The impact of changes in COVID‐19 lockdown restrictions on alcohol consumption and drinking occasion characteristics in Scotland and England in 2020: an interrupted time‐series analysis

Abstract Background and Aims Early evidence suggests that COVID‐19 lockdown restrictions affect alcohol consumption. However, existing studies lack data on how drinking practices changed as restrictions disrupted people’s work, family life and socializing routines. We examined changes in consumption and drinking occasion characteristics during three periods of changing restrictions in Scotland/England. Design Interrupted time‐series analysis of repeat cross‐sectional market research data (assessing step‐level changes). Setting Scotland/England, January 2009–December 2020. Participants Scotland: 41 507 adult drinkers; England: 253 148 adult drinkers. Measurements Three intervention points: March 2020 lockdown, July 2020 easing of restrictions and October 2020 re‐introduction of some restrictions. Primary outcome: mean units consumed per week (total/off‐trade/on‐trade; 1 unit = 8 g ethanol). Secondary outcomes: drinking > 14 units per week, heavy drinking, drinking days per week, solitary drinking, drinking with family/partners, drinking with friends/colleagues, own‐home drinking, drinking in someone else’s home and drinking start times. Findings In Scotland, March 2020’s lockdown was associated with a 2.32 [95% confidence interval (CI) = 0.61, 4.02] increase in off‐trade (i.e. shop‐bought) units per week, a −2.84 (95% CI = −3.63, −2.06) decrease in on‐trade (i.e. licensed venues) units per week, but no statistically significant change in total units per week. July 2020’s easing of restrictions was associated with a 1.33 (95% CI = 0.05, 2.62) increase in on‐trade units per week, but no statistically significant total/off‐trade consumption changes. October 2020’s re‐introduction of some restrictions was not associated with statistically significant consumption changes. Results for England were broadly similar. Lockdown restrictions were also associated with later drinking start times, fewer occasions in someone else’s home and with friends/colleagues, more own‐home drinking and (in Scotland only) more solitary drinking. Conclusions Reductions in on‐trade alcohol consumption following COVID‐19 lockdown restrictions in Scotland/England in 2020 were mainly offset by increased own‐home drinking. This largely persisted in periods of greater/lesser restrictions. The shift towards off‐trade drinking involved significant changes in the characteristics of drinking occasions.


