Multinomial Probit Models of Military Enlistments: A Comparison of Alternative Solution Algorithms

The increasing popularity of qualitative response models in recent years has led to the development of sophisticated algorithms to solve them. The relative efficiency and accuracy of various algorithms is a subject of considerable importance to the Army, because the results derived from probit models may be used increasingly by Army decision makers. This paper compares the accuracy and cost of several types of algorithms used to model the decision whether or not to enlist in the military. The paper concludes that econometricians who undertake analyses using large databases for cross-sectional studies will gain very little by using some of the recently developed algorithms such as BHHH. These new methods represent elegant and impressive theoretical achievements, but in practice they do not justify their enormous cost.