Reporting of contacts.

Contacts that occurred on the day prior to the survey were reported in 2 ways: individual contacts and group contacts. First, participants were asked to list each contact and their characteristics separately (“individual contacts”). Second, we asked participants to report the total number of contacts they had at work, school, or other settings for the age groups 0 to 17, 18 to 59, and 60+, both overall and for physical contacts only (“group contacts”). We define direct contact as anyone who met the participant in person with whom at least a few words were exchanged or physical contact was made. Questions on group contacts were included at the end of the survey, and they were added to surveys from the May 14, 2020 to accommodate individuals—such as those working in patient- or public-facing roles—who could not record details of all individual contacts that they made. Since August 2020, participants were also asked to describe the average time spent with group contacts and any transmission-related precautions they implemented, including distancing, wearing face coverings, and performing hand hygiene.

Demographic information.

The survey captures information about participant demographics, employment status, and whether participants attended work (or school/university for participants <18 years old and self-reported students) in the previous week and on the day they recorded contacts. We group participants by age into the groups 0 to 4, 5 to 11, 12 to 17, 18 to 29, 30 to 39, 40 to 49, 50 to 59, 60 to 69, and over 70. We group areas into 6 regions: East of England, Greater London, Midlands, North East and Yorkshire, South East, and South West. Participants were asked to report how they describe their gender, with the options of “Female,” “Male,” “In another way,” or “Prefer not to answer.” For analyses, due to the small number of participants answering the gender question with “In another way” or “Prefer not to answer,” we grouped these with missing values into the category “Other.” A socioeconomic category was assigned to participants based on their reported occupation (see S1 and S3 Text for more information).

Risk perception, status, and mitigation.

Participants were asked questions about their performance of risk mitigating activities and asked to respond to statements regarding their perception of risk. Participants were asked to respond to the following statements: (i) “I am likely to catch coronavirus”; (ii) “I am worried that I might spread coronavirus to someone who is vulnerable”; and (iii) “Coronavirus would be a serious illness for me” with the Likert scale of “Strongly agree,” “Tend to agree,” “Neutral,” “Tend to disagree,” and “Strongly disagree”. Participants self-reported whether or not they considered themselves to be high risk based on definitions given in the survey, which changed between survey versions as government advice changed (see questionnaires for details). Participants were also asked whether they wore a face covering and in which settings, whether they had washed their hands in the 3 hours prior to the survey, and whether they sanitised their hands in the 3 hours prior to the survey.

Descriptive.

We calculated summary statistics of the age, gender, socioeconomic status, household size, and National Health Service (NHS) region for participants for each analysis time period and survey panel. While parents answer as proxies for children in the study, we describe the designated child as the “participant” where applicable. We calculated the number and percentage of participants that completed 1, 2 to 3, 4 to 5, and 6 or more rounds of the survey by participant characteristics.

Mean contacts.

We calculated the mean number of contacts and associated confidence intervals (CIs) with 1,000 samples using clustered bootstrapping [15]. Each participant was sampled with replacement and then all observations for selected participants were included in a bootstrapped sample to account for dependency from repeated observations of the same participants. We calculated the mean number of contacts with a moving window over 2-week, overlapping intervals to increase the sample size per estimate and to include all participants from simultaneously running panels. While the initial panels were recruited to be representative of the UK population, we used poststratification weights of the mean by age group and gender (if available) to address bias introduced by differences between each sample and the UK population [16]. We report contacts by age groups for preschool-age children (<5 years old), school-age children (5 to 17 years old), adults (18 to 59 years old), and the elderly (60+ years old).

Post-stratified weights were assigned by the age groups 0 to 4, 5 to 11, 12 to 17, 18 to 29, in 10 year age bands from 30 to 69 and 70 years old and over. We used the World Population Prospects 2019 standard projections overall and by gender for the 2020 UK population [17]. Participants with missing ages were not included in this analysis. Estimates of the 2-week intervals are presented with the data points aligned to the central time point of each survey round, and, therefore, each data point shown is derived from information up to 1 week before and after the labelled date. We plotted hospitalised cases of COVID-19 in England alongside mean contact data by age and setting to illustrate the relationship between mean contacts and cases. We used hospitalisation data from the UK government online coronavirus dashboard and filtered for cases only reported in England [18], which we acquired using the covidregionaldata R package [19]. We repeated the same process to calculate mean contacts using the POLYMOD data for comparison [13], although we note that CoMix data was censored to 50 contacts per participant based on previous analyses, while the POLYMOD participants in the UK were limited to 30 contacts when completing the survey.

