A large-scale and a small-scale simulation of the reallocation of students into and out of an originally-selected treatment group were performed. It was postulated that both the scores on a selection test and teacher ratings are measures of student needs in the Title I Evaluation and Reporting System (TIERS). In the norm-referenced model, three conditions of reallocation with three separate gains estimates were studied using second grade students data from the Study of the Sustaining Effects of Compensatory Education. Gain estimates obtained under the various reallocation conditions were compared to the gain that would have been obtained under correct implementation of this variation of the model of selection on the pretest with a strict cutoff and correction for regression. Gain scores in the large-scale simulation after reallocation differed from the benchmark by as much as .2 standard deviation units. The small-scale analysis showed that large biases can result at the local level if even one student is moved. (Author/CM)