Research for Computational Methods

In computerized adaptive testing, an examinee's ability is often estimated using Bayesian procedures. Its calculation is greatly simplified if one uses Owen's approximation. However, when the test length is moderate (e.g., 15 items), the accuracy of Owen's approximation is inadequate for calculating the examinee's score. This is illustrated using the nine power subtests in the Armed Services Vocational Aptitude Battery. The recommended procedure is to use Owen's approximation for adaptive item selection, and to compute the exact Bayesian estimate after the entire test has been administered.

The rate of productivity growth in the U.S. economy and the extent and effects of government regulation have been matters of great concern in recent years. This study is an empirical investigation of the effect of regulation on productivity in the auto and steel industries during the period 1958-1980. The first step in the study was to develop numerical measures of regulations. These measures were then employed in two empirical models. In the single-equation model, the level or rate of growth of productivity was estimated as a function of output, technology, regulation, and other variables. In the multiequation model, cost and input cost shares were jointly estimated as a function of input prices, output, technology, and regulation. The estimate effect of regulation on productivity varied, depending on the model and the measure of regulation used. For example, the multiequation model indicated a positive effect of regulation on productivity in the steel industry. For the auto industry, however, it indicated that regulation generally had the expected negative effect on productivity. Over the period 1973-1980, regulation was estimated to have lowered the annual rate of productivity growth in the auto industry by 0.55 to 2.00 percentage points, depending on the measures of output and regulation. For Additional Information See 02 052702 00.

This paper discusses the structure and theory of the algorithm. The discussion focuses on the features that differ from a standard sequential quadratic programming algorithm. Part II describes the actual implementation of the algorithm and presents test results for several problems.

This paper presents some mathematical methods for evaluating the performance of a distributed data base system (DDBS).