Enduring consequences of experimentally induced biases in interpretation

Abstract

Previous research has demonstrated that it is possible to induce biases in the interpretation of ambiguous text passages by training. Participants consistently trained to interpret emotionally ambiguous passages in either a negative or positive direction show training-congruent effects when presented with new ambiguous material. These training effects are demonstrated by participants’ subsequent recognition ratings for disambiguating sentences, which represent both possible meanings of the novel ambiguous test passages.

In this series of experiments, we investigated the durability of these training effects over time spans of up to 1 day and found them to be robust. The findings encourage us to believe that induced biases may serve as a useful analogue to those observed clinically.

Introduction

Over the last two decades considerable research energy has been devoted to investigating the biases in cognitive processing that occur in those with clinical anxiety disorders and sub-clinical anxiety (Mathews & MacLeod, 1994; Williams, Mathews, & MacLeod, 1996; Williams, Watts, MacLeod, & Mathews, 1997). Until recently research has had to rely on the natural occurrence of these biases, but now new work suggests that the experimental induction of biases may provide an additional method of investigation (Hertel, Mathews, Peterson, & Kintner, 2003; MacLeod, Rutherford, Campbell, Ebsworthy, & Holker, 2002; Mathews & Mackintosh, 2000; Mathews & MacLeod, 2002; Yiend & Mathews, 2002). This work has shown that it is possible to induce processing biases in non-anxious individuals that mimic those observed in anxious populations. Not only do these induced biases appear to generalise to new material, but they also lead to increases in vulnerability to anxious mood, thereby providing the first direct evidence that cognitive biases have a causal role in anxiety disorders (MacLeod et al., 2002; Wilson, MacLeod, & Mathews, 2003). Here we report three experiments that, although not looking at anxiety vulnerability per se, extend the findings on laboratory-induced biases in interpretation by exploring the duration of their effects. Before explaining the rationale behind the current studies the recent literature on induction of interpretation bias will be briefly reviewed.

Two parallel lines of empirical investigation have looked at the induction of attentional and interpretative biases, respectively. Here we will focus only on the latter, but more comprehensive reviews are provided by Mathews and MacLeod (2002) and Yiend and Mathews (2002).

One variant of training biases in interpretation has made use of emotional homographs (Grey & Mathews, 2000). Grey and Mathews presented participants with emotionally ambiguous words such as ‘stroke’, and forced them to consistently select either the threatening or the positive/neutral meaning using a word fragment completion task for disambiguations of the original word (e.g. fragment of ‘cat’ or ‘disease’). The efficacy of training was then tested using tasks in which participants had to spontaneously disambiguate new material. For example, participants were faster to resolve new fragments of words consistent with the valence of their training (e.g. for positive training, ‘batter: pancake’), compared to fragments of inconsistent valence (e.g. for positive training ‘batter: assault’). Grey and Mathews concluded that it is possible to train biases in interpretation and that these generalise to novel, emotionally ambiguous material.

More importantly for the data we present here, they also addressed the role that active generation of meanings might play in securing the training effects observed. In initial experiments, participants had to actively generate the training-congruent meaning of the homograph by completion of the word fragment. In order to test whether such generation was critical to the efficacy of training, a variant of training was used in which the valenced associate (‘cat’ or ‘disease’, in the above example) was presented first and was followed by the ambiguous homograph (‘stroke’). Participants merely had to verify whether or not the two words were related. This method also produced significant training-congruent effects on new material, leading Grey and Mathews to conclude that engaging in the process of meaning generation during training was unnecessary and that simple exposure to the appropriate meanings was sufficient to induce biases in subsequent processing.

Mathews and Mackintosh (2000) introduced an alternative way of inducing biases in interpretation using passages of socially ambiguous text rather than single words (see Hirsch & Mathews, 1997). The last word was always a fragment to be completed, which then resolved the emotional meaning of the text in either a positive or negative direction (see Section 2.1.3 for similar examples). In this way, participants were forced to make predominantly negative or positive/neutral interpretations, according to their assigned training condition, and this interpretation was reinforced with a subsequent comprehension question.

To test the effects of this training Mathews and Mackintosh used a recognition test similar to that described in Eysenck, Mogg, May, Richards, and Mathews (1991). Ambiguous social texts were presented, as in training, but fragment completions and questions now maintained the ambiguity of meaning. Participants were later shown a series of disambiguating sentences that they had to rate according to their similarity to these previous texts. An induced interpretative bias would be reflected by higher recognition ratings for sentences that disambiguated in the same direction as training, than for those of the opposite direction. Each ambiguous test passage had two disambiguating target sentences that reflected accurately the two possible meanings of the preceding text (threatening or benign), but also two foil sentences, that were positively and negatively valenced, but whose content differed from the test passage. If training had induced a general priming effect of all congruent material, this would show up in a consistent recognition difference for foil as well as target sentences. Mathews and Mackintosh found significant training effects for both foils and targets, with the latter effect typically being significantly larger. It was concluded that, while this training method did produce a general affective priming effect, it also produced specific biases in the interpretation of newly encountered ambiguous items. They also found that state anxiety measures given before training commenced and again after the final test showed significant changes in anxious mood, congruent with the direction of training.

Following Grey and Mathews (2000), Mathews and Mackintosh addressed the importance of the active generation of meaning for the efficacy of training, by comparing ‘active’ and ‘passive’ training versions. The latter involved presenting passages that were already disambiguated to make them threatening or benign, followed by neutral fragment completions and questions. As predicted from Grey and Mathews, training effects persisted under these passive exposure conditions; however, the congruent state anxiety changes observed in generation training were no longer present. This suggests that the active deployment of interpretative processing during bias induction is crucial in producing mood change.

The developing body of the literature on bias induction shows unequivocally that it is possible to experimentally manipulate an individual's interpretative processing style. More importantly, they provide the first direct support for cognitive biases causing mood change, which is consistent with the long held hypothesis that these biases could be causally implicated in clinical disorders. The successful induction of cognitive processing biases also raises the question of potential therapeutic benefits, which is one of the long-term aims of work in this area. Specifically, if biases can be induced in non-anxious populations and produce congruent effects on mood, is it possible that they could also be reduced in anxious populations, with consequent beneficial effects on mood? However, we feel it is important to understand the characteristics of the processes being induced before moving to explore potential clinical or therapeutic applications. Similarly, we did not set out to demonstrate the causal role of induced biases in producing mood change and indeed our methodology, with its absence of a baseline measure of biased processing, does not allow strong conclusions to be drawn on this issue.

Instead, we focus on exploring the temporal characteristics of induced interpretative bias. The experiments we describe here used Mathews and Mackintosh's procedures for training biases in interpretation, to explore both the duration and nature of the cognitive effects produced. Experimentally induced biases could provide a convenient means of exploring in the laboratory the biases seen in clinical populations. However, more work is needed before we can assume that these experimentally induced biases are closely analogous to those which occur naturally. Here, we make the first step along this road by asking how transient are the consequences of induced interpretative biases?

In Experiment 1, we introduced a 20-min delay between training and test in order to examine the durability of both the induced cognitive bias and the associated mood change. In Experiment 2, we explored whether the type of training used influenced the durability of these effects. Specifically, we contrasted active training (participants learn to generate the valenced meaning of the text themselves) with passive (they are simply exposed to valenced information). Finally, we investigated the residual effects of active interpretative training after a much longer delay of 24 h.