Article Text

Original research
Differential impact of minimum unit pricing on alcohol consumption between Scottish men and women: controlled interrupted time series analysis
  1. Jürgen Rehm1,
  2. Amy O'Donnell2,
  3. Eileen F S Kaner2,
  4. Eva Jane LLopis3,4,
  5. Jakob Manthey5,
  6. Peter Anderson2
  1. 1 Social and Epidemiological Research, CAMH, Toronto, Ontario, Canada
  2. 2 Population Health Sciences Institute, Newcastle University, Newcastle upon Tyne, UK
  3. 3 Department of Health Promotion, Maastricht University, Maastricht, Limburg, The Netherlands
  4. 4 ESADE Business School, Ramon Llull University, Barcelona, Spain
  5. 5 Institute of Clinical Psychology and Psychotherapy, Technische Universität Dresden, Dresden, Sachsen, Germany
  1. Correspondence to Dr Peter Anderson; peteranderson.mail{at}gmail.com

Abstract

Objective To assess the immediate impact of the introduction of minimum unit pricing (MUP) in Scotland on alcohol consumption and whether the impact differed by sex, level of alcohol consumption, age, social grade and level of residential deprivation of respondents.

Design Primary controlled interrupted time series analysis and secondary before-and-after analysis of the impact of introducing MUP in Scotland using alcohol consumption data for England as control.

Setting Data from Kantar Worldpanel’s Alcovision survey, a continuous retrospective online timeline follow-back diary survey of the previous week’s alcohol consumption.

Participants 53 347 women and 53 143 men.

Interventions Introduction of a minimum price of 50 pence per UK unit (6.25 pence/g) for the sale of alcohol in Scotland on 1 May 2018.

Main outcome measures Number of grams of alcohol consumed per week, in total, in off-trade (eg, at home) and in on-trade (eg, in pubs, restaurants).

Results Primary interrupted time series analyses found that the introduction of MUP was associated with a drop in reported weekly total alcohol consumption of 5.94 g (95% CI 1.29 to 10.60), a drop in off-trade consumption of 3.27 g (95% CI −0.01 to 6.56) and a drop in on-trade consumption of 2.67 g (95% CI −1.48 to 6.82). Associated reductions were larger for women than for men and were greater among heavier drinkers than for lighter drinkers, except for the 5% of heaviest drinking men for whom an associated increase in consumption was found. Secondary before-and-after analyses found that reductions in consumption were greater among older respondents and those living in less deprived areas. The introduction of MUP was not associated with a reduction in consumption among younger men and men living in more deprived areas.

Conclusions Greater policy attention needs to be addressed to the heaviest drinking men, to younger men and to men who live in more deprived areas.

  • PUBLIC HEALTH
  • SOCIAL MEDICINE
  • Substance misuse

Data availability statement

Data may be obtained from a third party and are not publicly available. Kantar WorldPanel data cannot be shared due to licensing restrictions.

http://creativecommons.org/licenses/by-nc/4.0/

This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.

Statistics from Altmetric.com

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Strengths and limitations of this study

  • The study uses a large commercial dataset surveying the previous week’s alcohol consumption of 106 490 adults in Scotland and England.

  • The study uses location-controlled interrupted time series analyses of the potential impact of the introduction of minimum unit pricing (MUP) in Scotland, with the alcohol consumption of residents of England (and, in sensitivity analysis, residents of Northern England) as control.

  • The study assesses how the potential impact of MUP might differ by sex, level of alcohol consumption, age, social grade and level of residential deprivation of respondents.

  • The sample of respondents is not a random sample but rather a quota sample and cannot claim full representativeness of all adult residents in Scotland and England.

  • The study only assesses the immediate rather than the long-term impact of the introduction of MUP.

Introduction

The use of alcohol is one of the major risk factors for burden of disease and mortality found in global and European comparative risk analyses.1 2 Alcohol control policies are put in place to reduce this attributable harm. The WHO has identified the three so-called ‘best buys’ as the most effective, cost-effective and easy-to-implement policies: (1) policies to increase the price of alcohol via taxation increases or via floor pricing; (2) restrictions on availability of alcohol; and (3) bans on marketing of alcohol.3 Despite the demonstrated effectiveness of the best buy policies,4 other policies such as drink-driving or educational campaigns seem to be preferred by governments in Europe5 and elsewhere. However, following the lead of Scotland and some Eastern European countries (including Armenia, Belarus and Russia), floor-pricing policies (ie, policies where alcoholic beverages cannot be sold under a threshold price) are currently gaining support.6 7 Therefore, an evaluation of current policies and their impact is crucial to inform governments in other countries that are planning to institute such policies.8–10

This paper aims to evaluate the impact of a specific floor-pricing policy, the introduction of a minimum unit price (MUP) for all alcohol products in Scotland below which they cannot legally be sold. The MUP was set to be 50 GB pence per unit (8 g) of pure alcohol (ethanol) sold (6.25 pence/g) beginning on 1 May 2018.6 The rationale for introducing MUP as part of a larger national alcohol strategy in Scotland was to reduce hazardous and harmful alcohol consumption, targeting drinkers at the greatest risk of harm, those who tend to consume the cheapest alcohol, often purchased off-premise in supermarkets and shops where prices are comparatively lowest. Prior econometric modelling studies11 suggested that a MUP is likely to produce greater reductions in alcohol-related inequalities than either taxation on a volumetric basis (based on product strength/ethanol content) or an ad valorem basis (proportionate to product value). Part of this effect relies on preventing producers and retailers from absorbing some of the tax increases by further reducing prices, especially at the lower price points.12

