An Exploration of the Causal Effects of Alcohol Consumption on Income: A Case Study of the Mekong River Delta of Vietnam t to Phan Van Phuc ng hi Can Tho University, Vietnam ep Abstract w n Research on the relationship between alcohol consumption and economic outcomes is relatively lo underdeveloped This paper presents an analysis of the nature of the relationship between alcohol ad consumption and income per capita in the rural Mekong River Delta of Vietnam Applying a three-stage least y th squares estimator to the Vietnam Household Living Standard Survey data, this study finds that income per ju yi capita was detrimentally affected by alcohol use over the period 2010–2014 This research suggests that the pl Government should consider deploying more legal restrictions and reforms with respect to alcohol production n va Introduction n ua al and consumption ll fu Since the beginning of Doi Moi (Reform) in the late 1980s, the extraordinary economic progress has fostered oi m significant growth in income per capita In the period 2000–2015, the Vietnamese average income grew rapidly from less than US$400 to over US$2000 (World Bank 2017) It is also not surprising to observe a simultaneous expansion in alcohol consumption per capita as incomes have increased The General Statistics of Office of Vietnam (GSO) reported that, on average, individual consumption of beer and wine increased from 0.7 to 0.9 at nh z litres (wine and beer together) per month over 2006–2014 (GSO 2016) However, it is argued that unrecorded drinking could mean that the level and growth of expenditure and consumption of alcohol is much higher than these official statistics (e.g Bao Giang et al 2013, Lincoln 2016) Several studies have focused on the reasons behind the binge drinking decisions and damages related to z jm ht vb k alcohol use For instance, Dee (2001) argued that economic status causes binge drinking Rather than enjoying free time with beer or liquor, individuals drink more due to their bad economic conditions (e.g unemployment induces the likelihood of drinking) However, to the best of our knowledge, research of the effects of alcohol l.c gm om use on economic outcomes is still in its relative infancy Focusing on the consequence of drinking, the literature paints an opaque picture of alcohol consumption influences on economic outcomes; existing studies show conflicting directions of the causality between the alcohol consumption and income On the one hand, Pu et al (2008) argued that the consumption of alcohol an Lu ey t re 27 n In contrast, other studies found that drinking alcohol did not lead to any observable adverse economic consequences for either gender after employment was taken into account (Feng et al 2001) Ormond and Murphy (2016) even explored a positive correlation between income and the level of alcohol consumption The aforementioned controversies in empirical studies conceive the fact that the alcohol consumption – income va and tobacco together reduces living standards and overall wellbeing; the least wealthy groups are most severely affected by consuming these commodities In the United States, Bouchery et al (2011) estimated that the economic loss to income per capita due to drinking was US$746, of which the damage resulting from labor productivity decline was the largest (over 70%), followed by medical expenses (10%) and court costs (9%) nexus is relatively complex and may be conditional on other factors The lack of consensus induces us to test a hypothesis whether there is a negative effect of expenditure of alcohol in the case of Vietnam In this paper, to shed light on the alcohol consumption – income link, the rural Mekong River Delta (MRD) t to ng hi of Vietnam over the period 2010–2014 serves as our case study With a much less productivity than the national average1, MRD is still agriculture-based region where agriculture, forestry and fishery contribute a lion’s share to the total income This salient feature of the economy reinforces a low living standard of MRD households Nonetheless, alcohol consumption in the region as well as the whole country tends to be persistently going up ep (e.g Bao Giang et al 2013, Lincoln 2016) We base our investigation of the alcohol use – income association on an application of a system of simultaneous equations, being the three-stage Least Squares model (3SLS hereafter) A key reason behind the w n chosen methodology is that 3SLS can account for a potential endogeneity caused by bidirectional correlation between income and alcohol use Data for our study is extracted from the Vietnam Household Living Standard Survey (VHLSS) which is multi-purpose and most reliable data in the case of Vietnam The remainder of paper