I. Introduction
Research in health, epidemiology, and nutrition has long suggested a link between moderate alcohol consumption and health. Early studies (Klatsky et al., Reference Klatsky, Friedman, Siegelaub and Gérard1977; Marmot et al., Reference Marmot, Shipley, Rose and Thomas1981) established a “U-shaped” relationship between alcohol consumption and health outcomes and mortality. That is, alcohol confers health-protective benefits up to a point, but excessive consumption leads to adverse health effects. Renaud and de Lorgeril (Reference Renaud and de Lorgeril1992) were among the first to highlight wine as having specific health benefits beyond those attainable through beer or spirits consumption.
However, two lines of research in recent decades have raised questions regarding the exceptionalism of wine as an alcoholic beverage with potential protective effects. In one, studies have found a robust inverse relationship between mortality or the incidence of adverse health effects with beer or alcohol, broadly defined. For example, Brenner et al. (Reference Brenner, Rothenbacher, Bode, März, Hoffmeister and Koenig2001) found evidence for a strong protective effect against coronary heart disease (CHD) of alcohol among beer drinkers in Germany. The other line of research (e.g., Johansen et al., Reference Johansen, Friis, Skovenborg and Grønbæk2006) has shown that wine drinkers tend to have healthier diets and lifestyle choices, as compared with beer drinkers and spirits drinkers. Therefore, perhaps some of the protective benefits of wine may have been overestimated due to confounding effects.
Although specific linkages and mechanisms still need to be identified, evidence is growing to support the notion that beer may confer specific health-protective benefits. Much as the antioxidants specific to wine, including resveratrol, have been found to support health and reduce mortality, beer is likely to contain its own protective agents. There is reason to believe that, in this regard, not all beers are the same. Craft beer differs fundamentally from the typical American adjunct lager macrobeer (e.g., Budweiser), in that it typically features significantly more hops per gallon, higher alcohol content, and silicone and other minerals from malted barley, and it is generally nonpasteurized allowing for the digestive aid of brewer's yeast, among other factors. Although not all of these factors are necessarily health protective in nature, the brewing process and content of craft beer differ substantially from those of macrobeer, and hence there is reason to expect that the health effects of consuming these beverages may differ.
We use household-level scanner data on food and beverage purchases, merged with self-reported health outcomes, to investigate relationships between alcohol-purchasing behavior and CHD and type 2 diabetes (T2D). We measure alcohol-purchasing behavior by using annual expenditures. We are thus able to measure the associations between the incidence of CHD and T2D with shifts in expenditures among beer, wine, and spirits. Additionally, we quantify the associations between alcohol purchases, by type, and health outcomes, relative to nondrinkers. We differentiate between craft beer, macrobeer, and additional categories of beer purchases in an effort to identify the various health impacts.
Our results provide further evidence for the potential health-protective benefits of moderate alcohol consumption. Craft beer, macrobeer, and wine expenditures are all significantly associated with the decreased likelihood of CHD and T2D, after controlling for a large number of demographics, lifestyle choices, and other potential confounders. In many cases, the estimated craft beer effect is at least as strong as that found for either wine or macrobeer.
II. Background and Literature Review
Researchers throughout many disciplines have long sought to identify the factors associated with mortality and adverse health outcomes. Empirical findings in this respect have informed audiences including the medical profession, policy makers, and the general public. Likewise, alcohol use has been a subject of health-related research for readily apparent reasons. According to the National Institute on Alcohol Abuse and Alcoholism, National Institutes of Health (2016), in 2014 71% of American adults reported that they had consumed alcohol within the past year. Alcohol abuse in the United States is estimated to be responsible for $249 billion in costs annually, and approximately 88,000 American adults die each year from alcohol-related causes.
Early on, researchers in health and epidemiology established that alcohol use shares a U-shaped, or parabolic, relationship with morbidity and a number of other prominent ailments (Klatsky et al., Reference Klatsky, Friedman, Siegelaub and Gérard1977; Marmot et al., Reference Marmot, Shipley, Rose and Thomas1981). Moderate alcohol use is more associated with good health than is abstinence. This finding has proved robust across the years and a wide range of methods, data sets, ailments, and countries. Roerecke and Rehm (Reference Roerecke and Rehm2014) conduct a review of meta-analyses, citing more than 100 scholarly articles, and determine that alcohol consumption shares a clear parabolic relationship with heart disease.
