Predicting Primary Flight Grades by Averaging Over Linear Regression Models: Part 2
Abstract
Linear regression models are commonly used to predict some criterion variable of interest. Here, the criterion variable of interest is a grade assigned to Navy and Marine Corps student aviators at the end of Primary flight training. It would be of value to be able to predict such a flight grade based on prior existing information. The prior information is inserted into a linear regression model through some number of predictor variables. We are currently focusing our efforts on predictor variables arising from scores on the Aviation Selection Test Battery (ASTB), a final grade from Aviation Pre-Flight Indoctrination (API), and performance on a psychomotor test battery. We then calculated the posterior probability of all possible linear regression models. Some models had an insignificant posterior probability and we may safely ignore them in any averaging over models. The central result of such an exhaustive analysis is the identification of all those models, together with their constituent predictor variables and regression coefficients, that do, in fact, possess a significant posterior probability. The predictions of flight grades for any given values of the predictor variables can then be averaged over these models. A major advantage of this approach is the assessment of the variability of the predictions. Of more immediate concern, these results can be incorporated into the Pilot Prediction System (PPS) at the Naval Aerospace Medical Research Laboratory. The PPS is used to make statistical inferences about the training outcome of a student in flight training given information from a data base. We routinely use the PPS to make predictions about the probability of failure in some post-API phase of training given scores from the ASTB and API. The results from this paper could be inserted into the PPS to make predictions about the flight grade in Primary training.
Document Details
- Document Type
- Technical Report
- Publication Date
- Aug 01, 2000
- Accession Number
- ADA531714
Entities
People
- A. O. Albert
- D. J. Blower
- H. P. Williams
Organizations
- Naval Aerospace Medical Research Laboratory