Determinants of female alcoholism in Europe: multilevel approach

Размещено на http://www.allbest.ru/

1/34

Determinants of female alcoholism in Europe: multilevel approach

Introduction

World Health Organization (WHO) names drinking abuse as the third widespread (after cardiovascular and oncological diseases) cause of death in the world population and also as a factor that puts a serious strain on general well-being of the society (WHO, 2014). Levels of ethanol intake correlate positively with relative risk of developing 15 alcohol-attributed diseases and incidents, including cirrhosis, cancer of digestive system, essential hypertension, chronic pancreatitis, heart attacks, injuries and violence (Corrao, Bagnardi, Zambon, & La Vecchia, 2004). Some of them, such as cardiovascular diseases, correlate even with lowest levels of intake (Shield et al., 2013). This statistics makes drinking behavior an extremely important research topic, especially in Europe, which has the highest rates of alcohol abuse in the world (WHO, 2016).

In this region, difference in patterns of alcohol use accounts for 10-30% difference in life expectancy between males and females (Leon, 2011). The highest difference is observed in eastern countries, including Russia and former communist states, where hazardous drinking is most widespread (McKee & Shkolnikov, 2001). It seems thatfemales are not prone to excessive alcohol use; males outnumber them in drinking literally in every culture (Ahlstrцm, Bloomfield, & Knibbe, 2001; Allamani, Voller, Kubicka, & Bloomfield, 2000; Dawson, 1993; Marsh & Kibby, 1992; Jьrgen Rehm et al., 2003), and no reverse trend has been observed yet (Holmila & Raitasalo, 2005).

To some extent, this difference may be explained by biological factors, such as sex-specific differences in ethanol metabolism (Gandhi, Aweeka, Greenblatt, & Blaschke, 2004; Baraona, E., Abittan, C. S., Dohmen, K., Moretti, M., Pozzato, G., Chayes, Z. W., & Lieber, C. S, 2001; Ammon, Schдfer, Hofmann, & Klotz, 1996; Niaura, Nathan, Frankenstein, Shapiro, & Brick, 1987).

Nevertheless, if difference in pharmacokinetics is the only factor that defines levels of ethanol intake among men and women, levels of female alcohol consumption or at least gender difference in alcohol use would be roughly the same across European countries. Against this expectation, levels of alcohol consumption vary both between sexes and within sexes across Europe. Depending on the country, females drink from 2 to 14 grams of pure ethanol per day (Kim et al., 2005; see Figure 1), and males consume from 20 to 140% more alcohol than women (Mдkelд, Gmel, Grittner, Kuendig, Kuntsche, Bloomfield, & Room, 2006).

Figure 1. Graph built on data from Gender, Alcohol and Culture: an International Study

Because of this statistics, female drinking is not evaluated as a socially significant problem in European countries. Nevertheless, non-acceptance of female drinking may lead to underestimation of females' alcohol abuse and imbalance in national alcohol control policy. While heavy drinkers are more often males than females (Jensen, Andersen, Sшrensen, Becker, Thorsen, & Grшnbжk, 2002), alcohol-related harm is usually more pronounced for females. For instance, rates of cognitive recovery after alcohol intake are lower for them (Niaura et al., 1987; Mahalik et al., 2015). Moreover, females' binge drinking is usually characterized by more drastic social conditions, as it interacts with lower education andhigher social inequality (Grittner, Kuntsche, Gmel, & Bloomfield, 2012).

So far, international surveys have grasped some cultural determinants of females' alcohol use (Bloomfield, Gmel, & Wilsnack 2006; Room & Mдkelд 2000; Room 2010), but comprehensive description of drinking cultures is still to be explored. There is a call for a systematical description that would join gender determinants of alcohol use into a single scheme (Bloomfield, Stockwell, Gmel, & Rehn, 2003; Room, 2001; Robin Room, 2001; Wilsnack, Vogeltanz, Wilsnack, & Harris, 2000).

To fill this gap, I conduct multilevel analysis of drinking patterns with special reference to gender-specific behaviour. Research question is as follows: “Which country- and individual-level factors correspond with larger ethanol intake among females?” The answer to this question would shed some light on the mechanisms of social control on females' behavior. With this paper we contribute to the analysis of drinking patterns in Europe and explore correlations between female drinking and country-level parameters.

The data used for this paper is from Gender, Alcohol and Culture: an International Study (GENACIS), and is not available in public access. It was provided by special agreement from Dr. Sharon Wilsnack (University of North Dakota, USA), director of the GENACIS project, and project coordinators: Dr. Kim Bloomfield (Aarhus University, Denmark) and Dr. Thomas Greenfield (Public Health Institute, USA). We thank GENACIS data coordinators and our supervisors, Dr. Gerhard Gmel (Addiction Switzerland Research Institute, Switzerland) and Dr. Sandra Kuntsche (La Trobe University AER Centre for Alcohol Policy Research, Australia) for the help with data proceeding and general comments on the paper.

