LAUSR

We issued the call for the set of papers published in this special section in 2019 with the goal to illuminate innovative state-of-the-art methods for nonparametric measurement. The original idea for this call goes back to a series of potential topics for methodological special sections in our journal collated by our previous Co-Editor in Chief, Frans J. Oort, which we discussed and developed further in our editorial meetings at the International Society for Quality of Life Research (ISOQOL) Annual Conferences in Philadelphia (2017) and Dublin (2018). And I am very grateful to our two excellent guest editors, Klaas Sijtsma and Andries van der Ark, who helped us to finalize the call and who facilitated the project. We received 13 expressions of interest and six full papers were submitted, of which five are published in this special section. While parametric item response models are widely used (for example more than 7007 search results in PubMed titles and abstracts at the date of writing; 264 in Quality of Life Research), a search for terms specifically related to nonparametric item response models reveals only 392 results (15 in Quality of Life Research, including the papers of this issue). Nonparametric item response models have been around at least as long as the parametric model family as for example seen in Guttman’s work on scalogram analysis [1]. But owing to their versatility and solutions for practical problems, parametric models were the main focus of research and application until a renewed interest in the 1980s [2]. This may explain some of the differences in uptake, but originally limited options for the modelling of polytomous responses [3], delayed development of a minimal set of standard analyses and that the approaches may have appeared less “canonized” and were less accessible than their parametric alternatives [4, 5] may have contributed as well. Today, nonparametric models offer a theoretical framework to study human response processes, to explore the underlying assumptions of item response models, and procedures to explore data quality [2, 6]. While the last years have seen a swiftly increasing popularity of nonparametric item response theory (NP-IRT) to investigate health-related quality of life (HRQL) data, applications remain limited and the full potential of these models has not been explored for our field. The introductory paper by the guest editors [7] starts with a mention of previous applications and moves on to provide an overview of connections to parametric models and a series of steps required for applying NP-IRT. The steps are illustrated with a worked example and the text provides references to more detailed introductions and tutorials. The authors illustrate well that NP-IRT models are flexible and that this could be a distinct advantage when modeling assessments of HRQL and patient-reported outcome data, where our theories may not yet have matured enough to justify the choice of particular parametric models. The section starts with the application of Mokken scale analysis to the EQ-5D-5L using data from people living with and without chronic conditions in six western countries [8]. The paper firstly demonstrates the use of the methods that are the focus of the special section with data relevant to our field. But it secondly also asks an important question about whether there is potential in analyzing a measure that derives its validity mainly from valuation studies estimating preference-based scoring weights with methods that could in contrast potentially support the use of a summed score for further analysis. The result is favorable in this application, but with respect to the anxiety/depression dimension the results underscore the importance of reference samples when evaluating psychometric properties. The team suggests that this item might need to be evaluated separately. The authors discuss results and approaches in detail, and they highlight that the ultimate decision about scoring methods needs to be based in a theoretical framework: Positive results obtained with a psychometric procedure do not provide an answer what a score “means”. The paper is accompanied with a detailed set of R syntaxes. * Jan R. Boehnke [email protected]