GEOGRAPHIC IMPACTS OF FEDERALLY FUNDED STATE-BASED OBESITY PROGRAMS ON ADULT OBESITY PREVALENCE IN THE UNITED STATES By Keumseok Koh A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of Geography Doctor of Philosophy 2016 GEOGRAPHIC IMPACTS OF FEDERALLY FUNDED STATE-BASED OBESITY PROGRAMS ON ADULT OBESITY PREVALENCE IN THE UNITED STATES By Keumseok Koh The third study partitioned the variance in obesity prevalence between blacks and whites into explainable and unexplainable portions of obesity using a reweighting decomposition technique to further understand these disparities. The findings from this research identified where programs have been successful in controlling obesity and where to target future interventions to reduce obesity, reduce racial disparities in obesity and improve population health. The translation of this knowledge will also be helpful to reduce obesity in other countries, particularly those countries experiencing a transition toward obesity in their populations. iv v vi Table 1.1. States with Highest and Lowest Obesity Prevalence Rates in .9 Table 4.1. Estimated Mean BMI and Obesity Using Ordinary Least Squares and Logit Regression and Explanatory Risk Factors, United States, 2010 vii Table 4.2. Proportions of Black and White Racial Groups and Other Descriptive Statistics, United States, 2010.87 Table 4.3. Differences in Mean BMI and Obesity Prevalence for Blacks and Whites, United States, 20109 Table 4.4. Mean BMI Gaps between Blacks and Whites, United States, 2010 Table 4.5. Gaps in Mean Obesity Prevalence Rates per 100 Population (%) between the Blacks and the Whites, 2010 viii 1 1. 1.1. 1.1.1. 2 3 4 5 1.2. 1.2.1. 6 1.2.2. North Carolina State Center for Health Statistics, 2012).7 8 Source: Fryar et al. (2012). In addition to the epidemic levels of overall obesity there are distinct geographic inequalities by states and counties within the United States (Figure 1.3). In 1990 all states had less than 15% of obesity prevalence. In 2000 most Southern and Midwestern states experienced rapid increases in obesity prevalence (i.e., up to 25 per 100 population) (Flegal et al., 2002). In 2010 all states had an obesity prevalence of 20 or higher with many Southern states including Alabama, Arkansas, Kentucky, Louisiana, Mississippi, (CDC, 2013b). Table 1.1 lists five top states with the highest and lowest obesity prevalence in 2010. 9 Figure 1.3. Obesity Prevalence1 by States, United States 1990-2010. Source: CDC (2013b) with permission to use. 1 Rates per 100 population Table 1.1. States with Highest and Lowest Obesity Prevalence Rates1 in 2010. Rank Highest Rates Lowest Rates 1 Mississippi 34.0 Colorado 21.0 2 West Virginia 32.5 Nevada 22.4 3 Alabama 32.2 Connecticut 22.5 4 South Carolina 31.5 Utah 22.5 5 Kentucky 31.3 Hawaii 22.7 Source: CDC (2012). 1 Rates per 100 population Since most obesity comorbidities are the leading causes of preventable death obesity has enormously influenced mortality in the U.S: Mokdad et al (2004) found that obesity caused approximately 400,000 premature deaths, more than 16% of deaths in the United States, which was second to tobacco-related causes of mortality in 2000. Masters et al (2013) found that approximately 18% of the deaths for adults aged 40 to 85 years in the United States were associated with obesity in 1986 to 2006. These studies demonstrate the urgency to reduce obesity as a major public health problem in the United States. 10 1.3. 11 1.3.1. 12 1.3.2. 13 14 1.3.3. Ewing et al. (2003) studied a sample of adult residents from 1998 to 2000 using those years of BRFSS datasets and found that residents of sprawling counties were more likely to have higher BMI compared to those who lived in densely populated counties because physical activity was diminished in suburbia. Adults living in a more compact county (i.e., one standard deviation above the mean county sprawl index) were less likely to be obese OR = 0.9 (95% CI: 0.86-0.95) than adults living in a more sprawling county. Another study by Frank et al. (2004) using a travel survey of 10,878 participants in Atlanta found that the likelihood of obesity could increase by 6% with one additional hour per day spent in vehicles. They also found that a quartile increase in mixed land-use index developed in terms of four different types of land use (residential, commercial, office, and institutional) was attributable to a 12.2% decrease in the likelihood of obesity. Mixed land-use planning and the promotion of public transit were also found to be good policy ew York City (Rundle et al, 2008). However Vojnovic et al. (2013) recently found that the traditional relationship between higher densities, mixed land uses, higher connectivity, and greater accessibility do not guarantee higher pedestrian activity and lower BMI in declining inner-city neighborhoods in Lansing, Michigan. Neighborhood safety is an important risk factor that may impact BMI. Fish et al. (2010) found that Los Angeles residents who considered their neighborhoods unsafe in terms of crime victimization have OR = 2.81 (95% CI: 0.11-5.52) higher BMI than those 15 who perceived their communities safe using the 2000-2001 Los Angeles Family and Neighborhood Survey. Importantly, the characteristics of the environment and the processes by which individuals interact with their environment often determines healthy, since, individuals with lower socioeconomic status often live in poor neighborhoods, which are less favorable built environments for health (Darden et al., 2009). Especially in these poor environments there may be less access to high quality foods because of few to no grocery stores. The investigates the social conditions that economically and socially disadvantaged people suffer from higher food prices and the paucity of food stores in their neighborhoods (Zenk et al., 2005; Raja et al., 2010). However LeDoux and Vojnovic (2013) refuted the food desert hypothesis by showing that residents living in disadvantaged neighborhoods actually purchased their groceries from supermarkets in suburban locations outside of their neighborhoods due to the disproportionately with unhealthful food choices like convenience and party stores. 