ABSTRACT
This chapter introduces the principal features of a bootstrap validation method for the ever-popular logistic regression model (LRM). The standard use of the Hosmer–Lemeshow (HL) test, confirming a fitted LRM is the correct model, is not reliable because it does not distinguish between nonlinearity and noise in checking the model fit. Also, the HL test is sensitive to multiple datasets. The bootstrap validation method provides a measure of sensitivity of LRM predictions to changes in the data and a procedure for selecting the best LRM if it exists. Bootstrap samples are randomly different by way of sampling with replacement from the original data. The model builder performs repetitions of logistic regression modeling based on various bootstrap samples. The chapter uses the bootstrap validation method to yield reliable and robust results along with the final LRM tested on a fresh dataset.
