HUMEDITロゴ

[Paper] Impact of Regional and National COVID-19 Measures on Social Contacts in the UK: A Longitudinal Natural Experiment

[Paper] Impact of Regional and National COVID-19 Measures on Social Contacts in the UK: A Longitudinal Natural Experiment

Summary

Background

The UK's response to COVID-19 transitioned from a national lockdown to localized interventions. In response to increasing cases, these measures were supplemented by national restrictions on contacts (the rule of six), a 10 pm closure for bars and restaurants, and the encouragement of working from home. Following these, a three-tier system applying different restrictions in different regions was rapidly introduced. As cases continued to rise, a second national lockdown was declared. A national survey was used to quantify the impact of these restrictions on epidemiologically relevant contacts.

Methods

The authors compared setting-specific contacts before and after the implementation of each restriction using paired measurements. They tested for differences using paired permutation tests on the mean changes and proportions of contacts.

Results

After the implementation of each measure, individuals tended to have fewer contacts than before. However, the magnitude of change was relatively small and varied. For example, the early closure of bars and restaurants did not appear to have a measurable impact on contacts, but working from home reduced the average daily work contacts by 0.99 (95% confidence interval CI 0.03-1.94). The rule of six reduced non-work and school contacts by an average of 0.25 (0.01-0.5) per day. The tier 3 measures also seemed to reduce non-work and school contacts, while the evidence for the impact of less stringent restrictions (tiers 1 and 2) was much weaker. Additionally, there might be evidence of a saturation effect when tier 1 individuals (the least restricted) significantly reduced their contacts upon entering lockdown, which was not reflected in similar changes among those already under stricter restrictions (tiers 2 and 3).

Conclusion

From summer to autumn of 2020, various local and national measures were implemented in the UK, leading to a gradual decrease in the number of contacts. However, these changes were smaller compared to the initial lockdown in March. This might be because many individuals already had a low number of contacts to begin with.

Keywords:

COVID-19 "Contact Survey", Lockdown, Pandemic, Disease Outbreak, Non-Pharmaceutical Interventions, UK, United Kingdom

Background

On March 23, 2020, the UK, along with the rest of the UK, entered national restrictions in response to COVID-19. This allowed leaving home only for essential shopping, medical needs, or one form of exercise per day. Educational institutions and non-essential retail stores were closed, as were the leisure and hospitality sectors. Many European countries also implemented national lockdowns, leading to significant reductions in the number of contacts, mobility, and transmission, ultimately resulting in decreased daily case numbers and fatalities.

As the incidence of cases decreased, national restrictions were eased. The UK transitioned to localized responses, applying stricter regulations only to specific areas with increasing case numbers. The first of these regional measures was announced in Leicester on June 29, followed by other regions, primarily in the north of England. The scope of regional regulations varied but could include early closures, take-away services only for bars and restaurants, bans on gatherings with other households, and travel restrictions.

In parallel with regional regulations, some national measures were introduced in response to rising cases. On September 14, the "rule of six" was announced in the UK, preventing gatherings of more than six people. On September 24, pubs and restaurants were required to close at 10 pm, and individuals were encouraged to work from home. As cases continued to rise, the government implemented a three-tier system, combining several restrictions ranging from Tier 1 (medium risk) to Tier 3 (very high risk). Subsequently, a second national lockdown was imposed from November 5 to December 2.

The impact of these less stringent measures remains unclear, as cases continued to rise in most regions even after their implementation. It is expected that detecting (possibly moderate) changes in cases, hospitalizations, or deaths following the introduction of restrictions will be challenging. This study addressed these issues by estimating the impact of these measures, if any, on setting-specific contacts epidemiologically relevant to individuals, using repeated measurements before and after the restrictions.

Methods

Ethical Guidelines

Participation in this opt-in survey was voluntary, and all data analysis was conducted using anonymized data. This study was approved by the ethics committee of the London School of Hygiene & Tropical Medicine.

Data

Data from UK CoMix survey participants and information about local and national regulations from the UK government were integrated. Details of the CoMix trial, including the study protocol and survey instruments, have been previously published. In summary, CoMix is an online survey where individuals record details of all direct (i.e., potentially risky) contacts from the day before the survey. Direct contact is defined as meeting someone with at least one word exchanged or having any form of skin contact with them. Contacts of individuals under 18 years old were collected with responses from their parents. Data were collected weekly from two alternating, broadly representative panels (each approximately 2,500 people), with each individual surveyed every two weeks. Between August 31, 2020, and December 7, 2020, data on the start and end dates of restrictions and their locations were extracted from the UK government. CoMix participants were considered to be affected by regional restrictions if they reported living in a lower-tier local authority area (an administrative region in the UK) with restrictions different from those applied nationally. To allow for two weeks of complete responses, data were restricted to 16 days before and after each restriction was implemented. The nearest survey responses to each restriction date were then extracted. Participants with missing survey responses on any aspect of restriction start were excluded, resulting in two records per person.

