Purposeful selection of variables in logistic regression

Purposeful selection of variables in logistic regression

16 December 2008 | Zoran Bursac*, C Heath Gauss, David Keith Williams, David W Hosmer
This article introduces an algorithm for purposeful selection of variables in logistic regression, which automates the process described by Hosmer and Lemeshow. The authors compare this algorithm with three well-documented variable selection procedures (FORWARD, BACKWARD, and STEPWISE) using a simulation study. The simulation studies evaluate the performance of the purposeful selection algorithm under various conditions, including sample size, confounding, and non-candidate inclusion levels. The results show that the purposeful selection algorithm performs well, especially in retaining important confounding variables, leading to a potentially richer model. The algorithm is further illustrated using the Worcester Heart Attack Study (WHAS) data. The authors conclude that the purposeful selection algorithm is a valuable tool for analysts who need to retain both significant and confounding covariates in their models. The article also discusses the limitations of the algorithm and provides recommendations for its use.This article introduces an algorithm for purposeful selection of variables in logistic regression, which automates the process described by Hosmer and Lemeshow. The authors compare this algorithm with three well-documented variable selection procedures (FORWARD, BACKWARD, and STEPWISE) using a simulation study. The simulation studies evaluate the performance of the purposeful selection algorithm under various conditions, including sample size, confounding, and non-candidate inclusion levels. The results show that the purposeful selection algorithm performs well, especially in retaining important confounding variables, leading to a potentially richer model. The algorithm is further illustrated using the Worcester Heart Attack Study (WHAS) data. The authors conclude that the purposeful selection algorithm is a valuable tool for analysts who need to retain both significant and confounding covariates in their models. The article also discusses the limitations of the algorithm and provides recommendations for its use.
Reach us at info@study.space