Not everyone benefits equally from Sunday alcohol sales bans: socioeconomic differences in alcohol consumption and alcohol-attributable mortality

Abstract

We examined socioeconomic variations in the association of off-premises Sunday alcohol sales bans and alcohol consumption and alcohol-attributable mortality in the United States. We analyzed associations between Sunday sales ban presence and alcohol consumption patterns, allowing for a differential effect by education in fixed-effects regression models using data from the 2000–2019 Behavioral Risk Factor Surveillance System. Mortality data from the National Vital Statistics System (2000–2019) were analyzed in interrupted time-series analysis to test the effect of lifting the Sunday sales ban in Minnesota (07/01/2017) on alcohol-attributable mortality. Regression analyses indicated lower alcohol consumption when Sunday sales bans were in place, with an overall stronger effect on those with high education. The repeal of the Minnesota ban resulted in a significant mortality increase, especially among individuals with high education. While overall effective, off-premises Sunday alcohol sales bans appear inadequate to address socioeconomic inequalities in the alcohol-attributable health burden.

Introduction

Restricting the availability of alcoholic beverages is one of the most cost-effective policies for lowering the health burden attributable to alcohol (Chisholm et al., 2018). Globally, alcohol accounted for 2.5 million deaths in 2019, putting it among the leading behavioral risk factors for premature mortality (GBD 2019 Risk Factors Collaborators, 2020). In the United States (US), the alcohol-attributable mortality has increased by one fifth since the millennium (age-standardized mortality rate, 2000: 23.7 per 100,000, 95% confidence intervals [CI]: 19.7–28.2, 2019: 28.1 per 100,000, 95% CI: 24.2–32.9; GBD 2019 Risk Factors Collaborators, 2020). This increase has been accompanied by an accelerating liberalization of effective alcohol control policies, including availability restrictions. For example, the number of states with an off-premises Sunday sales ban have decreased from 20 to eight between 2000 and 2019 (National Institute on Alcohol Abuse and Alcoholism (NIAAA), 2022). Previous research has demonstrated that this liberalization of temporal availability restrictions contributed to an increase in alcohol use and alcohol-related health harms in the US (Popova et al., 2009; Yörük, 2014).

Importantly, the health burden attributable to alcohol is unequally distributed within the population. Across different SES indicators, low-SES men and women were found to have a 2.9 to 12.3 and 1.8 to 4.8-fold increased risk of dying from an alcohol-attributable cause, respectively, compared to their same sex high-SES counterparts (Probst et al., 2015). These socioeconomic inequalities in the alcohol-attributable mortality significantly contribute to growing health inequalities in the US and a substantial gap in life expectancy of 9 (women) and 12 years (men) comparing individuals with low and high education (Probst et al., 2022).

Strengthening alcohol control policies might be suitable for reducing alcohol-attributable health inequalities, while progressive liberalization may exacerbate them, as evidenced by research on pricing policies. For example, education-based inequalities in all-cause mortality narrowed after a marked increase in the alcohol excise taxes in Lithuania in 2017 (Manthey et al., 2023). In Scotland, the introduction of a minimum unit price (i.e., setting a floor price for a unit alcohol) in 2018 led to a decrease in hospitalizations and deaths fully attributable to alcohol in more socioeconomically deprived groups (Wyper et al., 2023). According to a recent systematic review, these changes may be driven by a greater responsiveness of those with low incomes to pricing policies and a subsequent larger reduction in alcohol use (Kilian et al., 2023). However, such consumption changes were not observed when using other SES indicators and no study has yet examined alcohol consumption changes by SES following policies regulating alcohol availability. Such differential effects were, however, suggested by one study investigating the repeal of the Sunday alcohol sales ban in Pennsylvania in 2003, where crimes near liquor stores were found to have significantly increased in low but not high SES neighborhoods after the policy change (Han et al., 2016).

Given the continued trend towards liberalizing Sunday alcohol sales and rising health inequalities in the US, it is important to study whether such policy changes disproportionally affect alcohol consumption and harm in lower socioeconomic groups. To close this pertinent knowledge gap, we conducted two studies investigating the differential association of off-premises Sunday alcohol sales on alcohol consumption (study 1) and alcohol-attributable mortality (study 2) across socioeconomic groups in the US. Specifically, in study 1, we examine whether the association between state-level Sunday sales bans and alcohol consumption patterns differs across education groups using repeated cross-sectional survey data. In study 2, we used the example of Minnesota to investigate the impact of repealing the Sunday sales ban on 100% alcohol-attributable mortality, stratified by education groups, using interrupted time-series analysis.