Limitations

Regional restrictions included a series of rules applied inconsistently across different areas. Most regional regulations fell into four categories: travel restrictions, non-essential closures, prevention of indoor mixing, and curbing overnight stays. Travel restrictions allowed only essential travel, discouraged travel in general, and prohibited residents from leaving the area. Non-essential closures included places of worship, non-essential retail stores, gyms, public buildings, care services, art venues, and tourist attractions.

The rule of six prevented individuals from meeting in groups of more than six people indoors and outdoors. The 10 pm closure required hospitality venues to close and ensure all customers left by 10 pm. Working from home was encouraged for individuals where possible.

The three-tier system was established on October 14, with each tier building on the previous one, Tier 1 being the least strict and Tier 3 the most stringent. Tier 1 (medium risk) generally corresponded to the "rule of six," "working from home," and the 10 pm rule, with an additional requirement to close businesses hosting music and dance events at night. In Tier 2, indoor gatherings between households were prohibited, travel was restricted, and more venues were closed. Tier 3 prevented private outdoor gatherings with non-household members, restricted restaurants and bars to table service only, and allowed alcohol to be served only with substantial meals.

The second national restrictions, while not as strict as the first since schools remained open, included closures of pubs, restaurants, gyms, non-essential shops, and stay-at-home orders.

Study Design

This study is a longitudinal natural experiment. For each participant, there is one observation before the implementation of restrictions and one observation after. The observation period is up to 16 days following the restriction start date. This allows participants to serve as their own controls, thereby reducing the impact of long-term temporal trends as well as inter-individual variability. The types of reported contacts were categorized into home, workplace, school, and other settings.

The authors evaluated the impact of (i) regional restrictions, (ii) three national restrictions: (1) the rule of six, (2) the 10 pm closures, and (3) working from home, (iii) entry into Tiers 1, 2, and 3, and (iv) the transition from Tiers 1, 2, and 3 into national lockdown by comparing the number of contacts before and after the implementation of restrictions. To assess the effects of various restrictions, the authors focused on changes in setting-specific contacts. For instance, regional restrictions and the tier system primarily targeted leisure contacts, while the rule of six did not apply to workplaces or schools. Therefore, for these two restrictions, the authors analyzed changes in contacts excluding work and school. The 10 pm closure rule required restaurants, pubs, and bars to close early, so it was not expected to directly impact contacts at home, work, or school. Thus, contacts in these settings were excluded as a result of this restriction, with the remaining contacts referred to as "other contacts." To evaluate the effect of working from home advice, the focus was on work-related contacts among employed respondents. During the second national lockdown, since schools remained open, school contacts were excluded when assessing the impact on overall contacts.

Statistical Analysis

All analyses were performed using R version 4.0.0, and the code and data are available on GitHub (see the "Data and Materials Availability" section). Descriptive and graphical summaries of participant characteristics, including age, gender, employment, and socioeconomic status, were created for each restriction, along with changes in the average number of contacts and the spatial and temporal changes of the restrictions. The uncertainty of the average contacts was calculated using cluster bootstrapping per person, rather than at the observation level, to preserve the correlation structure of the data.

The authors compared contacts before and during the restrictions by calculating the mean, median, and interquartile range (IQR). Changes in the number of contacts were categorized as an increase, no change, or a decrease. They evaluated uncertainty by calculating the mean of the paired differences in contacts before and during the restrictions and constructing 95% confidence intervals (95% CI) from 10,000 bootstrap samples of the paired differences.

For each restriction, the authors performed paired permutation tests with 50,000 permutations per test. Permutation tests were chosen because they are robust to the assumptions about the distribution of the underlying data. To preserve the study structure, the paired differences were calculated by subtracting the reported number of contacts during the restriction from the reported number before the restriction, then randomly altering the sign of each pair. In practice, this meant generating a vector of random values by multiplying the contact changes by a vector of -1 and 1 of the same length as the number of participants.

