Hi everyone,
I performed a Shapiro-Wilk Test to check the normality of my 60 samples (n=4). If just 5 of my samples are not normally distributed, can I still perform an ANOVA?
I also performed a Levene´s Test to check the homogeneity of each sample in different times of analysis. Some of them are significantly different and some are not. In this case, would be better to perform only non-parametric (Kruskal-Wallis) test for all my samples instead of using ANOVA?
Thank you in advance for your help :)
Veronica
Have you checked the residuals for normality? That would be the first step I should think. Depending on the results of that, and given the small sample size, I'd think you'd want to consider using a bootstrapping technique rather than rely on non-parametric tests.
Hello Veronica,
Do you mean that you have 60 batches of cases, each batch with 4 cases)? If so, the sample size per group is too small, in my opinion, to warrant checking for normality (and, as Edward G. Brown 's answer implies, you would be ideally checking for normality within groups, if groups was the lone independent variable).
If you mean you have 60 cases, each measured 4 times, then normality is checked within each measure (after accounting for any other IVs).
Finally, if I've missed the design of your study completely, here's two pieces of generic advice: (a) you can always use exact/resampling tests even if you think the usual parametric assumptions are not satisfied; (b) the parametric methods are pretty robust with respect to mild assumption violations--the only drawback is that you'll never know when robustness fails.
Good luck with your work!
Thank you Edward G. Brown and David Morse for your replies.
I have 15 samples in 4 different times (60 cases), each measured 4 times (a total data of 240).
One more question, is it valid if in the same experiment I apply ANOVA for some samples (with ND and homogeneity) and non-parametric test for the ones that do not have ND and homogeneity?
Best,
V
It sounds like your measurements are not independent, so you may need to try another approach to deal with the repeated measures. Unless I'm misunderstanding what you're saying.
If your small samples are dependent use Friedman two way analysis of variance. If you use repeated measures you can check our mice experiment over generations.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA