Moser et al (2008) conducted a study, investigating interpretation bias in high and low socially anxious individuals. Participants were assessed for their behavioral responses and brain activity (using Electroencephalogram) to ambiguous sentences ending in positive or negative words.
The authors build their study on the long-standing research that anxious individuals interpret information in a negative/threatening way. This interpretation/ attention bias towards threat has been shown to maintain anxiety. (Hirsch & Clark, 2004). On the other hand, low anxious individuals have a positive view of themselves and the world. Thus, it would not be surprising that given an ambiguous situation, LA individuals would expect and accept positive outcomes as benign compared to the HA who live in the fear of a negative occurrence. This is what the authors hoped to find through their study by measuring the EEG activity of their participants as they responded to positive/negative endings of ambiguous scenarios. While most of the EEG research has focused on finding negative bias in the high anxious (Richards et al 2013, Mogg et al 2007), this study focuses on the absence of ‘positive bias’, or non-expectation of positive outcomes, as differentiating the high anxious from the low anxious.
Early research in interpretation bias in anxiety has mainly been behavioral (Eysenck 1991, Amin 1998), or in the form of self-report. The use of EEG was justified in the light of these studies, as it shows direct, time-bound and sequential brain activity even before response is elicited. Moser et al were keen to examine the psychophysiological correlates of interpretation bias in social anxiety which was not done before, as most EEG studies focused on attentional bias (Schmitz et al, 2012) or emotional face processing (Moser et al 2008, Richards et al 2013).
Choice of ERP
The Event Related Potential under investigation was P600, occurring 600ms after stimulus onset. This ERP is known to reflect violation of expectancy and is involved in unexpected and salient events. (Coulson et al, 1998). However, in such a grammatical decision task, P600 could be confounding as it is elicited even when confronted with mere ungrammatical stimuli. The researchers hypothesized that the LA group would show greater P600 activity towards negative endings which were unexpected by them , owing to their positive bias, whereas the HA would show no P600 difference for positive or negative due to their lack of positive bias. The researchers found no significant difference on behavioral measures but EEG results were as expected.
The sample included participants in and around University of Pennsylvania who were screened using the Social Phobia Inventory. Although the sample size was adequate (34), and fairly equally divided (16- HA, 18 ‘LA), it was not representative. Most of the participants were students, in the age range of 20s. The groups were equal only on demographic variables of age and gender. The authors failed to mention the sample’s visual ability, which is necessary when subjects perform a visual task. The method of scanning was also disputable. They were screened via phone, which is indicative of less control in establishing a common setting for undertaking the test.
The aim was to examine differences in processing of the positive and negative endings of the sentences. This was implicitly measured using EEG. Participants heard the sentences on speaker and saw the sentence-terminal words on the computer screen. Participants merely had to indicate whether the ending fitted the sentence grammatically. This ensured that they paid heed to and remembered the context of the sentence. The sentences were non-social, social (positive/negative/neutral endings) and neutral. However, the authors failed to mention counterbalancing of sentence and sentence terminals, which is important for controlling order effects. (Cozby, 2009).
In addition to SPIN, they were administered other test measures like LSAS, SIAS, BFNE, and DASS-21.Research shows consistency in scores of these tests. (Caballo et al, 2013). The difference between the scores obtained by the groups on these tests, was significant. The ERP chosen was P600 measured at point Cz.
Only correct trials were used for analysis. It would have been fair to include incorrect trials as well, to examine whether the brain activity, owing to the valence of the sentence terminals, was still the same. Estimates of effect sizes were calculated using partial eta square and Cohen’s d. An independent samples t test was conducted on behavioral and ERP data to confirm that the groups did not differ in their baseline performance. The hypothesis was tested using 2(group- low and high anxious) X 2 (sentence ending- positive and negative) ANOVA. In case of significant interaction of Group X Sentence ending, a follow up paired samples t test was conducted. Non-social sentences were not included in the analysis. This was a drawback as it would have been helpful to know if the anxious participants lacked a positive bias in social situations only or in mundane life activities as well.
Results showed no difference in baseline performance between groups. With respect to behavior measures, although the RT was lesser for negative endings in HA and greater in LA, the ANOVA and follow up t test did not yield significant difference. In case of ERP data, although main effect of group was not significant, interaction of Group X sentence ending was found to be significant. The follow-up t test showed larger P600 for negative endings in LA than positive endings. No P600 difference was seen in HA. In addition to the measured ERP, the researchers also analyzed data obtained at N400 and P300. They found similar but less robust results in these time windows, where, only the interaction effect was significant, as seen by a larger activity to negative endings in LA with no difference in HA. However the authors only state the temporal but not the spatial aspect of the EEG results.
Discussion and Conclusion
The researchers concluded that LA group showed positive interpretation bias, wherein they expected positive endings and were baffled by negative endings as seen by their greater P600 activity towards the latter. On the other hand, the HA group who were believed to lack positive bias, showed no difference in activity indicating that they did not expect positive endings. The results were discussed as consistent with previous (Hirsch & Mathews, 2000, Coulson et al 1998, Moser et al 2008) and have been supported by later findings. (Gu et al, 2010). The authors also made it a point to state that resource allocation did not bias results as it has in other studies. (E.g.: Eysenck, et al, 2007). The authors did not succumb to circular reasoning by assuming that the HA showed negative bias, rather they stated that HA group showed no bias at all. The EEG findings solely investigated presence/absence of positive bias, although negative bias was seen in the RTs. The authors have also thought of outcome of their research in terms of treatment strategies (Murphy et al 2007).
The study could have taken a broader view by examining negative bias occurring at later ERPs, as more recent studies have shown that negative bias occurs at a later stage of processing. (Richards et al, 2013). EEG findings could also have been supplemented by asking the participants to rate the valence of the sentence terminals. The study could have done better with using other types of stimuli to compare findings, measuring activity at more than one point, testing other types of anxiety and negative moods. Lastly, as the sample tested was not clinical, the generalizability of the findings could be questioned. Future research could use different neuroimaging methods due to limited ability of EEG to record subcortical sources. Despite the flaws, it was the first to examine the psychophysiological correlates of interpretation bias in anxiety and has no doubt, contributed to advancing the model of anxious pathology.