Estimating the Proportion of Children Who Can Walk to School

Overview 

The purpose of the study was to estimate the proportion of Georgia children in early elementary school (grades K–2), late elementary school (grades 3 to 5), middle school (grades 6 to 8), and high school (grade 9 only) who lived within a reasonable and safe walking distance from their neighborhood public school. Grade-specific enrollment of public schools in the 1999–2000 school year provided by the Georgia Department of Education was used to estimate the denominator. The numerator was estimated by using demographic information collected in the 2000 U.S. Census. Information about the type and location of schools came from the Georgia Department of Education.

School Location 

The Georgia Department of Education provided addresses and contact information for all 1887 public schools. Using ArcGIS, version 9 (Earth Systems Research Institute, Redlands CA, 2006) and the Georgia TIGER/Line® 2004 file, 1170 (62%) schools were address matched. The location of the remaining 717 (38%) schools was verified by telephone calls to schools, Internet-based mapping sites, and aerial photo interpretation. During the manual verification of the positional accuracy of all 1887 schools, 257 (22%) of the 1170 originally matched schools were relocated because the original mapping was found to be >500 feet from the actual school location. After verification, the location of each school was set at the point of the school’s main entrance, usually where school buses enter and exit the premises.

Inclusion Criteria for Schools 

The focus of the project was on schools that were primarily neighborhood schools. Therefore, 128 schools whose names suggested that the school was not a neighborhood school such as special education, night, magnet, and vocational were excluded. An additional 13 schools with less than seven students per grade in the pertinent grades and six schools with unusual grade groupings, such as kindergarten only and grades 2 and 3 only, were excluded as well. The final analysis included 1060 schools with early elementary grades (K–2), 1052 schools with late elementary grades (3 to 5), 413 middle schools, and 284 high schools. Most elementary schools were included in assessments for both early elementary grades and late elementary grades because they included at least two grades in each grouping.

Walkable Areas or Pedestrian Catchment Areas 

Eight pedestrian catchment areas were drawn for each school using standard geographic information system (GIS) techniques. Two pedestrian catchment areas (PCAs) were circles with a 1-mile and a 0.5-mile radius from the school’s main entrance (radial PCAs). Two PCAs were irregular polygons representing the land areas from which students could reach school traveling along the existing street network for no more than 1-mile or 0.5-half mile (street network PCAs). Two PCAs were irregular polygons representing the land areas from which students could reach school using only streets with speed limits ≤35 mph and traveling no more than 1 mile or 0.5 mile (35-mph PCAs). The final two PCAs were irregular polygons including only streets with speed limits ≤25 mph and students traveling no more than 1 mile or 0.5 mile (25-mph PCAs). Posted speed limits were used as a proxy for safety because of the correlation between lower posted speed limits and reduced pedestrian fatality and injury rates.24

Potential Walkers 

The estimated number of potential walkers for each school, grade group, and PCA were developed using single-age population estimates from 2000 census-block data. Population estimates for block groups not fully contained within the PCA were calculated using area-weighted proportionate analysis. The fraction of the census block within the PCA was calculated and the population, assumed to be evenly distributed throughout the census block, was multiplied by that fraction. The population estimates from these partially contained census blocks were added to the population of the fully contained census blocks to obtain the population counts within each PCA.

Students in a specific grade are not all the same calendar age. To estimate the total number of children in a given PCA who would be in early elementary grades, the estimated number of children aged 5 to 7 was summed together. Similarly, to estimate the number of children in late elementary grades and middle school, the number of children aged 8 to 10 and 11 to 13 was summed together, respectively. The percentage of high school students who could walk to and from school was estimated using enrollment data for 9th grade only because only about 60% of Georgia ninth-grade students graduate from high school.25 The number of potential walkers in the ninth grade was estimated by summing the number of children aged 13 to 15 years in the PCA and dividing by 3.

