Comparing reflective and formative measures: New insights from relevant simulations

Woojung Chang, George R. Franke, Nick Lee

Research output: Contribution to journalArticlepeer-review

24 Scopus citations

Abstract

Previous simulations comparing formative and reflective models specify formative population models as the only correct model for a given construct, and compare them with various mis-specified reflective models. However, this approach does not generalize to situations where both reflective and formative specifications can work well to assess constructs. To address this limitation, this study presents simulations in which both formative and reflective specifications fit the underlying population data equally well. The results show that reflective specifications generate less biased and more powerful results than formative specifications, and make a strong case for considering standardized rather than unstandardized coefficients for both specifications. Therefore, conceptual and empirical consequences of using reflective models for constructs that could also be modeled as formative are less dire than past research has suggested.

Original languageEnglish
Pages (from-to)3177-3185
Number of pages9
JournalJournal of Business Research
Volume69
Issue number8
DOIs
StatePublished - 1 Aug 2016

Keywords

  • Construct specification
  • Formative measurement
  • Monte Carlo simulation
  • Reflective measurement

Fingerprint

Dive into the research topics of 'Comparing reflective and formative measures: New insights from relevant simulations'. Together they form a unique fingerprint.

Cite this