0
$\begingroup$

I'm comparing an intervention using a pre and post knowledge survey. My pre sample size is n=37, my post is n=32. The variance for the pre is 7.08 and 6.94 for post. The histogram of the pre is skewed right with a Shapiro-Wilk p-value of .048, the histogram of post looks more normal with a SW p-value of .087.

As my sample sizes are >30 I'm thinking it's fine to violate the normality assumption for the t-test (as my pre isn't normal)? As the sample sizes are similar and variances are similar I'm thinking I could use a independent sample t-test with equal variances. Is this the case or is there a strict rule that even if variances are similar the sample sizes HAVE to be the same for an equal variance t-test, in which case I could just use an unequal variance t-test?

$\endgroup$
  • $\begingroup$ Please edit your question as opposed to my answer. $\endgroup$ – Demetri Pananos Jun 25 '19 at 15:56
0
$\begingroup$

You're fine to use a t-test in this case. The assumption about homogeneity of variance is made at the population level. In your case, the differences in variances are so small, they are likely do to random variability.

The t-test can be surprisingly robust against violations of the normality assumption. I would say so long as your data don't look like an exponential distribution, and so long as the data are roughly symmetric about the mean, a t test is a good idea.

| cite | improve this answer | |
$\endgroup$

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.