Introduction to Subgroup Analysis

In Delphi studies, you often work with different groups of experts. These groups can have diverse backgrounds and perspectives:

• Clinicians vs. tech developers
• Policy makers vs. industry leaders
• Optimists vs. pessimists

Comparing these groups is essential to understand whether their ratings really differ in a meaningful way.

Why Use the Mann-Whitney U-Test?

The Mann-Whitney U-Test is one of the most popular methods for advanced subgroup comparison. Here's why it's widely used:

Non-parametric:

It doesn't assume your data are normally distributed – which is perfect for Delphi surveys that rely on Likert scales (e.g., 1–5 ratings).

Robust with small samples:

Even with modest participant numbers in each group, you can still test for differences reliably.

Clear interpretation:

The test shows whether the distribution of scores in one group is systematically higher or lower than in the other.

How the Mann-Whitney U-Test Works

In simple terms:

The Mann-Whitney U-Test compares whether the scores of two independent groups come from the same distribution.

Instead of comparing means, it compares ranks:

All ratings are pooled, sorted, and assigned a rank. Then, the test calculates whether the ranks in one group tend to be higher or lower than the other group.

Because it uses ranks, it is very robust against outliers and skewed data.

Example Scenario

Let's look at a concrete example:

Research Question:

Do AI experts and clinicians rate the likelihood of "autonomous diagnostics" adoption equally?

Data:

• AI Experts: Ratings of 4, 5, 5, 4, 5
• Clinicians: Ratings of 2, 3, 3, 2, 3

Interpretation without a test:

At first glance, it seems AI experts are more optimistic.

Why test it statistically?

Because you want to know:

• Is this difference significant, or could it just be due to chance?
• How strong is the evidence that the groups really disagree?

How the Mann-Whitney U-Test helps:

• You rank all ratings from lowest to highest.
• You calculate the U-statistic.
• You check whether the probability of observing this difference by chance is below your threshold (commonly p < 0.05).

Result:

If p < 0.05, you conclude:

The difference is statistically significant. This means the two groups really do rate differently.

Why Not Use a t-Test?

Many people know the t-test, but it requires normally distributed interval data. Delphi data are usually ordinal (Likert scales) and skewed, which makes the t-test less appropriate.

The Mann-Whitney U-Test is better suited for Delphi research because it:

• Works with ordinal scales
• Tolerates non-normal distributions
• Is robust against outliers

Example Interpretation

Imagine your test output:

• U = 2.0
• p = 0.01

You would conclude:

AI experts rate autonomous diagnostics significantly higher than clinicians (Mann-Whitney U = 2.0, p = 0.01). This supports the interpretation that professional background influences expectations about this technology.

Conclusion

The Mann-Whitney U-Test is a powerful tool for subgroup analysis in Delphi studies. Durvey.org helps you:

• Detect real differences between expert groups
• Strengthen your findings with statistical evidence
• Communicate your results confidently

By combining qualitative insights with rigorous statistical tests, you make your Delphi research more credible, transparent, and actionable.