Figure 1 displays a choropleth map of regional average mental health (GHQ Caseness) in the UK. London’s residents experienced the worst subjective wellbeing for the entire duration of the dataset, including during the pre-pandemic period. Immediately after the declaration of the pandemic in March 2020, we observe a radical decline in people’s mental wellbeing. April 2020 is clearly the worst month. Generally, mental wellbeing showed an improvement as summer approached, which brought with it an easing of lockdown restrictions; this was followed by a deterioration again in winter with lockdowns reinstated. These panels in Fig. 1 portend our main results from the more sophisticated analysis – that increased mobility restrictions due to the lockdowns likely resulted in poorer mental health.
We find a very strong correlation between movement reduction and average mental health outcomes (prevalence in mental health issues). Figure 2 presents the linear fit between the mobility measure and mental health outcomes aggregated by regions and survey waves (averages). The correlation coefficients of regional-level mobility change and mental health are 0.61 and 0.87 for the two derivative measures of mental health (GHQ in a Likert scale, which ranges from 0 to 36 (Fig. 2a), and GHQ Caseness score, which ranges from 0 to 12 (Fig. 2b); in both cases, higher values correspond to worse mental health), respectively.
The largest change in the duration spent at home is experienced in the earlier parts of the pandemic (April and May 2020, for instance, have the darker shades of green, which appear toward the right of each panel). This period is also associated with poorer mental health. A decrease in mobility at the societal level is also strongly associated with poorer mental health across all 12 regions. In Fig. 3, each dot represents the monthly average within each region, with the change in time spent at home (the horizontal axis) representing the deviation from the baseline at January 2020. Darker colours represent later waves.
Residents of London again display the strongest relationship (ρ = 0.97) between the lockdown restrictions and mental health. For instance, people living in London experienced the most radical change in the duration spent at home – about 30% at its most extreme (in April 2020), which is associated with the worst average GHQ Caseness score in Fig. 2. Visually, the other regions are fairly similar to each other, with correlation ranging from 0.77 (Northern Ireland) to 0.95 (West Midlands).
In Table 1, we show the main regression results using individual level observations with the GHQ Caseness score as the outcome variable. As a benchmark, we begin with a bivariate model (Column 1) and subsequently include COVID test results and socio-demographic variables (Column 2), pre-existing mental health or long-term illness history (Column 3), COVID-19 prevalence and government stringency level in the last seven days (Column 4), region fixed effects (Column 5), wave fixed effects (Column 6), and individual fixed effects (Column 7). The estimated relationship between movement reduction and mental health deterioration is statistically significant at least at the 5% level. The main results are also robust to all specifications including the alternative derivative measure of mental health (i.e., Likert; Supplementary Table 2), using the past-14 days average for mobility, COVID-19 cases, and stringency measures (Supplementary Table 3), and clustering standard errors at the region or region and survey wave levels (Supplementary Table 4).2F2F
Table 1
