PAGE 1
1 ESSAYS IN THE ECONOMICS OF EDUCATION By KATIE ANN SHOWMAN A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY U NIVERSITY OF FLORIDA 2009
PAGE 2
2 2009 Katie Ann Showman
PAGE 3
3 To my husband
PAGE 4
4 ACKNOWLEDGMENTS I could not have completed this body of work without the assistance of many individuals. I th ank the members of my com mittee: Larry Kenny, David Denslow, Julia Graber and Sarah Hamersma. I also thank Steven Slutsky for attending my final defense. Their intelligent teachings thoughtful ideas and endless support have been invaluable to me. I extend thanks to thank my parents Richard and Carol Sherron for a lifetime of encouragement and unconditional love Finally, I would like to thank my husband, Brian, for giving me the courage to tackle the unknown. Without him, I would have never take n on or complete d this chal lenge
PAGE 5
5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ ............... 4 LIST OF TABLES ................................ ................................ ................................ ........................... 7 LIST OF FIGURES ................................ ................................ ................................ ......................... 9 ABSTRACT ................................ ................................ ................................ ................................ ... 10 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .................. 12 2 DOES THIS TEST MAKE ME LOOK FAT ? ECONOMIC CONSEQUENCES OF ADOLECENT WEIGHT AND HEIGHT ................................ ................................ .............. 14 Introduction ................................ ................................ ................................ ............................. 14 Data ................................ ................................ ................................ ................................ ......... 16 Methods ................................ ................................ ................................ ................................ .. 20 Empirical Results ................................ ................................ ................................ .................... 23 Effects of H eight and Weight on Test Score U sing OLS ................................ ................ 23 Effects of H eight and Weight on Test Score U sing IV ................................ ................... 24 Effects of Height and Weight on Test Sore U sing FE ................................ ..................... 25 Effects of Height and Weight on Studen t Absences U sing OLS ................................ .... 26 Effects of Height and Weight on Student Absences U sing IV ................................ ........ 27 Conclusion ................................ ................................ ................................ .............................. 27 3 EVIDENCE ON THE EFFECTS OF INTER SCHOOL DISTRICT COMPETITION: COMPARISIONS OF STATE LIMITED SCHOOL DISTRICT STATES WITH OTHER STATES ................................ ................................ ................................ ................... 39 Introduction ................................ ................................ ................................ ............................. 39 Literature Review ................................ ................................ ................................ ................... 41 Empirical Model ................................ ................................ ................................ ..................... 43 Data ................................ ................................ ................................ ................................ ......... 46 Results ................................ ................................ ................................ ................................ ..... 48 Conclusion ................................ ................................ ................................ .............................. 52 4 ANALYSIS OF FLORIDA CONSTITUTIONAL AMENDMENT ELIMINATING THE OFFICE OF PUBLIC SCHOOL TRUSTEES, 1956 ................................ ..................... 59 Introduction ................................ ................................ ................................ ............................. 59 Sc hool District Trustees and School Boards ................................ ................................ .......... 59 Senate Joint Resolution No. 638 ................................ ................................ ............................. 61 Data ................................ ................................ ................................ ................................ ......... 61
PAGE 6
6 Hypotheses ................................ ................................ ................................ .............................. 61 Results ................................ ................................ ................................ ................................ ..... 62 5 CONCLUSIO N ................................ ................................ ................................ ....................... 67 LIST OF REFERENCES ................................ ................................ ................................ ............... 68 BIOGRAPHICAL SKETCH ................................ ................................ ................................ ......... 71
PAGE 7
7 LIST OF TABLES Table page 2 1 on Median Height for Females ................................ ................................ .......................... 29 2 2 of Overweight, At Risk for Overweight, and Underweight Based on Median Height for Males ................................ ................................ .............................. 29 2 3 Summary Statistics for 1997 Cross Section ................................ ................................ ....... 30 2 4 .............. 31 2 5 Cross section regression results, effect of weight on test score; dependent variable is grade n ormalized PIAT math test score ................................ ................................ ............. 32 2 6 OLS regression results, effect of weight on test score; dependent variable is grade normalized PIAT math test score ................................ ................................ ....................... 33 2 7 IV regression results, effect of weight on test score; dependent variable is grade normalized PIAT math test score ................................ ................................ ....................... 34 2 8 Panel regression results, effect of weight on test score; dependent variable is grad e normalized PIAT math test score ................................ ................................ ....................... 35 2 9 OLS regression results, effect of weight on student absences; dependent variable is number of absences in the 1997 fall semester ................................ ................................ ... 36 2 10 IV regression results, effect of weight on student absences; dependent variable is student absences ................................ ................................ ................................ ................. 37 3 1 Characteristics of states with an e xogenous number of school districts ............................ 53 3 2 MarketSize variables summary statistics ................................ ................................ ........... 53 3 3 Variable summary statistics ................................ ................................ ............................... 54 3 4 Regression results, the effects of limiting competition on student learning; dependent variable is grade normalized PIAT math test scores ................................ .......................... 55 3 5 Regression results, education production function; dependent variable is grade normalized PIAT math test scores. ................................ ................................ .................... 56 3 6 Sample of MSAs negatively impacted by mandates restricting inter county or inter state competition among public schools. ................................ ................................ ........... 57 4 1 General election votes on amendment to abolish the office of school district trustees ..... 65
PAGE 8
8 4 2 Regression Results ................................ ................................ ................................ ............. 65
PAGE 9
9 LIST OF FIGURES Figure page 2 1 Students actual weight and students trying to gain or lose weight ................................ .... 38 3 1 Impact of the market size on student learning as the market size increases. ..................... 58
PAGE 10
10 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy ESSAYS IN THE ECONOMICS OF EDUCATION By Katie Ann Showman August 2009 Chair: Lawrence Kenny Major: Economics In this work, I prese nt three separate studies. In the first study, I examine the impact of childhood overweight on classroom outcomes. I investigate whether height and/or weight have an impact on student test scores. I find that overweight students receive lower test score s than healthy weight students. While I find a negative impact of overweight among all adolescents, white females are especially at a disadvantage, receiving test scores 0.19 to 0.29 standard deviations lower than white females of recommended weight. I a lso find that students in the bottom fifth percentile for height receive lower test scores and that growth spurts are positively related to test scores among white males. In the second study, I consider the impact of restricting school district size on stu dent outcomes. Tiebout theory and empirical research tell us that within a metropolitan area people will sort themselves based on income and preference for schooling. Often, however, local and state government policies limit Tiebout choice. I contribute to the literature by using a unique method of dealing with the endogenous nature of the number of school districts. I compare student learning among states that exogenously limit Tiebout choice with student performance in other states. I find strong evi dence that restricting competition among public school districts negatively impacts student learning. Binding laws that mandate county or state wide school
PAGE 11
11 districts negatively impact student test scores by 0.09 to 0.40 standard deviations. In the larges t school districts with the most students, these restrictive mandates are especially harmful. In the final study, I explore the evolution of the system of voter representation in school decisions after Florida adopted countywide districts. I consider the vote on a 1956 amendment to the Florida constitution. The amendment eliminated a school administrative office and eliminated representation for portions of each county wide school district. I find that the support for the amendment was strongest in rural heterogeneous counties.
PAGE 12
12 CHAPTER 1 INTRODUCTION I conduct three separate studies in this paper: (1) Does this test make me look fat? Economic consequences of adolescent weight and stature, (2) Evidence of the effects of inter school district competi tion: comparisons of state limited school districts with other states, (3) Analysis of Florida constitutional amendment, eliminating the office of school district trustees, 1956. In the first study, I examine the impact of childhood height and weight on test scores. Adolescents receive differential treatment from peers based on outward physical characteristics. This attention may preoccupy the student, causing her to absorb less information in the classroom. I use the National Longitudinal Survey of Yo uth (NLSY97) to investigate whether height and/or weight have an impact on student test scores. I find that overweight students receive lower test scores than healthy weight students. While I find a negative impact of overweight among all adolescents, wh ite females are especially at a disadvantage, receiving test scores 0.19 to 0.29 standard deviations lower than white females of recommended weight. Regression results also reveal that students in the bottom fifth percentile for height receive lower test scores and that growth spurts are positively related to test scores among white males. In my next study, I examine the impact of limiting the number of public school districts on student outcomes. Tiebout theory and empirical research tell us that within a metropolitan area people will sort themselves based on income and preference for schooling. Often, however, local and state government policies limit Tiebout choice. The difficulty in assessing the impact of limiting Tiebout choice on academic performa nce is that the degree of competition within a market is usually endogenous. I contribute to the literature by using a unique method of dealing with the endogenous nature of the number of school districts. I compare student learning among
PAGE 13
13 states that exo genously limit Tiebout choice with student performance in other states. I find strong evidence that restricting competition among public school districts negatively impacts student learning. Binding laws that mandate county or state wide school districts negatively impact student test scores by 0.09 to 0.40 standard deviations. In the largest school districts with the most students, these restrictive mandates are especially harmful. In my final chapter I explore voting on a Florida constitutional amendme nt. The 1956 amendment that I study eliminates the position of school district trustees. This change reduces representation of certain communities within each school district. I find that in 1956, support for this amendment was related to county homogen eity. Specifically, counties that strongly supported abolishing school district trustees were predominately wealthy, white, urban counties where Tiebout sorting was complete.
