Associations between dimensions of behaviour, personality traits, and mental-health during the COVID-19 pandemic in the United Kingdom

The study investigates how the COVID-19 pandemic affected mood and mental health in the UK and which population factors modulated these effects. The authors note unprecedented changes due to illness, distancing, and socio-economic restructuring, and emphasize that effects are likely widespread but idiosyncratic. Expert opinion suggested modulation by demographics (age, ethnicity), social networks, financial/occupational status, shielding/carer responsibilities, pre-existing mental health, maladaptive online technology use, personality traits, and compulsivity. Prior research on COVID-19 and mental health often focused narrowly, omitted the breadth of psycho-socio-economic variables, and rarely accounted for self-perceived impact (positive and negative), leaving unclear which groups were most affected and how population perspectives align with expert views. Recruitment biases in prior studies were also highlighted. The research question asks: Did standard mental health measures change from pre- to mid-pandemic? Were changes larger for specific subpopulations? What are the principal dimensions of self-perceived pandemic impact on daily life? Do these dimensions predict standard mental health measures? Which population, contextual, and trait factors predict self-perceived impact?

The paper situates its work within calls for urgent research on mental health impacts of COVID-19 and factors moderating such impacts. It notes limited pre-existing knowledge from prior pandemics and highlights limitations of early COVID-19 studies: narrow focus on specific mental health outcomes, limited consideration of diverse psycho-socio-economic variables, and neglect of self-perceived positive/negative impacts. Many prior studies used recruitment explicitly mentioning COVID-19 and targeted established cohorts, risking non-representative sampling. The authors reference work linking mental health risk to demographics, social networks, occupational/financial circumstances, shielding/carer roles, pre-existing psychiatric symptoms, maladaptive technology use, personality traits, and compulsivity, as well as literature on loneliness, abuse, and trajectories of anxiety/depression during lockdown. They also draw on broader literature linking access to outdoor/green space with reduced stress and better mental health, and on problematic internet use and compulsivity as potential vulnerabilities.

Design and participants: Data were collected via an online study platform (https://gbit.cognitron.co.uk) launched December 26, 2019, with major UK promotions on January 1, 2020 (BBC platforms) and May 2, 2020 (aligned with BBC2 Horizon). The dataset comprised 379,875 respondents (~90% UK residents). For mental health analyses, three epochs were defined: pre-pandemic (Dec 25–Jan 31; N=243,875), early-pandemic (Feb 1–May 1; N=10,003), and mid-pandemic (May 2–Jun 30; N=125,177). After exclusions (<16 years due to abbreviated questionnaire; completion time <4 minutes), the analytic sample included 233,268 pre-pandemic, 9,049 early-pandemic, and 112,046 mid-pandemic participants. Due to age skew in the early-pandemic sample, analyses focused on pre- vs mid-pandemic epochs, which were closely matched on sociodemographics (sex, handedness, first language, UK residence, ethnicity, education, earnings, occupational status). Measures: Standard mental health self-assessment items assessed frequencies over recent weeks of anxiety, depression, concentration problems, insomnia, hours slept, and tiredness. Responses were transformed to a numeric scale 0 (never) to 6 (>hourly). In May–June 2020, a bespoke PanDemic General Impact Scale (PD-GIS) probed self-perceived current and longer-term impacts via agreement with positive and negative statements; 79,779 of 112,046 mid-pandemic respondents completed PD-GIS. Statistical analysis: - General linear models (GLMs) were used to adjust mental health scores for age, sex, handedness, education, first language, country of residence, occupational status, and income. Residualized scores were compared between pre- and mid-pandemic epochs using F tests, reporting effect sizes in SD units (Cohen’s d) per Sawilowsky’s criteria. Day-by-day analyses over 31 days post-promotion in January and May assessed stability of effects. - Interaction GLMs examined whether pre-to-mid differences varied by population factors (age, sex, handedness, ethnicity, employment status, first language, country, education level, earnings), partialling out main effects. - PD-GIS item-level analyses summarized agreement distributions. Principal component analysis (PCA) with permutation testing determined dimensionality; seven components were retained and varimax-rotated, labeled based on loadings. - Associations between PD-GIS component scores and standard mental health measures were examined via bivariate correlations and canonical correlation analysis (CCA), with significance assessed using Bartlett’s chi-squared test with Lawley’s modification; train-test validation assessed overfit. - GLMs related PD-GIS components to sociodemographic/economic variables, home context, cohabitees, and work arrangements. - Additional GLMs examined PD-GIS differences for participants reporting pre-existing psychiatric, neurologic, and medical conditions predisposing to COVID-19 vulnerability, controlling for population variables; disorders with <100 cases were grouped. - A train-test linear modeling pipeline quantified multivariate associations of online technology use, personality traits (Big Five-like facets and others), compulsivity (reward drive, cognitive rigidity), and technology-related stress/addiction with PD-GIS components, after factoring out sociodemographics. Predictive performance was evaluated by correlations of predicted vs observed PD-GIS scores on held-out test data across 100 random splits. Software details and supplementary figures/tables provide full analytical specifications.