Wine has been the subject of many studies on alcohol and health. Renaud and de Lorgeril (Reference Renaud and de Lorgeril1992) authored one of the earliest influential studies in this vein, investigating the so-called French paradox, whereby French adults consume more saturated fats yet experience less CHD than many other comparable populations. The authors identified red wine intake as a key factor in resolving this apparent contradiction. It is widely understood that resveratrol, a compound found in grapes and wine but not in other alcoholic beverages, is partially responsible for the perceived health benefits of wine. Resveratrol has a number of anti-inflammatory, anticarcinogenic effects and serves as an antioxidant (Gehm et al., Reference Gehm, McAndrews, Chien and Jameson1997). A handful of studies have compared estimated impacts across alcohol types and found wine to have the strongest protective impacts, including Criqui and Ringel (Reference Criqui and Ringel1994) with respect to CHD, Arranz et al. (Reference Arranz, Chiva-Blanch, Valderas-Martínez, Medina-Remón, Lamuela-Raventós and Estruch2012) with respect to cardiovascular disease and cancer, and Klatsky et al. (Reference Klatsky, Friedman, Armstrong and Kipp2003) with respect to adult mortality.
Beer has its own health-protective attributes, unique among alcoholic beverages. Both Keil et al. (Reference Keil, Chambless, Döring, Filipiak and Stieber1997) and Brenner et al. (Reference Brenner, Rothenbacher, Bode, März, Hoffmeister and Koenig2001) found the familiar U-shaped relationships between alcohol consumption and health outcomes, based on samples of predominantly beer-drinking German adults. Via hops, a key input to the beer-brewing process, beer contains the flavonoid xanthohumol, which has been the subject of a number of health studies. Stevens and Page (Reference Stevens and Page2004) found it has the potential to impart cancer-preventive properties through beer drinking. In laboratory tests, Vanhoecke et al. (Reference Vanhoecke, Derycke, Van Marck, Depypere, De Keukeleire and Bracke2005) showed xanthohumol to inhibit the growth and spread of cancer cells. Magalhães et al. (Reference Magalhães, Carvalho, Cruz and Barros2009) note that although questions remain regarding the specific mechanisms, the evidence to date suggests broad protective effects through beer consumption. Kaplan, Palmer, and Denke (Reference Kaplan, Palmer and Denke2000) and Bamforth (Reference Bamforth2002) reviewed the extensive literature on beer and health and synthesized a body of research demonstrating that beer is similarly effective to wine in preventing CHD and that beer features a greater composition of B vitamins, minerals, and fiber than does wine. Gerhäuser et al. (Reference Gerhäuser, Alt, Klimo, Knauft, Frank and Becker2002) conducted a series of laboratory analyses to show that beer is a source of several potential anticancer agents, including but not limited to xanthohumol.
Additionally, a wealth of research over the past 30 years has shown that there is a robust and significant impact of moderate alcohol consumption, defined generally, on health and longevity.Footnote 1 This includes Djouseé and Gaziano (Reference Djouseé and Gaziano2007) studying the incidence of heart failure, Brien et al. (Reference Brien, Ronksley, Turner, Mukamal and Ghali2011) examining levels of high-density cholesterol and the risk of CHD, Valmadrid et al. (Reference Valmadrid, Klein, Moss, Klein and Cruickshanks1999) studying CHD among people with older-onset diabetes, Ronksley et al. (Reference Ronksley, Brien, Turner, Mukamal and Ghali2011) on the risk levels for multiple cardiovascular outcomes, Williams et al. (Reference Williams, Cherkas, Spector and MacGregor2005) on bone mineral density in women, and Wang et al. (Reference Wang, Lee, Manson, Buring and Sesso2010) comparing moderate drinking with not drinking in the study of weight loss and overweight status among women. In studies particularly relevant to our own, Koppes et al. (Reference Koppes, Dekker, Hendriks, Bouter and Heine2005) and Joosten et al. (Reference Joosten, Grobbee, van der A, Verschuren, Hendriks and Beulens2010) showed that moderate alcohol intake is associated with reduced risk of T2D, compared with nondrinkers.
