2
ABSTRACT
Worldwide increasing overweight and especially obesity place a heavy burden on the individual’s physical and emotional well-being, but also pose an externality - large public health costs. In an attempt to reduce the magnitude of this problem, thereby yielding significant health and economic benefits, this paper introduces the concept of a tax based on Body Mass Index, which provides a financial incentive to improve health behaviour and internalizes the costs of poor health decisions. To make the first step towards assessing the potential of a BMI tax, this paper uses MEPS US data sets to study the relationship between experiencing direct costs from overweight/obesity and BMI compared to not experiencing direct costs from overweight/obesity and BMI. Evidence suggests BMI levels are lower for individuals who face the direct costs of their poor health decisions when controlling for, and matching on, demographic factors. Using panel data to perform a differences-in-differences analysis over the period of one year did not yield an explanation for the difference in BMI levels between those experiencing and those not experiencing direct costs of poor health behaviour. Further research that uses long-term, larger-scale panel data and includes factors on nutrition and exercise is recommended.
Key words: overweight, obesity, policy, BMI, tax
4
1. INTRODUCTION
Whether you see it in the streets, on television, or in the mirror: obesity has grown big and attracts large attention everywhere. Applying the World Health Organization definition, individuals with a Body Mass Index (BMI) that is equal to or exceeds 25 are overweight and when meeting or exceeding a BMI of 30, one is considered obese (WHO, 2012). While this definition is widely accepted, unity does not exist in the discussion on the causes, remedies and actors to be held responsible for overweight and obesity. Rivalling discourses blame either the individual’s behaviour (Huffman and Rizov, 2007), environmental factors (Jackson, 2003; WHO, 2003), or the individual’s genetic make-up (Hill and Peters, 2002; Taubes, 2008). Neither does consensus exist on obesity being a disease (American Medical Association, 2013), a risk factor (Heshka and Allison, 2001) or otherwise (Campos, 2004).
Notwithstanding the different outlooks on obesity, just like hips, numbers don’t lie. Worldwide, obesity has more than doubled since 1980 (WHO, 2012). In 2014, more than 1.9 billion adults (≥18 years of age) fell into the category overweight and over 600 million adults were obese. Recent research by the NCD Risk Factor Collaboration (2016) studying the trends in adult-body mass index in 200 countries from 1975-2014 finds that, if post-2000 trends continue, global obesity prevalence will reach 18% in men and surpass 21% in women by 2025.
In the literature, two comprehensive arguments prevail as to why these numbers are alarming. First, overweight and obesity place a heavy burden on human physical and emotional well-being (Doll et al., 2000; Fontaine and Barofsky, 2001) and bring about social and economic disadvantage for the overweight or obese individual (Lee, Harris & Lee, 2013; Sarlio-Lähteenkorva and Lahelma, 1999). Second, at times the employer, yet in general the taxpayer, eventually bears the external costs of obesity (Gates et al., 2008), which reduces public welfare.
5
reduce overweight and obesity rates have yet to prove their effectiveness (Capacci et al., 2012).
In an attempt to encourage policy makers to think outside the box, this paper assesses the idea of a new policy that may have potential to attack the obesity epidemic: a BMI tax. With a BMI tax, rational individuals pay the costs of their decisions regarding food intake and exercise; the external costs are being internalized. When faced with direct costs of one’s own health behaviour in the form paying more tax thereby reducing net income when being overweight, or obese, one may be incentivized to change health behaviour and lower BMI towards normal weight. Posing a BMI tax would mean deriving one’s income tax as a function of the deviation of BMI from a set target value. For the sake of simplicity, the target value will be in line with the WHO (2012), thereby set at a BMI level of 25 and 30. Thus, in addition to income tax, a BMI tax would be imposed on individuals with a BMI higher than 25. Starting at a BMI of 25, the tax percentage increases linearly per BMI point until a BMI of 30, from which every additional BMI point will increase the tax linearly yet steeper than when BMI is between 25 and 30.
In order to evaluate whether a BMI-based tax could be successful in bringing down BMI, this paper addresses the research question: what is the relationship between experiencing
direct costs from overweight/obesity and BMI compared to not experiencing direct costs from overweight/obesity and BMI? In other words, the empirical question is: how do consumers respond to BMI-related income shocks? Studying this relationship requires a suitable context
and measure.
Although obesity is no longer strange to developing countries, developed countries still account for the lion’s share. In the European Union, more than 30% of the adult population classified as overweight and an additional 17% as obese in 2014 (OECD/EU, 2014). In 2015, 34.9% of US adults were obese (Centers for Disease Control, 2015). Because of the US’ prominent stake in the obesity epidemic and the consequently extensive data availability, this paper is written in the context of the United States. It is reasonable to assume that the results should apply to similar developed countries.
6
not be explained by individuals who were confronted with the direct costs consistently losing weight, neither by individuals who were not confronted with the direct costs consistently gaining weight. Nor was evidence found that individuals who switched from not being confronted with the direct costs of poor health behaviour to being confronted with these costs immediately lose weight, and no evidence was found that individuals who make an opposite switch immediately gain weight. Without discussing details regarding the implementation of a BMI tax, neither addressing all ethical and political challenges, this paper aims to challenge the conventional wisdom.
This paper is organized as follows.
In section 2, I first elaborate on the motive behind a BMI tax. This starts with the reasons to intervene, including the negative effects of overweight/obesity on the overweight/obese individual, and the negative externalities caused by overweight/obesity. Next, I explain why one would use a tax to change health behaviour, discussing the use of public policy in general and financial incentives in particular, thereby referring to experience with food and beverage tax. Then, I clarify the idea behind taxing on BMI. Second, I review the effects of a BMI tax, thereby distinguishing a negative income effect and substitution effect. Thereafter, I discuss the factors that influence the trade-off between the effects, including income level, information availability and accessibility to healthy foods and exercise. Eventually, I briefly mention the role the tax revenue plays in this trade-off. Third, I explain the measure - how health insurance serves as a proxy to empirically test how consumers respond to BMI-related income shocks, resulting in testable hypothesis.
In section 3, I specify the data sources and data collection process, the sample, the variables, and the econometric techniques I apply to empirically test the hypotheses.
The results are presented in the fourth section.
In the fifth section the results are discussed, and limitations and suggestions for further research are presented.
7
2. LITERATURE REVIEW
2.1 The motiveWhy do we need to intervene in the first place?
v Negative effects of overweight/obesity on the overweight/obese individual
Let us begin with the consequential burden on human well-being. Overweight and obesity are associated with lower health-related quality of life (Fontaine and Barofsky, 2001) and poor levels of subjective health status, particularly in terms of physical well-being (Doll et al., 2000). Excess weight is associated with an increased risk of cardiovascular disease, type 2 diabetes (Colditz et al., 1995; Manson et al., 1992) hypertension (Witteman et al., 1989), stroke (Rexrode et al., 1997), dyslipidemia, osteoarthritis, and several cancers (Burton et al., 1985; Folsom et al., 1989; Lee et al., 1992; Sellers et al., 1992). Being obese compares to 20 years of aging when looking at the association with chronic health conditions (Sturm, 2002). Additionally, obesity has been found to decrease overall life expectancy (Peeters et al., 2003; Bhattacharya and Neeraj, 2011).