Movement restrictions worsen mental wellbeing
| (1) | (2) | (3) | (4) | (5) | (6) | (7) |
Change in duration of time spent at home (%) | 0.046*** | 0.047*** | 0.047*** | 0.046*** | 0.047*** | 0.037** | 0.040** |
(0.00131) | (0.00135) | (0.00136) | (0.00186) | (0.00188) | (0.0128) | (0.0126) |
COVID-19 positive | | 0.21*** | 0.20*** | 0.13*** | 0.14*** | 0.13*** | 0.11** |
| | (0.0343) | (0.0337) | (0.0348) | (0.0349) | (0.0354) | (0.0384) |
Age | | -0.018*** | -0.014*** | -0.015*** | -0.015*** | -0.015*** | |
| | (0.00202) | (0.00187) | (0.00187) | (0.00187) | (0.00187) | |
Female | | 0.74*** | 0.56*** | 0.55*** | 0.56*** | 0.56*** | |
| | (0.0406) | (0.0367) | (0.0367) | (0.0367) | (0.0367) | |
Marital status | | | | | | | |
Married/civil partnership | | -0.31*** | -0.19** | -0.18** | -0.18** | -0.18** | |
| | (0.0679) | (0.0622) | (0.0622) | (0.0621) | (0.0621) | |
Separated/divorced/widowed | | 0.0074 | -0.014 | 0.0017 | -0.0029 | -0.0047 | |
| | (0.0847) | (0.0752) | (0.0752) | (0.0753) | (0.0753) | |
Living with partner | | -0.29*** | -0.21*** | -0.21*** | -0.21*** | -0.21*** | -0.16** |
| | (0.0457) | (0.0440) | (0.0440) | (0.0440) | (0.0440) | (0.0558) |
Education | | | | | | | |
No qualification | | -0.24* | -0.32** | -0.31** | -0.32** | -0.31** | |
| | (0.114) | (0.0979) | (0.0979) | (0.0979) | (0.0978) | |
Other qualification | | -0.11 | -0.14† | -0.13† | -0.13 | -0.13 | |
| | (0.0924) | (0.0801) | (0.0801) | (0.0802) | (0.0802) | |
A level | | -0.016 | 0.0050 | 0.0023 | -0.00077 | -0.00088 | |
| | (0.0667) | (0.0596) | (0.0596) | (0.0596) | (0.0596) | |
Other higher degree | | 0.077 | 0.093 | 0.095 | 0.092 | 0.092 | |
| | (0.0733) | (0.0652) | (0.0652) | (0.0652) | (0.0652) | |
Degree | | 0.18** | 0.19*** | 0.19*** | 0.20*** | 0.20*** | |
| | (0.0596) | (0.0535) | (0.0535) | (0.0538) | (0.0538) | |
Live in rural area | | -0.070 | -0.043 | -0.044 | -0.046 | -0.047 | |
| | (0.0451) | (0.0405) | (0.0406) | (0.0420) | (0.0420) | |
Housing status | | | | | | | |
Mortgage | | 0.27*** | 0.12** | 0.11* | 0.12* | 0.12* | |
| | (0.0531) | (0.0481) | (0.0481) | (0.0481) | (0.0481) | |
Renting | | 0.76*** | 0.33*** | 0.33*** | 0.34*** | 0.34*** | |
| | (0.0698) | (0.0617) | (0.0617) | (0.0619) | (0.0619) | |
Employment | | | | | | | |
Unemployed | | 0.54*** | 0.37*** | 0.37*** | 0.37*** | 0.37*** | 0.43*** |
| | (0.0473) | (0.0450) | (0.0451) | (0.0451) | (0.0451) | (0.0663) |
Self-employed | | 0.21*** | 0.22*** | 0.22*** | 0.22*** | 0.22*** | 0.048 |
| | (0.0603) | (0.0569) | (0.0570) | (0.0570) | (0.0571) | (0.0927) |
Household composition | | | | | | | |
Aged 0–4 | | 0.024 | 0.074 | 0.074 | 0.075 | 0.075 | |
| | (0.0511) | (0.0493) | (0.0492) | (0.0492) | (0.0492) | |
Aged 5–15 | | 0.062* | 0.062* | 0.063* | 0.062* | 0.063* | |
| | (0.0291) | (0.0273) | (0.0273) | (0.0273) | (0.0273) | |
Aged 70 or older | | -0.10* | -0.071† | -0.071† | -0.070† | -0.071† | |
| | (0.0433) | (0.0412) | (0.0412) | (0.0412) | (0.0413) | |
Pre-COVID GHQ (Caseness score) | | | 0.38*** | 0.38*** | 0.38*** | 0.38*** | |
| | (0.00825) | (0.00825) | (0.00825) | (0.00825) | |
Long-standing illness or impairment | | | 0.48*** | 0.49*** | 0.49*** | 0.49*** | |
| | (0.0414) | (0.0414) | (0.0414) | (0.0414) | |
Case per 1,000 people | | | | 0.34*** | 0.33*** | 0.22† | 0.27* |
| | | | (0.0443) | (0.0445) | (0.131) | (0.128) |
Stringency index | | | | 0.0014* | 0.0013* | 0.0019 | 0.0014 |
| | | | (0.000600) | (0.000605) | (0.00220) | (0.00216) |
Constant | 1.73*** | 2.01*** | 1.17*** | 1.08*** | 1.12*** | 1.31*** | 2.05** |
| (0.0277) | (0.125) | (0.115) | (0.117) | (0.152) | (0.300) | (0.669) |