While the evaluations of the Scottish MUP thus far have been positive, showing a general decrease in alcohol purchases, use and heavy drinking,8–10 many of the evaluations are based on alcohol sales or household expenditures which did not, or could not, differentiate by the sex of the drinker. However, such differentiation is necessary to determine if the underlying assumption of an appropriately targeted policy holds true, especially since a lot of the modelling before implementation was based on sex-unspecific price elasticities or general assumptions. Only very recently has sex-specific modelling of MUP been undertaken, which predicted larger reductions in men than in women.13 For example, a 0.5 pence MUP was predicted to lead to a 5.3% reduction in consumption and a 4.1% reduction in hospital admissions for men, but to a 0.7% reduction in consumption and a 1.6% reduction in hospitalisations for women. The Kantar Worldpanel (KWP) Alcovision survey,14 a continuous retrospective online timeline follow-back (TLFB) diary survey, allows us to specifically investigate the gender-based impact of MUP in Scotland using England as a control group. In addition to allowing us to disaggregate consumption by sociodemographic characteristics, a further strength of the Alcovision survey, which has been used in previous alcohol-policy related analyses,15 16 is its large sample size—approximately 30 000 different respondents from Great Britain (England, Scotland and Wales) each year.

Based on current empirical evidence and modelling-based assumptions, we would expect the following:

  1. The introduction of the MUP in Scotland would lead to a reduction in overall consumption.

  2. The reduction in consumption would be more pronounced for heavy drinkers with scarce resources; in Scotland this would be men from lower socioeconomic strata who would be most affected by MUP.

Methods

Study design

As a primary analysis we undertook location-controlled interrupted time series regression of the short-term associated impact of the introduction of MUP on the off- and on-trade alcohol consumption of Scottish men and women using consumption of English men and women as controls. We analysed immediate and level changes in consumption rather than changes in trends (slopes), in line with the findings of our previous analyses.9 10 We undertook a sensitivity analysis, repeating the interrupted time series regression using men and women resident in Northern England as control, rather than all of England, noting that residents in Northern England are more likely than residents from all of England to have a similar drinking culture to residents in Scotland. As a secondary analysis we undertook before-and-after analyses to investigate in more detail the potential impact of MUP by individual age of respondent and by individual residential deprivation ranking of where the respondent lived.

Data sources

Our data source is the KWP Alcovision survey,14 an ongoing cross-sectional online TLFB diary survey of the previous week’s alcohol consumption, with an annual sample of approximately 30 000 individuals aged 18+ years in Great Britain. Participants provide detailed data on their drinking occasions during the previous 7 days, including details on brands and volumes drunk, and whether these are consumed off-trade (eg, at home) or on-trade (eg, in a bar, pub or restaurant) for each occasion. Participants complete the survey only once, without repeated surveys. Quota samples based on age, sex, social grade and geographical region are drawn from Kantar’s managed access panel.14 Invitations to participate are sent out on set dates and timed such that completion dates of the survey occur during every month, and each day of the year is represented in the data. Weights based on age-sex groups, social grade and geographical region are constructed using UK census data. Based on client requests, Kantar oversamples residents from Scotland and those aged 18–34 years from both England and Scotland (see online supplemental figures 1 and 2, page 1). In the dataset we analysed, drink diaries were completed by 106 490 respondents from England and Scotland during the 4 years from 2015 to 2018, with an average of 512 diaries per week (SD 173), a rate which remained stable over the 4-year period (F=0.544, p=0.462).

Supplemental material

We received truncated postal code data, which we used to identify respondents as being residents of Scotland, England or Northern England (regions of North-West England, North-East England, and Yorkshire and Humber). We used the English17 and the Scottish18 Indices of Multiple Deprivation to group respondents into levels of residential deprivation (for details see online supplemental pages 2–5 and online supplemental figures 3–7).

The number of drinks consumed were recorded separately for on- and off-trade, with information given on serving sizes in millilitres (mL). In the dataset analysed we had records of all drinks consumed during the 7-day time period but not specified by day of week. Drinks were categorised within 19 categories, which we collapsed, grouped and coded as beers, ciders, wines, spirits, fortified wines and ready-to-drink products. In the dataset we analysed detailed product description was provided for beers, including alcohol-free beers, but not for other beverages. For non-beer products, the alcohol by volume (ABV) averages of the categories obtained from household purchase data over the same 4 years (2015–2018) were used.19 For beer products, the brand-specific ABVs from the household purchase data were used.19 Volume was combined with ABV to calculate grams of alcohol (1 mL alcohol=0.79 g pure alcohol). We summed consumption into grams of alcohol by drink group per week for each individual survey respondent.