proceeds as follows Section introduces the background of alcohol consumption lo ad y th ju and its multiple impacts in Vietnam Section focuses on data and methodology in more detail Results and discussion is in Section 4, following by suggestions in the conclusion yi pl n ua al A Brief Review of Alcohol Consumption and Its Consequences in Vietnam n va For the case of Vietnam, there is a dearth of economic research on alcohol consumption effects Less than ten relevant publications on alcohol, of which none are in the economic area, in the Vietnamese language and limited international studies on the economic cost of alcohol consumption were found The literature focuses fu ll primarily on the drinking level, suggestions for and implications of binge drinking prevention without any economic assessment of the alcohol consumption effects on income and household expenditures For example, Nguyễn Hiền Vương and Phạm Việt Cường (2016) found that, at the mean, a drinker consumed 12.4 litres in 2013 and adults aged 25–34 drank the most (15.76 litres) The consumption of alcohol doubled in only five oi m at nh z years (3.8 and 6.6 litres in 2005 and 2010 respectively) These figures substantially differ from the officially statistical data that demonstrates a steady in alcohol use over the period 2006–2014 GSO (2016) reported that urban and rural residents drank the same total amount, at approximately 10.8 litres per month in 2014 Higher income quintile consumed alcohol notably more than the lowers z jm ht vb k Table Official data on drinking and the Vietnamese alcohol market 2006 2008 2010 gm 2014 8.4 8.4 10.8 10.8 10.8 Annual liquor supplied (million litres) 330.9 318.1 312.7 Annual beer supplied (million litres) 2420.2 2978.7 Year 2012 om l.c Annual alcohol consumption per capita (litres) 3287.2 an Lu Sources: (GSO 2016, pp.216&790-92) ey 28 t re It is only about one-third of the whole economy’s average (GSO 2017, p.146) n As such, buyers only take a 10-minute walk for a deal (Trần Thị Đức Hạnh et al 2016) Lincoln (2016) claimed that more than 250 million litres of unregistered strong liquor were produced at home in 2007 Home-made alcohol soared by 15% in 2006–2010 This untaxed and underestimated production maintains cheap prices, and thus stimulates the demand for the alcoholic products va Free access to the alcohol market is one of the reasons for the increased consumption over time It is critiqued that there was an absence of efficient official prohibition on the alcohol sales to young drinkers although several laws and regulations for alcohol production and sales were illustrated (Bao Giang et al 2013) Regarding the consequences of drinking, Nguyen Huu Minh (2016) argued that binge drinking was one of the fundamental causes of marriage problems Male married drinkers are more likely to lose their self-controls and make a half of the total number of domestic violence In addition, Bao Giang et al (2013) found that, in the t to ng households having heavy drinkers, drinking accounted for a greater share of the total consumption expenditure than the average; it means that they sacrificed other items in the consumption portfolio for the drinking purpose hi ep Data and Methodology w 3.1 Data n lo Data are extracted from the Vietnam Household Living Standard Survey (VHLSS) spanning 2010–2014 We choose VHLSS because of two main reasons First, although unofficial data from several hospitals reported that drinking caused the majority of traffic accidents and fatalities, exclusive data for alcohol consumption is not readily available for the whole country, including MRD In addition, several studies using anthropological ad ju y th yi approach claims that interviewed drinkers were reluctant to point out possible bad effects of drinking but the information on drinking decisions instead (Luu et al 2014) Second, VHLSS provides multiple characteristics of households including variables of interest (i.e alcohol use), education, employment and income that suit an analysis of alcohol consumption – income relationship pl ua al n The panel data structure of VHLSS provides us an additional examination of the alcohol consumption effect as well In this paper, the data are used in two ways We firstly use the full cross-sectional data which demonstrate n va fu ll the research outcomes year-by-year separately The sample size for each wave fluctuates around 1400 households for MRD Then, from the full sample size, we select retrospective households who were interviewed twice continuously and group them in the two pooled subsamples 2010–2012 and 2012–2014 The subsample sizes however fall considerably to only over 400 observations; therefore, they are treated as pooled oi m nh at cross-sectional