Psychometric and Machine Learning Approaches to Reduce the Length of Scales
Oscar Gonzalez, 2021
Multivariate Behavioral Research
citation: 12
Paragragh 1: An important aim in the health and social sciences is to minimize respondent burden in questionnaire research (Gibbons et al., 2008)
Paragragh 2: However, the potential advantages of giving respondents fewer items might not offset the disadvantages.
Paragragh 3: Two common approaches to give fewer items to respondents are to administer either a short-form or a tailored test. … However, Selecting items from a longer item set could be treated as a feature selection task, which is a common machine learning problem(Kuhn & Johnson, 2013)
Paragragh 4: The purpose of this paper is to review and provide a pedagogical illustration of pscychometric and machine learning apporaches to select fewer items and estimate a score on a unidimensional scale. In this paper, scales that estimate scores on reflective constructs are discussed, as opposed to scales primarily used for classification. Challenges to select items from scales that assess formative constructs are considered in the supplementary materials.
Paragragh 1:For the purposes of illustrating the statistical procedures to select items, it is assumed that researchers have met the previous methodological considerations – mainly that the measure is relevant to the specific population to which it is administered and that we have access to a computer that can administer the tailored test.
→ This paper assumes that researchers already know the target population of the test and the examinee can access computing environments.
Paragragh 1: … A modern psychometric apporach to build static short-forms or adaptive test is based on item information functions estimated with item response models (Edelen & Reeve, 2007)
→ IIF of IRT is used to build short-form and adaptive test.
Paragragh 1: … The GRM is appropriate for Likert-type items commonly encountered in psychology.