Introduction

Across the past three decades, cognitive models of anxiety vulnerability and dysfunction have steadily grown in influence. A shared prediction of these accounts is that an attentional bias to negative information plays a causal role in the tendency to experience unduly intense anxiety responses (c.f. Mathews and MacLeod 2005). There is now substantial evidence that an attentional preference for negative information is a reliable characteristic of elevated anxiety vulnerability, whether indicated by heightened trait anxiety, or by the presence of anxiety pathology (c.f. Bar-Haim et al. 2007). Indeed, this finding is so robust that in their recent meta-analysis, Bar-Haim et al. concluded that a further 11, 339 studies which consistently reported null results would be required to reduce this effect to non-significance.

These findings confirm that an attentional preference for negative information is associated with elevated anxiety vulnerability, but they do not permit the conclusion that this processing bias makes a causal contribution to such vulnerability. More direct evidence for this contention has come from studies that have made use of training variants of attentional assessment tasks, which are designed to systematically induce an attentional bias away from, or sometimes towards, negative information, to test the prediction that the modification of such bias will significantly alter anxiety vulnerability (e.g. Amir et al. 2008; Eldar et al. 2008). These cognitive bias modification tasks targeting attention (CBM-A) involve the introduction of a training contingency, whereby the ease of task performance will be enhanced if participants come to alter their pattern of attentional response to negative information.

The most commonly used CBM-A procedure is based on the dot probe task (MacLeod et al. 1986). In the assessment version of this task, probes are displayed in the location of negative and neutral members of stimulus pairs with equal probability, and relative speeding to discriminate probes in the location of the negative stimuli reveals attentional bias towards negative information. However, in the training version of this task, a contingency between the probe position and negative stimulus position is introduced, such that probes either are always displayed in the opposite location to where the negative stimuli appeared (avoid negative training), or are always displayed in the same location as where the negative stimuli appeared (attend negative training). No explicit instruction to alter attentional selectivity is given, but it is intended that, after repeated practice on this training task, participants will come to develop an attentional bias consistent with the training contingency through incidental learning.

MacLeod et al. (2002) exposed participants to this CBM-A procedure, with half receiving the training contingency designed to induce attentional avoidance of negative information, and half receiving the training contingency designed to induce attentional preference for negative information. Participants were not informed of this training contingency, and were not directed to exploit it. To determine whether it nevertheless altered attentional selectivity, an assessment version of the probe task, which did not contain a training contingency, was delivered immediately after completion of the attentional training. In this assessment task, the old stimulus pairs used in the attentional training task, and new stimulus pairs that had not previously been encountered, were presented. This enabled the investigators to determine whether any induced group difference in attentional bias generalized to new emotional stimuli. The investigators found that the differing CBM-A contingencies successfully induced the intended group difference in attentional response to negative information, which generalized to the new stimulus material. Of most importance, when participants were exposed to a subsequent anagram stressor, those who had just completed the avoid negative CBM-A showed an attenuated anxiety response to the stressor, compared to those who had just completed the attend negative CBM-A. Thus, these findings provide compelling evidence for the hypothesis that attentional bias to negative information can causally contribute to anxiety vulnerability. A number of studies have subsequently confirmed the reliability of this finding, using the same incidental learning approach within the laboratory setting (e.g. Amir et al. 2008; Hazen et al. 2009; Heeren et al. 2012). It has commonly been found that the induced attentional bias generalizes to new emotional stimuli (e.g. Heeren et al. 2011; MacLeod et al. 2007), and some studies have demonstrated that it generalizes also to attentional assessment tasks that differ from the approach used to train attention (e.g. Dandeneau and Baldwin 2004).

The focus of much recent work has been to investigate whether more extended exposure to this CBM-A procedure can reduce anxiety dysfunction in real-world settings. There is a rapidly growing body of evidence which highlights the potential therapeutic benefits of extended delivery of CBM-A designed to reduce attention to negative information. Again, this training contingency is not explicitly communicated to recipients, who are not instructed to alter their attentional selectivity, but this CBM-A has been shown to attenuate the symptoms of generalised anxiety disorder (Amir et al. 2009a; Amir and Taylor 2012), social anxiety disorder (Schmidt et al. 2009; Amir et al. 2009b), subclinical obsessive–compulsive symptoms (Najmi and Amir 2010), and separation anxiety disorder (Eldar et al. 2012). A recent meta-analysis by Hakamata et al. (2010) found that the effect size of reduction in anxiety symptoms in response to avoid-negative CBM-A is comparable to that produced by cognitive behavioural therapy and SSRI medication. Thus, these findings demonstrate the exciting potential of CBM-A as a tool that may be exploited for therapeutic gain.

There is now persuasive evidence that CBM-A procedures are capable of modifying attentional bias to negative information. Furthermore, the consistent finding has been that, when CBM-A successfully attenuates attention bias to negative information, so it also serves to attenuate anxiety vulnerability (c.f. Beard et al. 2012). It is worth noting, however, that not all CBM-A studies have successfully attenuated attentional bias to negative information, and perhaps not surprisingly, when no such attentional change has been elicited then no attenuation of anxiety vulnerability has been observed (Julian et al. 2012; Carlbring et al. 2012; Rapee et al. 2013). Clearly, this pattern of findings is fully consistent with the hypothesis that attentional bias to negative information can make a causal contribution to anxiety vulnerability. Nevertheless, they emphasize the need for investigators to focus on delineating the factors that may impair or enhance the efficacy of CBM-A (Beard 2011; MacLeod and Mathews 2012). In a recent methodological critique of this field, MacLeod et al. (2009) called, in particular, for future researchers to evaluate the role that explicit instruction may potentially play in the effective delivery of such bias modification. Noting prior researchers’ exclusive reliance on incidental learning, MacLeod et al. raise the possibility that explicitly instructing participants to practice the intended pattern of processing selectivity may potentially enhance the efficacy of CBM-A, while also cautioning that there are equally compelling grounds to anticipate that such explicit instruction could compromise the efficacy of CBM-A. For example, while there is evidence that explicit instruction can enhance cognitive training on some tasks (e.g. Foorman et al. 1998; Smeeton et al., 2005), it also has been found to impair such training effects on other tasks (e.g. Berry and Broadbent 1984; Farrow and Abernethy 2002; Green and Flowers 1991). Despite the call for researchers to investigate this issue, no CBM-A study has yet explicitly instructed participants to actively practice the pattern of attentional selectivity which the training contingency is designed to encourage. Hence, it remains unknown how the provision of such instructions would affect the capacity of CBM-A procedures to alter attentional selectivity and anxiety vulnerability.

The studies reported in this article were designed to address the two key questions pertaining to this issue. First, if participants are informed of the training contingency and instructed to actively practice the pattern of attentional selectivity this contingency is designed to encourage, then will the CBM-A procedure still effectively modify attentional response to negative information? Second, if participants are informed of the training contingency and directed to practice the target pattern of attentional selectivity, then will the CBM-A procedure still effectively alter anxiety reactivity to a subsequent stressful experience?

