Student weights that are much larger than the median weight for a particular jurisdiction will produce sample estimates which have a higher variance (see, for example, Kish (1992)). The general procedure in the 2002 state assessment trimmed those weights that were greater than 3.5 times the median weight value. Although introducing some bias, trimming weights results in a lower overall mean square error, given the reduction in variance.
The trimming groups were defined as unique jurisdictions, except for the five states with Trial Urban District Assessment (TUDA) districts. For these five states, the TUDA part and non-TUDA part were the two trimming groups. The procedure computed the median of the nonresponse-adjusted student weights in the trimming group. Any weight that was more than 3.5 times the median received a trimming factor:
It should be noted that ineligible students were not included in the calculation of the median.