Apart from negative impacts on physical health, overweight and obesity contribute to social disadvantage. Obese people are often being exposed to prejudices, over-all in a negative way (Schermer, 2011), and associated with absence of close friends outside the family circle (Sarlio-Lähteenkorva and Lahelma, 1999). Also, heavier teenagers have been found to be less likely to date (Cawley, Joyner & Sobal, 2006). In addition, they are more vulnerable to discrimination in the areas of employment, education and health care (Puhl and Brownell, 2001), which translates into economic disadvantage. With regard to this economic disadvantage, obese people have been found to be less capable of effectively finding and keeping jobs and the earnings and wages they receive are likely to be lower, especially for women (Cawley, 2004). In women, overweight was associated with current unemployment and obesity with long-term unemployment, while both were associated with low individual earnings (Sarlio-Lähteenkorva and Lahelma, 1999). In addition to lower income, Mora, Gil and Sicras-Mainer (2013) find a positive and statistically significant impact of BMI, obesity and overweight on annual medical costs. The consequent health care costs for obese individuals are $395 per year higher than for their normal weight counterparts (Sturm, 2002). Overweight and obesity do not only bring about costs for the person in question, however.
v Negative externalities caused by overweight/obesity
8
annually (Cawley et al., 2007). Obese people have also been found to be less productive while at work, costing employers $506 per obese worker annually (Gates et al., 2008). In addition, obese employees put employers up with higher insurance premia, which are rather large in aggregate (Ford Runge, 2007).
Finally, obesity affects expenditures by local, state, and national governments, which raises public morality and turns obesity into an issue of public concern. When programs such as Medical Assistance, unemployment insurance, state-run health insurance, Veterans hospitals, Medicare, and Medicaid compensate for or cover some of the private and workforce costs of illness and unemployment, obese imposes higher health costs onto present and future taxpayers (Ford Runge, 2007). Each year, obesity induces preventable chronic diseases and healthcare costs in the US estimated to range from $147 billion to nearly $210 billion (Bassett and Perl, 2004; Cawley and Meyerhoefer, 2009; Finkelstein et al., 2009). Current trends project the costs associated with treatment of preventable diseases that result from obesity to increase by $48–66 billion per year in the US (Wang et al., 2011).
Concluding, everyone is affected by the increasing prevalence of obesity, either personally, economically, or both, which decreases public welfare (Novak and Brownell, 2012). This provides rationale for intervention aimed at changing health behaviour.
Why use a tax to change health behaviour?
v Public policy as a tool to change health behaviour
The high levels of economic development in developed countries enable the formation of a welfare regime (Wilensky 1975). Welfare states have responsibility for public health, with policy being the most powerful tool to influence the population’s health. Policies aimed at improving society’s health behavior can be categorized as supporting more informed choice, interventions changing the market environment, interventions not explicitly targeted at healthy eating, or generic interventions. Assessing 129 policy interventions, Capacci et al. (2012) find a strong bias towards less controversial informational and educational actions rather than market-level interventions. Evidence on the effectiveness of the studied policies suggests that the impact of education and information campaigns on attitudes and intentions is significant (Capacci et al., 2012). However, changing the perception of risk does not necessarily translate into behavioral response in relation to that risk (Treby and Clark, 2004), which is reflected by limited estimated actual effects on behavior.
9
Therefore this paper turns to the relatively unexplored area of the use of market-level interventions to change health behaviour. Because as the old adage has it: “When all else fails, try money.”
v The use of financial incentives
The theoretical rationale for the use of incentives to facilitate weight loss comes from operant learning theory and behavioural economics; both theories predict that changing the immediate consequences of body weight or behaviours determining body weight will induce changes of behaviours and changes in weight over time (Jeffery, 2012).
In essence, the two possible means of changing the population’s dietary habits using financial incentives are positive (subsidies) and negative incentives (taxes).
Positive financial incentives have found to be effective in limited circumstances where simple tasks have to be completed within a limited time-frame, while less effective where the behavioral change required is complex (Jochelson, 2007). Given that losing weight is a complex and time-consuming process, positive financial incentives seem to be inadequate. In addition, positive financial incentives require large funding sources and previous research finds that removal of the incentives is being followed by weight regain (John et al., 2012). Although it is difficult to compare the outcomes of positive and negative financial incentives, previous studies on smoking cessation and weight loss suggest that negative financial incentives are more effective (Jeffery et al., 1983; Jeffery et al., 1984; Jeffery et al., 1987; Jeffery, 2011; Goldsmith and Dhar, 2013). Therefore, this paper is confined to the use of taxes for inducing improved health behavior.
Research on the (long-term) efficacy of market-level interventions in the form of taxes is limited since their use in the real world is relatively recent. The most extensive body of evidence concerns food and non-alcoholic beverage taxes. Drawing upon experience with food and beverage tax provides useful insights into both the implications and potential of a BMI tax.
v Body of evidence – food and beverage tax
10
taxes in various states in the US found no significant association between taxes and the prevalence of obesity, which could be due to the period of the studies being too short to actually observe an effect on population health and/or the tax rate being too low (Lin et al., 2011).
The latter argument has re-emerged in multiple other studies, which can be explained by the price elasticity of food and beverages (Powell, 2009; Franck et al., 2013). The price elasticity of foodstuffs generally suggests that, on average, prices need to double to get a 10% reduction in consumption, as food products individually are relatively cheap (Cornelsen and Carreido, 2015). Andreyeva et al. (2010) reviewed 160 studies on the price elasticity of demand for major food categories and find elasticities for foods and nonalcoholic beverages ranging from 0.27 to 0.81 (absolute values). The categories that have been found to be most responsive to price changes (elasticity of 0.7-0.8) are food outside the house, soft drinks, juice, and meats. Another recent study finds that taxes on foods would need to be at least 20% to have a meaningful impact on health outcomes (Cornelsen et al., 2014).
A relationship between the price of food and body weight has been shown by numerous studies. Chou et al. (2004) find a negative relationship among adults between on one hand the real fast-food restaurant price, the real food-at-home price, and the real full-service restaurant price and on the other hand weight. Fast food prices have also been found to be negatively related to adolescent BMI (Powell, 2009). Goldman et al. (2009) find very modest short-term effects of price per calorie on body weight, as it takes a long time for the effect to reach its full scale: within 30 years, a 10% permanent reduction in price per calorie would lead to a BMI increase of 1.5 units (or 3.6%). The long term effect is an increase of 1.9 units of BMI (or 4.2%). From a policy perspective, these results suggest that, when interested in reducing body weight, using a tax to improve health behaviour may have little effect on weight in the short-run, but could achieve significant weight reduction in the long-run (Goldman et al., 2009).
