NSF Award
2120174
This research project will develop methods for longitudinal and multidimensional models with partially ordered sets of responses or posets. Precise and valid measurement of psychological and social constructs are key to progress in the social sciences. However, many psychological constructs such as personality, self-efficacy, and emotional intelligence are not directly observed. These constructs are measured indirectly through questionnaire instruments with a designated response format, such as multiple-choice questions. Scores are assigned to individuals in a way that communicates quantitative information about the construct. In many measurement situations, however, scoring is a challenge and there may be no best answer. The resulting responses often result in data that are a mix of ranked responses and responses that cannot be ranked. Such a structure forms a partially ordered set or poset. This project will advance measurement science by including poset response as a new category of response format. Publicly available software will be developed. The project will train and mentor graduate students with the support of an established psychometric graduate program at the University of North Carolina. Collaborative activities with the Educational Testing Service will broaden the impact of this project.<br/><br/>This research project will develop a flexible class of measurement and predictive models for poset responses within complex settings that include multivariate, multidimensional, and longitudinal data. Because of the mix of different measurement scales, posets are notoriously challenging to model. This has been reflected in the predominantly non-model-based methods that have so far been used for handling posets; e.g., collapsing categories or summarizing by weighted means. The project's approach to poset measurement will be based on the theory of latent variable modeling. This method will enable researchers to test and falsify the model to advance the science. The poset model will allow measurement errors to be quantified, a feature that is often lacking in other methods. The model will be validated using simulation experiments and real-world applications across a broad range of areas, including cognitive tests, posets derived from latent class or cluster analyses, situational judgment tests, and attitudinal surveys.<br/><br/>This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