- Population-level changes pre- to mid-pandemic: Effect sizes were small to very small. Anxiety increased most (+0.28 SD; F(1,3.4425e+05)=6090.60; p<0.0001). Depression slightly decreased (-0.08 SD; F=537.75; p<0.0001). Problems concentrating slightly increased (+0.08 SD; F=507.88; p<0.0001). Insomnia change was negligible (+0.003 SD; F=0.628; p=0.4280). Hours slept increased (+0.14 SD; F=2035.70; p<0.0001). Tiredness decreased (-0.15 SD; F=1797.30; p<0.0001). These differences were stable across days within January and May observation windows. - Moderation by population factors: Age strongly moderated changes. Anxiety increase rose linearly with age (+0.40 SD for ages 60–80 vs +0.10 SD for teens). Depression increased slightly in older adults (+0.14 SD for 80+) but decreased in teens/young adults (-0.12 SD). Despite smaller change, baseline pre-pandemic anxiety and depression were much higher in teens/young adults than older adults (16–26 vs 76–86: anxiety +0.89 SD; depression +0.82 SD). Sleep increase was larger in younger adults (~+0.49 hours at age 16) versus ~0 SD at 60+. Occupational status: anxiety increases were largest for retired (+0.38 SD), homemakers (+0.31), and workers (+0.29). By sex: Other (+0.39 SD) > Female (+0.33) > Male (+0.22) for anxiety increase. Students showed the largest increase in hours slept (+0.37 SD). Effects of earnings, handedness, first language, and country were generally very small to negligible. - PD-GIS item-level perceptions: High agreement with positives: spending less/saving more (>70%), more time due to less commuting (~50%), more relaxed (>45%), enjoying simpler things (>65%), more time to read (~55%), more in touch with loved ones via apps (>80%), more pleasant environment (~75%), more wildlife (~70%), greater sense of community (>65%), belief in rapid advances in tech/healthcare (>50%), and that change is not necessarily for the worse (>60%). Strong disagreement with many negatives: loss of employment (~65%), loss of productivity (~70%), reduced attention to hygiene (~65%), loss of access to basics (~70%), increased conflict at home (~65%). Notable negatives: concern for loved ones' health (>85%), own health (>50%), loss of leisure/health activities (>75%), disconnectedness (~60%), loss of daily structure (>45%); ~30% reported drinking more, ~40% worried about less healthy lifestyle. - PD-GIS dimensionality: PCA with permutation testing identified seven components: 1) More time and less stress/tiredness; 2) Disrupted lifestyle (incl. loneliness); 3) Increased health concerns; 4) Positive outlook; 5) More conflict at home; 6) Improved environment; 7) More time for people at home. - PD-GIS and mental health associations: Bivariate correlations showed strongest links of anxiety with Health concerns and Disrupted lifestyle; depression and insomnia also related most to Disrupted lifestyle and Health concerns; irritability with Conflict at home. Positive components had small beneficial associations. CCA revealed seven significant modes (largest canonical r=0.56; smallest significant mode p=6.5e-26), confirming substantial multivariate relationships; train-test analyses indicated no overfit. - Predictors of self-perceived impact: Age had medium to very large associations with all PD-GIS components, often non-linear: older adults lower on More time/less stress and higher on Health concerns; younger people higher on Disrupted lifestyle; teens/early 20s higher on Conflict at home; working-age adults less Positive outlook; older working age higher on Improved environment; middle working age higher on More time for people at home. Occupational status and cohabitees were prominent: healthcare workers benefited least in More time/less stress and had greater Health concerns but reported better sleep/engagement and less Disrupted lifestyle; furloughed and home workers reported more time/less stress but, with students, more Disrupted lifestyle. Disabled/shielded participants reported least benefit (low More time/less stress), highest Health concerns, and least Improved environment; retired also had heightened Health concerns and least More time for people at home. Living with small children associated with lowest More time/less stress and highest Conflict at home; living alone linked to highest Disrupted lifestyle and lowest More time for people and Conflict at home. Quality of outside space (but not house type) predicted more positive reports across components (more time/less stress, less disruption, fewer health concerns, less conflict, more improved environment). Higher education predicted lower Positive outlook. Males reported fewer Health concerns than females/other; ethnic differences were small except higher Health concerns among Asian, Hispanic, and Other. - Pre-existing conditions: Disrupted lifestyle was higher for those with depression. Health concerns were higher for COVID-19 risk conditions (diabetes, lung, heart, weakened immune system) and higher still for anxiety disorders and OCD (~+0.49 SD). ADHD was associated with higher Conflict at home (~+0.18 SD). Improved environment tended to be lower across conditions, notably bipolar (~-0.22 SD) and Parkinson’s disease (~-0.25 SD). - Traits and technology: Train-test linear models showed substantial variance in PD-GIS components explained by personality, compulsivity, and technology-use traits after controlling sociodemographics: mean r=0.32 for Health concerns, 0.24 for Disrupted lifestyle, 0.22 for Improved environment (all p<0.0001). Negative components (Disrupted lifestyle, Health concerns) were positively associated with technology addiction, stress from technology, reward drive, and cognitive rigidity, and negatively with self-security and conscientiousness. Negative impact related more to maladaptive online behavior than to time spent online per se; reading news was a prominent stressor, while using technology to stay connected was beneficial.

