Getting NonEstimable values when crossing variables due to control setup

eberndl — Fri, 10 Nov 2023 18:48:35 GMT

Hi,

I'm working on a dataset where we are testing the effects of 2 treatments and the delay between them, and I'm trying to figure out how to line up my variables so I can do a Dunnett's test against various controls.

We are giving animals a dose of drug A (Variable A), waiting some time (T), and then giving them a dose of drug B . If the animal is NOT given drug A (= 0), then T doesn't have a meaning, so I've been setting it as -1. Similarly, there is no meaning if the animal does not receive drug B, which I've ALSO been setting at T=-1. Therefore all what I'm considering to be the controls (either no A or no B) have a T=-1

I've been creating my graph using graph builder and then launching a standard least squares analysis. If I use A, B and T to model the effects, I get values, but if I want to look at the interactions, everything falls apart and all my means become NonEstimable.

I understand that this is because there is no overlap between my T=-1 and my other time points, but I don't know how to say that a T=-1 for A=0, B=0 is the same value as a T = 0, 1 or 2. Do I just replicate my data with multiple T's? Do I need to put weighting on that pseudo data to make sure the standard deviations are accurate?

topic Getting NonEstimable values when crossing variables due to control setup in Discussions

Getting NonEstimable values when crossing variables due to control setup