Article Text

Tobacco expenditure and its implications for household resource allocation in Cambodia
1. Rijo M John,
2. Hana Ross,
3. Evan Blecher
1. International Tobacco Control Research, American Cancer Society, Atlanta, Georgia, USA
1. Correspondence to Rijo M John, International Tobacco Control Research, American Cancer Society, 250 Williams Street, Atlanta, GA 30044, USA; rijo.john{at}cancer.org

Abstract

Objectives To assess the determinants of smoking behaviour and to estimate the impact of tobacco consumption on the consumption of other commodities by Cambodian households.

Methods To assess the determinants of smoking in Cambodia, the authors used a logistic regression model that estimated the probability of an individual smoking, given a set of socioeconomic and demographic characteristics. A Seemingly Unrelated Regression method was used to assess the impact of tobacco consumption on the consumption of other commodities. The nationally representative 2004 Cambodia Socio-Economic Survey, collected by the National Institute of Statistics of the Ministry of Planning in Cambodia, was used for the analysis.

Results Smoking in Cambodia is influenced by a variety of factors such as gender, marital status, age, ethnicity, literacy, health status and perceptions about the health consequences of tobacco use. The authors found that spending on tobacco crowds out expenditures on education and clothing at the national level and expenditures on food for low- and middle-income households.

Conclusions The first analysis of the study showed that increased education is associated with lower daily smoking, and the second analysis revealed that expenditures on tobacco crowds out expenditures on education. Combining these two results points to a vicious circle where low education means higher likelihood of smoking, which in turn results in lower spending on education. Such budget allocation clearly has negative intergenerational consequences.

• Tobacco
• smoking
• expenditures
• crowding out
• Cambodia
• economics
• public policy

Introduction

Tobacco consumption is a serious public health concern in Cambodia since 49% of men and 21% of women use tobacco (both smoking and chewing).1 Estimates from the 2004 Cambodia Socio-Economic Survey (CSES)2 indicate that 33% of men and 3% of women (15 years and older) are daily smokers, while 37% of the non-smoking respondents were exposed to secondhand smoke.3 There is a large gender disparity in chewing tobacco use since 17% of the women consume chewing tobacco, while only 1% of the men do so. On the other hand, 48% of men smoke cigarettes, while only 4% of women do so. Of the 1.9 million adults who currently use tobacco in Cambodia, 1.2 million are men who smoke, 560 000 are women who chew tobacco and the rest are female smokers and male chewers of tobacco.1 Previous research shows that the smoking prevalence in Cambodia increases with age, is higher among ethnic minorities and is influenced by older smoking relatives.1 Smoking prevalence decreases with increased education and is lower in urban areas.2