1. Defining female alcoholism and its determinants

1.1 The concept of female alcoholism

Excessive drinking is an important characteristic of alcohol consumption and significant predictor of acute consequences of alcohol use (Kuntsche, Rehm, & Gmel, 2004). Nevertheless, despite the significant meaning of binge drinking in statistical surveys on alcohol consumption, there is no commonly accepted definition of the term. Some scholars define this concept as risky single occasion drinking (RSOD) and rest upon some quantitative limits - such as five drinks per occasion (Hanson & Engs, 1992) or even eleven (Anderson & Plant, 1996) - which may seem arbitrary. Moreover, these definitions are introduced regardless of gender difference and hence may be oversimplified. In other classifications, RSOD is defined as five or more drinks per occasion for men and four or more drinks per occasion for women (Wechsler et al., 1994). However, the whole idea of defining binge drinking as risky single episodes is debatable. Gmel et al. (2011) note that regular excessive drinking contributes to acute consequences of alcohol use not less than RSOD. Bondy (1996) also points that frequency and volumes of lower doses of ethanol are also an important predictor of alcohol-related harm: in Southern Europe, where alcohol is considered to be an essential part of daily life, volumes of ethanol consumption are much higher than in Nordic states, which indicates higher risk of alcohol-related damage. Other authors demonstrate thathigh volumes of ethanol intake may be reached either by regular intake or by drinking large amounts of alcohol occasionally, and both measures should be used in studies on alcohol-related harm (Dawson, 2011; Gual, 2011).

Additionally, smaller amounts of alcohol and shorter periods of drinking have more drastic effect on females than on males, so, thresholds of binge drinking for females will be lower (Beccaria & Guidoni, 2002). In GENACIS (Kim et al., 2005), median and modal annual frequency of binge drinking equals 0 days in the majority of European countries (see Table 1).

Table 1. Annual frequency of RSOD among females (in days)

Those countries where median annual frequency exceeds zero demonstrate very different patterns of females' alcohol use, especially with reference to males' alcohol consumption (see Figure 2). Though in all these countries proportions of male and female drinkers are higher than European average, the ratio of male drinkers to that of females is quite different.

It seems that binge drinking the measure that works perfectly well for males but is absolutely not applicable indicator of female alcoholism, because there is no country in Europe where female would score high on it.

Figure 2. Gender patterns of ethanol intake (from GENACIS data)

From what was stated above, one may conclude that exhausting and commonly accepted definition of female alcoholism is still lacking. Moreover, as Table 1 indicates, the existing distribution of RSOD frequency among females is skewed to such extent that conventional statistics as average, mode etc. become inapplicable. For this reason, I further focused on factors which increase volumes of individual alcohol intake rather than drive it to any arbitrarily assigned threshold.

1.2 Determinants of females' drinking

It has been shown that, to some extent, gender patterns of alcohol use can be explained by biological factors. Numerous surveys on both model animals (such as primates, see Green, K. L., Szeliga, K. T., Bowen, C. A., Kautz, M. A., Azarov, A. V., & Grant, K. A., 1999) and humans (Baraona, E., Abittan, C. S., Dohmen, K., Moretti, M., Pozzato, G., Chayes, Z. W., & Lieber, C. S, 2001; Ammon, Schдfer, Hofmann, & Klotz, 1996; Niaura, Nathan, Frankenstein, Shapiro, & Brick, 1987) indicated that females exhibited faster elimination of alcohol compared with males. However, data on differences in peak alcohol concentrations in blood of females vs. males are controversial. For example, Baraona et al. (2001) found that the same amount of consumed alcohol leads to higher blood alcohol concentration (BAC) in females and explained it with slower gastric metabolism. Oppositely, no differences in peak concentrations between different genders were found by Green et al. (1999, also see papers cited therein). As most of ethanol is converted mainly by hepatic enzymes, some authors explain gender differences in ethanol dynamics by hormone-dependent (and hence gender-specific) activity of these enzymes (Gustafsson, Mode, Norstedt, & Skett, 1983). Such parameters as rate of gastric emptying, body weight index, excretion activity etc. also contribute to ethanol elimination (and overall pharmacokinetics) and exhibits significant gender differences(Gandhi, Aweeka, Greenblatt, & Blaschke, 2004).