In addition to the five deprivation groups, we also grouped individuals into: (1) four age groups (18–24, 25–44, 45–64 and 65+ years) and (2) four occupation-based social grade groups (AB (‘highest’), C1, C2, DE (‘lowest’)) based on the National Readership Survey.20

For the interrupted time series analyses we prepared weekly data by averaging consumption across all respondents for each of the 208 weeks in the study period, separately for men and women, and separately for total consumption, off-trade consumption and on-trade consumption. We plotted the seasonally adjusted total consumption over time (study week) by England and Scotland (online supplemental figure 8, page 6). We observed parallel trends between England and Scotland prior to the introduction of MUP, illustrating the appropriateness of England as a control area (tests for parallel trends, see online supplemental table 1, page 6).

To analyse the potential impact of MUP in reducing alcohol consumption by levels of consumption, we calculated, separately for men and women and for each country (Scotland and England) and for each week (from week 1 to week 208), the average consumption for separate percentiles of consumption ranging from 5% to 95% within 5% intervals.

Statistical analyses

Primary interrupted time series analyses

As primary analyses, interrupted time series regressions21 were undertaken with the weekly consumption data averaged across all respondents, and separately for men and women, over the full 208 weeks, where week 1 is the first week of 2015 and week 208 is the last week of 2018. As with our previous analyses,9 10 we created three new dependent variables of Scotland minus England (net effect) for each of the weeks for: (1) the average consumption of all grams of all alcohol per week, separately for men and women; (2) the average consumption of all grams of all alcohol per week consumed off-trade (eg, at home), separately for men and women; and (3) the average consumption of all grams of all alcohol per week consumed on-trade (eg, in pubs, bars or restaurants), separately for men and women.

For each of the three dependent variables, we examined the distribution visually and with Q–Q plots and found all variables, being the differences Scotland minus England (net effect) for the means of consumption by respondent for each of the 208 weeks, to be normally distributed (see online supplemental figure 9, page 7). We adjusted the dependent variables for any seasonality using the ratio-to-moving-average method.22 Based on Durbin–Watson tests23 (range 1.53–2.18), there was no evidence of autocorrelation, and based on augmented Dickey–Fuller tests,24 the series were found to be stationary (see table 1). We examined the immediate and permanent level changes due to the event, the introduction of MUP in Scotland, at week 174. The event variable was entered as a dummy variable, coded with 0 for each week before the event and with 1 for each week from the event forwards. Thus, in our generalised linear regression models, which we ran separately for men and for women, the dependent variables were the difference in reported consumption of grams of alcohol between Scotland and England (net effect). The independent variables were the dummy variable event and time (each week from 1 to 208). Interrupted time series regression equation 1 and SPSS syntax is presented in online supplemental box 1, page 8.

Table 1

Unstandardised coefficients from interrupted time series analyses (95% CI) for all respondents and for men and women separately by total consumption, off-trade consumption and on-trade consumption with Durbin–Watson statistic (value should be near 2.0) and augmented Dickey–Fuller test (p value should be <0.05) of models added

To test if MUP had an associated differential impact by sex of respondent we re-ran interrupted time series regression equation 1 for the total sample (both men and women), adding sex of respondent and the interaction term sex*introduction of MUP to the model (see online supplemental box 1, page 8).

We repeated interrupted time series regression equation 1 separately for each of the four age groups, four social grade groups and five deprivation groups (thus, comparing the same groups in England and Scotland). For these analyses, we transformed the continuous variables into their z-scores and used the z-scores as the dependent variables, so that the results could be compared between groups in terms of SD rather than original units. This allowed us to compare the relative importance of the regression coefficients, and thus changes, across the sociodemographic characteristics of the respondents.

For the analyses by the separate consumption percentiles, for each separate percentile we also created a difference in consumption by subtracting the mean consumption, Scotland minus England. We repeated interrupted time series regression equation 1 separately for each of the 19 percentiles (from 5% to 95%) and plotted the coefficient and 95% CIs associated with the event (introduction of MUP) by the percentile, separately for men and women.

Sensitivity analysis

We repeated interrupted time series regression equation 1 using men and women resident in Northern England as control for Scotland, rather than residents from all of England.

Secondary before-and-after analyses

The secondary before-and-after analyses were done with individual respondent 7-day consumption data summed across each week separately for men and women to better understand variation in the associated impact of MUP by age and deprivation, for each individual age and each individual deprivation score rather than by the four age groups and the five deprivation groups used in the interrupted time series analyses. For these analyses, we did not compute a new dependent variable (Scotland minus England) but rather used the original data by country. We examined the distribution of the dependent variables and found them to be highly dispersed (see online supplemental figures 10 and 11, page 11). We excluded all respondents with zero consumption during the previous week and then took the natural log of the consumption data, resulting in a normal distribution of the natural logged data (see online supplemental figures 12 and 13, page 12). In our models the independent variables were: the event variable (introduction of MUP), coded as a dummy variable as above for the interrupted time series analysis; country as a factor (England or Scotland); age as a dummy coded variable for each individual age year; deprivation as a dummy coded variable for each deprivation score rounded to an integer; and time (weeks) as a covariate. For each of the dependent variables we ran two separate models, one for age and one for deprivation score. Before-and-after analysis regression equation 2 and the SPSS syntax are presented in Supplement Box 2, pages 8–9.