Methods

Two distinct approaches were used to evaluate the policy’s association with alcohol consumption (study 1) and alcohol-attributable mortality (study 2). For study 1, we analyzed repeated cross-sectional survey data from the Behavioral Risk Factor Surveillance System (BRFSS; Centers for Disease Control, 2022). For study 2, we conducted an interrupted time-series analysis to quantify the impact of permitting Sunday sales on 100% alcohol-attributable mortality in Minnesota using quarterly aggregated mortality data from the National Vital Statistics System (NVSS; National Center for Health Statistics, 2022). Minnesota lifted its Sunday sales ban on wine and spirits on 07/01/2017 and was chosen as case study for pragmatic reasons based on the timing of the policy (see Supplement S1).

We have chosen education as SES indicator as it was available in both databases and shown to have a stronger impact on alcohol-attributable health inequalities than other indicators (Probst et al., 2021). We distinguished three levels of educational attainment: high school diploma or less (low), some college but no bachelor’s degree (medium), and bachelor’s degree or more (high).

Description of data sources

Study 1:

The BRFSS is an annual repeated cross-sectional survey using a multistage-cluster sampling design. The survey collects representative data from non-institutionalized US residents in 50 states plus Washington DC. We included individual-level data from 6,989,274 adults (18+ years) participating between 2000 and 2019 with complete reports on relevant variables. A sample description and information on missing data is available in the Supplement S2.

Study 2:

Mortality data was obtained for the years 2000 to 2019 (National Center for Health Statistics, 2022). Information on sex, age, education, and cause of death were extracted from the Multiple Cause-of-Death Files of the NVSS. The data contained 785,307 adult deaths in Minnesota. Information on sex and age was complete, while education was missing for 0.9% of the deaths (see Supplement S3). To avoid underestimation of deaths, this missing information was estimated by randomly allocating deaths to each education level based on the observed proportion of deaths by education, sex, age, year, and cause of death.

Data on off-premises Sunday sales policies were obtained for each US state from the Alcohol Policy Information System (National Institute on Alcohol Abuse and Alcoholism (NIAAA), 2023). We distinguished between no (‘0’) and any Sunday sales ban (‘1’). The latter includes partial bans, where beer sales are allowed, and local options, authorizing local authorities to permit Sunday sales despite a state-wide ban (Supplement S4).

Definition of outcome variables

Study 1:

We studied three alcohol consumption variables: (1) any alcohol use within the past 30 days (yes/no), (2) average grams of pure alcohol consumed per day among past-month alcohol users, and (3) any hazardous alcohol use among past-month alcohol users (yes/no). The average grams of pure alcohol consumed per day were calculated based on the drinking frequency and the average number of drinks per drinking occasion. The alcohol intake of respondents indicating daily consumption levels above 200 grams of pure alcohol (0.1%) was set to this threshold. Hazardous alcohol use was defined as any alcohol use exceeding an average daily intake of 40 or 60 grams of pure alcohol in women and men, respectively, which is the equivalent of about 3 and 4 US standard drinks.

Study 2:

The age-standardized 100% alcohol-attributable mortality rate was used as outcome and calculated using population counts from the American Community Survey (ACS; Ruggles et al., 2022) Mortality rates were age-standardized using 2010 ACS population counts by sex and education. The following causes of deaths were considered to be fully attributable to alcohol (ICD-10 codes (World Health Organization, 1992): acute intoxication (F10.0), finding of alcohol in blood (R78.0), alcohol poisonings (T51, X45, X65, Y15, Y90), alcohol use disorders (F10.1-F10.9), alcohol-induced pseudo-Cushing’s syndrome (E24.4), degeneration of nervous system due to alcohol (G31.2), alcoholic polyneuropathy (G62.1), alcoholic myopathy (G72.1), alcoholic cardiomyopathy (I42.6), alcoholic gastritis (K29.2), alcoholic liver diseases (K70-70.4, K70.9), alcohol-induced pancreatitis (K85.2), alcohol-induced chronic pancreatitis (K86.0).