For each permutation and restriction, the authors calculated two test statistics: (1) the proportion of individuals with reduced contacts after the restriction was implemented, and (2) the mean change in contacts before and after the restriction. The proportion of reductions is robust to large values, treating differences of -1 and -1000 similarly in a skewed distribution. This measures the relative effect of the restrictions but does not estimate effect size. The mean difference estimates the absolute effect but is affected by skewed data. To mitigate the impact of skewness, they limited the total number of contacts to 200 per person per day for comparisons of means.

The authors conducted further evaluations of the rule of six and the 10 pm rule by age group, as these restrictions may have a greater impact on younger individuals who are more mobile, asymptomatic if infected, and less shielded. These analyses were stratified by age groups 5–17, 18–39, 40–59, and 60+.

Results

Participant Characteristics

The analysis of the rule of six included 3,884 participants, the 10 pm closures included 3,887 participants, working from home included 1,408 participants, and regional restrictions included 572 participants (Table 1). There were 2,415 entries for Tier 1, 1,654 for Tier 2, and 368 for Tier 3. Additionally, 2,095 individuals transitioned from Tier 1 to the national lockdown, 1,445 from Tier 2 to the national lockdown, and 323 from Tier 3 to the national lockdown.

The age distribution of samples for the rule of six, 10 pm closures, regional restrictions, tier entry, and national lockdown were similar, with over 30% of each sample aged 50–69 years across all nine analyses. The working-from-home category, by definition, included only participants aged 18 and over, with nearly 70% of participants aged 30–59. The gender split was approximately 50% for all restrictions. Apart from the working-from-home analysis, around 40% of participants were employed under each restriction, reflecting the broad age range of the sample, including children and older adults.

Socioeconomic status was consistent across all restriction analyses, with the majority falling into the C1-lower middle class category, and fewer participants in the A-upper middle class and E-lowest subsistence categories (Table 1).

All Adult Contacts and Restrictions

From March to June, nationwide restrictions were applied across the UK (Figure 1). During the summer, restrictions were eased, and regional restrictions on travel, non-essential closures, indoor mixing, and overnight stays were implemented (Figure 1c). These regional restrictions were primarily applied in the north of England (Figure 1a). The easing of restrictions in August coincided with an increase in the average number of contacts among adults. From September to November, as restrictions became stricter and more widespread, the number of contacts gradually decreased (Figure 1b). Following the second national lockdown, the average number of daily reported contacts returned to levels similar to those in July. Figure 1 shows the average number of contacts over time for adults only, as data on children were not collected during the study period, and the second national lockdown did not include school closures.

Distribution of Setting-Specific Contacts

Setting-specific contacts were positively skewed for all restrictions (Figure 2a, Table 2). The rule of six and regional restrictions showed similar distributions, with the modal response being one contact before the restriction. In contrast, working from home and the 10 pm rule had the majority of individuals reporting no contacts. The distribution for tier entry and exit from tiers to lockdown was nearly identical, with a median of 2 and an IQR of 1–3, except for tier 1 exits, which had an IQR of 1–4.

Overall, the magnitude of changes in the number of contacts was small, and the reported number of contacts did not change significantly after each restriction was introduced (Table 2). To illustrate the patterns in the data, axes were limited, and zero values were removed in Figures 2b, 3a, and 3b. These graphs are replicated in Additional File 1: Figures S1A and S1B and Additional File 2: Figures S2A and S2B, without removing zero values or limiting axes for comparison.

National Restrictions

Rule of Six

The authors compared non-work and non-school contacts for 3,884 individuals before and after the implementation of the rule of six. There was strong evidence suggesting that the number of contacts (excluding work and school) decreased for more individuals than expected by chance. Specifically, 1,314 individuals (33.8%) reduced their number of contacts, whereas 997 individuals (25.7%) increased their contacts (p<0.001). However, the majority of participants, 1,573 individuals (40.5%), recorded the same number of contacts, with a median of 2 contacts (IQR 1–3) before and after the restriction was introduced. There was a slight indication that the average number of non-work and non-school contacts per day decreased marginally (-0.25; 95% CI -0.5 to -0.01) (p = 0.05) (Table 2). Age group analysis (Table 3) suggested that the rule of six had the greatest impact on the contact patterns of young adults (18–39 years), reducing non-work and non-education contacts by an average of -0.59 (95% CI -1.09 to -0.04).