In contrast to those studies cited previously demonstrating the greatest health benefits from wine among all alcoholic beverages, a number of studies have reached comparable conclusions for beer. Rimm et al. (Reference Rimm, Klatsky, Grobbee and Stampfer1996) conducted a review of studies analyzing both beer and wine impacts. The reviewed case-control studies collectively do not suggest that wine is more protective than beer, and prospective cohort studies are evenly split in finding stronger health effects for beer or wine. Renaud et al. (Reference Renaud, Guéguen, Siest and Salamon1999) conducted a prospective cohort study and found that wine was more strongly associated with reduced mortality, but that beer and wine have comparable effects in reducing CHD. Costanzo et al. (Reference Costanzo, Di Castelnuovo, Donati, Iacoviello and de Gaetano2011) found that beer and wine had essentially identical overall impacts on cardiovascular health. Using the same data as our own, Adjemian, Volpe, and Adjemian (Reference Adjemian, Volpe and Adjemian2015) found that wine-purchasing behavior was most strongly, negatively associated with heart disease, but that beer and wine were equally associated with the reduction in T2D. Data limitations are likely a factor behind this ambiguity, as robust dietary data, including alcoholic intake differentiated by type, are rarely seen in alcohol-health studies.
We study the health impacts of beer and wine separately due to the possibility that diet and lifestyle have potentially confounded previous studies. Research on alcohol and diet have reached a consensus that wine drinkers, on average, consume healthier foods than do other drinkers and nondrinkers. Tjønneland et al. (Reference Tjønneland, Grønbæk, Stripp and Overvad1999) evaluated diet quality according to alcohol preferences and found that, among Danish adults, wine drinkers eat significantly healthier diets. Mortensen et al. (Reference Mortensen, Jensen, Sanders and Reinisch2001) and Barefoot et al. (Reference Barefoot, Grønbæk, Feaganes, McPherson, Williams and Siegler2002) argued that differences in diet and lifestyle explain much of the perceived differences between beer and wine impacts, as wine drinkers eat more fruits and vegetables, fewer saturated fats, and are less likely to smoke. The former study also found that wine drinkers also have higher average incomes and IQs. Breslow, Guenther, and Smothers (Reference Breslow, Guenther and Smothers2006) examined alcohol intake and diet with the National Health and Nutrition Examination Survey (NHANES). They found that adults consuming at least three alcoholic beverages per day ate significantly healthier diets, on average, than those consuming fewer than one drink per day.
To measure health impacts, our study focuses on CHD and T2D. These ailments were chosen due to data availability, but also because they are among the most studied phenomena in the literature cited previously. Both ailments have been linked to alcohol preferences for decades, are prevalent in the United States, and are responsible for substantial health care expenditures. High cholesterol and high blood pressure are associated with a number of heart ailments, notably CHD, that are estimated to be responsible for more than $300 billion in health care costs (Centers for Disease Control and Prevention [CDC], 2011). An estimated 29 million U.S. adults have diabetes, causing an estimated $176 billion in direct medical costs in 2012 (CDC, 2014).
III. Data and Statistics
To measure purchasing behavior, we use the Information Resources Incorporated (IRI) Household Panel, 2008–2012, which includes comprehensive point-of-sale records for all UPC-coded food and beverage purchases. Participants use handheld scanners to scan the bar codes for all purchases made for at-home consumption, and purchase records include prices, quantities, detailed product descriptors, any promotional activity, and the retailers. The household data are reported by purchase frequency, meaning each shopping trip yields observations, coded by date. We aggregate food and beverage purchases annually.
