2

I have a dataset with n = 76 independent test subjects (rows) and 3 dependent variables (columns) (which are related as a process taking place over time). For the example below I will assume I have two different conditions (condition 1 and condition 2). I have some questions about how to analyse this dataset correctly.

1) From a statistical point of view, are there any potential problems with slicing the same dataset in different ways (i.e., based on the two different conditions) and then perform statistical tests for each condition and for each of the 3 dependent variables. Example: first the n = 76 participants are divided into group 1 with n = 30 and group 2 with n = 46 (condition 1) and 3 statistical tests are used to understand if the dependent variables are different, then the same n = 76 participants are divided into group 1 with n = 21 and group 2 with n = 55 (condition 2) and another 3 statistical tests are used to understand if the dependent variables are different. The test used in all cases (if that would matter) is the Mann-Whitney U non-parametric test.

2) What I have understood is that there may be an idea to account for multiple comparisons (e.g., with the Bonferroni Correction) if repeated tests are performed on the same sample (even if it seems that there are different opinions on this). For the case described above, do I need to account for multiple testing and in which way does it matter that the 3 dependent variables are related? In 1), above, a total of 6 different tests are performed (3 for each dependent variable for condition 1 and 3 for each dependent variable for condition 2).

I hope I made myself clear enough. Thanks!

tankenot
  • 21
  • What are the "3 statistical tests" that are used on each condition? If you just trying to figure out if the two groups are different, that could be done with a single Mann-Whitney U, instead of 3. – Underminer Dec 10 '19 at 13:51
  • I perform three statistical tests (Mann-Whitney U), since for each condition I want to understand if three dependent variables differ. Does that make sense? – tankenot Dec 10 '19 at 19:50
  • Yes that makes sense. I missed that the first time. The binferroni correction would be appropriate for the three tests. There may be a more elegant solution other than splitting the same dataset into the two conditions. You could possibly be ignoring an interaction between the two splitting variables – Underminer Dec 10 '19 at 21:44

0 Answers0