Abstract
The COVID-19 pandemic triggered profound shifts in social dynamics and mental health globally. This study investigates the psychological impact on Sri Lankan citizens across various socioeconomic backgrounds. Using a nationwide survey of 3,020 households, researchers applied factor analysis and unsupervised clustering to identify how different population subgroups coped with the crisis. The analysis revealed two distinct groups: a "socially connected" group that maintained external ties and experienced better mental well-being, and a "socially reserved" group that relied on social media as a coping mechanism but suffered from increased loneliness and distress. These findings suggest that genuine social interaction, rather than digital substitutes, served as the primary buffer against pandemic-related psychological strain.
Addressing the Mental Health Gap in Sri Lanka
While the global mental health impact of COVID-19 is well-documented, much of the existing research focuses on high-income nations. There is a significant gap in understanding these dynamics within low-income, culturally distinct regions like Sri Lanka. This study addresses this by examining how social interactions and social media usage influenced psychological distress—specifically loneliness, emptiness, and fear of death—across diverse socioeconomic backgrounds. The researchers aimed to discover if the population naturally divided into subgroups based on their psychological responses rather than traditional demographic categories.
A Data-Driven Approach to Psychology
The researchers employed a hybrid approach combining traditional statistics with machine learning to analyze the survey data:
- Factor Analysis (FA): Using Principal Component Analysis (PCA) with Varimax rotation, 11 psychological variables were condensed into four core factors: Coping Mechanisms, Symptoms, Work-Life Balance, and Peer Interactions/Connections.
- Spectral Clustering: To determine the natural number of population subgroups, the authors used a "Sigma Sweep" to observe the "eigengap," mathematically identifying that the population split into exactly two clusters ($K=2$).
- K-means Clustering: Following the identification of the optimal cluster count, K-means was used to assign each of the 921 respondents to a specific group for detailed behavioral analysis.
Dataset and Geographic Reach
The study utilized a nationwide, face-to-face survey conducted in late 2021. The original dataset comprised 3,020 households across 20 districts in Sri Lanka. After filtering for completeness and ensuring a symmetrical Likert scale for machine learning models, a processed dataset of 921 responses was used for the final analysis.
As shown in Figure 1, the distribution of the survey population closely mirrored the actual population distribution of Sri Lanka, ensuring that the findings are representative of the national context.
Social Connection vs. Social Media
The study's results provide critical insights into the nature of social interaction during a crisis:
- Social Media as a "Band-Aid": One of the most striking findings was that social media usage did not group with "Peer Interactions." Instead, it loaded onto the Coping Mechanisms factor alongside hobbies and substance use. This suggests that individuals used social media to distract themselves from distress rather than to build meaningful human connections.
- The "Socially Connected" Advantage: Group 01 (the "Socially Connected") reported increased social connectedness and better relationships with neighbors during the pandemic. This proactive maintenance of external ties led to significantly lower levels of loneliness and fear of death compared to Group 02.
- Demographics vs. Behavior: Surprisingly, the two clusters were not distinguishable by income, education, age, or ethnicity. The primary differentiator was an individual's inherent tendency toward being "socially connected" or "socially reserved."
- Work-Life Balance Friction: Factor analysis showed that relationships with office colleagues and family members often acted as "opposite elements," highlighting the psychological strain of the "new normal" where the home became a site of both professional and domestic obligation.
Key Results and Validation
The four identified factors explained 63.68% of the total variance in the psychological data. Confirmatory Factor Analysis (CFA) yielded a Goodness of Fit Index of 0.962, validating the four-factor structure. The "Socially Reserved" group (Group 02) showed significantly higher mean values for loneliness and emptiness, with high social media usage strongly correlated with this group, reinforcing its role as a substitute for real-world connection.
Moving Forward
The authors acknowledge limitations, including potential questionnaire bias and sample attrition during data cleaning (reducing the dataset to 30% of its original size). However, the findings offer a clear directive for future public health crises: policymakers should focus on facilitating "quality peer interactions"—even if virtual—rather than just digital entertainment to support mental health. The researchers have published the digitalized dataset online to encourage further AI-driven psychological assessments.
Summary updated