Generalisation of modified interpretive bias across tasks and domains

Abstract

Patients with an anxiety disorder are characterised by a tendency to impose threatening interpretations on ambiguous information. Past research has examined the causal relationship between experimentally modified interpretive bias and its effects on anxiety. Effective modification of interpretation bias is typically shown on two specific tasks: an on-line reaction-time task and a post-training “recognition task”. Both tasks measure accessibility of negative or positive interpretations in a specific domain (social anxiety). From a theoretical and clinical perspective, it is important to know whether the effect of altered interpretation bias generalises to other tasks or domains. Therefore, in the present experiment, both the generalisation of Cognitive Bias Modification (CBM) to other tasks (a vignette and a video task) and the transfer to another domain (academic performance) were investigated. Results showed that the modified interpretive bias did not generalise to the other tasks, while it did transfer to another domain.

Keywords:

• Generalisation
• Interpretive bias
• Cognitive bias modification

Acknowledgements

We are grateful to Joseph Constans for supplying the vignettes and Nader Amir for supplying the video fragments. Furthermore, we want to thank Jeroen Scholten for his help in converting the video fragments and Yora Overes and Saskia Lunenborg for collecting the data. We thank the anonymous reviewers for their valued contribution to an earlier version of this paper.

Notes

¹Due to technical problems, one video clip containing an ambiguous comment could not be played. Hence, 71 clips were presented.

²We found this finding counterintuitive, since the vignette task only differs from the recognition task to a small extent. A second CBM-I experiment was conducted to replicate this null finding and the results were identical. Again a group of students with an average anxiety level was recruited. A total of 68 (60 female/8 male) participated and their mean age was 20.7 (SD=2.3). The positive and negative CBM-I groups did not differ in trait anxiety, t(66) = 0.34, ns. A 2 (Group) × 2 (Valence) × 2 (Target) ANOVA on the standard recognition task data revealed the predicted Group × Valence × Target interaction effect, F(1, 66) = 6.6, p<.01, =.09. Subsequent separate analyses of possible interpretations and foil sentences revealed Group × Valence interaction effects for both; for Interpretations, F(1, 66) = 26.0, p<.001, =.28 and for Foils, F(1, 66) = 17.1, p<.001, =.21. Positively trained participants interpreted information as significantly more positive than negative, while the negatively trained group did not differ in interpretations (see Table 2). Regarding the vignette task, again a main effect of Item-type was found, F(1, 66) = 47.4, p<.001, =.42, while the Group × Item-type interaction effect again failed to reach significance, F(1, 66) = 2.0, p=.16, =.03 (see Table 3).

³As a final step, both vignette datasets were pooled. A 2 (Group) × 2 (Dataset) × 2 (Item-type) ANOVA was conducted and the crucial Group × Item-type interaction effect was again not significant, F(1, 105) = 2.3, p=.13, =.02. A power analysis based on the combined dataset with an alpha level of .05, a correlation between repeated measures of .19, and a total sample size of 109 revealed that the power to detect large effect sizes was 1.00, the power to detect medium effect sizes was 0.98, and the power to detect small effect sizes was 0.36.