10pm Closures

For 3,887 participants, the number of "other" contacts (excluding home, work, and school) was compared before and after the implementation of the 10 PM closure rule. There was strong evidence suggesting that more individuals decreased their number of contacts than expected by chance. Specifically, 990 individuals (25.5%) reduced their number of contacts, compared to 843 individuals (21.7%) who increased their contacts (p<0.001). However, more than half of the participants (52.8%, or 2,054 individuals) recorded the same number of contacts, with a very low median of 0 (IQR 0–1) before and after the 10 PM rule was implemented. The data was consistent with no absolute effect (p = 0.915), and the estimated average change in "other" contacts was 0.01 (95% CI -0.23 to 0.23) (Table 2). Subgroup analyses suggested inconsistent patterns by age group (Table 3), which is expected when there is no evidence of a change in contacts for this indicator overall.

Work from Home

More than two-thirds of the participants, 933 individuals (66.3%), had the same number of work contacts before and after being encouraged to work from home. Nonetheless, the data strongly suggested that more individuals reduced their work contacts after the restriction was implemented than would be expected by chance (p = 0.001). The differences in the number of contacts were highly skewed, with 40 participants reporting a difference of 50 or more contacts, while the 25th and 50th percentiles of the differences were zero (Figure 2, Table 2). The data were consistent with a decrease in the average number of work contacts (p = 0.05). There was considerable uncertainty around the point estimate, but it was significantly different from zero (mean change in work contacts per day -0.99, 95% CI -1.94 to -0.03, Table 2).

Table 1: Characteristics of CoMix Survey Participants by Restriction

Restrictions   Entry into Exit from tier to lockdown
  Rule of Six 10 pm closure Work from home Local lockdown Tier 1 Tier 2 Tier 3 Tier 1 Tier 2 Tier 3
  N (col %) N (col %) N (col %) N (col %) N (col %) N (col %) N (col %) N (col %) N (col %) N (col %)
Total 3884 3887 1408 572 2415 1654 368 2095 1455 323
Age groups
0–4 129 (3.3%) 142 (3.7%) 0 22 (3.9%) 96 (4.0%) 65 (4.0%) 12 (3.3%) 53 (2.5%) 49 (3.4%) 13 (4.0%)
5–11 214 (5.5%) 275 (7.1%) 0 54 (9.5%) 151 (6.3%) 117 (7.1%) 36 (9.9%) 120 (5.8%) 85 (5.9%) 31 (9.7%)
12–17 261 (6.8%) 291 (7.5%) 0 59 (10.4%) 202 (8.4%) 121 (7.4%) 48 (13.2%) 149 (7.1%) 101 (7.0%) 41 (12.8%)
18–29 364 (9.4%) 384 (9.9%) 222 (15.8%) 61 (10.7%) 197 (8.2%) 161 (9.8%) 23 (6.3%) 158 (7.6%) 124 (8.6%) 20 (6.2%)
30–39 462 (12.0%) 432 (11.2%) 308 (21.9%) 65 (11.4%) 240 (10.0%) 191 (11.6%) 42 (11.6%) 194 (9.3%) 153 (10.6%) 34 (10.6%)
40–49 495 (12.8%) 531 (13.7%) 363 (25.8%) 89 (15.6%) 326 (13.6%) 219 (13.3%) 47 (12.9%) 288 (13.8%) 200 (13.8%) 42 (13.1%)
50–59 708 (18.3%) 613 (15.8%) 322 (22.9%) 90 (15.8%) 402 (16.7%) 277 (16.9%) 63 (17.4%) 355 (17.0%) 262 (18.1%) 49 (15.3%)
60–69 723 (18.7%) 751 (19.4%) 174 (12.4%) 88 (15.5%) 449 (18.7%) 309 (18.8%) 50 (13.8%) 428 (20.5%) 291 (20.1%) 48 (15.0%)
70+ 506 (13.1%) 449 (11.6%) 19 (1.3%) 41 (7.2%) 341 (14.2%) 181 (11.0%) 42 (11.6%) 341 (16.3%) 182 (12.6%) 43 (13.4%)
Missing 22 19 3 11 13 5 9 8 2
Gender
Female 2013 (52.0%) 2004 (51.6%) 718 (51.1%) 277 (48.7%) 1252 (52.0%) 890 (54.0%) 179 (48.8%) 1072 (51.3%) 758 (52.2%) 156 (48.4%)
Male 1861 (48.0%) 1877 (48.4%) 688 (48.9%) 292 (51.3%) 1156 (48.0%) 759 (46.0%) 188 (51.2%) 1018 (48.7%) 694 (47.8%) 166 (51.6%)
Missing 10 6 2 3 7 5 1 5 3 1
Employed
Yes 1487 (38.3%) 1433 (36.9%) 1408 (100.0%) 220 (38.5%) 882 (36.5%) 608 (36.8%) 134 (36.4%) 761 (36.3%) 521 (35.8%) 112 (34.7%)
No 2397 (61.7%) 2454 (63.1%) 0 352 (61.5%) 1533 (63.5%) 1046 (63.2%) 234 (63.6%) 1334 (63.7%) 934 (64.2%) 211 (65.3%)
Socio-economic status
A - Upper middle class 200 (5.1%) 214 (5.5%) 72 (5.1%) 24 (4.2%) 143 (5.9%) 89 (5.4%) 14 (3.8%) 119 (5.7%) 79 (5.4%) 9 (2.8%)
B - Middle class 1061 (27.3%) 1033 (26.6%) 394 (28.0%) 161 (28.1%) 622 (25.8%) 418 (25.3%) 90 (24.5%) 554 (26.4%) 375 (25.8%) 85 (26.3%)
C1 - Lower middle class 1285 (33.1%) 1332 (34.3%) 536 (38.1%) 184 (32.2%) 812 (33.6%) 596 (36.0%) 130 (35.3%) 731 (34.9%) 529 (36.4%) 115 (35.6%)
C2 - Skilled working class 534 (13.7%) 529 (13.6%) 197 (14.0%) 85 (14.9%) 343 (14.2%) 227 (13.7%) 50 (13.6%) 278 (13.3%) 206 (14.2%) 40 (12.4%)
D - Working class 571 (14.7%) 556 (14.3%) 203 (14.4%) 82 (14.3%) 377 (15.6%) 221 (13.4%) 55 (14.9%) 304 (14.5%) 176 (12.1%) 47 (14.6%)
E - Lower level of subsistence 233 (6.0%) 223 (5.7%) 6 (0.4%) 36 (6.3%) 118 (4.9%) 103 (6.2%) 29 (7.9%) 109 (5.2%) 90 (6.2%) 27 (8.4%)