In addition to its impact on public health, tobacco use is associated with poverty among low-income households. This issue is particularly relevant in Cambodia where, in 2004, 35% of the population lived in poverty (defined as having <2351 Riel (US$0.59) income per person per day in the capital Phnom Penh, <1952 Riel (US$0.49) in other urban areas and 1753 Riel (US$0.44) in rural areas4). Cambodia ratified the Framework Convention on Tobacco Control in November 2005.5 A national action plan for its implementation has been developed for 2006–2010 and 2011–2015, respectively. However, research, which generates evidence for the support of public health policy, is limited. The objectives of this paper are (1) to assess the determinants of smoking behaviour and (2) to estimate the impact of tobacco consumption on the consumption of other commodities by Cambodian households. Data and methods We used data from the 2004 CSES4 collected by the National Institute of Statistics of the Ministry of Planning. The 2004 CSES is a nationally representative household survey collecting data from 15 000 households in the five regions of Cambodia: (1) Phnom Penh, (2) Plain Region, (3) Tonle Sap Region, (4) Coastal Region and (5) Plateau Mountain. We focused on expenditure data for >600 durable and non-durable goods and services along with households' and individuals' socioeconomic and demographic characteristics. After eliminating individuals with missing data and those under the age of 14 (smoking information was not reported for the under 14 age group), our analytical data set contained 14 984 households with 47 717 individuals. To assess the determinants of smoking in Cambodia, we used a logistic regression model that estimated the probability of an individual smoking given a set of socioeconomic and demographic characteristics (equation (1)).Pr(y=1|x)=xβ+ɛ(1)where the dependent variable y represents a binary indicator for an individual's smoking status (1 if daily smoker, 0 otherwise; only 6% of the smokers were infrequent smokers and were excluded from our analysis). The set of independent variables (x) included age, gender, marital status, whether the person is the head of the household, ethnicity, whether the person is staying in his/her permanent home at the time of interview, school attendance, literacy, work status, health status (a self-report assessment of health as good, bad, average or do not know) and perception of the harmful effects of smoking. All variables with the exception of age are discrete. Even though price is an important determinant of smoking,6 it was not included in this regression for two reasons. First, the price paid for cigarettes was not collected by the survey. Second, cigarette prices do not vary significantly across the country in any given cross section because cigarette taxes are uniform across the country. Moreover, the data here belong to only one time period and, hence, we do not expect any time variation in prices. Thus including an external price measure would play a role similar to a constant rather than a variable. The second model estimated the impact of tobacco consumption on household expenditures on other commodities, also referred to as the crowding out effect. The crowding out in the context of this paper is defined as reduced consumption of goods and services as a result of tobacco consumption. The 605 household expenditure items collected by the survey were classified arbitrarily into nine broad categories: tobacco, food, clothing, durable goods, healthcare, entertainment, education, housing and other. The difference in mean budget shares devoted to each expenditure category between households with and without tobacco expenditure was assessed by an independent Student t test.7 Variances were unequal for all expenditure categories except housing. A Welch t test8 was performed for all unequal variance cases. However, this test does not take into account households' socioeconomic and demographic characteristics that could influence their spending pattern. Therefore, we used a regression which predicted the budget share allocation to each expenditure category according to tobacco use status controlling for household-specific characteristics (equation (2)). The model is derived from a Quadratic Almost Ideal Demand System (QUAIDS)9 10 that allows the same good to be either a luxury or a necessity based on a household's income level. This is achieved by including a quadratic income term on the right-hand side of the equation.9 For example, certain durable goods or certain types of clothing may be necessities for higher income individuals, while the identical items may be considered a luxury for lower income individuals. Since the household budget allocation to one expenditure category is correlated with expenditures on other categories, the error terms in the budget share equations are likely to be correlated, which would lead to increased variance in the estimated coefficients. This may increase the SEs and thereby affect the statistical significance of the estimated coefficients. This is addressed by employing a Seemingly Unrelated Regression model11 that generates efficient regression coefficients.12 The Seemingly Unrelated Regression model estimates all regression equations simultaneously using a Feasible Generalized Least Squares method. We chose the ‘other’ expenditure category as a benchmark and dropped it from the system of simultaneous equations to ensure that summation restrictions are met. wi=α+βT+ϑX+ɛ(2) where wi is the budget share on the ith expenditure category (other than tobacco), T is a dichotomous variable with the value 1 if the household has positive expenditures on tobacco (0 otherwise) and X is a vector of household's socioeconomic and demographic characteristics (log of total expenditure, log of total expenditure squared, household size and characteristics of the household head (gender, marital status, ethnicity, literacy)). The control variables used in equation (2) are different from equation (1) since the first analysis was performed at the individual level, while this analysis used household-level data as the budget allocation occured at the household level. Given the different characteristics of the urban and rural life, the analyses of equations (1) and (2) were conducted separately for urban and rural areas. We also constructed a variable that categorises households according to the importance of tobacco in their total budget. A binary variable distinguishes households with and without tobacco expenditures. Conditional on having tobacco expenditures, households below the 25th percentile of budget shares devoted to tobacco were defined as low-tobacco spending households. Similarly, households between the 25th and 75th percentile and above the 75th percentile were categorised as moderate- and high-tobacco spending households, respectively. All households were also classified into three income groups (low, middle and high income) using the percentile approach described previously. These categories were all created to present descriptive statistics (table 1). The regression analyses did not use the income classifications within rural and urban locations due to lack of sufficient sample size. Table 1 Annual household expenditures and budget share allocated to tobacco by the residence and tobacco spending status in Cambodia, 2004 Results Table 1 summarises the household expenditure and budget shares allocated to tobacco by the rural/urban location and tobacco spending status. An urban household that consumes tobacco spends, on average, 3.6% of their annual budget on tobacco which amounts to 145 988 KHR (US$36.2). Its rural counterpart spends, on average, 2.8% of the budget on tobacco or 11 338 KHR (US\$2.8) per year. The budget share for tobacco ranges from 0.4% among low spenders to 8.4% among high spenders in urban areas and from 0.4% to 6.9% in rural areas. With the exception of high spenders in the high-income category, the average budget share spent on tobacco is similar for different income groups within the same spending category. Although similar results have been found in India,10 the poor are known to spend a higher share of their budget on tobacco than the rich in many other low-income countries.13–15

