Processing bias in anxious subjects andrepressors,measured by emotional Stroop interferenceandattentional allocation

Abstract

We hypothesized that repressors would show cognitive avoidance ofthreateninginformation in an attention deployment task, but an attentional bias for the sameinformation inan emotional interference task, while high anxious subjects would show athreat-related bias onboth tasks. A modified Stroop task and a visual probe task (VPT) wereused, with physical threatwords, social threat words, social positive words and general positivewords. The relationship ofthe response to the two tasks was also investigated. The resultsshowed that high state anxietywas related to greater Stroop interference of physical threat wordsas well as social words, boththreat and positive. No group effects were found for the Stroop, inspite of su.cient power. Incontrast, in the VPT high trait anxious subjects shifted attention onlytowards social threatwords, especially when these words were presented outside their attentionalfocus. No differenceinvolving the repressor group was present. There was a small positiveinter-task relation forsocial threat- related bias. It is suggested that the emotional biasesmeasured by the Stroop andthe VPT reflect automatic decisions about cognitive resourceallocation at subsequent phases ininformation processing, at which increasingly more specificaspects of the emotional informationare deciphered and used.

Introduction

There is a growing body of evidence that anxiety is associated with systematic cognitivebiases that favour the processing of threatening information (Broadbent and Broadbent, 1988; Williams et al., 1988; Mathews et al., 1989; Mineka and Sutton, 1992). In general, these biases are explained by assuming an allocation of attentional resources orprocessing resources for threat information. However, the nature of these anxiety-related biases isstill poorly understood. Moreover, it is questionable whether these biases are simply absent in lowanxious individuals or whether these individuals show the opposite pattern, i.e. actively avoidingthe processing of threat information. Several studies have produced an inconsistent picture of thepattern of biases in low anxious individuals (e.g. MacLeod et al., 1986; Dawkins and Furnham, 1989; Fox, 1993 Fox, 1994; Myers and McKenna, 1996) and theclarification of this picture may have consequences for the understanding of cognitive processingin anxiety (de Ruiter and Brosschot, 1994). Two problems seem to be at the heart of theconfusion and will be addressed in this study. One is that both low anxious groups and high anxiousgroups may be far from homogeneous with respect to their reactions to threat information. A secondcause for confusion is that the different tasks used to study anxiety-related cognitive bias do notactually measure the same process or measure different aspects of the same process.

First, results from a range of studies show that low anxious individuals are a ratherheterogeneous group with respect to their behaviour under stressful conditions. A large subgroupshows physiological and behavioural reactions that are not compatible with their low trait anxietyscores on a paper and pencil test. Such reactions also include performance on tasks that are used tomeasure cognitive bias for threat information. The subgroup that displays these contradictoryresponses is defined by low scores on trait anxiety as well as high scores on a measure ofdefensiveness (the Marlowe–Crowne social desirability scale: MC, Crowne and Marlowe, 1960). They are called repressors (REP) and are distinguished from so-called truly low anxious (TLA), subjects who also score low on defensiveness. High anxious subjects are similarly dividedinto high defensive/high anxious (HD/HA) subjects, who score high on both scales and highanxious (HA) subjects, who score high on anxiety and low on defensiveness. This division intosubgroups on the basis of defensiveness scores is based on a series of studies inspired by the workof Weinberger et al., 1979. Several of these studies have demonstrated that while subjectswith a repressor (REP) coping style seem to be low anxious, they are more like high anxioussubjects in their physiological and nonverbal reactions to stressors. The magnitude of their reactionsmight even exceed those of high anxious subjects (Weinberger et al., 1979; Asendorpf and Scherer, 1983; Jamner et al., 1988; Esterling et al., 1990; Bonanno et al., 1991). In other words, while objective measures among REP subjects indicate thepresence of anxiety, subjective measures do not.