Region FE | No | No | No | No | Yes | Yes | Yes |
Wave FE | No | No | No | No | No | Yes | Yes |
Individual FE | No | No | No | No | No | No | Yes |
Observations | 117134 | 111450 | 109865 | 109865 | 109865 | 109865 | 117062 |
Individuals (cluster) | 18617 | 17454 | 17141 | 17141 | 17141 | 17141 | 18609 |
R2-within | 0.016 | 0.017 | 0.016 | 0.018 | 0.018 | 0.019 | 0.019 |
R2-between | 0.005 | 0.077 | 0.258 | 0.259 | 0.259 | 0.259 | 0.001 |
R2-overall | 0.007 | 0.056 | 0.182 | 0.183 | 0.183 | 0.184 | 0.002 |
Prob. > F. | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 |
Notes: GLS regressions. Dependent variable: Mental wellbeing (GHQ Caseness score). Reference group: Male, Single, Not living with a partner, GCSE, Live in Urban area, Owned outright, and Employed. Standard errors (clustered at individual level) in parentheses. † p < .10; * p < .05; ** p < .01; *** p < .001. |
A 10-percentage-point increase in time spent at home (compared to pre-pandemic baseline January 2020) is associated with an average increase of 0.37–0.47 in GHQ Caseness score (which has a standard deviation of 3.35). In standardized terms, a one standard deviation decrease in residential mobility is associated with a 0.066–0.084 standard deviations increase in the GHQ Caseness score. The relationship holds true for GHQ in a Likert scale, albeit slightly weaker (Supplementary Table 2). In general, the relationship between the change in the time spent at home and mental distress is positive – that is, a stay-at-home order predicts worse mental wellbeing. The size of the effect is noticeable and is comparable or higher than the effect of other important observable characteristics, such as unemployment (+ 0.40 points in the GHP Caseness score) or marriage (-0.20 points). We also estimate the impact of lack of mobility on the single components of the GHQ Caseness score (Supplementary Fig. 3). Overall, decreasing mobility worsens all aspects of mental health, and the most noticeable effects are found on the inability to concentrate, to make decisions and to enjoy day to day activities. Interestingly, other aspects, such as feeling worthless and feeling under strain are less affected.
Other independent variables’ effects on mental wellbeing follow the literature in the field. Mental health seems to improve for older people and those who live with a partner, while women report on average worse mental health. Mental health also increases for those who are employed and have no pre-existing health conditions. Unsurprisingly, increases in the number of COVID-19 cases and the stringency of restrictive measures worsen mental wellbeing.
Heterogenous Effect Of Lockdown On Mental Health
In Fig. 4, we show how subjective wellbeing (measured as GHQ Caseness score) evolves over time for different socioeconomic and demographic groups identified in the data. Notably, females, children, young adults, single households, and those with a long-term illness have tracked much worse than other groups. Those renting their homes also experienced worse mental health. Nearly all of these groups felt improvements in mental wellbeing between April 2020 and June 2020, which may reflect an adaptation to the “new normal”, or public and private policies that improved individual and social wellbeing, or indeed the easing of lockdown measures. There is a notable deterioration of wellbeing around the time the second lockdown starts in England (November 2020), but mental health also improves as the country transitions away from the day when the lockdown was introduced – similar to the evolution after the initial lockdown in March 2020.