We calculated the mean number of contacts in various settings: home, work and school (all educational establishments, including childcare, nurseries, and universities and colleges), and “other” (mostly leisure and social contacts, including shopping). The mean number of contacts was influenced by a few participants who reported very high numbers of contacts (often in a work context) relative to the rest of the panel. The distribution of reported contacts are right skewed with high variance. The mean number of contacts shown here was calculated by censoring the maximum number of contacts recorded at 50 per individual per day to reduce the variance, meaning we counted any individual who reported more than 50 contacts as if they reported 50 contacts to reduce the weight of individuals reporting high numbers of contacts on the mean. We have found in previous analyses that censoring at 50 contacts most closely reflects changes in contacts relative to changes in reproduction number at time t (R_t) over time as estimated by the Real-time Assessment of Community Transmission (REACT) study, a large home testing study conducted in the UK with the aim of quantifying COVID-19 transmission and infections [20,21].

We report bootstrapped mean contacts using the method previously described by responses to questions about reported risk and risk perception and by employment and income categories. For Likert style questions, we group participant responses of “Tend to agree” and “Strongly agree” into one category of “Agree,” and we group the responses of “Tend to disagree” and “Strongly disagree” into one category of “Disagree.” Only adult participants are included in these analyses. For contacts by employment, we only include participants who recorded working on the day in which they were reporting contacts. Participants who declined to answer these questions were not included in the analysis.

Study periods.

We calculated relative differences in mean contacts between study periods using an individual-level generalised additive model (GAM) [22,23]. We assumed reported contacts followed a negative binomial distribution, modelled using a log link function, with a random effect for participants by age group (0 to 4, 5 to 17, 18 to 59, and 60 years and over) with poststratification weights for age and gender (when available) based on the UK population. We similarly calculate the relative difference in mean contacts between age groups within each study period using a similar GAM function, with only the study period and age group variables exchanged.

Face coverings.

We present the bootstrapped 95% CIs of the proportions of participants who reported wearing a face covering in any setting for all participants and separately for only those participants who reported contacts outside the household on the day of the survey. Participants who declined to answer these questions were not included in the analysis.

Contact matrices.

We constructed age-stratified contact matrices for 9 age groups (0 to 4, 5 to 11, 12 to 17, 18 to 29, 30 to 39, 40 to 49, 50 to 59, 60 to 69, and 70+ years old). For child participants and contacts, we did not record exact ages and therefore sampled from the reported age group with a weighting consistent with the age distribution of contacts for the participants’ own age group, according to the POLYMOD survey methods [13]. We fitted a negative binomial model censored to 50 per matrix cell, due to dispersion of the reported number of contacts, to calculate mean contacts between each participant and contact age groups. To find the population normalised reciprocal contact matrix, we first multiplied the columns of the matrix by the mean normalised proportion of the UK population in each age group [13,24]. Then we took the cross-diagonal mean of each element of the contact matrix. Finally, we divided the resulting symmetrical matrix by the population mean normalised proportion of the UK population in each age group.

We used this approach to construct a contact matrix for each of the analysis periods by filtering the contact data by date. For each time period (Table 1), we calculated the dominant eigenvalue of the infectiousness and susceptibility corrected contact matrix (C_SI), calculated from the measured contact matrix C_t, and assumed age-dependent relative susceptibility and infectiousness vectors s and i:

We calculated the relative difference in the basic reproduction number under controlled conditions to Lockdown 1 by taking the ratio of the dominant eigenvalue of the effective contact matrices from the period in question and the dominant eigenvalue from Lockdown 1 [7,10,25]. The value of R_c, the basic reproduction number under controlled conditions, is defined as the expected number of secondary cases resulting from an initial infection in a completely susceptible population adjusted to changing conditions, in this case the change in social contacts over time. This value does not account for immunity in the population and will therefore be higher than the actual reproduction number at a given time.

We applied 2 assumptions of age-dependent susceptibility and infectiousness. First, we assumed that all age groups are equally infectious and susceptible. Second, we applied a weight for relative susceptibility and infectiousness by age as estimated by Davies and colleagues [26] (S1–S4 Text).

