Empirical Comparison of Criterion Referenced Measurement Models
Abstract
The Army needs information about how well an individual can perform the tasks necessary for him to do his job. This information is often gathered by means of a 'criterion-referenced test,' a test made up of items directly related to the job of interest. The test results can be used in two ways. The first way is to sort individuals into two groups, one made up of those who can perform their job satisfactorily and the other made up of those who do not meet minimal job requirements. A second use of the test results is to estimate the 'true' capability of the examinees to do the task being tested. These two uses are clearly related. Several psychometric models are available for grouping the individuals and/or for estimating 'true' scores. For example, one may simply calculate the proportion of items correctly answered and use that proportion as an estimate of 'true' capability. Alternatively, a binomial error model for deriving the expression for the regression of 'true' score on observed score can be used and a 'true' score calculated for each individual. Other possible models include a Bayesian Model II approach and a latent trait model such as the Rasch one parameter logistic model. Each of these models yields a somewhat different estimate of 'true' capability for any given individual.
Document Details
- Document Type
- Technical Report
- Publication Date
- Oct 01, 1976
- Accession Number
- ADA076822
Entities
People
- Frederick H. Steinheiser Jr.
- Kenneth I. Epstein
Organizations
- U.S. Army Research Institute for the Behavioral and Social Sciences