​​​​​NAEP Technical DocumentationStratification of Public Schools for the 2022 Long-Term Trend Assessment

For the public school sampling frame file for each of the ages 9 and 13, separate implicit stratification schemes were used to sort schools in certainty primary sampling units (PSUs) and noncertainty PSUs. The implicit stratification was achieved via a "serpentine sort."

For certainty PSUs, the schools were hierarchically sorted by

  • American Indian/Alaska Native (AI/AN) stratum (two categories based on percentage of AI/AN students);
  • urbanization classification (four categories based on urban-centric locale);
  • race/ethnicity stratum; and
  • race/ethnicity percentage.

The two categories within the AI/AN stratum were defined as follows. High AI/AN schools were schools with at least five percent AI/AN students and at least five AI/AN students in the sample age. Low AI/AN schools were those not designated as high AI/AN.  

For schools in the high AI/AN stratum, if there were fewer than six expected sampled schools for a particular urbanization classification cell (nested within the AI/AN stratum and census region), the cell was collapsed with a neighboring urbanization classification cell. No race/ethnicity strata were generated. The final sort variable was total percentage of AI/AN students. Note the lower limit of six was chosen to facilitate the construction of nonresponse adjustment classes with sufficient numbers of schools and students in them.

For schools in the low AI/AN stratum, if there were fewer than six expected sampled schools for a particular urbanization classification cell (nested within the AI/AN stratum and census region), the cell was collapsed with a neighboring urbanization classification cell. If the expected sampled schools exceeded 12, then the race/ethnicity strata were defined based on the total percentage of Black, Hispanic, and AI/AN students. The strata were defined so that there were at least six expected sampled schools for each race/ethnicity stratum. Within each race/ethnicity stratum, the final sort variable was total percentage of Black, Hispanic, and AI/AN students. If the urbanization classification stratum had an expected sample size less than 12, no race/ethnicity strata were generated, and the final sort variable was total percentage of Black, Hispanic, and AI/AN students.

Schools in noncertainty PSUs were hierarchically sorted by

  • AI/AN stratum (two categories based on percentage of AI/AN students);
  • urbanization classification (four categories based on urban-centric locale); and
  • race/ethnicity percentage.

The collapsing of cells within the noncertainty PSUs was implemented in a fashion similar to that described for certainty PSUs. 

For schools in the high AI/AN stratum, the final sort variable was total percentage of AI/AN students. For schools in the low AI/AN stratum, the final sort variable was total percentage of Black, Hispanic, and AI/AN students.

Last updated 18 July 2024 (SK)