Distance from a Theoretically Ideal Statistical Classification Model Defined as the Number of Additional Equivalent Effects Needed to Obtain Perfect Classification for the Sample

Paul R. Yarnold

Optimal Data Analysis, LLC

A method for computing the distance between an empirically-derived statistical classification model and a corresponding theoretically ideal classification model is described. Use of the distance index to identify and to compare globally optimal classification models, within and between descendent families, is illustrated with an example using ethnicity to parse the incidence of different types of cancer.

View journal article

UniODA vs. Legacy Bivariate Statistical Methodologies

Paul R. Yarnold

Optimal Data Analysis, LLC

Research comparing the use of optimal versus legacy methods for analysis of data representing different experimental designs is on-going. This note discusses bivariate legacy statistical tools for which the alternative use of UniODA has already been demonstrated as an always valid, exact, maximum-accuracy statistical methodology.

View journal article

Evaluating Non-Confounded Association of an Attribute and a Class Variable Using Partial UniODA

Paul R. Yarnold

Optimal Data Analysis, LLC

Partial UniODA is a two-step procedure for: (a) identifying the exact statistical model that explicitly maximizes accuracy (normed against chance) achieved for the sample by using an attribute to classify observations’ actual class categories; while (b) simultaneously “controlling for” (eliminating) the effect of a confounding variable. Step One drops observations correctly classified using the confounder to predict class category: observations in the reduced sample weren’t correctly predicted by the confounder. Step Two investigates the non-confounded relationship underlying attribute and class variable using the reduced sample.

View journal article