PAGE 14
14 CHAPTER 2 DOES THIS TEST MAKE ME LOOK FAT? ECONOM IC CONSEQUENCES OF ADOLECE NT WEIGHT AND HEIGHT Introduction Economists have extensively investigated the effects of physical characteristics (including race, gender, height, weight, and beauty) on labor market outcomes. Sargent and Blanchflower (1994), Averett and Korenman (1996), and Cawley (2004) find that obesity is associated with lower wage rates. Persico, Postlewaite, and Silverman (2004) and Sargent and Blanchflower (1994) find a positive relationship between height and earnings. Hamermesh and Biddle (1994) find that plain men and women earn five to ten percent less than their average looking colleagues. Some of the wage premium received by healthy weight and tall people may be due to their being more productive than their counterparts. Hamermesh and Biddle (1994), Avere tt and Korenman (1996), and Persico et al (2004) attempt to control for productivity differences by including a variable for the highest grade attained in each of their regressions. 1 However, if individuals with less desirable physical characteristics lea rn less in any given grade of school, they should be less productive in the labor market and some portion of the estimated beauty premium would actually be due to productivity differences. There has been very little research on the effects of physical cha racteristics on educational outcomes. This would provide valuable evidence on whether those with healthy weight and those who are taller have better cognitive skills. I contribute to the discipline by filling in the gap in this literature. There are ma ny reasons that we may expect to see a negative relationship between height/weight and student outcomes. Adolescents receive differential treatment from peers based 1 Only Sargent and Blanchflower (1994) use a different measure, test scores, to control for education level.
PAGE 15
15 on outward physical characteristics. A student whose appearance is different from others in height and/or weight is often a victim of bullying and teasing (Jansen et al, 2004), which is likely to lead to more absences. In addition to negative attention from peers, the student may receive negative attention or less instruction from her teacher s and parents. One would expect a suboptimal learning environment to be reflected in student outcomes. It is also possible that poor student outcomes may cause weight gain. A student who performs poorly in school may experience stress, causing her to tu rn to food for comfort. Finally, there may be a third factor causing both poor student performance and an unhealthy weight. It may be that a disadvantaged home life causes both weight gain and low test scores. On the other hand, there could be a positi ve relationship between student weight and academic outcomes. A student whose appearance is different from others may choose to spend more time studying instead of socializing with peers. Her teachers may see her potential and give her more attention tha n her average looking peers. This may lead to better student outcomes. Let us turn to the few studies of the effects of appearance on academic outcomes. Falkner et al (2001) find that underweight adolescent boys are more likely than their healthy weigh t peers to dislike school and less likely to expect to finish college. Datar, Sturm, and Magnabosco rch is closest to that done in this paper. He finds that a two standard deviation increase in weight leads to a 9 percent decrease in grade point average (GPA) among white females, but was unable to find a consistent impact of weight on outcomes among mal es or nonwhite females on the GPA.