Statistical analysis

Study 1:

We ran sex-stratified fixed-effects regression models using the lme4 package in R (B. Bolker, 2022). We computed logistic and linear regression models for binary and continuous outcome variables, respectively. Given the right-skewed distribution of the outcome variable average consumption levels, we applied a log-transformation. To assess the differential association of the policy with alcohol consumption patterns across socioeconomic groups, we added the interaction term of the policy with the education variable. The regression models were adjusted for individual-level (age, race and ethnicity, marital status) and state-level covariates (control state, drinking culture (Kerr, 2010), unemployment rate (z-standardized logarithm)(US Bureau of Labor Statitics, 2023). State was added as fixed effect. As we did not observe any secular time trends in the outcome residuals, year was omitted. Estimated marginal means (EMM) were computed using the R package ggeffects to illustrate the interaction effect (Lüdecke, 2018).

We conducted three sets of sensitivity analysis: First, we repeated all analyses described above accounting for the complex survey design of the BRFSS using the R package survey (T. Bolker, 2023). Given a change in the BRFSS sampling approach in 2011, impacting the weighting methodology, we did not apply sample weights in the main analysis. In the sensitivity analysis including survey weights, a dummy variable was included indicating the period before (period 1: 2000–2010) and after this methodological change (period 2: 2011–2019). We further reiterated the main analysis excluding respondents from states with local options (sensitivity analysis 2) and excluding all states in which Sunday alcohol sales were permitted during the entire study period (sensitivity analysis 3). In other words, we restricted our sample in the latter sensitivity analysis to only “intervention” (i.e., states that repealed their Sunday sales ban) and “control” states (i.e., states that enacted a Sunday sales ban during the entire study period).

Study 2:

For the Minnesota case study, we conducted an interrupted time-series analysis—an appropriate methodology to assess intervention effects without requiring a control group, while accounting for historical trends, seasonality, and autocorrelation (Beard et al., 2019). We fitted generalized additive models for each sex and education group using the R package mgcv (Wood, 2022). A detailed description is available in Supplement S5.

In brief, we first established baseline models for the pre-intervention time period (2000/Q1-2017/Q2). Next, three candidate models were built for each sex-education-group including the full time-series (2000/Q1-2019/Q4) and additional terms indicating (1) a level change (i.e., step change coded ‘0’ before and ‘1’ after the policy change), (2) a slope change (i.e., post-intervention trend), and (3) a level and slope change. Candidate models were compared using Akaike and Bayesian information criterion and the best-fitting model was selected based on the largest variance explanation (R²). If there were similar fitting models, we preferred the model with fewer variables. The relative change in the age-standardized mortality rate was then calculated by dividing the observed mortality rate within one year after the policy change (2017/Q3 to 2018/Q2) with the predicted mortality rate under a counterfactual scenario of no policy change over the same period. By fitting the baseline model to the pre-intervention data and using this model to predict the counterfactual scenario for comparison with the observed data, the interrupted time-series analysis accounts for the historical trend in the mortality data.

The R codes are publicly available at the Figshare repository (doi: {will be added upon acceptance}).

Results

Changes in off-premises Sunday alcohol sales bans since 2000

Figure 1 illustrates changes in the legislation on Sunday sales between 2000 and 2019. The majority of states (n = 30) never banned Sunday sales, while seven states had a partial ban (i.e., beer sales are exempted and/or local options) and one a full ban on Sunday sales in force throughout the entire study period. The remaining 12 states lifted their partial or full bans between 2000 and 2019. There was no state introducing a Sunday sales ban.

Figure 1.

Off-premises Sunday sales ban on alcoholic beverages between 2000 to 2019. Black: no ban; green: partial or full ban; stripe pattern: ban was repealed during the study period. Minnesota is highlighted in yellow.

Study 1: The association between off-premises Sunday sales bans and alcohol consumption

The results from the regression models are shown in Table 1. Sunday sales bans were generally associated with lower alcohol consumption. Women but not men had a lower likelihood to report any alcohol use in states and years with a Sunday sales ban compared to those allowing Sunday sales. Moreover, with a Sunday sales ban, the average daily drinking levels were on average 4% and 8% lower among male and female past-month alcohol users, respectively, and the odds of any hazardous alcohol use were reduced by 13% and 28% in men and women, respectively. In the sensitivity analysis including survey weights (Supplement S5), these main effects of Sunday sales bans on alcohol consumption patterns were evident in women only (men: all p’s > .05).

Table 1.