Nevertheless, as it was mentioned above, biological factors cannot explain cross-country variation in gender patterns of alcohol use.Many authors suggest that this variation may be defined by cultural factors: type of mostly preferred beverage, rigidity of social control over drinking, alcohol-related expectations and so on (Bloomfield, Gmel, & Wilsnack 2006; Room & Mдkelд 2000; Room 2010). In these papers, gender patterns of alcohol use are analyzed not only as a significant indicator of national health and well-being, but also as a reflection of drinking culture and the system of gender roles in the society.

From sociological point of view, gender patterns of alcohol use are a part of broader system of social norms and values. In terms of phenomenological sociology, commonly recognized norms of conduct become implicit in individuals' attitudes in the result of socialization and interirization, and members of the same society have similar beliefs about deviations and socially approved practices (Schutz, 1967). In this chapter, I will analyze why society encourages males and females to behave differently and how cultural context defines gender patterns in individuals' drinking behavior. I will also review individual-level determinants of females' alcohol use.

1.3 Individual-level predictors

Series of surveys reveal that drinking pattern depends on socio-demographic characteristics and their interactions with gender. ESS reports indicate that people with higher education drink alcohol more often but tend to intake smaller amounts of ethanol per occasion (Grittner et al., 2012). Ahlstrцm et al. (2001) revealed that marriage and parenthood reduce males' consumption in countries where men are more involved in unpaid housework (such as Finland), and consumption ratio is lower in such states. Authors also show that unemployment affects both genders, but seems to have more adverse impact on females (ibid.).

Several studies have registered an interaction between gender, education status and social inequality. It has been shown that drinking in the population correlates positively with high GDP and HDI (Rahav et al. 2006). However, there is evidence that socio-economic status affects males and females in different way (Grittner et al., 2012). For males, the association between welfare and drinking was mostly negative. But for women the direction of effect depended on general well-being of the country: in economically developed countries female drinking correlated with higher socio-economic status, while in low income states the most risky group are women with adverse financial position (ibid.).

Some authors suggest that the main reason for females' moderate alcohol use is fertility concern. Sharma, Biedenharn, Fedor, & Agarwal (2013) show thatthe risk of infertility makes women opt for healthier lifestyle and moderation.

Other authors suggest that females are less prone to alcoholism due to psychological features, such as nurturance, disinclination to aggression and destructive behavior(Byrnes et al., 1999; Nolen-Hoeksema, 2004). Risky behaviour does not belong to classical feminine traits (Schur, 1984; Umberson, 1992), but, at the same time,constututes an essential part of masculinity. Meta-analysis of 150 gender studies shows that in almost every sphere of life, including drinking, risky behavior is more typical for males (Mahalik et al., 2015).It has also been registered that drinking behavior corresponds more with gender identity rather than with biological sex. `Male-typical' women consume more alcohol than their canonically feminine peers (Mahalik et al., 2015); males with inequitable gender attitudes tend to drink more (ICRW & Promundo, 2015).

These commonly held stereotypes on alcohol use are interiorized during the process of socialization, which can be divided into two phases. The first socialization venue is family; parental control and parents' drinking define patterns of alcohol use to great extent. According to the longitude survey of Dutch teenagers, rigid parental control lowers levels of alcohol consumption, but attachment between parents and children does not influence adolescents' drinking (van der Vorst, Engels, Meeus, & Dekoviж, 2006). Parents' alcohol abuse increases chances of early onset of drinking (Hill, Shen, Lowers, & Locke, 2000) and probability of heavy drinking among adolescents (Barnes, 1990; Barnes, Farrell, Cairns, & Farrell, 1986). During further socialization, ethanol intake is often an important part coming-of-age for males, ritual of transition into adulthood (Cooper & Huselid, 2017; Iwamoto & Smiler, 2013). Masculinity stereotypes and peers' pressure are significant factors of drinking among male adolescents (ICRW & Promundo, 2015). Female teenagers, on the contrary, regard drinking as socially prohibited practice (Chassin, Tetzloff, & Hershey, 1985).

In the result of socialization, alcohol-related expectations also differ by gender. For males, drinking is more often associated with sensation-seeking and deviant behavior (Bergmark & Kuendig, 2008) and males' expectations towards drinking are, by and large, positive (Rolfe, Orford, & Dalton, 2009). According to the survey of British binge-drinking females (Rolfe et al., 2009), for women alcohol serves quite a different task: it is more a coping mechanism than a relaxing activity and is associated with adverse living conditions.

