Improving measurement models in clinical epidemiology: time to move beyond the inherent assumption of an underlying reflective measurement model

Abstract

Objectives: The nature of a construct's measurement model, most decisively being predominantly reflective or formative, is essential for its development, validation, and use. Differentiating between these types of measurement models cannot be done based on statistics alone, but has to rely on expert judgment, preferably guided by checklists and theoretical assumptions. However, consideration and substantiation of the choices of the measurement models is lacking in most studies describing the validation of measurement instruments in the field of clinical epidemiology.

Study design and setting: A convenience sample of 96 clinimetric studies, published from 2017 up until May 17th, 2018 was scored on model use and (mis)specification.

Results: In over 50% of the identified studies in this sample, formative measurement models are considered and/or analyzed as reflective.

Conclusion: Misspecification of formative measurement models as reflective was found to be more rule than exception. It is therefore recommended that model selection and considerations on the theoretical nature of the measurement model should be classified, motivated, and discussed, for example, by using available checklists. Hereby, it can be ensured that the appropriate measurement models and corresponding statistics are used.

Keywords: Checklist; Clinimetrics; Formative constructs; Measurement models; Patient reported outcomes; Psychometrics; Reflective constructs; Validation.