Results from the sex-stratified mixed-effects regression models using data from the Behavioral Risk Factor Surveillance System 2000–2019.

	Men	Women
Outcome: any alcohol use 1
Education: medium (ref: high)	0.65 (0.64, 0.65) ***	0.62 (0.62, 0.62) ***
Education: low (ref: high)	0.44 (0.43, 0.44) ***	0.36 (0.36, 0.37) ***
Sunday sales ban (ref: no ban)	0.99 (0.97, 1.01)	0.98 (0.97, 0.99) **
Interaction Sunday sales ban X medium education (ref: no ban, high education)	1.10 (1.09, 1.12) ***	1.03 (1.01, 1.04) ***
Interaction Sunday sales ban X low education (ref: no ban, high education)	1.13 (1.12, 1.14) ***	1.01 (1.00, 1.02)
Outcome: average daily drinking levels 2
Education: medium (ref: high)	0.00 (−0.01, 0.01)	−0.10 (−0.11, −0.10) ***
Education: low (ref: high)	0.07 (0.06, 0.07) ***	−0.16 (−0.17, −0.16) ***
Sunday sales ban (ref: no ban)	−0.04 (−0.05, −0.03) ***	−0.09 (−0.10, −0.08) ***
Interaction Sunday sales ban X medium education (ref: no ban, high education)	0.03 (0.02, 0.04) ***	0.02 (0.01, 0.03) ***
Interaction Sunday sales ban X low education (ref: no ban, high education)	0.03 (0.02, 0.04) ***	0.04 (0.03, 0.05) ***
Outcome: any hazardous alcohol use 1
Education: medium (ref: high)	2.12 (2.06, 2.19) ***	1.55 (1.50, 1.60) ***
Education: low (ref: high)	3.42 (3.33, 3.52) ***	2.11 (2.05, 2.18) ***
Sunday sales ban (ref: no ban)	0.87 (0.82, 0.93) ***	0.72 (0.67, 0.78) ***
Interaction Sunday sales ban X medium education (ref: no ban, high education)	0.98 (0.92, 1.04)	1.02 (0.95, 1.09)
Interaction Sunday sales ban X low education (ref: no ban, high education)	0.96 (0.91, 1.02)	1.11 (1.04, 1.18) **

Note.

^**p < .01

^***p < .001.

95% confidence intervals indicated in brackets. Ref: reference. Models adjusted for age, marital status, race and ethnicity, drinking culture, annual state-specific unemployment rate (z-standardized), control state status, and state.

¹regression coefficients reflect odds ratios.

²regression coefficients are based on linear models and can be interpreted as factor for every one-unit increase in the independent variable if exponentiated.

The association between the Sunday sales ban and alcohol consumption varied by education (Figure 2): For any alcohol use, men and women with low and medium education were generally considerably less likely to report drinking compared to their same-sex counterparts with high education in both states with and without a Sunday sales ban (i.e., main effect of education). However, these education-based differences in any alcohol use were narrowed in states and years with a Sunday sales ban compared to states and years without a ban (i.e., interaction effect), driven by a higher likelihood of past-month alcohol use in respondents with low (men only) and medium education (Figure 2–1). In the survey-weighted sensitivity analysis, this interaction was only significant in men.

Figure 2.

Estimated marginal means (EMM) by education group and regulation of Sunday sales on (1) any past-month alcohol use, (2) average daily drinking levels, and (3) any hazardous alcohol use in (A) men and (B) women. Low: high school diploma or less, medium: some college but no bachelor’s degree, high: bachelor’s degree or more. *past-month alcohol users only.

For average daily drinking levels, a sex-specific perspective is warranted, as men reporting past-month alcohol use had generally higher daily drinking levels with low education compared to those with high education, while women had lower daily drinking levels with low and medium compared to high education (i.e., main effect of education). In men, the observed socioeconomic difference was more pronounced in states and years with a Sunday sales ban compared to states and years without a ban, driven by lower drinking levels in high-SES men when a Sunday sales ban was in place (i.e., interaction effect, Figure 2–2A). In women, however, the apparent socioeconomic difference was less pronounced (Figure 2–2B). In the survey-weighted sensitivity analysis, the interaction was only significant comparing men and women with low and high education.

