Linear Regression and ANCOVA

Question

Hello! 
I am working on running an analysis comparing # of mitochondria vs axon area in two different regions of a nerve (proximal & distal).

I tried to run a regression first, but since some data points are missing from my two groups, JMP says I cannot run a regression with missing data points.

unstacked but won't run regression b/c missing data points

I then stacked the data and it seemed to work so I can get 2 linear regressions on one graph but now I cannot run an ANCOVA because the data is stacked and is only recognized as one group. Can you advise me on how to run an ANCOVA on this data?

Dan_Obermiller · Answer

It sounds to me like you really have 2 input variables: proximal axon area and distal axon area. Your data should look something like this (with obvious made-up data):

Proximal Axon Area     Distal Axon Area     #

5.2                                 4.87                        2

5.4                                 5.1                          1

But does this make sense? In other words, I am assuming that for an object there is a proximal area and distal area on that same object so that they are "matched".

This would be multiple linear regression. There is no way to do "ANCOVA" because your distal and proximal counts are different. You MUST have a response for an observation to be used in a regression model.

Samantha · Answer

Thank you for your response. Sorry, the screenshot does not show it but the Proximal does have other values other than 1. But the issue is that I need to run an ANCOVA on this data to report the p-values for the main effect of locus (proximal/distal) and interaction of locus x axon area (regressor). The axon area is never the same proximally vs distally. That is why its either a data point for proximal OR distal, never both. There is no way to take this into consideration for ANCOVA & regression?

Dan_Obermiller · Answer

For your situation, you should use data in the stacked format. In other words, a column for your axon area, a column for your proximal/distal, a column for your # of mitochondria.

But looking at the data that I can see, your mitochondria count is always 1 if it is proximal (which is a problem). Further, from the split data format, I see that you only have a mitochondria count for either proximal OR distal, but never both. You must have data for both in order to see the effect of proximal/distal. That is why your regression is not working.

Remember that a missing value (represented by the dot) is NOT the same as a zero. Once you have your data filled in properly, you should be able to conduct your analysis.