Assessing Aircraft Noise Conditions Affecting Student Learning, Volume 1: Final Report (2014)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

CHAPTER 5. IMPLEMENTATION OF THE RESEARCH PLAN 5.1. Introduction The research plan selected by the Panel was to conduct a nationwide macro-analysis to identify the relationship between aircraft noise exposure and student performance, taking into account the effect of school sound insulation and other confounding factors. The study was designed to examine this relationship for schools exposed to aircraft noise in the years 2000/01 to 2008/09 at the top 50 US airports, selected by the number of schools exposed to DNL 55 dB and higher. Student performance measures are based on the standardized reading and mathematics test scores in Grades 3 through 5 at each school. Schools were classified into three categories: target, control, and insulated. Target schools are those exposed to aircraft noise, defined as Day-Night Level (DNL) 55 dB or greater in the year 2000. Control schools are those located outside the DNL 55 contours in the year 2000 and are considered not exposed to aircraft noise. Insulated schools are those that have received modifications to increase the structural noise reduction. 5.2 School Database School location information was obtained from the US Department of Education (ED) Common Core Database (CCD) at http://nces.ed.gov/ccd, which provides names and street addresses for public elementary and secondary schools in the US. Latitude and longitude information were extracted from CCD to determine school locations. For cases where such data were missing in CCD, address information from CCD was used to identify the latitude and longitude of the school. The schools existing within the years of interest and in the states included in the study were extracted from the CCD database, and the following spatial data were added for subsequent analysis. • Population density around each school using Block Group data from the 2010 census; • Distance to major road from each school; • Identification of schools that are within the Year 2000 DNL 55 noise contour (target schools); • Elevation of target schools; • The airports near each school and their distance from each school. The scope of this macro statistical study did not allow for consideration of school type or age, nor windows open or closed. The master school database which contains all the school information (identified by unique number, lat/long, whether open or closed, by year, and name) contains a total of 59,722 entries for elementary schools serving Grades 3, 4, and/or 5. This does not mean that there are that many elementary schools in the US - there are many redundancies as schools open and close, change their name or addresses, or move. For example, the number in the year 2000 database is 41,035. Of these, 972 were within the Year 2000 DNL 55 contours of the selected 50 airports. A number of these target schools were insulated either before 2000 or in the period from 2000 to 2009. 5-1

5.3 School Noise Data The exterior noise levels at target (noise-exposed) schools were estimated using the Integrated Noise Model (INM) together with input files developed for US airports as part of FAA’s Model for Assessing Global Exposure to the Noise of Transport Aircraft (MAGENTA). The FAA uses the MAGENTA model to perform an annual assessment of total US airport noise exposure. INM studies appropriate for each year in the period 2000 to 2009 were obtained from the airports. The data source for airport traffic for the period 2000 to 2009 was the Enhanced Traffic Management System (ETMS) database provided by the FAA. The ETMS database includes scheduled and unscheduled air traffic, which allows for more accurate modeling of freight, general aviation, and military operations. The ETMS also provides details on aircraft type for an accurate distribution of aircraft fleet mix. Using the Year 2000 ETMS data, noise contours were calculated for the top 96 US airports and superimposed over school locations obtained from the US Department of Education Common Core Database (CCD) to determine the number of target schools within the DNL 55 noise contour. A total of 50 airports were then selected with the highest number of target schools. Of the selected 50 airports, three were excluded for lack of suitable INM studies in the years of interest 2000 to 2009 (Indianapolis, Milwaukee, and Sioux Falls). Omaha was excluded as the test scores in Nebraska were not reported in forms comparable from district to district. As a result, the analyses were based on 9 years of airport noise data at 46 of the nation’s largest airports in 26 states. The airports and states are listed in Table 5-1. TABLE 5-1 Airports and Schools in Corresponding States Included in the Study. Airport State Airport State Airport State Albuquerque (ABQ) NM Newark (EWR) NJ Chicago O'Hare (ORD) IL Anchorage (ANC) AK Fresno (FAT) CA Philadelphia (PHL) NJ Atlanta (ATL) GA Fort Lauderdale (FLL) FL Phoenix (PHX) AZ Boeing Field (BFI) WA Honolulu (HNL) HI Providence (PVD) RI Birmingham (BHM) AL Houston (HOU) TX Reno (RNO) NV Boston (BOS) MA New York (JFK) NY Rochester (ROC) NY Buffalo (BUF) NY Las Vegas (LAS) NV San Diego (SAN) CA Baltimore (BWI) MD Los Angeles (LAX) CA San Antonio (SAT) TX Cleveland (CLE) OH New York (LGA) NY Louisville (SDF) KY Charlotte (CLT) NC Orlando (MCO) FL Seattle (SEA) WA Columbus (CMH) OH Chicago (MDW) IL San Francisco (SFO) CA Cincinnati (CVG) KY Memphis (MEM) TN San Jose (SJC) CA Dallas Love Field (DAL) TX Miami (MIA) FL St. Louis (STL) MO Dallas Ft Worth (DFW) TX Minneapolis (SP) MN Tulsa (TUL) OK Detroit (DTW) MI Ontario (ONT) CA Tucson (TUS) AZ El Paso (ELP) TX Full-year ETMS operations data were collected for calendar years 2000 through 2009. The dataset was processed to yield operations corresponding to school days and school hours, and included operations occurring Monday through Friday, between the hours of 7am and 3pm. The data were filtered on a per-year basis, and split into two parts for each calendar year: January through May in any given year, and September through December. Then the data for January 5-2

through May were appended to the previous year's data for September through December, and combined into a file representing one school year. The naming convention selected is such that the year corresponding to the September through December portion is the year to which the school year designation refers. So for example, the September through December data for 2004, plus the January through May data for 2005, added together, corresponds to the full 2004 school year file. Following the assignment of INM studies to specific school years, and the assignment of ETMS data to those INM studies, the US MAGENTA model was exercised to estimate exterior noise levels at each target school. The following noise metrics were calculated (all levels are A- Weighted): • Arithmetic Average LAmax for school hours, including all aircraft events greater than 65 dB (Lmax65) and greater than 70 dB (LAmax70); • Energy Average SEL for school hours, including all aircraft events greater than 70 dB (SEL70) and greater than 80 dB (SEL80); • Average noise level, Leq, for school hours; • Time Above a threshold noise level L, TA(L), - TA(55), TA(60), TA(65), TA(70), TA(75), and TA(80) for school hours; • Numbers of Events Above a threshold level L, NA(L), - NA(55), NA(60), NA(65), NA(70), NA(75), and NA(80) for school hours. Following the execution of the US MAGENTA model to generate noise data at all 46 airports, the results were collected into a single master database which was then joined with the modified CCD database to allow for the subsequent school-level and system-wide analyses. The basic statistics of the aircraft noise file are shown in Table 5-2. The sample size (N) reflects the number of years each target school was open, and not insulated, multiplied by the number of open schools. TABLE 5-2. Statistics for Airport Noise Measures in Schools Exposed to Aircraft Noise. Measure Units Min Max Mean StdDev* N Nmiss LAmax65db decibels 65 98.5 70.1 2.5 7152 0 LAmax75db decibels 75 100.8 78.4 1.8 7121 31 SEL70db decibels 70 102.2 78.7 5.1 6976 176 SEL80db decibels 80 103.1 84.4 3.5 5765 1387 Leq decibels 30.9 86.3 54.2 5.2 7152 0 NA55 Count/8hrs 0.7 628.2 106.6 84.0 7152 0 NA60 Count/8hrs 0.2 435.5 72.1 62.9 7152 0 NA65 Count/8hrs 0 380.5 43.1 47.6 7152 0 NA70 Count/8hrs 0 314.9 20.4 32.8 7152 0 NA75 Count/8hrs 0 314.7 8.5 21.4 7152 0 NA80 Count/8hrs 0 291.8 2.9 13.3 7152 0 TA55 Minutes/8hrs 0.2 427.8 49.8 45.1 7152 0 TA60 Minutes/8hrs 0 288.9 23.8 25.4 7152 0 TA65 Minutes/8hrs 0 202.2 9.9 14.1 7152 0 TA70 Minutes/8hrs 0 142.7 3.4 7.5 7152 0 TA75 Minutes/8hrs 0 99.1 1.1 4.0 7152 0 TA80 Minutes/8hrs 0 64.9 0.4 2.1 7152 0 5-3

Note: Entries under N are number of school years, up to 9 each, for schools open and exposed to airport noise. Nmiss is the number of records for which the noise variable was not recorded. * Standard Deviation The distribution of the exterior equivalent noise levels, Leq, at the target schools averaged over the years 2000 to 2009 is as shown in Figure 5-1, with a mean of 54.2 dB and standard deviation of 5.2 dB. It must be remembered that these are average levels for the school day from 7 am to 3 pm, and not day-night average levels, DNL, with a nighttime weighting factor applied. Figure 5-1. Distribution of Aircraft Noise Levels at Target Schools The increased operations occurring in the morning hours are included, but not those of the late afternoon. In fact, for most of the school day, from 10am to 3pm, aircraft operations at all but the largest airports are at a fairly low level. Although the aircraft Leq values at target schools range from about 30 dB to over 80 dB (the low values being at airports with mostly nighttime operations), Figure 5-1 indicates the limited range of noise levels for most target schools – 81 percent are exposed to exterior Leq levels in the 15 dB range between 45 dB and 60 dB, 13 percent are exposed to levels greater than 60 dB, and only 3 percent to levels greater than 65 dB. The noise levels presented above are averaged values taken over a school year of aircraft operations; in other words, they are based on the total number of operations occurring within school day hours for the school year divided by the number of school days in the school year. At airports where the runway usage is fairly constant (Los Angeles International is a good example), the noise exposure is constant day to day and is well represented by the number of annual average operations. However, at airports where changes in wind direction require changes in runway usage the actual aircraft noise exposure levels may vary considerably day to day. On some days the aircraft noise levels will be higher than the calculated average, on other days they will be close to zero. The effect (if any) on test scores of such respites from aircraft noise cannot be determined as both test scores and aircraft noise levels used in this analysis are annual averages. 0 5 10 15 20 25 30 35 40 45 50 30 35 40 45 50 55 60 65 70 Pe rc en ta ge , % Leq, dB 5-4

In using the annual average noise levels we are making the assumption that aircraft noise has a negative effect on learning that results in lower test scores, and not that the aircraft noise interferes with the tests themselves, although this may also be true. 5.4 Ambient Noise The potential effects of aircraft noise on student learning may also be influenced by the level of external ambient noise at a school site from non-aircraft sources. An aircraft noise level of 65 dB might be more disturbing in a quiet suburban area than in a noisy urban environment. An estimate of the ambient noise at each school was obtained from information on the population density surrounding the school. Schomer (2010) has shown from empirical data that the ambient noise level in an area removed from major sources of noise (such as airports and major highways) can be expressed in terms of the Day-Night Average Level (DNL) by the expression: Ambient DNL = 10 log(ρ) + 19, dB where ρ is the population density in people per square mile. This formula is based on measurements at locations selected to obtain diversity in nearness to a major road or arterial street with a balance of near to arterials, near to lightly travelled roads, and in between. Measurement locations were more than 300 meters from a freeway, and not near an airport. The DNL noise metric is an average value of the equivalent noise level, Leq, taken over 24 hours, with a 10 dB weighting applied to noise during the nighttime hours (10pm to 7am). For application to schools, which operate only in the daytime hours, it was necessary to develop a relationship between DNL ambient and daytime noise ambient. This was achieved by using community noise data presented in EPA’s seminal report on community noise (EPA 1974) that shows daytime ambient levels are about 1.5 dB less than the day-night average ambient level, DNL. With this adjustment, we have Daytime Ambient Leq ≈ 10 log(ρ) + 17.5, dB The ambient noise levels presented in Schomer 2010 were measured in a wide range of areas with population densities as low as 118 and as high as 40,000 people per square mile. At the lower values, the number of data points is small, and hence the ambient DNL is less certain. For this study, a lower limit of 100 people per square mile was established, representing a minimum ambient Leq of 37.5 dB. The average ambient noise estimate for target schools included in this study is 56.5 dB, with a standard deviation of 5.4 dB. The average ambient level for all other schools is 48.8 dB, a difference of 8.1 dB. The average ambient level for other schools in the same school districts as the target schools is 55.3 dB, 1.2 dB less than that for the target schools themselves. Note that these differences do not represent the influence of aircraft noise, as this is specifically excluded from the empirical data of Schomer 2010. Target schools, and other schools in the same school districts, which are near airports, tend to be in more urban-like areas, whereas schools away from airports are in more suburban and rural areas with lower ambient noise levels. The distribution of noise level differences, i.e. the difference between Leq from the aircraft and the ambient Leq, at target schools is shown in Figure 5-2. It is noteworthy that the ambient Leq level exceeds the aircraft noise level at more than half of the schools. It must be 5-5

