Introduction
The demand for alcohol is not just of interest in its own right, but also because alcohol consumption has important and significant societal costs through adverse effects on crime and health, for example [
1]. The traditional approach to the economic analysis of alcohol demand has been to consider the demand for total alcohol consumed in a given period [
2]. In taking this approach, it is assumed that the frequency (how often an individual drinks) and intensity of consumption (how much an individual drinks at each drinking occasion) have no bearing on the utility an individual receives. Frequency and intensity are in effect treated as perfect substitutes. This simplifies the analytical task but potentially hides important information about how individuals consume alcohol. The pattern by which individuals consume alcohol is an important factor determining health outcomes and risky health behaviors: evidence from the USA shows that binge drinkers are 14 times more likely to drink drive compared with non-binge drinkers [
3] and more generally the risk of death from acute alcohol-related illness has been found to linearly increase with frequency but exponentially increase with intensity [
4]. How individuals drink has a bearing on the societal costs of alcohol consumption. Evidence from Sweden, noting that binge drinkers are more likely to be heavy drinkers, found that “at least for health care costs, the cost is quite heavily concentrated in the heaviest drinking group” [
1]. Understanding how frequency and intensity decisions affect the overall quantity decision will allow greater understanding as to what influences an individual’s drinking behavior. A policy may have no effect on total quantity for example, but may have an effect on the frequency and intensity decisions that would be masked by solely assessing the quantity decision.
While the economic literature on alcohol has in general focused on modeling the quantity of alcohol demanded, a few studies have examined the determinants of frequency of consumption specifically for binge drinkers [
5,
6]. Naimi et al. [
7] have studied the socioeconomic factors associated with intensive binge drinking, finding age <35 years, male and less education amongst others as key risk factors. However, even fewer studies have examined both the frequency and intensity decisions together. Berggren and Sutton [
2] estimated a structural model of alcohol demand where frequency and intensity entered the budget constraint as a multiplicative term. The authors found that for spirit consumption in Sweden, frequency and intensity are indeed simultaneous sub-decisions of the overall quantity decision and that education and income are negatively associated with intensity but have no effect on frequency. Petrie et al. [
8] consider an alternate form of the problem, examining the determinants of the intensity frequency ratio. Rather than modeling the budget constraint as Berggren and Sutton [
2] and Petrie et al. [
8] define a multiplicative quadratic utility function. The results in Petrie et al. [
8] are consistent with Berggren and Sutton [
2] in that they find a negative relation of the intensity frequency ratio with education. A consequence of the assumed form of the utility function the authors made is that the intensity frequency ratio is related to neither price changes nor income differences. Given the importance of the budget constraint in defining an individual’s choice set, it is undesirable to assume a utility function that yields this result. More recent evidence from Australia has found a negative association between the price of alcohol and the number of days in which alcohol is consumed lightly, yet found no association between price changes and the number of days of high-intensity drinking [
9].
While the research by Berggren and Sutton [
2] was pioneering in breaking apart the quantity decision into its constituent parts of frequency and intensity, some empirical issues remained. Prices of alcohol were not included (because the data had no time element), which may be a key component of any demand analysis. The data requirements are also quite demanding when estimating, as the authors did, a structural model. A structural model of the form estimated by Berggren and Sutton [
2] for frequency and intensity requires instruments for frequency and for intensity. However, the choice of instruments in Berggren and Sutton [
2] are debatable. A priori, it is quite hard to think of many variables that predict the frequency decision and not the intensity decision (and vice versa) when so little is known about how these decisions are made.
The literature on the demand for frequency and intensity of alcohol consumption is under-researched even though it has been shown to be important to consider both frequency and intensity separately. We address this evidence gap in this paper by utilizing new data from Sweden that allows the consideration of the determinants of the demand for frequency and intensity across three particular alcohol types: beer, wine, and spirits. We split the analysis by alcohol type because different individuals drink different alcohol types and there may be systematic differences in the characteristics of these individuals. Importantly, this new data also allows the frequency and intensity decisions to be compared between all of those who drink and the subset who are binge drinkers giving new insights into the binge drinking decision. The dataset is large and allows further breakdown by gender, which is useful because there are important biological differences in how much alcohol women and men can tolerate and also preferences may differ in important ways across the genders. This level of detail ensures not only a better level of understanding of the socioeconomic factors related to alcohol drinking patterns but also yields results that should be of greater relevance internationally. The analysis of a female wine binge drinker in Sweden is more likely to yield results of relevance to other female wine binge drinkers from other countries than results not split by alcohol type, gender, and drinking style participation preference. The paper unfolds as follows: Sect.