From the results of the regression model and for each individual age and for each individual deprivation score, we took the difference in the marginal means (and the 95% CI of the differences) [Scotland*MUP*age /or/ deprivation score] minus [England*MUP*age /or/ deprivation score], this difference representing the added associated impact of MUP in Scotland over and above that in England for each individual age and for each individual deprivation score. We plotted the differences of the marginal means as above (with their 95% CIs) by each age and each integer deprivation ranking respectively, for men and women separately. We extracted the mean values of the changes (y-axes) from the plots and performed a linear regression of these values respectively by age and deprivation score, separately for men and women, to test how the differences in the marginal means between Scotland and England (net effect) differed by age and deprivation score. The before-and-after analysis regression equation 3 and SPSS syntax are presented in Supplement Box 3, page 9. We tested the difference in slopes between men and women for total consumption by repeating regression equation 3 for the total sample (both men and women), adding the interaction term sex*age/or/deprivation score as an additional independent variable to the model. Finally, given the relationship between age and deprivation score (Supplement Figure 9, page 5), we also tested if any relationship between changes in alcohol consumption associated with MUP and age of the respondent differed by deprivation group. We tested this by adding an interaction term age*deprivation group to the regression model (see Supplement Box 4, page 10).

Sensitivity analysis

We repeated before-and-after analysis regression equation 2 using a root-normal model, taking the square root instead of the log to normalise the consumption data. We tested if any relationship between changes in alcohol consumption associated with MUP and age and deprivation score of the respondent differed by the method of normalising the data. We tested this by adding an interaction term ‘type of normalisation (natural log or square root)’*age/or/deprivation score to the regression model (see online supplemental box 5, page 10).

Power calculations are reported in the online supplemental, page 13.

Analyses were performed with SPSS v26 (IBM Corp, 2019).25 For our regression models we used generalised linear models, procedure GENLIN.

Patient and public involvement

The research was done without public involvement. The public was not consulted to develop the research questions, nor was it involved in identifying the study design or outcomes. We did not invite the public to participate in the interpretation of results, nor in the writing or editing of this paper. There are no plans to directly involve the public in the dissemination of the research findings.

Results

Overall, 106 490 respondents (53 347 women and 53 143 men) contributed to the dataset (for details of numbers of respondents by country before and after the introduction of MUP and by sociodemographic characteristics, see Supplement Table 2, page 14). Although there were small differences prior to MUP between Scotland and England (proportion of female respondents and age and mean deprivation score of male respondents), these differences remained the same following MUP, except for the mean age of women (see Supplement Tables 3–5, pages 15–17). Whereas Scottish women in the sample were on average a little younger than English women before MUP, they were on average a little older than English women after MUP (Supplement Table 4, page 16).

For all respondents (English and Scottish), the mean reported consumption per week was 125.8 g for men (66.4% consumed off-trade) and 71.3 g for women (71.3% consumed off-trade; for details see Supplement Table 6, page 18). Consumption decreased with age similarly for both sexes, by 5.1 g per every 10 years of increasing age (95% CI 4.4 to 5.7) (see Supplement Figure 14, page 19). Consumption decreased by only a small amount with decreasing deprivation, similarly for both sexes, by 1.1 g per every 10 points (within a scale 1–100) of decreasing deprivation (95% CI 0.8 to 1.4) (see Supplement Figure 15, page 19).

Interrupted time series analyses: main findings

Figure 1 plots the differences in consumption of alcohol (g) Scotland minus England (net effect) for each of the 208 weeks, 2015–2018. Table 1 shows the results of the associated impact of MUP on alcohol consumption changes for all respondents and for men and women separately. For all respondents and for total consumption, the introduction of MUP was associated with a net drop in consumption (Scotland minus England) of 5.9 g per week (95% CI 1.3 to 10.6) (a 6.2% drop from the mean pre-MUP level in Scotland, 95% CI 2.3% to 8.4%). The reductions in consumption are largely driven by women (a reduction of 8.6 g per week, 95% CI 2.9 to 14.3) rather than by men (a reduction of 3.3 g per week, 95% CI –3.6 to 10.4). Supplement Table 7, page 20 gives the results of the models with the interaction terms (sex of respondent*event, the introduction of MUP). Based on the coefficient of the interaction term, women showed a greater reduction in consumption associated with MUP than men of 8.8 g per week (95% CI 1.9 to 15.7).

Figure 1

Plots of average weekly alcohol consumption Scotland minus England (net effect) for all respondents by week of study period for total alcohol consumption, off-trade consumption (eg, at home) and on-trade consumption (eg, in pubs, bars and restaurants) with T4253H smoothing.58 Black vertical line shows introduction of minimum unit pricing. Data used for primary interrupted time series analyses.

Interrupted time series analyses: sensitivity analyses

Table 2 shows the results of the sensitivity analyses using respondents from Northern England as control. For all respondents and for total consumption, the introduction of MUP was associated with a net drop in consumption of 5.9 g per week (95% CI 2.6 to 9.2) (Scotland minus England), a very similar finding to that when using all of England as a control (table 1). Based on the model with the interaction terms (sex of respondent*event, the introduction of MUP), women showed a greater reduction in consumption associated with MUP than men of 6.0 g per week (95% CI 1.0 to 11.0), a slightly lower level to that when using all of England as a control (see Supplement Table 8, page 20).

Table 2

Sensitivity analysis using Northern England as a control for Scotland

Associated changes in consumption following the introduction of MUP by characteristics of respondents

Figure 2 plots the associated changes in the difference in alcohol consumption (Scotland minus England) following the introduction of MUP by drinking percentile distribution of total alcohol consumption (for mean consumption by percentile see Supplement Figure 16, page 21, and for numerical data of figure 2 see Supplement Table 9, page 22, in which a footnote adds the average number of respondents per percentile). Up to the 45th percentile there was no associated reduction in alcohol consumption. From the 45th to the 85th percentile there were reductions in alcohol consumption associated with MUP, with the magnitudes of reduction greater for women than for men (regression coefficient (RC) 2.8 g per 5 percentile, 95% CI 2.0 to 3.6). For the 95th percentile the introduction of MUP was associated with an increase in consumption for men of 13.8 g (95% CI 5.8 to 21.5), but not for women (4.8 g, 95% CI −4.0 to 13.7).

Figure 2

Associated changes in the difference in consumption (Scotland minus England, net effect) following the introduction of minimum unit pricing (MUP) by drinking percentile distribution of total consumption. Blue lines: men; red lines: women. Thicker lines: means; thinner lines: 95% CIs. Horizontal black line set at zero (ie, no change). Results from primary interrupted time series analysis.

Figure 3 shows the associated changes in the difference in consumption following the introduction of MUP by age group (top graph), social grade (middle graph) and deprivation group (bottom graph), plotting standardised coefficients, allowing for relative rather than absolute comparisons across the groups (for numerical data, see Supplement Tables 10–12, pages 23–25).

Figure 3

Associated changes in consumption following introduction of minimum unit pricing (MUP) for all consumption, off-trade consumption and on-trade consumption (Scotland minus England, net effect) by age group (top graph), social grade group (middle graph) and deprivation group (bottom graph) for men (blue) and women (red). Consumption changes are standardised coefficients (units of SD) from primary interrupted time series analyses with 95% CIs.

By age group (figure 3, top graph), there was a pattern of greater associated drops in all consumption and in off-trade consumption for both men and women with increasing age. For younger men there was an increase in off-trade consumption, which was offset by decreases in on-trade consumption in the same group. There appeared to be no clear or consistent discernible pattern by social grade (figure 3, middle graph) or by deprivation group (figure 3, bottom graph). The secondary before-and-after analyses provide more detail of the associated impact of MUP by individual age and deprivation ranking.

Secondary before-and-after analyses

Figure 4 plots the associated changes in alcohol consumption (in grams of alcohol) following introduction of MUP for all consumption, off-trade consumption and on-trade consumption by gender and individual age. For men, reductions in consumption following the introduction of MUP became greater with increasing age for both total consumption (linear RC across age −0.088, 95% CI −0.094 to −0.083) and off-trade consumption (RC −0.092, 95% CI −0.097 to −0.088). For on-trade consumption, reductions in consumption became very slightly smaller with increasing age (RC 0.0038, 95% CI 0.0026 to 0.0050). For younger men (those aged <30 years), the introduction of MUP was not associated with a decrease in consumption, more so the younger the age, as upper 95% CIs were greater than zero. For women a similar pattern emerged, with reductions in consumption across all ages. Reductions in both total (RC −0.070, 95% CI −0.072 to −0.067) and off-trade consumption became slightly greater with increasing age (RC −0.087, 95% CI −0.090 to −0.085), whereas reductions in on-trade consumption became very slightly smaller with increasing age (RC 0.0179, 95% CI 0.0176 to 0.0182). The coefficient for the interaction term, sex*age (with women as reference category) was −0.019 (95% CI −0.025 to −0.013), indicating that the reduction in consumption was slightly greater with increasing age for men than for women.

Figure 4

Plots of the changes in alcohol consumption (g/week, with 95% CIs) associated with the introduction of minimum unit pricing (MUP) in Scotland, controlling for changes in England for each age year. Plots of men and women for total consumption, off-trade consumption and on-trade consumption. Thicker lines: means; thinner lines: 95% CIs. Horizontal black line set at zero (ie, no change). The changes are derived from the secondary before-and-after analysis regression equation 2. They represent, for each age, the difference in the marginal means (and 95% CIs of the differences) for [Scotland*event (introduction of MUP) *age (dummy coded variable for each age)] minus [England*event (introduction of MUP) *age (dummy coded variable for each age)].

Figure 5 plots the associated changes in alcohol consumption (in grams of alcohol) following the introduction of MUP for all consumption, off-trade consumption and on-trade consumption by gender and individual deprivation ranking. For men, reductions in consumption following the introduction of MUP became greater with less deprivation, more so for total consumption (RC −0.102, 95% CI −0.108 to −0.097) and off-trade consumption (RC −0.082, 95% CI −0.087 to −0.078) than for on-trade consumption (RC −0.020, 95% CI −0.022 to −0.019), with an indication that those living in the most deprived areas (bottom two-fifths) showed no decrease in consumption, more so the greater the deprivation (as upper 95% CIs were greater than zero). For women, a similar pattern emerged, with reductions in consumption across all deprivation scores. Reductions in consumption following the introduction of MUP became larger with less deprivation for total consumption (RC −0.050, 95% CI −0.051 to −0.049), off-trade consumption (RC −0.035, 95% CI −0.036 to −0.034) and on-trade consumption (RC −0.0151, 95% CI −0.01550.107 to −0.0147). The coefficient for the interaction term sex*deprivation score (with women as reference category) was −0.053 (95% CI −0.059 to −0.046), indicating that the reduction in consumption was slightly greater with less deprivation for men than for women.

Figure 5

Plots of the changes in alcohol consumption (g/week, with 95% CIs) associated with the introduction of minimum unit pricing (MUP) in Scotland, controlling for changes in England for each deprivation score (on a 100% scale). Plots of men and women for total consumption, off-trade consumption and on-trade consumption. Thicker lines: means; thinner lines: 95% CIs. Horizontal black line set at zero (ie, no change). The changes are derived from the secondary before-and-after analysis regression equation 2. They represent for each deprivation score (the higher the deprivation score, the less deprived) the difference in the marginal means (and 95% CIs of the differences) for [Scotland*event (introduction of MUP) *deprivation score (dummy coded variable for each deprivation score)] minus [England*event (introduction of MUP) *deprivation score (dummy coded variable for each deprivation score)].

The age-related patterns in figure 4 were independent of deprivation. Before-and-after analysis regression equation 4 found no interaction between age in years and deprivation group in the changes in total alcohol consumption (Scotland minus England, net effect) associated with the introduction of MUP: for men, the coefficient for the interaction was −2.2-5 (95% CI −5.5-3 to 5.4-3) and, for women, the coefficient was 1.6-3 (95% CI −1.1-3 to 4.2-3). In other words, the slopes between changes in alcohol consumption by age for men and women plotted in figure 4 were almost identical across the five deprivation groups.

Before-and-after analyses: sensitivity analyses

We repeated the before-and-after analyses using the square root (as opposed to logged) grams of alcohol consumption as the dependent variable, with similar patterns of findings to Figures 4 and 5 (see Supplement Figures 17 and 18, pages 26–27). There were, however, differences in the slopes. For total consumption, before-and-after analysis regression equation 5 found, with age, that the slope for logged grams of alcohol was slightly steeper for men (RC of interaction term ‘type of normalization*age’ −0.017, 95% CI −0.025 to −0.008), but slightly less steep for women (RC of the interaction term 0.082, (95% CI 0.078 to 0.087) than the slope for the square root of consumption. There were similar findings in the differences in slopes for dependence score, the slope for logged grams of alcohol being slightly steeper for men (RC of interaction term ‘type of normalization*dependence score’ −0.059, 95% CI −0.068 to −0.050) and slightly less steep for women (RC of the interaction term 0.040, 95% CI 0.038 to 0.043).

Discussion

We found that the introduction of MUP in Scotland was associated with a change in overall reported alcohol consumption in line with the predicted direction. Compared with respondents from England, Scottish respondents reported a 6.2% drop in alcohol consumption (95% CI 2.3% to 8.4%) associated with MUP. Sensitivity analyses using respondents from Northern England, with more similar drinking levels to Scotland than England as a whole,26 found an almost identical associated drop in alcohol consumption. The drop in consumption was larger for heavier drinkers than for lighter drinkers, with the exception of the top 5% of heaviest drinking men for whom there was an increase in consumption associated with the introduction of MUP.

Against expectations, we found that associated drops in consumption were greater for women than for men, both in the main (using all of England as a control) and in the sensitivity (using Northern England as a control) analyses. Men and women also responded differently by age. Based on both the interrupted time series analysis and the before-and-after analysis, the size of the associated drop in consumption for men became smaller with decreasing age, with younger men showing no associated decrease in consumption. For women, the associated drop in consumption also became smaller with decreasing age, although less so than for men.

We included two potential measures of socioeconomic disadvantage: social grade and an index of residential deprivation based on multiple measures of income, employment, education, health, crime, access to housing and environmental quality,17 18 noting that the risk of alcohol-related harm increases both the more socioeconomically disadvantaged the individual is, and over and above that, the more socially disadvantaged the residential area in which the individual resides.27 It should be noted that estimates of the indices of residential deprivation differ between Scotland and England and thus, in absolute terms, they may not be the same. However, in our analyses we compare relative deprivation; for example, comparing the bottom fifth of deprivation of Scotland with the bottom fifth of deprivation of England, noting that relative deprivation itself is a key determinant of ill health.28 Based on the interrupted time series analyses, for both men and women there was no discernible pattern by social grade or deprivation group. However, based on the secondary before-and-after analyses (both main and sensitivity), the size of the associated drop in consumption for men became smaller with increasing deprivation, with men living in the most deprived areas having no associated decrease in consumption. For women, the associated drop in consumption also decreased slightly with decreasing deprivation score, although less so than for men.