In the following two tables (Tables 2 and 3), we interact the change in home duration with a number of different individual characteristics to examine heterogeneity over groups. Recall from Fig. 4 that some groups experienced deeper declines in mental wellbeing – interacting the change in home duration with these variables allows us to demonstrate the kind of person that might be more adversely impacted by mobility restrictions, at least in terms of their effect on mental health. The set of control variables for these regressions with interaction terms are the same as those in Table 1.
Table 2
Interactions with age, gender, marital status, living with partner, and educational attainment
| (1) | (2) | (3) | (4) | (5) |
Subjective wellbeing (GHQ): Caseness score | Age | Gender | Marital status | Living with partner | Education |
Δ home duration | 0.072*** | 0.022† | 0.042** | 0.037** | 0.032* |
| (0.0135) | (0.0129) | (0.0130) | (0.0129) | (0.0131) |
Age*Δ home duration | -0.00070*** | | | | |
| (0.0000852) | | | | |
Female*Δ home duration | | 0.023*** | | | |
| | (0.00262) | | | |
Married/civil partnership*Δ home duration | | | -0.0078* | | |
| | | (0.00350) | | |
Separated/divorced/widowed*Δ home duration | | | -0.0086† | | |
| | | (0.00463) | | |
Living with partner*Δ home duration | | | | -0.0013 | |
| | | | (0.00305) | |
No qualification*Δ home duration | | | | | -0.013† |
| | | | | (0.00659) |
Other qualification*Δ home duration | | | | | -0.0096 |
| | | | | (0.00590) |
A level*Δ home duration | | | | | 0.0054 |
| | | | | (0.00436) |
Other higher degree*Δ home duration | | | | | 0.0018 |
| | | | | (0.00474) |
Degree*Δ home duration | | | | | 0.0075† |
| | | | | (0.00386) |
Age | -0.0042† | -0.015*** | -0.015*** | -0.015*** | -0.015*** |
| (0.00223) | (0.00187) | (0.00187) | (0.00187) | (0.00187) |
Female | 0.56*** | 0.21*** | 0.56*** | 0.56*** | 0.55*** |
| (0.0367) | (0.0523) | (0.0367) | (0.0367) | (0.0367) |
Marital status | | | | | |
Married/civil partnership | -0.18** | -0.18** | -0.066 | -0.18** | -0.18** |
| (0.0621) | (0.0622) | (0.0804) | (0.0621) | (0.0621) |
Separated/divorced/widowed | -0.00031 | -0.0046 | 0.12 | -0.0047 | -0.0047 |
| (0.0753) | (0.0753) | (0.102) | (0.0753) | (0.0753) |
Living with partner | -0.21*** | -0.21*** | -0.21*** | -0.20** | -0.21*** |
| (0.0440) | (0.0440) | (0.0440) | (0.0619) | (0.0440) |
Education | | | | | |
No qualification | -0.32** | -0.31** | -0.31** | -0.31** | -0.13 |
| (0.0978) | (0.0978) | (0.0978) | (0.0978) | (0.137) |
Other qualification | -0.13 | -0.13 | -0.13 | -0.13 | 0.0089 |
| (0.0803) | (0.0803) | (0.0802) | (0.0802) | (0.118) |
A level | 0.00072 | -0.0021 | -0.00023 | -0.00082 | -0.080 |
| (0.0596) | (0.0596) | (0.0596) | (0.0596) | (0.0864) |
Other higher degree | 0.093 | 0.091 | 0.092 | 0.092 | 0.065 |
| (0.0652) | (0.0652) | (0.0652) | (0.0652) | (0.0935) |
Degree | 0.20*** | 0.20*** | 0.20*** | 0.20*** | 0.084 |
| (0.0538) | (0.0539) | (0.0538) | (0.0538) | (0.0772) |
Constant | 0.53† | 1.31*** | 1.01*** | 1.08*** | 1.29*** |
| (0.307) | (0.301) | (0.303) | (0.301) | (0.305) |
Control | Yes | Yes | Yes | Yes | Yes |
Region fixed-effects | Yes | Yes | Yes | Yes | Yes |
Wave fixed-effects | Yes | Yes | Yes | Yes | Yes |
N | 109865 | 109865 | 109865 | 109865 | 109865 |
N (cluster) | 17141 | 17141 | 17141 | 17141 | 17141 |
R2-within | 0.020 | 0.020 | 0.019 | 0.019 | 0.019 |
R2-between | 0.259 | 0.259 | 0.259 | 0.259 | 0.259 |
R2-overall | 0.184 | 0.184 | 0.184 | 0.184 | 0.184 |
