Obesity Causes: The Role of Female Labor Force Participation and Caloric Intake, Study notes of Research Methodology

Download Obesity Causes: The Role of Female Labor Force Participation and Caloric Intake and more Study notes Research Methodology in PDF only on Docsity! 1 The Determinants of Obesity in OECD Countries: A Cross-sectional and Time-Series Empirical Analysis By: Rachel Mecklenborg Submitted to Dr. Jacqueline Khorassani Econ 421 Capstone 2 Abstract Obesity is a growing epidemic worldwide, particularly among developed nations. The two main causes of obesity growth include an increase in caloric intake and a decrease in physical activity. This study is a cross-sectional and time-series empirical analysis that covers 54 observations from 6 different OECD countries over time. The study includes different socioeconomic factors that may cause an increase in caloric intake or decrease in physical activity that would ultimately affect obesity. To measure each variable’s effect on obesity, a fixed effects model that accounts for autocorrelation is used. The results of this empirical analysis show that none of the socioeconomic factors accounted for in this model significantly affect obesity. Variables available on an international basis that more directly account for calorie intake and physical activity would be necessary for future studies. 5 Literature Review The literature referenced to in this study of international obesity growth consist of five different papers written between the years of 2001 and 2007. They consider time- series and cross-sectional data, some focusing on obesity in the United States and some on international obesity. The papers build on each other as information collected in early papers is considered and referenced in later papers. Because of this, it is important to address the literature in chronological order while noting the progress of the study of obesity in the United States as well as worldwide obesity. Tomas Philipson (2001) did not perform regression analysis, but explains the necessity for economic research of worldwide obesity. The paper offers variables that may affect the worldwide growth in obesity. These variables include female labor force participation, market production of food, race, and level of physical activity. Philipson argues as the rate of female labor force participation has increased fewer home-cooked meals have been produced, and reliance on market production of food (i.e. fast food and dine in restaurants) has increased—thus increasing obesity. An example of a typical fast food meal includes French fries and a cheeseburger from the world’s most popular fast food chain McDonald’s with a total of 980 calories, or nearly half of the suggested daily caloric intake (www.acaloriecounter.com). Compare this with a home-cooked meal of baked chicken, a baked potato (and a tbsp of margarine), and broccoli at 289 calories to demonstrate the caloric difference (www.acaloriecounter.com). He states that many times race coincides with socioeconomic status, and races with lower socioeconomic status have higher rates of obesity. Lower income households buy more inexpensive 6 foods, which tend to be unhealthier than higher priced foods. Therefore, races with lower socioeconomic status buy unhealthy foods increasing obesity. The empirical papers on the subject of obesity have studied both the obesity in the United States as well as in the rest of the world. All of the literature reviewed measure obesity by a Body Mass Index (BMI) greater than 30. The major shortcoming of using BMI to measure obesity is that it does not account for muscle mass and may not be fully descriptive of an obese person. Shin- Yi Chou, Michael Grossman, and Henry Saffer (2002) as well as Rashad (2005) use cross-sectional and time-series data to analyze the growth of obesity in the United States by drawing on data from all 50 states over time. Both papers use similar independent variables to determine obesity. The independent variables shared include: per capita number of restaurants, clean indoor air laws, and price/ tax of cigarettes. The number of per capita restaurants is used as a measure of a state’s calorie intake. The higher the per capita number of restaurants, the higher the calorie intake. Additional background helps to explain this: The U.S. Surgeon General warns that “physical inactivity and super-sized meals are leading to a nation of oversized people.” Furthermore, a specific study entitled “Fast food consumption and increased caloric intake” by R. Rosenheck of the Harvard School of Public Health found that “sufficient evidence exists for public health recommendations to limit fast food consumption,” and that “the scientific findings and corresponding public health implications of the association between fast food consumption and weight are critical.” Both papers find the correlation between the per capita number of restaurants and obesity was positive and statistically significant. This independent variable proved to be the most statistically 7 significant of all variables in both papers, and a possible reason for this suggested in both papers is the increased value of time, meaning less time for home cooked meals and more dinning out or fast-food. Both the clean indoor air laws and cigarette tax variables attempt to capture the consumption of cigarettes. The nicotine in cigarettes speeds up metabolism and therefore decreases obesity (About.com Health's Disease and Condition). As the price or