A CENTROID IDENTITY AS A ZERO-COST CONSISTENCY CHECK FOR PUBLISHED MULTIVARIATE URODYNAMIC REGRESSION MODELS

Abstract

Multivariate linear models relating a urodynamic outcome to several measured predictors are published as a list of fitted coefficients, usually accompanied by a table of group means and standard deviations and a statement that the model is significant by the Fisher criterion. A reader has no way to refit the model and therefore no way to detect a transcription error, a lost decimal exponent or a mismatch between the estimation sample and the summary table. We show that the published material is in fact sufficient for an exact arithmetic check, and we propose it as a routine step in preparing and reviewing such papers.