Figure 5-2. Distribution of Leq - Lamb at Target Schools. remembered, however, that the time histories for the two noise sources are quite different. Ambient noise, by definition, is a steady noise throughout the time period with only minor fluctuations. Aircraft noise, on the other hand, is characterized by discrete noise peaks corresponding to aircraft events, separated by periods of low noise. Individual jet aircraft noise levels are almost always greater than ambient levels even in the highest ambient conditions experienced in urban areas. Nevertheless, it is clear that target schools are not located in pristine areas, but are exposed to considerable noise from sources other than aircraft. One of those other sources is heavy vehicle traffic on nearby roads and highways, which is not included in the definition of ambient noise, and is excluded from the measured values of ambient noise as presented by Schomer (2010). Particularly in urban/suburban areas, the noise from individual trucks on local roads can approach the levels from aircraft flyovers. The influence of truck noise, if any, is not included in this analysis. 5.5. School Sound Insulation Database An important objective of the study was to determine whether test scores have been influenced by the implementation of sound insulation modifications. The information on whether or not schools had been sound insulated, and the year that the insulation was completed, was added to the master school database. The database on the total number of schools that have been sound insulated through the year 2010 has been developed from all of the inputs provided by the airports and/or consultants. The total number at the 46 selected airports is 388. Of these, 187 are either middle or high schools (for which test scores are unreliable, and are not being included in this study), or are parochial schools, many of which do not provide test score information. A further six were not in the CCD database, and 17 were outside the year 2000 DNL 55 noise contour, leaving a total of 173 target elementary schools. Of these, 145 were insulated before or during the year 2000 and thus were not available for analysis of before-and-after test score comparisons, although they were available for the insulated/non-insulated analysis. Thus, 51 elementary schools were 0 5 10 15 20 25 30 -25 -20 -15 -10 -5 0 5 10 15 20 25 Pe rc en ta ge Leq - Lamb, dB 5-6

insulated in the years between 2001 and 2009 as shown in Table 5-3 Of these, 29 were open for at least 2 years before and after being insulated, and had test scores available. TABLE 5-3 Numbers of Sound Insulated Elementary Schools at US Airports 5.6. School Test Score Databases The experimental design called for collecting test score data for the schools near the 46 largest commercial airports, which were located in 26 states. Most of the test scores for school years 1997-98 through 2004-05 and 2006-07 and 2008-09 were available from the National Longitudinal School-Level Test Score Database (NLSLSASD). Scores for school years 2005-06 and 2007-08 were extracted and transformed from four additional sources1. Longitude and latitude information is not available on CCD prior to the year 2000, and so test scores for years before 2000 were dropped from the database for the study. Also, test scores were not available for 2010, so the year identified as 2009 (the 2009-2010 school year) was not included in the study. As most standardized testing results in elementary schools during the decade of the study were in Grade 3, 4, and 5, the study focused on those three grades. Scores for Grade 6 were not included in the study because a large number of sixth grades are in middle schools or junior-high schools, which are not comparable to elementary schools. Grades at which tests were administered and results were available for this study are shown in Table 5-4. In a small number of states and years, there were no grade-by-grade test score data but there were aggregate cross- grade scores. These are indicated by an entry of “0” in Table 5-4. Because these are rare, they were not included in the analysis. 1 National Longitudinal School-Level State Assessment Score Database; GS: Great Schools; State standards mapping study, funded by USDE; Study of magnet schools, funded by USDE; Ohio Department of Education website. Airport Total ES Ins. Ins. After 2000 Airport Total ES Ins. Ins. After 2000 BOS 13 0 MDW 18 12 BUF 1 1 MHT 1 1 BWI 2 0 MSP 10 3 CLE 2 0 ONT 3 0 CLT 1 0 ORD 50 10 CVG 3 3 PVD 1 0 DAL 3 0 RNO 2 0 DTW 2 0 SAN 3 0 EWR 9 2 SAT 4 0 HNL 2 2 SEA 2 2 JFK 16 8 SFO 2 0 LGA 14 3 SJC 3 1 MCO 1 1 STL 5 2 Total 173 51 5-7

TABLE 5-4 Grades at Which Test Scores were Included for Each State in Each Year State Grades with Quantitative Tests Grades with Verbal Tests 2000 2001 2002 2003 2004 2005 2006 2007 2008 2000 2001 2002 2003 2004 2005 2006 2007 2008 AK 34..|345.|.4..|3...|345.|345.|345.|345.|.4.. 34..|345.|.4..|3...|345.|345.|345.|345.|.4.. AL 345.|34..|345.|.4..|.4..|345.|345.|345.|.4.. 345.|345.|345.|.4..|.4..|345.|345.|345.|.4.. AZ 345.|345.|345.|3.5.|3.5.|....|345.|345.|.4.. 345.|345.|345.|3.5.|3.5.|....|345.|345.|.4.. CA 345.|345.|345.|345.|345.|345.|345.|345.|.4.. 345.|345.|345.|345.|345.|345.|345.|345.|.4.. FL 345.|345.|345.|345.|345.|345.|345.|345.|.4.. 345.|345.|345.|345.|345.|345.|345.|345.|.4.. GA 345.|345.|.4..|345.|345.|....|345.|345.|.4.. 345.|345.|.4..|345.|345.|345.|345.|345.|.4.. HI ....|3.5.|3.5.|3.5.|345.|....|345.|345.|.4.. ....|3.5.|3.5.|3.5.|345.|....|345.|345.|.4.. IL 3.5.|3.5.|3.5.|3.5.|3.5.|345.|345.|345.|.4.. 3.5.|3.5.|3.5.|3.5.|3.5.|345.|345.|345.|.4.. KY ..5.|..5.|..5.|..5.|..5.|....|345.|....|.4.. .4..|.4..|.4..|.4..|.4..|....|345.|....|.4.. MA .4..|.4..|.4..|.4..|.4..|345.|345.|345.|.4.. 34..|34..|34..|34..|34..|345.|345.|345.|.4.. MD 3.5.|3.5.|3.5.|345.|345.|345.|345.|345.|.4.. 3.5.|3.5.|3.5.|345.|345.|345.|345.|345.|.4.. MI .4..|.4..|.4..|.4..|.4..|....|345.|....|.4.. .45.|.45.|.4..|.4..|.4..|....|345.|....|.4.. MN 3.5.|3.5.|3.5.|3.5.|3.5.|345.|345.|345.|.4.. 3.5.|3.5.|3.5.|3.5.|3.5.|345.|345.|345.|.4.. MO .4..|.4..|.4..|.4..|.4..|345.|345.|....|.4.. 3...|3...|3...|3...|3...|....|....|....|.4.. NC 345.|345.|345.|345.|345.|345.|345.|345.|.4.. 345.|345.|345.|345.|345.|345.|345.|345.|.4.. NJ ....|.4..|.4..|34..|34..|345.|345.|....|.4.. ....|.4..|.4..|34..|34..|345.|345.|....|.4.. NM ....|....|345.|.4..|345.|345.|345.|345.|.4.. ....|....|345.|.4..|345.|345.|345.|345.|.4.. NV .4..|3.5.|345.|345.|345.|345.|345.|345.|.4.. .4..|345.|345.|345.|345.|345.|345.|345.|.4.. NY .4..|.4..|.4..|.4..|.4..|345.|345.|345.|.4.. .4..|.4..|.4..|.4..|.4..|345.|....|345.|.4.. OH .4..|.4..|.4..|.4..|34..|...0|345.|...0|.4.. .4..|.4..|.4..|.4..|345.|...0|345.|...0|.4.. OK ..5.|....|..5.|....|345.|....|.4..|345.|.4.. ..5.|....|..5.|....|345.|....|.4..|345.|.4.. PA ..5.|..5.|..5.|..5.|3.5.|345.|345.|345.|.4.. ..5.|..5.|..5.|..5.|3.5.|345.|345.|345.|.4.. RI .4..|.4..|.4..|.4..|....|345.|345.|345.|.4.. .4..|.4..|.4..|.4..|....|345.|345.|345.|.4.. TN ....|345.|....|....|345.|....|345.|...0|.4.. ....|345.|....|....|345.|....|....|...0|.4.. TX 345.|345.|345.|345.|345.|....|345.|345.|.4.. 345.|345.|345.|345.|345.|....|345.|345.|.4.. WA 34..|34..|34..|34..|34..|345.|345.|345.|.4.. 34..|34..|34..|34..|34..|345.|345.|345.|.4.. To increase the potential analysis sample size of schools exposed to aircraft noise, schools serving only a subset of Grades 3, 4, or 5 were later added to the database. Anticipating the unavailability or unreliability of scores from very small schools, schools with fewer than 5 students in at least one of the Grades 3, 4, or 5 were omitted. Finally, to avoid clearly non- comparable schools in the analysis, only regular schools, schools not focused primarily on vocational, special, or alternative education, (type=1 on CCD) were included in the analyses. 5.7. Preparation of Test Score Data The first step in preparing test score data was to extract the school average test scores from different sources, identify test scores that could be considered as indicators of verbal (i.e., reading, language arts, or in a few cases, writing) and quantitative (i.e., mathematics) achievement, and remove duplicate records and other extraneous data. The tests themselves and the methods of scoring varied both between states and, in many states, across years. The second step was to create indexes from the standards-based scores. In most states, following the passage of No Child Left Behind in 2002, school-level scores have been reported as the percentages of students meeting each of several standards. For this study, the multiple measures for each test were combined to a single index, multiplying the percentage meeting the lowest standard by 1, the percentage meeting the next higher standard by 2, and so on. While these test scores cannot be compared across states, which set quite different achievement standards (McLaughlin et al. 2007; Bandeira de Mello, Blankenship, and McLaughlin, 2009), they can be compared across schools in the same state and year. The types of school-level test 5-8