Contacts by setting.

For adults, contacts made at home mostly reflected household size (S1 Fig) and were consistently below a mean of 2 contacts per day over the study period, with little change in reported contacts across each of the analysis time periods (Fig 1C). Work and other contacts followed a similar pattern to adults: staying low but steadily increasing towards the end of the Lockdown 1, increasing in August 2020, decreasing slightly, and then returning to levels similar to the Lockdown 1 during the Lockdown 2 in November, and then reducing again over Christmas and throughout Lockdown 3.

During the first lockdown, schools were closed to all except vulnerable children and the children of essential workers, and recorded children’s contact rates were very low (Fig 1B and 1C). From early June 2020 until the third week of July 2020 (when schools were closed for the summer vacation), there was a limited reopening of schools, but most parents reported that their children continued to be educated from home. Average recorded contact patterns among children remained very low during this period (Fig 1B and 1C). When schools reopened fully in September 2020, the number of contacts rapidly increased for both school-age (5 to 17) and preschool-age children (0 to 4), although the increase in contacts in the latter age group was smaller. Children’s contacts declined significantly during the “half-term” vacation at the end of October 2020 but remained high during the second national lockdown (November 2020) as schools remained open. Schools were closed for the Christmas period, remained closed during the third national lockdown, and reopened on March 8, 2021. However, preschools were the first educational setting to reopen during the relaxation of the first lockdown and were not closed during the third lockdown. The contact patterns of 0 to 4 year olds reflect this, with mean rates of contact for this age group being higher than other children during the periods when preschools were open but primary and secondary schools were closed.

Contacts by study period.

We compared the relative difference of mean contacts using a GAM with Lockdown 1 as the reference period, as this was the beginning of the survey and the period of most stringent lockdown measures. The dates for lockdown periods can be found in Table 1. Contacts remained at similar levels to Lockdown 1 (the reference period) through Lockdown 1 easing for all age groups until the Reduced restrictions study period (Fig 2, Table 4). The relative difference was highest for adults over the 12-month study period during the period of Reduced restrictions with the relative difference of 1.59 (95% CI 1.54 to 1.64) for adults ages 18 to 59 years and 1.51 (95% CI 1.45 to 1.57) for adults aged over 60 years. For children, the relative difference was highest while schools were open during the Schools open, Lockdown 2, Lockdown 2 easing, and Lockdown 3 with schools open study periods, ranging from 1.74 (95% CI 1.54 to 1.97) for ages 0 to 4 to 3.03 (95% CI 2.82 to 3.26) for ages 5 to 17.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Relative difference in mean contacts by study period and age group with 95% CIs.

Relative differences calculated using a GAM with Lockdown 1 as the reference period for each age group adjusted to the UK population by age and gender (when available) for the age groups 0 to 4, 5 to 17, 18 to 59, and over 60 years old. Note the facets have different scales on the y-axes. Table 1 provides corresponding dates for each study period. CI, confidence interval; GAM, generalised additive model.

https://doi.org/10.1371/journal.pmed.1003907.g002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Relative difference in mean contacts by study period with 95% CIs.

https://doi.org/10.1371/journal.pmed.1003907.t004

Additionally, we compared the relative difference of mean contacts using a GAM with age 18 to 39 years old as the reference age group to compare the mean number of contacts between age groups within each study period, using the age groups 0 to 4, 5 to 17, 40 to 59, and 60+ years as the comparison groups. We chose 18 to 39 years as it was the midrange age group, which could then be compared to children and to older adults. The greatest relative difference of the mean number of contacts compared to 18 to 39 year olds was during Lockdown 2 for ages 5 to 17 with a relative difference of 3.67 (95% CI 3.48 to 3.87), while older adults had a relative difference of 0.67 (95% CI 0.64 to 0.70) (S4 and S5 Figs). The relative difference in contacts was lower during periods of school closure than when schools were open.

Precautionary behaviours and risk perception.

