0
$\begingroup$

I have a questionnaire which contains 8 questions on a 4-point Likert scale. This questionnaire is repeated after 8 weeks and a training course about the subject. There are approximately 30 - 35 people answering the questionnaire. I want to determine whether the training had any effect - but a t-test seems very inappropriate, as the data is clearly ordinal and not even closely a normal distribution. Simultaneously, I cannot pair them, as the answers have to be completely anonymous. And additionally, I would like to determine which of three added questions - education level, work experience, and work place - have the most influence on the answers. Any and all suggestions will be taken under consideration. Thank you!

$\endgroup$
1
  • 2
    $\begingroup$ Do you object when people add Likert items for a Likert scale? (which is kind of the point of Likert's scales)? Note that adding items already treats the items as interval. $\endgroup$
    – Glen_b
    Jun 27 '19 at 4:30
1
$\begingroup$

You can use ordinal logistic or probit regression with the scores on the pre- and post-tests as the outcome and time and your other variables as the predictors. Ordinal regression implicitly models a latent logistic or normally distributed variable, which is discretized into the ordinal variable you observe. The coefficients can be interpreted as the linear effect of the covariates on the latent variable.

To examine whether the covariates modify the effect of time, you can include interactions between the covariates and the time variable.

$\endgroup$
4
  • $\begingroup$ As far as I understand the ordered logistic regression, it will tell me whether the test results are influenced by the predictors, but not whether the pre- and post-test differ significantly. Am I mistaken? $\endgroup$ Jun 29 '19 at 16:44
  • $\begingroup$ If time is a covariate, then the effect of time is the difference between pre- and post-test. $\endgroup$
    – Noah
    Jun 29 '19 at 20:26
  • $\begingroup$ That would require the identification of the tests to match them - which cannot happen due to privacy concerns. In the first part - determining whether there is a difference between the pre- and the post-test - I am looking for the ordinal equivalency of the paired t-test. $\endgroup$ Jul 2 '19 at 15:53
  • $\begingroup$ How can you do a paired t-test if you can't link pre- and post-tests? What I was describing is like an unpaired t-test, which is what I assumed you wanted. $\endgroup$
    – Noah
    Jul 2 '19 at 16:32

Your Answer

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

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