preprints.org > computer science and mathematics > probability and statistics > doi: 10.20944/preprints202404.1606.v1

Preprint Article Version 1 Preserved in Portico This version is not peer-reviewed

Version 1 : Received: 23 April 2024 / Approved: 24 April 2024 / Online: 24 April 2024 (14:06:37 CEST)

Regression models are highly popular in empirical research and come in many different forms to fit virtually any distribution, variable, or research question. Usually, these models also compute how much variation in the outcome variable can be explained by all predictors, which is relevant to understanding whether the predictors are, in sum, able to explain the outcome or whether other potentially unobserved factors are more relevant. This aspect is crucial for interventions and policy as even statistically significant regression coefficients can be meaningless if they have little influence on the outcome overall. Besides having a measurement to judge the overall goodness of fit, ranking predictors by their relative importance is also relevant. For such analyses, it is necessary to decompose the total explanatory power of a model and partition it so that each explanatory variable is assigned a share. This enables the computation of a predictor variable's absolute and relative influence. Dominance analysis is a statistical approach to achieve this goal.

dominance analysis; regression; variance decomposition

Computer Science and Mathematics, Probability and Statistics

* All users must log in before leaving a comment