For my study, I'm investigating whether student's scores in a task (expressed as a percentage between 0-100%) influence their responses to a survey. The survey has 19 questions, and they're Likert scale answers, from:

  • 5 - very effective
  • 4 - moderately effective
  • 3 - neither effective nor ineffective
  • 2 - moderately ineffective
  • 1 - very ineffective

My supervisor suggested clustering the 19 questions into 4 or 5 groups of related questions, and then applying a Kruskal-Wallis test for all of the groups (i.e. 4 or 5 Kruskal-Wallis tests.) She also suggested doing an analysis with two independent variables: the question number, and the scores for each participant. The dependent variable is then the participant's responses for each question. I'm not versed in stats at all, but she reckoned that this analysis would show whether there was a relationship between both of the independent variables and the dependent variables, and between the two independent variables.

But from what I've been reading, the KW test can only use one independent variable. Is there a way to retool this test for my purposes, or is there a different test more suitable to use? Thank you.

  • 2
    $\begingroup$ You might be interested in multinomial regression (multiple regression for a categorical dependent variable), or ordered logit regression (multiple logistic regression with an outcome that is ordinal). $\endgroup$ – Alexis May 7 '18 at 17:15
  • 2
    $\begingroup$ Whoops! Forgot to link ordered logit regression in my previous comment. $\endgroup$ – Alexis May 7 '18 at 18:33
  • $\begingroup$ Thanks. So given that I'm using a Likert scale, that would be ordinal, yeah? And my two independent variables would be categorical and continuous, I'm guessing that would be suited for a ordered logit regression... $\endgroup$ – Lou May 7 '18 at 18:49
  • $\begingroup$ Technically categorical variables get coded for their separate categories using effect coding, dummy coding, or some other coding scheme. $\endgroup$ – Alexis May 7 '18 at 18:56
  • $\begingroup$ As I said, I'm new to stats entirely - what would coding involve, and why is it important to do? Is it necessary for Likert-type responses? $\endgroup$ – Lou May 7 '18 at 20:51

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.