|
Article Excerpt
Obesity has become a major public health issue in developing and developed countries alike. Overweight and obese individuals are at increased risks for a long list of physical ailments, including hypertension (high blood pressure [BP]), hypercholesterolemia (high blood cholesterol), diabetes, coronary heart disease, stroke, cancer, poor reproductive health, and psychological problems such as depression and eating disorders (Stunkard and Wadden 1993, p. 224; National Institutes of Health [NIH] 1998; Jensen et al. 2004; Costa-Font and Gil 2003). Obesity accounts for approximately 365,000 deaths in the United States per year, second only to tobacco use (Mokdad et al. 2005). The total direct and indirect costs attributable to overweight and obesity amounted to US$117 billion in 2000 (U.S. Department of Health and Human Services [USDHHS] 2001). As the then Surgeon General David Satcher noted, these problems, if left unabated, may soon cause as many preventable diseases and deaths as tobacco use (USDHHS 2001).
Ogden et al. (2002) noted that the prevalence of overweight children in the United States had continued to increase, especially among Mexican American and non-Hispanic black adolescents. Kuchler and Variyam (2003) and Mokdad et al. (2003) confirmed the deleterious effects of an overweight U.S. population. Given the increased prevalence of obesity and mounting evidence of its deleterious health effects, the Surgeon General called upon citizens to work together "in finding solutions to this public health crisis" (USDHHS 2001, p. xiv).
The increasing prevalence of overweight and obese people has created an urgent need to identify the causes, which can provide a framework to develop programs and policies to constrain and ultimately reduce the incidence. Lifestyle, health knowledge, social policies, and neighborhood characteristics have been hypothesized as the underlying factors of the obesity epidemic within and outside of the United States. Lifestyle variables have long been identified as contributing factors (e.g., Huffman and Rizov 2007). Kuchler and Variyam (2003) found nonsmokers more likely to become obese than smokers. Chen, Yen, and Eastwood (2007) examined the relationship between smoking and bodyweight and found that smoking may not cause sustained weight loss. Chou, Grossman, and Saffer (2004) found that the number of fast-food and full-service restaurants, food consumed at home, and prices of cigarettes and alcohol were related to obesity. Lin, Huang, and French (2004) examined the relationship among eating behavior, dietary intake, physical activity, attitude toward diet and health, sociodemographic variables, and body mass index (BMI) among women and children in the 1994-1996 Continuing Survey of Food Intakes by Individuals (CSFII). Significant correlations between women's BMI and age, race, dietary patterns, TV watching, and smoking were found.
Kan and Tsai (2004), using a sample from Taiwan, found a relationship between individuals' knowledge concerning the health risks of obesity and their tendencies to be obese. Nayga (2000) found positive effects of health knowledge on obesity among U.S. adults after controlling for education using the Diet and Health Knowledge Survey component of the 1994 CSFII. TV watching, video games, and computer uses have been blamed for childhood obesity in Switzerland (Stettler, Signer, and Suter 2004) and in the United States (Vandewater, Shim, and Caplovitz 2004).
Recent analyses of cross-sectional data indicate that women receiving food stamps are more likely to be overweight and obese and also weigh more than nonparticipating eligibles (Fox and Cole 2004; Chen, Yen, and Eastwood 2005). Using data from National Health and Nutrition Examination Surveys, Ver Ploeg et al. (2007, p. 10), however, found that the effect of FSP has vanished because "the BMI of the rest of the population has caught up to the BMI levels of food stamps recipients."
Declining food prices associated with long-run technological change have been found to contribute to the rising proportions of overweight and obese people in high-income countries (Philipson and Posner 1999; Lakdawalla and Philipson 2002; Cutler, Glaeser, and Shapiro 2003). Chen and Meltzer (2008) concluded that relative income had contributed to obesity in rural China.
