Assessing change with longitudinal and clustered binary data

Annu Rev Public Health. 2001;22:115-28. doi: 10.1146/annurev.publhealth.22.1.115.


Investigators often gather repeated measures on study subjects to directly measure how a subject's response changes with changes in explanatory variables. This paper focuses on several statistical issues related to assessing change with longitudinal and clustered binary data. Many popular approaches for analyzing repeated binary outcomes measure cross-sectional or between-subject, rather than within-subject, effects of covariates. The class of models known as cluster specific measures within-subject effects of covariates on responses but are subject to additional statistical complications. It is useful to decompose covariates into between- and within-cluster components. This paper describes several approaches that yield consistent estimates of the within-subject covariate effects of interest. Example data from three studies illustrate the results.

Publication types

  • Research Support, U.S. Gov't, P.H.S.
  • Review

MeSH terms

  • Humans
  • Linear Models
  • Longitudinal Studies
  • Markov Chains
  • Matched-Pair Analysis
  • Monte Carlo Method
  • Statistics as Topic / methods*