Weighted Single-Level and Multilevel Analyses in Large-Scale Assessment Surveys (Poster 11)

• In Event: AERA Poster Session Five
  In Poster Session: Division D Section 2 Poster Session

Abstract

Incorporating complex sampling design and weights in model analyses is recommended in Large-scale assessment survey (LSAS) data, but there is little research evidence and practical guidance. This study compared the weighted single-level model and five weighted multi-level models with different sampling weights and scaling options plus the unweighted multilevel model based on three LSAS data. Findings reveal that, in the context of LSA data, the weighted single-level model and multi-level models may have similar statistical inference in general. In addition, for weighted multilevel models, size scaling and effective scaling estimators have similar performance, and school only weights approach performs equally well compared with the two scaling approaches. Practical recommendations for using sampling weights in LSAS data are provided.

Author

• Ting Shen, Missouri University of Science & Technology
  Presenting Author