Hi Erwin,

Can I suggest that you don't bin the data?

I think you've actually got data that you might want to put a bit of thought into before you throw your analysis at it. Let me explain.

I am assuming that your error bars are from repeated stimulation at the same intensity in the same slice (otherwise how could they have the same fibre volley amplitude). I am then a little concerned that you seem to have no error bars in the X direction (because the fibre volley isn't going to be exactly the same). I assume, however, that there must be some variance in the X direction as well as the Y direction. If my assumptions are correct so far, I believe then that you have essentially four sources of variance in both your X and Y measurements: variance due to the two groups (i.e. the purple and yellow groups), variance due to increasing stimulation intensity, biological variation (e.g. quantal variance and variance in the number of axons activated etc) and variance due to electrical noise (potentially you also have another source of.

It is my belief that you are only interested in two those sources of variance, that due to increasing stimulation intensity and that due to your two groups. So I think you would be best to average all over your recordings at the same stimulation intensity (this would reduce/eliminate the source of variance due to biological variation and electrical noise). Then you can compare your purple and yellow data sets with linear regression.

If you don't like my suggestion, then you NEED to be very careful how you analyze your data, as when you are repeating your stimulation at the same intensity, you are creating psuedo-replicates. I believe nested-analysis is needed to deal with that.

Perhaps I'm getting your data all wrong, if so, I apologize.

Hey Erwin,

So we're getting away from your original question, so no worries if you just want to ignore me.

So you are collating all the fiber volley and fEPSP amplitudes at a given stimulus intensity and calling that a data point. So my question would be, why? Why not just plot every single recording as it is? Yes, this will give you a lot of data points, but that's good! It will give the reader a better view of your real data. The "why" question above comes from two places, the first is, I personally believe that, for instance, a scatter plot is better than a bar graph, as showing the reader more of your data is better than showing them less. I personally have more faith in results when I don't think the author is trying to hide things from me. The second point is perhaps more fundamental: why do you think that a fiber volley/fEPSP from one slice recorded at 10uA should be lumped with another one from another slice? That is to say, you wont be stimulating the exact same number of axons, you wont be recording from the exact same number of postsynaptic elements, your electrodes wont be in the exact same place (depth/layer). I appreciate you've probably tried to make those things as similar as possible, but we both know it is impossible to do that, so why lump the data at all? I will also add that this approach will make using regression to test for a difference in slope more legitimate, as your X value is supposed to be a known value with no error.

If I was making the graph, I would plot all of the data points, and use a "washed out" colour (i.e. instead of your purple, something like Red = 220, Blue = 200, Green = 220). And then plot the regression line in your dark purple. (Perhaps a sigmoid might be a better function to fit than a line, though I am a big fan of using simple statistical models).

But those are just my thoughts. If you've really got your heart set on binning the data, I'm happy to write you a Matlab (or Python) script to do it for you.