0

I have two groups (control and experimental) with 16 subjects each. Their pre-test scores are statistically different from a normal distribution so I opted to do nonparametric tests. I need to compare the learning gains of the students (post test score minus pre test score) but don't know what nonparametric test to use. I know though that if I were to use parametric tests, the ANCOVA is the test that I should use. What is ANCOVA's nonparametric test counterpart?

EDIT: Can I use Quade's test for this? This is a quasi-experimental study by the way, with the objective of determining whether a learning intervention enhance student achievement.

Thanks in advance!

kjetil b halvorsen
  • 77,844
Teacher Researcher
  • 11
  • You do not need to have normally distributed data to apply an ANCOVA. Use the "asymptotic chi-square" test instead of the F-test. – AdamO Mar 09 '18 at 14:11
  • The goal of ANCOVA is to account for variability caused by covariates. Do you have covariates? – Dave Nov 18 '19 at 11:04
  • 1
    https://stats.stackexchange.com/questions/3466/best-practice-when-analysing-pre-post-treatment-control-designs – kjetil b halvorsen Mar 21 '20 at 13:45

1 Answers1

1

If you are willing to subtract the pre-score from the post-score, and consider that your dependent variable, then you can just use a Mann-Whitney test between the two groups.

Otherwise, I think you would have to use an aligned ranks approach to stay in the nonparametric realm.

Sal Mangiafico
  • 11,330
  • 2
  • 15
  • 35
  • Not sure why you qualified your first statement. Taking the difference between the pre-score and post-score would be the standard approach. – Eli Sep 06 '22 at 13:29
  • @Eli , I was thinking that in order to subtract the scores, the observations have to be treated as at least interval in nature. If the observations are treated as ordinal observations, subtraction couldn't be used. But in this case, mixed effects ordinal regression could be used. – Sal Mangiafico Sep 06 '22 at 14:45