Prob. > F. | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 |
Notes: Control variables include COVID positive, living in rural areas, housing status, employment, household composition, pre-COVID mental health, long-term illness, COVID case statistics, and stringency index. Reference group: Male, Single, Not living with a partner, and GCSE. Standard errors (clustered at individual level) in parentheses. † p < .10; * p < .05; ** p < .01; *** p < .001. |
Table 3
Interactions with urbanity, home ownership, employment status, and other health conditions
| (6) | (7) | (8) | (9) | (10) |
Subjective wellbeing (GHQ): Caseness score | Live in rural area | Home ownership | Employment status | Pre-existing mental health | Long-term illness |
Δ home duration | 0.037** | 0.031* | 0.040** | 0.040** | 0.043*** |
| (0.0128) | (0.0129) | (0.0128) | (0.0128) | (0.0128) |
Live in rural area*Δ home duration | -0.0081** | | | | |
| (0.00301) | | | | |
Mortgage*Δ home duration | | 0.014*** | | | |
| | (0.00287) | | | |
Renting*Δ home duration | | 0.0012 | | | |
| | (0.00408) | | | |
Unemployed*Δ home duration | | | -0.012*** | | |
| | | (0.00280) | | |
Self-employed*Δ home duration | | | 0.0041 | | |
| | | (0.00479) | | |
Pre-COVID GHQ (Caseness score)*Δ home duration | | | | -0.0014* | |
| | | (0.000566) | |
Long-standing illness*Δ home duration | | | | | -0.018*** |
| | | | | (0.00289) |
Live in rural area | 0.070 | -0.047 | -0.047 | -0.046 | -0.046 |
| (0.0589) | (0.0420) | (0.0420) | (0.0420) | (0.0420) |
Mortgage | 0.12* | -0.090 | 0.12* | 0.12* | 0.12* |
| (0.0481) | (0.0627) | (0.0481) | (0.0481) | (0.0481) |
Renting | 0.34*** | 0.32*** | 0.34*** | 0.34*** | 0.34*** |
| (0.0619) | (0.0873) | (0.0618) | (0.0618) | (0.0618) |
Unemployed | 0.37*** | 0.37*** | 0.53*** | 0.37*** | 0.37*** |
| (0.0451) | (0.0451) | (0.0599) | (0.0451) | (0.0451) |
Self-employed | 0.22*** | 0.22*** | 0.16† | 0.22*** | 0.22*** |
| (0.0571) | (0.0571) | (0.0889) | (0.0570) | (0.0571) |
Pre-COVID GHQ (Caseness) | 0.38*** | 0.38*** | 0.38*** | 0.40*** | 0.38*** |
| (0.00825) | (0.00825) | (0.00825) | (0.0121) | (0.00825) |
Long-standing illness or impairment | 0.48*** | 0.48*** | 0.48*** | 0.48*** | 0.75*** |
| (0.0414) | (0.0414) | (0.0414) | (0.0414) | (0.0587) |
Constant | 1.08*** | 1.17*** | 1.03*** | 1.04*** | 0.99*** |
| (0.301) | (0.302) | (0.301) | (0.300) | (0.300) |
Control | Yes | Yes | Yes | Yes | Yes |
Region fixed-effects | Yes | Yes | Yes | Yes | Yes |
Wave fixed-effects | Yes | Yes | Yes | Yes | Yes |
N | 109865 | 109865 | 109865 | 109865 | 109865 |
N (cluster) | 17141 | 17141 | 17141 | 17141 | 17141 |
R2-within | 0.019 | 0.020 | 0.019 | 0.019 | 0.020 |
R2-between | 0.259 | 0.259 | 0.259 | 0.259 | 0.259 |
R2-overall | 0.184 | 0.184 | 0.184 | 0.184 | 0.184 |
Prob. > F. | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 |
In Table 2, the own-effect of the change in mobility is consistently positive, although lacking in statistical significance when we interact it with the gender of the respondent (Column 2). The interaction with the female variable, however, shows that women suffered more than men over the period. Older respondents and those who were partnered were more resilient (Columns 1, 3, and 4). Finally, more educated individuals suffered more (Column 5).
