Description
Correlate two rating-scale items by treating each response as a cut of a continuous latent attitude, with thresholds showing how the scale points spread.
Short rating scales restrict the values Pearson's r can take, so r understates and blurs association between items. The polychoric correlation assumes each respondent holds a continuous, normally distributed value and picks the scale point whose interval contains it. It estimates the correlation between the two latent variables and the threshold (tau) cut points between adjacent scale points; for binary items it reduces to the tetrachoric correlation. It is read like r, and the thresholds show whether categories discriminate across the latent range or bunch together. A matrix of polychoric correlations is the appropriate input for factor analysis of Likert-type items. Chapman and Feit show the estimate exceeding Pearson's r for two five-point satisfaction items (0.62 against 0.59) and recommend a jittered scatterplot first because ordinal pairs overplot.
Attributes
- Discipline
- Survey & Psychometrics
- Loss type
- Latent
- Stage
- Evaluation, Definition
- Data
- Quantitative
- Effort
- Hours
Steps
- Plot the two items against each other with jitter to see the joint distribution.
- Keep respondents who answered both items.
- Compute Pearson's r for reference.
- Compute the polychoric correlation (for example psych::polychoric in R) for the pair or for a set of items.
- Interpret the coefficient as you would r.
- Inspect each item's thresholds to judge whether scale points spread evenly across the latent continuum, and watch for warnings from sparse cells.
Sources
- R for Marketing Research and Analytics (Chapman & Feit, 2019)
- A Beginner's Guide to Structural Equation Modeling (Schumacker & Lomax, 2004)
- Structural Equation Modeling: A Bayesian Approach (Lee, 2007)
- Continuous Multivariate Distributions, Vol. 1 (Kotz et al., 2000)
Also known as
- Polychoric correlation
- Latent correlation of ordinal items
- Tetrachoric correlation (binary case)
Categories
- Analytics & Behaviour
- Survey & Psychometrics