PAGE 16
16 This paper uniquely contributes to the literature in three ways. First, student test scores are within the school. There have been no studies of the effect of appearance on test scores for students past kindergarten. Second, to my knowledge I am the first to explore the impact of hei ght on student outcomes. Finally, I explore the impact of student height and weight on student absences, which provides important evidence on the mechanism by which appearance affects test scores. This paper uses various measures of weight and three diff erent statistical methods: ordinary least squares (OLS), instrumental variables (IV) and fixed effects (FE) models. I find that students w ith undesirable physical traits indeed learn less in school than other students. Specifically, weight negatively im pacts test scores for white females, black females, and white males. Additionally, holding weight constant, height positively impacts test scores for white females, black females, and white males. The beauty premium found in the labor market thus reflect s some productivity differences. I also find that those with healthy weight and who are taller miss fewer days of school; numerous studies have found that students who miss fewer days have higher test scores. Data The 1997 National Longitudinal Survey of Youth (NLSY) follows a sample of 8,984 students in their transition from school to work beginning in 1997. This group is still being followed annually, and the most recent data available are from 2005. Only in the initial 1997 survey were parents intervi ewed and all students tested. A fraction of students took a standardized math exam in subsequent years. During each interview students were asked their
PAGE 17
17 height, and weight. 2 Because much more information is available for students in 1997, I analyze the cross section in addition to a five year panel. Many of the independent variables explained below are used only in the cross sectional analysis because data are not available in following years. The outcomes that I use are self reported absences from sc hool in the fall of 1997 and the PIAT is a computer administered exam that has been found to reliably measure mathematic achievement. During the first interview, studen ts were asked to take the math exam if they were in or below the ninth grade; 6,046 students born 1980 1984 took the exam in 1997. However, in subsequent interviews, only students born in 1984 retook the test. The exam begins with a few questions of vary prior answers were wrong. Raw exam scores are reported on a scale of zero to o ne hundred. As progress over time and to be able to compare students across grades, I normalize the raw score with respect to grade. 3 That is, I take each stu Therefore, within grade level, the learning measure that I use has a mean of zero and standard deviatio n of one. 2 While adults tend to under report weight and over report height, no similar studies have been conducted concerning the reliability of adolescent self reported height and weight. 3 I also used raw scores as the dependent variable in all regressions, including grade dummies. I do not report these results here, as they were very similar.
PAGE 18
18 The independent variables of interest are measures of weight, height and body mass index (BMI). 4 A child is conside red overweight 5 by the Centers of Disease Control and Prevention (CDC) if she has a BMI ( ) greater than 9 5% of her peers. 6 She is considered at risk for becoming overweight if she is at or above the top 85 th percentile of BMI for age and underweight if it is at or below the bottom 5 th percentile. The CDC stature for age and weight for age growth charts wer e originally created in 1977 and were based on information gathered using the second National Health Examination Survey (NHES II, 1963 1965) and the second National Health and Nutrition Examination Survey (NHANES II, 1976 1980). In 2000, the CDC issued re vised the stature for age charts and introduced BMI for age charts. 7 The revised charts employ the same data as the 1977 charts, but use slightly different statistical methods. Since the CDC growth charts and definitions of overweight are based on a trim mer population of adolescents, currently fifteen percent of th 8 A fourteen year old and is at risk for becoming overweight if she 132 pounds. She is underweight if she weighs less than 89 pounds. Further examples of overweight, at risk for overweight, and underweight for both males and 4 I discard observations with improbable height and weight. I do not use observations when reported height is greater than 78 inches or less than 48 inches or when BMI is greater than 140 or less than 11. 5 To avoid stigma, the NCHS avoids characterizing any youth as obese. However many researchers classify youth as obese and overweight when they are in the 95 th and 85 th percentiles of BMI for age, respectively. Throughout this paper, I will 6 Separate growth charts are used for males and females. 7 The weight for stature chart was also revised in 2000, but the NCHS recommends using the BMI for age char t to assess weight in relation to stature for children ages two through 20. 8 See Philipson (2001) or Hedley et al (2004) for further information about recent trends in overweight and obesity.
PAGE 19
19 females are presented in Table 2 1 and Table 2 2 I use the national BMI for age charts to assign each student in my sample to one of the four weight categories. 9 ny times 10 These questions may be viewed as indicators of both school quality and student effort. Surprisingly, these two variables have a correlation coefficient of only 0.16, and I am able to include both in m y analysis. The other gauges of school quality are the student teacher ratio and school size, which NLSY interviewers obtained directly from the school. One parent from each household was interviewed in 1997. About ten percent of students surveyed ha biological mother was the respondent for 80% of completed parent interviews. The parent was asked to report her own height weight, and s bio height, weight, education level and whether he still lived at home. The interviewer also asked if the biological mother or another family member attended PTA meetings or volunteered at the ). In addition to height and weight, the NLSY parent interview provides biological mother education instead. S ummary statistics for the 1997 sample are presented in Table 2 3. 9 Stature for age and body mass index for age charts are avai lable at http://www.cc.gov/growthcharts. 10 I discard observations where a student reports having something stolen or being threatened over 90 times in a semester.