Grade Enrollment 

School- and grade-specific enrollment data for the 1999–2000 school year were provided by the Georgia Department of Education. If a school included only two of the three pertinent grades (e.g., only grades 6 and 7), the number of potential walkers was adjusted accordingly (i.e., multiplied by two thirds).

Calculations 

For a given PCA and grade group, the estimated fraction of children who could walk to school was the number of children in a given age group (e.g., ages 5 to 7) living within the PCA divided by the number of children enrolled in the corresponding grades for that school (e.g., grades K through 2). For some PCA and school-grade groups, the number of potential walkers exceeded the number of enrollees. Most of these schools were elementary schools in such close proximity to another elementary school that they shared portions of their PCAs. When this occurred, the number of potential walkers was set to equal the enrollment for the corresponding grade group, making the estimated percentage of potential walkers 100% for that grade group at that school.

Potential walker estimates were made at the student and school levels. For students the overall percentage of potential walkers for each type of PCA was calculated by summing the potential walkers of a given age group and dividing by the sum of enrollees in the corresponding grade group for all schools in the analysis. In the school-level analysis, schools were categorized into those with ≥30% estimated potential walkers and those with <30%.

Regression Analysis 

To identify characteristics of schools with high percentages of potential walkers, a multivariate linear regression analysis was performed for each grade group. Independent variables in the analysis included school enrollment, demographic information, and measures of neighborhood connectivity. For school enrollment, enrollment for the grade group was used. Population density (total population per square mile), median household income, percentage of non-white residents (total population), percentage of housing built before 1950, and urbanization classification for each school developed by the National Center for Education Statistics (urban, suburban, town, and rural)26 were used for demographic variables. For each PCA, the area-weighting proportionate analysis technique was used for all variables except median household income and urbanization classification. The median household income was calculated by averaging the median household income of all block groups within or partially within the PCA. Street density (street length/PCA area in square miles), intersection density (three- to five-way intersections/street length in miles), ratio of three- to five-way intersections to three- to five-way intersections plus cul-de-sacs, and ratio of street network PCA area to radial PCA area were used for connectivity variables. The ratio of street network PCA area to radial PCA area represents the land area from which students could reach school traveling 1 mile along the existing street network divided by the area with a 1-mile radius (Figure 1). Neighborhoods with a high ratio of street network to radial PCA area tend to have more gridded street patterns and fewer cul-de-sacs.

View Large Image Download to PowerPoint

Figure 1. 1-mile radial and 1-mile street network pedestrian catchment areas for an elementary school in Georgia.

To better meet the assumptions of linear regression, all variables were log transformed except for metropolitan statistical area (MSA) classification, intersection density, and ratio of three- to five-way intersections to three- to five-way intersections plus cul-de-sacs. Independent variables were first tested in bivariate analysis to determine if they were associated with the dependent variable. Variables significantly associated with the dependent variable (p<0.05) were entered into the multivariate linear regression model. Backward stepwise regression was used to identify variables significantly associated with the dependent variable. Exploratory data analysis and regression modeling were performed using SAS, version 9.1 (SAS Institute Inc., Cary NC, 2006).

Assumed Preferences 

For this article, the assumed PCA preferences for students in grades K-5 require them to walk no farther than 0.5 mile along streets with speed limits ≤25 mph. For students in grades 6 to 9, the preferred PCA requires them to walk no farther than 1 mile along streets with speed limits ≤35 mph.

Student Estimates 

The estimated percentage of students in Georgia who, in 2000, could walk to school varied substantially by grade and type of pedestrian catchment area (PCA) ranging from 51% for early and late elementary school students using a PCA with a 1-mile radius and no other restrictions to 1% for high school students using a PCA with a 0.5-mile distance along streets with speed limits ≤25 mph (Table 1). Using the preferred PCA definitions of 0.5 mile along streets with speed limits ≤25 mph for elementary school students and 1 mile along streets with speed limits ≤35 mph for middle and high school students, the estimated percentage of students who could walk to school were 6% for early and late elementary school students, 11% for middle school students, and 6% for high school students.

