The impact of ignoring multiple-membership data structures

This study was designed to investigate the impact of multiple-membership data structures in multilevel modeling. Multiple-membership arises when lower level units (e.g., students) are nested within more than one higher level unit (e.g., schools). In this case, more than one school will contribute to students' academic achievement and progress. In reality, it is inappropriate to assume a pure nesting of a student within a single school. While use of HLM requires either deletion of the cases involving multiple-membership or exclusion of prior schools attended, MMREM includes students who attend multiple schools and controls for the effect of all schools on student outcomes. The simulation study found level two variability underestimation and corresponding level one variability overestimation when multiple membership data structures were ignored. The study also revealed that when HLM failed to include multiple membership data structures, it underestimated school level predictor. With an increased numbers of mobile students under the No Child Left Behind (NCLB) Act, researchers need to understand MMREM and correctly apply it to multiple membership data structures. This MMREM approach will help improve the generalizability of findings and will improve the validity of the statistical results.