# Likert Analysis - Dunn's Test (Stata)

Similar to previous questions but I'm still a little confused as to whether I've followed the right process/understood the results correctly.

For a research project I'm conducting I used Likert scales for questions and then calculated the weighted average for each group. After doing this I wanted to see if there was a stasticaly significant difference in the groups. To do so I conducted a Kruskal-Wallis test that included each of the questions.

One question did have a p-value of 0.02 and so I wanted to look into that further to see if one group is particularly polarising (or so forth). And so I followed up with dunntest (by Alexis Dinno) in Stata and below is what came up.

From this it looks like group C has some sort of significance but I don't quite know how to say it and to report the p-value (or if this is even needed). I'm just a little lost (sorry if this was a basic question). Would really appreciate any advice/suggestions.

. dunntest Question_3, by(PanelField) ma(none) Results below :

• Note that the list option in dunntest may give output that is a little easier to follow than the tabled output you show above if you are not accustomed to such tables. Commented Nov 9, 2019 at 3:23
• Ah, thank you! I see yeah that does make things a bit easier. I'm still a bit confused though with how to interpret as the p value is actually comparing 2 groups not 1 group vs rest. I could be generic and say overall group C is more likely to be disagree in comparison to the rest of the groups but I'm not sure I can state a specific p-value as such. Commented Nov 9, 2019 at 3:55
• Each pair (e.g., A vs B) gets its own separate p-value. Commented Nov 9, 2019 at 4:27

The comparisons suggest that values from C tend to sit higher than the other three (A,B, D), which are all fairly close together

I strongly suggest you plot the data when interpreting these things - a dotplot for each group with common axis makes sense.

This will make it easier to get the relationships straight

• Thank you so much for your suggestion! I just tried it and the data reads so much better than in the boxplot i had previously :) Commented Nov 9, 2019 at 4:00