In both of two experiments we exposed participants who reported mid-range anxiety vulnerability to a probe-based CBM-A task, designed to differentially modify attentional responses to negative information. This task was closely based on the original CBM-A procedure employed by MacLeod et al. (2002), and our first study was essentially a replication of this earlier study, in which participants were provided with no information about the training contingency, and were not instructed to actively practice the target pattern of attentional responding to the negative stimuli. In the second study, however, participants were informed of the training contingency, and were explicitly instructed actively practice the pattern of selective attentional responding to negative stimuli that the training contingency was designed to foster. In both studies we assessed the attentional impact of CBM-A condition, by examining probe discrimination latencies on attentional assessment trials from which the training contingency was eliminated. In both studies we also assessed the impact of CBM-A condition on anxiety vulnerability, by subsequently exposing participants to an anagram stress task and examining the magnitude of the resulting anxiety response to this stressor. Additionally, in both studies we sought to assess the generalization of the CBM-A induced change in attentional selectivity not only to new stimuli, but also to a quite different type of attentional bias assessment task employing a flanker procedure (Sanders and Lamers 2002). Specifically, we examined whether the probe-based CBM-A procedure modified the relative degree to which negative and neutral flanker words interfered with the execution of a central task.

Study 1

Method

Participants

Sixty-four introductory psychology students from the University of Western Australia accepted an invitation to participate in this study. We sought participants who were mid-range with respect to dispositional anxiety, in keeping with our intention of closely replicating the original design of MacLeod et al. (2002), where this approach to participant selection was adopted to avoid the prospect of ceiling or floor effects compromising the experimental manipulation of either attentional bias or anxiety reactivity to the stressor. Participant selection was guided by the outcome of a mass screening procedure of approximately 700 introductory psychology students on the trait scale of the State-Trait Anxiety Inventory (STAI; Spielberger et al. 1983). Only students who scored in the middle third of the STAI-T scores obtained by this population (i.e. between 34 and 45) were invited to participate in the present study.

Participants were randomly allocated to one of two training conditions. Half of the participants (9 males) were assigned to a training contingency designed to induce attentional avoidance of negative information (avoid negative training). The mean STAI-T score of participants in this condition was 40.16 (SD = 3.06), and their mean age was 18.53 years (SD = 1.52). The remaining participants (7 males) were exposed to the training contingency designed to induce attentional vigilance for negative information (attend negative training). The mean STAI-T score of participants in this condition was 40.22 (SD = 2.70), and their mean age was 18.22 years (SD = 1.13). This gave rise to a between-group factor of Training Condition (Avoid Negative Training vs. Attend Negative Training). Participants assigned to the two training conditions did not differ in terms of trait anxiety scores, t (62) = .008, p = .931, age, t (62) = .87, p = .355, or gender ratio, χ2 (1, n = 64) = .33, p = .564.

Materials

Experimental Stimuli

Emotional Word Pairs

The same set of 96 word pairs used by MacLeod et al. (2002) was employed in the present study. One member of each word pair had a negative emotional valence, while the other member had a neutral emotional valence. This set of 96 word pairs was divided into the same two word pair subsets used by MacLeod et al., each containing 48 word pairs. For each participant, only one word pair subset was employed in the attentional bias modification trials, while both subsets were used in the attentional assessment trials. Hence, word pairs in assessment trials could be either Old (previously exposed in training) or New (not previously exposed in training).

Anagram Task Letter Strings

We employed the same 40 letter strings used by MacLeod et al. (2002) in their anagram stress task. 20 of these letter strings were soluble anagrams, in that the letters could be rearranged to form a legitimate English word, though these anagrams had been rated as very difficult to solve. The remaining 20 letter strings were ‘insoluble’ anagrams, as the letters could not be rearranged to form any legitimate English word.

Questionnaire Instruments

State-Trait Anxiety Inventory

The trait scale of the Spielberger State-Trait Anxiety Inventory (STAI; Spielberger et al. 1983) was employed to assess dispositional anxiety. The STAI-T is a widely used measure of dispositional anxiety, and has good reliability and validity (Barne set al 2002; Oei et al. 1990).

Visual Analogue Mood Scale

To measure levels of anxious mood state before and after completion of the anagram stressor task, an analogue mood scale assessing anxiety was programmed for computer delivery. This scale consisted of a 15 cm line, divided into 30 equal partitions, with the terminals labelled relaxed and anxious. Using the mouse, participants used the scale to record their anxiety level by moving the cursor to a position on the line that corresponded to their current level of anxiety, and pressing the left mouse button to register their response. This resulted in a score between 1 and 30, with a higher score indicating a greater level of anxiety.

Contingency Awareness Questionnaire

To assess the degree to which participants were aware of the training contingency they had been exposed to during the attentional bias modification task, we developed a Contingency Awareness Questionnaire. This first asked the participant to describe what, if anything, they believed predicted the location of the probe. Responses were coded as ‘correct’ if the answer indicated the participant had knowledge of the training contingency. Following this open question, participants were required to select, from a set of seven candidate options, whether a particular word characteristic predicted the location of the probe (e.g. word length, or part of speech), and if so, in which location the probe appeared more often. Word valence was the only one of the 7 options that actually influenced probe location. The participant response was coded as ‘correct’ only if the following two criteria were both met: (i) the participant accurately identified valence as influencing probe location, and correctly indicated the contingency; and (ii) the participant endorsed no more than 3 of the 7 candidate options.Footnote 1 The order of these 7 options was fixed, ensuring that individual differences in participant responses could not be attributed to variation in presentation order.

Experimental Hardware

A Hewlett Packard PC with a 22 inch monitor, and a standard two button mouse, was used to present the attentional bias modification and assessment tasks, as well as the anagram stressor task. A National View Finder video camera (Panasonic Ltd., Seoul, South Korea) also was employed to increase the subjective stressfulness of the anagram task.

Experimental Tasks

Attentional Bias Modification Task

Each trial commenced with the 500 ms presentation of a fixation cue (+++) in the centre of the screen. The screen was then cleared, and a word pair was displayed for 500 ms, with one member appearing 1.5 cm above, and the other member 1.5 cm below the prior location of fixation cue, subtending slightly less than a 3° visual angle of separation. The negative member of the word pair appeared in the upper and lower screen location with equal probability. Immediately after termination of the word pair display, a small visual probe appeared in the screen position previously occupied by either one of the words. This probe could be either a single red dot, or a pair of adjacent red dots, with equal probability. For participants in the avoid negative condition, the probe always appeared in the screen location where the neutral word had just been presented, while for participants in the attend negative condition, the probe always appeared in the screen location where the negative word had just been presented. To encourage participants to anticipate and direct attention towards this locus, the probe was exposed only very briefly, and participants were required indicate the discriminate the identity of the probe as quickly as possible. They made their response by pressing the left mouse button to indicate a single dot, and the right mouse button to indicate a double dot. The screen was cleared when the participant’s response was detected, probe discrimination latency was recorded, and the next trial began after a 500 ms inter-trial interval.