tax of cigarettes increase, the consumption of cigarettes declines. As a result there is more obesity because of less smoking. Similarly, more clean indoor air laws are expected to positively affect obesity. Other independent variables used by Chou, Grossman, and Saffer (2002) include years of formal schooling completed and marital status on an individual basis. Chou, Grossman, and Saffer argue that married people eat out less therefore decreasing obesity while single people eat out more increasing obesity. Another independent variable used by Rashad (2005) is the gasoline tax over states. If the tax on gasoline is high, more people will be willing to use other modes of transportation besides cars to keep transportation costs low. These papers drew on different data sources: Chou, Grossman, and Saffer (2002) used data from The Behavioral Risk Factor Surveillance System while Rashad (2005) utilized data from The Census of Retail Trade and the Facts and Figures on Government Finance. These papers focused on obesity in the United States laying the ground work for further worldwide obesity regression analysis. The remaining papers focus on international obesity. Maria L. Loureiro and Rodolfo M. Nayga (2005) as well as and Bleich (2007) both draw on OECD countries’ cross-sectional and time-series data on obesity. The common independent variable shared between these two papers is the female labor force participation across countries 10 Table 1: Independent Variables included in Equation 1 along with Expected Effect on Obesity Independent Variable Definition Expected Effect on BMI FLFPR Female Labor Force Participation Rate + PNI Per Capita Income + EDU Tertiary attainment for age group 25- 34 - FOOD CPI of Food / Overall CPI (base year 2000) - GAS CPI of Gas/ Overall CPI (base year 2000) - CO2 Carbon Dioxide Emissions in millions of tons + The variables included in this model are chosen according to relevancy in the previous papers reviewed, with specific attention to the international models of obesity because this paper has an international focus. As Philipson (2001) suggested, as the rate of female labor force participation has increased time has become a scarce resource, and therefore fewer home-cooked meals have been produced. Because of this, reliance on market production of food (i.e. fast food and dine in restaurants) has increased—thus increasing obesity. Therefore, female labor force participation is expected to have a positive effect on obesity. The per capita income is included in Equation 1 to capture the effect of economic development on obesity. All else equal, an increase in a country’s per capita income increases the ability of an average resident to purchase more food. Therefore, I expected the sign on the coefficient of per capita income to be positive. The attainment of tertiary education measures the percentage of all persons in the age group between 25-34 that have completed a high school (or equivalent) education. 11 The expected effect of an increase in the attainment of tertiary education would be a decrease in obesity because more “educated” individuals would know the causes and harms of obesity. The explanatory variable relative CPI of food measures the average price of food and non-alcoholic beverages, excluding purchases in restaurants, relative to the average price of all goods and services. This variable is predicted to have a negative effect on obesity. As food prices relative to other things decrease, consumers are expected to purchase more food and less of other goods and services. With an increase in food consumption, or calorie intake, obesity increases. According to the OECD Factbook, the relative CPI of energy includes fuels for motor vehicles, heating and other household uses. The relative CPI of energy for each county is included in the model to account for the amount of physical activity of a country. With a higher relative price of energy compared to other goods and services, people may opt not to use modes of transportation high in energy consumption like cars and instead walk or bike to save money on energy. Because of this, the predicted effected on obesity is negative. The CO2 emissions variable also accounts for a measure of the amount of physical activity in a country. The greater the CO2 emissions, the more transportation a country is using. With less CO2 emissions, citizens are using alternative modes of transportation like walking or biking, so the expected effect of CO2 on obesity is positive. 12 Descriptive Statistics Table 2: Minimum and Maximum Value of Each Variable and the Average Value of Each Variable Variable Minimum Maximum Mean BMI 2.7 Japan 1997 18.6 Luxembourg 2005 11.97 GDP $22,122.00 Finland 1997 $63,945.00 Luxembourg 2005 $27,717.05 FLFPR 45.5% Luxembourg 1997 73.5% Sweden 2001 61.84% EDU 20.2% Luxembourg 1997 53.2% Japan 2005 34.37% FOOD .94 Netherlands 2005 1.04 Luxembourg 2003 1.01 GAS .85 Luxembourg 1998 1.23 Sweden 2005 .99 CO2 7 Luxembourg 1999 and 2000 1223 Japan 2004 336 As Table 2 displays, on average about 12 percent of the population across the six countries of Japan, Sweden, Finland, United Kingdom, Luxembourg, and Netherlands over the period of study (1997-2005) were obese. The average per capita income across all six countries throughout the same period is $27,717.05, and on average over 60 percent of the adult female population were active in the labor force. The mean percentage of those attaining tertiary education among these countries was