The majority (around 50%) of participants answered “Neutral” to a statement indicating that they were likely to catch coronavirus, and this remained fairly consistent over the course of the study (S3 Fig), among all adult age groups. Survey participants who agreed with a statement that they were likely to catch coronavirus recorded higher mean contacts (Fig 3A), especially in August 2020 and during the period following the second lockdown. Mean contacts for those who disagreed or were neutral were very similar. Participants who agreed with a statement indicating that they were worried that they might spread coronavirus to others generally had a higher mean number of contacts between the first and second lockdowns than those who disagreed with the same statement (Fig 3B). For example, those age 18 to 59 years who felt they were likely to spread COVID-19 reaching the mean number of contacts 6.20 (95% CI 5.27 to 7.07) in mid-September 2020, while those who disagreed during that time period was 3.47 (95% CI 3.00 to 4.01). In the same time period (mid-September), the mean number of contacts for those who reported that they did not worry that they would spread COVID-19 to someone who is vulnerable was 3.50 (95% CI 2.97 to 4.06) compared to those who reported they were worried 4.96 (95% CI 4.51 to 5.43). During all 3 lockdowns, the mean contact CIs overlap for participants in all 3 categories (Agree, Neutral, and Disagree) responding to this question.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Mean contacts by risk perception or risk category by adult age groups of 18 to 59 and 60 or more years with 95% CI of bootstrapped mean weighted by age, gender, and weekday.

Participants answered a series of questions about their risk perception with Likert scale response options. Answers of “Strongly agree” and “Somewhat agree” were combined into a category of “Agree, as were answers of “Strongly disagree” and “Somewhat disagree” to “Disagree.” (A) Answers to the statement “I am likely to catch coronavirus.” (B) Answers to the statement “I am worried I might spread coronavirus to someone who is vulnerable.” (C) Answers to the statement “Coronavirus would be a serious illness for me.” (D) Participant reported they were an individual at high risk for complications as defined in the questionnaire. CI, confidence interval; LD, lockdown.

https://doi.org/10.1371/journal.pmed.1003907.g003

Survey participants aged 18 to 59 years who disagreed that coronavirus would be serious for them reported slightly higher contacts than those who agreed with the statement, while participants over 60 years of age who disagreed were few in number and reported a wide range of contact behaviours (Fig 3C). Participants who were not high risk generally reported more contacts on average than those who were high risk in both age groups, especially during periods outside of lockdown and towards the end of the third lockdown, with the differences being more pronounced in the over 60 age group (Fig 3D). The highest mean number of contacts for those who did not report that COVID-19 would be a serious illness for them was 5.65 (95% CI 5.12 to 6.20) in mid-August 2020 compared to 4.62 (95% CI 4.20 to 5.04) for those who did agree that COVID-19 would be a serious illness for them in the same time period.

In terms of protective behaviour, the reported use of facemasks at least once on the previous day was low (approximately 12% for 18 to 59 year olds and approximately 3% for 60+ years olds) at the end of March 2020 for participants who reported contacts outside of the household (Fig 4). The proportion who self-reported wearing masks increased gradually for both age groups through June 2020, and a sharp increase in mask use was reported in late July and early August 2020, shortly after mask wearing became mandatory for entering shops on July 24, 2020 [27]. From August 1 through March 26, 2021, mask wearing ranged between 73% and 86% for adults 18 to 59 and between 70% and 84% for adults over 60 among participants with contacts outside the home.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Proportion of adult participants who report wearing a mask by age category with 95% CI of bootstrapped proportion.

Proportions plotted for all participants and for participants who reported any nonhousehold contacts, with the start date of face covering mandates in some settings indicated on July 24, 2020. CI, confidence interval; LD, lockdown.

https://doi.org/10.1371/journal.pmed.1003907.g004

Employment and income.

Participants who were employed part time consistently reported more contacts on average than full-time or self-employed participants with a wider range of contacts, and full-time workers reported contact means similar or slightly higher than self-employed workers in between lockdowns (Fig 5A).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Mean contacts by employment and income status.

Mean contacts of participants who worked on the previous day and their workplace was open on the previous day weighted by age, gender, and weekday. (A) By employment type: full time, part time, or self-employed. (B) By annual income level: less than £20k, £20k to £44.9k, and over £45k. LD, lockdown.

https://doi.org/10.1371/journal.pmed.1003907.g005

Mean contacts for adult participants over 18 years of age grouped by annual income levels (less than £20k, £20k to £44.9k, and over £45k), were similar and follow consistent patterns of decreasing during lockdowns and increasing slightly between the first and second lockdown (Fig 5B).