Hidden Markov modelling of eye movements in social anxiety : a data-driven machine-learning approach to eye-tracking research in psychopathology

Research output: Conference PapersRGC 32 - Refereed conference paper (without host publication)peer-review

View graph of relations

Author(s)

Related Research Unit(s)

Detail(s)

Original languageEnglish
Publication statusPublished - Mar 2019

Conference

Title2019 Anxiety and Depression Association of America Conference, ADAA 2019
PlaceUnited States
CityChicago
Period28 - 31 March 2019

Abstract

Background: Theoretical models propose that attentional biases in the presence of potential threats might account for the maintenance of social anxiety symptoms. However, eye-tracking studies have yielded mixed results. One reason for this is that these studies use arbitrary criteria to quantify eye-movements within predefined time windows and regions of interest, which either capture they dynamic nature of attention nor consider individual differences in eye-movement. Thus, a data-driven machine-learning method (Eye Movement analysis with Hidden Markov Models [EMHMM[ that has previously identified two face-viewing strategies (analytic/eye-centred vs holistic/nose-centred) in the general population was adopted. As the eye regions which covey more information about social interactions are perceived as especially threatening in people with social anxiety, an analytic pattern might reflect hypervigilance towards threats, while a holistic pattern resembles avoidance.
Methods: 60 adults (32 females) high and low in self-reported social anxiety participated in an eye-tracking experiment during which they freely viewed angry and neutral faces. Subjects' personalised face-viewing patterns were visualised by EMHMM and were clustered in analytic and holistic groups. This novel approach was them compared to traditional eye-movement indices including proportion of first fixations, total dwell time and latency of first fixations on the eye region.
Results: EMHMM analyses revealed heightened social anxiety in participants who were analytic for both angry and neutral faces compared to those who were analytic for neutral faces and holistic for angry faces, t(26) = 2.68, p = .013, d = 1.02. Moreover, participants who used the same face-viewing strategy for both angry and neutral faces showed higher social anxiety than those who used different strategies when viewing angry versus neutral faces, t(58) = 2.13, p = .013, d = 0.57. These relations between eye-movements and psychopathology were unique to social anxiety rather than general depressive or anxious symptoms, and were unattainable with traditional analyses.
Conclusion: This is the first study to adopt a data-driven machine-learning approach to investigate attentional biases in psychopathology. Our findings show that both hypervigilance to and avoidance of eye region are evident in socially anxious individuals, suggesting that social anxiety is characterised by attentional inflexibility to switch between strategies in response to different face emotions, instead of a single direction of attentional bias. This also implies that attentional bias modification that directs people's gaze away from threats might no benefit socially anxious individuals who are already avoidant. Future interventions should aim at training attentional flexibility that might improve the capability to transition between attentional strategies in different social situations.

Bibliographic Note

Information for this record is supplemented by the author(s) concerned.

Citation Format(s)

Hidden Markov modelling of eye movements in social anxiety: a data-driven machine-learning approach to eye-tracking research in psychopathology. / Chan, F.; Barry, T.; Chan, A. et al.
2019. Paper presented at 2019 Anxiety and Depression Association of America Conference, ADAA 2019, Chicago, Illinois, United States.

Research output: Conference PapersRGC 32 - Refereed conference paper (without host publication)peer-review