ESTIMATION TECHNIQUES FOR DEPENDENT LOGIT MODELS
Abstract
The report is a discussion of the estimation techniques of logit models whose response variables are not necessarily statistically independent. The techniques developed and used include the Generalized Least Squares approach of Zellner; stepwise regression and its relevance in estimating logit model parameters; and the 'contingency table' concept of fixed marginal probabilities, which is used to develop a bivariate logit model that possesses certain statistical and probabilistic properties concerning the correlation coefficient and the summation of individual cell probabilities to unity. The techniques are designed for application to two Air Force problems: The first problem uses concomitant information to estimate the probability that first-term airmen reenlist; the second, information concerning flying missions to estimate the probabilities that interdependent aircraft subsystems fail during these missions. Some applications to the problems are presented, but this study is mainly directed toward development of the necessary techniques. An example illustrating the various techniques is given.
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 01, 1968
- Accession Number
- AD0679780
Entities
People
- A. A. Cook Jr.
- A. J. Gross
Organizations
- RAND Corporation