instead of panel data to guarantee the sample size representation This choice is also consistent with a 3SLS model z z vb 3.2 Problems of Endogeneity jm ht k Before embarking on an analysis of the relationship between alcohol consumption and household income, this paper addresses the issue of potential endogeneity The endogenous bias may ensue from the dependent variable (household income) and independent variable (alcohol consumption) dual relationship, missing variables or/and measurement errors A consensus is that income causes drinking and vice versa (Ormond and l.c gm om Murphy 2016) Another potential cause of endogeneity is observable or omitted variables that correlate with the white noise error term Data constraint could be a weakness with respect to carrying out endogeneity The measurement errors are also a trouble which affects the estimate of correlation coefficients Endogeneity in the Ordinary Least Squares (OLS) estimators violate the assumption of non-covariance between independent an Lu ey 29 t re 3SLS model tackling the endogeneity problems is feasible, which is discussed in the following section n A common econometric strategy to avoid biased and inconsistent estimates caused by endogeneity is to find an instrumental variable that correlates with the explanatory variable but not with the dependent variable (Gujarati 2004, Gujarati and Porter 2009) This strategy is called Instrumental Variable (IV) model However, the appropriate instrumental variable is less likely available in practice due to insufficient data Therefore, a va variables and the disturbance term Once this condition is not held, econometric results are biased and inconsistent in OLS (Gujarati 2004, Gujarati and Porter 2009, Wooldridge 2009) This problem is called simultaneity bias 3.3 Model and Variable Specification The bidirectional relationship between alcohol consumption and income is carried out by the 3SLS t to ng hi estimators developed in Zellner and Theil (1962) This system concisely consists of three steps: First, it computes predicted values of each endogenous variable (functioned as the dependent variable) on the exogenous variables in the system This stage is identical to the first step in the Two-stage Least Squares (2SLS) estimators Then, using the residuals of structural equations obtained from the first step, the metrics generates ep a consistent calculation for the covariance matrix of equation disturbance Finally, applying a generalised least squares method to the covariance matrix obtained in the second stage, 3SLS estimates the parameters of righthand-side endogenous variables using the instrumented values (Stata Corporation n.d.) To specify the income per capita – alcohol use model, the system of two simultaneous equations is w n lo constructed as follows: (1𝑎) ad ln(𝑦𝑖 ) = 𝛼𝑖 𝑞𝑖 + ꞵ0 + 𝑋𝑖 ꞵ + 𝑢𝑖 where: ln(𝑦𝑖 ): the logarithmic form of real monthly income per capita; ju y th yi 𝑞𝑖 : household alcohol consumption (litres) per month, the variable of interest, endogenous with ln(𝑦𝑖 ); 𝑋𝑖 : a vector of household characteristics; ꞵ : the correlation coefficients of X on ln(𝑦𝑖 ); pl al n ua ꞵ0 : the model intercept; 𝛼𝑖 : the correlation coefficient of alcohol consumption on household income; 𝑢𝑖 : the disturbance term; and 𝑞𝑖 = 𝛾𝑖 ln(𝑦𝑖 ) + ξ0 + 𝑍𝜉 + 𝑣𝑖 (1𝑏) where: 𝛾𝑖 : the correlation coefficient of household income on alcohol consumption; ξ0 : the model intercept; n va ll fu oi m nh at 𝜉: the correlation coefficients of Z on 𝑞𝑖 ; Z: the control variables including education, household size, the age and gender of household head; 𝑣𝑖 : the disturbance term In the system of equation (1a) and (1b), 𝑦𝑖 and 𝑞𝑖 are jointly determined or they are endogenous variables z z ht vb k jm For the first direction, the central focus of this paper is the effect of alcohol consumption on income We want to check the extent to which alcohol drinking affects on income However, the correlation coefficient will be biased if the estimation is only based on equation (1a) due to the inverse effects The second direction, as gm om l.c described in equation (1b), concedes a probability that higher income could nurture consumption expenditure on alcohol The statistical yearbook of Vietnam reveals that the wealthiest quantile of population drank twice as much as the poorest one (GSO 2016) Lincoln (2016) emphasised that the rate of alcohol consumption escalated monotonically as a result of increasing wealth Thus, the dual directional effects are estimated an Lu simultaneously in this analysis In addition to endogenous variables, the control variables, X and Z, which are assumed to be exogenous2 are added in the model n va ey t re This condition will be checked with the Hansen-Sargan