In total, 576 trials were presented across the attentional bias modification task. For each participant, only one word pair subset was used in this training task. Across participants, however, each subset was used an equal number of times in these attentional training trials. During attentional training trials, each of the 48 word pairs was exposed 12 times in total. Order of word pairs was randomised, with the constraint that each word pair was exposed once before any word pair was presented for a second time, and was only exposed for a third time after all word pairs had been repeated once, and so on.

Attentional Bias Assessment Tasks

The assessment tasks determined whether the attentional bias modification procedure induced a group difference in attentional response to negative information on an assessment version of the probe task, and whether any such induced group difference generalized to a new attentional assessment procedure employing a flanker task. Each of these two attentional bias assessment tasks will be described in turn.

Conventional Probe Attentional Assessment Task

The temporal and physical characteristics of the trials in the conventional probe attentional assessment task were equivalent to those in the attentional training task, except that probes now appeared in the location of the previously presented neutral word, and the location of the previously presented negative word, with equal frequency, for all participants. It remained the case that participants were required to indicate the identity of the probe as quickly as possible, using the mouse. In total, 96 trials were presented across this attentional assessment task. For each participant, each word pair from both of the word pair subsets was exposed once, in a random order. Using the probe discrimination latency data obtained from the probe position attentional assessment trials, we calculated an index of attentional bias to negative information. This index reflected the speeding for probes in the locus of negative words relative to probes in the locus of neutral words, and was computed using the following equation:

$$\begin{aligned} {\text{Attentional Bias to Negative Information Ind}}{\text{ex}} &= {\text{RT for probes opposite negative word locus}} \\ &\quad- {\text{RT for probes in negative word locus}} \\ \end{aligned}$$

A higher score on this index reflects a greater attentional preference for negative, relative to neutral, information. To determine whether the induced attentional bias generalized to the new stimulus material, we computed this index of attentional bias to negative information separately for trials that presented word pairs previously used in the bias modification trials (i.e. “old” word pairs), and for trials that presented words not previously exposed in these modification trials (i.e. “new” word pairs). If the attentional impact of bias modification condition is affected by this word pair status factor, then consideration of the index computed for new words alone will reveal whether the training effect generalized, in the sense of being evident on word pairs unused in the attentional training procedure.

Flanker Assessment Task

Each trial commenced with the presentation of a target probe in the centre of the screen, and a single flanker word displayed both above and below this central target probe. The target probe was either a single red dot, or a pair of adjacent red dots, with equal frequency. On half of the trials the flanker word had a negative emotional valence, while on the remaining trials the flanker word had a neutral emotional valence. Using the mouse, participants were required to rapidly indicate the identity of the probe. The target probe and flanker word remained on screen until the participant’s response was detected. The screen was then cleared, and the next trial commenced 500 ms later. In total, 96 trials were presented across this attentional assessment task. For each participant, half of the negative words, and half of the neutral words, from both of the word pair subsets were presented.

Using the latency data obtained from the flanker attentional assessment trials, we calculated an index of attentional bias to negative information. This index reflected a slowing to probes when a negative flanker word was present, relative to probes when a neutral flanker word was present, and was computed using the following equation:

$$\begin{aligned} {\text{Attentional Bias to Negative Information Index }} &= {\text{RT for probes when negative flanker word present}} \\ &\quad- {\text{RT for probes when neutral flanker word present}} \\ \end{aligned}$$

A higher score on this index reflects a greater attentional preference for negative, relative to neutral, information. Once more, to determine whether the induced attentional bias generalized to new stimulus material, we computed this index separately for trials presenting “old” word pairs and “new” word pairs. Again, if this word pair status factor moderates the attentional impact of the bias modification procedure, it will be important to determine whether the training effect is significant for new word pairs, as will be the case if the training effect generalizes to stimuli not used in the attentional training procedure.

Anagram Stress Task

We employed the same anagram stress task used by MacLeod et al. (2002) to assess anxiety reactivity subsequent to the attentional bias modification procedure, in keeping with our desire to closely replicate this earlier study. The use of such anagram stress tasks to assess anxiety reactivity has been common in prior research investigating the relationship between attentional bias and emotional vulnerability (e.g. Haeffel et al. 2012; Grafton et al. 2012). The task is designed to expose participants to internally generated threat cues (concerning the negative implications of failure) and to external threat cues (e.g., a video camera supposedly recording this failure, for future viewing by others), permitting the possibility that differential attention to such negative information may potentially moderate anxiety reactivity to the stressor.

Participants were informed that the anagram task was part of ongoing research into the relationship between performance on cognitive tasks and academic achievement, and was designed to measure individual differences in the ability to solve anagrams. They were told that they would be video-taped during task performance, and that their video might be used for teaching purposes in forthcoming psychology tutorials, if they scored in the lower or upper 10th percentile of task performance. However, unknown to participants, the video camera did not actually record during task performance. Participants were given 3 min to solve as many as possible, recording their answers on a supplied response sheet, and were told to click the left mouse button to move onto the next anagram. Participants were instructed that, if they could not solve an anagram, they should click the right mouse button to receive the next anagram. Once the 3 min time period had expired, a message was displayed on the screen stating that the participant’s performance was in the lower 10th percentile, and that their video would be used for later demonstration purposes. Anxious mood state was assessed on the visual analogue anxiety scale immediately before and immediately after the anagram task.

Procedure

Each participant was tested individually. The test session began with the participant being seated approximately 60 cm in front of the computer screen, and provided with instructions for the attentional task. A brief practice task was delivered before the attentional bias modification procedure. The participant was told to discriminate the probe identity as quickly as possible, without compromising accuracy, and was not provided with any information whatsoever concerning the existence of a training contingency, nor with any direction concerning what attentional response they should make to the negative stimuli. Both the attentional bias modification task and the attentional bias assessment task were delivered sequentially, without the participant being informed of the transition between these two tasks. Half the participants who completed each of the two training conditions subsequently completed the conventional probe attentional assessment task, while the other half completed the flanker attentional assessment task. Following completion of the attentional assessment task, the participant was provided with instructions concerning completion of the anagram task. Once they had completed this anagram task, the participant filled in the Contingency Awareness Questionnaire, before being debriefed about the purpose of the study and thanked for participation.

