Multivariate graphical methods provide an insightful way to formulate explanatory hypotheses from limited categorical data

J Clin Epidemiol. 2012 Feb;65(2):179-88. doi: 10.1016/j.jclinepi.2011.06.007. Epub 2011 Sep 1.

Abstract

Objective: Graphical methods for generating explanatory hypotheses from limited categorical data are described and illustrated.

Study design and setting: Univariate, bivariate, multivariate, and multiplicative graphical methods were applied to clinical data regarding very ill older persons. The data to which these methods were applied were limited as to their nature (e.g., nominal categorical data) or quality (e.g., data subject to measurement error and missing values). Such limitations make confirmatory inference problematic but might still allow for meaningful generation of new explanatory hypotheses in some cases.

Results: A striking feature of the graphical results from this study's major illustrative application was that posttraumatic stress disorder (PTSD) after intensive care unit discharge occurred rarely and nearly always co-occurred with two or more other mental health conditions. These results suggest the explanatory hypothesis that PTSD in this context is less attributable to single traumatic causes than to acute illnesses contributing to a cascade of mental health decrements.

Conclusion: Illustrative applications of a sequence of graphical procedures yield more informative and less abstract representations of limited data than do descriptive statistics alone, and by doing so, they aid in the formulation of explanatory hypotheses.

MeSH terms

  • Aged
  • Aged, 80 and over
  • Critical Care
  • Data Display*
  • Female
  • Humans
  • Logic*
  • Male
  • Middle Aged
  • Multivariate Analysis
  • Statistics as Topic
  • Stress Disorders, Post-Traumatic / complications