Alcohol-related expectations mediate the relation between volumes of consumption and the severity of alcohol-related consequences (Simpura and Karlsson 2001). Study conducted by MCM research group reveals that drunken behavior often depends on commonly held expectations on the effect of alcohol on demeanor (Marsh & Kibby, 1992). Marsh and Kibby (1992) report that in Scandinavian region, Great Britain, United States and Australia drunken individuals can attribute their anti-social behavior to the effect of beverage, not to their usual tendency to violence and aggression. In Mediterranean countries, however, intoxication is not associated with destructive behavior, and substance abuse proceeds less harmfully.

1.4 Country-level predictors of female drinking

Gender stereotypes about alcohol use are nested within attitudes to drinking in general. In societies where alcohol is not well-integrated into everyday life and drinking is regarded as deviation, female drinking is a stigma, while in countries where drinking is a morally neutral everyday activity, social control on females' alcohol use should be less tough. The most popular classification of drinking cultures is the theory of `dry' and `wet' regions (Rehm, Ashley, Room, Single, Bondy, Ferrence, & Giesbrecht, 1996), or, stated in another way, `Mediterranean' and `Nordic' states (Marsh & Kibby, 1992). This theory applies several factors that constitute drinking behavior and commonly held attitudes towards drinking: the proportion of drinkers in the population; type of mostly preferred beverage; frequency and volumes of drinking; alcohol-related expectations. In Nordic countries, the rate of drinkers is low, but those who drink alcohol usually opt for spirits, are prone to binge-drinking and demonstrate high intoxication rates often followed by destructive behavior (Rahav et al., 2006). Therefore, in `dry' societies alcohol is regarded as a dangerous disinhibitor and definitely not a feminine attribute. In Mediterranean countries, on the contrary, alcohol is well-integrated into daily life; the most popular pattern here is moderate wine drinking. Alcohol is regarded not as a deviation but as a social binding and a relaxing practice there (Felson, Savolainen, Bjarnason, Anderson, & Zohra, 2011). It was suggested that female drinking is more common in Mediterranean cultures, because wine is more often chosen by females and is associated with moderation, while spirits are more typical for male drinkers (Klatsky, Armstrong, & Kipp, 1990). This classification of drinking cultures had been applied in many studies (e.g. Marsh & Kibby, 1992; Allamani, Voller, Kubicka, & Bloomfield, 2000; K. Bloomfield, Stockwell, Gmel, & Rehn, 2003), but was contested with later research. Simpura & Karlsson (2001), Rahav et al. (2006) and Tigerstedt (2007) pointed that drinking patterns in Europe have converged to some extent, making `dry' and `wet' division irrelevant. After denial of `dry' and `wet' cultures and debates on convergence theory, no other systematical description of drinking patterns was suggested. Further studies focus on distinct parameters which define alcohol use at country level.

Some authors note that drinking is more socially acceptable for men (Cooper & Huselid, 2017) and social sanctions for females' abuse are more tough (Byrnes, Miller, & Schafer, 1999). At macro-level, drinking is considered to be inconsistent with feminity and the traditional caretaking role of women in the society and female drinking is usually discouraged (Holmila & Raitasalo, 2005). At the same time, in many cultures and folklore, drinking is portrayed as a traditionally male feature (Iwamoto & Smiler, 2013). Moreover, females are regarded as moderators of males' drinking; they control alcohol use of their partners, who often admit lack of self-efficacy in control of their alcohol use (ibid.).

Difference in gender patterns of alcohol use can also be explained by social roles theory, which focuses on women's involvement in social and political life. Rahav et al. (2006) show that in states where women more often receive higher education, participate in labor force, take places in parliament, etc. female drinking is more common. Grittner, Kuntsche, Gmel, & Bloomfield (2012) note that Gender empowerment and Gender equality indices are associated with smaller gender gap in ethanol intake, while Gini coefficient is inversely related to the gender difference in alcohol consumption.

Drinking behavior is also regulated by value systems that are translated by social institutes, such as religion and law. Religious doctrines are, probably, one of the most powerful regulators of alcohol use. Islam totally prohibits alcohol consumption, and even those Muslim immigrants who violate this ban demonstrate much lower levels of intake than local European population (Valentine, Holloway, & Jayne, 2010). Protestantism does not forbid drinking but substantially restricts ethanol intake. In UK, Canada, New Zealand, Scandinavia and Iceland, Protestantism is strongly associated with temperance movement that emphasizes restrictive measures, advocacy of self-control and moderation (Levine, 1993; Peele, 2010; Room, 2010). It has been shown that the effect of religion on alcohol use may be stronger for females, because women often are more religious than men (Brown, Parks, Zimmerman, & Phillips, 2001; Michalak, Trocki, & Bond, 2007).