Table 1. Percent of children in public schools who are potential walkers by grade and definition of potential walker, Georgia, 2000

	Definition of potential walker	Grade
		Grades K–2 (n=329,921)	Grades 3–5 (n=330,791)	Grades 6–8 (n=304,870)	Grade 9a (n=108,390)
	≤1 mile
	Radius from school	51	51	23	13
	Via street network	30	31	12	7
	Via street network using only streets with speed limit ≤35 mph	29	29	11b	6b
	Via street network using only streets with speed limit ≤25 mph	19	19	7	4
	≤mile
	Radius from school	19	19	7	4
	Via street network	9	9	3	2
	Via street network using only streets with speed limit ≤35 mph	8	9	3	2
	Via street network using only streets with speed limit ≤25 mph	6b	6b	2	1

a Analysis restricted to 9th-grade students due to the large number of dropouts from grades 10 through 12. b Preferred safety and distance parameters for grade group.

School Estimates 

The estimated proportion of schools to which ≥30% of enrolled students could walk also varied substantially by grade and type of PCA, ranging from 63% for early and late elementary school students using a PCA with a 1-mile radius and no other restrictions to 0% for high school students using all half-mile PCAs (Table 2). Using the preferred PCA definitions for safety and distance, ≥30% of enrolled students could walk to school at only 5% of schools with early elementary grades, 6% of schools with late elementary grades, 13% of middle schools, and 2% of high schools.

Table 2. Percent of public schools where ≥30% of students are potential walkers by school level and definition of potential walker, Georgia, 2000

	Definition of potential walker	School level
		Early elementary (n=1060)a	Late elementary (n=1052)a	Middle school (n=413)a	High school (n=284)a
	≤1 mile
	Radius from school	63	63	29	10
	Via street network	43	43	14	2
	Via street network using only streets with speed limit ≤35 mph	41	41	13b	2b
	Via street network using only streets with speed limit ≤25 mph	27	28	7	<1
	≤mile
	Radius from school	25	26	5	0
	Via street network	8	10	2	0
	Via street network using only streets with speed limit ≤35 mph	8	9	2	0
	Via street network using only streets with speed limit ≤25 mph	5b	6b	1	0

a Number of schools in the analysis. b Preferred safety and distance parameters for grade group.

Regression Analysis 

Schools with a higher proportion of walkers were characterized by high population density, low school enrollment, and high ratio of street network PCA to radial PCA area (Table 3). All ten demographic, school, and connectivity measures were associated with higher proportion of potential walkers for one or more grade groupings. However, the contribution of population density, school enrollment, and ratio of street network to radial PCA area as measured by partial R2 values was ≥10-fold larger than the contribution of the other variables to the elementary and middle school models. For high schools, school enrollment was not associated with higher proportion of potential walkers, and the contribution of population density and ratio of street network to radial PCA were three-fold or greater than the contribution of other variables in the model.

Table 3. Characteristics of neighborhoods associated with higher percentages of potential walkers within 1-mile street network pedestrian catchment area by grade group, Georgia, 2000