1.4. As the obesity epidemic has risen in the United States over the past few decades there has been the need for public health programs and policy implementation. There is no doubt that governments are important actors in curbing the obesity epidemic because they are responsible for enhancing public health by providing public goods and services (Gortmaket et al., 2011). All levels of government, i.e., federal, state and local have been involved in various programmatic and policy interventions to reverse the increasing trends of obesity prevalence in the United States (Khan et al., 2009). The U.S. federal government identified obesity as a key public health priority through the 16 the One of the most notable public health interventions to address and understand the obesity epidemic in the U.S. was the establishment of the Division of Nutrition, Physical Activity, and Obesity (DNPAO) at the CDC with the approval of the U.S. Congress in 1999. The goal of the CDC-DNPAO prograb). The CDC-DNPAO environments because chronic energy imbalance involving both dietary intake and physical activity are recognized as major causes of obesity (Hamre et al., 2008; Gortmaker et. al, 2011). The CDC-DNPAO program is a competitive cooperative-agreement with participating state health departments applying for these grants to and concerns. State health departments can strengthen their ability to provide better health promotion, to implement effective nutrition and physical activity interventions, and to accumulate scientific evidence on obesity and its risk factors using the program funding (Hamre et al., 2008). The theoretical framework within which CDC-DNPAO programs are designed is a five-level Social-Ecological Model (SEM), first proposed by McLeroy et al. (1988), which implies that human behavior can be influenced by distinct yet intertwined levels of society (Hamre et al., 2007; Brown, 2011). From the SEM perspective a society has five levels of interactions that include intrapersonal, interpersonal, organizational, community, and society levels (CDC, 2013c): 17 Individual: Different food intake and physical activity habits by each person can have an impact changes in practices which have been formulated through Interpersonal: Interpersonal interactions includes any social network and support system among people with a shared relationship in society. The common examples of interpersonal groups are families, friends, neighbors and work groups (McLerey et al., 1988). An interpersonal group is usually formed informally but sometimes it can be built through a formal organization such as a club or around a common interest. Individuals can expect physical and/or emotional support and reliance from informal members (Brown, 2011). Most norm and rules within interpersonal groups are naturally made and shared among members. Organizational: A society has various types of organizations generally formulated and governed by official rules and regulations. These organizations consist of individuals and interpersonal groups. For example people usually spend time in educational institutions (e.g. primary and secondary schools) and workplaces interacting with other members. While organizational characteristics and structures can also share unofficial and unconscious experiences among them (McLerey et al., 1988) which may lead to promotional or untoward personal behaviors and/or interpersonal interactions. Community: The concept of community can vary by definition and context. A community usually includes families, informal social networks, neighborhoods, civic groups, and churches within which peo18 (McLerey et al., 1988). In communities people share common values and experiences through social interactions. The CDC emphasizes the role of a community for obesity prevention environment to give residents the best possible access to healthful foods and communities to address obesity include changes to zoning ordinances, improvements to parks and recreation facilities and creating ways to distribute free Societal: At the highest level societal or macro-level interventions can also be implemented to reduce obesity. Regulatory policies, interventions and laws implemented by local, state, or federal government may impact population-health outcomes or behaviors. The five important evidence-based strategies for the CDC-DNPAO program include (1) balancing caloric intake and expenditure, (2) increasing physical activity, (3) increasing the consumption of fruits and vegetables, (4) decreasing television-viewing time, and (5) increasing breastfeeding (Yee et al., 2006). Since obesity has a complex, multifaceted etiology, it is essential to have evidence-based strategies for implementing obesity prevention programs (Economos & Irish-Hauser, 2007). Obesity prevention and control interventions generally involve developing nutrition, physical activity, and environment plans to balance caloric intake and expenditure through public and private partnership (Hamre et al., 2008). In Michigan, for example, the Michigan Department of Community Health (MDCH) is working with county 19 health departments and community coalitions through several obesity prevention improving walking trails and bicycle facilities, and promoting healthy lifestyles through partnerships with non-profit organizations (CDC, 2012b). The national budget for the CDC-DNPAO program in 2010 was $90 million with the average annual state grant, approximately $756,000 (National Alliance for Nutrition and Activity, 2010). After granting the programmatic funding CDC requests all funded states to submit performance reports on the effectiveness of the CDC-DNPAO program in their state. Based on these reports and new requests from other states, CDC-DNPAO will decide to continue the support for existing participants or provide new funding for current non-participating states (Yee et al., 2006; Hamre et al., 2008). Recent studies have also shown that CDC-DNPAO programs have provided funded states with momentums to develop statewide partnerships, establish health promotion infrastructures, implement policy interventions, and enacting obesity-related legislations (Hersey et al., 2011; Yee et al., 2006). Through these environmental interventions on obesity, CDC-DNPAO programs may contribute to control overall obesity prevalence in funded states. While these reports and studies provide important information on how well each state is performing, it is still unknown what the cumulative impacts are of state-based interventions on the geography of adult obesity and its racial inequalities in the United States. 1.5. The Need of This Study Obesity is a major concern to public health worldwide. There are also regional variations in obesity prevalence. Obesity and many comorbidities contribute to a decrease in healthy life, premature mortality, and the increase in public health spending. 20 Western dietary habits and sedentary lifestyles and are major causes of obesity. Some developing countries are also experiencing the paradoxical dual burden of obesity and malnutrition under rapid westernization, demographic shifts, and underdeveloped public health system. Other micro- and macro- causes of obesity are still under investigation. In the United States, obesity prevalence has more than doubled since the 1980s following a moderate increase in obesity during the 1970s. Currently two-thirds of U.S. adults are obese or overweight. Geographically Southern states have higher obesity prevalence than other states. In terms of race, blacks and American Indians are more likely to be obese compared to whites and other racial and ethnic groups. The CDC has been actively involved in monitoring obesity prevalence by conducting annual public health surveys e.g., BRFFS and NHANES and implementing CDC-DNPAO programs. Researchers also have extensively investigated obesity risk factors in terms of population, behavior, and environmental perspectives. The most important individual and population-based risk factors for obesity are diet coupled with high fat and sugar and sedentary life styles. The effect of demographic characteristics, socioeconomic status and environments on obesity varies by population groups and regions. This dissertation research addresses the obesity epidemic by focusing on three perspectives, studies that have not yet been addressed in the obesity literature and can broaden our understanding of the high obesity prevalence in the United States. The first study investigated the spatial and spatio-temporal prevalence of obesity at the county level. The obesity literature has been reporting obesity prevalence only at the national or state levels and there is an immediate need to investigate the prevalence at the county level to inform future obesity interventions. There are also no studies, to my 21 knowledge, that use a spatial microsimulation methodology to calculate obesity prevalence across counties in the United States. The second study evaluated the impact of CDC-DNPAO programs on obesity prevalence at the county level within and across states over time using the output from the first study. To date no national study has been conducted to evaluate the CDC-DNPAO using simulated county-level obesity data. This study will indirectly evaluate the effectiveness of state-level obesity interventions on changing obesity prevalence. The third study focused on the racial inequalities in obesity prevalence. Partitioning the underlying causes of racial inequalities in obesity into known and unknown portions will be valuable to further understand why these disparities exist and how to implement future interventions to reduce the racial gaps. 1.6. The purpose of this research is to investigate the impacts of CDC-DNPAO statewide intervention programs on the geography of adult obesity prevalence in the United States to identify where programs are successful and where to target future interventions to reduce obesity and improve population health. The specific objectives of this research include: To visualize and explore the spatial and spatio-temporal patterns of obesity prevalence across counties in the United States (1998-2010) using the BRFSS datasets. A spatial micro-simulation approach will be implemented to calculate obesity prevalence estimates within and across counties. To evaluate the impacts of state-level CDC-DNPAO programs on county-level obesity prevalence (1998-2010) using a quasi-experimental modeling to identify 22 counties where state programs are more or less protective of obesity and to identify counties in need for future intervention; and To partition the variance in obesity prevalence between blacks and whites using a Blinder-Oaxaca Decomposition Technique into explainable and unexplainable causes of obesity to improve our understanding of racial disparities in obesity prevalence in the United States. 1.6.1. Study Hypotheses 1.6.2. Study Design 23 24 Figure 1.4. Theoretical and Conceptual Framework to Study Obesity. Note. Social-Ecological Model (SEM) is the theoretical model for the CDC-DNPAO state interventions. Disease-Ecology summarizes the interactions between obesity and its risk factors from behavioral, population and environmental perspectives. 25 2. STUDY I: SPATIAL AND SPATIO-TEMPORAL PREVALENCE OF ADULT OBESITY AT THE COUNTY LEVEL IN THE UNITED STATES: A SPATIAL MICROSIMULATION APPROACH 26 27 2.1. 28 2.1.1. 29 2.1.2. 30 31 2.2. 2.2.1. 2.2.2. 32 2.2.3. 33 34 35 36 2.3. 2.3.1. 37 38 39 40 2.3.2. 41 2.3.3. 42 43 44 45 46 47 48 49 50 51 2.4. 52 53 54 2.5. 55 56 3. STUDY II: IMPACTS OF FEDERALLY FUNDED STATE OBESITY PROGRAMS ON ADULT OBESITY PREVALENCE IN THE UNITED STATES, 1998-2010. 57 3.1. 58 59 60 61 3.2. 3.2.1. Study Area The study area in this research includes the 50 states and the District of Columbia with a total sample (n=2,774,697) representing nearly 2 billion (n=1,965,992,351) of U.S. adults aged 18 years and older from 1998 to 2010. 3.2.2. Data The data used for this study which collects information on health risk behaviors, preventive health practices, disease outcomes and health care access (CDC, 2014). Since 2010, the CDC and each state collaborated to perform telephone interviews to collect self-reported behavioral health risk information and to manage the survey data. Since 2011, CDC adopted a new survey methodology to collect data by both landline telephones and cell-phones to include more diverse demographic groups, especially low-income and young adults. The CDC therefore, recommends not to compare the BRFSS datasets collected before 2010 and the ones collected thereafter due to this change in sampling methodology. For this study, the BRFSS data regarding the implementation of CDC-DNPAO in each state from 1998 to 2010 were collected by the authors. This study assumes that all residents living in CDC-DNPAO participating states were exposed to protective effects of CDC-DNPAO based on social-ecological model as aforementioned. Individual-level obesity risk factors used in the current study included sex, age group, race, marital status, educational attainment, household income level and smoking behaviors. Sex was classified male (reference) and female. Age group was categorized 62 into 18-24 years old (reference), 25-34 years old, 35-44 years old, 45-54 years old, 55-64 years old, and 65 years old and older. Race was divided into Whites (reference), Blacks, Asians/Pacific Islanders, American Indian/ Alaska Native, and Others. Marital status was grouped into Married (reference), Divorced, Widowed, Separated, Never married and Unmarried couple. Educational attainment was categorized into Under high school, High school, Some college (reference), and College and higher. Household income level was classified into Under $25,000, $25,000-35,000, $35,000-50,000, $50,000-75,000 (reference) and $75,000+. Smoking behavior was categorized into Current daily smoker (reference), Current occasional smoker, Former smoker, and Never smoked. 3.2.3. Analysis Following work by Monheit et al. (2011) and Pande et al. (2011), this study utilized a set of a logistic modeling and a quasi-experimental analysis to evaluate overall effect of CDC-DNPAO before and after its implementation. The first analysis uses logistic regression to estimate the overall effects of the CDC-DNPAO on obesity prevalence using the total sample of this study from 1998 to 2010. The default year of implementing CDC-DNPAO is 2000, which means the effects of CDC-DNPAO on the odds of obesity is examined before and after 2000. Obesityist = a1 + a2 STATEs + a3 YEARt + a4 TRENDt + a5 (STATEs x TRENDt) + a6 DNPAOist + a7 DURATIONist + a8Xist + eist In this model, the dependent variable (Obesityist) is a binary outcome of the ith individual in state s at time t (Obesityist = 1 when one is obese; otherwise 0). Obesity is 63 defined by body mass index (BMI) of 30 (kg/m2) or greater. The coefficients for STATE is state-specific fixed effects to account for time-invariant differences across states that may result in differences in obesity. YEAR controls for year-specific fixed effects possibly contribute to obesity. TREND is a linear time trend to account for secular changes in obesity apart from CDC-DNPAO implementation and state effects. The interaction term (STATE x TREND) accounts for any time-varying state-specific changes in obesity. DNPAO is set to 1 for all years that a state was funded from CDC-DNPAO in a year and is 0 otherwise. DURATION is the total number of years a state participated in CDC-DNPAO program (coded 1 in 2000 and 11 in 2010). The vector X contains a set of important obesity