Determinants of smoking

The regression results of equation (1) are shown in table 2. For a continuous variable, the OR shows the effect of a unit change in the value of that variable, whereas for a categorical variable, it shows the effect of a change from the reference category to the category under consideration. The results confirm some of the findings from the prior research.1 3 16 As age increases by 1 year, the odds of an individual becoming a daily smoker increase by two and three percentage points in urban and rural areas, respectively. In urban areas, the odds for a man to be a daily smoker are 18 times higher than that of a woman and 25 times higher for men in rural areas. Married and widowed individuals are significantly more likely to smoke than those never married (in both areas), while those divorced and separated are only more likely to smoke in rural areas. Being a household head increases the probability of smoking in both rural and urban areas. The Khmer are the largest ethnic group in Cambodia accounting for approximately 90% of the population. All ethnic groups other than the Khmer have significantly higher odds of smoking. Specifically, compared with the Khmer, the odds of smoking among the indigenous hill tribes (such as the Kouy, Souy and Chaaray) are 35 times higher.

Table 2

Determinants of smoking in Cambodia

Individuals who never attended school have a 60% and 22% higher probability of smoking in urban and rural areas, respectively, than those who attended school. However, illiteracy has an even larger effect on the probability of smoking (91% and 54% in urban and rural areas, respectively). Working in the previous week (which is a proxy of income) significantly increased the probability of smoking in both rural and urban areas. Increased disposable income may have contributed to this effect. The perception of an individual's own health status as ‘good’ is associated with a lower probability of smoking compared with those who perceived their health as ‘average’. Conversely, those who perceived their health as ‘bad’ were significantly more likely to smoke. The perception of the impact of smoking on one's health is also correlated with smoking status. Those perceiving smoking as ‘not harmful’ to their health and those who ‘did not know’ if smoking was harmful have a significantly higher probability of smoking than those who perceive smoking to be ‘harmful’.

Crowding out effect of tobacco consumption

Table 3 shows the results of the Student t test for the differences in budget allocation among households with and without tobacco expenditures for the nine budget categories. The null hypothesis is that the difference between these two household types is zero. In urban areas, tobacco-consuming households allocate significantly lower budget shares to clothing and education than non-tobacco-consuming households. In rural areas, tobacco-consuming households allocate significantly lower budget shares to food and education and a higher budget share to entertainment than non-tobacco-consuming households.

Table 3

Student t test for the difference in budget allocation between non-tobacco and tobacco using households

Although the t test provides a formal test for the differences in mean shares, it does not control for other household-specific characteristics that may affect budget allocation. We use the regression model in equation (2) to control for these characteristics. Table 4 reports the coefficients of the dichotomous tobacco consumption variable representing tobacco expenditures for each budget share equation in the system. Coefficients of the control variables are not reported for the sake of brevity. A negative coefficient signals a lower budget allocation among tobacco-consuming households than non-tobacco-consuming households.

Table 4

Crowding out effect of tobacco consumption in Cambodia by place of domicile

Tobacco-consuming households allocate significantly lower budget shares to education and clothing, although the coefficient for clothing is only statistically significant in urban areas. On average, the budget share allocation to education is 1.6 and 0.6 percentage points lower among tobacco-consuming households in urban and rural areas, respectively. In urban areas, tobacco spending also led to a 0.5 percentage point reduction in the budget share allocated to clothing. Tobacco-consuming households in rural areas allocate a significantly higher budget share to entertainment and housing.

We conducted a secondary analysis similar to the one presented in table 4 but instead of breaking down the results by urban and rural areas, we broke down the result by income groups (table 5). The results show that expenditure on tobacco significantly reduces the budget allocation to food among low- and middle-income households but not among high-income households. Spending on tobacco reduces the budget share devoted to education; however, the results are not statistically significant for the low-income households. High-income households also reduce their budget allocation to clothing as a result of their tobacco consumption. Tobacco-consuming households tend to devote larger budget shares to entertainment (results significant for low- and high-income households only) and to housing (results significant for low- and middle-income households only).

