Supplemental Online Appendices Air pollution around Schools Affects Student Health and Academic Performance BY Paul Mohai Byoung-suk Kweon Sangyun Lee Kerry Joy Ard Prepared for Health Affairs May 2011
Supplemental Online Appendices Table of Contents Data and Method 3 Regression Analysis of School Attendance Rates in a School 5 On-line Exhibit 1 6 Regression Analysis of Percent of Students in a School Not Meeting Michigan Educational Assessment Program (MEAP) Standards in English and Math 7 On-line Exhibit 2 9 On-line Exhibit 3 10 End Note 11 2
Data and Method RSEI-GM is modeled from emissions data in EPA s Toxic Release Inventory (TRI) and is used by EPA to compare relative risks to human health imposed by TRI facilities. 1 TRI data only include information for industrial facilities that are within the North American Industry Classification System (NAICS)and are of a minimum size and capacity (employing 10 or more fulltime equivalent employees and manufacturing, processing or otherwise using regulated chemicals above their designated thresholds). 2 The TRI does not contain information about pollution from other sources, including from small industrial establishments and automobiles. Beginning in 2006, the model used to create the RSEI-GM data was a Gaussian-plume fate-and-transfer model (the American Meteorological Society/EPA Regulatory Model). 3 Taking into account factors associated with the emitted chemicals (e.g., stack height, exit-gas velocities, and prevailing wind current), this model determines the transportation rates and fate of the chemicals into areas adjacent to the facility. Specifically, the quantities of the emitted chemicals (in micrograms per cubic meter) are estimated for every one-kilometer grid cell in a 101 x 101 kilometer square area centered at the facility. In each grid cell, each of the chemicals are furthermore weighted by 3
their toxicity to human health and the weighted quantities of all the chemicals are then summed to produce a total toxic concentration score for the cell. Toxicity weights are set by the EPA Science Advisory Board which aims to make all TRI chemicals (both carcinogenic and non-carcinogenic) comparable. 4 In our analyses, we used the 2006 RSEI-GM data (modeled from the emissions data reported in the 2006 TRI) as these were the most recent data available. MEAP scores are available from the Michigan Department of Education s website and are broken out by school, grade, and subject (reading, writing, math, science and social studies). Reading and writing scores are further combined into an English Language Arts score. The science test is taken only by 5 th and 7 th graders while the social studies test is taken only by 6 th and 9 th graders. Each subject score is categorized into four different performance levels (advanced, proficient, partially proficient and not proficient). The percentage of students in a school who did not meet the MEAP standard included the percentage of students in the school whose MEAP scores were either partially proficient or not proficient. These percentages were determined for English and math in each grade from 3 rd to 8 th. 4
Regression Analysis of School Attendance Rates in a School Online Exhibit 1 shows that attendance rates were lower for schools with greater concentrations of pollution around them. This relationship was not linear, so we sorted the schools into quintiles based on the total estimated air pollution concentration within two kilometers. Although attendance rates did not vary appreciably for schools in the first three quintiles, we found statistically significant decreases in these rates for schools in the fourth and fifth quintiles. This was true even after we controlled for confounding variables, such as the rural, suburban, or urban location of the school; average expenditure per student; size of the student body; studentteacher ratio; and percentage of students enrolled in the free lunch program. 5
On-line Exhibit 1: Regression Analysis of School Attendance Rates in a School N=3417 schools Model 1 Model 2 Model 3 Beta Sig Beta Sig Beta Sig 1 st Quintile -- -- -- -- -- -- 2 nd Quintile -0.021-0.002-0.012 3 rd Quintile -0.028-0.018-0.042 *** 4 th Quintile -0.099 *** -0.095 *** -0.080 *** 5 th Quintile -0.243 *** -0.191 *** -0.129 *** Rural -- -- -- -- Urban -0.153 *** -0.086 *** Suburbs 0.120 *** 0.066 ** Expenditure by pupil 0.042 ** #Students 0.026 Student/Teacher Ratio -0.113 *** %Free lunch students -0.249 *** R 2 0.049 *** 0.106 *** 0.160 *** * p<0.05; ** p<0.01; *** p<0.001 SOURCE: Authors analysis of geographic microdata for 2006 from Note 22 in text, and the attendance rate for 2007 and school demographic data for 2007 from Note 24 in text. NOTES: The dependent variable is the attendance rate. The number of schools was 3417. The quintiles are based on the total estimated air pollution concentration within two kilometers of the schools. Beta means beta coefficient. Sig means statistical significance. For an explanation of the models, see the text. R 2 is the percent (expressed as a decimal) of the variation in the dependent variable explained by the independent variables. 6
Regression Analysis of Percent of Students in a School Not Meeting Michigan Educational Assessment Program (MEAP) Standards in English and Math Online Exhibit 2 displays the results of this analysis for English. Model 1 includes only the dummy variables representing the five levels of air pollution concentration (with the first quintile as the reference category). As online Exhibit 2 shows, the patterns shown in Exhibit 4 were statistically significant. Schools in the fourth and fifth quintiles of total toxic air concentration were statistically significantly more likely to have higher percentages of students who failed to meet the state standards in English (Model 1). Controlling for school attendance reduced the size of the beta coefficients or the strength of the relationship between air pollution quintiles and percent of students in the schools failing to meet the MEAP standards in English for the fourth and fifth quintiles of total air pollution concentration (Model 2), but these remained statistically significant, suggesting that air pollution had a direct effect on student performance, in addition to indirect effects through reduced school attendance. Furthermore, the fourth and fifth quintiles of pollution remained statistically significant even after controlling for school locations, average expenditure per student, size of the student body, student- 7
teacher ratio, and percentage of students enrolled in the free lunch program (Models 3 and 4). Similarly, schools in the fourth and fifth quintiles of total air pollution concentration were statistically significantly more likely to have higher percentages of students failing to meet the state standards in math (Model 1, online Exhibit 3). Although the fourth quintile was no longer statistically significant (using a standard of p < 0.05) after controlling for attendance rates, school locations, expenditures, and other characteristics (Model 4), the fifth quintile was. 8
On-line Exhibit 2: Regression Analysis of Percent of Students in a School Not Meeting Michigan Educational Assessment Program (MEAP) Standards in English N=2633 schools Model 1 Model 2 Model 3 Model 4 Beta Sig Beta Sig Beta Sig Beta Sig 1st quintile -- -- -- -- 2nd quintile 0.034 0.015-0.036-0.013 3rd quintile -0.002-0.008-0.048 * -0.019 4th quintile 0.173 *** 0.128 *** 0.093 *** 0.042 * 5th quintile 0.385 *** 0.285 *** 0.213 *** 0.052 ** Attendance rates -0.338 *** -0.252 *** -0.097 *** Rural -- -- Urban 0.272 *** 0.100 *** Suburban -0.034 0.071 *** Expenditure per pupil -0.010 Number of students 0.068 *** Student-teacher ratio 0.030 * %Free lunch students 0.661 *** R 2 0.135 *** 0.241 *** 0.315 *** 0.575 *** * p<0.05; ** p<0.01; *** p<0.001 SOURCE: Authors analysis of geographic microdata for 2006 from Note 22 in text, and MEAP scores for 2007 and school demographic data from Note 24 in text. NOTES: The dependent variable is the percent of students (in third to eighth grades combined) not meeting the MEAP standards in English. The number of elementary and middle schools was 2633. The quintiles are based on the total estimated air pollution concentration within two kilometers of the schools. Beta means beta coefficient. Sig means statistical significance. For an explanation of the models, see the text. R 2 is the percent (expressed as a decimal) of the variation in the dependent variable explained by the independent variables. 9
On-line Exhibit 3: Regression Analysis of Percent of Students in a School Not Meeting Michigan Educational Assessment Program (MEAP) Standards in Math N=2633 schools Model 1 Model 2 Model 3 Model 4 Beta Sig Beta Sig Beta Sig Beta Sig 1st quintile -- -- -- -- 2nd quintile 0.025 0.007-0.039-0.026 3rd quintile -0.005-0.011-0.045-0.029 4th quintile 0.149 *** 0.106 *** 0.077 ** 0.022 5th quintile 0.361 *** 0.265 *** 0.201 *** 0.051 * Attendance rates -0.324 *** -0.242 *** -0.113 *** Rural -- -- Urban 0.253 *** 0.097 *** Suburban -0.043 0.025 Expenditure per pupil -0.004 Number of students 0.130 *** Student-teacher ratio 0.053 *** %Free lunch students 0.566 *** R 2 0.120 *** 0.218 *** 0.286 *** 0.478 *** * p<0.05; ** p<0.01; *** p<0.001 SOURCE: Authors analysis of geographic microdata for 2006 from Note 22 in text, and MEAP scores for 2007 and school demographic data from Note 24 in text. NOTES: The dependent variable is the percent of students (in third to eighth grades combined) not meeting the MEAP standards in math. The number of elementary and middle schools was 2633. The quintiles are based on the total estimated air pollution concentration within two kilometers of the schools. Beta means beta coefficient. Sig means statistical significance. For an explanation of the models, see the text. R 2 is the percent (expressed as a decimal) of the variation in the dependent variable explained by the independent variables. 10
End Note 1. For more information about these data, see Ash M, Fetter TR. Who Lives on the Wrong Side of the Environmental Tracks? Evidence from the EPA's Risk-Screening Environmental Indicators Model. Social Science Quarterly. 2004;85(2):441-62. 2. For more information, see http://tri.supportportal.com/ics/support/default.asp?deptid=2 3021 and http://www.dec.ny.gov/chemical/8434.html 3. Information about the modeling methodology can be found at: http://www.epa.gov/opptintr/rsei/pubs/index.html 4. Details about EPA s method of determining relative toxicities can be found at: http://www.epa.gov/opptintr/rsei/pubs/technical_appendixa_tox icity.pdf 11