PHD Discussions Logo

Ask, Learn and Accelerate in your PhD Research

Question Icon Post Your Answer

Question Icon

What non-parametric or rank-based statistical tests are most robust for assessing the association or monotonic relationship between two ordinal-scaled variables, considering assumptions of distribution and measurement level?

In my survey research, both my key variables like "satisfaction level" and "willingness to recommend" are measured on ordinal Likert scales. Given the uncertainty about the underlying distribution and the fact that the intervals between scale points aren't guaranteed to be equal, I want to avoid parametric tests like Pearson correlation and use a more distribution-free approach.

 

All Answers (1 Answers In All)

By Taylor Answered 9 months ago

When dealing with two ordinal variables, I consistently recommend rank-based measures over parametric ones. Spearman's rho is a strong, widely understood choice that assesses monotonic association. However, in my work with survey data, I often prefer Kendall's tau-b, especially with smaller sample sizes or when you have many tied ranks on your Likert scales it tends to give a more conservative and, in my view, a more realistic estimate of the association's strength. Both are robust to non-normality. Your key assumption is that the data can be reliably ranked, which is perfect for your Likert items. Always report the coefficient alongside its confidence interval.

Your Answer