The drop in consumption of 6.2% is a little lower than the 7.6% drop we found in our previous analysis of household purchase data in both the short9 and medium term.10 As with the present study based on survey data, our previous analyses of household purchase data also found that drops in consumption were greater among households with higher rather than lower usual purchases of alcohol.9 10 However, with our previous analyses of household purchase data, we could not test the impact of MUP on purchases by age or gender as the purchase data were for the household as a whole and not attributable to individual household family members. Nor did those analyses report the impact of MUP by the social grade of the household or the level of deprivation in which the household was located. The findings presented in this paper thus provide a more nuanced understanding of the differential impact of MUP on different population subgroups. Specifically, what we identified in the present analysis is the top 5% of heavy drinking men did not reduce their consumption in association with MUP; rather, our results suggest an increase in associated consumption among this group. For women, there was an upturn in changes in alcohol consumption in the heaviest drinking percentiles (figure 2); that the lower 95% CI for women did not cross zero could be due to the relatively small numbers of respondents in each of the 19 consumption percentiles (Supplement Table 9, page 22).

We do not know why, for both younger men (those aged <32 years) and for those living in residential areas in the bottom two-fifths of deprivation, there was no decrease in consumption associated with MUP compared with older men and those living in less deprived areas. It has been suggested that some very heavy drinkers (as we found for the top 5% of heavy drinking men) would be less prone to the potential impact of MUP,29 and in potential need of additional support to cope with the impact of MUP.30 Responses to MUP might vary by individual and psychosocial factors including socioeconomic disadvantage, which may interact with the situational availability of alcohol.31 This is clearly an area for further study.

Before we discuss the implications of the results, it is important to mention potential strengths and limitations of our study. We based our analysis on a large sample of 53 347 women and 53 143 men from England and Scotland which— apart from the oversampling of 18–34 year-olds—was, in general, representative of the sex and age structure of the population (Supplement Figures 1 and 2, page 1). The sample was neither more nor less deprived than the population of England or Scotland as a whole (Supplement Figure 3, page 3). A strength of the interrupted time series analyses is the large number of data points (weekly consumption) before (n=173) and from the introduction of MUP onwards (n=25), considered more than sufficient for interrupted time series analyses.22 A second strength overall and for the before-and-after analyses is the large sample size (88 894 respondents prior to the introduction of MUP and 17 596 respondents thereafter). A third strength is the use of a location control, both all of England and Northern England in the sensitivity analysis. Location controls allow for other extraneous factors beyond the intervention to be controlled for, such as an unusual heat wave during the months of June, July and August that affected all Great Britain.32

For limitations, first, all results are based on subjective reports of drinking. While such subjective reports tend to underestimate consumption as measured by sales or other recorded data in general in all European countries,33 there is no reason to believe that under-reporting should differ by country or region or before or after the introduction of the MUP. The timeline follow-back survey method has been criticised for the limited time period of drinking it covers, thus missing heavy episodic drinking occasions among participants with a low frequency of such occasions. This limitation for classifying individuals is actually a strength when it comes to the characterisation of population averages; however, the shorter the time period, the smaller the biases due to memory and the more accurate the population average.34 Second, as with all survey-based research on alcohol, this research cannot claim full representativeness.35 Statistical theory stipulates such representativeness needs to be based on probabilistic sampling design (ie, all residents from England and Scotland need to be assigned a probability >0) combined with high response rates unaffected by systematic non-response.36 However, these conditions can no longer be reached in modern surveys involving alcohol, no matter which methodology is used.35 37–39 Instead, post-stratification based on sex, age, social grade and geographical region was used to allow for generalisations to be made for the general population. The quota sample was derived from Kantar’s managed access panel. Data were not available and not attainable on the number of respondents approached to achieve the 30 000 respondents surveyed each year, and this information is not mentioned in existing publications based on the Alcovision survey, for example.15 16 Unlike the household purchase data which record purchases wherever they are made and thus account for cross-border purchases, we are unable to account for any cross-border purchasing or drinking the respondents might have engaged in. If this was significant (and a study on licensing compliance would suggest that it is not40), one might hypothesise that the estimated sizes of the associated impact with MUP in reducing alcohol consumption would differ between using Northern England or all of England as a control, which was not the case. Finally, as we only had data to the end of 2018, we have been unable to examine the impact of MUP beyond the immediate term.

In our analysis we used both interrupted time series analysis and before-and-after analyses. With the interrupted time series analysis, we used England (or Northern England) as a location control, creating new dependent variables, the differences between Scotland and England. Interrupted time series analysis is an appropriate methodology for investigating the impact of a newly introduced natural experiment (the introduction of MUP) that takes into account seasonal variation and autocorrelation of the data over time.22 The before-and-after analysis is simply comparing the means before and after the introduction of MUP. Results of before-and-after analyses are often presented along with interrupted time series analyses, as we have done previously with household purchase data.9 While we add in an interaction term of country*event (introduction of MUP), which should take into account common events outside of MUP that occurred in both Scotland and England, our analyses are unable to control for seasonal variation when comparing the longer time period before the introduction of MUP and the 8-month period following the introduction of MUP.

Externally validated indicators35 39 using sales41 42 or household purchasing data as the basis9 10 corroborate our results that, in comparison with England over the same and longer time periods, the introduction of the MUP was associated with a decrease in alcohol consumption. Finally, the reductions in alcohol consumption in Scotland were part of an overall national strategy or framework for alcohol policy, where all measures had already been extensively covered in the press. It cannot be excluded that the actual reductions may have been due in part to the media reports surrounding the introduction of the MUP rather than to the floor pricing itself (for an example of an alcohol policy measure where the media impact seems to be stronger, see Møller43). However, it is highly unlikely that media reports would produce exactly this abrupt and permanent pattern—that is, a drop in consumption starting exactly at the date of introduction of MUP and lasting for the time period studied, in comparison to a control group.