risk factors to control for in the logistic models, including sex, age group, educational attainment, racial group, marital status, household income level, and smoking habit. Finally, eist is a stochastic error term. The current study also examined the effects of natural occurrence in obesity policy by using a quasi-experimental analysis design, which compared overall obesity prevalence before-after the implementation of CDC-DNPAO. In this analysis three states (Massachusetts, North Carolina, and Texas) with CDC-DNPAO program in all study period (from 2000-2010) as a treatment group were compared with a pool of the thirteen states never funded from CDC-DNPAO as a control group. Considering their geographic proximity before-after obesity prevalence rates of three treatment states were compared with their corresponding control states within the same census regions, i.e. Massachusetts vs. Connecticut, North Carolina vs. Delaware and Virginia, and Texas vs. Alabama, Louisiana, and Mississippi. All the analyses included BRFSS sampling weights and post-stratification 64 adjustments to account for differences in probabilities of sampling selection and nonresponse, and to adjust for noncoverage of households without landline telephone (CDC, 2014). Stata version 14 was used for the all analyses performed in this current study (StataCorp, 2015). 3.3. 3.3.1. 65 66 67 68 3.3.2. 69 70 3.4. 71 72 3.5. 73 74 4. STUDY III: EXPLAINED AND UNEXPLAINED RACIAL AND REGIONAL INEQUALITY IN OBESITY PREVALENCE IN THE UNITED STATES. ABSTRACT There are substantial racial and regional inequalities in obesity prevalence in the United States. This study partitioned the mean Body Mass Index (BMI) and obesity prevalence rate gaps between non-Hispanic blacks and non-Hispanic whites into the portion attributable to observable obesity risk factors and the remaining portion attributable to unobservable factors at the national and the state levels in the United States. This study used a simulated micro-population dataset combining common information from the BRFSS and the U.S. Census data to obtain a reliable, large sample representing the adult populations at the national and state levels. It then applied a Blinder-Oaxaca reweighting decomposition method to decompose the black-white mean BMI and obesity prevalence inequalities at the national and state levels into the portion attributable to the differences in distribution of observable obesity risk factors and the remaining portion attributable to black-white differences in effects of risk factors. The mean racial difference in BMI was 18.5%. The racial difference in obesity prevalence was 20.6%. These differences represent the disparities in obesity between non-Hispanic blacks and non-Hispanic whites due to known obesity risk factors. There were substantial variations in how much the differences in distribution of known obesity risk factors explained the black-white gaps in mean BMI (-67.7% to 833.6%) and obesity prevalence (-278.5% to 340.3%) across states. The results from this study demonstrate that known obesity risk factors explain a small proportion of the racial, ethnic and regional inequalities in obesity prevalence in the United 75 States. Future etiologic studies are needed to further understand the causal factors underlying obesity and racial, ethnic and geographic inequalities. 76 4.1. Introduction The World Health Organization (WHO) defines obesity as a medical condition of abnormal or excessive adipose tissue accumulation that increases the risk of other health problems (World Health Organization, 2011). The body mass index or BMI is the 2). The BMI is the most frequently used measure to diagnose and describe obesity in medical and population studies. The BMI thresholds defined by the WHO are 18.5 to 24.9 for normal weight, 25.0 to 29.9 for overweight and 30.0 or greater for obesity (WHO, 2011). Obesity is now referred to as a chronic disease that can result in reduced quality of life and premature death due to its strong association with many comorbidities, including but not limited to cardiovascular disease, Type-II diabetes, osteoarthritis, stroke and certain types of cancers. Obesity is a major public health problem in the United States. The obesity prevalence rate for American adults was approximately 14 per 100 population in the early 1970s but it reached 36 in 2010 (CDC, 2010). There are also large racial and regional inequalities in obesity prevalence in the United States (CDC, 2010) with a significantly higher obesity prevalence among non-Hispanic blacks (herein after referred as blacks: 35.7, 95% CI: 35.6-36.3) compared to non-Hispanic whites (herein after referred as whites: 23.7, 95% CI: 23.5-23.9) in 2006-2008 (Rate Ratio: 1.5). This black-white obesity prevalence gap was consistent across U.S. states but it varied substantially, ranging from Oklahoma (5.4) to the District of Columbia (23.9) in 2006-2008 (CDC, 2009). Today obesity is responsible for 216,000 preventable deaths each year (Danaei et al., 2009) and the national health care spending directly and indirectly incurred from obesity is about 77 $190 billion. To reverse the increasing trends in obesity prevalence in the United States the Centers for Disease Control and Prevention (CDC) Division of Nutrition, Physical Activity, and Obesity (DNPAO) has funded participating state health departments to implement programs to reduce obesity prevalence since 2000 (CDC, 2012b; 2016). With this program funding, state health departments strengthened their ability to provide better health promotion, to implement effective nutrition and physical activity interventions, and to accumulate scientific evidence on obesity and its risk factors (Hamre et al., 2008). The obesity literature has shown little evidence of biological differences between blacks and whites to explain the racial disparities in obesity prevalence. More than 300 human genes or gene markers are potentially involved in causal obesity pathways but as of yet, genetics do not explain racial disparities in obesity (Bouchard et al., 2003; Health Central, 2015). Importantly, obesity researchers consider race as a social construct and therefore, focus on social factors that may contribute to the high rates of obesity among blacks in the United States. These researchers measure racial inequalities in obesity using the following approaches. First, mean obesity prevalence rates or mean BMI between blacks and whites is simply compared after stratifying known obesity risk factors. For example, Seo and Torabi (2006) found that the mean BMI of black women with a high school diploma was 31.0 compared to college graduates 27.7, while their white counterpart groups were 28.1 and 25.3, respectively. Second, an index or a measure is designed and used to calculate the racial inequalities in obesity. For example, Zhang and Wang (2004) use the Concentration Index to assess the degree of inequality in the distribution of obesity across socioeconomic status (SES) levels using the 1988-1994 National Health and Nutrition Examination Survey (NHANES) dataset. In their study, 78 lower SES was significantly associated with higher obesity prevalence rates for black men compared to white men but not for black women compared to white