This data set affords a number of advantages, relative to dietary recall data (e.g., NHANES). The IRI purchase records are comprehensive, providing longitudinal data and precluding the need to infer diet quality based on periodic survey responses. With purchase records at the UPC level, we are able to distinguish alcoholic beverages at a granular level, separating out wine, beer, spirits, and subcategories within. The same is true for food, allowing researchers to examine dietary choices in conjunction with multiple measures of diet quality, food groups, or consumption recommendations. The disadvantages of the IRI household data are that we observe purchase only, not consumption, and that we have no records of food or beverage purchases for consumption away from home (e.g., restaurants). Our analysis assumes parity between purchase and consumption at the household level, and that consumers engage in relatable behavior at home and away from home.
To measure health outcomes, we use the 2010–2012 IRI MedProfiler data. This includes an array of self-reported ailments and health concerns at the individual level. Participants in MedProfiler are asked to respond to questions regarding common ailments with one of five possible responses: suffer but do not treat, suffer and treat with a prescription, suffer and treat with over-the-counter medication, suffer and treat with both prescription and over-the-counter medications, or do not suffer. For both CHD and T2D, we created annual binaries equal to 1 if the respondent indicated that he or she suffers at all. MedProfiler also includes responses to a number of questions that provide insight into lifestyle choices and overall healthiness. These include the frequency of exercise, eating dessert, eating fast food, eating organic foods, or experiencing stress. Approximately 30,000 households participated in both the Household Panel and the MedProfiler data.
The IRI data sets provide rich information on demographics and other important descriptors, providing us with key controls, as suggested by past studies on alcohol and health. The Household data include household-level characteristics, including annual income, the education level and occupation of the household head(s), race, geographic location, marital status, and the age and number of children. The MedProfiler data include individual descriptors, including height and weight (with which we can calculate body mass index [BMI]), age, and gender.
To control for diet quality, we calculate the annual USDAScore, by household. The USDAScore was devised by Volpe and Okrent (Reference Volpe and Okrent2012) and measures the extent to which shopping baskets conform to the Dietary Guidelines for Americans (DGA). The U.S. Department of Agriculture (USDA) Center for Nutrition Policy and Promotion (CNPP) Thrifty Food Plan is intended to inform households on any budget how to best apportion food dollars in order to meet the DGA. The CNPP organized all foods and nonalcoholic beverages into 24 comprehensive categories and assigned recommended expenditure shares, by category and by age and gender. The USDAScore is calculated by comparing observed expenditure shares with household-specific recommendations, and higher scores reflect greater adherence to the DGA. The full details of the variable construction are available from Volpe and Okrent (Reference Volpe and Okrent2012).
Our alcohol preference variables are based on expenditure shares. For each household and year, we calculate total alcohol expenditures and the share of alcohol expenditures attributed to beer, wine, and spirits. We also use total alcohol expenditures, by year, to classify all households as light, medium, heavy, or nondrinkers. Due to potential health effects of hops and alcohol, two beer ingredients that can vary depending on the brewing process, we further distinguish beer as either macrobeer, craft beer, imported beer, or malt beverages and cider.Footnote 2 To enrich further our understanding of the associations between alcohol and health, we also create subcategories for wine. Much as the concentration of fiber or B vitamins is likely to differ across beer varieties, production methods and inputs vary with the quality of wine. Moreover, wine demand can change substantially based on price points (Cuellar and Huffman, Reference Cuellar and Huffman2008; Nelson, Reference Nelson2013). Hence, we use price per liter to categorize all wine purchases in our data as jug, table, premium, or ultrapremium.Footnote 3
Given that our data set pairs household-level expenditures with individual health outcomes, care must be taken in interpreting the results. With our data set, we are able to provide insights into the associations between residing in households exhibiting alcohol expenditures and the likelihood of reporting T2D and CHD. Nevertheless, it remains the case that our data set and preferred estimation approach are subject to measurement error due to the merger of household and individual characteristics. We take care to examine and discuss this point with robustness checks.
We also calculate the annual ratio of total alcohol expenditures to total food expenditures. This variable, AlcFoodRatio, serves two purposes. First, it is an additional lifestyle control, intended to capture households’ overall preferences toward alcohol consumption. Second, it allows us to identify potentially problematic outliers, such as households that considerably underreport food purchases, overreport alcohol expenditures, or have made disproportionately large alcohol purchases for hosting or entertaining purposes. Table 1 reports definitions and summary statistics for the variables used in our analysis.