Results and Discussion

Impact of Attentional Bias Modification Procedure on Attention

Effect of CBM-A on Conventional Probe Attentional Assessment Task

To determine whether the training manipulation induced a group difference in attentional bias, as assessed by the conventional attentional probe task, latencies to correctly discriminate probes in the assessment version of this task were used to calculate attentional bias index scores. These indices, which are shown in leftmost two columns of Table 1, were computed as described in the Method section, after excluding those that fell further than 2.58 SD from the participant’s mean probe discrimination latency within each experimental condition. The avoid negative training and attend negative training conditions did not differ in terms of the percentage of latencies excluded (0.72 % and 0.85 % respectively, F < 1). A higher index score reflects a greater attentional bias towards negative, relative to neutral, stimuli. We inspected the distribution of these index scores across participants using a Shapiro–Wilk test, and because this revealed evidence of non-normality, W (32) = .93, p = .029 we applied the recommended square-root transformation (Osborne 2002) which served to normalize the data, W (32) = .98, p = .713. Throughout this manuscript, we follow the convention of reporting test statistics obtained when using transformed data, and summary statistics obtained using the untransformed data (Manikandan 2010).

Table 1 Mean and standard deviation of attentional bias index scores, and analogue mood scores, obtained from the attentional assessment tasks, and anagram stressor task, in Study 1
Full size table

These attentional bias index scores were subjected to a mixed-design 2 × 2 ANOVA that considered the between-group factor Training Condition (Avoid Negative Training vs. Attend Negative Training), and the within-group factor Word Pair Status (Old Word Pair vs. New Word Pair). This analysis revealed a significant main effect of Training Condition, F (1, 30) = 4.88, p = .035, partial η2 = .14. This reflected the fact that, on this post-attentional training assessment, participants who had completed the attend negative CBM-A condition displayed higher scores on the index of attentional bias to negative information (M = 66.27, SD = 102.93), than did participants who had completed the avoid negative condition (M = −52.60, SD = 172.20). This confirms that the two CBM-A conditions induced a group difference in attentional response to negative stimuli on this assessment version of the probe task employed in the training procedure. No other significant effects emerged from the analysis. Given the absence of any trend towards the main effect of Training Condition being modified by Word Pair Status in a two-way interaction, F (1, 30) = .02, p = .883, partial η2 = .001, this induced group difference in attentional bias was no less evident on the new word pairs than on the old word pairs, consistent with the generalization of the attentional training effect to emotional stimuli not previously encountered in the attentional training trials.

Effect of CBM-A on Flanker Assessment Task

In order to determine whether an impact of the attentional bias modification procedure could be detected on a new attentional bias assessment task that differed in format from the training task, flanker bias index scores were calculated using the latencies to correctly discriminate probes in the flanker task. These index scores, which are shown in the middle two columns of Table 1, were computed as described in the Method section, again excluding latencies that fell further than 2.58 SD from the participant’s mean latency within each experimental condition. The avoid negative training and attend negative training conditions did not differ significantly in terms of the percentage of latencies excluded (2.41 vs. 2.80 % respectively, F (1, 30) = 1.26, p = .271). A higher flanker bias index score reflects a greater attentional bias towards negative, relative to neutral, stimuli, on the flanker task. Inspection of these index score distribution revealed evidence of non-normality, W (32) = .91, p = .014, and so we again applied a square-root transformation, which normalized the data W (32) = .94, p = .072.

These flanker bias index scores were subjected to a mixed-design 2 × 2 ANOVA that considered the between-group factor Training Condition (Avoid Negative Training vs. Attend Negative Training), and the within-group factor Word Status (Old Word vs. New Word). The only significant effect to emerge from this analysis was a main effect of Word Status, F (1, 30) = 8.01, p = .008, partial η2 = .21, reflecting the fact that, across participants, higher index scores were observed on trials presenting words not previously exposed in training, (M = 11.17, SD = 3.55), compared to trials presenting words previously exposed in training (M = 8.57, SD = 2.92). There was no evidence, however, that that training condition exerted any influence on flanker task performance. Neither the main effect of Training Condition, F (1, 30) = .13, p = .721, partial η2 = .004, nor the interaction of Training Condition × Word Status, F (1, 30) = 3.08, p = .089, partial η2 = .09, reached statistical significance. Hence, there was no evidence that the influence of CBM-A condition generalized to this very different attentional assessment task.

Impact of Attentional Bias Modification Procedure on Anxiety Reactivity

The anxiety scores obtained from the analogue mood scale given before and after the anagram stressor are shown in rightmost two columns of Table 1. A Shapiro–Wilk test indicated that the distribution of these scores did not violate normality, W (64) = .99, p = .603. These anxiety scores were subjected to a mixed-design ANOVA that considered the between-group factor Training Condition (Avoid Negative Training vs. Attend Negative Training), and the within-group factor Anxiety Assessment Point (Pre-Anagram Assessment vs. Post-Anagram Assessment). This revealed a significant main effect of Assessment Point, F (1, 30) = 82.88, p < .001, partial η2 = .57, reflecting the fact that anxiety level was significantly elevated post-anagram stressor (M = 22.11, SD = 10.31) compared to pre-anagram stressor (M = 11.67, SD = 7.84). This confirms that the anagram stressor experience did serve to elevate anxious mood state, as intended. Of greater importance, this main effect was subsumed within a two-way interaction of Training Condition × Anxiety Assessment Point, F (1, 30) = 4.64, p = .035, partial η2 = .07, which was the only other significant effect to emerge from the analysis.

As can be seen in Fig. 1, the magnitude of this anxious response was attenuated in participants who had just been exposed to avoid negative CBM-A condition, compared to those who had just been exposed to the attend negative CBM-A condition. Thus, the induction of the group difference in attentional response to negative information resulted in a corresponding group difference in anxiety vulnerability, revealed by attenuated anxiety reactivity to the anagram stressor in participants given the avoid negative attentional bias modification procedure.

Fig. 1
figure 1

Impact of anagram stress experience on anxious mood state after completion of the attentional bias modification training in Study 1

Full size image

Awareness of Attentional Training Contingency

The data obtained from the Contingency Awareness Questionnaire (CAQ) was examined to determine whether participants’ awareness of the training contingency was limited, consistent with the assumption that this conventional CBM-A procedure exerts an implicit impact on attentional selectivity. We first analysed participant responses to the open-ended question in Section A of the CAQ. Only 21.9 % of participants answered this question in a manner that mentioned the possibility of probe position being related to word valence. Thus, the substantial majority of participants evidenced no explicit awareness of the training contingency. We went on to examine participant responses to Section B of the CAQ, which invited participants to endorse any of 7 possible contingencies, only one of which implicated word valence. The mean score obtained in this section of the CAQ was 1.75 (SD = 1.05), and the scores ranged from 1 to 5. Only 37.5 % of participants endorsed the correct option without endorsing a total of more than 3 candidate options. Hence, despite this very liberal criterion, it remained the case that the majority of participants showed no awareness of the training contingency, in that they could not discriminate the correct training contingency from incorrect alternatives.

