First, attentional bias scores were calculated from the dot probe data, and the lexical decision task was analyzed to ensure that subliminal cues were truly outside conscious awareness. Second, the impact of threat cue relevance on attentional biases was explored using a three-way mixed design ANOVA, with threat type and exposure duration as within subject factors and stress level as a between subjects factor. Third, the capacity of attentional control to moderate the impact of fearful temperament on attentional bias was tested using regression analyses. Finally, ways in which attention to social threat, social stress, fearful temperament, and attentional control interact to predict symptoms of depression and anxiety were tested using regressions.
Dot Probe Data Preparation
Before analysis, raw data were evaluated for inaccurate and outlying responses. For all participants, response latencies for incorrect responses about probe location were dropped (less than 3.5% of the data), and reaction times less than 200 ms or greater than 2000 ms (less than 0.5% of the data) were removed as error. Once attentional bias scores were calculated, nine participants with one or more bias scores falling 2.5 SDs above or below the group mean bias score for each condition were removed, following procedures used by Boyer and colleagues (2006)
. Inspection of scores for these 9 participants suggested that they were similar to the remaining sample for Fearful Temperament, Attentional Control, Depression, and Anxiety, but were lower in Social Stress (M
= 9.0 versus 14.9).
For each participant, attentional bias scores representing the average response time difference for targets occurring in neutral versus threat locations were calculated for subliminal physical threat, subliminal social threat, supraliminal physical threat, and supraliminal social threat conditions. Bias scores were based on the standard equation, ½[(UpLt-UpUt) + (LpUt-LpLt)] (MacLeod & Mathews, 1988
), with U = upper position, L = lower position, p = probe, and t = threat. Thus, UpLt indicates the mean response latency for trials in which the threat word is in the lower position and the probe replaces the neutral word in the upper position. A positive score indicates bias toward threat; a negative score indicates bias away from threat; a score of zero indicates the absence of bias.
Mean bias scores were 4.48 (SD = 69.38) for subliminal physical cues, 3.40 (SD = 69.88) for subliminal social cues, 2.03 (SD = 68.23) for supraliminal physical cues, and 3.60 (SD = 64.07) for supraliminal social cues. None of these scores was significantly different from zero, indicating no attentional bias in the full sample.
Lexical Decision Task
Accuracy scores on the lexical task were analyzed to ensure that participants were not able to discriminate between real or nonsense words presented subliminally. The percent of trials in which students correctly identified words or non-words was calculated for each participant. Mean response accuracy was 47% (SD = 8%), and no individual performed reliably better than chance, indicating that individuals could not distinguish between real and nonsense words presented subliminally. This makes it reasonable to consider responses to subliminal threat cues in the dot probe task a reflection of cognitive processes taking place without conscious awareness.
Threat Cue Relevance
A Threat Type (social, physical) × Stress (low, medium, high) × Exposure Duration (subliminal, supraliminal) mixed design ANOVA was conducted to look at threat related biases at differing levels of social stress. Attentional biases for social threat cues were expected to be more pronounced in individuals facing high levels of social stress than in individuals facing low levels of social stress, and these biases were expected to be specific to social threat cues. Because it was expected that personal relevance of threat cues would affect automatic stages of attentional orienting (e.g., Bar-Haim et al., 2007
; Boyer et al., 2006
), this pattern of attentional bias was expected to be most evident in subliminal threat cue presentations. Thus, the predicted effects would be demonstrated either in a two-way interaction of threat type and stress, or in a three-way interaction of threat type, stress, and exposure duration. Main effects for threat type, stress level, and exposure duration were all non-significant, indicating that the magnitude of threat-related attentional biases did not differ for social and physical threat cues, high and low social stress groups, or subliminal and supraliminal exposures in the full sample. The Threat Type × Exposure Duration and Threat type × Stress level interactions were also non-significant. However, the three-way interaction of threat type, stress level, and exposure duration was significant, F
(2, 109) = 3.39, p
The specific nature of the interaction was explored using Social Stress × Exposure Duration ANOVAs for physical and social threat words separately. The medium stress group was omitted, as the primary interest was in the difference between high and low stress groups. As can be seen in , the pattern of results suggests that the three-way interaction was driven by social threat words, with the simple interaction of social stress and exposure duration approaching significance for social threat words, F (1, 79) = 2.31, p = .12, but not physical threat words. Individuals high in social stress tended to be vigilant to subliminal social threat (M = 18.12, SD = 68.06), but not supraliminal threat (M = -4.84, SD = 68.98), while individuals low in social stress showed no attentional biases toward social threat at either the subliminal (M = -1.25, SD = .13) or supraliminal (M = 7.31, SD = 60.68) levels.
