Title: From psychometric research to implementation and back: selected
examples

Author: David Magis

Affiliation: University of Li‘ge, Belgium

Abstract: Current psychometric research is most often supported by computer
software.  New research perspectives often imply intensive simulation
studies to validate the tested theories or hypotheses, and therefore
require accurate implementation as e.g., R packages.  However, it may
happen that unexpected psychometric phenomena are detected almost
accidentally, through such implementations with basically totally different
purposes.  This talk will illustrate this phenomenon by means of two recent
examples from item response theory (IRT) with polytomous models: (a) the
equivalence between the weighted likelihood estimator (WLE) and the Bayes
modal estimator with Jeffreys prior, and (b) the relationships between
observed and expected information functions.  Rather than focusing on the
technical details, the purpose of this talk is to highlight how the results
were identified first through R implementation, then confirmed by
theoretical derivations.  The talk concludes by advocating for flexible and
stable open-source implementations (such as R packages) to support current
and ongoing psychometric research.