Figure 1

図1

Local Restrictions

There was strong evidence suggesting that participants' non-work and non-school contacts decreased more than expected by chance when following local restrictions (p<0.001). Among the 572 participants, 197 (34.4%) reported a decrease in the number of contacts, 123 (21.5%) reported an increase, and 252 (44.1%) reported the same number of contacts. On average, participants reported 0.69 fewer non-work and non-school contacts (95% CI 0.17–1.25; p = 0.004) compared to before the restrictions, corresponding to a relative decrease of 21% (95% CI 5–40%).

Tier Entry

The authors compared non-work and non-school contacts for 2,415 individuals entering Tier 1, 1,654 individuals entering Tier 2, and 368 individuals entering Tier 3 (Table 2). Although the proportion of individuals entering Tier 1 who reduced their contacts was higher compared to those entering other tiers, there was strong evidence suggesting that more individuals reduced their contacts after entering each tier than would be expected by chance.

In fact, the data indicated that the changes in reported contacts were close to zero after entering Tier 1 or Tier 2, with no significant decrease in the average number of contacts. After entering Tier 3, however, there was a suggested decrease in daily contacts from 3.09 to 2.32 (p = 0.067), though this observation was based on only 368 individuals. The median number of daily contacts remained fixed at 2 (IQR 1–3) before and after entering any of the tiers.

Figure 2

図2

Exit from Tiers to National Lockdown

The authors compared non-school contacts before and during the transition to national lockdown for 2,095 individuals moving from Tier 1, 1,455 individuals from Tier 2, and 323 individuals from Tier 3. The data were consistent with a greater-than-expected reduction in contacts by chance.

The largest difference was observed among those transitioning from Tier 1 to lockdown, with 750 individuals (35.8%) reducing their contacts compared to 390 individuals (18.6%) who increased their contacts (Table 2). This was strongly supported by evidence that the average number of daily contacts decreased by 1.40 (95% CI 0.85–2.03) after moving from Tier 1 to lockdown (p < 0.001). The impact of transitioning from Tier 2 or Tier 3 to lockdown was less pronounced, with minimal observations for the estimated effects from Tier 3 to lockdown.