INTRODUCTION
COVID-19 has led to many countries implementing 'lockdown' measures to reduce social contact, e.g. by closing work-places/businesses/schools and restricting people's movements and freedom to meet non-household members face-to-face. This has significantly impacted upon health outcomes and health-related behaviours [1][2][3][4].
With regard to alcohol use, lockdown measures can affect drinking practices in various ways, with hospitality closures and changes to people's work, family life and socializing routines all likely to alter consumption patterns. Consumption may increase due to stress arising from uncertainty, social isolation and/or loss of work during the pandemic [5] or due to less structured days for those in job retention schemes [6]. Conversely, consumption may decrease due to reduced disposable incomes, less socializing and hospitality closures affecting alcohol availability. Lockdowns may also act as a catalyst for some people to attempt health improvements by re-evaluating their relationship with alcohol [7].
In the United Kingdom, lockdown restrictions were in place for longer than in many other countries in 2020 due to its comparatively high COVID-19 case/mortality rates [8]. Restrictions were first introduced in March 2020 and have subsequently changed significantly over time (in response to fluctuating case numbers) and varied between the United Kingdom's four devolved nations. Key lockdown policy developments in Scotland/England in 2020 are detailed in Table 1. To summarize, the United Kingdom introduced a strict national lockdown in March 2020 to close on-trade premises (e.g. pubs, bars, clubs and restaurants) and prevent people from leaving their homes except for essential work/shopping or short periods of exercise. Alcohol remained available throughout the year to purchase 'off-trade', i.e. via supermarkets, off-licences or on-line deliveries. The UK's national and devolved governments began significantly easing lockdown restrictions in July 2020 to allow limited indoor gathering and re-opening of on-trade premises.
However, restrictions (including closures of some on-trade premises) were then gradually re-introduced throughout late September-October 2020, with new tiered systems of localized restrictions being introduced, meaning restrictions varied between local areas. England (but not Scotland) then had a second national lockdown in November/December 2020, after which the tiered system returned. In general, lockdown measures tended to be slightly stricter in Scotland than in England, apart from during England's second lockdown (Table 1).
Existing evidence regarding the impact of lockdown restrictions on drinking is mixed. Sales data suggest that despite higher off-trade sales, overall alcohol sales decreased by 6% in both Scotland and England/Wales during the March-July 2020 lockdown [11]. Descriptive analysis of market research data also suggests that overall consumption fell in England, although did not statistically significantly change in Scotland during the early stages of the pandemic [12], while household shopping panel data suggest that British households did not buy more alcohol than expected for the time of year during the March-July 2020 lockdown [13].
Other UK surveys have tended to suggest that during the initial lockdown up to a third of people were drinking more than before, with a similar proportion drinking less [14], but there has been concern over potential increases in high-risk/binge drinking [1,15]. This is backed up by more recent research highlighting a polarization in drinking in England during the pandemic, with some people drinking less than before but heavy drinkers consuming more, and evidence of increased alcohol-related harm [16]. Internationally, some research suggests that lockdowns are associated with decreased alcohol consumption [17] and some suggests increased hazardous alcohol use [18], but the majority of surveys highlight consumption increasing among some groups but decreasing among others [19][20][21]. However, many of these surveys have limitations, such as changing data collection methods during the pandemic and using weak measurements [22]. Existing studies have also been unable to investigate how drinking occasion characteristics may have changed. This may have public health implications, given that risks of alcohol-related harm varies between drinking contexts [23,24]. Finally, so far studies have tended to only include data on the early months of the pandemic, so do not provide insight into whether consumption changed further as restrictions were eased and/or re-introduced. This is important to understand, given ongoing speculation concerning whether people revert to pre-pandemic drinking once restrictions are relaxed or if there are 'new norms' [7,25].
The current study has the following objectives: 1. To assess the impact of introducing, easing and re-introducing lockdown restrictions on alcohol consumption in Scotland and England in 2020.

2.
To assess the impact of introducing, easing and re-introducing lockdown restrictions on drinking occasion characteristics (in terms of where people drank, who with, and start times of occasions) in Scotland and England in 2020.
T A B L E 1 Key dates/policy developments in the introduction, easing and re-introduction of COVID-19 lockdown restrictions in Scotland and England in 2020 This table has focused upon policy developments affecting all or large parts of Scotland and England, but there have also been local restrictions introduced for specific areas (e.g. Aberdeen/Leicester) prior to the tier system or the other localized restrictions outlined being introduced. Sources of information are the BBC News Website [9] and The Health Foundation's COVID-19 policy tracker [10].

Research design
The natural experimental conditions of lockdown policy, arising from three periods of changing restrictions over time, were exploited to analyse the impact of introducing, easing and re-introducing restrictions using interrupted time-series (ITS) analysis. Scotland and England were analysed separately due to differences in the nature of their restrictions (see Table 1).

Data
Kantar Alcovision data from January 2009-December 2020 was used.

Intervention points
Key dates in lockdown policy developments were used to specify three England as, although restrictions varied between countries, timings of policy developments were broadly similar (see Table 1).

Outcome measures
To provide an overall picture of consumption the primary outcome was mean units per week, which was split into total/off-trade/on-trade to offer insight into substitution between settings. This is the mean number of units that respondents reported consuming in their 7-day drinking diary. Alcovision records consumption in 'serves'.

