Elsevier

Journal of Health Economics

Volume 33, January 2014, Pages 180-187
Journal of Health Economics

Estimating the price elasticity of beer: Meta-analysis of data with heterogeneity, dependence, and publication bias

https://doi.org/10.1016/j.jhealeco.2013.11.009Get rights and content

Highlights

  • Precise estimates of alcohol price elasticities are required for tax policy.

  • I examine a sample of 191 beer price elasticities from 114 empirical studies.

  • Previous meta-analyses have neglected to correct for publication bias.

  • Meta-regressions indicate that the average beer price elasticity is about −0.20.

  • The corrected elasticity is about half of the previous consensus average.

Abstract

Precise estimates of price elasticities are important for alcohol tax policy. Using meta-analysis, this paper corrects average beer elasticities for heterogeneity, dependence, and publication selection bias. A sample of 191 estimates is obtained from 114 primary studies. Simple and weighted means are reported. Dependence is addressed by restricting number of estimates per study, author-restricted samples, and author-specific variables. Publication bias is addressed using funnel graph, trim-and-fill, and Egger's intercept model. Heterogeneity and selection bias are examined jointly in meta-regressions containing moderator variables for econometric methodology, primary data, and precision of estimates. Results for fixed- and random-effects regressions are reported. Country-specific effects and sample time periods are unimportant, but several methodology variables help explain the dispersion of estimates. In models that correct for selection bias and heterogeneity, the average beer price elasticity is about −0.20, which is less elastic by 50% compared to values commonly used in alcohol tax policy simulations.

Introduction

Excessive consumption of beverage alcohol, especially by youth and young adults, can be addressed by a variety of policy tools, including demand-reducing policies in the form of higher alcohol taxes and prices (Phelps, 1988, Cook, 2007, Cawley and Ruhm, 2012). The economic justification for higher alcohol taxes is based on two broad principles. First, there are external costs associated with excessive consumption such as drink-driving, violence, property destruction, and drunken behavior. Higher alcohol taxes reflect social costs not contained in market prices, which restores economic efficiency. Other options include restrictions on time, place or manner of consumption, and severe penalties for some costly behaviors. Second, some individuals are poorly informed about addictive, intoxicating or adverse health effects of excessive alcohol use, and their current consumption may not be optimal due to information costs. Higher alcohol taxes also address this market failure, although better health information or restricted sales are policy alternatives. However, a broad, population-based tax policy is difficult to sustain if there are two types of consumers: heavy drinkers who account for the majority of social costs but whose demands are relatively price inelastic; and light/moderate drinkers with more elastic demands. In the worst case scenario, demands by heavy-drinking individuals are perfectly inelastic, so higher taxes or prices do little to reduce consumption and social costs, and end up imposing welfare losses on moderate drinkers.1 The taxation paradox is often challenged by reviewers citing particular studies suggesting heavy drinkers are highly responsive to prices (e.g., Xu and Chaloupka, 2011), but this view is challenged on empirical grounds by Ayyagari et al. (2011), Manning et al. (1995), Nelson, 2013a, Nelson, 2013b, and Ruhm et al. (2012). Taxation policies are clearly tied to the price elasticity (or inelasticity) of demand for alcohol, but elasticity estimates exhibit substantial dispersion across drinking patterns, beverages, countries and econometric models and methods, making precise summary estimates difficult to obtain.

The objective of this paper is to summarize and synthesize estimates of the price elasticity of beer, while accounting for dispersion and potential biases due to heterogeneity (factual and methodological); dependence of estimates due to methodology and sampling; and publication selection bias. Meta-analysis and meta-regression analysis are used to address these data issues, and I first discuss deficiencies in several previous meta-analyses that failed to analyze fully the dispersion of estimates for beer price elasticities. The importance of this investigation for public policy is illuminated by prior studies of drinking patterns for what is termed the “beer-drinking subculture.” First, Greenfield and Rogers (1999) and Rogers and Greenfield (1999) present survey evidence showing that the top 5% of US drinkers by volume account for 39–42% of alcohol consumption, men and young adults are overrepresented among heavy drinkers, and beer accounts for the bulk of alcohol consumed by heavy drinkers (80% compared to 4% for wine and 16% for spirits). Second, beer is more often involved with drink-driving and other risky behaviors, regardless of confounding variables such as age, gender, and education (Berger and Snortum, 1985, Hennessy and Saltz, 1990). Third, beer accounts for two-thirds of all alcohol consumed by binge drinkers (Naimi et al., 2007), and generally is the preferred beverage among males, young adults, and college students (Dawson, 1993, Kerr et al., 2004, Snortum et al., 1987). As noted by Cook and Moore (1994, p. 53), there is no basis for claiming that beer is the “drink of moderation.” A refined or targeted tax policy could levy higher taxes on beer compared to wine and spirits, but success of this policy depends on the price elasticity of beer (and cross-price elasticities with wine and spirits). Overall, tax policies that address social costs of excessive consumption must carefully consider price responsiveness of alcohol demands for beverages and consumers, especially the price elasticity of beer drinkers.