below half of the population of these countries at 34.72%. The relative food and gas CPI are measured by the ratio of food and gas CPIs respectively to the overall CPI. The average relative 15 correlation coefficients between each pair of independent variables included in Equation 1. Table 3: Correlation Matrix for All Explanatory Variables in Equation 1 As indicated in Table 3, there is no multicollinearity problem. The absolute values of all correlation coefficients are below .80 demonstrating no severe multicollinearity. CO2 EDU FLFPR FOOD GAS GDP CO2 1 0.662135138 -0.614373908 -0.271856642 -0.005221110 -0.338214687 EDU 1 -0.272118431 -0.011174176 0.407398751 -0.384373050 FLFPR 1 0.331845971 0.099000143 0.540106291 FOOD 1 0.577256975 0.505768335 GAS 1 0.236743437 GDP 1 16 Model Specification Choice of Estimation Method There are four options in estimating data that is both cross-sectional and time- series. These methods include using a pooled model with a common intercept, using the seemingly unrelated regression method, using the fixed effects model, or using the random effects model. The most appropriated estimation method for this data set is the fixed effect model. A pooled model assumes that there is no variation in intercept or the slope coefficients of Equation 1 across countries over time. Consequently, the deficiency of this model is that it ignores differences over time or among countries. Despite this fact, given the small number of observations (54) using pooled data gives the most degrees of freedom of all estimation options. The seemingly unrelated regression method uses separate regressions for each country. This allows the intercepts and slope estimates to differ, but the degrees of freedom for each regression would be less than 40. This would deter the estimation process from working properly and therefore would not work well for this data set. The fixed effects model and random effects model both allow different intercepts for each country, but the slope coefficients are the same for both models. A random effects model assumes that the differences among the intercepts of Equation 1 across various nations are due to random factors. A fixed effect model, on the other hand, assumes that the differences among the intercepts of Equation 1 across various nations are caused by structural differences. 17 For this data set, the fixed effects method or the pooled data model could be most appropriate. If the intercepts for each variable are the same, then using the pooled data would work best because there would be higher degrees of freedom. If the intercepts for each variable are different, then the fixed effects method would be best because this model acknowledges that the nations under study have structural differences. To determine which test is the best fit, an F-test is performed. An F-test can help us determine whether or not the intercepts of Equation 1 are significantly different across nations. The null and alternative hypotheses are as follows: Ho: B1=B2=B3=B4=B5=B6 Ha: B1, B2, B3, B4, B5, B6, B7 are not all equal The relevant F-stat is calculated as follows: Pooled – Fixed effect 636-26 610 F= q = 6 = 102 = 189 Fixed effect 26 .541 n-k 54-6 F-stat= 189 Fcritical= 3.9 At a 1% significance level, F-stat (189) is greater than Fcritical (3.9) , so we reject the null hypothesis that the intercept is the same for all five variables. This verifies that a fixed effect model should be used because the F- test found that the intercept for all five variables are significantly different from each other. 20 Empirical Estimation Results Table 3 below shows the estimation results of Equation 1 using three different techniques. The first technique relies on the pooled model specification. The second technique utilizes the fixed effect model, and finally the last technique adjusts the fixed effect model for autocorrelation. Table 3: Coefficients and T-Statistics of Independent Variables included in Equation 1 in Pooled Model, Fixed Effect Model, and Fixed Effect Model Adjusted for Autocorrelation Variable Pooled Data Fixed Effect (no adjustment for autocorrelation) Fixed Effect (with adjustment for autocorrelation) C -8.467803 N/A N/A CO2 0.006304 (3.189618) 0.003657 (0.356466) -0.002576 (-0.301572) EDU -0.378377 (-3.999143) -0.003475 (-0.065861) -0.015329 (-0.400162) FLFPR 0.639789 (4.963181) 0.222054 (2.037691) -0.194760 (-1.299686) FOOD -4.434043 (-0.126434) -3.575604 (-0.431141) -1.204261 (-0.157551) GAS -22.93483 (-2.067082) -0.136415 (-0.047565) 1.083145 (0.447114) PNI 0.000578 (5.466517) 0.000169 (2.932755) 4.41E-05 (0.819936) Adjusted R- squared 0.626140 0.982605 0.991468 Number of observations 54 54 54 Finland N/A -3.604463 27.12578 Japan N/A -14.42337 17.86290 Luxembourg N/A 1.239366 28.06259 21 Netherlands N/A -6.479647 23.58900 Sweden N/A -7.897111 24.56615 United Kingdom N/A 3.355760 38.24890 *Those black-bolded values express significance at a 1% level of significance **Those blue-bolded values express significance at a 5% level of significance The results in Table 3 display each variable’s coefficients and t-stats in a pooled data regression analysis, fixed effects regression analysis, and a fixed effects regression analysis corrected for autocorrelation. Many times pooled data will not take into consideration the effect of structural differences between countries. Because an F-test is performed, we learn that each country in this model is structurally different. The fixed effect model fails the