women, controlling for differences in age, low education and low income. This finding demonstrated that the role of SES may vary across gender within race. Third, racial disparities in obesity prevalence may be estimated using regression modeling. For example, Wen and Kowaleski-Jones (2012) using the 2003-2008 NHANES found that blacks were 1.2 times (odds ratio = 1.2) more likely to be obese than whites, controlling for differences in education and poverty levels. Fourth, obesity researchers have focused on the hypothesizes that economically, socially disadvantaged populations have been exposed to higher food prices, and limited access to food stores in their neighborhood (Raja et al., 2010). For instance, Zenk et al. (2005) found that distance to the nearest supermarket was 1.1 miles further from impoverished black-predominant census tracts than from white-predominant census tracts in Detroit metropolitan area (Wayne, Oakland, Macomb counties) and this inaccessibility may have been a factor that explained the higher obesity -city areas. Moreover, Jetter and Cassady (2006) reported that higher prices of groceries were a hindrance to consume healthier food in low-income neighborhoods using market-basket surveys conducted in 25 stores in Los Angeles and Sacramento, California. Lastly, the differences in cultural norms or attitudes may also explain racial inequalities in obesity (Robert and Reither, 2004). In terms of culture, Millstein et al (2008) using the National Physical Activity and Weight Loss Survey (2002) found that black women were more accepting of larger body sizes as perceived ideal body; while Jackson and McGill (1996) found that black male college students preferred 79 females with larger body types compared to white male college students. Caprio et al. (2008) argued that these attitudes and perceptions on body image are shared and transmitted from black parents to their children. In sum, these previous approaches to measure racial disparities in obesity improve our understanding of the problem by contextual risks. There is a need to partition the racial gaps in obesity in itself the portions explainable and unexplainable with known risk factors in order to target future obesity policy and programmatic interventions. This study therefore, used an advanced Blinder-Oaxaca decomposition technique to decompose the black-white obesity gap into the explained portioni.e. due to the differences in the values of covariates; and the remaining portion that was unexplainablei.e., due to differences in the effects of covariates. There are a handful of studies using Blinder-Oaxaca decomposition techniques to examine racial gaps in obesity among population groups. For example, Dutton and McLaren (2011) used a standard decomposition technique to study the regional disparities in mean BMI at the province level in Canada. Using data from the 2004 Canadian Community Health Survey, these researchers (the highest mean BMI in Canada) provinces were explained by the differences in known obesitregions were mostly explained by the unexplainable differences in effects of obesity risk factors on BMI. Johnston and Lee (2011) used the 2003-2006 NHANES and found that differences in energy intake explained approximately 48% of the difference in the black-white mean BMI, 44% of the average waist-to-height ratio differences between blacks 80 and whites and 38% of the obesity prevalence differences between black and white females aged 20-74 years. They also found that differences in energy expenditure contributed to 13% of the black-white mean BMI difference, 16% of the average waist-to-height ratio difference between blacks and whites and 11% of the obesity prevalence difference, between black and white females 20-74 years. Finally, Sen (2014) using a sample drawn from the BRFSS found that the mean BMI gap between black and white females in Alabama and Mississippi was 4.07 BMI units, only 8% of which was explained by demographic and health behavioral variables. Interestingly Sen also found that there was no statistically significant difference in the mean BMI between black and white males in these same states suggestive of some differences in obesity underlying causal factors While the previous studies using Blinder-Oaxaca decomposition techniques have begun to shed light on the racial or regional gaps in obesity, they are limited by their population diversity and geography distribution: i.e. Johnston and Lee (2011) used only women in their analysis; Sen (2014) analyzed a sample only from Alabama and Mississippi; and the study of Dutton and McLaren (2011) was for Canadian populations and provinces. This study thus aims to partition the black-white gap in BMI and obesity prevalence into the explained and unexplained portions of contributing factors at the national and state levels in the United States by adopting a Blinder-Oaxaca reweighting decomposition method. 81 4.2. Methods 4.2.1. Study Area The study area includes all 50 states and Washington D.C. in the United States using the 2010 county-census boundaries. 4.2.2. Data create a simulated dataset by which to calculate obesity prevalence. The BRFSS is a state-based, self-reported health survey system collected by the CDC to gather information on disease outcomes, health risk behaviors, preventative health practices, and health care access (CDC, 2013). However, the sample sizes of some population groups in the BRFSS are too small to estimate stable race-stratified obesity prevalence rates across all states. In addition the county identifiers for some rural or sparsely-populated counties are not released in the BRFSS to protect the confidentiality of respondents. To address these problems, a simulated population dataset was generated by using a spatial microsimulation technique (Ballas et al., 2005; Rahman, 2009; Lovelace and Ballas, 2013; Koh et al., 2015). older at the county level across states in the United States. There were a total of over 211 million records of adults (blacks, n= 29,903,955 (14%); whites, n=181,225,439 (86%)). The variables of interest to study the racial gap in obesity included the black and white mean BMI and their respective obesity prevalence rates. The individual variables were categorized into: sex (male and female); age (18-24, 25-34, 35-44, 45-54, 55-64, 65 years old and older); marital status (never married, married, separated, widowed, and divided); 82 educational attainment (under high school, high school, some college, and college and higher); household income levels (under $35,000, $35,000-50,000, $50,000-75,000, and $75,000 and higher); and smoking behaviors (current daily smoker, current occasional smoker, former smoker, and never smoked). The contextual-level variables to evaluate the environmental differences between blacks and whites included county level poverty rates, county Gini coefficients of income inequality (the 2010 U.S. Census), and the number of healthy grocery stores (USDA, 2011). In addition, the implementation (yes or no) and total duration (0-10 years) of the from and the literature, region (ref. Northeast, Midwest, South and West) for each state was also included for general geographic reference (the 2010 U.S. Census). The predicted power of these individual and contextual-level variables was validated with a linear regression model of estimating individual level BMI or obesity as the outcome variable (Table 4.1). 