I want to compare two cumulative distributions (Interevent interval and amplitude) of mEPSCs using KS test. Running the test with the original data returns me p values less than 0.001, but if I bin the set (for example using 40 bins) the returned p value is above 0.05.
In that regard, should I bin the data or not?
If I should, I wonder what would be the "correct" number of bins, that would be considered not to be "arbitrary".
Note that the hypothesis the analyst would like "is normal" requires a large p-value not small because of the definition of the D stat. My PhD students missed this a lot. Check it out. 2 draw the cdf graphs just like a basic stats book says. These 2 things will probably help and next time please indicate your hypothesis so that we could be sure we are answer ing the right question. Best, D. Booth
Hello Cesar,
If the variables may genuinely be considered continuous in form, then I'm not sure I understand why binning (which in this case is tantamount to discarding some of the information in the data set) would be considered as a wise choice. What's more troubling is that the idea opens up the door to "p-value fishing," looking for the result that someone might have desired.
Beyond that thought, I agree with David Eugene Booth .
Good luck with your work.
for answering better! can I ask, what’s your exact mean about binning the data?
- Badges
- Science topic
- Similar topics
- Engineering
- Chemical Engineering