Data and Methodology

Data Sources for Primary Analysis

To create the school district border dataset used in this analysis and mapping tool, we obtained data from the following sources:

• School district geography: School district boundary data for the 2020–21 school year comes from the Education Demographic and Geographic Estimates Program (EDGE), Composite School District Boundaries File.
• School district revenues: Revenues from federal, state, and local sources for the 2021 fiscal year come from the U.S. Census Bureau, Annual Survey of School System Finances, also known to researchers by the survey number F33. The following adjustments were made to the revenues for some or all school districts:

1. We exclude revenue for capital outlay and debt service programs from state revenues, because it can contribute to large fluctuations in district revenues from year to year. Similarly, we exclude money generated from the sale of property from local revenues.
2. We subtract from state and local revenues the total amount of money sent to separate charter local education agencies (LEAs)—an expenditure category included in the F33 survey—divided proportionally across the local, state and federal revenue categories based on the percent of each district’s revenues that come from these sources. This adjustment addresses the fact that, in just under 2,000 districts, revenues received by local school districts include funds that are transferred to charter schools that are operated by charter LEAs. This artificially inflates the per-pupil revenues in these school districts, because this pass-through charter funding is included in the district’s revenue, but the students educated by these charter schools are not counted in enrollment totals.
3. In the state of Arkansas, some revenues that are collected locally are categorized as state revenues. Before analysis, we subtracted the values of these collections from state revenues and added them to local revenues, The misattribution of revenue for each district is described in the F33 documentation as C24, Census state, NCES local revenue.

See the Annual Survey of School System Finances: Public Elementary-Secondary Education Finance Data Technical Documentation (2021) for state-specific notes in relation to education finance data.

• School district enrollments and racial composition: School district enrollment characteristics for the 2020–21 school year come from the U.S. Department of Education, National Center for Education Statistics, Common Core of Data (CCD). When data for 2020–21 are not available, we use the previous year’s data. In conjunction with the above finance data, these enrollment statistics are used to calculate per-pupil revenue figures.
• School district school-age poverty rates: School district-level data on poverty rates among school-age children in 2021 come from the U.S. Census Bureau’s Small Area Income and Poverty Estimates (SAIPE).
• School district community indicators: School district-level economic, demographic, housing and social indicators, including median household income and median value of owner-occupied homes from the U.S. Census Bureau’s American Community Survey 5-year estimates (2017–2021). The data are provided by Education Demographic and Geographic Estimates (EDGE), drawing upon data from the U.S. Department of Education, National Center for Education Statistics, and U.S. Census Bureau.
• Native American reservations and other Native lands: The boundary data for Native lands come from the American Indian Areas/Alaska Native Areas/Hawaiian Home Lands Boundary File from the Census Bureau’s MAF/TIGER geographic database. This includes shapefiles for all federally and state-recognized American Indian reservations, off-reservation trust lands, Tribal-designated statistical areas, state-designated Tribal statistical areas, Alaska Native village statistical areas, Oklahoma Tribal statistical areas, and Hawaiian Home Lands.

Methodology for Primary Analysis

We conducted a spatial analysis of all unified, secondary, and elementary districts in the United States. This process identified all pairs of school district neighbors that share a border. Only districts that share land borders and borders along linear bodies of water were considered to be neighbors. Districts whose shared borders exist entirely along wider bodies of water, such as lakes, were not considered to be neighbors. Pairs were excluded from this neighbor list if their shared boundary was less than 500 feet long or if the two districts were located in different states.

Each neighbor pair was identified by their shared school district border and joined to the data from the SAIPE, CCD, and ACS described above. To determine the degree of economic segregation between the districts separated by each border, we calculated the difference in their school-age poverty rates. To determine the degree of racial segregation between the districts separated by each border, we aggregated the enrollment percentages for all racial groups other than non-Hispanic white into a new category representing each district’s percentage of students of color enrolled and took the difference in these enrollment percentages. After making the exclusions outlined below, we ranked each border in our dataset by the degree of both racial and economic segregation it enforces. Similarly, we computed the dollar amount difference in local, state, and the combined value of local and state revenues between districts; this information was not used for ranking but is provided for context.

School District Exclusions

We employed several exclusion criteria in compiling our borders dataset. Our analysis includes only districts that meet our standard requirements for a geography-based analysis. Therefore, any district that does not have a defined geographic area and is not included in the Composite School District Boundaries File was excluded. We also excluded districts from the U.S. territories. Further, because we only identify within-state school district neighbors, Hawaii and the District of Columbia were excluded from the neighbor-pair analysis, as they each have only one school district.

There are three types of school districts: unified, elementary, and secondary. Our analysis was confined to certain categories of district pairs in order to avoid comparing resources across districts of different types. These pairings include unified to unified, unified to secondary, secondary to secondary, and elementary to elementary.

We additionally excluded school districts where the student population is at least 75 percent Native, or where more than 75 percent of the area overlaps with American Indian reservation land. For the purposes of this report, American Indian reservations are not considered to include off-reservation trust lands or Tribal statistical areas, as neither are considered to be sovereign administrative units. Trust lands are administered by the federal government, and often have relatively small Native populations. State and federal designated Tribal statistical areas, including Oklahoma Tribal statistical areas and Alaska Native village statistical areas, encompass areas with significant Native populations, but whose Tribal majority do not have a reservation or trust lands. School districts where 75 percent of the area overlaps with any other kind of reservation land are excluded from this analysis. For more information on the reason for this exclusion, please see the section “Divided Districts and Native Students” above.

Since the school-age poverty rates are estimates, they are not always reliable for school districts with very small school-age populations. Therefore, we removed districts where the student population is less than 200 and did not analyze or rank the borders they share with neighboring districts.

Finally, we removed districts with a student density of less than or equal to 0.5 students per square mile, as these districts often face unique geographic considerations due to the extremely low student density. The borders of these districts with their neighbors were not analyzed or ranked.

After applying the above exclusions, we analyzed, ranked, and mapped the resulting database of 24,658 pairs of district neighbors.

Additional Calculations and Data Sources

Throughout this report, we supplement our borders dataset with additional state-level data in order to provide context to our findings.

Median household income is also used to measure economic disparity between districts in specific instances. We calculate these disparities between neighboring districts as the ratio of the median household income in the higher-income district to that in the lower-income district. While we do not report these disparities in the national data explorer, we note the differences for specific geographic areas in the full report.

The full report also includes brief discussion of analyses of assessed property valuations and per-pupil revenues in select states: Arkansas, Connecticut, Mississippi, and Ohio.

Connecticut aggregates assessed property values at the town level in its Equalized Net Grand List dataset. Though the default public school district in Connecticut also serves town units, there are 17 regional districts that serve multiple municipalities. Using a dataset provided by the School + State Finance Project, we were able to link the towns served by each district with both the state and national LEA codes. This connection allowed us to link our borders dataset with assessed property values.

To compute assessed property values per pupil in Connecticut, we first calculate the share of each regional district’s enrollment from each of its constituent towns. As previously mentioned, only regional districts draw from more than one town. We then multiply the total assessed value for each town by its enrollment share in a regional district. We obtain the total assessed value per pupil by summing across each of a district’s constituent towns and dividing by its enrollment.

We also calculate assessed valuations per pupil in Arkansas, Mississippi, and Ohio, each of which aggregate the total value of assessed property at the school district level. We divide this value by a district’s enrollment.

To determine if a statistical relationship is present between differences in assessed property value per pupil and the degree of segregation between districts, we separately regress our measures of racial and economic segregation on assessed property value per pupil. We report a relationship as statistically significant if it is valid at the ∝=0.01 level.