In line with these findings, Dawkins and Furnham, 1989 showed that REP subjectsdemonstrated more threat-induced decrement in reaction time for threat words on a modified Strooptask, even more than commonly found among HA subjects. In fact, the interference by threat wordswas twice as high for REP subjects than for HA subjects, while there was no interference at all inTLA subjects. Fox, 1994 repeated this experiment, but she did not find the same interactionbetween group and word type. An explanation for this non-replication of the Dawkins and Furnhamresults could be that Fox administered words of different valence randomly and in a trial-by-trialfashion, while Dawkins and Furnham used blocked presentations of similar word types with thetraditional format of the Stroop task. In a blocked trial, repeating words of the same category maylead to stronger priming (Broadbent and Gathercole, 1990). Thus, only when blocked trialsare used, may the interference effects be robust enough to detect the influences of the REP copingstyle.

If the Dawkins and Furnham results are replicated, they will add support to the view that REPsubjects are in fact high anxious subjects who claim not to be anxious on self-report measures. Anemotional interference effect as shown in their study may then be seen as proof of a cognitive bias for threat words in REP subjects, consistent with the usual interpretation of these resultsin anxious subjects (MacLeod and Mathews, 1991). On the other hand, other similarcognitive experimental studies suggest an alternative interpretation. Myers and McKenna, 1996, using socially threatening words and a more extreme selection of repressors, reported theexact opposite of Dawkins and Furnham. In their study, all groups except the REP subjectsshowed higher interference for socially threatening words. Myers and McKenna interpreted thesefindings suggesting that REP subjects avoid the processing of social-emotional information. In twoother studies REP subjects show the opposite of the HA reaction pattern: again, REP subjects seemto avoid and even actively inhibit threat information, when compared with other groups of subjects (Fox, 1993, Fox, 1994). In one of these experiments, Fox, 1993 used a taskfor attentional deployment, the visual probe paradigm developed by MacLeod et al., 1986.In this task, two words are presented on a computer screen, while the subject has to name aloud thetop word and ignore the bottom word. Following the display of the words, a visual probe (a blackdot) may appear on the location of a word and the subject is asked to push a button as soon as thishappens. Attentional allocation is believed to be measured by the reaction time (RT) to this visualprobe. The rationale is that the faster the subject reacts to a given dot location, the more the subjectsattention is drawn by that location. On the other hand, the longer it takes to push the button for agiven dot location, the more the subjects attention is drawn by the opposite location. When boththese conditions are fulfilled for threat words, attentional bias for threat words is thought to bepresent. Using this paradigm, MacLeod et al., 1986 and Mogg et al., 1992 found thatanxious patients showed an attentional bias for threatening social and physical words, while controlsubjects showed the reverse pattern. Fox, 1993 replicated this finding for high trait anxioussubjects (i.e. HA subjects), but only for social threat words. She also showed that low trait anxioussubjects, that were also REP subjects, shifted their attention away from social threat words,while truly low trait anxious TLA individuals did not show a consistent attentional pattern.

In another study, Fox, 1994 showed a specific threat-related negative priming effectfor REP subjects, which was not present in the HA subjects or TLA subjects. In a negative primingparadigm, a reaction to a target stimulus is delayed when the same stimulus or a semantically relatedstimulus has to be ignored in the previous trial. This delay (negative priming) is interpreted as ameasure of cognitive inhibition. By using stimuli of neutral and threat categories, it wasdemonstrated that the reactions of REP subjects, when compared to the other subjects, were moredelayed if the repeated stimulus was a threat word than when it was a neutral word. HA subjects,on the other hand, had slower reactions to all stimuli after a threat stimulus that had to be ignored.Fox concluded that in this task REP subjects seem to inhibit threat information more effectively thaneither HA or TLA subjects. In contrast, HA subjects seemed to have difficulty inhibiting the threatinformation — at least until the next trial.

