2014 - Compositional Effects, Segregation and Test Scores: Evidence from the National Assessment of Educational Progress

Attribution: Munk, Tom E., McMillian, Monique, & Lewis, Nicole R.
Researchers: Monique McMillianNicole R. LewisTom E. Munk
University Affiliation: Westat; Morgan State University; University of Hawaii at Manoa
Email: Tommunk@westat.com
Research Question:
How is the expected test score of an average student predicted by the economic and/or ethnic composition of her or his school?
Published: Yes
Journal Name or Institutional Affiliation: Review of Black Political Economy
Journal Entry: Vol. 41 Pp. 433-454
Year: 2014
Findings:
  1. Controlling for ethnicity, free lunch students scored about 21 points lower than full-price lunch students.
  2. Controlling for subsidized lunch status, black students scored on average about 27 points lower than white students; Hispanic students scored about 19 points lower, American Indian students about 16 points lower, and Asian students about 6 points higher.
  3. Controlling for peer economic level, the percentages of black and Hispanic students in a school had a positive relationship with grade 8 adjusted school mean test scores, while the percentage of Asian/ Pacific Islander students had a decidedly negative relationship with those scores. Only the percentage of American Indian/ Alaskan Native students failed to demonstrate a significant composition effect.
  4. With all other factors held constant, a nationally average grade 8 student in a 100% subsidized-lunch had an expected math score 24.9 points lower than the same student in a 100% full-price lunch school.
  5. Projection of the data suggests that if the nation’s schools were completely desegregated economically (but not at all ethnically), the test-score gap between free lunch students and students paying full price for lunch would decline by 25 %.
  6. Ethnic compositional effects for black, Asian/Pacific Islander (API), and Hispanic students were reversed from their within-school effects, with positive effects for students in schools with larger proportions of Black and Hispanic students and a strong negative effect for students in schools with larger proportions of Asian/Pacific Islander students.
Scholarship Types: Journal Article Reporting Empirical ResearchKeywords: Academic AchievementCompositionEthnicitySegregationRegions: NationalMethodologies: QuantitativeResearch Designs: Secondary Survey DataAnalysis Methods: Descriptive StatisticsOLS (Ordinary least-squares) regressionUnconditional two-level model Sampling Frame:8th grade students
Sampling Types: NationalAnalysis Units: SchoolStudentData Types: Quantitative-Cross Sectional
Data Description:
  • This study uses data from the Grade 8 Main NAEP 2003 Mathematics database. NAEP is the only nationally representative and continuing assessment of what students know and can do in mathematics and other subjects and is increasingly used as a check on state testing programs. Weights are provided with the sample. The results are representative of all grade 8 students in the nation. The study includes 6,334 targeted schools and 162,730 targeted students. The unconditional model two-level model had a sample size of 156,740 students. The compositional effects model had a sample of 134,840 students.
  • The two key independent variables were family income level and ethnicity. Family income level is represented by subsidized lunch status. The authors coded lunch status as 1 for students that are eligible for free lunch, 0.5 for those that are eligible for reduced-price lunch, and 0 for those who are not eligible for this service. A set of dummy variables is used to represent student ethnicity. For the purposes of NAEP, schools categorize students into one of six mutually exclusive ethnic categories: White, Hispanic, Black, Asian/Pacific Islander, American Indian/Alaska Native, and Other. At the school level, the NAEP dataset includes reports of the percentages of the student body that are the particular ethnicities.
  • The outcome variable, NAEP mathematics proficiency, is a plausible value for the student’s mathematics scale score. In order to keep tests short for individual test-takers while covering all important topics, test-takers are presented only a small portion of the mathematics items, making accurate estimation of their individual proficiencies impossible. Missing-data imputation techniques are combined with IRT proficiency estimation methods to generate a set of five plausible values for each student.
Theoretical Framework:
Relevance:Factors Related to STEM Readiness
Archives: K-16 STEM Abstracts