# How to find groups inside the dataset (test for bi-multimodality)

I have non-normally distributed dataset, a record of a parameter, e. g. 50 subjects reacted to stimulation, (9 periods of time in total). So I have a matrix of 50x9 numbers, 9 medians for each time-period for each volunteer. When I plot the above data, subjectively I see that there is some division, a part of subjects reacted to stimuli with a delay (differences in parameter change in 4th period rather than 3rd). I am trying to find a statistical way to check if there are these groups (to check for correlations with other parameters later( psychological stats)). Actually, should I test for bi or multimodality? I am using Matlab and Statistica

I will be grateful for any ideas.

Upd.: The record of several physiological parameters (sweating, temperature, etc, changes in those as a response to emotionally-negative footage). There were rest-stimulation-rest phases, the recording of physiological data was performed continuously

• I added the clustering tag to help you: you can link through it to thousands of qualitatively similar problems. You can narrow the possibilities by describing your experiment in more detail, because it sounds like it might be a longitudinal or "repeated measures" study. – whuber May 12 at 16:21
• @Whuber thank you! It was monitoring of several physiological parameters (sweating, temperature, etc, changes in those as a response to emotionally-negative footage). There were rest-stimulation-rest phases, the recording of physiological data was performed continuously. – Vitto Titto May 12 at 16:31
• What you should likely test is whether it is not unimodal. We don't know what the data looks like - Gaussian? Then try a goodness of fit test with a Gaussian. If you cannot reject the hypothesis that it is a single Gaussian, then the deviation you see are likely random. – Anony-Mousse May 14 at 5:55
• @ Anony-Mousse thank you very much! – Vitto Titto Jun 8 at 10:13