Apart from religious practices and commonly held stereotypes, alcohol use is substantially restricted by national policy. European states apply various restrictions, such as control of alcohol production and sales; limitations on advertising; maximum blood alcohol concentration (BAC) level permitted for drivers; age limits for buying alcohol, etc. Basing on these indicators, Karlsson and Osterberg (2001) developed an index that varies from zero to 20 and measures strictness of alcohol control policy across 15 European countries. They revealed that Southern countries (Greece, Portugal, Spain, Italy, and France) demonstrate highest consumption rates in Europe, but social control of alcohol use is the least rigid in these countries. Scandinavian region, on the contrary, scores lowest on levels of intake but highest on the degree of alcohol-related problems and strictness of alcohol control policy (ibid.). Here I use Karlsson and Osterberg's data to analyze the dynamics of strictness index from 1950 to 2000 (see Figure 2).

Figure 3. Strictness of alcohol control policy in 15 European states

In this data, we can see an evidence for convergence theory: during the last decades, Mediterranean region has tightened control on alcohol use, while for Northern countries the opposite is true. Nevertheless, it would be premature to claim that European drinking patterns have converged: average score of strictness index still varies significantly.

In the graph (Figure 2), Denmark is an interesting outlier. Geographically, it belongs to Northern region, by religious denomination - to the Lutheran Church, but the pattern of alcohol control policy is clearly Mediterranean in this country. Demant and Krarup (2013) highlight that liberal attitudes towards alcohol use have developed as the result of Denmark's comparatively early entrance to the European Union, which led to greater latitude in economic aspect of alcohol policy and to the rise of liberal values in the society. Demant and Krarup (ibid.) review various studies that show that in Denmark alcohol is associated with leisure activities and youth lifestyle, and that adolescents in this country score highest on binge drinking in Europe.

Studies mentioned above represent a set of interesting correlations between patterns of drinking and various economical and socio-demographic variables. Unfortunately, these surveys were conducted in different time periods and on different samples, so one cannot say how various socio-economical and culturological determinants of drinking may correspond to each other. Results of these studies resemble a puzzle where each paper highlights one or two details, while the whole picture remains far from being comprehensive. It was suggested that single-dimensional analysis of drinking patterns cannot be efficient, and that problem should be tackled with multidimensional approach (Bondy, 1996). We suggest clustering countries by drinking behavior and then analyze the link between groups and country-level features.

Our hypothesis is that female alcohol consumption should be defined at two levels:

Macro-level factors:

Ш Economic development and social inequality: Gini coefficient and higher GDP per capita should be correlated with larger amount of ethanol consumed by females;

Ш Typical drinking pattern: in wine-preferring countries, females may drink more than in the countries where spirits and liquor are the most popular beverages;

Ш Religion: Protestant countries should score lower on volumes of female alcohol intake;

Individual-level factors:

Ш Caretaking role: marriage and cohabitating should be inversely correlated with female drinking;

Ш Higher education may be connected with more moderate drinking;

Ш Individual beverage preferences: wine-preferring females can be consuming more alcohol than those who usually opt for spirits.

In this chapter, I highlight that female alcoholism should be measured not as number of RSOD per time period, but as a tendency to drinking larger volumes of ethanol. Possible country- and individual-level predictors of females' alcohol use were reviewed; on the basis of literature review I build on hypotheses on the link between various factors and levels of consumption; these factors will be further tested in multilevel regression models.

2.

3. Multilevel analysis of drinking patterns in Europe

3.1 Sample description

The data used in this paper is from the project Gender, Alcohol and Culture: An International Study (GENACIS). GENACIS is a collaborative international project affiliated with the Kettil Bruun Society for Social and Epidemiological Research on Alcohol and coordinated by GENACIS partners from the University of North Dakota, Aarhus University, the Alcohol Research Group/Public Health Institute, the Centre for Addiction and Mental Health, the University of Melbourne, and the Research Institute of Addiction Switzerland. Support for aspects of the project comes from the World Health Organization, the Quality of Life and Management of Living Resources Programme of the European Commission (Concerted Action QLG4-CT-2001-0196), the U.S. National Institute on Alcohol Abuse and Alcoholism/National Institutes of Health (Grants R21 AA012941 and R01 AA015775), the German Federal Ministry of Health, the Pan American Health Organization, and Swiss national funds. Support for individual country surveys was provided by government agencies and other national sources.

GENACIS project seems to be the most comprehensive source of information on the topic, covering not only most detailed characteristics of volumes and levels of consumption, but also data on drinking contexts, respondents' employment and social bonding, as well as measurements of drinking consequences including alcohol-related harm. It provides information on drinking volumes in 36 countries, including 17 European states. Though the focus of this paper is on European countries, we include all the data in the analysis for more accurate statistical inference.