Thus, both Foxs visual probe study and her negative priming study support the notion thatREP subjects are characterized by cognitive avoidance of threat information. Although thisis clearly consistent with the usual interpretation of the repressor coping style, it is not immediatelyclear how these results can be reconciled with the higher threat-induced bias of REPsubjects in a Stroop task (Dawkins and Furnham, 1989). This might be explained if onetakes into account the stage of information processing at which these tasks are directed and thespatial properties of the tasks. One of the differences between the Stroop task and the other tasks —the visual probe task and negative priming task — is that the Stroop task measures reactions tosimultaneously presented stimuli, while the other two tasks measure reactions to the stimuluspresented immediately after the threat stimuli. Considering that in the visual probe task and thenegative priming task the critical measures occur 300 to 500 ms after the presentation of the threatstimulus, these tasks are probably measuring different mechanisms than the Stroop task does.Another important difference between the tasks is that in the Stroop task, the critical threateninginformation is integrated in the attended stimuli, while in the VPT half of the threatening andattended stimuli are spatially separate (but all pairs of threatening and attended stimuli are separatedin time). Both these differences might explain the contrasting effects found for repressors in thetasks. It may be possible that compared to HA subjects, REP subjects have as much (or even more)difficulty inhibiting threat information during reacting to the — integrated — target, reflectedin higher threat-related interference in the Stroop task (Dawkins and Furnham, 1989).However, in contrast to HA subjects, they may succeed in cognitive avoidance in a laterprocessing stage, especially when the target is spatially separated from threat. This would bereflected in an attentional bias away from threat information in the visual probe task and a betterinhibition of threat information in a negative priming task (Fox, 1993, Fox, 1994).

In summary, it seems that the presence of a selective bias for threat information in lowanxious persons is dependent on their defensiveness. Part of the low anxious subjects arecharacterized by an avoidant or repressive style of reacting to stressors. An attentional bias for threatinformation (resembling the bias found in high trait anxious subjects (Dawkins and Furnham, 1989)) as well as avoidant processing of threat information has been found for theserepressors (Fox, 1994). This avoidant style may be detected depending on which task isused to tap it.

The present study serves two main goals. One is to try to replicate the differential findings ofprocessing bias and cognitive avoidance for REP, HA and TLA subjects. This is done with the aidof the two paradigms that are most frequently used in cognitive bias studies in anxiety, the visualprobe task (VPT; Fox, 1993) and the modified Stroop task (Dawkins and Furnham, 1989; Fox, 1994). Our other aim was to investigate the degree to which a different cognitivemechanism is tapped by these two tasks. To our knowledge, such a comparison within subjects ofthe two tasks has not been previously undertaken. For this purpose, we have studied the convergenceof the biases measured by both tasks for different emotional words. Such an analysis would showwhether or not and to what extent, the two tests measure a similar mechanism.

We had several expectations. Firstly, in line with Dawkins and Furnham, 1989, weexpected that interference by threat words would be higher in REP subjects than in HA subjects andhigher in the latter than in TLA subjects. Secondly, we expected that REP subjects would shift theirattention away from social threat words in the VPT, in comparison to physical threat words (Fox, 1993) and, additionally, to positive words, while HA subjects were expected to shift theirattention toward social threat words. Thirdly, we thought TLA subjects would show no selectivebias in either task. Fourthly and exploratively, we correlated response latencies for the emotionalStroop trials with the VPT indexes that are used to express emotional bias in the VPT.

As an additional check, we also used positively valenced social words in both tasks, inaddition to general positive words, in order to investigate whether the expected biases for socialwords pertain to the semantic category of general social information, instead of to the specificthreatening content (see also Mathews and Klug, 1993). Furthermore, unlike Dawkins and Furnham, 1989 and Fox, 1993, we took care to include the HD/HA group.Most studies using Weinberger et al., 1979 repressor operationalization were performedwithout this group, mainly because it is difficult to find enough HD/HA subjects. It is obvious,however, that without this category no final conclusions can be drawn with regard to theindependent contributions of defensiveness and trait anxiety. Finally, we included state anxiety in the analyses, since attentional bias for threat is most often found when trait as well asstate anxiety is high (Broadbent and Broadbent, 1988; MacLeod and Mathews, 1988).