Table of Contents  |  Search Technical Documentation  |  References

NAEP Weighting Procedures → 2002 Weighting Procedures and Variance Estimation → State Assessment Weights → State NAEP School and Student Nonresponse Adjustments → Computation of Student Nonresponse Adjustments → Development of Student Nonresponse Cells

NAEP Technical DocumentationDevelopment of Student Nonresponse Cells

Preliminary weighting cells were formed for each student based on student with disability/limited English proficient (SD/LEP) status, school nonresponse cell, age, subsidized lunch status, and gender. SD/LEP status was collapsed into a dichotomy: students who were either SD or LEP in one group, and students who were neither in the other group. The school nonresponse cells were the final school nonresponse cells c = 1,...,C generated in the school nonresponse adjustment step. Age was collapsed into a dichotomy: students who were “older”,¹ and students who were “normal age or younger.” Weighting cells had to meet the following criteria:

• at least 20 sampled students in the cell;

• at least 15 sampled schools in the cell with nonzero replicate weights, for each replicate;

• maximum student nonresponse adjustments and replicate student nonresponse adjustments of 2.0; and

• replicate student nonresponse adjustments no larger than 1.5 times that of the corresponding replicate adjustment.

When preliminary cells did not satisfy these criteria, they were collapsed with other cells. New automated software was used in this collapsing process in the 2002 State NAEP assessment. The software automatically collapsed any cells that violated the criteria and had a clearly defined neighbor cell to collapse with. The input parameter stream controlling the collapsing algorithm listed the variables which determine the preliminary cells, in order of 'importance' (which variable was collapsed first, as its levels were the least important to preserve, which was collapsed last, as its levels were the most important to preserve, and all variables in between ordered by their assigned 'importance'), and listed the levels for each of these variables. In State NAEP 2002, the variables in order of importance from most important to least important were SD/LEP status (two levels: yes or no), school nonresponse cells (many levels), age group (two levels: 1-older and 2-normal age or younger¹), free lunch status (three levels: free-lunch status, reduced-price lunch status, and neither), and sex (two levels: male and female).   

If no cells in a branch were deficient, then no collapsing was done. If at least one cell was deficient in a given branch, then the two sex levels (male and female) were collapsed first within the hierarchy, and the next dimension (Free Lunch Status) was only collapsed if there were still deficiencies after collapsing across sex. This collapsing was done separately within the cells determined by the higher-level dimensions. In some of the final branches, sex may have been collapsed; in others, not. If there were deficient cells after all the Free Lunch Status cells were collapsed, then age group was collapsed. A collapse at a higher level might entail the collapsing of other lower-level cells which were not heretofore collapsed.

Dichotomous variables such as Sex had a simple collapsing scheme: male was collapsed with female. For variables with three or more levels, the scheme defined which levels were collapsed first. For example for Free Lunch Status, the scheme was defined so that free-lunch status and reduced-priced lunch status cells were collapsed together first, and only if there was still a deficiency were all three levels then collapsed. 'Scores' were defined in the input parameter stream which determined this internal collapsing scheme for variables with three or more levels.

¹ For the fourth grade, older students are those born earlier than October 1991, and for the eighth grade, older students are those born earlier than October 1987.

Printer-friendly Version