Finally, men and women with low or medium education reporting any past-month alcohol use had generally higher odds of any hazardous alcohol use compared to their same-sex counterparts with high education (i.e., main effect of education). A significant interaction between education and Sunday sales ban was observed in women only: women with low education had higher odds of any hazardous alcohol use in states and years with Sunday sales ban compared to those with high education and states and years without a ban (all other p’s > .05, see Table 1). In the survey-weighted sensitivity analysis, this interaction was not significant, but men with low or medium education had lower odds of any hazardous alcohol use when a Sunday sales ban was in place.

Taken together, our analyses suggest that, while Sunday sales bans are generally associated with lower alcohol consumption, the policy’s favorable association with drinking was less pronounced in the low and medium education groups. These findings were corroborated by our sensitivity analyses (Supplement S7, S8). Specifically, for any alcohol use, the main effect of Sunday sales ban, as well as the interaction effect of the ban and low education were significant in both sensitivity analyses for both men and women (compared to just women and just men, respectively, in the main analysis). For average daily drinking levels and any hazardous alcohol use among current alcohol users, we further observed similar significant main and interaction effects as in the main model. The only exception to this is men with medium education having slightly lower average daily drinking levels compared to those with high education in the sensitivity analysis restricting the sample to states that permitted Sunday alcohol sales (Supplement S8).

Study 2: The impact of permitting off-premises Sunday sales on alcohol-attributable mortality in Minnesota

The trends in the age-standardized alcohol-attributable mortality rates in Minnesota by sex and education are depicted in the Supplement S9, revealing an increase in mortality rates over the full study period across all groups. Taking the time point of the intervention (07/01/2017) as reference, visual inspections suggest an increase in the mortality rates in the post-intervention years, which appears to be most pronounced for alcohol-related liver cirrhosis.

This observation is supported by the interrupted time-series analyses for all groups but men with low education (Table 1, Supplement S9). For men with medium or high education, there was a significant step increase in the age-standardized alcohol-attributable mortality rate: compared to a counterfactual scenario without the policy change, the mortality rates increased by +19.0% (medium) and +51.8% (high education) within the first year of permitting Sunday sales. In women, a similar increase was observed, with significant slope (low and medium education) and step increases (high education) in mortality rates following the policy change. Mortality rates increased by +9.5%, +11.4% and +49.4% in women with low, medium, and high education, respectively. Variance explanation (adjusted R²) in the final models varied between 23.3% and 74.8% for women and 45.8% and 59.9% for men with high and low education, respectively.

Discussion

In this study, we provide novel evidence on the differential association of off-premises Sunday alcohol sales and alcohol consumption and mortality by SES. Our analysis of the 2000–2019 BRFSS revealed that across education groups there was an overall lower alcohol consumption in states and years with Sunday sales bans. However, compared to those with high education, individuals with low or medium education appear to be less likely to engage in more favorable consumption patterns (i.e., no past-month alcohol use, lower average daily drinking levels, no hazardous alcohol use) in the presence of Sunday alcohol sales bans. Consistent with these findings, the interrupted time-series analysis suggests an increase in alcohol-attributable mortality rates after Minnesota repealed its ban on Sunday alcohol sales in the second half of 2017. This increase in mortality rates were most pronounced in men and women with high education.

In line with prior research, we found that Sunday alcohol sales bans were associated with lower alcohol consumption in the US. The effectiveness of Sunday sales bans in reducing alcohol consumption was recently supported by a systematic review and meta-analysis, where the restriction of alcohol sales on one day a week was linked to a significant reduction in alcohol consumption levels by 3.6% (95% CI: −5.0%, −2.2%) (Kilian et al., 2023; see also Sherk et al., 2018). It further aligns with US studies evaluating the effect of repealing the Sunday sales ban on alcohol consumption. For example, the repeal of the Sunday sales ban in Delaware (2003), Pennsylvania (2003), and New Mexico (1995) was found to be associated with a significant increase in per capita consumption levels, although no significant change was observed in Massachusetts and Rhode Island who lifted their bans in 2003 and 2004, respectively (Yörük, 2014). The absence of changes in per capita consumption following the repeal of the ban in some states might be caused by the day-specific nature of the policy. Specifically, day-specific sales bans were found to impact the day-specific rather than the overall consumption of alcohol over longer periods (Carpenter & Eisenberg, 2009). This is mirrored by our study’s findings, which show limited support for Sunday sales bans being associated with the past-month prevalence of alcohol use in men. However, among current alcohol users, the presence of a Sunday alcohol sales ban was associated with lower daily drinking levels and lower odds of reporting hazardous alcohol use, which under consideration of the available evidence and in the absence of other potential explanatory variables, could be an indication of a possible effect of the Sunday sales ban policy.

