Mokken Scaling: A Nonparametric Route to Stronger Survey Measurement

Why Survey Measurement Matters

Wijbrandt van Schuur addresses a common challenge in survey-based political research: how to measure an underlying ordinal trait (for example, attitudes or values) with multiple items when strict parametric assumptions may not hold. The article revisits the Guttman tradition and places Mokken scale analysis as a probabilistic, nonparametric alternative grounded in Item Response Theory (IRT).

What Mokken Scaling Is

Mokken scaling is presented as an ordinal, unidimensional measurement framework built on two nonparametric models: Monotone Homogeneity and Double Monotonicity. These models relax the stronger distributional and functional-form assumptions of parametric IRT (notably the Rasch/one-parameter logistic model) while retaining probabilistic interpretations of item ordering and respondent latent traits.

How the Article Compares Methods

• The paper contrasts Mokken scaling with Classical Test Theory approaches (reliability and factor analysis) and with parametric IRT, highlighting differences in assumptions and inference.
• Procedures are given for applying Mokken analysis to both dichotomous (yes/no) and polytomous (multi-category ordinal) survey items, showing how scalability and item ordering can be assessed without specifying a logit link or item characteristic curve.

Empirical Illustration

• Two scale analyses of items from the World Values Survey are used to demonstrate the approach in practice. These examples show how researchers can use Mokken procedures to evaluate item sets, detect departures from monotonicity, and build ordinal scales appropriate for substantive analysis.

What This Means for Political and Survey Researchers

Mokken scale analysis provides a practical middle ground: it keeps the conceptual strengths of IRT (latent traits, item ordering) but imposes fewer parametric constraints than models like Rasch. For political scientists working with ordinal survey data—especially public-opinion and values measures—Mokken methods offer a transparent, assumption-light toolkit for scale development and validation that complements both factor analysis and parametric IRT approaches.