Hypothesis testing for two-group incomplete-data growth curve problems

James D. Leeper, Hang Ling Chang

Research output: Contribution to journalArticlepeer-review

3 Scopus citations

Abstract

Using simulation techniques, the null distribution properties of seven hypothesis testing procedures and a comparison of their powers are investigated for incomplete-data small-sample growth curve situations. The testing procedures are a combination of two growth curve models (the Potthoff and Roy model for complete data and Kleinbaum's extention to incomplete data) and three estimation techniques (two involving means of existing observations and the other using the EM algorithm) plus an analysis of a subset of complete data. All of the seven tests use the Kleinbaum Wald statistic, but different tests use different information. The hypotheses of identical and parallel growth curves are tested under the assumptions of multivariate normality and a linear polynomial mean growth curve for each of two groups. Good approximate null distributions are found for all procedures and one procedure is identified as empirically most powerful for the situations investigated.

Original languageEnglish
Pages (from-to)3715-3728
Number of pages14
JournalCommunications in Statistics - Theory and Methods
Volume16
Issue number12
DOIs
StatePublished - Jan 1 1987

Keywords

  • EM algortithm
  • Monte carlo
  • missing data
  • small samples

Fingerprint

Dive into the research topics of 'Hypothesis testing for two-group incomplete-data growth curve problems'. Together they form a unique fingerprint.

Cite this