2 first presents the data material and following this presents the estimation strategy, Sect.
3 reports the results and Sect.
4 discusses the results and concludes.
Results
The results are estimated for males and females separately due to the very different patterns of alcohol consumption observed between the genders. The regression results are presented for males, and where differences are observed between the genders, these are discussed (results for females are found in Appendix
1). In general, the regression results show that between the years 2004 and 2011 there has been a downward trend in drinking participation, participation in binge drinking, frequency, and intensity of drinking episodes as estimated by a linear trend function (controlling for covariates). More men than women drank alcohol, and on average men and women had similar wine drinking patterns, but men drank more beer and spirits and were more likely to binge drink (see Table
1).
The results of the alcohol participation decision are presented in Appendix
1 (Table
5 for males and Table
7 for females) and are presented as average partial effects. The results show that income and education were positively correlated with the binge drinking decision across all types of alcohol, but less so in comparison to all drinkers. The highest income group was 27 % more likely to participate in beer consumption in comparison to the lowest income group. This likelihood is slightly higher for wine participation but about half as large for binge drinkers. A much greater distinction between binge drinkers and all drinkers is observed for the covariates of both age and economic activity where being younger and economically inactive are both stronger predictors for binge drinking compared to all drinkers. All covariates are fairly similar in magnitude across the alcohol types for binge drinkers, suggesting binge drinkers distinguished to a lesser extent between the alcohol types during the decision process to binge drink. Women and men had statistically different values for the explanatory variables in the wine participation equation, but these were fairly similar in size in a broader economic sense. Women had much shallower income and education gradients for beer and spirits consumption participation. Turning to price, the excise duty rate increase in January 2008, as assessed by interrupted time-series, had a negative impact on participation across the alcohol types (varying between −1 and −2 % depending on alcohol type and drinking group), with the exception of wine, which consistently saw no significant impact by gender and drinking group. The additional regressions for participation that include price, but not the excise duty rate dummy variable (see Tables
6 and
8 in Appendix
1), find generally positive but largely insignificant own price elasticities for all alcohol types and drinker types and the effect sizes are very large, ranging between −9 and 37 % for a 1 % change in price levels (controlling for a linear trend).
The results for the demand for frequency and intensity
conditional on participation are shown in Tables
2–
4 (Tables
10–
12 in Appendix
1 for females). The fitted values of the inverse Mills ratio (IMR) have a positive and statistically significant effect on the beer and wine frequency decision but for men only. Sample selection appears to be a bigger concern for binge drinkers with much larger IMR values (again only for males). The IMR values for binge drinkers suggest that those that select to be binge drinkers have a higher frequency of beer and spirits consumption, lower beer intensity consumption, and a higher spirits intensity consumption compared to a random draw from the population. As shown in the participation equations [Appendix, Tables
5,
6 (males) and
7,
8 (females)], there are mixed successes in the Probit model’s ability to predict extreme low and high probabilities, especially for women. Therefore, where the IMR is not significant (in Tables
2–
4), this does not necessarily suggest that unobserved heterogeneity is not an issue. It is possible that there remains unobserved heterogeneity that is correlated with the errors due to sample selection for the equations where the IMR is non-significant.
