Computational Psychometrics: New Methodologies for a New Generation of Digital Learning and Assessment
Esther Ulitzsch · Psychometrika · 2022
Innovations in technology-enhanced learning and assessment systems have evolved rapidly in recent years.Facilitating the (real-time) collection of process data that provide detailed documentation of how test takers and learners engage with the administered tasks-such as keystrokes, mouse movements, clickstreams, or audio traces of subjects interacting with each other-technologyenhanced learning and assessment systems entail vast opportunities.These range from a deepened understanding of how persons interact with the administered tasks through the assessment of capabilities that manifest themselves more in the process than in the final outcome-such as problem-solving strategies, collaboration, or inquiry-to the possibility to dynamically adapt the systems as well as ones' beliefs on the targeted capabilities as test takers' and learners' interactions with the tasks evolve.How evidence on test takers' and learners' capabilities can be extracted and synthesized from this usually enormous and possibly unstructured amount of data, however, is not self-evident.In fact, many characteristics of process data violate the very assumptions of traditional psychometric models, such as unidimensionality or local independence.To accommodate the sequential and unstructured nature of process data and leverage its potential, the computational psychometrics paradigm calls for enriching theory-grounded psychometrics with stochastic process theory and data-driven techniques originating in computer science (von Davier, 2017).The volume "Computational