Linear Transformations for Multivariate Binary Data
Peter Bloomfield · Biometrics · 1974
The interpretatioii of statistical data may often be simplified by a preliminary transformatioii. In the conitext of contingency tables, one way of achieving this would be to relabel the possible outcomes, or in other words to permute the cells of the table. For a 2d table, certaiui permuitations have the property that a loglinear model for the cell probabilities transforms in a simple way. These are, in a sense, linear transformations of the original variables. The aim of making such a transformation is to fit the transformed data by a simple model, such as a low-order hierarchical model or one in which certain variables are inidependent of others. A 24 table has beeni analyzed with this end in view. All the models were fitted to the original data, anid to do this a computer program has been developed which will fit iionhierarchical models by iterative scalinig.