Table 2
Beer frequency and intensity demand equation estimates, males
Linear time trend | −0.01 | −0.00 | −0.02* | 0.01 |
(0.00) | (0.00) | (0.01) | (0.01) |
Alcohol duty change 08 | −0.05** | 0.04*** | −0.07** | 0.07*** |
(0.02) | (0.01) | (0.03) | (0.02) |
INCOME1 (reference) |
INCOME2 | 0.07 | 0.00 | 0.12* | −0.06 |
(0.04) | (0.03) | (0.06) | (0.05) |
INCOME3 | 0.23*** | −0.03 | 0.28*** | −0.14** |
(0.07) | (0.06) | (0.09) | (0.07) |
INCOME4 | 0.37*** | −0.02 | 0.46*** | −0.23** |
(0.11) | (0.08) | (0.14) | (0.10) |
INCOME5 | 0.47*** | −0.08 | 0.56*** | −0.36*** |
(0.12) | (0.09) | (0.16) | (0.12) |
INCOME6 | 0.48*** | −0.16* | 0.57*** | −0.47*** |
(0.13) | (0.09) | (0.15) | (0.12) |
COMPULSORY SCHOOL (reference) |
COLLEGE | 0.07*** | −0.01 | 0.15*** | −0.11*** |
(0.03) | (0.02) | (0.05) | (0.04) |
UNIVERSITY | −0.02 | −0.14*** | 0.01 | −0.18*** |
(0.02) | (0.01) | (0.02) | (0.02) |
LNAGE | −0.15* | −0.83*** | −0.54** | −0.25 |
(0.08) | (0.06) | (0.24) | (0.18) |
EMP (reference) |
INACTIVE | −0.12** | −0.17*** | −0.29** | 0.04 |
(0.05) | (0.04) | (0.12) | (0.09) |
UNEMP | 0.05 | 0.08*** | 0.03 | 0.09** |
(0.03) | (0.02) | (0.05) | (0.04) |
STUDENT | −0.02 | −0.10*** | −0.07 | −0.03 |
(0.03) | (0.02) | (0.05) | (0.04) |
COHABIT | −0.02 | −0.20*** | −0.11*** | −0.11*** |
(0.01) | (0.01) | (0.03) | (0.03) |
IMR | 0.58** | −0.18 | 0.97*** | −0.71*** |
(0.25) | (0.19) | (0.34) | (0.26) |
CONSTANT | 1.25*** | 4.86*** | 2.24*** | 3.59*** |
(0.08) | (0.06) | (0.48) | (0.36) |
Participation observations | 59,251 | 59,251 | 59,251 | 59,251 |
Freq/intens observations | 35,650 | 35,650 | 19,297 | 19,297 |
Proportion drink/binge | 60 % | 60 % | 33 % | 33 % |
Table 3
Wine frequency and intensity demand equation estimates, males
Linear time trend | −0.02*** | −0.01*** | 0.04** | −0.01 |
(0.01) | (0.00) | (0.02) | (0.01) |
Alcohol duty change 08 | −0.01 | 0.02* | 0.03 | 0.01 |
(0.02) | (0.01) | (0.06) | (0.02) |
INCOME1 (reference) |
INCOME2 | 0.05 | 0.00 | −0.27** | −0.03 |
(0.05) | (0.03) | (0.14) | (0.07) |
INCOME3 | 0.29*** | 0.03 | −0.52** | −0.02 |
(0.06) | (0.03) | (0.23) | (0.12) |
INCOME4 | 0.52*** | 0.04 | −0.83** | −0.01 |
(0.08) | (0.05) | (0.39) | (0.20) |
INCOME5 | 0.82*** | 0.05 | −0.96* | 0.01 |
(0.11) | (0.06) | (0.51) | (0.26) |
INCOME6 | 1.06*** | 0.05 | −0.81 | 0.02 |
(0.12) | (0.07) | (0.52) | (0.27) |
COMPULSORY SCHOOL (reference) | |
COLLEGE | 0.65*** | −0.00 | −0.14 | 0.08 |
(0.06) | (0.04) | (0.21) | (0.11) |
UNIVERSITY | 0.96*** | −0.04 | 1.37*** | 0.11 |
(0.05) | (0.03) | (0.28) | (0.14) |
LNAGE | 0.14*** | −0.14*** | 0.74*** | −0.12 |
(0.02) | (0.01) | (0.26) | (0.14) |
EMP (reference) | | | | |
INACTIVE | 0.08* | 0.08*** | −0.10 | 0.08** |
(0.04) | (0.02) | (0.09) | (0.04) |
UNEMP | 0.42*** | −0.04 | −0.06 | −0.03 |
(0.06) | (0.03) | (0.11) | (0.05) |
STUDENT | 0.28*** | −0.07*** | 0.18*** | −0.08*** |
(0.03) | (0.02) | (0.03) | (0.01) |
COHABIT | −0.11*** | 0.03 | 0.30* | 0.01 |
