Data Scientist Salt Lake Community College, United States
Session Abstract: It is easy to build regression models where the continuous predictors have a linear relationship to the outcome variable. However, it sometimes happens that an important variable has a NONlinear relationship to some outcome of interest. A student's age, for example, may be related to academic outcomes in complex ways: students under 18 are different from students of traditional college age, who are different from working adults, who are different from students near or past retirement. In situations like these, neither simple linear regression nor a standard transformation (e.g., taking the log of the predictor variable) provides an adequate model. This session presents four techniques for handling nonlinear predictors like age: binning, piecewise linear regression, polynomial regression, and splines. For each approach, we will discuss its mathematical basis, its pros and cons, and how the technique can be implemented in R.