Statistical analysis
First, descriptive analysis was conducted to assess how alcohol consumption and drinking occasion characteristics varied throughout 2020. This was performed by plotting each outcome over time in 2020 and, for comparison with recent years, comparing with the 2016-19 average for the same month.
The impact of changes in lockdown restrictions was then more formally evaluated via ITS analysis, using data from the full January 2009-December 2020 time-series. July 2017 was missing, which was handled using Kalman filtering [26]. The analytical process involved constructing a monthly time-series, whereby individual-level Alcovision data were aggregated to give monthly averages for each outcome. Seasonal autoregressive moving average (SARMA) modelling was used to estimate the effect of each change in lockdown restrictions on each outcome, adjusting for autocorrelation, seasonality and trend. Candidate SARMA models were selected by testing for autocorrelation and non-stationarity using auto-correlation function (ACF) and partial auto-correlation function (PACF). The most appropriate models were then selected by using Akaike information criteria (AIC) and Bayesian information criteria (BIC), and testing model assumptions. All analyses were undertaken on weighted data and conducted using Stata/MP version 16.1. This analytical protocol was not pre-registered. As such, results should be considered exploratory. Table 2 gives descriptive statistics on the mean values of each outcome pre-lockdown and during each period of changes in restrictions.

RESULTS
It also provides information on the number of individuals in the sample in each country-period.
The descriptive analysis results are provided in Figure 1 (for the primary outcome) and Supporting information, Appendix B (for secondary outcomes). The ITS analysis results are provided in Tables 3-6. Changes in average weekly consumption

Changes in other alcohol consumption measures
The results for the secondary consumption measures were largely consistent with the primary analysis. In the off-trade, the ITS analysis   Pre-lockdown covers the period January 2009-February 2020, initial lockdown covers the period March 2020-June 2020, restrictions eased covers the period July 2020-September 2020 and some restrictions re-introduced covers the period some off-trade occasions took place in locations other than respondent's own home or someone else's home (e.g. outdoors, in holiday accommodation, outdoors or an event).
However, these were not included in the analysis due to making up a very small proportion of off-trade occasions overall. includes all diary weeks from October 2020 (when the tier system was introduced, and also covering the period of England's second lockdown from November 2020) onwards. Results for the AR terms and seasonal AR terms included in each model are reported underneath the main results. L refers to number of lags, and these were selected following an iterative process involving autocorrelation function/ autocorrelation function (ACF/PACF) plots and model fit statistics.