Existing studies have often focused on one weight category (overweight). However, the effects of sociodemographic and lifestyle variables on body weight may differ by weight category and such differentiated effects help identify target populations for public policy intervention. Audrain, Klesges, and Klesges (1995), for instance, found that smoking increased resting energy expenditure in both normal-weight and obese smokers, but the metabolic effect was larger and lasted longer among normal-weight smokers. The theoretical biology literature confirms that unobserved energy expenditure, used to maintain basic metabolism, increases with bodyweight (Christiansen, Garby, and Sorensen 2005), which leads to our hypothesis that the relationship between obesity and its underlying factors may differ across weight categories.
This study addresses the differentiated effects of sociodemographie and lifestyle variables on BMI by weight category. We first examine the effects of sociodemographic and lifestyle variables on body weight by estimating ordinary least-squares (OLS) regressions for weight-segmented subsamples and the pooled sample. These regression estimates allow testing of the hypothesis that the relationship between obesity and the underlying factors is constant across weight categories. On rejection of this "equal relationship hypothesis," we tested the next hypothesis that determination of weight categories is exogenous. A comparison among the segmented sample results motivates the use of a more sophisticated statistical model, namely the endogenous switching regression model (SRM), which permits a test of the exogeneity hypothesis. (1)
In the next section, we present a theoretical framework that motivates our empirical specification and describe the dataset and variables used. We then compare OLS estimates of the BMI regressions by weight categories. A multiregime endogenous SRM is then developed. Results for the SRM are presented next, which is followed by concluding remarks.
METHOD
Theoretical Framework
The empirical specification is based on the consumer utility maximization theory. Conditional on a set of sociodemographic and lifestyle variables [S.sub.1], an individual derives utility from body weight (W), physical activity or exercise (E), and levels of food (F) and other goods (C) consumed, with corresponding prices [P.sub.E], [P.sub.F], and [P.sub.C]. Body weight is a function of food consumed and exercise, conditional on another set of sociodemographic and lifestyle variables [S.sub.2]. The utility function
U = U(W (F,E;[S.sub.2]), E, F, C; [S.sub.1]) (1)
is maximized subject to an income (I) constraint limiting the amount of money spent:
[P.sub.F]F + [P.sub.E] E + [P.sub.C] C = I (2)
This setup is similar to that of Philipson and Posner (1999) and Schroeter, Lusk, and Tyner (2008). Solving the constrained utility maximization problem in (1) and (2) yields the optimal levels of food ([F.sup.*]), exercise ([E.sup.*]), other consumer goods ([C.sup.*]), as well as body weight ([W.sup.*]). The optimal weight can therefore be expressed as
[W.sup.*]= [W.sup.*]([F.sup.*]([P.sub.F], [P.sub.E], [P.sub.C], I; [S.sub.1], [S.sub.2]), [E.sup.*]([P.sub.F], [P.sub.E], [P.sub.C],I;[S.sub.1], [S.sub.2]);[S.sub.1]) (3)
[W.sup.*] is a function of [F.sup.*] and [E.sup.*] and, importantly, these two variables are endogenous (i.e., functions of prices, income, and sociodemographic and lifestyle variables). Note that while prices are not available in a single cross section, some of the price variations are likely explained by the sociodemographic variables [S.sup.1] and [S.sup.2], especially regional and urbanization variables. Multiple food items, used in the empirical specification below, can be accommodated by allowing F to be a vector. We treat the determination (endogenization) of [F.sup.*] and [E.sup.*] differently because inclusion of many food items in a simultaneous-equations framework would require too many variables and would be further complicated by censoring (observed zero values) in these foods items. We take a simpler approach of predicting the food quantities with a censored (Tobit) regression model (Amemiya 1985). These predicted quantities are used as regressors in the weight category and level equations discussed below. This procedure to circumvent the endogeneity issue was used by Bryant (1988) and Yen (1993). The other endogenous variable, exercise, is incorporated by including a probit...
|