For ease of interpretation, we also graphically represent the estimation results of Table 2 in Fig. 5. For almost the entire range of the percentage change in time spent at home, women are worse off than men. The gradient is also consistent for the interaction with age: the larger the change in home duration, the worse off people are, but this relationship is much stronger for younger people than for older people.
In Table 3, we continue with the interactions of the change in time spent at home with the following independent variables: living in an urban area, homeownership, employment status, a measure of pre-existing mental health (pre-COVID), and an indicator for having a long-term illness. Those living in rural areas, the unemployed, and those who had a long-term illness before COVID-19 started are less affected by the change in mobility. Similarly, individuals with better pre-pandemic mental health suffered more from restrictions to mobility.
Similar to Fig. 5, we also graphically represent these results in Fig. 6. Those who are paying off a mortgage and renting – perhaps because of increased financial pressure – show a larger deterioration in their mental health than those who own their domicile (Fig. 6a). The self-employed are also more adversely affected than those who are employed, and the unemployed are hardly affected at all (Fig. 6b). Across the range of the change in home duration, the effect of restrictions to mobility on mental health is less for those who have previous mental health issues or long-term illness (Figs. 6c and 6d).
In Fig. 7, we graphically represent the estimated coefficients on three-way interactions of mobility, gender, and age group (Fig. 7a) and mobility, gender, and an indicator for having a child aged 5–15 in the household (Fig. 7b). Younger women are more adversely affected than younger men, although the size of the differential declines as we move to older age groups. In addition, having a child in the household amplifies the negative relationship between mental health and being female during periods of increased mobility restrictions.
These estimates show how changes in average mobility impact the average mental health outcome. The main limitation of this approach is that, especially when we interpret interaction effects, the underlying assumption is that the movement change is of the same magnitude for the whole population. This is unlikely to be the case, as older people or people with long term illnesses were less mobile even before the pandemic started. Therefore, caution should be used when interpreting these estimates, as this type of measurement error could create attenuation bias in the results, which could overestimate the effects for groups of individuals whose movement change is less than the population average, and underestimate the effects for those whose mobility was higher.
However, the negative impact of lack of mobility may arise from several sources apart from restrictions to individuals’ movement. For example, it is possible that the lack of mobility in society, which is in turn reflected in a decrease of available services and overall social interaction, may have a negative impact on individual wellbeing, in addition to the effects due to the restrictions on the individual mobility.
[3] We obtain robust and qualitatively similar results when we 1) cluster the standard errors at the household level, 2) use a subsample of individuals who completed all nine COVID-19 waves (40.5% of the full sample) or 3) exclude those subjects who did not participate in the latest pre-COVID wave (wave 10 in 2019; 5.72% of the full sample).