The General Monotone Model (GeMM) is a statistical algorithm for predicting rank orders from a set of k predictors. As shown in Dougherty and Thomas (2012, Psychological Review), GeMM is unaffected by any monotonic transformation of the criterion variable, unaffected by non-linear monotone relationships, unaffected by violations of normality of errors, and shows better power and predictive accuracy than Least-Squares regresison procedures in a variety of real-world contexts. GeMM is mathematically equivalent of the Maximum Rank Correlation estimator developed by Han (1987), which has been shown to be general regression procedure that encompasses ordinary least squares, Box-Cox regression, and logistic regression, amongst many others, as special cases. Whereas Least-Squares procedures can yield invalid conclusions when applied to messy data that includes a small number of outliers, our work shows that GeMM is robust to outliers.

• Article comparing GeMM to various Least Squares alternatives on real and simulated data (Dougherty & Thomas, 2012, Psychological Review)
• Article showing GeMM's robustness to outliers comparison to OLS, Robust LS, and GLM (Dougherty, Thomas, Brown, Chrabaszcz, & Tidwell, 2015, Sociological Methodology)

Link to data and R code for simulations

• Article describing GeMM with Order Constrained Linear Optimization and equivalance with Han's (1987) Maximum Rank Correlation estimator (forthcoming)
• Article describing extention of GeMM that can accomodate non-monotonicities (forthcoming)
• Non-technical article describing why GeMM works so damm well (forthcoming)

R Package for GeMM

SAS code GeMM

Download Matlab-GeMM code (this is dated and slow -- please use the R package)

Download Matlab-GeMM tutorial

Contact

Michael R. Dougherty, PhD
Professor of Psychology
Email: mdougher@.umd.edu
Office: Biology-Psychology Building, 1145C
Phone: (301) 405-8423