Why tax on BMI?
11
frequently will not generate any externalities. In addition, contradictory to for example smoking, consuming junk food does not directly impair someone else’s well-being.
Unlike a food or beverage tax, which impacts those who enjoy soft drinks or a candy bar in moderation as well as over-consumers, taxing people on the basis of their BMI will directly target the individuals whose health behaviour decreases public welfare by pushing public healthcare costs; they are the ones that need to change behaviour.
The idea behind a BMI tax holds that overweight and obese persons will be directly confronted with the costs resulting from their BMI being beyond the threshold. Their net income will be lower than their normal weight counterparts. Adults (fully) paying for their body weight decisions creates a financial incentive to reduce BMI and reach a normal weight.
2.2 The effects
Up to this point, hardly any research has been done on people’s behavioural response when being directly confronted with the (financial) consequences of their (health) behaviour. Although it concerns a different kind of energy, natural experiments by McClelland (1980) find that conversions of property from master to individual metering lead to three to four times as much energy savings; suggesting that people are capable of changing behaviour when given the power and responsibility to be in charge of their own costs. Following this line of thought, a BMI tax may be the push overweight individuals need to start changing their health behaviour. Whether this change in health behaviour benefits a healthier lifestyle depends on which of the two theoretical effects of a BMI tax will dominate: the negative income effect or the substitution effect.
Negative income effect
The negative income effect the BMI tax is expected to have on the overweight/obese individual can become evident in two ways:
On the consumption side:
Ø Less disposable income causes people to consume less expensive foods.
12
healthier one actually decreases costs (Mitchell et al., 2000; Raynor et al., 2002). Hence, caution is required in comparing the cost of healthy versus less healthy food, as the use of different price metrics leads to opposing conclusions.
Measuring the price per calorie ($/calorie), research find that healthier diets cost $1.54/2000 kcal ($1.15 to $1.94) more than less healthy diets (Rao et al., 2013). By this metric, vegetables and fruits are low in calories and therefore a relatively expensive way to purchase food energy. Vice versa, less healthy foods are generally high in calories and tend to have a low price per calorie, especially when high in saturated fat and added sugar.
Contradictory, measuring on the basis of edible weight ($/100 edible grams) or average portion size ($/average portion), Carlson and Frazão (2012) find that healthier foods including grains, vegetables, fruit, and dairy foods tend to be less expensive than most protein foods and products high in saturated fat, added sugars, and/or sodium.
Yet, in reality, it is not only the price of the food itself; the availability or absence of kitchen facilities, cooking skills, and time are factors contributing to (the perception of) the price of a healthy diet compared to a unhealthy diet (Drewnowski and Eichelsdoerfer, 2009). On the income side:
Ø One needs to work more in order to maintain the pre-tax standard of living, thereby gives up leisure.
Abramowitz (2016) finds that for American workers in non-strenuous jobs, increasing working hours by 10 is associated with an increase in BMI of 0.424 for women (2.5 pounds) and 0.197 for men (1.4 pounds). No association was found for strenuous jobs. The increase in BMI is likely to be the result of greater work-stress, which has been found to result in more frequent fast food consumption (Bauer et al., 2012), while reducing physical activity levels (Ng and Jeffery, 2003).
Substitution effect
The substitution effect can also reflect itself in two manners: On the consumption side:
Ø Consumers switch from unhealthy to healthy foods.
13
strength primarily depends on the extent of information about the relation between consumption and weight available to consumers.
On the income side:
Ø As working becomes relatively less profitable than leisure, the overweight individual gives up working hours.
Conversely to the argument made earlier, decreasing working hours may reduce working-stress levels, thereby fast food consumption, which could result in lower BMI. However, increased leisure time does not necessarily translate into increased physical activity. Americans spend more than half of their leisure time in front of the television, which is over 5 times as much as the time they spend on sports and exercise (Bureau of Labour Statistics, 2014). Hence, depending on how leisure time is spent and the type of work the overweight/obese individual does, this substitution effect could either lower or increase BMI.
Which effect will dominate?
Three interrelated key factors determine which effect will dominate: income level, information availability, and accessibility to healthy food and exercise.
v Income level
Whether the negative income effects and substitution effects add up to improved health behaviour of the targeted individual is largely dependent on income-level.
First of all, with a higher income, the option to reduce working hours thereby increasing leisure is more within reach than when being dependent on a lower income. In the latter situation, one may be forced to work more hours or take on another job in order to feed a family.
Moreover, low-income populations spend a relatively larger share of their disposable income on foods compared to their wealthier counterparts (Green et al., 2013). Given that low-income populations’ food choice is primarily determined by the amount of money available (Burns, Cook, & Mavoa, 2013), (their perception of) the price of a healthy diet plays a more prominent role than for their wealthier counterparts (Blanck et al., 2009; Glanz et al., 1998).
14
prevalence of obesity among low-income groups, though varying by demographic and lifestyle characteristics (Scheibehenne, Miesler, & Todd, 2007), are psychological stress (Dallman, Pecoraro, & la Fleur, 2005) and difficulty finding balance between work demand and feeding a family as time to cook a meal from scratch is often absent or limited (Devine, Connors, Sobal, & Bísogní, 2003; Devine et al., 2006). This favours the choice of tasty, readily available, energy-dense but nutrient-poor sweets and fats, which in addition reduce waste, spoilage, and cooking costs (Drewnowski and Darmon, 2005; Drewnowski and Eichelsdoerfer, 2009). Another large contributor is poor nutrition knowledge among lower-income groups (Hendrie, Coveney, & Cox, 2008). The latter is related to inadequate information availability.
v Information availability
In order for a BMI tax to be effective in triggering improved health behaviour, the overweight/obese target group must be aware of what actions to take to lower BMI: this information must be available.
The price metrics previously discussed are not easily accessible to the consumer at the point of sale, which makes comparing total daily/weekly/monthly cost of a healthy versus unhealthy diet difficult. In addition, lack of knowledge on nutrition and physical activity creates barriers to a healthier lifestyle (Shepherd et al., 2006). Nutrition knowledge is important for health and diet decisions (Worsley, 2002) and has been found to be predictive of dietary change (Smith, Baghurst, & Owen, 1995). Therefore, education is required to increase knowledge and awareness (Klohe-Lehman et al., 2006). Yet, knowledge alone is not sufficient: in order to apply knowledge to food-related decision making, motivation is needed (Miller and Cassady, 2012). The costs imposed by a BMI tax is likely to do just that.
v Accessibility to healthy foods and exercise
Whether overweight and obese individuals have access to healthy foods and possibilities to exercise is closely related to the environment on two levels: built environment, referring to the physical design, and the food environment (Lake and Townshend, 2006).
15
pavement provision and monotonous, uninteresting views, which discourage walking and cycling (Frumkin, Frank, & Jackson, 2004).