scores included in the study in addition to these indexes are scale scores, raw scores, median percentile scores, and normal curve equivalent scores. As the tests are different in different states, and may vary with years, the only valid comparisons of test scores are between schools in the same state in the same year. Scores in a school in one state cannot be compared directly with scores of another school in another state, and the scores in a school in one year cannot be compared directly with scores in that school in another year, except in terms of a change in its rank among schools in its state. Because no comparisons of raw test scores are made across states or across years within states, no information is lost by rescaling (normalizing) the test scores in each year across all the schools in each state to have a common mean and standard deviation. The advantage of that rescaling is to give equal weight to test scores in all states in computing summary statistics. Therefore, the means of the test scores in each year in each state were rescaled to 50, and the standard deviations were rescaled to 10. Thus, a school with test scores at the average of schools in its state with the same demographics would have an adjusted score of 50, and a school with test scores one standard deviation above (or below) schools with the same demographics would have an adjusted score of 60 (or 40). This standard procedure (also employed in the RANCH study) provided a valid basis for combining the within-state comparisons across 26 states. The third step was to check to ensure the test scores have construct validity by correlating them with demographic and resource measures. (See Appendix E for additional details) Many states reported more than one verbal achievement measure for each grade in each year, such as reading comprehension, language arts, and writing. The fourth step was to select either a single best measure, or an average of measures, for each state and year to use as the primary reading score and for the primary mathematics score. The selection was based on the construct validity of the scores measured in the preceding step. 5.8. Accounting for Demographic Factors2 In assessing the possible relationship between aircraft noise exposure and student test scores, it is necessary to consider other contributing (or confounding) factors. Families that live near schools exposed to substantial airport noise are generally less affluent than families living near other schools in the same area, and such demographic factors have been shown to account for variation in school achievement, even without taking the noise levels into account. Therefore, demographic effects must be controlled to isolate direct effects of airport noise on achievement, To reduce the extent to which differences in test scores between target schools and control schools might be related to demographics and resources associated with those schools, the scores in each state and for each school year were adjusted to account for four measures of demographic and resource characteristics, namely: 1. The fraction of students eligible for the free and reduced price lunch program; 2. The fraction of the school’s enrollment of children who are members of minority groups (African American, Hispanic American, and Native American); 3. The pupil-teacher ratio, a measure of school resources; and 4. The average enrollment per grade in the school. 2 See Appendix F.1 for additional details. 5-9

Demographic data were available from the government’s Common Core of Data (CCD) for the large majority of the schools with test scores. As a pre-analysis step, occasional missing values for these factors were filled in using statistical methods documented in McLaughlin (2003) CCD Data File:Thirteen-Year Longitudinal Common Core of Data Non-Fiscal Survey Database. The demographic adjustment was based on empirical data on the relationships between demographic measures and achievement among schools not near major airports in each state, and hence not exposed to aircraft noise. The predicted test scores for grade 3, 4, and 5 reading and mathematics, based on demographics, were computed for each school and subtracted from the actual scores to produce the demographic adjustment. A separate demographic adjustment to school-level scores was implemented in each year in each state. There are, of course, a number of other factors why scores might differ between schools close to airports and other schools. For example, it is possible that more effective teachers tend to choose not to teach in schools near airports. It is also possible that other aspects of the neighborhood in which the school is located make it less supportive of student achievement than other schools in the district. However, measures of teacher effectiveness and other factors affecting the learning environment are not systematically available on a large sample of schools. The fact that the measures used in the study accounted for half of the variance in school-level scores suggests that a large portion, if not all, of the demographic and resource differences between schools close to airports and other schools in the same district are eliminated by the adjustment. 5.9. Selection of Data Analysis Procedure The data to be analyzed were adjusted school-level achievement scores related to the subjects of mathematics and reading in grades 3, 4, and 5. Individual student scores were not available, so the analyses made use of average scores of students in each grade and subject in each school for each year. Target and control schools were included for each of 26 states, and scores for any particular school were included for one or more of the 9 years of the study. The factors in the design are states, school districts, schools, years, subjects, and grades. In design terms, schools are “nested” within school districts, and school districts are “nested” within states. Years, subjects, and grades are “crossed” with schools within districts within states. Schools were either exposed to airport noise from a major airport (target schools) or not exposed to airport noise (control schools). The important properties of these data that determined the approach to the analyses are as follows. 5.9.1. Missing data The testing policies in the 26 states included different grades in different years, sometimes including only reading or only mathematics. As a result, large amounts of data were systematically, as well as randomly, missing. (Table 5-4 records the changing patterns of testing in different states.) Furthermore, some schools were open for only a subset of the years of the study. Procedures that either require a complete dataset or omit observations (schools in this case) where data is missing are unacceptable, as they introduce potentially serious bias in conclusions to be drawn from the analysis. Because more than half of the schools would be excluded if only schools with all six scores (reading and mathematics in 3 grades) were included, separate analysis was carried out 5-10