Variable Definitions and Summary Statistics
Note: a The classifications based on BMI are taken from the Centers for Disease Control and Prevention.
Relative to the general population, the IRI households sampled for our study are older, have higher incomes, and are better educated. The average age in our sample is 53, and the average annual household income is $63,663. Approximately 18% of respondents completed some graduate school. Our data set is also 84% white. These aspects of our sample should be kept in mind while interpreting the results and considering further research.
With respect to health, approximately 10% of respondents have CHD and T2D, respectively. Over a third suffer from high cholesterol and hypertension, respectively. The average BMI is nearly 29, indicating that the average respondent is nearly clinically obese. Almost 72% of respondents indicated that they experienced stress throughout the time series. Hence, we have a range of health risks and indicators within the sample.
Among alcoholic beverages, wine dominates slightly, with an average expenditure share of 23%. Thus, households in our sample spend the most on wine among all alcoholic beverages. Macrobeer is second, with an average 18% expenditure. Spirits are third with 13%, followed by craft beer with 3%, malt beverages/cider with 2%, and imported beer with 1%, on average.
IV. Methodology and Results
We employ a two-pronged approach to estimate potential health impacts from alcohol consumption, as proxied by expenditures. First, following Adjemian, Volpe, and Adjemian (Reference Adjemian, Volpe and Adjemian2015), who also used IRI data and similarly controlled for overall diet quality, we estimate logistic regressions to identify the impacts of alcohol preference on the prevalence of CHD and T2D. Our indicators for CHD and T2D are binaries, reporting 1 for sufferers and 0 for nonsufferers; therefore, limited dependent variable regression is appropriate for our purposes. The likelihood of suffering from the ailment of interest (CHD or T2D) for individual i in household h at time t is modeled as follows:
$$\eqalign{{\rm Ailment}_{iht} = \;& {\rm \theta} _1 + {\rm \theta} _2 {\rm USDAScore}_{ht} + {\rm \theta} _{\rm A} {\bf AlcoholShares}_{ht} + {\rm \theta} _{\rm H} {\bf HealthIndicators}_{it} \cr &+ {\rm \theta} _{\rm D} {\bf Demographics}_{ht} + {\rm \theta} _{\rm L} {\bf Lifestyle}_{it} + {\bf e}_{iht}.} $$
Thus, the determinants of the incidence of these ailments are dietary quality and a series of vectors. AlcoholShares is the vector of alcohol preference variables, given by the expenditure shares by alcohol type. In this setting, we are restricted to estimating potential health impacts among broad categories of drinking behavior. HealthIndicators is the vector of measures of health and well-being, as drawn from the MedProfiler data, including age, BMI, stress, hypertension, and high cholesterol. Demographics is the vector of household-level descriptors drawn from the Household Panel, including income, education levels, marital status, the presence of children, race, and a vector of geographic variables including urban versus rural and regional dummies. Consumption craft versus macro beer has been shown to be geographically dependent (Elzinga et al., Reference Elzinga, Tremblay and Tremblay2015). Lifestyle is a vector of variables measuring an individual's stated lifestyle and preferences, in our effort to account for the unobservable demand for health that is likely correlated with health outcomes. These include exercise frequency and consumption of fast foods, organic foods, and dessert.