Our Minnesota case study further adds to the existing evidence that repealing Sunday sales bans are not only associated with alcohol consumption but are also likely to increase alcohol-attributable mortality (see also Chisholm et al., 2018). This outcome derived from interrupted time-series analysis is strengthen by the absence of other alcohol policies at the time of the intervention possibly explaining the observed mortality increase (National Institute on Alcohol Abuse and Alcoholism (NIAAA), 2023). To the best of our knowledge, there are no other factors that could explain the observed increase in alcohol-attributable mortality in Minnesota in 2017, supporting our empirical finding.

The association between Sunday alcohol sales bans and alcohol consumption and mortality differed across socioeconomic groups defined by educational attainment. While there have been studies reporting socioeconomic differences for pricing policies (Manthey et al., 2023; Kilian et al., 2023; Wyper et al., 2023) and complex policy indicators (Silver et al., 2022), this study provides new insights into the possible differential effects of an alcohol availability restriction policy. Investigating the association between state-level alcohol policies and alcohol use using data from the 2011–2019 BRFSS, Silver and colleagues found that differences in any past-month alcohol use and heavy drinking (≥15/8 drinks per week in men/women) among current alcohol users narrowed between education groups with stricter state-level alcohol policies (Silver et al., 2022). For any past-month alcohol use, this mirrors our contradictory finding of men with low education drinking more when a Sunday sales ban was in place (or with stricter state-level alcohol policies) compared to those with high education and no sales ban (or less strict alcohol policies (Silver et al., 2022)). Importantly, any past-month alcohol use was lowest in the low education group. For any past-month hazardous alcohol use, we found a statistically significant greater education gap in the event of the stricter policy in women with low versus high education only. While this oppose the findings of Silver and colleagues, we acknowledge that there are notable differences in the definition of this drinking pattern and the policy variable under consideration, as well as differences in methodology. We therefore emphasize that these findings must first be corroborated by further research, including studies using other data.

Given the observed differential associations of Sunday sales bans with alcohol consumption and mortality across socioeconomic groups, the question arises as to what drives these differences. Prior research has found that bans on Sunday alcohol sales are associated day-specific alcohol use in particular, while not necessarily affecting overall consumption levels (Carpenter & Eisenberg, 2009; Yörük & Lee, 2018). Compared to weekday drinking, alcohol use is generally higher on Saturdays and Sundays (Liang & Chikritzhs, 2015). This within-week variation in consumption is mirrored by higher rates of alcohol-related deaths on weekend days, including Sundays (Udesen et al., 2023). Against this backdrop, it seems plausible that a ban on Sunday alcohol sales contributes to stopping the typically elevated drinking on weekends one day earlier. This hypothesis is supported by the observed changes in alcohol-attributable mortality: Descriptive explorations of the mortality trends suggest that increases in alcohol-related liver cirrhosis may drive the overall increase in mortality seen after the repeal of the Sunday sales ban in Minnesota. As alcoholic liver cirrhosis is linked to daily heavy alcohol use (Llamosas‐Falcón et al., 2023), the availability of alcoholic beverages on one additional day in the week may result in sizeable increases in liver cirrhosis mortality. Such day-to-day effects of policies regulating alcohol availability were also reported in a recently published evaluation study on reduced sales hours in Lithuania, where binge drinking related excess mortality decreased on weekends and Mondays following a significant reduction in alcohol sales hours (Stumbrys et al., 2023). It seems plausible that differences in day-specific alcohol consumption across education groups may explain the observed differential effectiveness of Sunday sales bans. However, empirical data that systematically compare within-week purchasing or consumption across socioeconomic groups is lacking though required to understand the underlying policy mechanism. Future research should therefore seek to explore the day-to-day impact of alcohol availability restrictions on alcohol purchases, consumption, and mortality in greater detail, accounting for differences across socioeconomic groups.

