A New Statistical Method for Mixed Variates Analysis in Epidemiological Research |
| |
Authors: | Dr. H.-P. Wortha |
| |
Affiliation: | Institut für Biostatistik und Medizinische Informatik Martin-Luther-Universität Krausenstraße 14 Halle, DDR- 4020 |
| |
Abstract: | Epidemiologically oriented research often may not do without observational or only partially controlled studies. In many such situations both qualitative characteristics and quantitative ones are observed. In literature there are different methods of handling such problems. The paper presents a method for analyzing dependencies resp. associations between random variates of any kind. The model concerned fullfills the whole field between analysis of variance, analysis of covariance and contingency table analysis. The method is named MIVA or mixed variates analysis, bases on the class of Conditional Gaussian Distributions of the exponential family and results in a unique system of mixed and unmixed measures of association–of pairwise, partial, multiple and global type. These measures are easy to be estimated and tested on significant deviation from zero. They may be used describing or analyzing dependence structures in many epidemiological studies but also in other fields. |
| |
Keywords: | Mixed measures of association Analysis of variance Random subclass-numbers Contingency coefficients Conditional Gaussian distribution |
|
|