TL;DR
This paper introduces the Biometric Attachment Test (BAT), a novel, multimodal approach to assessing adult attachment security by analyzing physiological, linguistic, and behavioral responses to standardized stimuli. The BAT shows promise in providing a more objective and automated measure of attachment, validating its core assumptions and paving the way for more efficient and accessible mental health assessments.
What Problem Does This Paper Address?
Attachment theory, a cornerstone in understanding mental health, development, and relationships, has traditionally relied on psychometric instruments. However, these instruments often suffer from limitations inherent to traditional psychometrics, such as self-report biases and a lack of objective measurement. While existing methods like questionnaires and interviews have their strengths, they also present challenges: questionnaires can be prone to subjective biases, and interviews are costly, time-consuming, and can be influenced by interviewer subjectivity. This paper addresses the need for a more objective, automated, and multimodal assessment of adult attachment security, leveraging recent advancements in sensor technology and machine learning.
Methods
The core of this research is the development and initial validation of the Biometric Attachment Test (BAT). The BAT is an exposure-based test where participants are exposed to a short (9-minute) standardized set of visual and music stimuli. During this exposure, their immediate reactions and verbal responses are captured using multiple sensor modalities:
- Physiological: Heart rate (HR) and electrodermal activity (EDA) were recorded using an Empatica E4 wristband. Features like heart rate variability (HRV) and skin conductance response (SCR) were extracted.
- Behavioral: Facial expressions (using FACET's Emotient), head movement, and gaze direction (using OpenFace) were captured via webcam.
- Linguistic/Paralinguistic: Speech was recorded using a USB microphone, and paralinguistic features were extracted using COVAREP. Automated transcripts were generated using Google's Cloud Speech API, and linguistic features were analyzed using the LIWC dictionary.
The BAT utilizes "themes" as placeholders for stimuli, allowing for flexibility and rotation of visual content while maintaining the core narrative. This approach aims to overcome issues like stimulus leakage and priming effects common in fixed-stimulus tests. The study involved an exploratory analysis using correlation matrices to identify features associated with attachment security, followed by a confirmatory analysis using a composite "effects index." Repeated measures ANOVAs were used to assess the stability of these features across different stimuli sets.
Datasets
The study did not introduce a new benchmark dataset. Instead, it utilized existing resources for stimulus selection:
- Nencki Affective Picture System (NAPS)
- International Affective Picture System (IAPS)
- Geneva Affective Picture Database (GAPED)
- iStockPhoto (for additional images)
- A music set that provides scores on discrete perceived emotions.
The participant data was collected through a web-based survey and direct interviews.
Key Insights and Ideas
- Multimodal Fusion for Attachment: The core idea is that attachment security isn't a single observable trait but a complex construct best captured by integrating data from multiple modalities (physiology, behavior, language). This moves beyond traditional self-report or interview-based assessments.
- Themes as Flexible Stimuli: The concept of "themes" as placeholders for stimuli is a novel approach to psychometric testing. It allows for dynamic stimulus sets, preventing test obsolescence and enabling longitudinal studies without priming effects.
- Objective Measurement of Subjective States: By automatically analyzing physiological and behavioral cues, the BAT aims to provide objective markers for subjective internal states related to attachment security, addressing a key limitation of current methods.
- Theory-Driven Feature Selection: The study leverages established attachment theory to hypothesize which features should correlate with attachment security, guiding the feature extraction and analysis process.
Key Results
- Exploratory Analysis: 46 theory-consistent correlations were found between BAT features and attachment security measures (AAQ, AAP, AMMI). Examples include:
- Positive facial expressions correlating with attachment security (AAP).
- Inverse correlation between AMMI security (father) and low-frequency HRV.
- Inverse correlation between AAQ security (partners) and anger facial expressions.
- Confirmatory Analysis: A composite effects index was significantly correlated with attachment security in the AAQ and AAP, supporting the overall validity of the multimodal approach.
- Stimuli Set Stability: 84% of the theory-consistent features remained stable across three different BAT stimuli sets, suggesting that participants respond to the underlying themes rather than specific stimulus details. This supports the viability of rotating stimuli.
Limitations and Future Work
- Limited Sample Size: The study acknowledges that the small sample size for the AMMI measure might explain the lack of correlation with the composite effects index.
- Theory-Inconsistent Findings: A notable portion (29%) of significant correlations in the exploratory analysis went against theoretical predictions, suggesting these features might need to be discarded or further investigated.
- Future Algorithm Development: The paper highlights the ongoing work on developing automated scoring and classification algorithms for the BAT. This includes finding the optimal fusion formula for multimodal features to accurately predict attachment scores and classifications.
- Broader Attachment Dimensions: The BAT was designed to assess not only attachment security but also attachment anxiety, avoidance, and disorganization, as well as emotion regulation and relational trauma. Future work will focus on empirically validating these dimensions.
- Clinical Validation: The authors envision the BAT being deployed online for remote assessment, offering cost-effective and efficient mental health assessments.
Summary updated