Comparing the performance of multivariate multilevel modeling to traditional analyses with complete and incomplete data

Ryoungsun Park, Keenan A. Pituch, Jiseon Kim, Hyewon Chung, Barbara G. Dodd

Research output: Contribution to journalArticlepeer-review

6 Scopus citations

Abstract

A multivariate multilevel model (MVMM) extends standard multilevel modeling by including multiple dependent variables and thus could be used in place of traditional multivariate analyses. For a two-group study with two correlated dependent variables, a simulation study was conducted to compare the performance of MVMM to traditional MANOVA and a series of univariate analyses. The results showed that MVMM provides for virtually always greater power than other analyses, even for conditions that have been previously shown to favor univariate analysis. Further, this power advantage can be substantial even when no missing data are present and is often much greater when data are missing. While the Type I error rate for the overall multivariate null hypothesis can be somewhat elevated with MVMM, especially with small sample size and a large proportion of missing data, the Type I error rate for the test of a specific dependent variable is accurate.

Original languageEnglish (US)
Pages (from-to)100-109
Number of pages10
JournalMethodology
Volume11
Issue number3
DOIs
StatePublished - Oct 2015
Externally publishedYes

Keywords

  • Experiments
  • Missing data
  • Multilevel models
  • Multivariate analysis
  • Simulation study
  • Univariate analysis

ASJC Scopus subject areas

  • Social Sciences(all)
  • Psychology(all)

Fingerprint Dive into the research topics of 'Comparing the performance of multivariate multilevel modeling to traditional analyses with complete and incomplete data'. Together they form a unique fingerprint.

Cite this