and presented for each of the six scores. The pattern of missing test scores precludes multivariate analyses of systematic relations among the six scores. 5.9.2. Non-Comparable Dependent Variables The student testing in different states and, in many cases, in different years in the same state, often employed different measures of mathematics and reading, of different types and on different scales. As a result, it is not possible to compare test results between states. An analysis procedure that attempts to compare results in different states is not appropriate for this study, due to this non-comparability of test scores. This means that it is not possible to develop a simple relationship between raw test scores and noise level that is representative across states. 5.9.3. Non-Constant Data Variances (Heteroscedasticity)3 The school test database contains widely varying numbers of students tested in different schools in different years. As a result, the random variation in the school-level achievement measures was very large for small schools and much smaller for large schools. Because the data were school-level scores, they did not satisfy the assumptions of many analytical methods, namely, that scores have roughly the same standard error. The only way to address this issue is to weight each score by the number of students included in the score. A school with ten times as many students tested would then count ten times as much in computing overall averages. Failing to employ such weights would substantially reduce the power of the study to identify significant effects. The selected analysis procedure must not require homoscedasticity (equal standard errors) and must allow for appropriately employing case weights. 5.9.4. School District Effects4 All public schools are located in school districts (“local education agencies”). Even removing demographic factors, achievement scores may be affected by factors associated with a school district and its local administration. Thus it was necessary to take into consideration the variation between school districts in a state. After verifying that achievement in school districts with schools near major airports differed significantly from other school districts in the same state, a decision was made to limit the sample of comparison, or control, schools to schools in the same district as a target school exposed to airport noise. With this limitation imposed to reduce district effects, the overall database for the analyses consisted of adjusted test scores in 6,198 schools in 104 school districts, and of these schools 917 (= 905 + 12) were classified as exposed to significant airport noise, i.e. target schools, as shown in Table 5-5. A few schools were included as target schools some years and comparison schools in other years, due to changes in either school location or airport operations. 3 See Appendix F.2 for additional details 4 See Appendix F.3 for additional details 5-11

TABLE 5-5 Total Number of Schools and Districts Included in the Test Score Database Schools in District Districts Exposed Schools Comparison Schools Both* At least 2 exposed and 2 comparison schools 104 905 5281 12 1 exposed and at least 2 comparison schools 44 46 641 1 Fewer than 2 comparison schools 129 214 69 15 Total 277 1165 5991 28 *Schools exposed to airport noise some years and not in others. Test scores were, on average, available for 5 or 6 of the 9 school years. The numbers of schools included in the main analysis for each grade and subject are shown in Table 5-6. TABLE 5-6 Number of Schools and Districts Included in the Main Analyses for Each Grade and Subject Test Score Districts Exposed Schools Comparison Schools Both* Reading Grade 3 92 695 4473 8 Reading Grade 4 97 851 4904 9 Reading Grade 5 89 670 4445 9 Math Grade 3 89 683 4451 8 Math Grade 4 99 857 4934 10 Math Grade 5 89 682 4466 8 These schools were in districts with at least two exposed and two comparison schools with the specified test scores. For Grade 3 Reading, for example, the comparisons were based on 703 (= 695 + 8) airport noise-exposed schools and 4481 (= 4473 + 8) comparison schools in 92 school districts. 5.9.5. Correlations Across Years5. In each grade, a school has different students each year, but other aspects remain constant from year to year. Thus, average test scores in a school, compared to other schools in the state, after taking demographic factors into account, might be correlated from year to year. i.e. knowing that a school’s scores were higher than average one year would tell you whether it was 5 See Appendix F.4 for additional details. 5-12

likely to be above average another year. If they are correlated, then the results obtained from any analysis method would underestimate the standard errors of the quantities in the model. Analyses verified that the scores were substantially correlated across years (See Appendix F.4). Therefore a standard procedure was employed to adjust the standard errors by computing a “design effect,” the ratio of the reported standard errors to the true values of those standard errors. 5.9.6. Analysis Procedure In summary, the details of the test score data available for the study were carefully considered, and steps were taken to ensure that the results were valid and replicable. It was necessary to accurately reflect the multi-level structure of the data, measures in different grades, subjects, and years at schools in different school districts, in the analysis. Specifically: • Due to patterns of missing test scores stemming from state testing policies, separate analyses were performed for the six combinations of grade and subject. • School average test scores were weighted to reflect the number of students tested, in lieu of availability of individual student data, to maximize the power of the analyses. • To deal with the non-comparability of state tests, between-state variance components were removed from the analysis by standardizing each state’s scores to the same mean and standard deviation. • Because the size of the comparison sample within the same districts as schools exposed to airport noise was more than adequate, there was no advantage to introducing school district as a separate variance component. All test score comparisons were between schools in the same school district. The analysis method selected, the SAS general linear model procedure (GLM), was dictated by the unusual properties of the dataset. Alternative procedures within the SAS system (CALIS, MIXED, NESTED. ANOVA) were explored to determine whether they were appropriate for this dataset, but none were capable of matching the properties of the dataset. The multi-level modeling approach employed in the RANCH study (Stansfeld 2005) is not appropriate for analyzing this dataset with its many missing observations, and, if used, could have introduced a potentially serious bias in the results. The GLM procedure in SAS handles missing data appropriately. 5.9.7. Analysis Design The basic analysis design was to compare average adjusted test scores in schools exposed to various levels of aircraft noise (target schools) to average adjusted test scores in control schools in the same district that were not exposed to aircraft noise. The analysis procedure was essentially a linear regression that combines mean adjusted test scores as the dependent variable, and aircraft noise values and an ambient noise estimate as the predictors, for each school and year, from 2000-01 to 2008-09, for which data were available. Separate comparisons were conducted for mathematics and reading in Grades 3, 4, and 5. Although the results were generally similar across the two subject areas and three grades, there were variations because different states conducted testing in different grades in different years. 5-13