Model assumptions and sensitivity analyses
T A B L E 4 Estimated impact of the introducing, easing and re-introducing COVID-19 lockdown restrictions on alcohol consumption measures in England (step changes) peoples' routines and the amount of stress/boredom they face [28].
This study suggests that when under lockdown restrictions drinkers substituted occasions in someone else's home with more occasions in their own home, while there were also fewer occasions with friends/ colleagues. Lockdown restrictions were also associated with spikes in solitary drinking occasions in Scotland, but not England. This may reflect Scotland's disproportionately high number of one-person households [29]. Finally, the results suggest that drinking occasions shifted to later in the day under lockdown restrictions, which goes against concerns raised early in the pandemic that home-working and job retention schemes would increase daytime drinking due to less structured days and a disconnection from employers. One explanation for this finding is that occasions starting earlier are traditionally associated with longer drinking occasions, often involving groups of friends and a mixture of on-trade/off-trade locations [30]. Lockdown restrictions prevent this type of occasion from taking place.
The key strength of this study is that, while many surveys changed their previously established methods during the pandemic [22], Alcovision continued its pre-COVID data collection methods unchanged. Furthermore, while existing studies have tended to only have data on the early stages of the pandemic and focus upon alcohol consumption only, this study included monthly data up to the end of 2020, and provides insight into drinking occasion characteristics.
However, there are some limitations to note. First, like many large-scale alcohol surveys [31] Alcovision relies upon quota sampling from an on-line panel rather than random sampling. This has known limitations relating to selection bias [32], although we used a 'raking' technique to increase representativeness. Secondly, ITS analysis works best with a substantial amount of pre-and post-intervention data [33]. As COVID-19 is still a recent phenomenon, our analysis is limited by its relatively small post-intervention sample size, although our large pre-intervention sample allows models to account for preexisting trends. Thirdly, the three intervention points used in the analysis were based on calendar months, so do not perfectly align with restriction changes occurring mid-month. This means that the modelling may underestimate the impact of changing restrictions. Fourthly, on-trade consumption did not fall to zero when premises were closed during lockdowns. This phenomenon has also been seen in Australian survey data [34]. It may occur due to people misreporting take-away alcohol as on-trade consumption, illegal on-trade consumption taking place or hotels legally selling alcohol to guests residing there for essential work [12].
To conclude, COVID-19 lockdown restrictions represent a substantial intervention, with important implications for drinking practices.
This study has highlighted that restrictions in Scotland and England Estimates are adjusted for autocorrelation, seasonality and trend. B = coefficient, 95% confidence intervals (CI) and P = P-value.
'Initial lockdown' includes all diary weeks from March 2020 onwards, 'Restrictions eased' includes all diary weeks from July 2020 (when on-trade premises re-opened and restrictions on visiting other households were relaxed) onwards and 'Some restrictions re-introduced' includes all diary weeks from October 2020 (when the tier system was introduced, and also covering the period of England's second lockdown from November 2020) onwards. Results for the AR terms and seasonal AR terms included in each model are reported underneath the main results. L refers to number of lags, and these were selected following an iterative process involving autocorrelation function/ autocorrelation function (ACF/PACF) plots and model fit statistics.
T A B L E 5 Estimated Impact of the introducing, easing and re-introducing COVID-19 lockdown restrictions on drinking occasion characteristics (who with, off-trade location and start time) in Scotland (step changes) Estimates are adjusted for autocorrelation, seasonality and trend. B = coefficient, 95% confidence intervals (CI) and P = p-value.
'Initial lockdown' includes all diary weeks from March 2020 onwards, 'Restrictions eased' includes all diary weeks from July 2020 (when on-trade premises re-opened and restrictions on visiting other households were relaxed) onwards and 'Some restrictions re-introduced' includes all diary weeks from October 2020 (when the tier system was introduced, and also covering the period of England's second lockdown from November 2020) onwards. Results for the AR terms and seasonal AR terms included in each model are reported underneath the main results. L refers to number of lags, and these were selected following an iterative process involving autocorrelation function/ autocorrelation function (ACF/PACF) plots and model fit statistics.
T A B L E 6 Estimated impact of the introducing, easing and re-introducing COVID-19 lockdown restrictions on drinking occasion characteristics (who with, off-trade location and start time) in England (step changes) Estimates are adjusted for autocorrelation, seasonality and trend. B = coefficient, 95% confidence intervals (CI) and P = p-value.
'Initial lockdown' includes all diary weeks from March 2020 onwards, 'Restrictions eased' includes all diary weeks from July 2020 (when on-trade premises re-opened and restrictions on visiting other households were relaxed) onwards and 'Some restrictions re-introduced' includes all diary weeks from October 2020 (when the tier system was introduced, and also covering the period of England's second lockdown from November 2020) onwards. Results for the AR terms and seasonal AR terms included in each model are reported underneath the main results. L refers to number of lags, and these were selected following an iterative process involving autocorrelation function/ autocorrelation function (ACF/PACF) plots and model fit statistics.
However, one concern is the observed increase in 'home drinking'.
While 'home drinking' is currently a relatively under-researched topic [7,35], its increase is likely to have contributed to the high levels of alcohol-related harm during the pandemic [16]. There is a need to monitor this further in the future to ascertain whether 'home drinking' habits picked up during 2020 become a 'new norm' within people's drinking behaviour.

DECLARATION OF INTERESTS
None.