Attentional bias for Low and High Social Stress Groups.
Attentional Control as a Moderator of Relations between Fearful Temperament and Attentional Bias
Regression analyses were used to test the hypothesis that attentional control would moderate relations between fearful temperament and attention to social threat. To take advantage of the continuous nature of the temperament measures, moderation was tested by constructing a regression equation that included fearful temperament, attentional control, and a multiplicative term representing the interaction of the two traits (Aiken & West, 1991
). All predictors were centered to maximize interpretability and minimize potential problems with multicollinearity. Because social stress was related to attentional biases, all regressions also controlled for total social stress. Separate regressions were performed for sub- and supraliminal social threat cues. Fearful temperament was expected to predict stronger attentional biases. Attentional control was not expected to show any relationship to subliminal social bias, which reflects purely automatic attentional allocation, but was expected to interact with fearful temperament to predict supraliminal social bias. The subliminal social model explained 3% percent of the variance in bias to subliminal social threat cues, F
(4, 107) = 0.87, p
= .48. Neither temperament trait, nor the interaction of fearful temperament with attentional control was a significant predictor.
The supraliminal social model accounted for 10% of the variance in supraliminal social bias, F (4, 107) = 3.01, p < .05. Once again, neither fearful temperament nor attentional control was a significant predictor. However, as expected, attentional control did interact with fearful temperament to predict attention to supraliminal social threat cues, t = 2.90, p < .01. shows the simple slopes for the relation between fearful temperament and supraliminal social bias plotted at high (+ 1 SD) and low (-1 SD) levels of attentional control. The slope at high levels of attentional control was not significantly different from zero, indicating that there is no relationship between fearful temperament and allocation of attention to supraliminal social threat for individuals with a strong ability to regulate attention. However, for individuals low in attentional control, only those who were also low in fearful temperament attended to supraliminal social threat cues. Higher levels of fearful temperament were linked to greater avoidance of supraliminal social threat cues.
Figure 2 Interaction of Fearful Temperament and Attentional Control in predicting Supraliminal Social Bias. Simple slopes are presented for values 1 SD above and below the mean for Attentional Control. Values depicted are unstandardized regression coefficients; (more ...)
Relations between Attentional Bias and Distress
Despite a well-established relationship between threat cue vigilance and psychological distress, it is not clear whether attentional biases are harmful to all people across all circumstances. A more nuanced understanding of the interplay between social context and dispositional traits, along with consideration of differences between automatic and controlled attention, is essential. Regression analyses were used to test relations between social threat biases and symptoms of distress, and to explore ways in which these relationships might be moderated by social stress, fearful temperament, and attentional control. One set of two regressions was run predicting anxiety from either subliminal or supraliminal social threat bias scores plus all possible two-, three-, and four-way interactions of bias, social stress, fearful temperament, and attentional control (see ). Another set of two regressions was run predicting depression from these same variables. For simplicity, because this paper focuses on attentional bias, interactions not involving bias scores will not be discussed. As above, interactions with bias scores were explored using simple slope analyses for high (+1 SD) and low (-1 SD) levels of each variable.
Interactions of attentional bias, Fearful Temperament, Attentional Control, and Social Stress in predicting Anxiety and Depression
Main effects of predictors
In both the sub- and supraliminal models, bias scores alone were typically not direct predictors of distress. The one exception to this was the weak, but significant, association between bias for subliminal social threat and depression. As expected, fearful temperament predicted greater anxiety and depression in all analyses. Social stress was correlated positively with anxiety and depression, but this relationship was only significant in the supraliminal models. Attentional control was significantly and negatively associated with symptoms of depression.