Despite these potential limitations, most research corroborates the results of our study that the MUP resulted in a reduction of overall alcohol consumption compared with England or Northern England.9 10 41 42 Overall, research was based on a number of designs including purchasing data from households or sales records. Our results here were based on a control group design, where the intervention was only introduced in one group, thus strengthening our confidence in a real effect.44

When the Minister for Public Health, Sport and Wellbeing introduced the 2018 alcohol policy framework,6 he emphasised that the implementation of the MUP was strongly motivated by an interest in decreasing health inequalities through a reduction in alcohol consumption among the heaviest and most vulnerable drinkers. Our results indicate that this goal may not be fully realised: first, we found that women, who are less heavy drinkers in our data and in almost all surveys worldwide to date,45 reduced their consumption more than men; second, the 5% of heaviest drinking men had an increase in consumption associated with MUP; and, third, younger men and men living in more deprived areas had no decrease in consumption associated with MUP. These results are surprising as modelling studies would have suggested otherwise.11 14

We can only speculate about the reasons for the increase in the 5% of the heaviest drinking men. Several studies have found that overall, heavier drinkers—including people with alcohol use disorders—react less to price than the general population (ie, they react more price inelastic and their consumption is determined by other factors46 47). However, while this may explain lower reductions, it cannot explain an increase in consumption. Such a polarisation with increasing consumption of the heaviest drinkers in overall decreasing consumption levels has now been observed in several studies, often in adolescents and young adults.48 49 These studies indicate that such polarisation means a deviation from the standard collective theory of all subgroups changing in the same direction,50 but fall short on good explanations as to why this is the case.

The results may also imply a diminished impact on alcohol-attributable hospitalisations and mortality, which have been shown to be strongly associated with heavy drinking in men and in those of lower socioeconomic status.51–54 Indeed, a large controlled study on emergency department visits following the introduction of MUP did not show any reduction in alcohol-related emergency department visits.55

Before any further conclusions can be drawn, we need to corroborate our sex-, age-, heavy drinking- and socioeconomic status-related findings in different studies. This seems important as different conclusions about the impact of MUP may result for other countries. If indeed the findings of our study are corroborated, then additional and/or different pricing mechanisms may need to be considered to reduce alcohol-attributable hospitalisations and mortality. For instance, several harms from alcohol use are specifically linked to on-trade drinking, such as public disorder and violence.56 Recent experiences in Lithuania have shown substantial reductions in all-cause mortality following a taxation increase, which mainly affected men.57

Data availability statement

Data may be obtained from a third party and are not publicly available. Kantar WorldPanel data cannot be shared due to licensing restrictions.

Ethics statements

Patient consent for publication

Ethics approval

Ethical approval was not required as our analyses are based on a publicly available commercial dataset.

Acknowledgments

We thank Kantar Worldpanel for providing the raw data and reviewing the method description as it describes the data collection. Kantar Worldpanel provided the raw data at no cost to Newcastle University under a direct contract. Through its own contracted work as a market research company, Kantar Worldpanel received reimbursement from AB InBev to cover the costs of the data. Kantar Worldpanel has similar commercial relationships with other customers who pay to have data collected on food and non-food items available for sale in supermarkets and other retail outlets covered by the WorldPanel. Kantar Worldpanel and no other entity had any role in the study design, data analysis, data interpretation or writing of the manuscript. Professor Kaner, supported by a National Institute for Health Research (NIHR) Senior Investigator award, is Director of the NIHR Applied Research Collaboration, North East and North Cumbria. Dr O’Donnell is a National Institute for Health Research (NIHR) Advanced Fellow. The views expressed in this article are those of the authors and not necessarily those of NIHR or the Department for Health and Social Care.

References

Supplementary materials

  • Supplementary Data

    This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.

Footnotes

  • Twitter @JakobManthey

  • Contributors JR conceptualised the paper and prepared the initial draft of the introduction and discussion. PA undertook the analyses and prepared the draft of the methods and results and submitted the paper. PA, AO’D, EJ-L, JR, JM and EK refined the various versions of the full paper and approved the final manuscript for submission. The corresponding author attests that all listed authors meet authorship criteria and that no others meeting the criteria have been omitted. All authors had access to the data used for analyses, and PA, AO’D and EJ-L verified the raw data sets received from Kantar WorldPanel and are the guarantors for the data used for the analyses.

  • Funding The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.

  • Competing interests All authors have completed the ICMJE uniform disclosure form at www.icmje.org/coi_disclosure.pdf and declare no support from any organisation for the submitted work within the previous five years. PA declares receipt of funds from AB InBev Foundation to provide evidence-based public health comment on the proposed content and evaluation of the Foundation’s global drinking goals, outside of the submitted work. The remaining authors declare no financial relationships with any organisations in the previous five years that might have an interest in the submitted work. All authors declare no other relationships or activities that could appear to have influenced the submitted work.

  • Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.