. De Ruiter, C., Brosschot, J.F., 1994. The emotional Stroop interference effect in anxiety: attentional bias or cognitive avoidance? Behav. Res. Ther. 32, 315–319. Ebner, N., Riediger, M., Lindenberger, U., 2010. FACES—a database of facial expressions in young, middle-aged, and older women and men: development and validation. Behav. Res. Methods 42, 351–362. Ekman, P., Friesen, W., 1976. Pictures of Facial Affect [Slides]. Consulting Psychologists Press, Palo Alto, CA. Faul, F., Erdfelder, E., Lang, A., Buchner, A., 2007. G*Power 3: a flexible statistical power analysis program for the social, behavioral, and biomedical sciences. Behav. Res. Methods 39, 175–191. First, M., Spitzer, R., Gibbon, M., Willams, J., 2002. Structured Clinical Interview for DSM-IV-TR Axis I Disorders, Research Version, Patient Edition. Biometrics Research. New York State Psychiatric Institute, New York, NY.


Garavan, H., Ross, T.J., Stein, E.A., 1999. Right hemispheric dominance of inhibitory control: An event-related functional MRI study. P. Natl. Acad. Sci. USA, 96, 8301–8306. Glaser, W.R., Glaser, M.O., 1989. Context effects in Stroop-like word and picture processing. J. Exp. Psychol.: Gen. 118, 13–42. Gotlib, I.H., Joormann, J., 2010. Cognition and depression: current status and future directions. Annu. Rev. Clin. Psychol. 6, 285–312. Gotlib, I.H., Kasch, K.L., Traill, S., Joormann, J., Arnow, B.A., Johnson, S.L., 2004a. Coherence and specificity of information-processing biases in depression and social phobia. J. Abnorm. Psychol. 113, 386–398. Gotlib, I.H., Krasnoperova, E., Yue, D.N., Joormann, J., 2004b. Attentional biases for negative interpersonal stimuli in clinical depression. J. Abnorm. Psychol. 113, 127–135. Hallion, L.S., Ruscio, A.M., 2011. A meta-analysis of the effect of cognitive bias modification on anxiety and depression. Psychol. Bull. 137, 940–958. Hofelich, A.J., Preston, S.D., 2012. The meaning in empathy: Distinguishing conceptual encoding from facial mimicry, trait empathy, and attention to emotion. Cogn. Emot. 26, 119–128. Holmes, D.S., 1974. Investigations of repression: differential recall of material experimentally or naturally associated with ego threat. Psychol. Bull. 81, 632–653. Izard, C.E., 1994. Innate and universal facial expressions: evidence from developmental and cross-cultural research. Psychol. Bull. 115, 288–299. Kellough, J.L., Beevers, C.G., Ellis, A.J., Wells, T.T., 2008. Time course of selective attention in clinically depressed young adults: an eye tracking study. Behav. Res. Ther. 46, 1238–1243. Koster, E.H., Fox, E., MacLeod, C., 2009. Introduction to the special section on cognitive bias modification in emotional disorders. J. Abnorm. Psychol. 118, 1–4. Leyman, L., De Raedt, R., Schacht, R., Koster, E.H., 2007. Attentional biases for angry faces in unipolar depression. Psychol. Med. 37, 393–402. Mathews, A., Ridgeway, V., Williamson, D.A., 1996. Evidence for attention to threatening stimuli in depression. Behav. Res. Ther. 34, 695–705. Mayr, U., Keele, S.W., 2000. Changing internal constraints on action: the role of backward inhibition. J. Exp. Psychol.: Gen. 129, 4–26. McCabe, S.B., Gotlib, I.H., 1995. Selective attention and clinical depression: performance on a deployment-of-attention task. J. Abnorm. Psychol. 104, 241–245. Mogg, K., Bradbury, K.E., Bradley, B.P., 2006. Interpretation of ambiguous information in clinical depression. Behav. Res. Ther. 44, 1411–1419. Mogg, K., Bradley, B.P., 2005. Attentional bias in generalized anxiety disorder versus

505

depressive disorder. Cogn. Ther. Res. 29, 29–45. Mogg, K., Bradley, B.P., Williams, R., 1995. Attentional bias in anxiety and depression: the role of awareness. Br. J. Clin. Psychol. 34, 17–36. Nolen-Hoeksema, S., 1991. Responses to depression and their effects on the duration of depressive episodes. J. Abnorm. Psychol. 100, 569. Papageorgiou, C., Wells, A., 2000. Treatment of recurrent major depression with attention training. Cogn. Behav. Pract. 7, 407–413. Posner, M.I., Petersen, S.E., 1989. The Attention System of the Human Brain. DTIC Document. Preston, S.D., Stansfield, R.B., 2008. I know how you feel: task-irrelevant facial expressions are spontaneously processed at a semantic level. Cogn., Affect., Behav. Neurosci. 8, 54–64. Richardson, J.T.E., 2011. Eta squared and partial eta squared as measures of effect size in educational research. Educ. Res. Rev. 6, 135–147. Shaver, P., Schwartz, J., Kirson, D., O'Connor, C., 1987. Emotion knowledge: further exploration of a prototype approach. J. Personal. Soc. Psychol. 52, 1061–1086. Shestyuk, A.Y., Deldin, P.J., Brand, J.E., Deveney, C.M., 2005. Reduced sustained brain activity during processing of positive emotional stimuli in major depression. Biol. Psychiatry 57, 1089–1096. Sonnby–Borgström, M., 2002. Automatic mimicry reactions as related to differences in emotional empathy. Scand. J. Psychol. 43, 433–443. Taylor, J.L., John, C.H., 2004. Attentional and memory bias in persecutory delusions and depression. Psychopathology 37, 233–241. Wegner, D.M., Wenzlaff, R.M., 1996. Mental control. In: Higgins, E.T., Kruglanski, A. W. (Eds.), Social Psychology: Handbook of Basic Principles. Guilford Press, New York, pp. 466–492. Wenzlaff, R.M., Bates, D.E., 1998. Unmasking a cognitive vulnerability to depression: how lapses in mental control reveal depressive thinking. J. Personal. Soc. Psychol. 75, 1559–1571. Wenzlaff, R.M., Rude, S.S., Taylor, C.J., Stultz, C.H., Sweatt, R.A., 2001. Beneath the veil of thought suppression: attentional bias and depression risk. Cogn. Emot. 15, 435–452. Whitmer, A.J., Banich, M.T., 2007. Inhibition versus switching deficits in different forms of rumination. Psychol. Sci. 18, 546–553. Williams, J.M., Broadbent, K., 1986. Autobiographical memory in suicide attempters. J. Abnorm. Psychol. 95, 144–149. Williams, J.M.G., Watts, F.N., MacLeod, C., Mathews, A., 1997. Cognitive Psychology and Emotional Disorders, 2nd ed. John Wiley & Sons, New York.