Hey everyone! I want to compare the means of some variables of two groups (those, who completed all measurement points and those who did not). Completers are roughly 600-700 people per sex. Non-completers are around 100 people per sex. As one assumption of t-tests is that variances have to be homogenous and Andy Field strongly recommends not to use Levene-Test for this, but to perform correction in any case (Discovering statistics using SPSS p. 259), I won't perform Levene-Test. Yet, I want to perform bootstrapping, since most variables don't follow a normal distribution. I've recently read the article by Zimmerman (also recommended by Andy Field in his book), who says in the last sentence, that Welch's t-test should be performed when handling unequal sample sizes. However, I've also recently read in a discussion thread, that students' t-test is robust against unequal sample sizes when they have a certain size. Do my sample sizes already belong to this "certain size" or would you recommend performing Welch's test nonetheless?
Given the large sample sizes you mentioned (600-700 completers and 100 non-completers per sex), it is generally safe to use the Student's t-test even with unequal sample sizes. The Student's t-test is known to be robust against unequal sample sizes when the sample sizes are sufficiently large. However, if you have concerns about the assumption of equal variances and want to be more conservative, you can still opt for Welch's t-test. Welch's t-test is specifically designed for situations with unequal variances and can handle unequal sample sizes effectively. So, while the Student's t-test is likely to be suitable for your scenario, using Welch's t-test would provide extra assurance in case of any violations of the equal variances assumption.
"Yet, I want to perform bootstrapping,..."
If you anyway are going to bootstrap the distribution of your test statistics - why should you care about t-tests -- be it Student's or Welch's ?
Hello Sophia,
The problem with arguing robustness is, one never knows when it has failed for some statistical method. Like Jochen, I vote for bootstrap.
Good luck with your work.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA