2
$\begingroup$

I am trying to compare the paired pre to post-responses in a student survey to see if gender plays a role. I was told to do a wilcoxon signed ranked test, and I can get the mean and p-value by each gender, but it won't compare the genders just the pre-post data.

Am I using the wrong test?

$\endgroup$
4
  • 1
    $\begingroup$ What is your outcome variable? $\endgroup$ Commented May 1, 2023 at 22:50
  • $\begingroup$ The pre and post are both acting as my outcome variable since we are attempting to see the difference between the responses after educational training $\endgroup$
    – Cam
    Commented May 2, 2023 at 0:41
  • $\begingroup$ Are they total scores or scores for a single item? $\endgroup$ Commented May 2, 2023 at 1:49
  • $\begingroup$ I have 47 participants who each have a pre and post variable that is used to do pairings. $\endgroup$
    – Cam
    Commented May 2, 2023 at 2:36

1 Answer 1

0
$\begingroup$

A least as an initial answer: Yes, you are using the wrong test.

A simple approach would be to use the differences between the paired observations as the dependent variable, and compare genders using a test for independent samples (t-test or Wilcoxon-Mann-Whitney, if there are two genders).

But this won't directly address if there is a difference between pre and post. You could do a separate test to determine if the differences calculated above are statistically different from zero (one-sample t-test, one-sample Wilcoxon test, one-sample sign test).

A more sophisticated --- and probably preferred approach --- would be to create a single model which includes both Time and Gender as independent variables. This would use repeated measures since the same subject has both a pre and post response.

$\endgroup$
2
  • $\begingroup$ Why not use an ancova? This is the suggested approach for such data, as per Bland and Altman. ANCOVA has the additional benefit of not assuming the regression coefficient for the pre variable is 1, an implication of the change score approach you mention. $\endgroup$ Commented May 2, 2023 at 4:58
  • $\begingroup$ Thanks, @DemetriPananos . Analysis of covariance (ancova) is a good solution. A single model including Pre, Post, and Gender. $\endgroup$ Commented May 2, 2023 at 14:11

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.