NAEP Technical Documentation2020 Public School Long-Term Trend Assessment

			Target Population Sampling Frame Stratification of Schools Sampling of Schools Substitute Schools Ineligible Schools Student Sample Selection

The NAEP 2020 sample design yielded nationally representative samples of public school students at ages 9 and 13 for long-term trend (LTT) through a three-stage approach:

• selection of primary sampling units (PSUs);
• selection of schools within strata; and
• selection of students within schools.

The sample of schools was selected with probability proportional to a measure of size based on the estimated age enrollment in the schools.

The 2020 sampling plan was designed to assess 14,400 students (7,200 for each of the two age-specific samples) in public schools for LTT. These students were allocated among tests in mathematics and reading. Target sample sizes were adjusted to reflect expected public school and student response and eligibility.

Schools on the sampling frame were explicitly stratified prior to sampling by PSU type (certainty/noncertainty). Within certainty PSUs, schools were implicitly stratified by census region, American Indian/Alaska Native (AI/AN) stratum, urbanization classification, race/ethnicity stratum, and race/ethnicity percentage. Within noncertainty PSUs, schools were implicitly stratified by PSU stratum, AI/AN stratum, urbanization classification, and race/ethnicity percentage. Note that the use of the AI/AN stratum as an implicit stratification variable helped ensure that a reasonable number of schools with sufficient numbers of AI/AN students in them were selected. 

From the stratified frame of public schools, systematic random samples of age-eligible schools were drawn with probability proportional to a measure of size based on the estimated age enrollment of the school for the relevant age.

Additionally, AI/AN, Black, and Hispanic students were oversampled at moderate rates as follows. First, schools in a high AI/AN stratum (i.e., schools with at least five percent AI/AN students and at least five AI/AN students in the sample age) were sampled at four times the rate (by quadrupling their measure of size) as schools not in a high AI/AN stratum to implement oversampling of AI/AN students. Second, schools not in a high AI/AN stratum but in a high Black/Hispanic stratum (i.e., schools that were not oversampled for AI/AN students and with at least 15 percent Black/Hispanic students and at least 10 Black/Hispanic students in the sample age) were sampled at twice the rate (by doubling their measure of size) as schools not in a high Black/Hispanic stratum to implement oversampling of Black and Hispanic students. This approach is effective in increasing the sample sizes of AI/AN, Black, and Hispanic students without inducing undesirably large design effects on the sample, either overall or for particular subgroups.

Finally, schools in the Honolulu PSU were oversampled at twice the rate (by doubling their measure of size) as schools not in the Honolulu PSU. This was done to ensure at least one school was sampled from this PSU. At least one school was sampled because the total measure of size for all schools in Honolulu exceeded the sampling interval. The PSU was selected with certainty not due to its size, but because it is unique.

Each selected school in the public school sample provided a list of age-eligible enrolled students from which a systematic sample of students was drawn. Within each school, students of the same race/ethnicity were selected with equal probability.

Printer-friendly Version