Generalized Estimating Equations and motivations for open science

Sep-28 : In this lesson, we will understand how generalized estimating equations work and discuss motivations for open GIScience.

Readings

1. Hanley, J. A., Negassa, A., Edwardes, M. D. de B., & Forrester, J. E. (2003). Statistical analysis of correlated data using generalized estimating equations: An orientation. American Journal of Epidemiology, 157(4), 364–375. https://doi.org/10.1093/aje/kwf215

Resources

1. Rmarkdown walkthrough of the analysis in Hanley et al (2003): Learn-GEE

Key Ideas and Concepts

• Correlation statistics may be based on normal continuous data (Parametric), using Pearson’s R, or based on ordinal ranked data (Nonparametric) using Spearman’s Rho. The resulting R or Rho range from -1 to 1 where: --1 is perfect inverse correlation
  • 0 is no correlation
  • 1 is perfect positive correlation
• Linear models consist of:
  • dependent variable
  • independent variables
  • constant / intercept
  • coefficients
  • estimated values
  • residuals / errors
  • standard error
  • 95% confidence intervals
• While linear models assume that each observation is independent, spatial autocorrelation violates this assumption.
• GEE approaches the autocorrelation problem by
  • defining clusters
  • assuming that correlation can be expected within clusters, but that clusters are independent of one another
  • estimating correlation within clusters
  • weighting observations based on correlation within clusters and cluster size
  • with no correlation within clusters, weights still equal 1
  • with perfect correlation within clusters, weights approach 1/k where k is the cluster size

Main Page