Create Robust Linear Models Using Generalized Linear Regression (GLR)
Aug 13, 2015 6:45 AM
Scott Wise, JMP Principal Systems Engineer, SAS Brady Brady, JMP Technical Enablement Engineer, SAS
Creating Robust Linear Models Using Generalized Linear Regression (GLR)
Penalized regression techniques are a viable alternative to traditional linear modeling approaches, particularly in cases where the typical assumptions of orthogonality, homoscedasticity or response normality are violated. They can also be used to automatically select which predictors are most useful in predicting a given response. While these techniques are not new, their use in applied settings has historically been limited, due both to their computational expense and their relative obscurity outside of academia. Today, the proliferation of affordable, powerful computers makes it feasible to implement these techniques in desktop analytic software. This session will explain penalized regression techniques and explore the use of the Generalized Regression personality of the JMP Fit Model platform. Attendees will find that these models are easy to set up, run and analyze, yet are robust in the presence of outliers, correlated inputs, and non-normal and/or zero-inflated responses.