What it’s about

In this paper, we examine whether very short financial literacy tests can be improved by modeling not only whether responses are correct, but also how respondents answer when they are incorrect or uncertain.

Abstract

Very brief financial literacy tests are often used in large surveys, but conventional right/wrong scoring can limit their reliability and validity. Using data from the Dutch LISS panel, this study compares standard binary item-response models with models that recover information from distractors and non-substantive response options such as “don’t know” and “would rather not say.” Particular attention is given to the nested logit model, which preserves the distinction between correct and non-correct responses while allowing non-correct options to carry additional information. Results suggest that this approach improves reliability, especially at low to average levels of financial literacy, and yields scores that are more strongly associated with external validity criteria. The paper shows how richer modeling of response patterns can improve measurement in short survey-based knowledge tests.