PAGE 20
20 Methods I first use ordinary least squares ( OLS ) to examine the effect of physical appearance on student absences, employing an education production function: (2 1) I consider two learning measures: score on the PIAT and self reported absences from school in the fall of 1997. The student inputs that I am most interested in are measures of height and weight. I use three different specificatio ns of height and weight. I look at BMI, height and weight in pounds, and finally I use three dummy variables underweight, at risk for overweight, and overweight. Throughout my study, I estimate separate regressions for white females, black females, white males, and black males since I expect personal appearance to impact these groups differently. 11 Averett and Korenman (1999) find that white women have a lower ideal BMI and report having more negative feelings about their bodies than black women. I expec t the coefficient on the weight variables to be bigger in magnitude for white females than black females. I also expect weight to impact females more severely than males. Adolescent females, especially those who are overweight, experience greater preoccu pation with their weight and express dissatisfaction with their bodies more often than males (Rosenblum and Lewis, 1999). I find that the same is true for the NLSY sample. Each year of the NLSY students were asked what they were currently doing about th eir presented in Figure 2 1. Approximately two thirds of the fe males in the sample want to lose 11 Cawley (2004) and Sabia (2007) also estimate separate regressions for different sex race g roups. Both authors reject the null hypothesis that the coefficients on BMI are equal across groups.
PAGE 21
21 weight, while only one third of males want to lose weight. Student answers also reveal that white females have a lower ideal weight than black females: a greater proportion of white females are trying to lose weight even t hough overweight is 50% more prevalent among the black females in my sample. All education production function regressions include student, school, and parental inputs. Parental inputs must include measures of both quality and quantity of time that pare nts contribute the parent quality. Quantity of time the parents contribute to learning is measured by indicators of the biological father living with the student a parent volunteering in the classroom, parent PTA membership, and a variable for the number of children under eighteen living in the household. School quality is measured by student teacher ratio, school size, number of times a student has had somethi ng stolen from school, and number of times he was threatened at school. The NLSY surveyors ask students how many times the student has had something stolen from school in the semester. Students indicating that they have had items stolen multiple times ar e likely attending a lower quality school and will therefore learn less, all else equal. The number of times a student was threatened is also an indicator of school quality and student learning. Threatened students would learn less since they would be pr eoccupied with their safety. The NLSY gathered the student teacher ratio and enrollment numbers directly from the school that the student attended in 1997. However, the student teacher ratio and school size are assigned only to a range. I deal with this by assigning the midpoint to each group. The two school variables are missing 12% of the time; I include flags to account for this. A complete list of independent variables and the expected signs of their OLS coefficients is presented in Ta ble 2 4
PAGE 22
22 As explained above, endogeniety of the weight variable is a concern. That is, test scores could cause weight change. I use an instrumental variables method to deal with simultaneity. A good instrument will eliminate any bias in the coefficient estimate and must satisfy two assumptions. The first assumption is that the instrument is correlated with the weight variable. The second assumption is that the instrument is not correlated with learning except through weight. I use parental overweight as my ins trument. This variable takes a value of one if either biological parent has a BMI greater than or equal to twenty five. The variable has a value of zero if n either parent is overweight. For the instrumental variables regressions, I drop observations whe re no biological parent BMI is recorded. Studies have shown that biological parent weight is strongly correlated with child weight; approximately half of the variation in weight between child and parent can be explained by genetic variation (Comuzzie and Allison 1998). Cawley (2004) uses biological sibling weight as an instrument for respondent weight. Sabia (2007) uses parental self reports of obesity as an instrument for student weight. bi ological father or mother This study improves on the work by Sabia in that I use a more objective measure of overweight and am able to calculate BMI overweight. One may be concerned that unobserved family characteristics such as motivation and work ethic cause both child and parent weight. However, adoption studies have shown that it is only genetics and not family environment that contributes to weight. 12 That is, the weight of adopted children is not correlated with the weight of their adoptive parents or siblings. Twin studies have 12 Stunkard et al 1986
PAGE 23