(0.03) | (0.02) | (0.16) | (0.08) |
IMR | 1.04*** | −0.07 | −2.08** | 0.14 |
(0.17) | (0.10) | (0.92) | (0.48) |
CONSTANT | −3.98*** | 1.58*** | −0.95** | 0.94*** |
(0.41) | (0.24) | (0.42) | (0.21) |
Participation observations | 59,251 | 59,251 | 59,251 | 59,251 |
Freq/intens observations | 31,925 | 31,925 | 14,414 | 14,414 |
Proportion drink/binge | 54 % | 54 % | 24 % | 24 % |
Table 4
Spirits frequency and intensity demand equation estimates, males
Linear time trend | −0.01 | −0.02** | −0.07*** | −0.07*** |
(0.01) | (0.01) | (0.03) | (0.02) |
Alcohol duty change 08 | −0.06** | 0.01 | −0.25*** | −0.12 |
(0.02) | (0.02) | (0.09) | (0.08) |
INCOME1 (reference) |
INCOME2 | −0.06 | −0.00 | 0.29* | 0.27** |
(0.06) | (0.05) | (0.16) | (0.13) |
INCOME3 | −0.00 | 0.07 | 0.64** | 0.47** |
(0.12) | (0.10) | (0.25) | (0.21) |
INCOME4 | 0.06 | 0.12 | 1.08*** | 0.77** |
(0.17) | (0.15) | (0.39) | (0.32) |
INCOME5 | 0.12 | 0.12 | 1.45*** | 0.96** |
(0.22) | (0.19) | (0.49) | (0.40) |
INCOME6 | 0.18 | 0.09 | 1.55*** | 0.92** |
(0.24) | (0.20) | (0.50) | (0.41) |
COMPULSORY SCHOOL (reference) | |
COLLEGE | −0.00 | −0.28*** | 0.05 | −0.33*** |
(0.03) | (0.03) | (0.06) | (0.05) |
UNIVERSITY | 0.26*** | −0.43*** | −1.53** | −1.75*** |
(0.04) | (0.03) | (0.64) | (0.53) |
LNAGE | 0.18*** | 0.01 | −0.41 | −0.42** |
(0.02) | (0.02) | (0.25) | (0.21) |
EMP (reference) |
INACTIVE | 0.10** | 0.13*** | 0.11 | 0.11 |
(0.04) | (0.03) | (0.12) | (0.10) |
UNEMP | 0.11** | −0.04 | −0.04 | −0.22** |
(0.06) | (0.05) | (0.12) | (0.10) |
STUDENT | −0.05* | −0.12*** | −0.26*** | −0.31*** |
(0.02) | (0.02) | (0.09) | (0.07) |
COHABIT | −0.05 | −0.07 | −0.65*** | −0.44** |
(0.09) | (0.08) | (0.21) | (0.17) |
IMR | −0.05 | 0.38 | 3.14*** | 2.57*** |
(0.50) | (0.43) | (1.09) | −0.89 |
CONSTANT | 0.07 | 2.76*** | 2.93*** | 5.12*** |
(0.35) | (0.30) | (1.06) | (0.87) |
Participation observations | 59,251 | 59,251 | 59,251 | 59,251 |
Freq/intens observations | 31,243 | 31,243 | 16,641 | 16,641 |
Proportion drink/binge | 53 % | 53 % | 28 % | 28 % |
The impact of the alcohol duty rate changes in 2008 reduced the frequency of beer consumption by 5 % but raised the intensity of drinking episodes by 4 %. An impact is also observed for beer binge drinkers where frequency reduced by 7 % but intensity increased by 7 %. It appears that the excise duty change for beer also either had an income effect and subsequent substitution effect or complementary effect as it negatively impacted on the demand for frequency and intensity of spirits. The impact of own prices on frequency and intensity appears to be negative for frequency (see Table
9 in Appendix
1). Significant and large impacts of own prices are observed for frequency of binge drinkers of wine and beer for males but not for females. Similarly to the participation equations, the impacts of own prices are very large in comparison to the impacts of the alcohol duty changes and other covariates.