The second component of our approach seeks to identify potential health impacts of alcohol consumption, based on intensity and relative to not drinking. The likelihood of suffering from the ailment of interest (CHD or T2D) for individual i in household h at time t is modeled as follows:
$$\eqalign{ {\rm Ailment}_{iht} = \;& {\rm \theta} _1 + {\rm \theta} _2 {\rm USDAScore}_{ht} + {\rm \theta} _{{\rm AL}} {\rm AlcoholSharesLight}_{ht} \cr & + {\rm \theta} _{{\rm AM}} {\bf AlcoholSharesModerate}_{ht} + {\rm \theta} _{{\rm AH}} {\bf AlcoholSharesHigh}_{ht} \cr & + {\rm \theta} _{\rm H} {\bf HealthIndicators}_{it} + {\rm \theta} _{\rm D} {\bf Demographics}_{ht} + {\rm \theta} _{\rm L} {\bf Lifestyle}_{it} + {\rm} {\bf e}_{iht}.} $$
The setup is comparable to equation (1), though the households have now been classified as nondrinkers (the reference category), light, medium, or heavy. We calculated total alcohol expenditures, by household and year, and created three equal quantiles to create the categories. An example of an alcohol variable is CraftBeerShareLight, which is the share of alcohol expenditures among households in the lowest quantile for alcohol expenditures. In estimating equation (2), we also expand our classification of wine into the four subcategories, based on price per liter. This has the advantages of providing a more nuanced understanding of alcohol/health associations. However, the samples of households within the subcategories included in the model may grow small, potentially yielding imprecise coefficient estimates.
Given the size of the data set and the work required by respondents to record and input responses, we expect that the sample contains errors and outliers. We removed all households with total food expenditures, 2008–2012, less than $100, to eliminate households that significantly underreport food purchases. We also removed households with total alcohol expenditures at least twice the value of total food expenditures. Finally, we removed households with individuals who demonstrate likely errors in inputting their health outcome data. Any household with individuals who changed their ailment status twice between 2010 and 2012 was removed from the sample. We also removed all possible and probable outliers for any of the continuous variables, including BMI and dietary scores.
In settings such as these, endogeneity and dual causality are persistent concerns. It is easy to imagine that less healthful preferences for alcohol and food, as well as the proclivity toward CHD or T2D, are both determined by underlying preferences for health that cannot be captured by the model. To circumvent this and to establish a degree of causality, we treat alcohol preferences and diet as stock variables. Additional, this decision was made in light of Williams's (Reference Williams2005) finding that alcohol use is partially due to habit formation and therefore longer-term decisions. In our baseline estimation, we model health outcomes in 2012 as a function of each household's overall alcohol expenditure shares and USDAScore for the entire time period during which they participated in the Household Panel. In many cases, this constitutes 5 years of data. To increase the sample size, we also relax this restriction and estimate equations (1) and (2) such that health outcomes for 2010–2012 are functions of aggregate purchasing behavior for all years leading up to and including the year of the outcome. The results for CHD are reported in Table 2.
Logit Regression Results for Coronary Heart Disease
Notes: Asterisks (***, **, and *) indicate estimated odds ratio is statistically different from 1 at the 0.01, 0.05, and 0.10 levels, respectively. P values in parentheses. BMI, body mass index; T2D, type 2 diabetes.
Unless otherwise noted, we restrict our discussion to the 2012-only results, our preferred specification for both equations (1) and (2). With few exceptions, the results are closely comparable across our two specifications of equation (1), lending robustness to the results. The coefficient estimates are odds ratios. A value below 1 indicates factors associated with decreased risk, whereas a value above 1 indicates heightened risk for the ailment of interest. In our preferred estimation of equation (1), among alcohol variables only the coefficient on wine share is statistically different from 0, and it indicates the expected protective effect. However, the coefficient on craft beer share is the smallest in magnitude and is very near being statistically significant at the 0.10 level, providing evidence for a potential protective effect for CHD. Although insignificant, the coefficients on spirits, macrobeer, and import beer are all below 1, whereas the coefficient for cider and malt beverages indicates a potential harmful effect, though we stress more data are needed to study this category.
As expected, our estimates for equation (2) are less precise, as there are fewer households in each drinking category. Here we find that light craft beer drinking, relative to not drinking, is associated with the decreased incidence of CHD. We also find significant protective effects for heavy macrobeer expenditures and heavy jug and premium wines. The results to this point are beginning to flesh out the narrative that wine and beer exhibit the strongest protective effects with respect to CHD, a finding consistent with much of the research to date on alcohol and health.
