Dear Ralph, here the fine print matters.

Assume you have 1000 recordings from 1 cell. You do a  KS test with N=1000 independent samples. If significant, you demonstrate that for the very cell you recorded you have a significant effect. You can not make a statement on other cells.

Assume you have 1 recording from 1000 cells. You do a  KS test with N=1000 independent samples. If significant, you demonstrate that for the population of cells you sampled from you have a significant effect. You can not make a statement on individual cells.

You have 100 recordings from 10 cells. If you want to demonstrate an effect in individual cells, the first example applies, and you have N=100. But you do not have a statement on the general population. A possible result would be that in 8 out of 10 cells recorded you find a significant effect. I find that a reasonable way to go. 

If you want to make a statement on the general population, sorry about that, you have a N=10, the number of cells matters.

An alternative is a linear mixed model. It simultaneously considers variance within and between cell recordings. 

Best, Peter

Dear Ralph,

I would average the events in each cell first. That's what we have done in our previous papers comparing responses between WT and KO animals.

Best wishes,

Refik

Dear Peter & Refik:

Thanks for the rapid and thoughtful responses.  I understand the "individual cell" idea laid out by Peter, but how, then, do we compare the control cells with the treatment group cells?  We have cells from control animals, and cells from nicotine exposed animals.  So, we could look at acute effects of a drug WITHIN a cell using the K-S test on a cell-x-cell basis, but how do we compare, for example, the event frequency at baseline between control and nicotine-exposed cells?  I suppose Refik's idea to simply take an average in each cell would work there, though there would be more events in some cells than in others, which would reveal a "weighting" issue.

Oh my, stats are a necessary ulcer-inducer!

Thanks again,

Ralph

Dear Ralph,

Yes you are right, indeed stats can either be an ulcer- or migraine- inducer, and can hold a paper for years before acceptance. I think there are several things you can do to reduce the variability in your sampling. Such as using same age and sex of animals. I presume you are working with hypoglossal motoneurons-If you are, then you can try to include cells primarily from ventromedial region of the nucleus-just to increase the uniformity. Discard cells with too high (bad access) or too low (leaky) resistances, and include cells that have resting membrane potentials of -55 mV or lower (without spontaneous action potentials). For consistency, I usually analyse the first 60s of recording, soon after the whole-cell is established. n=10 is a pretty good, but increasing sample numbers may also further improve the case. There is no magic bullet but I hope that is some help. Best wishes, Refik

Thank you Refik--we do take many of those precautions with hypoglossal motoneurons, although we record for 2-3 minutes, which perhaps is unwise.

I am meeting with a biostatistics expert this afternoon, and will post the results of that meeting.

Ralph