Leveraging Spatial Interpolation Methods to Predict Socioeconomic Status for Neighborhoods Without Boundaries

• In Event: 54.032 - Measurement of Socioeconomic Status for American Education Surveys

Abstract

Objectives: The National Center for Education Statistics (NCES) is tasked with reporting academic progress by student socioeconomic status (SES). Free/reduced price lunch eligibility is used as an SES proxy, but is becoming less reliable and less available due to program changes so NCES has been examining alternative SES measures. The National Assessment Governing Board (2003) and an Expert Panel on SES (2012) recommend creating an SES composite that includes neighborhood-level data from the U.S. Census Bureau. Census data is collected consistently across the United States, and neighborhood characteristics influence educational outcomes of students (McWayne et.al., 2007).

While census geographies are often used to represent neighborhoods in research studies, their boundaries are often incongruent with residents’ perceptions of their neighborhoods (Campbell et.al., 2009; Coulton, 2012; Sampson et.al., 2002). For those census geographies that most closely align with neighborhood perceptions, the estimates may be unreliable due to limited sample. To address challenges with predefined census geographies, we examine creating estimates for school-based neighborhoods by using spatial interpolation methods. Can we define neighborhoods and create estimates based on a specific number of inputs surrounding a school location rather than rely on predefined geographies? Do values from spatial interpolation align with expectations across household groups and school characteristics? Are the resulting school-based neighborhoods an appropriate size?

Methods and Data: We test creating estimates with specific application of kriging called Empirical Bayesian Kriging (EBK). The goal of kriging is to predict an unknown value at a location based on the spatial relationships and values of the surrounding known values. The input data are the values of the ratio of income to poverty (a component of SES) for households with children from the American Community Survey (ACS) sample. The households are located at the centroid of their TIGER/Line blocks. EBK uses the input values and locations to create a prediction surface, and allows for predictions at any point location on the surface, which represent the anchor point of the neighborhood. We predict values of the ratio of income to poverty for nine household groups at almost 1,800 public school locations from the Common Core of Data.

Results: Summary statistics show the values of the ratio of income to poverty vary across the household groups as expected (e.g., two-parent households have larger values than single-parent households). The values also have appropriate relationships with other known characteristics of the schools. Although the neighborhoods are based on the same number of inputs, the sizes vary and seem appropriate for the school’s urbanicity. The median size of neighborhoods across the schools are reasonable (about 6/10ths mile) and much smaller than the median size of the census’s Zip Code Tabulation Areas.

Significance: Our preliminary findings suggest school-neighborhood estimates will benefit an overall SES composite for linking to academic progress. Spatial interpolation methods treat all locations equally by ensuring there are a constant number of inputs, whereas aggregates to predefined boundaries may not have sufficient sample for reliable estimates. These methods have potential to be extended to other characteristics and locations.

Author

• Laura Nixon, U.S. Census Bureau