The Impact of Sample Sizes on Multiple Membership Multilevel Modeling Estimation

국문 초록 (Abstract)

다중소속 다층자료가 교육현장자료에서 증가하고 있으나, 표본크기가 다중소속다층모형에 미치는 영향에 관한 연구는 매우 미흡하다. 이 시뮬레이션 연구에서는 표본의 크기가 다중소속다층모형의 모수 추정과 신뢰구간에 미치는 영향을 살펴보고자 하였다. 본 연구의 목적을 수행하기 위해 계층 내 상관계수 (ICC; 0.15, 0.25), 2수준 표본크기 (20, 30, 50, 100), 1수준 표본크기 (20, 30, 50, 100, 200)의 조건들이 고려되었다. 모든 실험조건에서 고정효과와 1수준 변량의 추정에서는 상대편의가 나타나지 않았다. 그러나 2수준 표본크기 조건에서는(<=50), 2수준 변량의 과대 추정되는 상대편의가 나타났다. 모든 실험조건에서, 고정효과와 변량의 95%신뢰구간 포함 비율은 명목수준인 5%보다 미만으로 나타났다. 본 연구에서는 연구자들에게 다중소속다층모형의 추정을 위해서 최소 50 그룹 및 각 그룹 당 30명 이상 참여자가 속해 있는 크기의 자료를 사용할 것을 제안하였다.

다국어 초록 (Multilingual Abstract)

Multiple membership data structures are increasingly encountered in the educational field, but few studies have investigated the impact of sample sizes on multiple membership multilevel modeling. This simulation study examines the parameter estimates and non-coverage rates of 95% credible intervals in multiple membership multilevel modeling under a variety of manipulated conditions including: the intraclass correlation coefficient (ICC; 0.15 and 0.25), the number of schools (20, 30, 50 and 100), and the number of students (20, 30, 50, 100 and 200). Across all conditions, no substantial relative bias was found for fixed effect and student level variance component estimates. However, in some of the smaller number of schools (<=50), school level variance component estimates were substantially positively biased. Across all conditions, non-coverage rates of 95% credible interval for the fixed effect and random variance component estimates were less than the nominal non-coverage rate (i.e., 5%). It is suggested that applied researchers should analyze sufficiently large dataset (at least 50 groups with 30 participants) for multiple membership multilevel modeling estimation.

목차 (Table of Contents)

• ABSTRACT
• Ⅰ. Introduction
• Ⅱ. Method
• Ⅲ. Results
• Ⅳ. Discussion and Suggestions for Future Research
• References
• <요약>

참고문헌 (Reference)

1 Wolff Smith, L. J., "The impact of using incorrect weights with the multiple membership random effects model" 10 (10): 31-42, 2014

2 Hyewon Chung, "The impact of ignoring multiple membership data structures in multilevel models" Wiley-Blackwell 65 (65): 185-200, 2012

3 J. J. HOOGLAND, "Robustness Studies in Covariance Structure Modeling: An Overview and a Meta-Analysis" SAGE Publications 26 (26): 329-367, 1998

4 Browne, W. J., "Multiple membership multiple classification(MMMC)models" 1 (1): 103-124, 2001

5 Goldstein, H., "Multilevel statistical models" Hodder Arnold 2010

6 J. L. Krull, "Multilevel Mediation Modeling in Group-Based Intervention Studies" SAGE Publications 23 (23): 418-444, 1999

7 U.S. Government Accounting Office, "K-12 education: Many challenges arise in educating students who change schools frequently" U.S. Government Printing Office 2010

8 Rasbash, J., "Handbook of Multilevel Analysis" Springer 2008

9 Ihrke, D. K., "Geographical Mobility: 2005 to 2010" U.S. Census Bureau 2012

10 U.S. Census Bureau, "Geographic mobility: 2012 to 2012"

11 Rasbash, J., "A User’s Guide to MLwiN, v2.26"