Numerous empirical studies estimate demand functions for beer. The present study reviews, summarizes, and synthesizes 191 price elasticity estimates – the effect size – drawn from 114 primary studies. Criteria for inclusion and exclusion of primary studies and estimates are described below. Estimates are based on several types of data for 24 different countries. However, six English-speaking countries – Australia, Canada, Ireland, New Zealand, the UK, and US – account for 143 estimates or 75% of the total. Estimates exhibit substantial dispersion and contain potential biases, which reflect several factors. First, there is variation due to factual heterogeneity, such as sample time period and country. There also are numerous methodological factors that vary from study to study, such as level of aggregation, econometric models, estimation methods, and measurement of prices and income. Second, investigators may report more than one elasticity estimate from the same data or different investigators may use identical or similar data sets. This means that effect-size estimates are not independent, which biases summary estimates and their standard errors (Borenstein et al., 2009, Nelson and Kennedy, 2009). Third, studies and estimates may be subject to publication selection bias, which distorts the magnitude and precision of reported effect sizes (Borenstein et al., 2009, Stanley and Doucouliagos, 2012). A complete analysis must consider all three sources of dispersion. Previous analyses have addressed heterogeneity and dependence, but this is the first meta-analysis to report summary estimates of beer price elasticities that correct for publication bias.

The remainder of the paper is divided into four sections. Section 2 contains a brief review of three previous meta-analyses of alcohol price elasticities, and comments on methods used here compared to earlier analyses. Several weighted-average estimates for price and income elasticities are reported, where weights are based on inverse variances. Section 3 contains an investigation of publication bias. Using several methods, corrected weighted-averages are reported. Section 4 examines the effects of heterogeneity within the set of 191 estimates, and presents meta-regressions that test for effects of independent variables describing factual differences, methodology, and publication bias. Section 5 contains a summary of results and discusses implications for alcohol tax policy. Overall, the demand for beer is less elastic than reported in previous meta-analyses or contained in many policy discussions of alcohol problems.

Section snippets

Meta-analysis: procedures in previous analyses and the present study

Meta-analyses in economics must cope with several problems that arise due to the observational nature of data used in econometric studies, reporting methods favored by academic researchers and journals, and potential biases arising from selective reporting and publication of empirical results in primary studies (Nelson and Kennedy, 2009, Stanley and Doucouliagos, 2012). These problems include heteroskedasticity, heterogeneity, outliers, dependence of estimates, and publication bias. Three

Publication bias: detection and treatment

Publication bias occurs when primary researchers search among elasticity estimates and select those with statistically significant coefficients, “correct” signs, and more elastic values. The result is a biased set of published estimates, with general expectations of a positive association between reported effect sizes and their standard errors, i.e., less precise estimates are more likely to be published if they have larger effects.2

Meta-regression analysis of heterogeneity and publication bias

Adding moderator variables to Eq. (1) yields a weighted least-squares (WLS) meta-regression model of heterogeneity and publication bias. Specifying all moderator variables as binary dummies preserves the interpretation of coefficients in Eq. (1), with the intercept as an estimate of the true effect size for a null case and precision coefficient as an index of distortion due to publication bias. As a correction for heteroskedasticity, weights are based on inverse variances of elasticity

Discussion

This study uses meta-analysis to correct summary averages of beer price elasticities for heteroskedasticity, heterogeneity, dependence, and publication bias. First, weighted means for a sample of 191 estimates are −0.23 and −0.35 for fixed- and random-effects. Second, correcting for publication bias using trim-and-fill yields estimates of −0.20 and −0.23. At the median precision, a cumulative meta-analysis yields weighted-means of −0.23 and −0.30. Third, correcting for heterogeneity and

Acknowledgements

Research leading to this paper was supported in part by the International Center for Alcohol Policies, Washington, DC. This paper presents the work product, findings, viewpoints, and conclusions solely of the author. The views expressed are not necessarily those of ICAP or any of ICAP's sponsoring companies.