83 Table 4.1. Estimated Mean BMI and Obesity Using Ordinary Least Squares and Logit Regression and Explanatory Risk Factors, United States, 2010. Dependent Variables BMI (OLS) Obese (Logit) Coeff. Std.Err. P>|t|2 Coeff. Std.Err. P>|t| Black (ref. White) 1.6113 0.0013 *** 0.4425 0.0004 *** Sex (ref.=Male) -0.4068 0.0009 *** 0.0091 0.0003 *** Age (ref. 18-24 yrs old) 2.6166 0.0017 *** 0.6800 0.0007 *** 3.5394 0.0017 *** 0.9388 0.0007 *** 3.5983 0.0017 *** 0.9433 0.0007 *** 3.6914 0.0018 *** 0.9686 0.0007 *** 2.0473 0.0019 *** 0.4524 0.0007 *** Marriage (ref.= Not Married) -0.1403 0.0012 *** -0.0008 0.0004 ** 0.1037 0.0021 *** 0.0591 0.0007 *** -0.8055 0.0021 *** -0.1852 0.0008 *** -0.3934 0.0016 *** -0.0671 0.0006 *** Education (ref.=Under High School) -0.1839 0.0014 *** -0.0387 0.0005 *** -0.2232 0.0014 *** -0.0709 0.0005 *** -1.2923 0.0016 *** -0.4257 0.0006 *** Household Income (ref.=Under $35,000) -0.6059 0.0013 *** -0.1654 0.0005 *** -0.7689 0.0013 *** -0.2229 0.0005 *** -1.4760 0.0013 *** -0.4491 0.0005 *** Smoking (ref.=Current Daily Smoker) 0.5334 0.0021 *** 0.0953 0.0008 *** 1.8569 0.0014 *** 0.5015 0.0005 *** 1.4614 0.0013 *** 0.3858 0.0005 *** 0.0111 0.0001 *** 0.0039 0.0000 *** -0.9274 0.0142 *** -0.3658 0.0052 *** -0.5385 0.0043 *** -0.1528 0.0016 *** -0.1374 0.0011 *** -0.0319 0.0004 *** 0.0029 0.0002 *** 0.0010 0.0001 *** 0.2589 0.0015 *** 0.0827 0.0005 *** 0.3007 0.0014 *** 0.0944 0.0005 *** -0.2972 0.0015 *** -0.0989 0.0006 *** 25.8399 0.0063 *** -1.5539 0.0023 *** R2/ Pseudo R2 0.0649 0.034 84 Table 4.1. . Notes: 1. Two regression models were used to validate the effects of covariates on BMI (OLS) and obesity prevalence rates (logit). The categorical variables and their reference groups (hereafter ref.) were race (ref.: whites); sex (ref. male); age (ref .: 18-24, 25-34, 35-44, 45-54, 55-64, 65 years old and older); marital status (ref.: never married, married, separated, widowed, and divided); educational attainment (ref.: under high school, high school, some college, and college and higher); household income levels (ref. under $35,000, $35,000-50,000, $50,000-75,000, and $75,000 and higher); smoking behaviors (ref.: current daily smoker, current occasional smoker, former smoker, and never smoked); and census region (ref. Northeast, Midwest, South and West). County level poverty rates, county Gini coefficients and the number of healthy grocery stores. the implementation (yes or no) and total duration (0-10 years) 2. *** p<0.005 3. DNPAO: Division of Nutrition, Physical Activity, and Obesity 4.2.3. Analysis A spatial microsimulation technique was used to generate the simulated population data. As aforementioned the original BRFSS may have blurred county identifiers for some population data for small geographic areas where existing survey and/or census data are generated through an iterative proportional fitting (IPF)-based deterministic spatial microsimulation method (Lovelace and Ballas, 2013). With this technique, each respondent in the 2010 BRFSS was replicated and allocated to counties based on the proportion to common demographic characteristics in the 2010 BRFSS and 2010 SF1 U.S. Census data. Racial and regional gaps in obesity were decomposed using an inverse probability weighting (IPW) decomposition method proposed by Elder et al. (2011). Unlike the original Blinder-Oaxaca decomposition technique to focus the mean differences of covariates between groups, this method assumes that the differences in distributions of 85 covariates (e.g. age and education) between population groups are attributable for the gap in the outcome variable (e.g. BMI or obesity prevalence). A population group (e.g. the whites) is reweighted so that it has similar distributions of covariates with the other population group (e.g. the blacks) in this method. This study reweighted the individual and contextual level characteristics (obesity risk factors) of whites to have a similar distribution as those of blacks. Suppose f (o | g) is the probability density of obesity for group g and F(o | g) is the cumulative distribution of obesity risk factors x for group g. Then f (o | g) for the whites and the blacks are defined as the equation (1) and (2): (1) f (o | g = W) = f ( o; go|x = W, gx = W); and (2) f (o | g = B) = f ( o; go|x =B, gx = B). The equation (3) represents the counterfactual condition when the whites have the characteristics and obesity: (3) f (o; go|x = W, gx = B) . The equation (4) and (5) explain how the counterfactual density in (3) can be calculated from a weighted function of the actual whites with the weights of WB (x): (4) f (o; go|x = W, gx = B) , where the weights of WB (x) are calculated from (5) WB (x) = x 86 (6) f (o | g = B) - f (o | g = W) = [f (o; go|x = W, gx = B) - f (o | g = W)] + [f (o | g = B) - f (o; go|x = W, gx = B)]. In the right hand of the equation (6), the first part defines the explainable portion of the obesity prevalence gaps between the blacks and whites and the latter part defines the unexplainable portion of the gaps with obesity risk factors used in this study. The Blinder-Oaxaca decomposition technique analyses were performed with STATA 13 (StataCorp, 2013). 87 Table 4.2. Proportions of Black and White Racial Groups and Other Descriptive Statistics, United States, 2010. Variables Mean Black White White Weighted* Sex Male 0.4058 0.4940 0.4175 Female 0.5942 0.5060 0.5826 Age 18-24 0.1693 0.1231 0.1685 25-34 0.2183 0.1576 0.2170 35-44 0.1706 0.1599 0.1703 45-54 0.1835 0.1908 0.1849 55-64 0.1399 0.1716 0.1404 65+ 0.1183 0.1971 0.1189 Education High School- 0.1895 0.1279 0.1956 High School 0.3114 0.2824 0.3026 Some College 0.3176 0.3230 0.3138 College+ 0.1815 0.2666 0.1880 Household Income $35,000- 0.6543 0.4198 0.6519 $35,000-$50,000 0.1271 0.1466 0.1267 $50,000-$75,000 0.1053 0.1576 0.1053 $75,000+ 0.1133 0.2761 0.1161 Smoking Behaviors Current-everyday 0.1455 0.1518 0.1464 Current-someday 0.0796 0.0522 0.0802 Former Smoker 0.1631 0.2760 0.1655 Never Smoked 0.6117 0.5200 0.6080 County Poverty Rates 17.4576 14.9819 17.3555 County Income Gini Coeff. 0.4617 0.4447 0.4622 Grocery Stores 0.2350 0.1978 0.2364 88 Table 4.2. . Variables Mean Black White White Weighted* DNPAO 0.5346 0.5783 0.5475 DNPAO Duration 5.7492 6.0206 5.8750 Census Region Northeast 0.1693 0.1815 0.1693 Midwest 0.1734 0.2360 0.1734 South 0.5683 0.3611 0.5683 West 0.0890 0.2213 0.0890 * White Weighted denotes the estimates obtained under the hypothetical condition that the whites pulation characteristics with its own associations with characteristics and obesity. Source: The Authors; U.S. Census Bureau (2010); USDA (2011). 