Each week, we publish layman’s abstracts of new articles from our prestigious portfolio of journals in statistics. The aim is to highlight the latest research to a broader audience in an accessible format.
A Bayesian nonparametric approach to dynamic item-response modeling: An application to the GUSTO cohort study. Statistics in Medicine. 2021; 40( 27): 6021– 6037. https://doi.org/10.1002/sim.9167, , , , , .
Psychometric questionnaires are are common tools used used for the quantification of mental health status. These questionnaires are usually composed of specific questions aimed at measuring certain symptoms of mental health in the subjects, such as anxiety or depression. The data collected from these questionnaires are characterized by their ordinal nature, and are often analysed using techniques from item response theory, also known as latent response theory. This framework involves statistical models for the analysis of the relationship between latent traits (i.e., the mental health status) and their outcomes (i.e., the answers to the questionnaires). With these models, it is possible to differentiate respondent profiles and to characterize the questions(items) included in the questionnaire via interpretable parameters. Despite being widely used, these models present some limitations. For instance, they are often used to evaluate the performance of the subjects, rather than other aspects such as mental health. They may also be employed in a cross-sectional manner, i.e. by analysing data collected at different time points independently. The motivating application of this work is the analysis of psychometric questionnaires taken by a group of mother sat different time points and by their children at a later time point. The data, avail-able through the Growing Up in Singapore Towards healthy Outcomes (GUSTO)cohort study, present complex features such as the dependence over time and among mother/child pairs. To flexibly model these data, a Bayesian semiparametric model is proposed which extends the current literature by: (i) introducing temporal dependence among questionnaires taken at different time points; (ii) jointly modelling the responses to questionnaires taken from different, but related, groups of subjects (in this case mothers and children), introducing a further dependence structure and there-fore sharing of information; (iii) allowing clustering of subjects based on the way the subjects answer the questions via the implementation of techniques from Bayesian nonparametrics. When applied to the GUSTO data, the proposed model is able to identify three main groups of mother/child pairs, characterised by specific answering profiles. An interesting outcome of this study is the identification of a maternal reporting bias effect strongly affecting the clustering structure of the mother/child dyads, suggesting the presence of differences in the mother and child’s perception of behaviour.