23 also supported these findings. Twins raised in different homes have body weight that is as highly correlated as twins who were raised together. 13 As a final test, I use panel data from 1997 and 1998 along with a fixed effects (FE) model to control for unobserved heterogeneity. The FE model controls for any unobserved individual level, time invariant characteristics. The downside of the FE model is that it uses individual changes in weight and test score to identify the impact of weight on learning and exacerbates any measurement errors present in the data There must be significant individual level variation in both weight and te st score to detect a statistically significant effect. Empirical Results Example coefficient estimates for control variables can be found in Table 2 5 In the remaining t ables I present the main findings of this paper. For the sake of space, I only pr esent coefficient estimates on the independent variables of interest. All control variables have the expected sign. 14 Effects of H eight and Weight on Test Score U sing OLS The results of OLS regressions with test score as the dependent variable are presen ted in Table 2 6 Three different regressions we re run for each group of students, using three different measures of weight and height. In Column 1, all three different measures produce statistically significant results for white females. A one standard deviation increase i n BMI is associated with a 0.011 standard deviation drop in test score. 15 The coefficient on pounds indicates that, for white females, a one standard deviation weight gain is associated with test scores 0.0 02 standard 13 Grilo and Pogue Geille 1991 14 One may hypothesize that the impact of weight will vary as the fraction of the population that is overweight varies. State level measures of adult overweight did not change regression results and were not statistically significant They were therefore excluded from final regressions. 15 The average white female in my sample has a BMI of 2 1.3 and raw test score of 73.61; this means that a one standard deviation increase in BMI ( to 25.3 ) is associated with a test score dr op to 73.43
PAGE 24
24 deviations lower. Also, this regression indicates that a one standard deviation increase in height is associated with a test score that is 0. 031 standard deviations higher. Finally, results show that white females who are overweight are expected to receive test scores 0. 279 standard deviations lower than their healthy weight peers. Using OLS, no statistically significant results were obtained for either black females (Column 2) or black males (Column 4). However, underweight white males receive test scores that are 0. 237 standard deviations lower than their healthy weight white counter parts. This females. Effects of H eight and Weight on Test Score U sing IV Results of instrumenta l variables (IV) regr essions are presented in Table 2 7 The sample size is slightly smaller than that used in OLS since observations must also have the height and weight of at least one biological parent. Coefficient estimates on height and weight varia bles for white females, black females, and white males are statistically significant and larger in magnitude than OLS estimates, indicating that IV has eliminated upward bias in the endogenous variable. In Column 1, results for white females are presente d. A one standard deviation increase in BMI is associated w ith a drop in test score of 0.057 standard deviations. A weight gain of 26 pounds (one standard deviation) is associated with a drop in test sore of 0. 010 standard deviations. Height is positive ly related to test score. A growth spurt of three inches (one standard deviation) is associated with a test score that is 0.06 5 standard deviations higher. Finally, overweight white females can expect to receive test scores 1.54 standard deviations lower than their healthy weight counterparts. Instrumental variable regressions indicate that weight is a predictor of student performance not only for white females, but for black females and white males as well. In fact, coefficient
PAGE 25
25 estimates are smaller f or white females than they are for black females and white males. Coefficient estimates in Column 2 indicate that a one standard deviation increase in BMI among black females is associated with a drop in test score of 0.74 standard deviations. A weight g ain of 33.7 pounds (one standard deviation) is associated with a drop in test score of nearly one standard deviation. An increase in height of three inches (one standard deviation) is associated with a test score 0.42 standard deviations higher. Results also indicate that overweight black females can expect to receive test scores 1.8 standard deviations lower than their healthy weight counterparts. Results for white males, presented in Column 3, are quite similar to results for black females. A one sta ndard deviation increase in BMI among white males is associated with a drop in test score of approximately one half standard deviation. A weight gain of 35.3 pounds (one standard deviation) is associated with a 0.78 decrease in test score and a one standa rd deviation increase in height (4 inches) is associated with a 0.46 increase in test score. Results indicate that overweight white males receive test scores that are 2.51 standard deviations lower than their healthy weight counter parts. Effects of Hei ght and Weight on Test Sore U sing FE Sample sizes used in the FE regressions are rather small since they require the student have test score and height/weight information in both 1997 and 1998. In order to obtain statistically significant results using a FE model, there must be enough individual level variation in both test score and weight. The great majority of students (80%) are a healthy weight in both 1997 and 1 998. Fixed effects regressions using the dummy variables, which are presented the last ro ws of Table 2 8 do not utilize the data to its full potential, identifying off of a very small number of observations. Because of the small identifying sample, coefficients on underweight, at risk for overweight, and overweight are not statistically signi ficant for any sex race group.
