A Multivariate Logistic Distance Model for the Analysis of Multiple Binary Responses

We propose a Multivariate Logistic Distance (MLD) model for the analysis of multiple binary responses in the presence of predictors. The MLD model can be used to simultaneously assess the dimensional/factorial structure of the data and to study the effect of the predictor variables on each of the re...

Full description

Saved in:
Bibliographic Details
Published in:Journal of classification 2018-04, Vol.35 (1), p.124-146
Main Authors: Worku, Hailemichael M., de Rooij, Mark
Format: Article
Language:English
Subjects:
Binary data
Bioinformatics
Constraint modelling
Data analysis
Empirical analysis
Graphical representations
Logistics
Marketing
Mathematics and Statistics
Multivariate analysis
Pattern Recognition
Psychometrics
Signal,Image and Speech Processing
Statistical Theory and Methods
Statistics
Variables
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:We propose a Multivariate Logistic Distance (MLD) model for the analysis of multiple binary responses in the presence of predictors. The MLD model can be used to simultaneously assess the dimensional/factorial structure of the data and to study the effect of the predictor variables on each of the response variables. To enhance interpretation, the results of the proposed model can be graphically represented in a biplot, showing predictor variable axes, the categories of the response variables and the subjects’ positions. The interpretation of the biplot uses a distance rule. The MLD model belongs to the family of marginal models for multivariate responses, as opposed to latent variable models and conditionally specified models. By setting the distance between the two categories of every response variable to be equal, the MLD model becomes equivalent to a marginal model for multivariate binary data estimated using a GEE method. In that case the MLD model can be fitted using existing statistical packages with a GEE procedure, e.g., the genmod procedure from SAS or the geepack package from R. Without the equality constraint, the MLD model is a general model which can be fitted by its own right. We applied the proposed model to empirical data to illustrate its advantages.
ISSN:0176-4268
1432-1343
DOI:10.1007/s00357-018-9251-4