Table 2 presents information on survey design in 17 European countries. More detailed description, as well as information on other countries, can be found on GENACIS website (GENACIS, 2005).

Table 2. Sample description

ECAS stands for European Comparative Alcohol Survey (ECAS), which is also a project of European Commission and constitutes a significant part of GENACIS survey.

To compensate for difference in sample sizes between countries, I apply population size weights recommended by GENACIS team (Kim et al., 2005).

For cross-countries comparisons I conduct multilevel regression analysis of volumes of ethanol intake. Multilevel modelling belongs to hierarchical methods and aims at comparing group effects on individual behavior, combining analysis of actor and context (Paccagnella, 2006; Snijder, 2011). This makes multilevel regressions the most appropriate tool for solving objectives of the present study.

In this paper, I use two blocks of GENACIS core questionnaire (GENACIS, 2001): drinking variables and sociodemographic. The main focus is on annual volume of ethanol consumption - metric variable measured in grams of pure ethanol. This indicator was not measured directly, but calculated by GENACIS specialists from two variables: frequency of drinking and usual quantity (see Application 1 for question wordings). For both indicators, original scales were transformed: ordinal frequency was recoded in number of days, number of drinks was postcoded as grams of pure ethanol (Kim et al., 2005). These procedures are described in Application 2. Multiplication of these recoded values gives annual volume of ethanol intake in grams of pure ethanol (ibid.).

3.2 Exploratory research

First of all, I am testing distribution of volume of intake and difference in drinking patterns between genders in GENACIS data. Brief check shows that the distribution of annual volume of ethanol intake is not homogeneous (see Figure 3). There are many outliers in the data, especially on the right tail of the scale. They may well distort the regression line, making it lay far from the majority of observations, which will lead to large residuals and poor model fit.

Figure 4. Distribution of annual volume of ethanol intake, GENACIS data

A possible solution might be deleting outliers, but removing contrasting cases from the sample contradicts with the aim to highlight country differences in drinking behavior.

Another option is to apply logarithmic transformation to the dependent variable, making it approximately normal (Gelman & Hill, 2007), which is what I have done. Distribution of natural logarithm is still not perfectly normal, but it becomes more symmetric and can be used for statistical inference (see Figure 4).

Figure 5.Distribution of natural logarithm of annual volume

To check the difference between gender patterns and between-country variance, I build almost empty multilevel model that predicts volumes of ethanol intake by gender and country. Basing on the previous studies on the topic (Ahlstrцm et al., 2001; Mдkelд et al., 2006a; Tedor et al., 2017), it is reasonable to expect that the effect of gender on volumes or intake would be co-directional in every country, but coefficients may differ by country. Therefore, slopes in the model are randomized. The regression equation looks as follows: ML_0<-lmer(ln_annuvol~(1+gender|country), data=drinkers, weights =weight).

From this model, it appears that, at б=0.05, the effect of gender is significant in every country from GENACIS sample (see Figure 4).Taking into account that males are coded as 1 and females as 2, we can conclude that females drink significantly less alcohol than males.

Figure 6. Random effect of gender by country

In log-linear models, 1-unit change in predictor describes e^в multiplicative change in the dependent variable (Benoit, 2011). In my model, European maximum of gender effect on ethanol intake is observed in Czechia and equals -1.52. This means that predicted gender gap in ethanol consumption equals e^1.52, meaning that males consume 4.57 times more alcohol than females. European minimum of gender effect is observed in Sweden and equals e^0.69 ~ 2-fold difference in ethanol intake between males and females. So, GENACIS data once again confirms the existence of gender gap in ethanol intake, the ratio varies from 2 to 4.57.

3.3Predicting females' drinking at country and individual level

In order to check which factors define females' drinking behavior and how they differ from predictors of males' behavior, I run multilevel regression models on two subsamples - for males and females separately - and compare coefficients, i.e. effect sizes.

According to the literature review, I am testing the following country-level predictors: GDP per capita, GINI coefficient and the type of mostly preferred beverage in the country. The scale for GDP differs a lot from scales for other predictors, so I take natural log in order to shrink it. For beverage type, I set wine as a baseline category. Individual-level predictors are as follows: age, presence of higher education (binary, yes/no) and marital status (married or cohabiting/single).