These results replicate and extend the findings of previous CBM-A studies that, like the present experiment, have not explicitly informed participants of the training contingency, and have given them no instruction to actively practice the attentional response to negative information which the training was designed to foster. Using this conventional instruction approach, the present bias modification procedure induced a group difference in selective attentional response to negative information, and the attentional impact of this CBM-A generalized to new stimuli, though not to the quite different flanker task measure of attentional bias. Most importantly, the CBM-A manipulation also induced a significant group difference in anxiety reactivity. Specifically, participants who received the avoid negative CBM-A condition displayed attenuated anxiety responses to the subsequent anagram stressor. The substantial majority of participants showed no evidence that they were aware of the training contingency, which is consistent with the assumption that it exerted a largely implicit influence on attentional selectivity.

This first study was not designed to determine whether exposure to a CBM-A manipulation induces change in attentional bias, and in anxiety reactivity, when participants are aware of the training contingency, and its power to adequately address such questions is limited given the relatively small numbers of participants who evidenced such awareness. Nevertheless, to enable provisional post hoc investigation of this issue, we repeated the ANOVAs previously carried out on the attentional bias scores, and the anxiety reactivity scores, when considering only those participants who displayed awareness of the training contingency, as evidenced by their combined responses to Section A and B of the CAQ. This revealed that, in these contingency-aware participants, the CBM-A manipulation still served to induce a significant difference in attentional bias, F (1, 12) = 9.31, p = .010, partial η2 = .44, but this attentional effect was not accompanied by any significant difference in observed anxiety reactivity to the anagram stressor, F (1, 22) = 1.86, p = .187, partial η2 = .08. Hence, this post hoc analysis suggests that when participants are aware of the CBM-A training contingency then, although this training contingency still exerts a significant impact on attentional selectivity, this attentional impact is not accompanied by any corresponding change in anxiety reactivity.

While not directly relevant to the key questions of whether or not participants who are aware of an CBM-A training contingency nevertheless display a significant attentional and emotional response to this CBM-A manipulation, we considered it of potential interest to also directly compare the impact of the CBM-A manipulation in those participants who did, and who did not, display awareness of the contingency in the present study. Hence, we repeated the ANOVAs previously carried out on the attentional bias scores, and on the anxiety reactivity scores, while including the between-group factor Contingency Awareness (Aware of Training Contingency vs. Not Aware of Training Contingency), again indexed by participants’ combined responses to the CAQ. This revealed that Contingency Awareness tended to increase the impact of the CBM-A manipulation on attentional bias, though this effect fell just short of statistical significance, F (1, 28) = 3.84, p = .06, partial η2 = .12. Contingency Awareness, however, did not increase the impact of the CBM-A manipulation on anxiety reactivity to the stressor, F (1, 28) = .24, p = .629, partial η2 = .004. These findings suggest that awareness of the training contingency may increase the attentional change elicited by CBM-A, but it does not increase the impact of CBM-A on anxiety reactivity (and, indeed, as noted in the previously reported analysis, for participants showing such awareness the CBM-A manipulation did not significantly influence anxiety reactivity in the present study).

In our second study we wanted to more directly address the two key questions under empirical scrutiny. Specifically, we sought to determine whether the same attentional training procedure employed in this first study would still prove capable of producing an impact on attentional bias, and on anxiety reactivity, when participants were explicitly informed about the training contingency, and instructed to actively practice the pattern of attentional selectivity this contingency was designed to encourage.

Study 2

Method

Participants

Sixty-four introductory psychology students from the University of Western Australia took part in this second study. They were selected in exactly the same manner as in Study 1, and again were assigned at random to the same two CBM-A training conditions. None of the participants had prior knowledge of the research topic, or had taken part in the first study. Participants exposed to the avoid negative training (8 males) had a mean STAI-T score of 39.28 (SD = 3.08), and a mean age of 19 years (SD = 2.95). Participants exposed to the attend negative training (7 males) had a mean STAI-T score of 38.69 (SD = 2.98), and a mean age of 18.44 years (SD = 1.13). Those assigned to the two training conditions did not differ in terms of trait anxiety scores, t (62) = .61, p = .436, age, t (62) = 1.01, p = .318, or gender ratio, χ2 (1, n = 64) = .09, p = .768.

Materials

All materials used in this second study were identical to those used in the first study.

Procedure

The procedure was identical to Study 1, with the sole exception that in this second study participants were informed in advance of the training contingency, and were instructed to adopt and practice the pattern of attentional selectivity their assigned training condition was designed to produce. Specifically, participants were told that a word pair would be presented on each trial, and that one member would have a negative emotional tone, while the other would have a neutral emotional tone. Those assigned to the avoid negative condition were informed that probes would generally appear in the locus of the neutral word in each pair, and were instructed that they should always quickly shift their attention away from the negative word and towards this neutral word, on each trial. Participants assigned to the attend negative condition instead were informed that the probes would generally appear in the locus of the negative word in each pair, and were instructed to quickly shift their attention towards negative member of each word pair and away from the neutral member, on every trial. Participants were informed that the purpose of the experiment was to investigate how the anticipation of probe location, by attending to the required word in line with the attentional instruction, would impact on their ability to identify the probe quickly and accurately. Participants were told nothing about our interest in the emotional consequences of this attentional manipulation.

Results and Discussion

Impact of Attentional Bias Modification Procedure on Attention

Effect of CBM-A on Conventional Probe Attentional Assessment Task

The attentional bias index scores were calculated from latencies to correctly discriminate probes on this task, in exactly the same manner as was adopted in Study 1, and using the same exclusion criteria. Again the percentage of data excluded did not differ significantly in avoid negative training and attend negative training conditions (0.98 % vs. 0.37 % respectively, F (1, 30) = 3.85, p = .06). The resulting attentional bias index scores are shown in the leftmost two columns of Table 2. Again, a higher score reflects a greater attentional preference for negative, relative to neutral, stimuli. A Shapiro–Wilk test indicated that the distribution of these scores did not violate normality, W (32) = .97, p = .377.

Table 2 Mean and standard deviation of attentional bias index scores, and analogue mood scores, obtained from the attentional assessment tasks, and anagram stressor task, in Study 2
Full size table

As in Study 1, these attentional bias index scores were subjected to a mixed-design 2 × 2 ANOVA that considered the between-group factor Training Condition (Avoid Negative Training vs. Attend Negative Training), and the within-group factor Word Pair Status (Old Word Pair vs. New Word Pair). This analysis revealed a very robust main effect of Training Condition, F (1, 30) = 22.06, p < .001, partial η2 = .42, reflecting the fact that the participants who had received the attend negative CBM-A condition displayed very much higher scores on the index of attentional bias to negative information (M = 152.07, SD = 129.18), than did those participants who had received the avoid negative CBM-A condition (M = −70.59, SD = 138.79). No other significant effects emerged from the analysis of the probe discrimination latencies. Again, it of importance to note that the main effect of Training Condition did not interact with Word Pair Status F (1, 30) = 1.31, p = .260, partial η2 = .04, indicating that the training induced group difference in attentional response to negative information was no less evident on the new word pairs than on the old word pairs. Hence, these results confirm that it is possible to effectively modify attentional responses to negative information when participants are informed of the training contingency, and explicitly instructed to practice the pattern of attentional selectivity this contingency is designed to foster. Furthermore, the attentional change induced when such explicit instruction is given continues to generalize to new stimulus materials not employed in the training procedure.