Analyses were implemented using the GLM procedure (PROC GLM) in the Statistical Analysis System (SAS (r)), version 9.1.3. This procedure is a generalization of linear regression that combines quantitative and categorical predictors and computes parameter estimates that best fit a dataset to a model. The best fit is defined as the fit that minimizes the variance of deviations between predicted and observed values of a dependent variable. The model used was: <adjusted target school test score>ijy = < mean school district adjusted test score>iy + A × <aircraft noise measure>ijy + B × <ambient noise measure>ijy + <error>ijy for school j in district i in year y, where control schools are in the same district as the target schools. The deviation of a target school test score from the district average is thus: Deviationijy = <adjusted target school test score>ijy – <mean adjusted district test score>iy = A × <aircraft noise measure>ijy + B × <ambient noise measure>ijy + <error>ijy A mean score was estimated for each school district in each year, and the deviation of a target school’s adjusted test score from the district mean in each year was fit to a linear combination of an aircraft noise measure and an ambient noise estimate. Estimates for the quantities A and B were then obtained for best fit of the 26-state database to the model. Aircraft noise measures were set to zero for control schools. Each observation in the dataset was weighted by the reported or estimated number of students tested. The sample was limited to school districts with at least two schools exposed to aircraft noise and two schools not exposed to aircraft noise, a total of 917 aircraft noise-exposed schools and 5,293 comparison schools in 104 school districts in 26 states. Analogous results obtained by also including districts with a single aircraft noise-exposed school and multiple comparison schools were very similar. Parameter estimates for the quantities A and B were scaled as percentages of a test score standard deviation per a 10dB change in noise level. The standard deviation was computed for school mean test scores in each state and year, prior to demographic adjustment. The statistical significance of the aircraft noise and ambient noise effects (i.e., the likelihood that the effect is different from zero) were estimated by the ratio of the parameter estimates to their estimated standard errors. 5.9.8. Data Presentation Because test procedures are different from state to state, and can vary within states for different years, it I not possible to simply present test scores as a function of noise level on a national scale. The test scores for each state have to be normalized before being combined for a national estimate. Having done that, the effect of noise is described by how much the combined, normalized test scores differ from the average. 5-14

Effects in this study are reported in terms of differences in average test scores of schools with different levels of exposure to aircraft noise6. These differences, or effect sizes, are reported in fractions of standard deviations of the distributions of school test scores across the schools in the various included states and years. Unless otherwise noted, all effect sizes displayed in the report are statistically significantly different from zero (p<0.05). As a reasonable approximation, the distribution of test scores in a state roughly follow the normal bell curve. That means that a school with a score of 60, or one standard deviation above the average, would rank at the 84th percentile of schools in its state, or higher than five out of six schools in the state, while a school with a score of 40 would rank at the 16th percentile, or lower than five out of six schools in the state. Table 5-7 shows the relationship between percentile ranks and standard deviation units. TABLE 5-7 Relationship Between Percentile Ranks and Standard Deviation Units Percentile rank 10th 20th 30th 40th 50th 60th 70th 80th 90th Standard deviation from mean -1.282 -0.842 -0.524 -0.253 0 0.253 0.524 0.842 1.282 Rescaled test score 37.18 41.58 44.76 47.47 50.0 52.53 55.24 58.42 62.82 Thus, if the effect of exposure to a given level of airport noise is associated with a rescaled test score deficit of 0.524 standard deviations, it would be as if it changed a 70th percentile school into a 50th percentile school or a 50th percentile school into a 30th percentile school, a change of 20 percentile points. Note that the relationship shown in Table 6 is reasonably linear over the range from the 30th to 70th percentile, but deviates from linear at more extreme percentile levels. To provide context to understand the size of the airport noise effect, consider the analogous association between poverty and test scores. A 10 percent greater number of students at a school who are eligible for the federal free and reduced price lunch program (e.g., 30 percent instead of 20 percent) is, on average, associated with a 0.170 standard deviation deficit in achievement relative to other schools in the state, other things being equal. This 0.170 standard deviation deficit in achievement corresponds to about a 7 percentile change in state ranking, say from the 50th to 43rd percentile. 5.10. Results The study considered a total of 6198 schools at 46 airports, 917 of which were identified as being within the Year 2000 DNL 55 dB noise contour (target schools), and in school districts containing at least two comparison schools that were not exposed to aircraft noise. The detailed results of the analyses are presented in the tables of Appendix G in terms of fractional changes of a standard deviation for given differences in the value of the selected noise metrics. This is the standard way in which results are presented in educational research studies. To gain a better understanding of the size of the aircraft noise effect the results are summarized in Table 5-8 in terms of the percentile change in state ranking resulting from the effect of aircraft noise based on the relationship in Table 5-7. 6 The use of the term “effects” is in accord with the use of linear regression models. However, these “effects” are not necessarily causal. They merely indicate that there is a relation between variation on the predictor measures and variation on the dependent variable 5-15

TABLE 5-8 Summary of Estimated Noise Effects for Various Aircraft Noise Metrics Noise Metric Percentile Decrease in State Ranking No. of Target Schools Affected1 Airport Noise, Leq <1 All Ambient Noise, Lamb 3 All Total Noise, Ltot 3-4 All Average Maximum Level, LAmax ~1 All Average SEL ~12 All 4 80 8 25 5 22 5 71 10 dB increase 6 103 15 dB increase 9 30 (All Target Schools) 20 dB increase 12 10 10 dB increase 9 103 15 dB increase 14 30 20 dB increase 18 10 10 dB increase2 5 103 15 dB increase2 8 30 20 dB increase2 10 10 10 dB increase Noise Measure 10 dB increase 10 dB increase 10 dB increase 10 dB increase Incremental Noise, ΔL = Ltot -Lamb Incremental Noise, ΔL = Ltot -Lamb (Same Target Schools) 20 minutes above 70 dB (2.5 min/hour) 100 events greater than 70 dB (12/hour) Number of Daily Events Above a Noise Threshold Daily Time Above a Noise Threshold 50 events greater than 70 dB (6/hour) All Students Non-Disadvantaged Students Disadvantaged Students 25 minutes above 65 dB (3 min/hour) 1 Number of affected target schools exposed to the stated condition (noise measure) in 2008. 2 Effect size not statistically significantly different from zero. The percentile decrease in state ranking is listed for all the selected noise metrics. In addition, the final column in the table includes the number of target schools (included in this study) that would meet the conditions listed under the ‘Noise Measure’ column heading for aircraft operations existing in the year 2008. The conclusions that can be made from these results are as follows. The association between the effects of aircraft noise and student test scores is statistically significant7 but small when aircraft noise levels are described by the simple decibel noise metrics, such as exterior Leq, LAmax, and SEL, by themselves. In fact, the influence of ambient noise and total noise Ltot (aircraft plus ambient noise) has more effect on test scores than aircraft noise described by these metrics. The lack of a correlation with Leq alone is maybe not surprising as the ambient level exceeds the aircraft noise level at more than half of the target schools. The correlation is improved somewhat by considering total noise Ltot (airport plus ambient). The single-event metrics, LAmax and SEL, perform poorly perhaps because, even though they describe the individual aircraft noise levels, they contain no information on the frequency of aircraft events. 7 The magnitude of the effect is small, but the result would be expected to arise simply by chance only in rare circumstances. Hence the result provides enough evidence to reject the hypothesis of “no effect”. 5-16