Findings indicate that while average changes in conventional mental health metrics from pre- to mid-lockdown were modest, there was substantial heterogeneity across the population. Age emerged as a dominant moderator: older adults experienced greater increases in anxiety and depression and fewer sleep-related benefits, whereas younger adults and teens remained at higher absolute levels of anxiety/depression and reported more disrupted lifestyles and home conflict. Self-perceived impact was multidimensional, with seven latent components capturing both positive and negative experiences. These components explained meaningful variance in standard mental health measures (canonical r up to 0.56), suggesting that interventions might target specific dimensions (e.g., mitigating health concerns and lifestyle disruption, or enhancing positive outlook and connectedness). Population context mattered: healthcare workers exhibited both burdens (less free time, more health concerns) and benefits (better sleep/engagement); furloughed/home workers experienced more free time but also disruption; living with small children and cohabiting with parents increased home conflict. Access to quality outdoor space was a robust protective factor. Vulnerabilities were accentuated among those with anxiety/OCD (health worries) and ADHD (home conflict). Personality, compulsivity, and maladaptive technology use were collectively important in shaping impact, highlighting modifiable behavioral targets (e.g., reducing problematic internet use and technology-related stress, bolstering self-security, and addressing compulsive traits). Overall, the results support a tailored approach to mental health strategies during and after the pandemic, informed by demographic, environmental, occupational, clinical, and trait profiles.

This study leveraged a uniquely large, broadly inclusive UK sample with pre-pandemic baseline and mid-lockdown data to characterize mental health changes and self-perceived impacts of COVID-19. Average changes were small, with anxiety increasing most, while sleep improved and tiredness decreased. However, impacts were heterogeneous and strongly moderated by age, occupation, cohabitees, and living environment. A seven-dimensional structure captured self-perceived impacts, which in turn explained substantial variance in standard mental health measures. Healthcare workers, disabled/shielded individuals, and those living with small children faced distinct challenges, while access to outdoor space and positive use of technology were beneficial. Personality, compulsivity, and maladaptive technology use collectively predicted negative impacts, underscoring potential intervention targets. Future research should: - Develop and evaluate tailored interventions addressing health concerns and lifestyle disruption, stratified by age and contextual factors. - Investigate strategies to promote healthy technology use and reduce technology-related stress and problematic use. - Examine environmental interventions (e.g., improving access to quality outdoor spaces) for resilience. - Conduct longitudinal and interventional studies to establish causality and assess long-term outcomes, integrating PD-GIS-like measures with clinical endpoints.

Primary limitations include the cross-sectional comparisons (within a broader longitudinal framework), which limit causal inference; reliance on self-report measures; and potential sampling biases inherent to non-probabilistic online recruitment. Nonetheless, large sample size, close demographic matching between pre- and mid-pandemic epochs, and rigorous multivariate modeling mitigate some concerns. Representativeness remains a consideration, although BBC-mediated broad outreach and bias analyses suggested negligible differences between PD-GIS completers and non-completers in mental health measures.