Effect of CBM-A on Flanker Assessment Task

To determine whether the attentional impact of the CBM-A manipulation generalized to the flanker task, we calculated flanker bias scores, as in the previous study, again using the same exclusion criteria. Once more, there was no difference in the percentage of data excluded from the avoid negative and attend negative training conditions (3.06 vs. 2.97 % respectively, F < 1). Flanker bias scores are presented in middle two columns of Table 2. A higher score reflects a greater attentional bias towards negative, relative to neutral, stimuli. Inspection of the index score distribution revealed evidence of non-normality, W (32) = .88, p = .002, and so we again imposed a square-root transformation, which normalized the data, W (32) = .94, p = .083.

These attentional bias scores were subjected to the same mixed-design 2 × 2 ANOVA employed in Study 1, that considered the between-group factor Training Condition (Avoid Negative Training vs. Attend Negative Training), and the within-group factor Word Status (Old Word vs. New Word). As in the previous study, the only significant effect to emerge from this analysis was a main effect of Word Status, F (1, 30) = 26.09, p < .001, partial η2 = .47, reflecting the fact that, across participants, higher index scores were observed on trials presenting words not previously exposed in training (M = 13.21, SD = 2.93), compared to trials presenting words previously exposed in training (M = 8.99, SD = 3.44). There was again, however, no evidence that the training condition exerted any influence on flanker task performance. Neither the main effect of Training Condition, F (1, 30) = .84, p = .365, partial η2 = .03, nor the interaction of Training Condition × Word Status, F (1, 30) = 1.09, p = .305, partial η2 = .04, approached statistical significance. Hence, although participants given the present explicit information and practice instruction displayed a strong attentional impact of the CBM-A procedure when attention was measured using an assessment version of the training task itself, there was no evidence that the influence of CBM-A condition generalized to this new attentional assessment task.

Impact of Attentional Bias Modification Procedure on Anxiety Reactivity

The anxiety scores recorded on the analogue mood scale, which was given pre and post the anagram stressor, are shown in the rightmost columns of Table 2. A Shapiro–Wilk test indicated that the distribution of these scores did not violate normality, W (64) = .98, p = .422. These anxiety rating data were subjected to the same mixed-design ANOVA used to analyse such data in Study 1. This considered the between-group factor Training Condition (Avoid Negative Training vs. Attend Negative Training), and the within-group factor Anxiety Assessment Point (Pre-Anagram Assessment vs. Post-Anagram Assessment). There was a robust main effect of Anxiety Assessment Point, F (1, 30) = 86.64, p < .001, partial η2 = .58 reflecting the fact that anxiety ratings obtained post-anagram stressor (M = 20.45, SD = 9.44), were substantially higher than those obtained pre-anagram stressor (M = 11.47, SD = 5.98), again confirming that the anagram stressor experience served to elevate anxious mood state, as intended. However, in striking contrast to Study 1, there was now no evidence whatsoever that this main effect was modified by Training Condition, with the interaction between the two factors not approaching significance F (1, 30) = .44, p = .509, partial η2 = .007. Hence, in the present study which informed participants of the training contingency, and instructed active practice in the target pattern of attentional selectivity, exposure to the avoid negative and attend negative CBM-A condition did not result in differing levels of anxiety vulnerability, as revealed by the intensity of anxiety reactions to the anagram stressor.

Awareness of Attentional Training Contingency

To confirm that participants in the present study showed awareness of the training contingency, as intended, the data from the Contingency Awareness Questionnaire (CAQ) were examined. In response to the open-ended question in Section A of the CAQ, 70.3 % of participants correctly implicated the position of the negative word as predicting the position of the probe. The mean score obtained in Section B of the CAQ was 1.95 (SD = 1.03), with scores ranging from 1 to 5. In contrast to Study 1, 81.3 % of participants correctly endorsed word valence as predicting probe position, without endorsing more than 3 of the candidate options provided. Thus, the great majority of participants in this second study showed awareness of the training contingency, in keeping with our experimental intention.

Comparison of the Impact of Attentional Bias Modification on Attention and Anxiety Reactivity Between Study 1 and 2

The analyses carried out on the attentional bias scores, and the anxiety reactivity scores, in this second study serve to answer the two key questions we sought to address in the present research. Specifically, the findings clearly demonstrate that, when participants were informed of the training contingency and instructed to practice the target pattern of attentional selectivity, in this second study, the CBM-A manipulation still effectively modified attentional bias to negative information, but such modification of attentional selectivity was not accompanied by any corresponding change in anxiety reactivity to the subsequent stressor. While the data from Study 2 alone serve to answer these two key questions under scrutiny, we considered it of additional interest to directly compare the findings obtained using the different CBM-A instruction approaches employed in the two studies.

First consider the impact of the CBM-A procedure on attentional bias. Comparison of the Training Condition effect size obtained in each study indicates that the magnitude of the attentional bias effect size obtained in Study 2, which explicitly informed participants of the training contingency and directed them to actively practice the target pattern of attentional selectivity, was more than twice as great as the attentional bias effect size obtained in Study 1, which used the conventional CBM-A instruction approach (partial η2 = .42 vs. .16, respectively). Inclusion of a between-group Study factor (Study 1 vs. Study 2) in an ANOVA carried out on all data from both studies showed that this difference in effect size was not statistically reliable, as the two-way interaction of Study × Training Condition fell short of statistical significance, F (1, 62) = 2.26, p = .138, partial η2 = .04. Nevertheless, it is clear that the CBM-A manipulation exerted a robust impact on attentional selectivity when participants were made aware of the training contingency in this manner. Given that the effect size pertaining to the attentional impact of the CBM-A manipulation doubled in size when participants were explicitly informed of the training contingency, and told to actively practice the target pattern of attentional selectivity, there is no suggestion from our data that the attentional impact of CBM-A was compromised by such conditions of awareness.