autocorrelation test, so the model has to be adjusted accordingly. The first column in Table 3 displays pooled data in which structural differences and the problem of autocorrelation are ignored. The coefficient on the level of CO2 emissions (CO2) is significant at a 1% level and has a positive affect on obesity, which is consistent with my expectations. The value of the coefficient of CO2 can be interpreted as for every addition 1 million tons of CO2 emissions, the percentage of population that is considered obese in a country will increase by .006%. The coefficient of the percentage of tertiary attainment (EDU) is also significant at a 1% level and has a negative affect on obesity as predicted. The value of the coefficient of the percentage of tertiary attainment means that as the percentage of tertiary attainment increases by 1% in a country the percentage of obesity in a country decreases by about .37%. Additionally, the coefficient of female labor force participation rate (FLFPR) has a 1% level of significance and, as expected, positively affects obesity. The interpretation 22 of the value of the coefficient for female labor force participation rate is for everyone 1% increase in female labor force participation rate of a country the percentage of obese people will increase by .67%. Lastly, the coefficient of per capita income (PNI) is also significant at a 1% level and shows a positive effect on obesity as predicted. The value of the coefficient of per capita income can be interpreted as if the per capita income increases by 1 dollar, the percentage of obese people in a country’s population increases by .0005%. Furthermore, the coefficient on the relative CPI of gas (GAS) is found to be significant at a 5% level and matches my prediction as negatively affecting obesity. The interpretation of the value of the coefficient of the CPI of gas is as the price index of gas relative to all other prices increases by 1 the percentage of obese people in a country’s population decreases by 22%. The only coefficient that is not statistically significant in the pooled data is CPI of food, which was predicted to have negative affect on obesity. A comparison between the fixed effects model and the pool data model suggests that once we account for structural differences among various nations, the coefficients on the following three variables become insignificant: CO2 emissions, tertiary attainment, and CPI of gas. The coefficient for female labor force participation rate is still significant at a 1% level and the affect on obesity remains positive. The coefficient for per capita national income is now significant at a 5% level and still positively affects obesity. When we adjust the fixed effect model for autocorrelation, the adjusted R-squared increases while the number of significant coefficients decreases. This is a sign that there may be an omitted variable bias in the model that affects the error term over time. The 25 Works Cited 1. Fast Food Restaurants and Nutrition Facts Compared. 2009. A Calorie Counter. Mar. 2009 <http://www.acaloriecounter.com/fast-food.php>. 2. Economic Analysis of Adult Obesity: Results from the Behavioral Risk Factor Surveillance System (2002). National Bureau of Economic Research. Oct. 2002. National Bureau of Economic Research. Jan. & feb. 2009 <nber.org>. 3. International Dimensions of Obesity and Overweight Related Problems: An Economics Perspective (2005). National Bureau of Economic Research. 2002. National Bureau of Economic Research. Jan. & feb. 2009 <nber.org>. 4. Obesity. 2009. World Health Organization. March & april 2009 <http://www.who.int/topics/obesity/en/>. 5. THE SUPER SIZE OF AMERICA: AN ECONOMIC ESTIMATION OF BODY MASS INDEX AND OBESITY IN ADULTS (2005). National Bureau of Economic Research. Aug. 2005. National Bureau of Economic Research. Jan. & feb. 2009 <nber.org>. 6. WHY IS THE DEVELOPED WORLD OBESE (2007). National Bureau of Economic Research. Feb. 2007. National Bureau of Economic Research. Jan. & feb. 2009 <nber.org>. 7. The World- Wide Growth in Obesity: An Economic Research Agenda (2001). National Bureau of Economic Research. 2001. National Bureau of Economic Research. Jan. & feb. 2009 <nber.0rg>. 26 Data Sources 1. CO2 Emissions from Fuel Combustion. 2008. OECD Factbook 2008. March & April2009 <http://titania.sourceoecd.org/vl=1386641/cl=18/nw=1/rpsv /factbook2009/ htm>. 2. Employment Rates: Women. 2008. OECD Factbook 2008. March & april 2009 <http://titania.sourceoecd.org/vl=1386641/cl=18/nw=1/rpsv/factbook2009 /index. htm>. 3. Macroeconomic trends - Gross Domestic Product (GDP) - National income per capita 2008. OECD Factbook 2008. March & april 2009 <http://titania.sourceoecd.org/vl=1386641/cl=18/nw=1/rpsv/factbook2009 /index. htm>. 4. Obesity Rates. 2008. OECD Factbook 2008. March & april 2009 <http://titania.sourceoecd.org/vl=1386641/cl=18/nw=1/rpsv/factbook2009 /index. htm>. 5. Prices - Prices and interest rates - Consumer Price Indices (CPI) Prices CPI of Food. 2008. OECD Factbook 2008. March & april 2009 <http://titania.sourceoecd.org/vl=1386641/cl=18/nw=1/rpsv/factbook2009 /index. htm>. 6. Prices - Prices and interest rates - Consumer Price Indices (CPI) Prices CPI of Gas. 2008. OECD Factbook 2008. March & april 2009 <http://titania.sourceoecd.org/vl=1386641/cl=18/nw=1/rpsv/factbook2009 /index. htm>. Tertiary Education for age group 24-35. 2008. OECD Factbook 2008. March & april 2009 <http://titania.sourceoecd.org/vl=1386641/cl=18/nw=1/rpsv/ 27 factbook2009/index. htm>.