Relations between social threat bias and anxiety
The subliminal social threat model accounted for 45% of the variance in self-reported anxiety. There were significant interactions between social stress and attentional control, and between social stress, fearful temperament, and attentional control, which will not be discussed. Most notable to the present hypotheses was the three-way interaction involving fearful temperament, attentional control, and subliminal social bias, t = 2.76, p < .01. As depicted in , individuals low in fearful temperament exhibited fewer symptoms of anxiety than those high in fearful temperament. For individuals low in fearful temperament, bias to subliminal social threat cues predicted anxiety only for those low in attentional control. In contrast, for individuals with low fearful temperament and high attentional control, subliminal attentional biases were unrelated to anxiety, suggesting that a high degree of attentional control is able to buffer the negative impact of subliminal vigilance for individuals with a low temperamental vulnerability to threat. For individuals high in fearful temperament and low in attentional control, patterns of attention to subliminal social cues were unrelated to anxiety, as these individuals showed high levels of anxiety regardless of avoidance or attention to subliminal threat cues. However, attentional patterns did matter for high fearful temperament individuals who were also high in attentional control. Although avoidance of subliminal social threat had no benefits for high fearful temperament/low attentional control individuals, for high fearful temperament/high attentional control individuals, avoidance was associated with low anxiety, with symptom levels comparable to those of low fearful temperament individuals. High attentional control, therefore, was not sufficient to protect individuals with fearful temperament from the negative impact of vigilance to subliminal social cues.
Figure 3 Interaction of Fearful Temperament, Attentional Control, and Subliminal Social Bias in predicting Anxiety. Simple slopes are presented for values 1 SD above and below the mean of Fearful Temperament and Attentional Control. Values depicted are unstandardized (more ...)
The supraliminal social model accounted for 43% of the variance in self-reported anxiety. Although bias alone was completely unrelated to anxiety symptoms (β = -.01), fearful temperament interacted with supraliminal social bias to predict anxiety, t = -1.93, p = .05 (see ). Level of fearful temperament appeared irrelevant for individuals attending to supraliminal social threat cues. However, avoidance of supraliminal social threat cues was associated with anxiety for individuals high in fearful temperament, but not for those low in fearful temperament.
Figure 4 Interaction of Fearful Temperament and Supraliminal Social Bias in predicting Anxiety. Simple slopes are presented for values 1 SD above and below the mean of Fearful Temperament. Values depicted are unstandardized regression coefficients; standard errors (more ...) Relations between social threat bias and depression
The subliminal social threat model accounted for 50% of the variance in depressive symptoms. Subliminal social bias interacted with social stress, t = 3.02, p < .01, and attentional control, t = 2.35, p < .05, to predict depression (see ). Analysis of simple slopes for the interaction of stress level with subliminal social bias indicated that attentional patterns were unrelated to depressive symptoms for those with low stress. However, for those facing high levels of social stress, vigilance to subliminal social threat predicted greater depression. Importantly, regressions demonstrated that when high stress was paired with avoidance of subliminal social threat cues, resulting symptom levels were comparable to those of individuals with low social stress. Analysis of the interaction of subliminal social bias and attentional control indicated that attentional allocation to subliminal threat was irrelevant for individuals low in attentional control, whose symptoms remained steady across attentional patterns. However, for individuals high in attentional control, subliminal threat cue avoidance was associated with lower levels of depressive symptoms.
Figure 5 Interaction of Subliminal Social Attentional Bias and Social Stress, and Subliminal Social Bias and Attentional Control in predicting Depression. Simple slopes are presented for values 1 SD above and below means for Social Stress. Values depicted are (more ...)
The supraliminal social threat model accounted for 46% of the variance in depressive symptoms. Although none of the interactions with supraliminal bias was significant, the three-way interaction of social stress, fearful temperament, and supraliminal bias approached significance, p = .08, suggesting a possible direction for future research.