	Characteristic	Grades K–2		Grades 3–5		Grades 6–8		Grade 9a
		Beta	Partial R2	Beta	Partial R2	Beta	Partial R2	Beta	Partial R2
	VARIABLES MAKING LARGER CONTRIBUTIONS TO MODELb
	Population density	1.005	0.166	0.981	0.129	0.909	0.468	0.652	0.099
	School enrollment	−0.887	0.094	−0.775	0.066	−0.856	0.189	—
	Ratio of street network to radial PCA area	0.972	0.044	0.890	0.034	1.083	0.066	1.272	0.067
	VARIABLES MAKING SMALLER CONTRIBUTIONS TO MODEL
	Percentage of non-whites	0.056	0.002	0.056	0.002	0.072	0.003	0.115	0.007
	Percentage of housing built before 1950	−0.035	<0.001	−0.073	0.003	−0.051	0.002	—
	Street density	−0.152	<0.001	−0.179	0.001	—		—
	Median household income	−0.119	<0.001	−0.088	<0.001	—		—
	Ratio of intersections to intersections plus cul-de-sacs	−0.176	<0.001	—		—		—
	Intersection density	—		—		—		0.076	0.020
	Urbanization
	Rural schools	reference		reference		reference		reference
	Town schools	—		—		—		−0.099	0.004
	Suburban schools	—		−0.025	<0.001	—		−0.183	0.011
	Urban schools	−0.028	<0.001	−0.031	<0.001	—		−0.236	0.011
	Adjusted R2		0.933		0.941		0.940		0.795

Note: Characteristics are significantly associated (p<0.05) with percentage of potential walkers if values for beta and partial R2 are provided.

a Analysis restricted to 9th-grade students due to large number of dropouts from grades 10 to 12. b Contribution of variables was three-fold or greater than the contribution of other variables as measured by partial R2 values.

Assumptions 

Several of the analytic decisions made are likely to have influenced the accuracy of the estimates. First, although great care was taken to ensure that schools were correctly located on the map, only one entry into the school grounds was mapped for each school. Many schools have more than a single entry point, and this decision likely lowered the estimated proportion of students who could walk. The second assumption was that children must use local streets to get to school. Pedestrian-only pathways may enable some students who live outside of the calculated PCAs to walk to school without exceeding the 1-mile or 0.5-mile limit. The third assumption was that the student population was evenly distributed throughout the census blocks that were only partly included in the walkable network. This may cause an over-estimation of potential walkers for some schools and an under-estimate for others; this issue was assumed to have little effect on the overall statewide estimates. Fourth, posted speed limits were the only limiting hazard for streets along which students would be required to walk. Other types of hazards may be present, such as absent sidewalks or crossing high-volume streets, that may even make streets with acceptable speed limits hazardous and make the estimates too high. Fifth, schools that appeared not to be neighborhood schools were excluded from the analysis. Few students can walk a reasonable distance to these schools because their catchment areas extend beyond the immediate neighborhood. This would cause the statewide estimates to be too high. Sixth, students who attend private, parochial, or home schools—approximately 8% of school-aged children in Georgia27—and who live within the PCA of a public school would be counted as potential walkers. This, too, would cause the estimates to be too high. Seventh, students who attend a magnet or special focus public school or who choose to attend a public school in another neighborhood for race-ethnicity or school quality reasons would be counted as potential walkers, also causing the estimates to be too high. Eighth, for some schools (most often urban or suburban elementary schools) the estimate of potential walkers exceeded the number of enrolled students. Most instances of this impossible result occurred because of overlapping PCAs. For all such schools the number of potential walkers was reduced to the level of enrollees, thereby setting the percentage of potential walkers for that school and grade group at 100%. Few schools would be expected to have 100% potential walkers; this maneuver would have led to an over-estimate of the percentage of potential walkers. The ninth assumption was all students are physically capable of walking to school. This, too, probably caused a small over-estimate in the proportion of potential walkers. Of these potential biases, only the first two would cause the estimated percentages of potential walkers to be too low; the next one is expected to have negligible impact on the estimates; the remaining six would cause the estimated percentages to be too high. Finally, arbitrary preferences were set for distances and safety conditions for the preferred networks. Other options are available and these estimates have been provided.

Generalizability 

The methods used in this study should be widely applicable. However, the variability in demographic, educational, and housing characteristics suggest that the percentages of potential walkers may vary among geographic areas. Georgia’s population growth since 1950, a time during which automobile-oriented suburban development has been popular, has been relatively rapid.28 States that are more densely populated or have grown less rapidly over the past half-century may have larger percentages of potential walkers.