Utility of Spatially Interpolated Neighborhood Income Estimates for National Education Surveys

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

Abstract

Obtaining accurate information on Socioeconomic Status (SES) for the purposes of national large scale surveys such as the National Assessment of Educational Progress (NAEP) has been and continues to be challenging. Although student eligibility in the National School Lunch Program (NSLP) is currently available for NAEP as a reporting variable, the relationship between NSLP eligibility and SES is somewhat tenuous due to many changes in the administration of the NSLP program over time. NAEP also collects student self-report data on SES, however, this information can also be problematic for reasons such as self-report inaccuracies, intrusiveness of questions, and non-negligible amounts of missing data.

The U.S. Census Bureau with support from the National Center for Education Statistics (NCES), developed a school-level socio-economic status indicator of the ratio of income-to-poverty of neighborhoods surrounding schools (POVPI). POVPI values are estimated with a spatial interpolation model, developed with data from the American Community Survey (ACS). The use of estimates based on a spatial interpolation model has a significant benefit over the majority of other socioeconomic indicators, in that estimates are available for every school with no missing data. Furthermore, the use of model predicted values as opposed to real data points allow publication of ACS information without risk of disclosure.

In the present study, the properties of POVPI were evaluated with data from the 2015 NAEP mathematics and reading assessments for grade 4 students in a heterogeneous U.S. state, as well as an Urban District that is participating in NAEP’s Trial Urban District Assessment (TUDA). The construct validity of the POVPI estimates were evaluated by examining whether the variable had the expected pattern of correlations with available external variables. Specifically, the relationships between POVPI and various NAEP contextual variables (e.g., urban locale, demographics, self-report survey questions including SES-related questions on household income) were estimated and found to be consistent with expectation.

POVPI was also related to reading and mathematics proficiency, to further examine the construct validity of the variable, the utility of the variable for research and reporting purposes, and the benefit of introducing the variable into national large-scale educational survey conditioning models. POVPI showed strong positive relationships with mathematics and reading proficiency, with a non-linear shape that became flatter at higher POVPI values. POVPI explained more than half of the between-school variance in NAEP proficiency, as well as over seventy percent of the proficiency differences between SES-related NAEP subgroups and major demographic variables. The POVPI showed reduced relationships with mathematics and reading proficiency and NAEP SES measures in the Urban District. Alternate hypotheses for this result were evaluated and discussed.

Overall the POVPI model showed great promise as a measure of neighborhood SES for research in general and specifically for American educational surveys. The approach of modeling U.S. Census ACS data to disseminate to the wider research community can be generalized to other SES-related variables such as occupation and educational attainment.

Authors

• Paul A Jewsbury, Educational Testing Service

• Yue Jia, Educational Testing Service

• Laura Jerry, Educational Testing Service