When studying the availability and accessibility to healthy food choices, supermarkets have been shown to generally offer healthier foods at lower cost compared to smaller food stores (Sallis et al., 1986; Horowitz et al., 2004). The same applies to chain supermarkets as opposed to non-chain supermarkets, since a significant premium is paid by those who shop in non-chain supermarkets (Chung and Myers, 1999). Not having access to (chain) supermarkets therefore complicates improving health behaviour. American research shows that lower-income and minority communities have fewer supermarkets (Morland et al., 2002) and the availability of chain supermarkets in low-income neighbourhoods is only 75% compared to the chain supermarkets available in middle-income neighbourhoods (Powell et al., 2007). Comparing neighbourhoods by race, the availability of chain supermarkets in African American neighbourhoods is 52% of that in White neighbourhoods and Hispanic neighbourhoods have access to only 32% as many chain supermarkets compared to non-Hispanic neighbourhoods. Hence, the food environment contributes to both racial and socio-economic health inequalities (Glanz, Saelens, & Frank, 2005).
The role of BMI tax revenue
The tax revenues resulting from the BMI tax offer opportunity to influence above key factors, thereby increasing chances of the tax to succeed in lowering BMI. Especially with regard to increasing information availability, and accessibility of healthy foods and exercise an important task awaits government institutions.
Resources are needed to improve information availability, as food advertising and marketing did more harm than good over the last decades. To give an idea: for every US$1 the WHO spends to improve nutrition, US$500 is spent by the food industry on the promotion of processed foods (Lang and Millstone, 2002). In addition, a BMI tax creates an opportunity to make sure the information reaches the target group, as the deeply rooted federal and state governments’ tax channels can be utilized. Given that weight change is inherently slow, the education process should be initiated before implementation of the tax reform; giving overweight and obese individuals the opportunity to start improving health behaviour thereby limiting/preventing the future tax.
16
The empirical question that remains is in which scenario we are living; how do consumers respond to BMI-related income shocks?
2.3 The measure
Health insurance as a proxy
In order to empirically assess the possible success of a BMI tax as a financial incentive to reduce BMI of overweight and obese people, one needs to examine the association between (not) experiencing additional costs due to poor health (being overweight/obese) and BMI. This requires a comparison between BMI of a population that directly experiences additional costs when overweight/obese and BMI of a population that does not directly experience additional costs when overweight/obese. To the best of my knowledge, the concept of a BMI tax is new and no similar policy exists. Therefore, this study addresses the relationship between different types of health insurance and BMI as a proxy.
Health insurance brings down the monetary cost that individuals pay for health care, yet the distribution of this perceived benefit is skewed. Depending on the type of coverage, health insurance functions as a subsidy for the obese. Health care expenditures are higher for obese individuals than for normal weight individuals, estimated at $732 annually (Lourairo, 2004). Obese individuals may fairly pay for their higher medical care expenses in the form of higher out-of-pocket medical costs, reduced wages or higher premiums for health insurance.
However, regarding the first, these surplus expenses do not seem to sufficiently cover the expenditures: Sturm (2002) finds that obese individuals’ spending on medical care is $395 more than their non-obese counterparts annually.
Secondly, some evidence has been found that obese workers receive lower wages than their non-obese colleagues (Pagan and Davila, 1997; Bhattacharya and Bundorf, 2005), yet obesity related pay differences merit more investigation.
Whether obese individuals pay higher premiums than their normal weight counterparts depends on the type of coverage, which creates a testing ground for this paper’s empirical question. We distinguish two main categories of health insurance coverage: either public or private.
17
medical expenditures among Medicaid and Medicare beneficiaries costs the average taxpayer $175 per year.
When considering private health insurance, employer-based-health-insurance may provide the same “favour” for those who make poor health decisions. As both obese and non-obese employees are enrolled into a single risk pool, individuals do not proportionately bear the health care costs of excess weight: the largest share is born by the employer. Unsurprisingly, employee health benefits are among the highest employer expenses (Fong and Franks, 2007).
When purchasing other types of health insurance like when self-employed, having non-group health insurance, or having coverage via the exchange market, the premiums do in essence internalize medical costs of weight gain: being in poor health directly increases the premium, the out-of-pocket costs (DiJulio and Claxton, 2010), or both, thereby decreasing disposable income - comparable to a BMI-related income shock. Internalizing the medical care costs of weight gain is what a BMI tax is expected to do, thereby striving to incentivize overweight and obese individuals to decrease expected medical care costs while improving health.
Hypotheses
Therefore this paper’s data analysis tests whether being confronted with the direct costs of overweight/obesity versus not being directly confronted with these costs have different effects on BMI in the context of health insurance. Throughout the rest of this paper, I will refer to the group of individuals that experience the direct costs of overweight/obesity as population A. Population A holds individuals that have either self-employed insurance, non-group coverage, or private insurance through an exchange. The group of individuals that does not directly experience the costs of poor health behaviour makes up population B. Population B thereby consists of individuals that have either public or employer-based health insurance. Based on the literature, this paper tests the following five hypotheses, with the fifth being split into a. and b.
18
Subsequently, the first hypothesis is:
H1: Individuals in population A have lower BMI levels than individuals in population B.
To prevent findings from being a coincidence, H1 will be tested for the last five years for which data is available (2010-2014).
As previously discussed literature has shown, demographic factors possibly affect BMI. To assess the potential of a BMI tax, I wish to isolate the effect of being confronted with the direct costs of poor health behaviour - a BMI-related income shock – on BMI. Therefore, the second hypothesis that will be tested is:
H2: Individuals in population A have lower BMI levels than individuals in population B when controlling for demographic factors.
Other studies found information availability and nutritional knowledge being particularly determinant for health and dietary decisions. Therefore, it is expected that when higher educated, being confronted with direct costs leads to lower BMI; the combination of knowledge and an incentive provides sufficient motivation to improve health behavior. In contrast, experiencing these costs may either not lead to lower BMI levels of lower educated individuals due to a lack of knowledge, or the BMI-decreasing effect could be smaller. Therefore, the third hypothesis that tests whether the relationship between being confronted with the direct costs of poor health behavior and BMI was different for higher than for lower educated individuals reads:
H3: Individuals in population A with higher levels of education have lower BMI levels than individuals in population A with lower education levels.
19
H4: Individuals in population A have lower BMI levels than individuals in population B when matching individuals on demographic factors.
Bhattacharya et al. (2009) find that moving from a lack of insurance to Medicaid induces an increase in BMI by over 2 points. Though far from definite proof, this suggests that no longer being confronted with direct costs of poor health behaviour – comparable with moving from population A to population B – results in poorer health decisions thereby increasing BMI over time. Vice versa, the income shock to overweight/obese caused by a BMI tax – comparable with moving from population B to population A – is expected to result in a lower BMI level over time. Subsequently, this paper’s final twin hypotheses are:
H5:
a. When an individual switches from population A to population B, change in his/her BMI level is positive.
b. When an individual switches from population B to population A, change in his/her
20
3. DATA & METHOD
3.1 Specification of data sources and data collection process
The data source used for this study is confined to the Household Component (HC) of the Medical Expenditure Panel Survey (MEPS). The MEPS is a continuing data collection effort which purpose is to track changes in health status, use of health care services, costs for health care, and access to health care. Each year MEPS selects a new sample of households to provide data for the study for the length of two full calendar years. The HC applies an overlapping panel design, including five rounds of interviews over the two-year period and obtains its sampling frame from National Health Interview Survey respondents - a survey conducted by the National Center for Health Statistics.