Table 5

Crowding out effect of tobacco consumption in Cambodia by household income groups

Discussion

Smoking in Cambodia is associated with a variety of factors such as gender, marital status, age, ethnicity, literacy, health status and perceptions about the health consequences of tobacco use. We found that spending on tobacco crowds out expenditures on education and clothing for rural and urban households and expenditures on food for low- and middle-income households. Similar results have been found in other low- and middle-income countries such as India,10 China,17 18 Indonesia19 and Taiwan.20

Our first analysis showed that increased education is associated with lower daily smoking, and the second analysis revealed that expenditure on tobacco crowds out expenditure on education. Combining these two results points to a vicious circle where low education means higher likelihood of smoking, which in turn results in lower spending on education. This budget allocation has clear negative intergenerational consequences.

The opportunity costs of tobacco use in Cambodia are significant. On average, households in Cambodia devote 4% of their total budget to tobacco. In a country where 35% of the population lives below the national poverty line such expenditure has negative consequences beyond the damage that tobacco use imposes on people's health. We found that the poor- and middle-income households reduce spending on food in order to buy tobacco. At the same time, buying just a half pack of a premium cigarette brand (such as 555) or two packs of the most popular domestic brand (ARA) will exhaust the daily budget necessary to be above the national poverty line.4 The amount of money spent on one pack of cigarettes can buy as much as 3500 food calories comprising a typical Cambodia daily diet according to CSES 2004.

Although this study sheds light on the determinants of smoking and the effect of tobacco expenditures on household budget allocations, it has some limitations. First, the price of cigarettes was not available at the individual level and could not be included among the determinants of smoking considered in our analysis. Second, CSES 2004 was the latest survey that was available to us. Although the 2009 CSES survey has been completed, we could not get the unit-level data to perform the analysis in this study. Third, given the aggregate nature of the expenditure categories, we were not able to study the effect of tobacco expenditures on specific items of consumption.

Even though Cambodia ratified the Framework Convention on Tobacco Control in 2005, advertising and promotion of tobacco products and sponsorship of cultural events by the tobacco industry have intensified.21 The National Law on Tobacco Control was proposed in 2007 but, as of August 2011, has not been adopted. Concerted efforts need to be made in order to integrate the obligations of the treaty into the legislative framework and to allow for its implementation locally. A number of tobacco control policies exist in Cambodia such as educational media campaigns, smoking cessation programs, health warning on cigarette packs and smoke-free areas in workplaces.22 However, the implementation and enforcement of these policies has been weak.23

Internationally proven methods of tobacco control such as raising the price of tobacco products by raising taxes24 must be given a high priority.22 Cigarette taxes in Cambodia are among the lowest in the ASEAN region.25 Increasing tobacco taxes will have multiple benefits such as reducing consumption, increasing government revenue and reducing the opportunity cost of tobacco use. Our study shows that poor households in Cambodia will benefit from reducing their tobacco consumption since they would have higher disposable income, which can be spent on food, education and clothing. Tobacco control, therefore, should be an integral part of the government's poverty alleviation strategy.

What is already known on this subject

Smoking is known to be associated with certain socioeconomic factors such as gender, age and ethnicity in Cambodia. Despite the high prevalence of smoking and high rates of poverty in Cambodia, there are no studies that quantify the effect of tobacco expenditures on the consumption of other commodities or the determinants of smoking in greater detail.

This study found that smoking in Cambodia is influenced by a variety of factors such as gender, marital status, age, ethnicity, literacy, health status and perceptions about the health consequences of smoking. It also found that spending on tobacco crowds out expenditures on education and clothing at the national level and expenditures on food for low- and middle-income households.

Acknowledgments

We would like to acknowledge Mr. Lundy Saint at the National Institute of Statistics, Cambodia, for providing us the data used in this paper and Dr. Yel Daravuth at the WHO Tobacco Free Initiative and Health Promotion for his comments.

View Abstract

Footnotes

• Correction notice This article has been corrected since it was published. The title has been amended to read ‘Tobacco expenditure and its implications for household resource allocation in Cambodia’.

• Competing interests None.

• Provenance and peer review Not commissioned; externally peer reviewed.

Request permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.