I have firing-rate data for neurons ranging from 40-60 in number for various experimental groups. I've seen both mean and median being used in the literature, but without much explanation for the choice. My boss thinks that usually for sample sizes > 4-5, mean is ok to use. But I would like to have a better justification for my choice since the results I get with the two quantities are quite different.
$\begingroup$
$\endgroup$
3
-
2$\begingroup$ There is no such rule. You can either show both or choose the one that is more relevant for scientific reason. $\endgroup$– Michael MApr 1, 2014 at 5:35
-
$\begingroup$ The relative efficiency depends mostly on the distributional shape. It does change a little with sample size of course (at n=1 and 2, for example, both sample means and sample medians are just the same!), but it's mostly not so much impacted by sample size. Why would (as it seems from your post) you use one at n=6 and a different one at n=3? If means are a problem at n=3, why wouldn't they be a problem at n=6? What is it you're actually trying to achieve by computing one or the other? What's the model? $\endgroup$– Glen_bApr 1, 2014 at 7:02
-
$\begingroup$ The answers to a very similar question at stats.stackexchange.com/questions/2547/… are relevant. $\endgroup$– whuber ♦Apr 1, 2014 at 16:27
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
Rather than compare means/medians why not fit a model to each and compare the distributions of parameter estimates for each group? This approach would seem to provide much more information.
Edit: I moved my "answer" to this question as it seemed somewhat offtopic here. How to fit this neuron firing model with R?