In addition to the nationally representative estimates of health status, health care use, expenditures, and health insurance coverage for the U.S. civilian noninstitutionalized population, MEPS includes demographic variables of each individual including age, sex, race, marital status, and educational attainment. The information collected from households is supplemented by surveys of medical and health insurance providers utilized by respondents (HC-MEPS, n.d.).
As the array of questions covered by MEPS is extensive, a system of skip patterns and section-grouped questionnaire modules is operationalized to ensure the number of questions asked of each respondent is minimized; a question is only asked when determined as applicable by the specific respondent’s skip pattern. Not all sections are included in in every round of data collection.
Data is collected with computer-assisted personal interviewing (CAPI) using a laptop computer and next to the questions, interviewing instructions are provided to respondents. At the end of each completed interview, participants receive a gift of $50.
To conduct the empirical analysis, I use the 2014, 2013, 2012, 2011, and 2010 Full Year Population Characteristics data files, which are available for download as ASCII data files via the MEPS website1. The 2014 Full Year Population is the most recently published MEPS data set by the time of this study. In addition, I use the MEPS Panel 17 Longitudinal Data File, which is the most recently published longitudinal data set, representing those who were in the MEPS population for all or part of the 2012-2013 period.
1
21
Each data file is accompanied by STATA Programming Statements needed to create a permanent Stata dataset.
3.2 Specification of the sample
To test the hypotheses that follow from the empirical question “how do consumers respond to
BMI-related income shocks”, I create a subsample out of the MEPS baseline sample. The
baseline sample consists of 30,000+ respondents each year, though varying per year by about 20%; considering the years 2010-2014 the amount lies between 32,846-38,973.
Studying the effect of experiencing a BMI-related income shock, individuals below the age of 18 are excluded as they are generally on a parent’s health insurance plan. In addition, I exclude individuals above 65 years of age due to their eligibility for Medicare and because it is unclear whether the same BMI recommendations apply to older adults (Winter et al., 2014). Attempting to isolate the effect of experiencing a BMI-related income shock, this study is confined to employed individuals. Being unemployed is likely to show a bias towards being obese (Finkelstein et al., 2007; Lakdawalla et al., 2004; Kinge and Morris, 2010) and towards being uninsured (KFF, 2015), which could reduce the validity of this paper’s analysis.
Moreover, I exclude women who were pregnant at the time of the survey as their BMI may not be representative of their health status, and individuals who were underweight (BMI<18.5) are excluded because of their increased risks for malignancies and eating disorders.
Drawing on this subsample, observations can be in either population A or population B, which is indicated by the dummy “confronted with direct costs” as will be explained below. 3.3 Specification of the variables
v Dependent variable
The key dependent variable is BMI, which is a measure of weight divided by height squared and is commonly used by health professionals and governments to determine what is a “healthy weight” for individuals. Respondents’ BMI was calculated by MEPS using the formula from the Centers for Disease Control and Prevention (www.cdc.gov.nl) website: BMI = [Weight in Pounds / (Height in Inches)2 ] * 703
v Explanatory variable
22
overweight/obese due to higher health insurance premiums he/she has to pay – thereby exposed to a BMI-related income shock.
The dummy takes a value of 1 when the individual:
• Directly purchased health insurance associated with a self-employed job where firm size is equal to 1. This is indicated with “yes” to the statement “self-employed and had health insurance through that business on 12/31/[year of survey]”.
• Has non-group private health insurance, which is nonpublic insurance that provided coverage for hospital and physician care. This is indicated with “yes” to the statement “had non-group insurance on 12/31/[year of survey]”.
• Is covered by private insurance through an exchange/marketplace, which provided hospital/physician services. This is indicated with “yes” to the statement “covered by private exchange insurance on 12/31/[year of survey]”.
If the above applies, the observation is part of population A. The dummy takes a value of 0 when the individual:
• Has public health insurance, which holds being either covered by Medicare, Medicaid, TRICARE, or other public hospital and physician coverage. This is indicated with “yes” to the statement “covered by public health insurance on 12/31/[year of survey]”. As the sample used by this paper excludes people age 65 and over, it is unlikely that the respondent’s indication of public health insurance refers to Medicare, with the exception of respondents with disabilities. The share of TRICARE, which is health insurance coverage provided for military dependents, is also negligible as the sample holds few militants. Medicaid is the health program for low income individuals and families and expected to make up for the lion’s share of sample’s public insured.
• Has employer-based health insurance, which means the jobholder has health insurance coverage through the employer. This is indicated with “yes” to the statement “employment/union group insurance on 12/31/[year of survey]”.
If the above applies, the observation is part of population B. v Control variables
The control variables used, include the following five demographic characteristics:
§ Gender. This is indicated by the variable descriptor “sex”. Gender is an indicator variable that takes the value of 1 when male and the value of 2 when female.
23
the value of 2 when not married. The latter is applicable when marital status is either “widowed”, “divorced”, “separated”, or “never married”.
§ Age. This is indicated by the variable descriptor “age as of 12/31/[year of survey]”. Age is a categorical variable holding five categories, taking the value of 1 when 18-24 years old; 2 when 25-34 years old; 3 when 35-44 years old; 4 when 45-54 years old; 5 when 55-64 years old.
§ Race. This is indicated by the variable descriptor “race/ethnicity”. Race is a categorical variable, which takes the value of 1 when Hispanic; 2 when white; 3 when black; 4 when Asian; 5 when other race.
§ Education level. This is indicated by the variable descriptor “year of education or highest degree”. Education level is a categorical variable, which holds four categories:
- Lower education: applicable when the year of education or highest degree is either “less than/equal to 8th grade” or “9-12th grade, no high school diploma”.
- Middle education: applicable when the year of education or highest degree is either “GED or equivalent”, “high school diploma”, or “some college, no degree”.
- High education: applicable when the year of education or highest degree is either “associate degree: occupational, technical, vocational”, “associate degree: academic program”, or “bachelor’s degree (BA, AB, BS, BBA)”.
- Top education: applicable when the year of education or highest degree is “Master’s, professional, doctoral degree”.
3.4 Overview of the applied econometric techniques
This study’s statistical analysis were performed by utilizing STATA software, version 14.1 (StataCorp, College Station, Texas). Stata has the capability to adjust for the complex sample design of MEPS, thereby being capable of producing appropriate standard errors for representative estimates.
24
assumption as the Central Limit Theorem ensures that the distribution of disturbance term will approximate normality.