PAGE 26
26 However, when the variables BMI or pounds are used instead of dummy variables, I do obtain small but statistically significant coefficients for both white females and white males. The results presented in Column 1 of Tab le 2 8 indicate that, among white females, a BMI increase of one standard deviation is associated with a decrease in test score of 0.02 standard deviations. A weight gain of 26 pounds is associated with a test score decrease of 0.05 standard deviations. The coefficients on BMI and pounds are positive for white males, indicating that heavier white males perform better than their thinner counter parts. Results indicate that a one standard deviation increase in BMI is associated with a 0.12 standard deviat ion increase in test score and a weight gain of 35 pounds is associated with a 0.17 increase in test score. Effects of Height and Weight on Student Absences U sing OLS In addition to student test scores, being over or underweight may affect student absences A student who is over or underweight may try to avoid school because of teasing from her peers. While missing school is a concern in itself, the negative impact of student absences on overall academic performance is well documented (Bos, Ruijlers, Viss cher, 1992; Lamdin, 1996). To see whether student absences are affected by height or weight I use OLS and IV regressions with student absences as the dependent variable, presented in Tables 2 9 and 2 10 16 Using OLS, I find statistically significant an d similar results among white females, black females, and white males. For these groups, a one standard deviation increase in BMI or in pounds is associated with missing approximately an additional half day of school per semester. Results also indicate t hat black females who are at risk for becoming overweight miss one and one half more days of school than their healthy weight counterparts. Surprisingly, results 16 Unfortunately, student absences are only reported in 1997 and further analysis is not possible with this data set.
PAGE 27
27 indicate that overweight white males are absent less than their healthy weight peers. These results are economically significant since the median number of student absences is only three. Effects of Height and Weight on Student Absences U sing IV Other results indicate that instrumental variables regressions are not at all similar to OLS regres sions. I estimate no statistically significant impact among white females, black females, or black males. Among white males, heavier students miss more school. The estimated impact is quite large in magnitude and statistically significant. A one standa rd deviation increase in BMI is associated with missing 1.5 additional days of school. A weight gain of 35 pounds is associated with missing 1.6 additional days of school. Finally, overweight white males miss nearly five and one half additional days of s chool per semester. All these results suggest that heavier students are absent more frequently. These results are economically significant, considering the median number of days absent is only three. Conclusion The strongest evidence that heavier or shor ter children do not do as well in school as their a significantly negative impact on her test score in OLS, IV, and FE specifications. The hypothesis that talle r students perform better is supported in the OLS and IV regressions. For black females, only the IV regressions provide statistical support for the overweight and height hypotheses. For both female groups, there is evidence in the OLS regressions that h eavier students are absent from school more often. These results suggest that at least part of the fall in test scores due to being overweight is attributable to less time being spent in school. Body features have less impact on male test scores. None of the weight and height measures are significant in the regressions for black males. The evidence is less consistent for white males. In the OLS regressions, underweight boys have lower test scores. In the IV
PAGE 28
28 regressions, heavier and shorter males have l ower test scores. On the other hand, heavier boys are estimated in the FE regressions to have higher test scores. There is stronger evidence in the white male regressions than in the female regressions that heavier students miss more school. The additio nal support comes from two of the three OLS regressions in which some measure of Overweight students may be victims of bullying by peers, they may receive differential treatment from teachers, or they may experience a l ess favorable home environment. Undoubtedly, students who do not perform to their potential in school will face academic as well as labor market consequences. The American Obesity Association reports that overweight children have a 79% chance of being ov erweight adults. 17 If overweight individuals learn less as solely due to employer discrimination. 17 http://www.obesity.org/subs/fastfacts/obesity_youth.shtml
PAGE 29
29 Table 2 1. d efinitions of o verweight, a t r isk for o verw eight, and u nder weight b ased on m edian h eight for f emales Age Median h eight Overweight (>95%) R isk for o verweight (>85%) U nderweight (<5%) 12 129 lbs 111 lbs 76 lbs 13 145 124 84 14 161 137 93 15 169 144 98 16 175 149 102 17 179 151 103 18 184 154 104 Table 2 2. NCHS d efinitions of overweight, at r isk for overweight, and underweight based on median height for m ales Age Median h eight Overweight (>95%) R isk for o verweight (>85%) U nderweight (<5%) 12 121 lb s 105 lbs 74 lbs 13 137 119 84 14 156 136 96 15 173 151 107 16 186 164 115 17 193 170 121 18 200 177 126
PAGE 30
30 Table 2 3 Summary s tatistics for 1997 c ross s ection Females N=2,155 Males N=2,526 Mean Min Max Mean Min Max PIAT s core 71.61 1 .00 100 .00 71.84 1 .00 100 .00 PIAT n ormalized by g rade 0.01 4.26 2.22 0.04 4.31 2.22 Days absent from s chool 4.41 0 .00 88 .00 4.35 0 .00 75 .00 BMI 21.47 14.13 39.57 21.60 14.06 39.87 BMI p ercentile 58.54 0.20 99.54 60.76 0.01 99. 66 Overweight 0.0 9 0 .00 1 .00 0.117 0 .00 1 .00 At risk for o verweight 0.1 4 0 .00 1 .00 0.1 5 0 .00 1 .00 Underweight 0.0 3 0 .00 1 .00 0.03 0 .00 1 .00 Age 13.60 12 .00 17 .00 13.70 12 .00 17 .00 Grade 7.83 6 .00 9 .00 7.77 6 .00 9 .00 Black 0.28 0 .00 1 .00 0.30 0 00 1 .00 ducation 12.65 2 .00 20 .00 12.71 1 .00 20 .00 ducation 12.71 2 .00 20 .00 12.76 1 .00 20 .00 Live w/ biological f ather 0.53 0 .00 1 .00 0.54 0 .00 1 .00 Parent volunteers at s chool 0.52 0 .00 1 .00 0.50 0 .00 1 .00 P arent m ember of PTA 0.69 0 .00 1 .00 0.69 0 .00 1 .00 Household m embers <18 2.47 1 .00 9.00 2.47 1 .00 9 .00 Threats at s chool 0.59 0 .00 25 .00 0.86 0 .00 25 .00 Theft at s chool 0.39 0 .00 20 .00 0.54 0 .00 20 .00 Student teacher r atio 17.0 0 10 .00 26 .00 16.9 0 1 0 .00 26 .00 School s ize 845 .00 50 .00 1250 .00 829 .00 50 .00 1250 .00
PAGE 31
31 Table 2 4 Independent v e xpected i mpact on t est s cores and s chool a bsences Independent v ariables Test s core School a bsences Student Inputs BMI + Pounds + Height + O verweight + Risk for o verweight + Underweight + Parent Inputs ears of e ducation + ducation + Live with f ather + Parent v olunteers at s chool + Parent m ember of PTA + Number of household m embers <18 + School i nputs Threats at s chool + Theft at s chool + Student teacher r atio + Sch ool s ize +
PAGE 32
32 Table 2 5 Cross section regression results, effect of weight on test score ; dependent variable is grade normalized PIAT math test score White f emales Black f emales White m ales Black m ales Underweight 0. 270 0. 016 0. 237*** 0.210 [0.120 ] [0. 280 ] [0.120 ] [0.193 ] R isk for o verweight 0.002 0.133 0.059 0.11 3 [0.067] [0.110 ] [0.058] [0.112 ] Ov erweight 0. 279*** 0.126 0.063 0.134 [0.095 ] [0.110 ] [0.066] [0.110 ] Mom's years of e ducation 0.069 0.078 0.062 0.079 [0.009]*** [0.022]*** [0.009]*** [0.020]*** Dad's years of e ducation 0.04 0.024 0.067 0.002 [0.009]*** [0.024] [0.008]*** [0.021] Live w/ biological d ad 0.139 0.027 0.04 0.099 [0. 048]*** [0.091] [0.046] [0.086] Parent volunteers at s chool 0.204 0.061 0.087 0.107 [0.049]*** [0.085] [0.045]* [0.080] Parent PTA m ember 0.038 0.014 0.05 0.003 [0.049] [0.099] [0.047] [0.088] Household m embers <18 0.006 0.091 0.01 0.028 [0.019] [0.028]*** [0.019] [0.027] Student threatened at 0.025 0.008 0.004 0.006 S chool [0.013]* [0.017] [0.007] [0.017] Something stolen from 0.007 0.014 0.048 0.006 S chool [0.023] [0.029] [0.016]*** [0.020] Student teacher r atio 0.007 0.027 0.008 0.002 [0.005] [0.009]*** [0.004]* [0.008] School s ize <100 0.357 0 .000 0.106 0 .000 [0.355] [0.001 ] [0.354] [0.001 ] 100