I want to compare ERP data from a single group of participants between 2 measurement times. I also want to compare the data from this group of participants with a group of control participants. I have used the jackknifing method to get cleaner averages. Sample size is 9 in both groups. I know that usually, t-tests are performed on ERP data even when the sample size is small. But I have been told by some people evaluating my last seminar that it was absolutely not appropriate for me to share (in my thesis) the results of parametric tests considering my sample sizes. I don't necessarily agree with them but I don't have much choice. So my questions are :
1) is it possible to perform non-parametric statistical analysis on jackknifed data? These analysis would mainly be non-parametric equivalents to t-tests (Wilcoxon and Mann-Whitney).
2) if it's possible, given that t values have to be adjusted before looking for the p values, how can I adjust the values I obtain in the non-parametric tests ?
3) if it's not possible, can you detail the reasons why it should not be done?
Non-parametric analysis is the best choice since biological data is usually not uniform distribution (bell-shaped). Usually non-parametric p -value is less significant than parametric analysis. So, your p-value can lose strength. For adjustment, you may use FDR, benfornni etc.
Hello Lydia,
If data conform to the assumptions for parametric tests, then parametric tests are more powerful (by a slight bit) than non-parametric tests for a given data set: the sample size is not the determinant for suitability.
If you are at all concerned about assumptions (as noted in Md Saifur Rahman Khan 's advice), then go with an exact test (with only 9 cases per group, the computational effort is not great). More statistical packages offer this as an adjunct or option to the more ancient parametric and non-parametric tests.
Good luck with your work.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size