Test scores show a much higher correlation with non-decibel airport noise metrics, such as Number of Events above a threshold, NA(L), and Time Above a threshold, TA(L). Most target schools are exposed to less than four events per hour greater than 70 dB during the school day, and only about 5 percent experience more than ten aircraft events per hour greater than 70 dB. The NA metric fails to provide information on the levels of the individual events, but TA does provide information on the length of time that a disturbance may occur. The influence of airport noise on test scores becomes more apparent when the levels are related to ambient noise levels, such as with the incremental noise level Ltot – Lamb, and the effects are relatively insensitive to ambient levels greater than 50 dB (at least as far as this statistical analysis allows). This result may not be surprising since, even though the individual maximum aircraft levels exceed the ambient, a certain amount of masking does occur. Schools exposed to airport noise are near airports, and many airports are in metropolitan areas with high ambient noise levels. It should be noted that since control schools are in the same school districts as the target schools, their ambient levels will be similar. Rather than define an absolute level above which aircraft noise can affect test scores, it may be more appropriate to base the onset of effects on the level of ambient noise. If the aircraft Leq is 5 dB greater than the ambient Lamb, then the incremental level Ltot – Lamb is 6 dB, and the percentile decrease in state ranking is about 3 percent and statistically significant. In Appendix G.3 it is estimated that this relationship is valid for ambient levels greater than 50 dB. As a result, a first cut at defining a threshold for the effect of aircraft noise on test scores would be when the aircraft noise is 5 dB greater than the ambient noise for ambient levels greater than 50 dB. When the analysis is repeated to evaluate the test scores as the aggregate statistic for subgroups of students in the same school (under exactly the same ambient noise and school conditions), it was found that the estimated effect of aircraft noise increment for non- disadvantaged students is almost twice as great as for disadvantaged students. A 10 decibel increment over ambient noise was associated with a 9 percentile change in state ranking for non- disadvantaged students, but for disadvantaged students the estimated association was much smaller and not statistically significantly different from zero. The mean adjusted scores of disadvantaged and non-disadvantaged students in schools exposed and not exposed to airport noise are shown in Table 5-9 from the same schools comparison analysis. Note that the scores for all grades and subjects drop with exposure to noise, and that the mean differences are all larger for non-disadvantaged students than for disadvantaged students. Although the statistical analysis process may identify that a relationship exists, it does not necessarily provide a rationale for that relationship. At this time, any hypotheses we might offer to explain the differences would be conjecture. TABLE 5-9 Mean Adjusted Test Scores for Subgroups of Students Disadvantaged Students Non-Disadvantaged Students Not Exposed Exposed Delta Not Exposed Exposed Delta Reading Grade 3 50.53 50.47 -0.06 50.08 49.39 -1.36 Reading Grade 4 50.44 49.83 -0.61 50.10 48.74 -0.68 Reading Grade 5 50.59 50.09 -0.50 50.16 49.48 -0.65 Math Grade 3 50.52 50.05 -0.47 50.04 49.39 -1.70 Math Grade 4 50.39 49.29 -1.10 50.02 48.32 -1.28 Math Grade 5 50.70 50.20 -0.50 50.24 48.96 -1.36 5-17

Finally, estimates were made of the effect of sound insulation on student test scores. The obvious method of comparing scores immediately before and after sound insulation was introduced was hampered by the naturally occurring variation in scores from year to year, and a sample size limited to 29 elementary schools that were insulated within the study years for which test scores were available. However, a comparison of the airport noise increment effects in 119 schools that were insulated (many before the period of the study) versus those that were not insulated indicated that the apparent effects of airport noise increments found for non-insulated schools disappeared when measured for insulated schools. In other words, the effect of insulation was essentially to cancel out the effect of the aircraft noise exposure. The mean adjusted scores, by school category, are shown in Table 5-10 showing that the scores in exposed schools after sound insulation is introduced approach and sometimes exceed those in non-exposed schools. The numbers in this table are averages over all levels of aircraft noise exposure. Schools with higher exposure may show larger improvement than those with lower exposure. TABLE 5-10 The Effect of Sound Insulation on Reading and Math Scores (119 Schools) School Category Reading Math Grade 3 Grade 4 Grade 5 Grade 3 Grade 4 Grade 5 Not exposed 50.0 50.0 50.1 50.0 50.0 50.1 Exposed and not insulated 49.6 49.0 49.7 49.6 49.0 49.7 Exposed and insulated 49.9 49.8 50.5 50.6 50.3 51.3 5.11. Grade Equivalent Scores In Europe, researchers have used the concept of grade equivalent scores and “months delay in reading skills” to describe the effects of noise on test scores. For example, when the national data relating to the reading comprehension tests were used, an 8 percent decrease in ranking (equivalent to a decrement of one-fifth of a standard deviation) was equivalent to an 8- month difference in reading age in the United Kingdom, and a 4-month difference in reading age in the Netherlands (Stansfeld et al 2005). The problem in applying this concept to the US is that one-fifth of a standard deviation might be 3 months in one state and 6 months in another, depending on the distribution of test scores in each state. The RANCH study results (Stansfeld et al 2005), for example, shows different relationships in the different countries. Percentile ranks of schools within states mean the same thing in all states. 5.12. Results Related to Study Objectives The questions to be answered in this study were listed in Section 3,1 of the report. The results outlined above provide the following answers: 6. To what extent is student learning affected by aircraft noise? Estimated 6 and 12 percentile decrements in state ranking for increases of 10 and 20 dB respectively in incremental aircraft noise level. 7. What is the most appropriate noise metric for describing aircraft noise as it affects learning? 5-18

The metric that shows the largest effect is the incremental level Ltot - Lamb for ambient levels greater than 50 dB. The Number of Events Above 70 dB, NA(70), and Time Above 70 dB, TA(70), metrics are also indicators of the effects of aircraft noise. 8. What is the threshold above which the effect is observable? An approximate threshold is when the aircraft noise Leq exceeds the ambient by 5 dB, based on a criterion of a decrement of 3 percentile points in state ranking. 9. Has insulation meeting existing classroom acoustic criteria improved student achievement? Insulated schools have better test scores than those with no insulation which may be an indication that insulation could contribute to improved scores by returning test scores to what they would have been with no aircraft noise. 10. How does aircraft noise affect learning for students with different characteristics? The study showed that the effect of noise was greater for non-disadvantaged students than for disadvantaged students, although the analysis process does not make it possible to provide a rationale for this result. This issue may be addressed in the upcoming ACRP Project 02-47 Assessing Aircraft Noise Conditions Affecting Student Learning - Case Studies 5.13. Cautionary Notes A note of caution is necessary in interpreting these results. Test scores were available for a large sample of schools (917 target and 5,281 control schools in 104 school districts), and data on airport operations were available for 46 of the largest airports in 26 states, over a period of 9 years, providing some reassurance that the effect of random errors was reduced. However, the results might well be distorted by three types of imperfections in the data. First, the adjustment for demographic and resource factors was limited to four measures. Schools exposed to airport noise may have differed in other ways from other schools in their districts (besides the noise factor itself) that were not captured by the measures used for demographic adjustments. Second, the government agency that provided the latitude and longitude information on the location of schools could not provide documentation of quality control on those data, and there was some evidence that some school locations may have been imperfect. Latitude and longitude data were checked for all of the 173 insulated elementary schools and were found to be reasonably true, and certainly not sufficiently inaccurate to account for more than one decibel in noise level. Third, the estimates of ambient noise were based on a single measure, population density, and although there is empirical data to support the relationship between population density and ambient noise levels, that estimate is subject to error. The noise from local heavy trucks on arterials is included in the definition of ambient noise. Noise from heavy trucks on freeways is not included in the estimate of ambient noise; neither is it included in estimating the effect of noise on test scores at target schools. This may have an effect on the results for schools close to freeways with heavy truck traffic. It should be noted that freeway noise, and specifically 5-19

individual truck noise, is localized and attenuates rapidly with distance due to shielding from buildings, unlike the noise from elevated aircraft. 5.14. Comparison of Results with Previous Studies The current study found statistically significant associations between airport noise and student mathematics and reading test scores, after taking demographic and school factors into account. Associations were also observed for ambient noise and total noise on student mathematics and reading test scores, suggesting that noise levels per se, as well as from aircraft, might play a role in student achievement. This study further adds to the increasing evidence base which suggests that children exposed to chronic aircraft noise at school have poorer reading ability and school performance on achievement tests, than children who are not exposed to aircraft noise (Clark, et al., 2006; Haines, Stansfeld, Brentnall, et al., 2001; Haines, Stansfeld, Job, Berglund, and Head, 2001a, 2001b; Hygge, Evans, and Bullinger, 2002; Stansfeld, et al., 2005). Overall, evidence for the effects of noise on children’s cognition is strengthening and there is increasing synthesis between epidemiological studies, with over twenty studies having shown detrimental effects of aircraft and road noise on children’s reading (Evans and Hygge, 2007). The current study has undertaken a series of analyses of separate datasets, together encompassing an extremely large sample of schools (917 target and 5281 comparison schools from 104 school districts), and presents convincing evidence from an extensive set of analyses that both airport noise and ambient noise levels influence children’s learning. In comparison, the largest comparable primary study to date, the RANCH study (Clark, et al., 2006; Stansfeld, et al., 2005) examined over 2000 children from a total of 89 schools in the Netherlands, Spain and the United Kingdom. While the current study does not take into account individual differences in reading and mathematical ability and individual level socioeconomic factors, the findings are striking, and if further individual level data were available, larger effects may be observed. The current study adds specifically to knowledge about aircraft noise effects within the US, which had not been studied in recent years. A strength of the study is that data from 26 states has been examined. A previous study of aircraft noise exposure and school and data from all elementary schools in Brooklyn and Queens, New York from 1972 to 1976 found that an additional 3.6% (95% CI 1.5-5.8%) of the students in the noisiest schools read at least 1 year below grade level. It is hard to compare the findings between these studies, as the current study has had to assess percentile change in state ranking of the school for reading and mathematics. However, both studies indicate a detrimental effect of aircraft noise on learning, despite being conducted over 30 years apart. The findings of the current study are more consistent in their conclusion than those of the FICAN pilot study (Eagan, Anderson, Nicholas, Horonjeff, and Tivnan, 2004; FICAN, 2007), which focused on whether abrupt aircraft noise reduction within classrooms, caused either by airport closure or newly implemented sound insulation was associated with improvements in test scores, in 35 public schools in Illinois and Texas. The pilot study found some evidence for effects of aircraft noise reduction and improved test results, but also, due to its small size, found some associations that were null and some that were in the opposite direction to that hypothesized. The current study is one of the first studies to quantify the potential impact of sound insulation on children’s learning achievement for aircraft noise exposure. The study found evidence from a sample of 119 elementary schools that the effect of aircraft noise on children’s 5-20

learning disappeared once the school had had sound insulation installed. The issue of the effectiveness of sound insulation in reducing learning deficits associated with aircraft noise was identified as a priority in the literature review undertaken for this project. Little prior research has tested whether sound insulation of classrooms lessens the effects of aircraft noise on children’s learning. One study by Bronzaft (Bronzaft, 1981) examined the effectiveness of track improvements and the installation of sound absorbing ceilings in 3 classrooms at school. Before insulation, school children in classrooms on the noisier side of the school had poorer reading achievement scores than children on the quieter side of the school. After the intervention, which resulted in a total noise reduction of 6-8 dB in train noise, there were no differences in reading achievement between children on the noisy side and those on the quiet side of the school. To date, only a few studies have examined associations between noise exposure and mathematical ability (Haines, Stansfeld, Head, and Job, 2002; Ljung, Sorqvist, and Hygge, 2009; Shield and Dockrell, 2008). Only one previous study has examined associations specifically for aircraft noise exposure, finding that the initial association between aircraft noise and mathematics was explained by socioeconomic factors (Haines, et al., 2002). The current study strongly suggests that aircraft noise exposure at school influences mathematical achievement and suggests that mathematics should be examined in further studies. Effects on achievement tests were found for the following aircraft noise metrics: Leq, Ltot, Lmax, SEL. Test scores also showed associations with non-decibel airport noise metrics such as Number of Events above a threshold NA(L) and time above a threshold TA(L). Associations were observed for ambient noise level at the school (Lamb) and total noise (Ltot). A few previous studies have examined effects of different noise metrics on children’s learning outcomes, but the current study is among the first studies to examine aircraft noise. Shield and Dockrell (Shield and Dockrell, 2008) found associations between external noise (predominantly road traffic noise) and poorer national test scores for children aged 7 and 11 years attending London primary schools, and also found an effect for LAmax. These studies, considered together, suggest increasing evidence that maximum noise levels are important for noise effects on children’s learning. The estimated effect of aircraft noise was found to be twice as great for non- disadvantaged students as for disadvantaged students (disadvantaged students were those whose family, social or economic circumstances hinder their ability to learn at school). The West London School study around Heathrow airport in London did not find any differences in the effect of aircraft noise on children’s reading when examining disadvantaged subgroups of children: English and Non-English as the main language spoken at home, children in employed and unemployed households, children in deprived and not deprived households (Haines, Stansfeld, Brentnall, et al., 2001). The findings of the current study need replication in further studies, preferably conducted in different countries, before conclusions about the influence of disadvantage on the association between aircraft noise and children’s learning can be drawn. 5-21