test after estimations 30 Results and Discussion 4.1 Descriptive Statistics t to ng We firstly observe the entire data from VHLSS for the rural MRD over three waves separately (2010, 2012 and 2014) The number of observations for this region fluctuates from 1199 to 1455, which accounts for 13–16% hi of the total VHLSS sample size These data show opposite trends in the growth of income per capita and wine consumption Income nearly doubled whereas wine consumption reduced significantly It is doubtful that the fall in wine consumption could be a statistical error as it contrasts the national trend reported by GSO Therefore, this study carefully checks with the pooled data from the two panels (2010–2012 ep w n and 2012–2014) Unlike the full sample data, the subsample pooled data 2010–2012 and 2012–2014 show that alcohol consumption remained nearly unchanged (about litres over weeks prior to the survey) irrespective of a similar pattern in the income growth It is noting that we could unable to create the only one 2010–2014 panel due to seriously observational drops that impacts the econometric modelling and research results lo ad y th ju Real income per capita increased rapidly from 1280 to 2238 (thousand Vietnamese Dong (’000 VND hereafter)) in 2010–2014, of which income rose in the first two years faster than in the later phase The regional average income was below the national level (from 1507 to 2497 respectively) as the agricultural activities were the main drivers of the household income The data from the two pooled subsamples also confirm an yi pl ua al n expansion in income, albeit with a less impressive rate Wine consumption per household indicates a sophisticated issue of drinking in the region It is used as a proxy for alcohol consumption in this study as (rice) wine is traditionally used on a regular basis in rural Vietnam n va ll fu While the entire sample shows a remarkable decline in wine consumption, it only decreased marginally according to data from the subsamples The drinkers in the rural MRD drank more than the average (1.85, 1.95 and 1.54 litres in 2010–20144) It is expected that wine consumption negates income Schooling of household head increased steadily over time from 5.2 to 5.5 years irrespective of the type of data oi m nh at used As the average years of schooling in national rural areas steadied from 6.6 to 6.7 5, there was a remarkable gap in educational achievements between rural Vietnam and MRD households Education is expected to contribute positively to income Gender of household head informs that three-fourths of the households were led by male in MRD, which z z vb k jm ht shares a similarity with the other rural areas across Vietnam The literature reveals ambiguous relationship between this characteristic and income as there is an equal number of evidences of both negative and positive association in the income – (fe)male head link Analogously, the household size may negate the household income per capita with respect to the number of household nonworking members; however, more employed gm om l.c working members could raise earnings Obviously, income can be proportionately influenced by household head’s age that is considered as the return of experience on income Thus, a positive effect of this independent variable on the dependent one is expected an Lu Table 2: Descriptive statistics for model variables 2010 2012 VHLSS 2010–2014 VHLSS 2010–2014 VHLSS 2010–2014 31 1895.1 1678.5 1881.5 1713.6 2168.9 1536.4 ey 1341.1 1152.1 t re Observation 1455 1199 1440 Monthly real per capita income (’000 VND) Mean 1280 1816.9 2238.3 SD 1078.9 1657.7 2026.3 2012 – 2014 panel 2012 2014 218 218 n 2010 – 2012 panel 2010 2012 231 231 2014 va Year Year 2010 2012 2010 – 2012 panel 2010 2012 227 253 8172 18095 2014 t to ng hi Min 113 253 199 Max 17118 18095 24872 Monthly wine consumption per household (litres) Mean 2.02 1.97 1.61 2.06 SD 3.75 2.89 3.18 2.78 Min 0 0 Max 60 30 50 20 Schooling of household head (ranging from to 12 years) Mean 5.2 5.41 5.53 5.24 SD 3.4 3.42 3.53 3.43 Min 0 0 Max 12 12 12 12 Gender of household head (male = 1; female=0) Mean 0.75 0.81 0.75 0.82 SD 0.43 0.39 0.43 0.38 Min 0 0 Max 1 1 Household size (the total number of household members) Mean 3.92 4.4 3.78 4.26 SD 1.50 1.6 1.48 1.45 Min 1 1 Max 12 11 10 10 Age of household head (years) Mean 49.06 50.79 51.88 50.03 SD 14.12 13.16 13.73 13.65 Min 11 24 18 23 Max 94 89 94 85 ep w n lo ad ju y th yi pl n ua al n va ll fu oi m 2.03 2.83 20 1.89 2.50 12 2.01 3.55 30 5.36 3.32 12 5.41 2.17 10.5 5.5 3.41 12 0.82 0.38 0.78 0.41 0.77 0.41 4.45 1.63 11 4.23 1.57 11 4.04 1.52 10 51.31 13.42 24 87 50.69 12.57 24 87 52.16 12.33 26 88 at nh Source: VHLSS 2010–2014 2012 – 