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

The COVID-19 pandemic (including lockdown) is likely to have had profound but diverse implications for mental health and well-being, yet little is known about individual experiences of the pandemic (positive and negative) and how this relates to mental health and well-being, as well as other important contextual variables. Here, we analyse data sampled in a large-scale manner from 379,875 people in the United Kingdom (UK) during 2020 to identify population variables associated with mood and mental health during the COVID-19 pandemic, and to investigate self-perceived pandemic impact in relation to those variables. We report that while there are relatively small population-level differences in mood assessment scores pre- to peak-UK lockdown, the size of the differences is larger for people from specific groups, e.g. older adults and people with lower incomes. Multiple dimensions underlie peoples’ perceptions, both positive and negative, of the pandemic’s impact on daily life. These dimensions explain variance in mental health and can be statistically predicted from age, demographics, home and work circumstances, pre-existing conditions, maladaptive technology use and personality traits (e.g., compulsivity). We conclude that a holistic view, incorporating the broad range of relevant population factors, can better characterise people whose mental health is most at risk during the COVID-19 pandemic.

home, better sleep due to spare time, greater sense of purpose in work, greater opportunity to exercise, improved natural environment, time to read for pleasure, work less stressful due to doing it from home, spending more time on hobbies, spending less and saving money, more social contact outside of the home, feeling less tired, feeling better connected with people at home, more wildlife, taking greater appreciation for the simple things in life, and being less stressed by daily responsibilities.
Full-text for the PD-GIS is as follows: Cue-Please indicate how well the following statements describe the impact of lockdown on you I am more concerned about my personal health I am more concerned about the health of my loved ones I feel more lonely than before There is an increased frequency or intensity of conflict at home I am preoccupied with consequences of getting COVID-19 Watching or reading the news brings on unpleasant emotions, or distressing thoughts that are hard to get rid of I have been grieving due to the loss of someone close to me I have lost employment, job opportunities or income I have lost leisure opportunities or activities important for my well-being I feel that my daily routine no longer has enough structure I have experienced changes in my sleep/wake patterns My lifestyle and/or daily routine has become more unhealthy I have not paid as much attention to my personal hygiene My productivity has gone down I have felt disconnected from important people in my life Infection control routines more than ever dominate my life I have lost access to essential goods, services or medication I am arguing more often with the people I live with I go online more to avoid the people I live with I feel that I have more time as am commuting less There is now more structure to my day I am happier as am able to spend more time with people within my home I am connecting online with people who I had trouble finding the time for before There is a greater sense of shared community I am working more efficiently/productively now I feel more relaxed as am spending more time at home I am sleeping better as have more spare time I feel a greater sense of purpose in the work that I do I am able to exercise more often When I go outside, the environment is quieter and more relaxing than it was before I have more time to read just for pleasure I find work less stressful now that I am doing it from home I am now spending more time on hobbies that I enjoy I am spending less and saving more money than before I have more social contact outside of my home I feel less tired now I am better connected now with the people I live with There seems to be more wildlife now I am enjoying the simple things in life more I feel less stressed by my daily responsibilities Cue -How will things change in the long term? I believe the world will be a better place than it was I believe the negative impact on the economy will be short lived Things will change but not necessarily for the worse I have more belief that we can cope with global problems like climate change Technology science and healthcare will improve more rapidly than before (e) Online Technology use Technology use was quantified by asking about frequency of use of the following, over the previous 4 week period: Smart Phone, Computer (Desktop or Laptop), Tablet Device, Gaming Console, Email, Social Media, reading the news, playing computer games, online gambling, working, learning/studying, shopping, streaming films or music, and searching for information online. Each question was responded to on a 7-point scale, from 0 (never) to 7 (more often than hourly every day).

(f) Stress from online technology
Stress from online technology was measured by asking the participants the following questions, regarding the past 4 weeks: When you checked Email, did it tend to make you feel stressed/unhappy or relieved/happy? When you used social media, did it tend to make you feel stressed/unhappy or relieved/happy? When you read the news, did it tend to make you feel stressed/unhappy or relieved/happy? When you played computer games, did it tend to make you feel stressed/unhappy or relieved/happy? The response options for each question were: "Mostly stressed/unhappy", "Mostly relieved/happy", "Both", or "Neither".

(g) Maladaptive ('Addictive') use of online technology
Maladaptive use of online technology was quantified using the following questions, which were based on expert consensus amongst the study team in the field of Problematic Usage of the Internet: How often did you check email or social media accounts after you went to bed? How often did you use internet related activities to block out disturbing thoughts or soothe yourself? How often did you choose to spend time on internet related activities to battle loneliness or boredom? How often did you suffer from negative financial consequences because of an online activity? How often did you check your email or social media account or equivalent before something else that you needed to do? How often did you try to stop an excessive online activity but feel a compulsion to continue? How often did you try to cut down the amount of time you spend on-line and fail? The questions asked about these areas over the preceding 4-week period. For the first question (using technology before bed), response options were 1 (never) to 5 (daily). For the other questions, response options were: 1 (never) to 7 (more than hourly every day).

