This page gives access to the R-package postHoc developed by R. Labouriau at the Applied Statistics Laboratory (aStatLab), Department of Mathematics, Aarhus University. The page is maintained by Rodrigo Labouriau (e-mail).
The R-package postHoc implements methods for post hoc pairwise comparisons and
clustering for standard linear models, generalized linear models, mixed models,
generalized linear mixed models. It is under implementation methods using some non-parametric tests
(Kruskal-Wallis and permutation tests applied to compare the distribution of sub-samples
defined by the levels of a classification factor) and some contingency table-related models.
The package constructs groups and clusters of parameters that are not statistical significantly different of each other (at a pre-specified significance level) using a graph-based representation where the vertices are the model parameters (typically the levels of a classification explanatory variable) and two vertices are adjacent in the graph if, and only if, they are not statistical significantly different. The groups of parameters are the maximal cliques in the graph described above; the clusters are formed by finding the largest subgraph contained in the representation graph. A vertice of the graph can belong to more than one group, but it belongs to only one cluster.
The package postHoc have methods defined for plotting the representation graph, produce tables with estimates, confidence intervals (bootstrap or Wald) and grouping/clustering, interaction line plots and barplots.
The source of recent versions of postHoc (yet experimental) are available at:
Total number of accesses: 2801
Last access at Fri, 03 Feb 2023 12:34:06 +0100