Table 2: Summary of Permutation Tests on Decrease in Number of Contacts and Mean Differences Pre- and Post-Restrictions

Comparison with proportion decreased        
Restriction Contacts N Adults Children Decreased Same Increased P value
Local exclude work and school 572 434 138 197 (34.4%) 252 (44.1%) 123 (21.5%) < 0.001
ROS exclude work and school 3884 3258 626 1314 (33.8%) 1573 (40.5%) 997 (25.7%) < 0.001
10 pm other 3887 3160 727 990 (25.5%) 2054 (52.8%) 843 (21.7%) < 0.001
WFH work 1408 1408 0 288 (20.5%) 933 (66.3%) 187 (13.3%) < 0.001
T1 entry exclude work and school 2415 1955 460 752 (31.1%) 993 (41.1%) 670 (27.7%) 0.017
T2 entry exclude work and school 1654 1338 316 468 (28.3%) 823 (49.8%) 363 (21.9%) < 0.001
T3 entry exclude work and school 368 267 101 103 (28.0%) 188 (51.1%) 77 (20.9%) 0.034
T1 exit to LD exclude school 2095 1764 331 750 (35.8%) 955 (45.6%) 390 (18.6%) < 0.001
T2 exit to LD exclude school 1455 1212 243 428 (29.4%) 732 (50.3%) 295 (20.3%) < 0.001
T3 exit to LD exclude school 323 236 87 85 (26.3%) 173 (53.6%) 65 (20.1%) 0.062
Comparison in mean difference Median (IQR)   Mean  
Restriction Contacts Before After   Before After Difference (95% CI) P value
Local Exclude work and school 2 (1 to 4) 2 (1 to 3)   3.18 2.49 − 0.69 (− 1.25 to − 0.17) 0.004
ROS Exclude work and school 2 (1 to 3) 2 (1 to 3)   2.9 2.66 − 0.25 (− 0.5 to − 0.01) 0.045
10 pm Other 0 (0 to 1) 0 (0 to 1)   1.37 1.38 0.01 (− 0.23 to 0.23) 0.915
WFH Work 0 (0 to 1) 0 (0 to 0)   4.62 3.62 − 0.99 (− 1.94 to − 0.03) 0.042
T1 entry Exclude work and school 2 (1 to 3) 2 (1 to 3)   2.79 2.66 − 0.13 (− 0.39 to 0.11) 0.305
T2 entry Exclude work and school 2 (1 to 3) 2 (1 to 3)   2.42 2.56 0.14 (− 0.17 to 0.55) 0.473
T3 entry Exclude work and school 2 (1 to 3) 2 (1 to 3)   3.09 2.32 − 0.77 (− 1.97 to − 0.03) 0.067
T1 exit to LD Exclude school 2 (1 to 4) 2 (1 to 3)   4.21 2.81 − 1.40 (− 2.03 to − 0.85) < 0.001
T2 exit to LD Exclude school 2 (1 to 3) 1 (1 to 3)   3.4 2.97 − 0.42 (− 1.13 to 0.33) 0.247
T3 exit to LD Exclude school 2 (1 to 3) 2 (1 to 3)   3.08 3.54 0.46 (− 0.28 to 1.41) 0.343

Two-sided p-values were calculated by counting the number of permutations where the test statistic was greater than the observed test statistic and dividing by the total number of permutations.

Figure 3

図3

Table 3: Summary of Permutation Tests on Decrease in Number of Contacts and Mean Differences Before and After 10pm, and Rule of Six by Age