When we instead compared the emotional impact of the CBM-A procedure across the two studies, we observed that the effect size reflecting the impact of the CBM-A manipulation on the anxiety reactivity index (i.e. the degree to which anxiety was elevated post vs. pre stressor) was exactly 10 times smaller in Study 2 than in Study 1 (partial η2 = .007 vs. .07, respectively). Again inclusion of a between-group Study factor (Study 1 vs. Study 2) in an ANOVA carried out on all data from both studies did not show this between study difference in effect size to be statistically reliable, as the three-way interaction of Study × Training Condition × Anxiety Assessment Point fell short of statistical significance, F (1, 124) = 1.49, p = .225, partial η2 = .01. Nevertheless, the fact remains that the CBM-A manipulation exerted no detectable impact on anxiety reactivity when participants were made aware of the training contingency and were instructed to actively practice the target pattern of attentional selectivity. Not only was this effect, which had been significant in Study 1, found to be non-significant in Study 2, but power analysis revealed that it would be expected to remain non-significant under this explicit instruction condition even if a further 600 participants had been added to this latter study.Footnote 2

Given that the interactions involving Study as a between-group factor did not reach statistical significance, we do not claim on the basis of our findings that the explicit provision of training contingency information and directed practice instruction to participants receiving CBM-A significantly modified the impact of this CBM-A on attentional bias or anxiety reactivity. Rather we conclude only that, while we replicated the previous finding that CBM-A delivered using conventional implicit instruction significantly modified anxiety reactivity, we were able to find no evidence of such an effect on anxiety reactivity when participants were informed of the training contingency and directed to actively practice the target pattern of attentional selectivity. The absence of such an anxiety reactivity effect under this instruction condition is underscored by the fact that power analysis revealed it would have remained non-significant even if the sample size in Study 2 had been increased tenfold. Furthermore, we conclude that the failure to obtain this anxiety reactivity effect when we employed such explicit instructions did not reflect the weakening of the attentional effect elicited by CBM-A under this instruction condition. The attentional impact of CBM-A remained every bit as great under this explicit instruction condition (and indeed this effect size was nominally doubled when such explicit instruction was used in Study 2). Thus, while CBM-A delivered using the explicit instruction condition exerted a robust impact on attentional bias, this attentional effect was not accompanied by any detectable impact on anxiety reactivity when explicit instruction was employed.

General Discussion

The growing evidence that cognitive bias modification procedures targeting attention can alter anxiety vulnerability (c.f. Hakamata et al. 2010) has led to increased interest in identifying how alternative approaches to CBM-A delivery might influence its effectiveness (c.f. MacLeod and Mathews 2012). The present research was motivated by the prior call to empirically establish whether informing participants of the training contingency and instructing them to actively practice the target pattern of attentional selectivity, rather than relying on incidental learning, eliminates or augments its efficacy (MacLeod et al. 2009). Our findings indicate that when participants receiving CBM-A are informed of the training contingency, and directed to practice the pattern of attentional selectivity this training contingency is designed to induce, then although the training exerts a robust impact on attentional selectivity, the impact of CBM-A on attentional bias is no longer accompanied by any corresponding change in anxiety reactivity to a subsequent stressor. Thus, although the use of such explicit instruction did not compromise the capacity of the CBM-A manipulation to modify attentional response to negative information, it eliminated the impact of the CBM-A manipulation on anxiety vulnerability.

Before considering the implications of this finding, it is appropriate to acknowledge some limitations of the reported research. One limitation is that, because attention and anxiety reactivity were assessed only after delivery of the CBM-A procedure, it was not possible to test whether change in attentional bias statistically accounted for the observed change in anxiety reactivity within Study 1, using meditational analysis as recommended by MacLeod et al. (2009). These investigators’ note the conclusion that an attentional bias to negative information causally contributes to anxiety vulnerability would be strengthened if it could be shown that the change in attentional bias produced by the training manipulation mediates the change in anxiety reactivity. Meta-analytic studies have confirmed such mediation in previous studies (e.g. Hakamata et al. 2010), and we believe it would be of value in future research to consistently adopt designs that enable the computation of the change scores needed to determine whether the emotional impact of CBM-A is mediated by its attentional impact. A second limitation is that, because we did not assess clinically anxious participants, we cannot draw the conclusion that explicitly communicating the training contingency, and instructing active practice in the target pattern of attentional selectivity, necessarily would eliminate the previously demonstrated therapeutic benefits of conventional CBM-A on clinical symptomatology (e.g. Amir et al. 2009a, b; Eldar et al. 2012; Schmidt et al. 2009). Future research should directly examine the impact that such instructions have on the capacity of CBM-A to ameliorate the symptoms of anxiety pathology. A third limitation of the present work is that we have examined how explicit provision of training contingency information and directed practice instructions influence only one particular cognitive bias modification task. It would be appropriate to also investigate how this instructional approach influences the cognitive and emotional effects of other types of attentional bias and interpretive bias modification procedures, before drawing general conclusion about this issue. Nevertheless, our findings indicate that when probe-based CBM-A is delivered to non-clinical participants, alerting them to the training contingency and instructing them to actively practice the target pattern of attentional selectivity, serves to eliminate the impact of this CBM-A manipulation on anxiety reactivity without attenuating its impact on attentional selectivity.

It is always appropriate to consider whether experimental demand may suffice to explain the patterns of emotional effects obtained in CBM-A studies. Although the possibility of demand effects can never be conclusively ruled out, we believe it fairly unlikely that the emotional impact of the CBM-A manipulation in Study 1, or the absence of such emotional impact in Study 2, plausibly reflect experimental demand. Attributing the findings obtained in Study 1 to demand effects would require the assumptions that participants first inferred our expectation that they would develop a particular attentional bias to negative information, then executed with considerable precision the pattern of probe discrimination latencies consistent with such bias. Next, on the anagram stress task, participants would need to have inferred not only that we expected this induced attentional bias to affect emotional responding, but also that we harboured the expectation that this emotional impact would be observed not on post-training mood state, but only on anxiety reactivity to the subsequent anagram stressor. And, of course, they would need to have been motivated to satisfy this perceived expectation, by simulating such an impact. Furthermore, if the emotional impact of the attentional training manipulation in Study 1 represented a demand effect, then alerting participants more explicitly to this training manipulation in Study 2 should have increased, rather than eliminated, this emotional impact. Perhaps it could be argued that demand effects only influenced Study 2, with participants wrongly inferring in this study alone that we expected the training condition designed to reduce attention to negative information to exacerbate rather than attenuate negative mood. However, once again, participants endeavouring to fulfil such an expectation would surely have reported differing mood scores immediately following exposure to the alternative CBM-A conditions, and there was no evidence that CBM-A condition influenced anxiety scores at all in Study 2. Hence, while we think it is fairly unlikely that the present findings result from expectancy effects, we suggest that future research of this nature should directly evaluate the potential contribution of expectancy effects by directly interrogating participants’ expectations concerning the anticipated impact of the CBM-A procedure on both attentional processing, and emotional reactivity. It also would be appropriate to include measures of anxiety reactivity that do not rely on self-report, such as observational measures of anxious behaviour and physiological measures of anxious responding during performance of the final stressor.

