Predicting With Logistic or Linear Regression: Will It Make a Difference in Who Is Selected for Pilot Training?

Abstract

Selecting individuals for pilot training is typically accomplished by determining what human characteristics are predictive of training success using ordinary linear regression. Frequently, however. the measure of success is passing or failing training. Some critics suggest that ordinary linear regression is inappropriate for such a dichotomous criterion and recommend using logistic regression. For the practitioner the question becomes: "Will it make a difference in who is selected for pilot training?" A pilot selection database is used to demonstrate that the ranking of pilot training applicants is nearly identical for both methods. The rank correlations between the methods was .9967. Therefore, both linear and logistic regression would select the same applicants. Other considerations favoring the use of linear regression are discussed.