Comparison with proportion decreased        
Restriction Contacts N Adults Children Decreased Same Increased P value
ROS
5–17 Exclude work and school 464 0 464 167 (36%) 179 (38.6%) 118 (25.4%) 0.0022
18–39 Exclude work and school 816 816 0 291 (35.7%) 343 (42%) 182 (22.3%) < 0.001
40–59 Exclude work and school 1193 1193 0 396 (33.2%) 488 (40.9%) 309 (25.9%) 0.0005
60+ Exclude work and school 1225 1225 0 403 (32.9%) 475 (38.8%) 347 (28.3%) 0.0219
10 pm
5–17 Exclude work and school 550 0 550 167 (30.4%) 239 (43.5%) 144 (26.2%) 0.1062
18–39 Exclude work and school 813 813 0 243 (29.9%) 376 (46.2%) 194 (23.9%) 0.0103
40–59 Exclude work and school 1134 1134 0 350 (30.9%) 507 (44.7%) 277 (24.4%) 0.002
60+ Exclude work and school 1196 1196 0 395 (33%) 453 (37.9%) 348 (29.1%) 0.0452
Comparison in mean difference Median (IQR)   Mean  
Restriction Contacts Before After   Before After Difference (95% CI) P value
ROS
5–17 Exclude work and school 3 (2 to 4) 3 (2 to 4)   3.93 4.09 0.17 (− 0.35 to 0.76) 0.5668
18–39 Exclude work and school 2 (1 to 3) 2 (1 to 3)   2.89 2.3 − 0.59 (− 1.09 to − 0.04) 0.0183
40–59 Exclude work and school 2 (1 to 3) 2 (1 to 3)   2.61 2.31 − 0.3 (− 0.67 to 0.03) 0.1055
60+ Exclude work and school 2 (1 to 3) 2 (1 to 3)   2.56 2.6 0.04 (− 0.39 to 0.52) 0.8797
10 pm
5–17 Exclude work and school 3 (2 to 4) 3 (2 to 4)   3.85 4.84 0.98 (0.28 to 1.81) 0.0115
18–39 Exclude work and school 2 (1 to 3) 2 (1 to 3)   2.7 2.57 − 0.13 (− 0.72 to 0.5) 0.649
40–59 Exclude work and school 2 (1 to 3) 1 (1 to 3)   2.4 2.54 0.14 (− 0.37 to 0.65) 0.6183
60+ Exclude work and school 2 (1 to 3) 1 (1 to 3)   2.54 2.09 − 0.45 (− 0.88 to − 0.12) 0.0039

Discussion

Like many other countries, the UK transitioned from a national lockdown approach to more localized interventions with less extensive national measures, and then back to a national lockdown in autumn 2020. We found that the impact of these measures varied: regional measures, which were highly diverse across different locations, and the rule of six likely led to small reductions in the number of contacts; the directive to work from home where possible led to larger reductions in contacts, while there was little to no evidence that the 10 pm closures of bars and restaurants had a significant effect. Similarly, Tiers 1 and 2 had little impact on average contacts, whereas Tier 3 (the most stringent) reduced the average daily reported contacts. The subsequent national lockdown appeared to reduce contacts among individuals previously in Tier 1 (the least restrictive), but the data suggested it was harder to further reduce contacts among those already under stringent restrictions (Tiers 2 and 3).

In absolute terms, the changes in average contacts were relatively small. However, the relatively small size of these absolute impacts in our study does not indicate that the restrictions were ineffective; rather, it suggests that the restrictions were applied when individuals had already reduced their contacts. For example, the move to work from home only reduced the average number of work contacts by about one per day, but this was likely difficult to achieve as working from home was already relatively widespread. Reflecting on these changes, the initial national lockdown in March led to an estimated reduction in the average number of daily contacts from 10.8 to 2.8. This 74% reduction decreased the effective reproduction number (R0) of COVID-19 from around 2.6 to around 0.6. The relatively small reductions in work contacts under the various restrictions discussed here would have much smaller impacts on R0.

Determining the epidemiological impact of restrictions has proven difficult. This is due to the lag between implementing measures and their effect on reported cases, hospitalizations, and deaths. Additionally, reported case numbers may be upwardly biased in regions with additional efforts for case detection and testing. It is also challenging to estimate the number of cases that might have occurred without the restrictions. For these reasons, evidence on the impact of local and national regulations is weak. This study takes a different approach: contact numbers are expected to change immediately following the implementation of restrictions and are less likely to be affected by changes in case detection. Furthermore, the longitudinal panel nature of the data allows individuals to act as their own temporal controls, making it easier to detect relatively small changes in contact patterns.

This study has several limitations. We had to group several types of measures used under regional restrictions, so the observed effects are of a combination of interventions. Individuals might not accurately report their contacts due to recall bias or social desirability bias. Another limitation is that the restrictions were not randomly assigned, so observed effects could be due to other confounding factors. However, confounders remain constant for individuals, and while they may affect generalizability, repeated measurements on the same individuals reduce inter-individual variation. Contact data being zero-bound and skewed means using averages could be a less relevant summary measure. Thus, we also conducted permutation tests focusing on the sign rather than the magnitude of differences. Additionally, we did not differentiate the length of time spent on different contacts. Finally, as the number of contacts decreases, it becomes harder for individuals to reduce social interactions further. Thus, changes in contacts are small, and very large sample sizes are required for precise quantification.

Despite these constraints, we aimed to provide insights into whether various restrictions in response to COVID-19 were effective and, if so, to what extent. We focused on one epidemiologically relevant measure, but the impacts of different restrictions likely have broader social implications that need to be considered for policy changes.