In contrast to Study 1, our second study alerted participants to the nature of the training contingency, while also instructing them to practice the execution of a specified pattern of attentional responding to the negative stimuli. Hence, it is uncertain whether the effect of CBM-A on anxiety reactivity was eliminated in Study 2 by making participants aware of the training contingency, or by having had them actively practice the target pattern of attentional selectivity. There is reason to believe that awareness of the training contingency alone may not suffice to explain this observed effect. While no previous research has directly instructed CBM-A recipients to actively practice the target pattern of attentional selectivity, one previous study by Krebs et al. (2010) has informed recipients of the training contingency, without instructing them to actively practice a specified pattern of attentional selectivity. Participants given this contingency information responded to the CBM-A manipulation by displaying a particularly robust change in attention to negative information, which was accompanied by a corresponding change in reported levels of worry during a subsequent worry induction task. Hence, the awareness of the training contingency alone did not serve to eliminate the emotional impact of the CBM-A manipulation.

Perhaps CBM-A training fails to influence anxiety vulnerability when it involves deliberately practicing the target pattern of attentional selectivity because the resulting CBM-A induced attentional bias is strategic in nature. In contrast, it is commonly believed that individual differences in anxiety vulnerability are characterized by variation in automatic attentional bias (c.f. Bar-Haim et al. 2007). For example, Teachman et al. (2012) conclude in their recent review that the attentional bias displayed by individuals with heightened anxiety vulnerability is uncontrollable, unconscious, and unintentional in nature, all hallmarks of an automatic process. It is not possible to determine from the present data whether the differential attentional bias induced by our CBM-A manipulation was less automatic when participants were directed to actively practice the intended patterns of attentional selectivity, than was the case when the CBM-A procedure did not instruct such active practice. Indeed, some recent evidence suggests that conventional CBM-A procedures may impact on strategic attentional responding to negative information (e.g. Eldar and Bar-Haim 2010; Koster et al. 2010; Browning et al. 2010), though it seems likely that such strategic change may become automatized with sufficiently extended practice. Future research could usefully examine whether CBM-A variants that do, or do not, instruct participants to actively practice adopting the target pattern of attentional selectivity differ in the degree to which they influence automatic patterns of attentional selectivity, and whether this may account for their differing impact on anxiety vulnerability.

Another possibility is that the attentional bias induced when participants are informed of the training contingency, and instructed to actively practice the target pattern of attentional selectivity, may be more narrow in its scope of operation than is the case when the conventional CBM-A approach is adopted. There are two ways in which the training effect may be more narrow. First, the attentional bias induced by former CBM-A approach could be narrow in the sense that, while enduring, it influences selective attention only under very constrained conditions that closely match the training task (e.g. when negative and non-negative information are spatially separated on a computer screen and exposed only for a brief duration). Second, a rather different way in which the induced bias could be narrow when the explicit approach is adopted is that it may be more transient, ceasing to operate beyond the temporal boundaries of the original CBM-A training procedure. In either case, the CBM-A procedure would become less likely to influence the pattern of attentional selectivity adopted during performance of the anagram stressor task, and so would be less likely to impact on emotional reactivity to this stressor.

The finding that the capacity of CBM-A to alter emotional vulnerability may be compromised by providing recipients with explicit information about the nature of the training manipulation, and having them engage in actively practicing the desired pattern of attentional selectivity, gives rise to an important practical dilemma. Clinical researchers have recently reported that some patients receiving multi-session CBM-A express frustration over the absence of an explicit rationale that enables them to understand the nature of the training procedure, and to more actively engage in the intended therapeutic process (Beard et al. 2010). Given the evidence that the efficacy of therapeutic interventions often can be augmented by improving patient acceptability (Kaltenthaler et al. 2008), such patient feedback invites the development of CBM-A approaches that explicitly communicate the attentional training rationale, and actively involves recipients in the process of altering attentional responding to negative information. However, the present findings cast doubt on the clinical merit of such an approach to CBM-A delivery. Future research should directly investigate whether the symptoms of clinical anxiety can be effectively attenuated by more extended exposure to CBM-A training that provides such explicit information and directed practice instruction. Although the present single session of CBM-A may have produced intentional, or narrow attentional change, with the attendant consequence that this attentional change does not influence anxiety vulnerability, this may not be the case following extensive practice. With more extended exposure to such training the resulting attentional change may become less intentional, and perhaps consequently less narrow, with the result that it then comes to exert the desired impact on anxiety vulnerability. Optimism that such an approach may prove effective is buoyed by the evidence that initially intentional cognitive change produced by explicit instruction can become unintentional through extended exposure to continued practice (Mullen et al. 2007).

Alternatively, it may be better not to give patients explicit information about the training contingency, or instructed practice in the desired pattern of attentional responding to negative information, in order to preserve the therapeutic impact of CBM-A, and to instead try to increase the acceptability of CBM-A to patients in alternative ways. For example, future researchers may attempt to make CBM-A tasks sufficiently enjoyable, perhaps by making them more game-like in design, so that motivation to complete them does not necessarily depend on the patients’ understanding their intended mechanisms of operation. Such an approach may be especially useful with younger participants, as parents could be provided with the CBM-A rationale, but would be able to encourage their children’s participation by emphasizing the entertainment value of the task, and socially reinforcing high scoring performance.

The present observation that providing explicit information about the training contingency, and engaging participants in active practice of the target attentional change, eliminates the desired impact of attentional change on anxiety vulnerability, raises the intriguing possibility that the emotional benefits of CBM-A may be optimized if active steps were taken to reduce patient insight into the training contingency, and to preclude such active practice in attentional change. Such steps could include, for example, masking the emotional stimuli in order that there is no possibility of consciously apprehending the training contingency, or introducing a secondary cognitive load to disrupt strategic processing during completion of the CBM-A task. Paradoxically, the adoption of this type of approach could, in principle, permit the possibility of being fully candid with patients about the rationale underpinning CBM-A, without compromising the potentially crucial automaticity of the resulting attentional bias modification.

Given the evidence that CBM-A can therapeutically attenuate dysfunctional anxiety, it is of critical clinical importance to now identify how best to deliver CBM-A to maximise its therapeutic benefits. In the present research, we have begun to address whether explicitly communicating the CBM-A training contingency, and having recipients actively practice the pattern of attentional selectivity this contingency is designed to foster, can influence its efficacy. This represents only one of many issues that investigators must address to ensure the most effective translation of the CBM-A approach into successful clinical procedures. It will be equally important to determine, for example, the optimal numbers of training sessions, the optimal scheduling of these sessions across time, and the optimal combination of CBM-A tasks with other evidence-based clinical techniques. For the moment, however, the present results suggest that the provision of explicit information and instructed practice in the desired pattern of attentional selectivity, while potentially enhancing patient acceptability with CBM-A, may eliminate its capacity to effectively alter anxiety vulnerability. Hence, when it comes to delivering CBM-A in clinical settings, it may well be the case that ignorance is bliss.