Broadly, across the alcohol types there was a positive income gradient to frequency demand and a negative income gradient to intensity demand. The highest income group had a 48 % higher frequency of beer consumption and 16 % lower intensity of beer consumption in comparison to the lowest income group. Binge beer drinkers had a steeper income gradient in both frequency and intensity. The income pattern for spirits drinkers was similar to that of beer consumption with regards to frequency. Unlike beer, spirits had a positive income gradient in intensity. Higher education levels predict lower levels of intensity for beer and spirits consumption, but higher levels of frequency of beer and wine consumption. Women differ slightly from this pattern in that there is no clear relationship between income, education and frequency or intensity for drinkers of beer and spirits.
Most often, where a positive gradient for income and education is observed for the quantity equation (see Appendix, Tables
14,
15,
16,
17), this is driven by the positive income and education gradients in frequency demand dominating the negative income and education gradients in intensity demand and vice versa. Where no statistically significant relation is observed for the quantity equation, this can obscure important responses. For example, the income response of beer binge drinkers shows a positive income gradient for frequency and negative gradient for intensity but no significant effect for quantity.
Discussion
This paper has broadened the evidence base regarding the determinants of frequency and intensity demand for alcohol using a large individual level dataset from Sweden that has allowed the analysis to be extended to different alcohol types, drinker types, and to be split by gender. The time period under analysis is interesting, as it is a period where a reduction in drinkers, binge drinkers, quantity, frequency, and intensity of alcohol consumption has been observed. These trends are important to note as they involved large reductions in consumption generally during a period of relatively small changes in the real prices of alcohol. Real (inflation-adjusted) prices of beer, wine, and spirits, in general, fell at the same time that the proportion of the population who drank and the intensity in which they consumed fell. With the entry to the EU, Sweden saw increasing liberalization of alcohol trade with other EU members and alcohol private import quotas were fully abolished by 2004. Alcohol consumption had been increasing up until 2004, but since 2004 overall alcohol consumption has been on the decline [
12]. Our results generally find large positive but largely insignificant own price elasticities. The participation price elasticities are significant for binge drinkers of wine and spirits, but this cannot be a realistic result. More likely is that preferences for alcohol by binge drinkers has changed in Sweden in a way that cannot adequately be explained by changes in price, controlling for other observables (including a linear time trend to capture the general decline in consumption patterns), and that we have not been able to explain these changing preferences adequately.
We now turn to the impact of the change in excise duty and its analysis using interrupted time-series analysis. It appears that the alcohol excise duty increase enacted in January of 2008 did reduce beer drinking participation and frequency of consumption but raised the intensity. It also appears to have reduced both intensity and frequency of wine and spirits consumption. The results of the excise duty rate change on beer consumption are similar to those of Byrnes et al. [
9] who found a negative price elasticity of frequency of light drinking days. However, we find that the impact on binge drinkers was similar to that for all drinkers, but stronger for males, and Byrnes et al. [
9] found no impact on frequency of heavy drinking days. Why there was a positive intensity response to the excise rate change is not clear. The price response to the excise duty change observed here suggests that individuals respond to increased prices by drinking less often but substitute somewhat by increasing how intensely they drink. We also observe that the excise rate change for beer had no measurable effect on quantity consumed, yet it increased the intensity of consumption, highlighting the importance of frequency and intensity analysis. It is important to note that interrupted time-series analysis attempts to capture the effect of a time-specific policy, but due to a lack of a control group, it is possible that other events occurred at the same time as the change in excise rates that we are unable to control for.
Previous analysis of the determinants of frequency and intensity of alcohol consumption [
2] concluded that income and education are negatively associated with intensity and that neither had an effect on frequency. The results of the current paper find that generally across different alcohol types and different types of drinkers there was a positive income and education gradient with frequency and a negative income and education gradient with intensity. Broadly, we find income and education had a positive gradient with drinking and binge drinking participation, quantity and frequency demanded, but a negative gradient with intensity demanded. The results presented here suggest the reason behind the positive income gradient with alcohol demand documented in Cook and Moore [
18] is due to a frequency relation. A potential explanation for the observed positive relationship between income and drinking frequency could be that higher-earning individuals can afford to socialize more often and socializing norms involve the consumption of alcohol.
