I have not worked with mice data before for paired designs and I could not invest much of thinking before this. For humans its simple we have same subject id and we have multiple measurements or paired design we will go for paired t test etc.
I have an understanding that mice are caged in groups and before today I was using groups comparison and it was easy to do unpaired t test for group-wise comparisons e.g. obese versus control KO for certain gene etc.
However recently, I have to analyze a data where there is this group of 5 mice group caged together. They are on Intermittent fasting food regimen where they are provided food 12 hours a day. two measurements are taken at day 1 pre fasting and post feeding. two more measurements are taken 4 weeks later again pre fasting and post feeding. Together they are 10 measurements for say variable X. where X can be any cytokine measurement or glucose or gene expression etc.
Now in order to do a comparison between prefast versus postfed at day 1 and week 4 shall I perform paired t test or unpaired t test and Why.
My guess is to perform unpaired t test because although the measurement are taken from same mice before and after we don't know which mice is which mice and therefore we cant get the differences of each mice ? which is needed to perform paired t test!
But then there would be violation of assumption of independence which confuses me further because then I can use paired t test? but how to get the differences of each mouse? Arbitrarily?
I have read about arbitrary pairing sort of thing but I am not able to comprehend the concept fully for lab animals experiments.
Please help in explanation why we should do either?
GenoMax @kevin I would appreciate your help particularly.
Thanks
Don't you give these mice ear marks or something to identify them? Anyway, if you cannot tell which mouse is which then it's unpaired, simple as that because the grouping information is lost. The dependent / paired tests have the advantage that they address the direct difference between the paired sample rather than caring about the individual differences between the individuals of one group. That increases the statistical power even if the groups itself are heterogeneous. But as you do not have that information it falls back to an unpaired test. This is an example where talking to an analyst before the experiment would have been a benefit as with your setup now you have reduced power compared to the paired test.
Thank You ATpoint for much clarification and explaining in detail. Actually I even dont know how they handle and can handle mice. I get to know by you that it is possible to do ear marks etc in mice experiment to get intraindividual variation sort of stuff. :( Person always talk so vaguely who deals with this stuff and the papers that I get to read for paired t tests are always so clear and were mostly on same organs from same mice type of experiment. I could not find literature where ppl have use paired t test for mice data so that I could dig more about how husbandary practice etc they explain in experimental part of methods. i tried a lot but could not find.
You simply need a way to distinguish the mice. You can also just color-spray a little dot onto the fur if that helps, as long as you know which mouse is which, because that enables the paired test.
And this is not hard for mice ppl and this should be kind of regular stuff for them? What I mean to ask is they can not start telling me - this is not how we usually can maintain mice like etc ? I am just trying to think through and gather all needed for pushing it for better statistical design implementation next time we do it and not fall for excuses etc.