References (46)

  • M. Borenstein et al.

    A basic introduction to fixed effect and random-effects models for meta-analysis

    Research Synthesis Methods

    (2010)
  • D.E. Card et al.

    Time-series minimum-wage studies: a meta-analysis

    American Economic Review

    (1995)
  • J. Cawley et al.

    The economics of risky health behaviors

  • K.W. Clements et al.

    Is utility additive? The case of alcohol

    Applied Economics

    (1997)
  • S. Cnossen

    Alcohol taxation and regulation in the European Union

    International Tax and Public Finance

    (2007)
  • P.J. Cook

    Paying the Tab – The Costs and Benefits of Alcohol Control

    (2007)
  • P.J. Cook et al.

    This tax's for you: the case for higher beer taxes

    National Tax Journal

    (1994)
  • D.A. Dawson

    Patterns of alcohol consumption: beverage effects on gender differences

    Addiction

    (1993)
  • H. Doucouliagos et al.

    Theory competition and selectivity: are all economic facts greatly exaggerated?

    Journal of Economic Surveys

    (2012)
  • S.J. Duval et al.

    Trim and fill: a simple funnel-plot-based method of testing and adjusting for publication bias in meta-analysis

    Biometrics

    (2000)
  • M. Egger et al.

    Bias in meta-analysis detected by a simple, graphical test

    British Medical Journal

    (1997)
  • J. Fogarty

    The demand for beer, wine and spirits: a survey of the literature

    Journal of Economic Surveys

    (2009)
  • C.A. Gallet

    The demand for alcohol: a meta-analysis of elasticities

    Australian Journal of Agricultural and Resource Economics

    (2007)
  • Cited by (41)

    • How elastic is alcohol consumption?

      2022, Economic Analysis and Policy
      Citation Excerpt :

      This literature has been summarised in several surveys and meta studies and Table 1 contains a summary. It is to be noted that a substantial number of observations lie behind the figures in the table; for example, from the first row, Nelson (2014) analysed 191 estimated own-price elasticities of beer and found the mean to be −0.20. Panel A shows that for beer the mean of the means over nine studies is −0.47, while the mean of the medians reported in four studies is −0.37.

    • The impact of information-based interventions on conservation behavior: A meta-analysis

      2020, Resource and Energy Economics
      Citation Excerpt :

      Beyond publication bias, other potential shortcomings can compromise model results. Robustness checks are essential to meta-analysis and can occur in several ways (Lindhjem and Navrud, 2008; List and Gallet, 2001; Nelson, 2014; Penn and Hu, 2018; Shrestha and Loomis, 2001). To deal with potential outliers, we report the results of a 98% dataset, in which the smallest 1% and the largest 1% of effect sizes are removed.9

    • Understanding the behavior of beer consumers

      2020, Case Studies in the Beer Sector
    • Cheap talk efficacy under potential and actual Hypothetical Bias: A meta-analysis

      2019, Journal of Environmental Economics and Management
      Citation Excerpt :

      Instead, field surveys, using auction-type mechanisms, and longer CT scripts demonstrate more evidence of lower CTCF, though each presented only modest evidence. Another strategy to balance out observations across studies is to use a limited number of effect sizes per study as in Nelson (2014) and Fogarty (2010). This robustness check method created a dataset of 102 effect sizes based on the 67 studies, with no more than two observations per study.

    • Minimum unit prices for alcohol

      2019, Journal of Health Economics
      Citation Excerpt :

      There is also controversy in the empirical literature. Nelson (2014) argues that meta-analyses face a number of biases due to factors such as heterogeneity, multiple estimates from a single study, and publication bias. He argues that when these biases are addressed, estimated elasticities are reduced.

    View all citing articles on Scopus
    View full text