America's Charter School Deserts

This website uses 2010 data from the U.S. Census Bureau and school-level data provided by the school rating organization GreatSchools (hereafter GS). Below we explain in detail how the GS data were processed to yield school-level student demographic and academic information.

Included Schools

Fordham staff downloaded GS school-level data files on March 1, 2018. We included in the sample all elementary and non-virtual traditional public schools and charter schools in the GS database. The final sample includes 60,843 schools.

Elementary Schools
Elementary schools are configured differently across the country and even within school districts. For this reason, we include schools that serve any grade from kindergarten to grade 5. This means that if a school serves, for example, 5^th through 12^th grade students, it is included, even though it may not meet the commonsense definition of an elementary school.

Virtual Schools
We attempted to remove virtual schools from the data file since by definition they serve students outside the geographic boundaries of neighborhood schools. Although GS does not identify virtual schools, we used two methods to eliminate them. First, we accessed published lists of virtual schools, flagging and eliminating them by hand. Second, we used data analysis software to flag schools whose names included words such as “Virtual,” “Online,” “Eschool,” “Distance,” “Computer,” “Cyber,” “Digital,” and “Correspondence,” then checked to confirm whether it was indeed a “Virtual” school. If confirmed, these schools were omitted from the final sample.

Maine Duplicates
GS data for Maine includes a large number of apparently duplicated schools, where schools that are located very near each other, have similar names, serve similar grade levels, yet one school is missing all enrollment, free or reduced-price lunch, and academic outcome data.

We have attempted to eliminate these duplicates in Maine by rounding latitude and longitude data to the nearest integer, classifying schools by whether they serve kindergarteners, and using the following rule: when there are two schools that have the same (rounded) latitude and longitude, have similar school names, and have the same kindergarten status, we preserve the school with more complete enrollment, free or reduced-price lunch, and academic outcome data.

School Sector
GS identifies whether a school is a traditional, charter, or private school. We omit private schools from the sample and preserve only traditional public and charter schools. GS also identifies whether each observation represents a school, district, or state. We eliminate district and state observations, preserving only school-level observations.

Missing Data

GS data include information on grade levels served, the percentage of students who qualify for free or reduced-price lunch, the number of students attending, and the percentage of students proficient in math and English language arts (ELA) for most schools. This information is available for various years as described below. Fordham processed the data with the goal of preserving the most recent and relevant information for each school.

Geolocation and Grade Levels
The files containing both geolocation data and the grade level of students served do not vary by year: we include the data available in these files at the time of download.

Enrollment and Free and Reduced-Price Lunch
We use the most recent year of data available for school enrollment and the percentage of students qualifying for free or reduced-price lunch. For all states, that year is 2014-15, with the exception of Colorado, where data are from 2015-2016.

Academic Outcomes
For academic assessment data, we restrict each state to math and ELA data as described below, then select the most recent year with the most complete data for elementary students.

In most cases, selecting the optimal year was a simple decision as it presented no ambiguity or tradeoffs. For example, Kansas has only math and ELA data for 2014-2015 and California has the greatest number of outcome observations in 2016-2017, which is also the most recent year in the GS files.

In some cases, selecting the best year of data entailed tradeoffs between preserving data that are more recent and data that are more complete. For instance, many states have fewer assessment observations in the most recent year, yet the drops from the prior years are either very small or occur as a result of changing tests or assessing fewer grades. For example, although Michigan has fewer math and ELA test observations in 2015-16 than the two previous years, we still chose to use this year because the drop mostly occurred as a result of high school students not being tested—not elementary school students, who are our focus. We also use the most recent year in cases where the number of observations declined less than 10 percent from the year prior and the grades and number of assessments were consistent.

The table off to the side shows the year of assessment data used for each state.

Assessment Aggregation

Since we restricted the sample to schools serving elementary grades, we include all outcomes for a given school, regardless of the grade levels they may apply to. The outcome of interest is the percentage of students scoring proficient and above, which, according to GS, is based on each state’s proficiency standards.

As indicated, the GS files include information on a number of math and ELA assessments. For example, a school may have outcomes for both reading and for writing, which are both ELA assessments. For math, a school may have outcomes for “Math 1,” “Math 2,” and “Algebra I.” To aggregate these subject tests for each school in a given year, we use one of three methods, described below in order of the most to least preferred. 

1. Outcome weighted by the number tested. For most schools, a variable indicates the number of students tested in each subject. To calculate the overall percentage of students who are proficient in math and ELA in each school, we multiply the proficiency rate of each subject test (e.g., reading and writing for ELA) by the number of students tested, sum the results for all assessments, then divide by the total number of students tested for all subject tests. Not counting schools missing all academic outcome data, we use this method for 77 percent of schools. 
    
2. Outcome weighted by the number of students enrolled according to NCES data. In cases where the number of students tested is missing, we perform the same calculation above, but using as the weight enrollment data for each grade  derived from the National Center for Education Statistics (NCES) Common Core of Data 2015-2016. Not counting schools missing all academic outcome data, we use this method for 22 percent of schools.
    
3. Outcomes aggregated using the information for “all grades.”  Some schools lack both data for the number tested and NCES identifiers. The GS academic outcomes data include a value of “all” for the grades variable. The problem with using this method of aggregation is that multiple assessments cannot be aggregated because there is no way to weight them. When using this method, we include only the assessments explicitly named “Math” and “English Language Arts,” which may exclude outcomes for some students from the school who take related assessments with other names. Not counting schools missing all academic outcome data, we use this method for less than two percent of schools.

***

We encourage users to submit feedback to the Fordham Institute for improvements to the site by emailing [email protected].