2014 panel 2012 2014 253 199 18095 10826 4.2 Income – Alcohol Consumption Relationship z z The effect of wine consumption on income is captured by a 3SLS model for both with the full three waves of VHLSS 2010–2014 and the subsamples of the two pooled 2010–2012, 2012–2014 Generally, this study ht vb k jm explores a reduction in income per capita due to consuming wine For the full sample of three waves, Table shows that rural MRD households’ income growth was statistically significantly hurt by alcohol drinking, albeit this effect cannot pass the over identification test for the year 2012 Both in the two years 2010 and 2014, we find a substantially significantly negative effect of drinking on income per capita at the 1% level In 2010, l.c gm om other things being equal, at the mean, an additional litre of wine consumption decreased 55% of income per capita growth This strong link between the two variables was still confirmed with a half of the initial level after four years However, Table shows a smaller pattern in the negative correlation coefficient between the alcohol an Lu ey 32 t re cut their self-control jobs on the field for the drinking purposes This also means that there were insufficient n This stringent impact of wine use could arise from the double interchangeable reasons People who drink frequently could be unable to work efficiently even when they are not heavy drinkers Specifically, in rural MRD where the majority of earnings depend on agricultural activities, households can easily make a delay or va consumption and income For example, in 2010, a litre of wine consumption rise, ceteris paribus, wiped out 20% of income growth Then, in 2014, it took out more a tenth of individual earning expansion At any time, individual earnings in rural areas across the country could be more seriously cut by alcohol consumption than in MRD, albeit the over identification test is not hold for whole country data sample employments in the agricultural sector More importantly, intermediate and heavy drinking causes various health problems which interrupt the employment or working time, and thus induce a fall in income growth 4.3 Other determinants of income growth t to ng hi Turning now into the income drivers First, the similar correlation coefficients from both the full and subsample data confirm that education is of importance to income growth Specifically, a year higher in household’s head educational level can explain for 7.4% and 5.2% increase in income per capita in 2010 and ep 2014 respectively This result reaffirms the fact that education of the household head’s schooling improves not only individual income but also the other family members’, especially their children’s earnings (e.g Nguyen et al 2007, Hoang and Yabe 2011) Badiani et al (2012) also argued that income has a strongly positive relationship to education, and thus improving educational level is a long-run strategy of poverty alleviation w n lo ad The gender of household head generally shows a statistically significant correlation to income growth With the full data, income growth was more likely to rise in the male-led households In particular, those who lived in the male-led households boosted 40% of their income compared with people belonging to female-led ju y th yi households The correlation coefficient between income growth and the household heads’ gender was less sizable in the year 2014, but income increased by over one-fifth in the households with the male head However, this income and household head’s gender association was complicated when using the data from subsamples For the pooled panel 2010–2012, the correlation coefficient was neither statistically significant nor pl ua al n positive This paper cannot find any rigorous support the exceptional outcomes for MRD when the model for the national rural areas was positively significant at the 1% level Nevertheless, a limited sample size could circumvent the validity of this computation It came as no surprise then that the proportionate correlation was statistically significant n va ll fu oi m Finally, the household size positively contributed to income extension in spite of no statistically significant results in 2014 It is argued that a greater ratio between working and non-working members could underlie this result The long-run family planning strategy of the Vietnamese Government reinforces the fertility rate fall, and hence the number of children per household reduced monotonically, including the examined period at nh z z Table 3: Results from 3SLS model for the full cross-sectional data vb Source: VHLSS 2010–2014 33 6618 ey results from equation (1b) are suppressed and could be provided upon requests 1440 t re Note: *: p