To test H1, H2, and H3, I perform a cross-sectional analysis.
H1, stating that BMI of population A is lower than BMI of population B, is tested using a two-sample t-test for each of the years 2010, 2011, 2012, 2013, and 2014 separately. To test the first hypothesis, I use the following equation:
𝐵𝑀𝐼!" = 𝛽!+ 𝛽!𝐷𝐶!"+ 𝑒!" (1)
H0: 𝛽!≥ 0 H1: 𝛽!< 0
In this equation, 𝛽! gives the intercept and indicator variable DC is defined to denote whether
it concerns population A or B.
1 if individual is in population A (= confronted with direct costs)
DC = 0 if individual is in population B (= not confronted with direct costs)
𝛽! captures the difference in BMI when in population A and therefore is the variable of
interest. An error term 𝑒!" is included.
H2, H3, and H4 are tested based on the 2014 data set.
To test H2, I first perform a regression with only the control variables. Based on the p-value, I determine per demographic factor whether it is a relevant control variable to include and drop the variables for which no relationship with BMI was found. Then, I add the explanatory variable “confronted with direct costs”. Performing a robust regression when controlling for relevant demographic characteristics enables us to evaluate whether there indeed is a negative relationship between being confronted with direct costs of poor health behaviour and BMI. To empirically test the second hypothesis, equation (2) is used:
𝐵𝑀𝐼!" = 𝛽!+ 𝛽!𝐷𝐶!"+ 𝛾!𝑋!"#+ 𝑒!" (2)
H0: 𝛽!≥ 0 H1: 𝛽!< 0
In this equation, the term 𝛾!𝑋!"# captures the total effect of the control variables on BMI.
𝛽! is the variable of interest, capturing the effect of being confronted with the direct costs of poor health behaviour on BMI.
25
In order to test H3, stating that the relationship between being confronted with direct costs of poor health behaviour (= being in population A) and BMI is stronger when having a higher education level than when being lower educated, I add an interaction term that multiplies the two predictor variables being in population A and education level. H3 is modelled as follows:
𝐵𝑀𝐼!" = 𝛽!+ 𝛽!𝐷𝐶!"+ 𝛽!(𝐷𝐶!" ×𝐸𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛𝐿𝑒𝑣𝑒𝑙!) + 𝑒!" (3)
Hereby I test whether 𝛽! is a) negative and b) larger for higher levels of education.
In addition to performing a regression, I compare BMI among the two populations using Propensity Score Matching, thereby testing H4. PSM is an approach that matches individuals in a treatment group to others who did not receive the treatment, but have comparable characteristics. By developing a single (propensity) score that covers multiple characteristics, PSM attempts to estimate the effect being directly confronted with the costs of poor health behaviour by accounting for the covariates that predict being in population A in the first place. Put differently, it mimics randomization by creating a sample of individuals from population A that is comparable on all observed covariates to a sample of individuals from population B. The observed covariates hold the relevant demographic characteristics. The advantage that matching plus regression has over regression alone, is that it does not rely on a specific functional form for the covariates. For the matching procedure, I take experiencing the direct costs of overweight/obesity as the treatment, which means that the individuals not experiencing these costs are in the control group. PSM yields an average treatment effect (ATE) and the p-value indicates whether the ATE is statistically significant.
26 ∆𝐵𝑀𝐼!,! = 𝛿! + 𝛾!"#$𝑆𝑤𝑖𝑡𝑐ℎ + ∆𝑒! (4) H0:𝛾!"#$≤ 0 H1: 𝛾!"#$> 0 ∆𝐵𝑀𝐼!,! = 𝛿!+ 𝛾!"#$𝑆𝑤𝑖𝑡𝑐ℎ + ∆𝑒! (5) H0: 𝛾𝑩!"#≥ 0 H1: 𝛾!"#$< 0
Indicator variable Switch is defined to denote whether it the individual changed health insurance coverage, thereby changing population, during the time frame of one year.
1 if individual switched from population A to population B or the other
Switch = way around
27
4. RESULTS
4.1 First stage estimation
One outlier was found in the 2014 data set (BMI = 135.1) and removed from the sample. In the years 2010-2014, BMI of population A was on average 5.2-6.7% lower than BMI of population B. The difference in means between the two samples was significant at the 99% confidence level for each year (Table 1), which leads us to reject H0. The cross-sectional analysis showed no clear trend in the difference in means of population A and B, yet mean BMI of population B appeared to be continuously increasing.
Table 1 – results two-sample t-tests, 2010-2014 Population A (Confronted with direct costs) Mean Population B (Not confronted with direct costs) Mean Difference p-value BMI 2010 N = 26.27 (4.83) 324 28.14 (6.00) 8,786 1.87 <0.01 2011 N = 26.53 (5.16) 316 28.22 (6.14) 9,337 1.69 <0.01 2012 N = 26.78 (5.52) 356 28.25 (6.21) 10,148 1.48 <0.01 2013 N = 26.51 (4.98) 338 28.40 (6.24) 9,606 1.89 <0.01 2014 N= 26.93 (5.50) 640 28.42 (6.22) 9,739 1.50 <0.01 Notes: Standard deviations appear in parentheses below their mean values.
4.2 Descriptive statistics
To explore the difference in BMI levels between population A and B in more detail, the 2014 data set was used for further analyses. Despite the large difference in sample size, populations A and B seemed rather similar on demographic characteristics: both populations held equally males and females, slightly more married than not married individuals, a relatively small share of 18-24 year olds, while including a particularly large share of white individuals, and the majority in both populations was middle educated (Table 2). Relatively notable differences can be found in the larger share of black, and 18-24 year old individuals in population B, while population B holds a smaller share of white, and 55-64 year old individuals.
4.3 Robust regression estimates
28
was found between marital status (p=0.13) and BMI, controlling for gender, age, education level, and race. Despite the expected relationship that was found between age and BMI, BMI levels were found to not significantly differ when in the age category 35-44, 45-54 or 55-64 years old. Therefore, these were merged into one category: 35-64 years old. Finally, being “other race” (p=0.440) did not have a different effect on BMI than being Hispanic. Hence, Hispanic and “other race” individuals were merged into one category: Hispanic/other race.
Table 2 – characteristics of MEPS respondents aged 18-64 & employed, 2014 Population A (N= 640) Population B (N=9,739) BMI (5.50) 26.93 (6.22) 28.42 Male 0.50 (0.50) (0.50) 0.50 Female 0.50 (0.50) (0.50) 0.50 Married 0.53 (0.50) (0.50) 0.53 Not married 0.47 (0.50) (0.50) 0.47 Age 18-24 (0.28) 0.09 (0.32) 0.12 Age 25-34 (0.40) 0.19 (0.43) 0.24 Age 35-44 (0.41) 0.21 (0.42) 0.23 Age 45-54 0.26 (0.44) (0.43) 0.24 Age 55-64 0.25 (0.43) (0.39) 0.18 Black (0.35) 0.14 (0.40) 0.20 White 0.50 (0.50) (0.50) 0.43 Hispanic (0.42) 0.23 (0.43) 0.25 Asian (0.32) 0.12 (0.28) 0.09 Other race (0.12) 0.02 (0.17) 0.03 Lower education (0.32) 0.12 (0.35) 0.14 Middle education 0.53 (0.50) (0.50) 0.56 Higher education (0.43) 0.24 (0.39) 0.19 Top education (0.32) 0.11 (0.31) 0.11
Notes: Standard deviations appear in parentheses below their mean values.