Regression equations are as follows:

ML_1m<-lmer (ln_annuvol~ln_GDP + GINI + bevtype + smst_bin + edu_bin + age + (1|country), data= subset(drinkers, gender=="1"), weights=weight)

ML_1f<-lmer (ln_annuvol~ln_GDP + GINI + bevtype + smst_bin + edu_bin + age + (1|country), data= subset(drinkers, gender=="2"), weights=weight)

The results are presented in Table 3. Despite low predictive power (model explains 9% of variance in volumes of intake for males and 17% for females), it is worth mentioning that country belonging influences females' intake much more than that of males'. Judging from the coefficients, GINI coefficient and GDP are non-significant. The type of most popular beverage in the country is also not informative - it would probably be better to check individual-level beverage preferences and income than country-level. Unfortunately, the data on respondent's income is not available in the dataset.

At individual level, marriage and cohabitating reduce volumes of intake for both genders; the effect is slightly larger for females, which confirms the idea that caretaking practices may be correlated with lower levels of drinking. However, we indispose data on respondents' parenthood, so, our knowledge about caretaking activities is not comprehensive.

Table 3. Multilevel model 1

In the second model, I check the effect of religion on the drinking. Since GENACIS does not include questions on religious belonging, I am using country-level data provided by Association of Religion Data Archives (ARDA). In GENACIS sample, Christian countries can be divided into three groups: mostly Roman Catholic, mostly Protestant, or countries with roughly the same proportions of these denominations.

To save all countries in the sample, I've tried to include shares of Catholics and Protestants in the predictors, but coefficients for them turned out to be insignificant. Nevertheless, when I select countries for which religious denomination can be defined unambiguously and divide them into Catholic and Protestant, Independent-Samples Mann-Whitney U Test shows that ratio of males' annual volume to that of females is significantly higher in Protestant Countries (H0 is rejected at б=0.006) . So, I include country's denomination in the regression equations, which look as follows:

ML_2m<-lmer(ln_annuvol ~ protestants2 + catholics + smst_bin + edu_bin + age + bevtype_ind +(1|country), data=subset(drinkers, gender=="1"), weights=weight)

ML_2f<-lmer(ln_annuvol ~protestants2 + catholics + smst_bin + edu_bin + age + bevtype_ind +(1|country), data=subset(drinkers, gender=="2"), weights=weight)

In the resulting model (see Table 4), Protestantism effects only females' drinking (б=0.01). So, in Christian countries, female drinking behavior is to some extent regulated by religion: in Catholic countries, females drink less than in Protestant countries. Thus, our initial hypothesis that Protestantism is associated with temperance movements and lower consumption is rejected. The effect of living in Protestant country is positive, meaning higher levels of consumption. Though we do not dispose data on respondent's individual beliefs, we can conclude that country-level religious background is strongly connected with gender patterns of drinking. This is in line with the idea that individual drinking patterns are affected by macro-social factors, such as religious values and commonly accepted standards of behavior.

Type of mostly preferred beverage was not meaningful at country level (see Table 3), meaning that modal choice of beverage does not affect individuals' pattern of alcohol use. This is another evidence against the theory of `dry' and `wet' countries that rests upon countrywide beverage preferences.

Nevertheless, the effect is significant at individual level (Table 4): females who opt for wine drink less than spirits-drinkers but more than beer-drinkers. So, it is not liquor drinking that poses a treat for females' health, but socially acceptable beverages, such as beer, because they constitute the major part of female ethanol intake.

Table 4. Multilevel model 3

Slight heteroscedasticity of model residuals is observed at left tail, showing that in some groups variation is slightly more than in others, meaning that robust standard errors should be applied instead of usual ones (Figure 10). Taking into consideration sample size, we can say that coefficient estimates approach real в_s and can be used for valid inference.

Figure 10. Testing heteroscedasticity

Conclusion and discussion

In the available dataset, there was no data on income and religious denomination at individual level. Meanwhile, comparing these effects on country and individual level would shed more light on the link between financial and religious factors and gender patterns of alcohol consumption. There is also lack of information about the relation between parenthood and drinking, so the idea of caretaking role as barrier to alcoholism cannot be properly tested.

In contrary to existing publications (Grittner et al., 2012; Rahav et al., 2006), economic development and social inequality seem not to affect excessive alcohol use among females: coefficients for Gini coefficient and GDP per capita turned out to be insignificant in the models on GENACIS data. The only registered macro-level effect is interaction of gender and Protestantism at country-level. Although data on individual religiousness are missing, it should be noted that religious traditions of certain country contribute significantly (if not decisively) to a system of views and custom in this country which defines a socio-cultural background of life of every citizen.

As demonstrated in earlier publications(Grittner et al., 2012), higher education is negatively associated with volumes of consumed alcohol, if evaluated in the whole sample. In females, this connection is inverted, as persons with higher education are significantly more prone to drinking. This may probably result from inflated expectations among educated women which conflict with existing possibilities in job, authority etc., but this association needs deeper examination involving additional socio-psychological predictors.