The key difference between men’s and women’s drinking patterns is that women predominantly drank wine. Relatively few women drank beer or spirits (in grams of alcohol terms) whereas men were more evenly split amongst the alcohol types. Wine is very different to beer and spirits in terms of who drinks it and who drinks it more frequently and intensely. Income and education were positively associated with both frequency and intensity demand for wine. Wine appears to be a luxury good, favored by higher earners and the more educated, especially women.
The results for the sub-group of binge drinkers show important differences to the population of drinkers as a whole. Participation in binge drinking is strongly predicted by young age and economic inactivity. Binge drinkers appeared to differentiate between the alcohol types to a lesser degree in comparison to average drinkers, possibly reflecting a different attitude to alcohol. Across all alcohol types the more educated bingers drank less intensely, possibly reflecting the increased opportunity cost of intense drinking episodes and/or greater health awareness, but they drank more often.
As set out in the introduction, binge drinkers are associated with higher social costs. The results presented here highlight the complexities associated with attempting to limit the social costs of this harmful drinking behavior. An alcohol-related policy targeted at the less educated would, for example, have to understand why the less educated are less likely to binge drink but the intensity of less educated individuals who binge drink is higher. Policy aimed at reducing socioeconomic-related health inequalities needs to consider the particular complexities of alcohol demand highlighted in this study. Previous research investigating various socioeconomic measures related to alcohol participation inequality in Sweden by Combes et al. [
20] has found alcohol participation to be positively associated with income. This is consistent with the findings of the current paper. However, the demand for intensity by binge drinkers is negatively associated with education and income. Given that binge drinkers who drink most intensely are the individuals who will have the worse wider health outcomes, the findings of this paper suggest it is in fact individuals with low education and low income levels who drink more intensely that should be the focus of policy, contrary to Combes et al [
20]. In fact, a newly designed randomized trial in the UK [
21] is focusing exactly on this group, highlighting how the results presented here could help focus policy interventions.
The results presented in this paper are robust associations and are useful for highlighting which groups of individuals drink in the most harmful way and where alcohol policy could most effectively be targeted in order to reduce alcohol-related harm and its negative impact on health. However, the results are not free from endogeneity or simultaneity. It therefore cannot be said that policy aimed at changing the factors observed in this study, such as education levels, will change alcohol drinking behavior as might be expected if the results were interpreted as causal effects. Addressing the potential endogeneity and two-way causality issues that potentially exist in the analysis presented here is a valuable line of investigation for future research.
The results of this paper are subject to a few further limitations that impact the potential interpretation of the results. While the data used in this paper are very detailed and have enabled us to examine in more detail the demand for frequency and intensity, the main limitation of the paper is the quality of the data we have used. This reflects the challenging nature of alcohol demand analysis. As noted in the data description, the Monitor project data saw a reduction in response rate with a response rate near the end of the survey period in the range of 35–45 %. There is also evidence of missing values bias where those who report an income are more educated, male, younger, and more likely to be employed and those who fail to report their drinking behaviors are younger, with lower income and who are male. The Monitor project also made no effort to counter the fact that heavy drinkers are less likely to respond to survey questionnaires. There has been research looking into the response rate of the survey data and this found no significant issues, but nonetheless, a response rate of 35 % will always be a concern, and the results found in this paper could just reflect a subset of the population who are willing to respond to such a survey. It is also concerning that missing values can be predicted by covariates, and beyond controlling for these covariates the results have to be interpreted with caution, as there may be other factors that both predict alcohol drinking behavior and non-response to a particular question. Finally, our definition of binge drinker is not the same as the more internationally recognized definition used in the AUDIT survey. This may potentially explain differences between the results found here and those found elsewhere, especially as the definition used in this paper does not distinguish between genders and the differences we observe between genders could be driven by our definition.