Future research could assess whether restrictions reduce the time spent with individuals and the effectiveness of the 10 pm rule. Examining the effects of restrictions on different age groups and the potential compliance with regional and national restrictions will help determine whether the lack of effect is due to the inefficacy of the restrictions or sampling bias.

Conclusion

Behavior monitoring has demonstrated that the impact of national and regional regulations on the transmission of COVID-19 can be evaluated quickly. While many of these restrictions appear to have led to changes in behavior, the magnitude of these changes seems to be small.

Abbreviations

CI: Confidence Interval

IQR: Interquartile Range

UK: United Kingdom

References

  1. Prime Minister ’s Office, Street 10 Downing. Prime Minister ’s statement on coronavirus (COVID-19): 23 March 2020. GOV.UK. 2020.
    https://www.gov.uk/ government/speeches/pm-address-to-the-nation-on-coronavirus-23-march-2 020. Accessed 11 Oct 2020.
  2. Jarvis CI, Van Zandvoort K, Gimma A, Prem K, CMMID COVID-19 working group, Klepac P, et al. Quantifying the impact of physical
    distance measures on the transmission of COVID-19 in the UK. BMC Med. 2020;18:124.
  3. Davies NG, Kucharski AJ, Eggo RM, Gimma A, Edmunds WJ. Centre for the Mathematical Modelling of Infectious Diseases COVID-19
    working group. Effects of non-pharmaceutical interventions on COVID-19 cases, deaths, and demand for hospital services in the UK: a
    modelling study. Lancet Public Health. 2020;5:e375 –85.
  4. Flaxman S, Mishra S, Gandy A, Unwin HJT, Mellan TA, Coupland H, et al. Estimating the effects of non-pharmaceutical interventions on
    COVID-19 in Europe. Nature. 2020;584:2570 –61.
  5. Prime Minister ’s Office, Street 10 Downing. PM announces easing of lockdown restrictions: 23 June 2020. GOV.UK. 2020.
    https://www.gov.uk/ government/news/pm-announces-easing-of-lockdown-restrictions-23-june-2 020. Accessed 11 Oct 2020.
  6. Department of Health and Social Care. Plans for managing the coronavirus (COVID-19) outbreak in Leicester. GOV.UK. 2020.
    https://www.gov.uk/ government/speeches/local-action-to-tackle-coronavirus. Accessed 11 Oct 2020.
  7. Department of Health and Social Care. Local restrictions: areas with an outbreak of coronavirus (COVID-19). 2020.
    https://www.gov.uk/government/collections/localrestrictions-areas-with-an-outbreak-of-coronavirus-covid-19. Accessed 11 Oct 2020. 8. Home Office. Rule of six comes into effect to tackle coronavirus. GOV.UK. 2020. https://www.gov.uk/government/news/rule-of-six-comesinto-effectto-tackle-coronavirus. Accessed 11 Oct 2020.
  8. Cabinet Office. Coronavirus (COVID-19): What has changed – 22 September. GOV.UK. 2020.
    https://www.gov.uk/government/news/coronavirus-covid-19- what-has-changed-22-september. Accessed 11 Oct 2020.
  9. Prime Minister ’s Office, Street 10 Downing. Prime Minister announces new local COVID Alert Levels. GOV.UK. 2020.
    https://www.gov.uk/government/news/primeminister-announces-new-local-covid-alert-levels. Accessed 2 Nov 2020.
  10. Prime Minister ’s Office, Street 10 Downing. Prime Minister ’s statement on coronavirus (COVID-19): 31 October 2020. GOV.UK. 2020.
    https://www.gov. uk/government/speeches/prime-ministers-statement-on-coronavirus-covid-1 9-31-october-2020. Accessed 15 Dec 2020.
  11. R Core Team. R: A Language and Environment for Statistical Computing. 2017. https://www.R-project.org .
  12. Dowle M, Srinivasan A. data.table: Extension of data.frame; 2020.
  13. Wickham H. ggplot2: Elegant Graphics for Data Analysis. 2016. https:// ggplot2.tidyverse.org .
  14. Efron B. Bootstrap methods: another look at the jackknife. Ann Stat. 1979;7: 1–26. https://doi.org/10.1214/aos/1176344552.
  15. Good PI. Permutation, parametric, and bootstrap tests of hypotheses: Springer Science & Business Media; 2006.
  16. Coronavirus and homeworking in the UK: April 2020 - Office for National Statistics.
    https://www.ons.gov.uk/releases/homeworkinginapril2020 . Accessed 16 Dec 2020.