Variable length testing using the ordinal regression model.

Health questionnaires are often built up from sets of questions that are totaled to obtain a sum score. An important consideration in designing questionnaires is to minimize respondent burden. An increasingly popular method for efficient measurement is computerized adaptive testing; unfortunately, many health questionnaires do not meet the requirements for this method. In this paper, a new sequential method for efficiently obtaining sum scores via the computer is introduced, which does not have such requirements and is based on the ordinal regression model. In the assessment, future scores are predicted from past responses, and when an acceptable level of uncertainty is achieved, the procedure is terminated. Two simulation studies were performed to illustrate the usefulness of the procedure. The first used artificially generated symptom scores, and the second was a post hoc simulation using real responses on the Center for Epidemiologic Studies Depression scale. In both studies, the sequential method substantially reduced the respondent burden while maintaining a high sum score quality. Benefits and limitations of this new methodology are discussed.