GLMM tree의 적용 가능성 탐색

Fitting the GLMM tree to educational data: focusing on differential effects of afterschool program and private tutoring participation

본 연구의 목적은 Generalized Linear Mixed-effects Model tree(GLMM tree)를 소개하고 이를 교육자료에 적용해봄으로써 교육 연구에서의 그 유용성을 보이는 것에 있다. GLMM tree는 Model-based recursive partitioning(MOB)을 다층자료에도 적용할 수 있도록 확장한 것으로, 처치-하위집단 상호작용에 따라 하위집단을 구분하고, 각 하위집단에서의 고정효과와 함께 각 관측치가 속하는 군집에 의한 무선효과를 추정한다. 본 연구에서는 GLMM tree를 적용하여 고등학생의 수학 성취도에 대한 방과후학교와 사교육 참여의 처치-하위집단 상호작용을 통해 차별적 효과를 비교하였다. 그 결과, 방과후학교와 사교육 참여 여부에 대해 각각 16개, 15개의 하위집단으로 구분되었고, 각 하위집단의 특성에 따라 고정효과가 다르게 나타났다. 이상의 결과들을 토대로 방과후학교와 사교육 참여의 차별적 효과 양상을 비교하였으며, 교육연구에서 GLMM tree의 적용가능성을 논하였다.

This study aims to introduce the GLMM tree and show its usefulness in educational research. The GLMM tree model is an extension of MOB to be applied to multilevel data, which detects subgroups with differential effects to estimate the fixed-effect in each subgroup and the random effect by the cluster to which each observation belongs. In this study, we identified the differential effects by detecting subgroups of afterschool programs and private tutoring participation in mathematics achievement of second grade high school students. Using the GLMM trees, students were divided into 16 and 15 subgroups respectively according to the participation of afterschool programs and the private tutoring. Also, cognitive and affective variables such as pre-mathematics achievement, interest in mathematics were selected as nodes. For both treatments, it was confirmed that the fixed-effect was estimated differently for each subgroup. Based on the results, we compared the differential effects of participation in afterschool programs and private tutoring and factors selected as nodes of models, and discussed the potential of the GLMM tree model in educational research.