Supplementary figure 7 Country of residence sample probability distributions
Proportion of participants per country of residence within the pre-, early-and mid-pandemic epochs. Y axis is proportion per epoch.

Supplementary figure 8 Earnings sample probability distributions
Proportion of participants per earnings bracket within the pre-, early-and mid-pandemic epochs. Y axis is proportion per epoch.

Supplementary figure 9 -Analysis of day-by-day mood self-assessment scores in January and May
Mean scores for mood self-assessment measures were contrasted separately for each of 31 days after the two promotion launches on January 1 st (blue, pre-pandemic) and May 2 nd (orange, mid-lockdown). Demographic variables including age, sex, handedness, ethnicity, first language, country of residence, education level, employment status and earning have been factored out. Y axis is in standard deviation units. X axis is days since launch. Shading represents the standard error of the mean for data collected on that day. The overall pattern of differences can be seen to be consistent throughout these two months with increased anxiety, increased sleep, reduced tiredness, and similar mean levels of depression, insomnia and problems concentrating. Therefore, the observed differences in mental health measures reflect sustained differences throughout these epochs, that is, as opposed to transient spikes in national mood on individual days. Source data are provided as a Source Data file.

Supplementary results 1 -PD-GIS mental health sampling bias analysis
One concern could be that people who opt to answer Cornonavirus-19 questionnaires are those for whom it is more relevant, e.g., due to their mental health status. To address this issue, we quantified sampling bias for the optional self-perceived impact sub-scale by analysing differences in mood measures for participants who did (79,736) minus did not (112046) opt to complete the PD-GIS. Differences in anxiety (0.018SDs t=2.0247 p=0.043), depression (-0.037SDs t=-4.4958 p<0.001), concentration (-0.073 SDs t=-8.4393 p<0.001), insomnia (0.032SDs t=3.4731 p<0.001), hours slept (-0.058SDs t=-7.1141 p<0.001) and tiredness (0.035SDs t=4.0017 p<0.001) scores were statistically significant but bi-directional with respect to valence, and critically, of negligible effect size scale. This accords poorly with the possibility of sampling bias towards people for whom mental health problems are most relevant during the pandemic in the context of PD-GIS analysis.

Supplementary figure 10 -PCA analysis of the PD-GIS items.
A MATLAB implementation of Horn's Parallel Analysis 6 was applied to estimate the number of significant components from the principal component analysis. PCA. This is a permutation-based approach whereby the true data are permuted and data reduced with PCA many times, producing distributions of variance explained by components at each index for statistical comparison to those observed for the unpermuted data. Estimated with 1000 permutations indicated 7 statistically significant components (greater than 95% of values within the corresponding null distribution) when the PD-GIS data were analysed in this manner. Application of the Kaiser convention would indicate 11 components with eigenvalues >1. Top left, cross correlation matrix for PD-GIS items. Top right, question-component loadings after varimax rotation. Bottom, scree plot. Note 11 th components places above the scree and prior to the 4 th inflection point. (All data and models are available for download from the UK Data Service). Source data are provided as a Source Data file.

Supplementary figure 11 -Sub sampling Train-Test pipeline to evaluate overfit in the PD-GIS by Mental Health Canonical Correlation analysis
Upper panel left. Bivariate Pearson's correlations between PD-GIS component scores and scores on Mood Self-Assessment items. Upper panel right. Canonical Correlation mode scores. Middle left. CCA mode scores for trained data sub sampling at different sizes (X axis is in thousands and Y axis is mode correlation value). Middle right, the same analyses conducted for data where the index of the X matrix was permuted, breaking the X-Y matrix linkage whilst retaining their inner structure. Note the near zero scores above 20K samples, indicating little overfit. Bottom left. Mode correlation scores when applying the trained CCA model to the held-out data, to which the model was naïve, with X axis corresponding to the number of participants in the trained set, whereby the held-out set comprises all other participants. Note that canonical r values approximate those of the trained set at higher sample size, indicating little overfit. Bottom right. The analysis of held-out data repeated for the permutated data. Source data are provided as a Source Data file.

Predicting individual differences in PD-GIS component scores from population variables
Supplementary

Supplementary table 24. PD-GIS by sociodemographic factors. Parameter estimates in standard deviation (SD) units
Parameter estimates for predictors in the GLM. p<0.05*, p<0..01**, p<0.001. All predictors are binary and can be interpreted as effect sizes in standard deviation units (apart from age, which is reported separately).

Continued overleaf -->
Parameter estimates for predictors in the GLM. p<0.05*, p<0..01**, p<0.001. All predictors are binary and can be interpreted as effect sizes in standard deviation units (apart from age, which is reported separately). Effects sizes highlighted in blue (negative) and green (positive).