Beverage type does not affect females drinking at country level but has effect on individual level. This contests the idea that country belonging to `dry' or `wet' culture defines levels of individual intake (K. Bloomfield et al., 2003; Felson et al., 2011; Rehm et al., 1996). Nevertheless, at individual level, the choice of beverage does affect females' volume of intake. The stronger (or the more socially disapproved) the beverage, the lower is the volume of ethanol consumed through drinking it. This may be regarded as the effect of commonly held stereotypes on inconsistence of femininity with drinking strong alcohol beverages.

Considering all listed tendencies, one may propose that increase in alcohol consumption among females is generally conditioned by conflict between expectations and possibilities which are open for women - possibly unequal with ones available for males. If higher education finds no application in traditionally constituted community where female is more expected to be family- rather than career-oriented, this may result in drinking. This can also explain higher rates of female drinking in Protestant countries compared with those where Catholicism prevails. The latter confession is associated with more `traditional' role of woman in family and society, so possibly a frequency of discouragement with disappointed expectations concerning career and `active' possibilities is lower in these countries. Otherwise stated, Catholic countries combine `traditional' views on females' role in both being less ambitious and unaccepting alcohol drinking. Oppositely, in Protestant countries, social stereotypes expect women to be more active in career while alcohol consumption by females is less condemned. Of course, this combination of factors, although feasible, needs more foundation. Moreover, this hypothesis is not fully exhausting. For example, this hardly explains the existing (comparatively small) gender gap in alcohol intake in Catholic Ireland (Figure 2). That is why none of previously postulated hypotheses (including one on `dry' and `wet' countries) cannot be fully neglected, as they seem complementing each other rather than contradicting.

Acknowledgements

This is to express gratitude to Dr. Christian Frцhlich (Sociological department, National Research University Higher School of Economics, Russia) for making arrangements on the access to GENACIS data. I also thank heartily Dr. Andrey Sinjushin (Biological faculty, Lomonosov Moscow State University, Russia) for the help with information on gender difference in metabolism of ethanol.

References

female alcoholism predicate

1.Ahlstrцm, S., Bloomfield, K., & Knibbe, R. (2001). Gender Differences in Drinking Patterns in Nine European Countries?: Descriptive Findings. Substance Abuse, 22(1), 69-85. https://doi.org/https://doi.org/10.1023/A:1026475910263

2.Allamani, A., Voller, F., Kubicka, L., & Bloomfield, K. (2000). Drinking Cultures and the Position of Women in Nine European Countries. Substance Abuse, 21(4), 231-247. https://doi.org/https://doi.org/10.1023/A:1026475910263

3.Ammon, E., Schдfer, C., Hofmann, U., & Klotz, U. (1996). Disposition and first-pass metabolism of ethanol in humans: Is it gastric or hepatic and does it depend on gender? Clinical Pharmacology and Therapeutics, 59(5), 503-513. https://doi.org/10.1016/S0009-9236(96)90178-2

4.Anderson, K., & Plant, M. (1996). Abstaining and carousing: Substance use among adolescents in the Western Isles of Scotland. Drug and Alcohol Dependence, 41(3), 189-196. https://doi.org/10.1016/0376-8716(96)01251-3

5.ARDA. (n.d.). The Association of Religion Data Archives. Retrieved May 27, 2018, from http://www.thearda.com/

6.Baraona, E., Abittan, C. S., Dohmen, K., Moretti, M., Pozzato, G., Chayes, Z. W., … Lieber, C. S. (2001). Gender Differences in Pharmacokinetics of Alcohol. Alcoholism: Clinical and Experimental Reseacrh, 25(4), 502-507.

7.Barnes, G. M. (1990). Impact of the Family on Adolescent Drinking Patterns. In R. L. Collins, K. E. Leonard, & J. S. Searles (Eds.), Alcohol and the Family: Research and Clinical Perspectives (pp. 137-161). New York: Guilford.

8.Barnes, G. M., Farrell, M. P., Cairns, A., & Farrell, M. P. (1986). Parental Socialization Factors and Adolescent Drinking Behaviors. Journal of Marriage and Family, 48(1), 27-36. https://doi.org/10.2307/352225

9.Beccaria, F., & Guidoni, O. V. (2002). Young people in a wet culture: functions and patterns of drinking. Contemporary Drug Problems, 29, 305-334.

10.Benoit, K. (2011). Linear Regression Models with Logarithmic Transformations. London School of Economics, 1-8. Retrieved from http://www.kenbenoit.net/courses/ME104/logmodels2.pdf