29
race, Hispanic and especially black individuals tend to have higher BMI levels than their white counterparts, with even larger deviations from Asian individuals. As expected, BMI levels decrease with higher education, which is closely associated with higher income. Surprisingly, middle educated individuals tend to have higher BMI levels than low educated individuals.
Including an interaction variable (Table 3, model 2) showed the negative effect of being confronted with the direct costs of poor health behaviour on BMI is stronger as education level increases, with the exception of top education (p=0.50).
Table 3 – Regression Results for 2014 BMI levels MODEL (1) (2) Constant 25.73*** (0.20) 25.67*** (0.20) POPULATION A (Confronted with direct costs) -1.38*** (0.22) 0.08 (0.68) AGE Age 25-34 2.63*** (0.21) 2.62*** (0.21) Age 35-64 3.55*** (0.18) 3.53*** (0.18) RACE White -0.33** (0.14) -0.33** (0.14) Black 1.13*** (0.19) 1.13*** (0.19) Asian -3.57*** (0.18) -3.57*** (0.18) EDUCATION LEVEL Middle education 0.57*** (0.17) 0.65*** (0.18) Higher education -0.77*** (0.20) -0.63*** (0.21) Top education -1.53*** (0.23) -1.50*** (0.23) INTERACTION TERM Population X Education level A X Middle education -1.62** (0.75) A X Higher education -2.26*** (0.77) A X Top education -0.60 (0.89) R-squared 0.0860 0.0868 No. observations 10,379 10,379
Notes: Notes: *** p< 0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses .
Population B (Not confronted with direct costs) was used as the base level. Age 18-24 was used as the base level.
Hispanic/other race was used as the base level. Lower education was used as the base level.
30
4.4 Propensity score matching
Using propensity score matching to match individuals in population A to individuals in population B yields an Average Treatment Effect of -1.23 (p<0.01), which suggests that average BMI if all individuals would be confronted with direct costs from poor health behaviour (= experience a BMI-related income shock) would be 1.23 BMI point less than the average that would occur if none were confronted with these direct costs.
The differences between the treated and control group (APPENDIX B) are not statistically significant (p=1.00) in the matched sample, which confirms that the matching procedure worked.
4.5 Differences-in-differences analysis
The results of the differences-in-differences analysis, presented in Table 4, show that neither the constant time effect for either population, nor the treatment effect for switching to either population are statistically significant different from zero. We fail to reject H0 for both H5a and H5b. Table 4: DiD Model for Panel 17 BMI levels A (Constant time effect for population A) -0.097 (0.345) AtoB (Switching from population A to B) 0.207 (0.705) B (Constant time effect for population B) 0.035 (0.056) BtoA (Switching from population B to A) -0.315 (0.617) Root mean squared error of the model 2.746
Note: Number of observations = 2,501 (Nstayed-in-A = 33, Nswitched-to-A = 20, Nstayed-in-B =
2,398, Nswitched-to-B = 20)
Results presented are estimated coefficients from ordinary least squares multiple linear regressions, standard errors in parentheses.
31
5. DISCUSSION
Literature increasingly acknowledges the public nature of the expanding obesity problem. However, previous policies attempting to improve people’s health behaviour were mainly confined to informational and educational programs, and a small part concerned food and non-alcoholic beverage. In addition to the limited effectiveness (or lack thereof) of these policies in lowering overweight and obesity, they do not directly target the overweight and obese. Therefore, this study adds to existing literature by suggesting that directly confronting people with the costs resulting from their poor health decisions is associated with lower BMI levels. Over the last five years, individuals that belonged to population A had, on average, lower BMI levels than individuals from population B. Using measures of the average American male and female (Centers for Disease Control and Prevention, 2012), this comes down to a difference in body weight between the two populations of around 10.4 – 13.2 pounds.
The difference in BMI level between the two populations was still statically significant, yet slightly smaller, when controlling for age, race, and education level. That the difference in BMI turns out to be partially explained by demographic factors is not surprising, as the existence of a relationship between these demographic variables and BMI has been repeatedly proven by previous studies. Surprisingly, this study’s findings suggest no relationship between gender and BMI. A potential explanation for the latter could be found in the sample specification; this study only includes employed individuals, which decreases differences in health status between men and women (Cohn, Livingston, and Wang, 2014). This study neither finds a relationship between marital status and BMI, which is in contrast with research by Verbrugge (1983), who finds that married people have better health than any nonmarried group. Yet, evidence by Berge et al. (2014) suggest that being married is associated with increased prevalence of overweight/obesity among young adult men, while being a protective factor for health-related behaviors associated with overweight/obesity for young adult women. Future research could shed light on the relationship between marital status and BMI.
32
Matching individuals from population A to individuals from population B confirms the difference in BMI level between both populations, yet once again a slightly smaller one than when controlling for these characteristics.
The finding that individuals who continuously face direct costs associated with poor health behaviour have, on average, lower BMI levels than demographically similar individuals who do not experience these costs, suggests that people are capable of pursuing a healthier lifestyle which requires time and effort when aware of the negative financial consequences that not taking these measures will have.
I explored the possibility that the difference in BMI between both populations is caused by unobservables by performing a difference-in-difference analysis. However, the results showed there is no difference in BMI pattern between groups; the differences-in-differences were not statistically significant different from zero. This finding implicates that the difference in BMI levels between the two populations found in the cross-sections cannot be explained by population A consistently losing weight, neither by population B consistently gaining weight. In addition, switching from either public or employer-based health insurance to self-employed, non-group, or exchange market coverage (a.k.a. switching from population B to population A) does not immediately result in weight loss, neither does a switch in the opposite direction directly result in weight gain. A BMI tax would create a negative income shock on overweight and obese individuals, which comes closest to switching from population B to population A (a.k.a. previously not being confronted with the direct costs of poor health behaviour, thereafter being confronted with the direct costs of poor health behaviour). Based on this study’s findings, I can therefore not conclude a BMI tax would succeed in bringing down BMI of the ones affected. However, these results must be interpreted with caution, as the “switching groups” available for study consisted of a very small number of individuals and the time span in which they could be studied was one year. It is unsurprising that no weight loss or –gain due to a change in type of health insurance coverage can be observed within the same year this change took place, as both losing and gaining weight are relatively slow process. Therefore, in order to determine what is causing the differences behind the previous cross-section analysis, one needs to monitor longitudinal data for a longer period of time and/or study a larger amount of “population-switchers”. Unfortunately, data sets that satisfy these conditions are – to the best of my knowledge – not available.