4.3. Results Table 4.2 reports the descriptive characteristics of the variables used in this study by racial gr4.2, the population characteristics of the whites became similar to blacks after the reweighting process. For example the proportions of males and females are 41% and 59% for blacks and 49% and 51% for whites, respectively but weighted whites have similar distribution in sex (male 41% and female 59%). Compared to whites, blacks had disadvantaged status in terms of age structure, education, and household income. Blacks have also higher mean county poverty rates and income Gini coefficients than whites. As summarized in Table 4.3, the mean BMI for whites and blacks at the national level was 27.6 and 29.6, respectively. The total BMI gap between whites and blacks was therefore, 1.9. The hypothetical (reweighted) BMI for whites was 28.02 under the counterfactual that the whites had the same population characteristic distributions as the blacks. This implies that 18.6% (0.4) of the mean BMI gap between whites and blacks 89 were explained by the differences in distributions of age, marital status, education, household income, smoking habit (individual-level variables), county poverty rates, county Gini coefficient, the number of grocery stores, DNPAO, the duration of DNPAO, and census region (contextual-level variables). At the national level obesity prevalence rates for whites and blacks were 28.3 and 40.4, respectively; only 20.1% of the difference in the black-white gap was explained by the differences in population characteristic distributions and known obesity risk factors. Table 4.3. Differences in Mean BMI and Obesity Prevalence for Blacks and Whites, United States, 2010. White (A) White Weighted (B) Black (C) Mean BMI 27.66 28.02 29.60 Total Gap (C-A) 1.94 Gap (%) Explained (B-A): 0.36 (18.56%) Unexplained (C-B): 1.58 (81.44%) Obesity Prevalence Per 100 Population 28.29 30.77 40.35 Total Gap (C-A) 12.06 Gap (%) Explained (B-A): 2.48 (20.56%) Unexplained (C-B): 9.58 (79.43%) Source: The Authors; U.S. Census Bureau (2010); USDA (2011). Table 4.4 list the gaps in the mean BMI and mean obesity prevalence rates between whites and blacks by state. Compared with whites, the mean BMI gaps for blacks were 3.0 or larger in the states of Oregon (B-W: 5.3), Washington D.C. (3.7), Hawaii (3.3), and Virginia (3.0), while whites had higher mean BMIs than blacks in Idaho (B-W: -1.98), New Hampshire (-0.6), Wyoming (-0.6), Utah (-0.5), Montana (-0.5), and Maine (-0.4). Over 40% of mean BMI gaps were explained by the differences in distributions of age, marital status, education, household income, smoking habit, county poverty rates, county Gini coefficient, the number of grocery stores, DNPAO, the duration of DNPAO, and 90 census region in four states, including Washington, Minnesota, Kansas, and Oklahoma. On the contrary, the known obesity risk factors used in this study could only explain 10% or less of the mean BMI gap in Florida, Delaware, South Carolina, Louisiana, Maryland, Oregon, New York, and New Jersey. 91 Table 4.4. Mean BMI Gaps between Blacks and Whites, United States, 2010 States Overall Mean BMI White BMI (A) White Weighted BMI (B) Black BMI (C) Explained Gap (B-A) Explained Gap % North Dakota 27.68 27.62 27.88 27.65 0.26 Nevada 27.36 27.46 28.44 27.72 0.98 Washington 27.57 27.74 28.19 28.51 0.45 Minnesota 27.34 27.37 27.64 27.98 0.27 Kansas 28.13 28.03 28.70 29.57 0.67 Oklahoma 28.17 27.96 28.55 29.36 0.58 West Virginia 28.34 28.38 28.68 29.19 0.30 Utah 26.67 26.62 26.43 26.09 -0.19 Georgia 28.02 27.68 28.16 29.06 0.48 South Dakota 27.96 27.76 27.97 28.43 0.21 Indiana 28.27 28.14 28.66 29.82 0.52 California 27.13 27.42 27.88 28.92 0.46 Iowa 27.93 27.96 28.27 29.00 0.31 Massachusetts 27.15 27.21 27.45 28.02 0.23 Alaska 28.04 28.11 28.45 29.41 0.34 New Mexico 27.61 27.31 27.75 29.17 0.43 Texas 28.17 28.04 28.44 29.82 0.40 Nebraska 27.86 27.85 28.10 29.01 0.25 Mississippi 28.97 28.29 28.70 30.38 0.41 Arizona 27.40 27.28 27.37 27.73 0.09 Colorado 26.71 26.60 26.84 27.96 0.24 Tennessee 28.27 28.07 28.38 29.93 0.31 Arkansas 28.41 28.18 28.46 29.96 0.28 Michigan 28.37 28.17 28.46 29.99 0.28 Connecticut 27.37 27.10 27.51 29.77 0.40 Missouri 28.26 28.11 28.33 29.72 0.22 Rhode Island 27.77 27.68 28.00 30.01 0.31 Alabama 28.62 27.98 28.32 30.55 0.34 North Carolina 28.14 27.65 27.96 30.09 0.31 Pennsylvania 28.11 27.93 28.19 30.00 0.26 Virginia 28.00 27.57 27.93 30.59 0.37 Illinois 27.46 27.17 27.46 29.62 0.28 Vermont 27.07 27.06 27.34 29.52 0.27 Kentucky 28.44 28.25 28.55 31.01 0.30 Ohio 28.14 27.95 28.16 29.93 0.21 Florida 28.09 27.76 28.02 30.35 0.26 92 Table 4.4. ( States Overall Mean BMI White BMI (A) White Weighted BMI (B) Black BMI (C) Explained Gap (B-A) Explained Gap % Delaware 28.15 27.79 27.99 29.83 0.20 South Carolina 28.37 27.71 27.94 30.20 0.23 Louisiana 28.54 27.91 28.06 29.85 0.16 Maryland 28.06 27.75 27.82 29.21 0.07 Oregon 27.65 27.73 27.90 33.02 0.17 New York 27.44 27.24 27.28 28.56 0.04 New Jersey 27.09 27.11 27.14 28.48 0.03 Wisconsin 27.81 27.73 27.71 29.85 -0.02 Washington D.C. 26.72 24.71 24.55 28.43 -0.17 Hawaii 26.61 26.30 26.04 29.60 -0.26 Idaho 27.43 27.37 27.56 25.38 0.19 Montana 27.21 27.09 27.27 26.64 0.17 Wyoming 27.44 27.44 27.66 26.88 0.22 New Hampshire 27.50 27.56 27.83 26.97 0.26 Maine 27.77 27.78 28.01 27.41 0.23 In Table 4.4, Oregon, Washington D.C., Alaska, Virginia, and Connecticut were among the top 4 states with the highest black-white obesity prevalence gaps while whites had higher obesity prevalence rates in South Dakota, Montana, North Dakota, Arizona, and Idaho. Over the 40% of black-white obesity prevalence gaps in Washington, Minnesota, Kansas and Oklahoma were explained with obesity risk factors whereas less than 10% of the gaps were explainable in Delaware, South Carolina, Louisiana, Maryland, Oregon, New York and New Jersey. 93 Table 4.5. Gaps in Mean Obesity Prevalence Rates per 100 Population (%) between the Blacks and the Whites, 2010. States Overall Obesity Rates White (A) White Weighted (B) Black (C) Explained Gap (B-A) Explained Gap % Maine 28.86 28.84 30.95 29.46 2.11 340.3% Washington 27.52 28.42 32.02 31.51 3.60 116.6% New Hampshire 26.51 26.81 29.58 30.44 2.77 76.4% Iowa 29.68 29.99 31.54 32.96 1.55 52.3% California 25.44 26.58 29.76 34.09 3.18 42.4% West Virginia 33.11 33.33 35.20 37.92 1.87 40.7% Arkansas 33.20 32.60 35.66 40.33 3.06 39.6% Nevada 25.61 25.35 28.40 33.44 3.05 37.7% Minnesota 26.59 26.61 29.35 34.07 2.74 36.7% Oklahoma 31.81 30.12 33.70 40.05 3.58 36.1% Kansas 31.65 30.64 34.86 42.66 4.22 35.1% Indiana 32.15 31.82 34.67 40.18 2.85 34.0% Louisiana 33.40 30.00 32.90 40.93 2.90 26.5% Vermont 24.62 24.60 26.63 32.49 2.03 25.8% Missouri 32.13 31.43 33.13 38.39 1.70 24.5% Massachusetts 24.98 24.91 26.34 31.08 1.43 23.2% Kentucky 33.79 33.07 35.70 44.95 2.63 22.2% Connecticut 25.73 24.11 27.94 41.41 3.83 22.1% Colorado 22.40 21.90 24.13 32.54 2.23 21.0% Delaware 31.26 28.83 31.27 40.83 2.44 20.3% Texas 31.93 31.13 32.81 39.42 1.68 20.3% Michigan 33.21 31.85 34.25 44.00 2.40 19.8% Rhode Island 28.51 28.12 30.25 38.93 2.13 19.7% Georgia 31.85 28.38 30.43 40.04 2.05 17.6% Virginia 30.55 27.83 30.95 45.70 3.12 17.4% North Carolina 30.69 27.59 30.25 43.25 2.66 17.0% New Mexico 27.90 25.46 27.69 38.70 2.23 16.9% Wyoming 25.85 25.91 26.70 30.77 0.79 16.2% Alabama 33.62 29.58 32.18 45.96 2.60 15.8% Tennessee 32.55 30.91 33.16 45.57 2.25 15.3% Pennsylvania 30.75 29.74 31.67 42.69 1.93 14.9% South Carolina 33.08 29.08 31.35 44.70 2.27 14.6% Florida 30.90 28.87 30.91 44.57 2.04 13.0% Nebraska 28.64 28.45 29.81 39.19 1.36 12.6% New Jersey 25.28 25.42 26.41 33.96 0.99 11.5% 94 Table 4.5. . States Overall Obesity Rates White (A) White Weighted (B) Black (C) Explained Gap (B-A) Explained Gap % Mississippi 35.77 31.88 33.19 44.16 1.31 10.6% Alaska 29.15 29.82 31.93 52.72 2.11 9.2% Maryland 30.01 28.17 29.00 37.45 0.83 9.0% Illinois 26.96 24.99 26.37 41.73 1.38 8.2% Wisconsin 29.24 28.54 29.56 44.38 1.02 6.5% Utah 23.06 22.59 23.21 33.42 0.62 5.7% Oregon 29.36 29.58 31.59 67.44 2.01 5.3% Ohio 30.88 29.47 30.14 43.67 0.67 4.7% Washington D.C. 21.43 9.07 9.50 32.13 0.43 1.8% New York 26.36 25.80 25.86 34.42 0.06 0.7% Hawaii 22.30 20.93 20.42 34.36 -0.51 -3.8% South Dakota 30.14 28.82 30.88 13.31 2.06 -13.3% Montana 24.87 23.90 25.83 14.21 1.93 -19.9% Arizona 26.88 26.54 28.13 21.76 1.59 -33.2% North Dakota 28.85 28.22 30.61 21.45 2.39 -35.3% Idaho 28.43 27.56 30.18 26.62 2.62 -278.5% 4.4. Discussion & Limitation This study documented that there were large gaps in mean BMI and mean obesity prevalence rates between whites and blacks at the national and state levels. This is the first research, to my knowledge, to report the gaps in mean BMI and obesity prevalence for all states in the United States without any omitted state. This study found that there were substantial differences in the distribution of obesity risk factors at the individual and contextual levels between blacks and whites. The whites appeared to have more protection from obesity (i.e., being married, college educated, and having high-household income) and less obesogenic environments (i.e., lower poverty, higher income equality), this difference in obesity risk factor distributions between two racial groups. 95 This study implies that the combined effect of all obesity risk factors on racial disparities in obesity varied by states. For example 59% of the state of mean BMI gap between blacks and whites were explained by known obesity risk factors. This implies that the black-white mean BMI gap could be narrowed if future obesity interventions could contribute to the differences in obesity risk factors in Washington. On -white mean BMI inequality was explained by these known risk factors, indicating there might exist other unknown obesity risk factors potentially influencing racial obesity inequality in New York. Therefore future obesity interventions need to examine the individual roles of each obesity risk factor on racial obesity inequalities to select target population or areas. In addition researchers need to focus finding other possible obesity risk factors from other perspectives including comorbidities, attitudinal and cultural norms, and/or other obesogenic environmental variables, including walkability and physical activities. This study had some limitations. First this study used a cross-sectional study of BRFSS data might be necessary to further understand the temporal variations in racial obesity gap in the United States. Second, while this study reported the overall gaps in mean BMI and mean obesity prevalence rates between blacks and whites the effect of each obesity risk factor on the gap(s) were not analyzed due to computational difficulty. Fourth, studies may need to investigate the role of each risk factor on racial disparities in obesity. Third the information of individual physical activity, one of the important obesity risk factors, was not included in the analysis model because the BRFSS over-simplied the information. Finally while this study focused on black-white obesity gap/inequality there might be obesity inequality 96 between other racial and ethnic groups that could be studied in the future. Despite these black-white obesity prevalence gaps from a qualitative approach. 4.5. Conclusion In 2010 there were large racial and regional inequalities in mean BMI and obesity prevalence in the United States and across states. This study found that that 19% of the mean BMI difference and 21% of the obesity prevalence inequality between blacks and whites were explained by known obesity risk factors. There are substantial variations in the mean BMI, obesity prevalence, and their decomposition results at the state level. The results from this study suggest that a small portion of known obesity risk factors are attributable to racial and regional inequalities in obesity prevalence in the United States. Future studies are needed to further explain the unknown portion of these the inequality in obesity prevalence between other races/ethnicities and geographies. 97 5. CONCLUSIONS 5.1. Overall contribution 98 99 100 5.2. Future Studies Obesity research has focused in the areas of individual and contextual risk factors, with most studies using public health surveys performed at the broader level of geography, e.g. national or state levels. In this regard it is very difficult to identify vulnerable populations at risk of obesity and to disentangle individual and environmental risk factors 101 at the local level, i.e. county or lower level. This study used a spatial microsimulation modelling approach has been shown to be an alternative option to study population health at the local level since it provides a simulated population dataset for any local geographic unit where public health surveys and census data are available. It is expected that other chronic diseases and their spatial patterns across time may be investigated with spatial microsimulation. In particular overlaying the spatial patterns of demographic and environmental measures and estimated disease prevalence at the local level would be helpful to investigate the associations between population, environment and disease prevalence. For example overlaying downscaled air pollution data and asthma prevalence at the local level will be useful to identify vulnerable populations at risk of chronic diseases. Evaluating the effect of public health policy is important when reviewing the effectiveness of public health interventions and designing future policies. A quasi-experimental study design is useful to investigate the effect of policy, which naturally divides intervention and control groups. Natural occurrence usually happens when policy interventions are implemented within and across population group(s) or geography. Future studies can utilize a quasi-experimental study design to parsimoniously examine the effect of policy interventions. Using a Blinder-Oaxaca technique is beneficial for researchers to investigate inequality in health among different population groups. Unlike other analysis methods a Blinder-Oaxaca technique decomposes the current health outcome differences into the portion explainable by known risk factors and the portion caused by unexplainable causes. By so doing it is possible for researchers to inequalities and the existing inequalities, where the future studies need to focus on to 102 reduce obesity, related chronic diseases and improve population health. This dissertation research demonstrated the utilization of new techniques in the study of obesity prevalence and the evaluation of populations, policies and programmatic interventions in the United States. It implemented these techniques to answer important theoretical and applied questions regarding obesity. Future research should continue to advance theory and new methods to address the important obesity and chronic disease epidemics in the United States and worldwide. 103 104 105 106 107 108 109 110 111 112 113 114 115