A Comparison of Linear and Equipercentile Test Equating Procedures in Large Samples
Abstract
The Armed Services Vocational Aptitude Battery (ASVAB) is used for selection and classification of enlisted personnel. New forms of the ASVAB are developed about every four years, and equated to the odd reference form 8a. The ideal outcome is that, during operational use of the ASVAB, the distribution of standard scores is the same for all forms. Equating for operational use is based on data collected during Initial Operational Test and Evaluation (IOT&E); sample sizes exceed 10,000 per form. Two equating procedures often used by psychometricians are equipercentile and linear. When samples are small, the equipercentile method has large random errors. Linear equating is more stable- that is, it has smaller random error. However, it suffers from bias, i.e., systematic errors at high and/low scores, if the two forms have score distributions with different shapes. Linear equating was used for forms 11, 12, 13 and for all subtests except one in forms 15, 16, and 17. As sample size increases, the superior stability of linear equating becomes less important while its bias remains the same. The question addressed in this paper is whether IOT&E samples are large enough to make equipercentile equating preferable to linear. For equipercentile equating in this study, score frequencies were smoothed by a five-point rolling average and a dogleg was used-i.e., the equating curve below the fifth percentile was replaced by a straight line. Keywords: Statistical analysis, Statistical samples, Scoring.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 01, 1988
- Accession Number
- ADA205878
Entities
People
- D. R. Divgi
Organizations
- Center for Naval Analyses