33
First, BMI levels are likely to be downwardly biased due to the use of self-reported weight, as women and heavier men (>100 kg) tend to under-report their weight, while men under 100 kg have been found to over-report their weight (Lakdawalla and Philipson, 2009). Moreover, evidence suggests the reporting bias is greater for non-whites. Accounting for this bias would increase the study’s validity. It is possible to gain access to the height and weight variables through the MEPS Data Center, which enables researchers to create their own BMI variable.
Second, it is important to bear in mind that BMI is not a perfect measure. Body Mass Index should not be regarded as a medical diagnosis, as people counted as obese based on BMI can be physically fit when using different measures. Yet, the ease of measuring body weight and height makes BMI the number one used measure across the globe. Using alternatives to BMI like Body Adiposity Index, Waist-to-Hip Ratio, or Body Fat Measuring may require more advanced methods and will bring about an additional administrative burden on health and government institutions. Above all, it will take (a long) time to collect the data on these alternative measures and be comparable across countries. Despite the challenges, I reckon it may be worth the effort when it could result in a more accurate picture of a population’s health and the actual risk on non-transitory diseases involved.
Third, I would preferably have included a variable that holds the individual’s income, yet data on income was not available in the MEPS 2014 data set by the time of writing this paper. One may interpret education level as an imperfect proxy for income, since a high correlation between education and income has repeatedly been found by previous studies (Strauss, 2012; Graham and Paul, n.d.).
A fourth limitation worth noting, is that this study’s models did not include behavioral factors known to be more proximal to BMI outcomes. The models could be enhanced by including nutrition factors (i.e. caloric intake) and physical activity variables (i.e., average step count), which were not part of the MEPS data set, therefore not included in this study.
Finally, when elaborating on the concept of a BMI tax, areas that merit further investigation include determining an effective tax rate, and eventually covering the ethical and political challenges that need to be overcome when considering implementation.
34
6. CONCLUSION
This paper represents a first step in understanding the potential and caveats of a BMI-based tax. There is little question that overweight and especially obesity pose a serious issue on personal physical and emotional health, subsequently bringing about increased health care costs. Yet, motivation for weight loss efforts among overweight and obese individuals is limited, resulting in failure to achieve weight loss (French, Jeffery, and Murray, 1999). In addition, people seem to be insufficiently concerned with the negative externality caused by their poor health decisions – large public costs that decrease public welfare. Simultaneously, governments seem to be either short in resources to enable the population to improve health behaviour by providing them with the necessary information and a supportive environment, or do not give sufficient priority to tackling the obesity epidemic.
35
7. ACKNOWLEDGEMENTS
The author wishes to thank Piet C. de Groen, MD, of the Department of Gastroenterology and Hepatology, Mayo Clinic, Rochester, Minnesota, for sharing his enthusiasm and ideas which immediately sparked the author’s interest in the area of health economics and made conducting this study a joy rather than a pain in the ass.
8. REFERENCES
Abramowitz, J. (2016) The connection between working hours and body mass index in the U.S.: a time use analysis. Rev Econ Household. Vol. 14, page 131–154
American Medical Association (2013) AMA Adopts New Policies on Second Day of Voting at Annual Meeting. Retrieved from:
http://www.ama-assn.org/ama/pub/news/news/2013/2013-06-18-new-ama-policies-annual-meeting.page [Accessed 1/6/2016]
Andreyeva, T., Michael W. Long, M. W., Brownell, K. D. (2010) The Impact of Food Prices on Consumption: A Systematic Review of Research on the Price Elasticity of Demand for Food. American Journal of Public Health. Vol. 100 (2), page 216–222
Bassett, M. T. and Perl, S. (2004) Obesity: The Public Health Challenge of Our Time. Am J Public Health. Vol. 94 (9), page 1477
Bauer, K.W., Hearst, M.O., Escoto, K., Berge, J.M., Neumark-Sztainer, D. (2012) Parental employment and work-family stress: associations with family food environments. Soc Sci Med. Vol. 75 (3): 496-504
Berge, J. M.., Bauer, K. W., MacLehose, R., Eisenberg, M. E., and Neumark-Sztainer, D. (2014) Associations between relationship status and day-to-day health behaviors and weight among diverse young adults. Families, Systems, & Health. Vol 32 (1), page 67-77
Bhattacharya, J. and Bundorf, K. (2005) The Incidence of the Healthcare Costs of Obesity. National Bureau of Economic Research Working Paper #11303
Bhattacharya, J., Budnorf, K., Pace, N., Sood, N. (2009) Does Insurance Make You Fat? Economic Aspects of Obesity [working paper]
36
Blanck, H. M., Yaroch, A. L., Atienza, A. A., Yi, S. L., Zhang, J. & Mâsse, L. C. (2009) Factors influencing lunchtime food choices among working Americans. Health Education and Behavior. Vol. 36(2), page 289-301
Bureau of Labour Statistics (2014) American Time Use Survey. Retrieved from: http://www.bls.gov/TUS/CHARTS/LEISURE.HTM [Accessed 3/6/2016]
Burns, C., Cook, K., & Mavoa, H. (2013) Role of expendable income and price in food choice by low income families. Appetite. Vol. 71, page 209-217
Burton B.T., Foster, W.R., Hirsch, J., VanItallie, T.B. (1985) Health implications of obesity: NIH consensus development conference. Int J Obes Relat Metab Disord. Vol. 9, page 155-169
Rao, M., Afshin, A., Singh, G., & Mozaffarian, D. (2013) Do Healthier Foods and Diet Patterns Cost More Than Less Healthy Options? A Systematic Review and Meta-Analysis. BMJ Open. Vol. 3
Campos, P. (2004) The Obesity Myth: Why America’s Obsession with Weight is Hazardous to Your Health. New York, NY: Gotham Books
Capacci, S., Mazzocchi, M., Shankar, B., Macias, J. B., Verbeke, W., Pérez-Cuento, F. J. A., Kozio-Kazakowska, A., Piórecka, B. (2012) Policies to promote healthy eating in Europe: a structured review of policies and their effectiveness. Nutrition Reviews. Vol. 70, No. 3, page 188-200
Carlson, A., and Frazão, E. (2012) Are Healthy Foods Really More Expensive? It depends on How You Measure the Price. U.S. Department of Agriculture, Economic Research Service. Economic Information Bulletin, no. 96
Cawley, J. (2004) The impact of obesity on wages. Journal Human Resources. Vol. 39, page 451-474
Cawley, J., Joyner, K., and Sobal, J. (2006) Size matters: The influence of adolescent’s weight and height on dating and sex. Rationality and Society. Vol. 18(1), page 67-94 Cawley, J., Rizzo J.A., Haas, K. (2007) Occupation-specific Absenteeism Costs Associated with Obesity and Morbid Obesity. Journal of Occupational and Environmental Medicine. Vol. 49 (12), page 1317-1324
