Using Kendall test to find correlation between Likert type data


I am doing a small experiment and I was wondering if I have the correct method for analysing the data. Two groups of participants (speakers and listeners) comment on a description of a famous person.

The speaker creates a description and answers the question: "I think a general reader will know who I mean" using Likert scale (Strongly agree, Agree, Neither agree nor disagree, Disagree, Strongly disagree).

The listeners get the description and after guessing the name of the famous person they answer the question: "I am sure I know who this description refers to." using the same scale.

For simplicity I refer to the questions as speaker/listener confidence. After looking at the methods of analysing Likert items (ordinal data) I concluded that I can use Mann-Whitney U test to check for difference in the confidence of speakers and the confidence of listeners.

Some more searching on the Internet suggests that the Kendall test could be suitable for finding any correlation between the two confidences. Is this valid method or should I keep looking?

Also Kendall requires the two sets to be the same length but more than one listener can get the same description. Does it make sense to pair the values by duplicating the speakers confidence? e.g. speaker's rating for the confidence is 4 and three listeners provided values 4, 4, 5 so my dataset would be
4 4
4 4
4 5

Last question: is the Kendall test still valid when there are lots of ties? With 500 descriptions and only 5 possible values (the Likert scale) there will be an extraordinary amount of ties. Especially when one considers that most of the answers will be skewed towards high confidence due to the nature of the task.

Thank you for your attention. I appreciate any comments on the data analysis.

Last edited: