0
$\begingroup$

I have 2 groups of medical students (50 in each group) that have each used 3 different techniques to undertake a procedure on 12 patients. So, 3,600 data points but accepting that they are not truly independent.

I would like to compare the performance of the two groups of students undertaking each of the 3 techniques. The outcome measures are (1) success as yes/no, (2) time in seconds, and (3) self-reported confidence on a 1-10 scale.

As there are 3 different techniques, I couldn't make a 2x2 table for (1) so used Kruskal-Wallis, which seems to permit comparison of more than 2 groups. My feeling is that I could do the same for the continuous outcome data in (2). However, when I try to read about this online, I understand that ordinal data (e.g. the confidence data in (3)) isn't appropriately analysed using Kruskal-Wallis.

Can anyone

  1. comment on whether I am on the right track with (1) and (2)

  2. and/or suggest an alternative approach to analysing the data in (3).

Improve this question
$\endgroup$
2
  • $\begingroup$ Is your interest in differences between the two groups of students, or between the three techniques, or both? What you would you do with a conclusion that the six things being tested (students$\times$techniques) suggesting they are not all the same? $\endgroup$
    – Henry
    Commented Jul 6, 2021 at 10:50
  • $\begingroup$ I am interested in both comparisons - types of student (who were randomised to the two groups then taught to perform the procedure either remotely or in-person) and types of technique. I would hope to be able to determine whether or not there is evidence to support (1) either in-person or remote training and/or (2) any of the three techniques for completing the procedure. $\endgroup$
    – David
    Commented Jul 6, 2021 at 11:08

1 Answer 1

Reset to default
2
$\begingroup$

Kruskal-Wallis is fine for ordinal Y. The only concern is whether it handles excessive ties accurately enough. You might instead use the proportional odds ordinal logistic model for all 3 outcomes. It reduces to the binary logistic model for (1) and handles arbitrarily many ties. It also leads to a more insightful Bayesian analysis.

Improve this answer
$\endgroup$
1
  • 1
    $\begingroup$ Thank you very much and to Nick Cox for editing my post. I had no idea that my amateur question would attract the attention of two people whose work I have relied upon many times in the past! Thanks for sharing your time. $\endgroup$
    – David
    Commented Jul 6, 2021 at 14:12

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.