The results of estimating equations (1) and (2) on T2D are reported in Table 3. Wine, spirits, craft beer, and macrobeer are all associated with decreased risk, among drinkers, according to the estimation results for equation (1). Among these four, there are no coefficient estimates statistically different from any others. Comparing drinkers with nondrinkers via the results for equation (2), we find consistent evidence of the protective effects of wine consumption. The estimated impacts of medium and heavy jug wine consumption, heavy table wine consumption, all degrees of premium wine consumption, and light ultrapremium wine consumption are statistically significant. Light and heavy spirits consumption demonstrate protective effects, and again we find no significant impacts for the relatively thin cider/malt beverage category. Among beer, heavy consumption of craft beer and the medium and heavy consumption of macrobeer are all associated with protective effects. Import beer consumption demonstrates no effect overall, but we caution that more data on import beer purchases are needed to better study the potential role of these beverages.
Logit Regression Results for Type 2 Diabetes
Notes: Asterisks (***, **, and *) indicate estimated odds ratio is statistically different from 1 at the 0.01, 0.05, and 0.10 levels, respectively. P values in parentheses. BMI, body mass index; CHD, coronary heart disease.
Broadly, the results support the long-standing contention that alcohol consumption (as proxied by purchase behavior in our study) has health-protective effects. Craft beer, wine, and macrobeer are all associated with the decreased likelihood of CHD and T2D. Moreover, spirits are also associated with decreased T2D risk. Among statistically significant associations, we find that beer and wine are comparable in their health-protective effects.Footnote 4 In most estimations, the coefficient for craft beer is the lowest among all beer types and is among the lowest among all alcohol variables. However, these differences among coefficients are not statistically significant.
Most of our controls have the expected signs and are, for the most part, significant. All of the health factors are positively and significant associated with CHD, and all but stress are significant determinants of T2D. Females are significantly less at risk for both ailments. Age is a significant risk factor for T2D, married couples are less at risk for CHD, and households with children present are somewhat less as risk for T2D. Income and dietary quality are negatively and significantly associated with both ailments.
The lifestyle measurements are jointly important as determinants. Exercise significantly decreases the risk factor for both CHD and T2D. Fast-food consumption, as expected, is positively associated with T2D incidence, and the opposite is true for organic foods. One somewhat counterintuitive finding is that dessert is shown to have a protective effect for T2D. However, there are multiple potential explanations for this. Dietary quality is in the model, and so sugars and sweets purchased to constitute desserts are already controlled for, meaning that this variable could be capturing, in part, the tendency to eat dinners as a family. Additionally, the IRI question on dessert does not specify the type of dessert, meaning this includes individuals eating fruit or cheese for dessert.
To aid in the interpretation of the odds ratios for our key alcohol variables, we calculated the marginal effects on the incidences of the ailments based on 10 percentage point increases in the respective expenditure shares. These values are reported in Table 4. As with the estimated odds ratios, the majority of the estimated effects are not statistically different from one another.
Marginal Impacts on the Incidence of Heart Disease and Type 2 Diabetes
Notes: The marginal impacts are calculated using the estimated odds ratios from Table 2, the regression results. For the alcohol variables, they report the estimated effect of increasing the respective shares by 10% on the likelihood of suffering from the ailments. For example, a 10 percentage point increase in the craft beer share is associated with a 3.4% decrease in the likelihood of reporting coronary heart disease.
Our findings suggest that, with respect to health-protective effects, beer consumption is comparable to wine and in some cases may be superior. Increasing the household craft beer expenditure share by 10 percentage points is associated with a 3.4% decrease in risk for CHD and a 5.5% decrease in the risk for T2D in our baseline estimations of equation (1). A similar adjustment for macrobeer is associated with approximately a 0.8% decrease in risk for CHD and a 4.5% decrease in T2D risk. For wine, the estimated marginal effects are a bit more disparate, at 1.4% for CHD and 5.2% for T2D. There is no evidence that malt beverages and cider or imported beer have protective health effects, but these beverages had low average expenditure shares in the data.
We also provide the estimated marginal impacts relative to nondrinkers, as drawn from our estimation of equation (2). As before, care must be taken in using these numbers, as model (2) yielded less precise estimates. However, the magnitudes of the protective effects for craft beer stand out among beer types. The potential reduction in the incidence or likelihood of CHD is as high a 9% as a result of craft beer drinking, relative to not drinking at all. For T2D, the effect is as high as 7%. In some cases, the estimated protective effects due to wine consumption may be as high as 10% to 11%.
Finally, to help contextualize these marginal impacts, we also report the estimated effects for some of our key control variables. A 1-point increase in BMI is associated with a 1% increase in the likelihood of CHD and a 7% increase for T2D, at the mean. Increases in exercise frequency, as measured by the scale used in the IRI questionnaire, have substantial implications for health effects. Adults who exercise regularly are between 11% and 17% less likely to suffer from either of these ailments, as compared with nonexercisers. Dietary quality and household income are both associated with economically significant protective effects as well.
A. Robustness Check: The Panel of Individuals Only
Recognizing the potential measurement error that may result from using household behavior to model individual health outcomes, we also estimate equations (1) and (2) using the sample of households in the panel consisting of individuals only. We identified these households using the household size variable and restricting our attention to only those with values equal to 1. The selected results are reported in Table 5.
Selected Logit Regression Results for Households of Size 1
Notes: Asterisks (***, **, and *) indicate estimated odds ratio is statistically different from 1 at the 0.01, 0.05, and 0.10 levels, respectively. P values in parentheses.
Our goal in conducting this robustness check is to investigate if the estimated effects, with respect to alcohol expenditures, are qualitatively comparable to those reported in Tables 2 and 3, using the full sample. There are two important reasons to expect that the magnitudes of the coefficients might differ considerably. The first is that the sample is very thin for this estimation, particularly for equation (2). The second is that people living alone in the sample differ, statistically, from the overall sample. They are considerably more likely to be either students or elderly and are more likely to live in urban areas, as examples.
Nevertheless, the results in Table 5 corroborate our findings from the general sample. Wine and craft beer stand out among alcohol types, exhibiting significant protective effects in four and three of the estimations of equation (1), respectively. Spirits and macrobeer demonstrate significant protective effects for T2D, whereas import beer and spirits show no significant effects. We therefore conclude that our full-sample results are not affected importantly by the merging of household and individual factors.
V. Conclusions
We investigate relationships between alcohol-purchasing behavior, dietary quality, and the incidence of CHD and T2D. Among U.S. adults who purchase alcohol, we find support for the contention that beer is comparable to wine in its protective effects for drinkers. We measure alcohol behavior based on expenditure shares and find that shifting alcohol expenditures to beer or wine has significant protective effects against both ailments. Moreover, we distinguish craft beer from other varieties of beer and find evidence that craft beer may have stronger protective effects than macrobeer or wine. We discuss a number of reasons as to why craft beer may be healthier than macrobeer, malt beverages/cider, or other choices.
Our findings also support the growing consensus in research across multiple disciplines that alcohol intake has health-protective effects, relative to not drinking at all. We find that beer, wine, and spirits are all associated with the decreased incidence of heart disease and diabetes, relative to nondrinkers. As was the case when we focused entirely on drinkers, the effects of beer and wine on health are closely comparable, lending weight to the notion that beer has its own health-protective attributes.
There are three avenues by which this research can readily and fruitfully be extended. One is to examine the impacts of alcohol-purchasing behavior on other health measures, including longevity and body weight. Examining shorter-term ailments, for example strokes, might also improve the identification strategy. Another is to examine consumption directly, in order to quantify adults according to the number of drinks consumed per day or per week. This would allow for consumers to be categorized according to their adherence to dietary recommendations with respect to alcohol intake, specific to gender and body weight. Related to this, it would improve our estimation considerably to observe alcohol consumption away from home. Finally, it would be worthwhile to investigate the prevalence of switching among alcohol types among U.S. consumers and to understand the determinants of this behavior among consumers. Research (Bray, Loomis, and Engelen, Reference Bray, Loomis and Engelen2009; Empen, Glauben, and Loy, Reference Empen, Glauben and Loy2012) has demonstrated that brand loyalty can be an important factor in determining purchases, at least among beers. Future work can help inform on the mechanisms that might drive consumers toward more health-protective alcoholic beverages.