Several limitations need to be acknowledged. First, we used self-reported alcohol consumption data from the BRFSS, which is generally prone to underreporting and self-reporting biases (Stockwell et al., 2014). However, these distortions would have affected our results only if there were systematic differences in these biases across sociodemographic groups or across survey years, for which there is limited evidence (Kilian et al., 2020). Second, legislations banning Sunday sales differ between states and there was a reasonable number of states having local exemptions to their state-wide ban. We tested the potential impact of states with local options in a sensitivity analysis (Supplement S7), supporting the robustness of our findings. Third, we did not account for the complex survey design of the BRFSS in our main analysis, given that the Centers for Disease Control and Prevention changed the weighting procedure from post-stratification to raking in 2011. In a series of sensitivity analysis, we included the survey weights and found that our overall conclusions were largely confirmed (Supplement S6). Finally, no causal conclusions can be drawn from the results of study 1 given the chosen methodology. Alternative statistical approaches that allow for a more causal interpretation, such as difference-in-difference models, were considered initially but discarded given that critical model assumptions were not met. For the interrupted time-series analysis, we acknowledge that, given the profound changes in alcohol use caused by the COVID-19 pandemic in 2020, we examined a short post-intervention period (n = 9 data points), which prevented us from identifying long-term trends. Moreover, in the absence of suitable controls, we were unable to conduct a controlled interrupted time-series analysis that could account for unobserved confounders, providing stronger empirical evidence for intervention effects. Nevertheless, our methodological approach allows us to conclude, in the absence of other potential explanatory variables, that the observed change in the alcohol-attributable mortality in Minnesota is likely caused by the repeal of the Sunday alcohol sales ban in 2017.

Public Health Implications

Our study corroborates the available evidence on the effectiveness of alcohol availability restrictions by demonstrating that off-premises Sunday alcohol sales bans are associated with lower alcohol consumption, while the permission of such Sunday sales give rise to alcohol-attributable mortality (Chisholm et al., 2018; Kilian et al., 2023). These changes in alcohol consumption and mortality, however, differ by education group, indicating that not everyone benefits equally from the more restrictive policy. As we found those with high education to have the most pronounced policy-driven changes in alcohol mortality, availability restrictions appear to be inadequate for lowering the disproportionately high alcohol burden in low SES groups. To mitigate alcohol-attributable health inequalities, a more comprehensive alcohol policy strategy is required, including but not limited to pricing policies that were found to lower consumption levels among less affluent and high-risk alcohol users in particular (Kilian et al., 2023; Vandenberg & Sharma, 2016). Moreover, reducing the spatial availability of alcoholic beverages may also prove suitable, though research on their effectiveness across different socioeconomic groups is scarce (Gruenewald et al., 2023; Han et al., 2016). Eventually, to approach health equity in the long-term, root causes need to be addressed, including improved healthcare access and affordability, as well as the alleviation of economic inequalities.

Table 2.

Results from the interrupted time-series analysis: repeal of Sunday alcohol sales ban in Minnesota (07/01/2017), outcome: age-standardized 100% alcohol-attributable mortality rate by sex and education based on the National Vital Statistics System (2000–2019).

Sex	Men			Women
Education	Low	Medium	High	Low	Medium	High
Intercept	5.01 (4.82, 5.19)***	2.53 (2.11, 2.95)***	1.82 (1.71, 1.93)***	0.95 (0.71, 1.18)***	0.98 (0.77, 1.20)***	0.52 (0.36, 0.69)***
Linear trend	Non-linear1	0.03 (0.02, 0.04)***	–	0.03 (0.02, 0.03)***	0.01 (0.01, 0.02)***	0.004 (0.00, 0.01)
Level change	–	0.90 (0.07, 1.73)*	0.81 (0.27, 1.35) **	–	–	0.41 (0.12, 0.69)**
Slope change	−0.03 (−0.17, 0.12)	–	0.09 (−0.01, 0.19)	0.19 (0.12, 0.26)***	0.15 (0.09, 0.22)***	–
Unemployment rate2	–	−0.24 (−0.48, 0.00)	–	–	–	–
Seasonal adjustment	–	–	Yes	–	–	–
Adjusted R2	59.9%	50.4%	45.8%	74.8%	55.2%	23.3%

Notes.

^*p < .05

^**p < .01

^***p < .001.

95% confidence intervals indicated in brackets. The best fitting models are shown. Models without seasonal adjustment: simple linear regressions, models with seasonal adjustment: generalized additive models with smoothing term using cyclic cubic spline. There was no autocorrelation detected in any model.

¹non-linear time trend fitted with smooth term using cubic spline.

²z-standardized logarithm.