Hi Doug,

Sorry the previous suggestion referred to a suggestion from Donald earlier in the thread as follows:

"You might consider the contrast

[0 1.5 .5 -.5 -1.5] across all 5 groups showing a linear effect of
disease severity. The first column being controls."

I will try that suggestion out, but I have also grouped all the patients and controls together in 1 group (as opposed to 5 different groups), and given that controls contains no endogenous factor giving them a covariate of 1, and then P1 a value of 2, P2 a value of 3, P3 a values of 4 and P4 a value of 5.  There are good biological grounds to suspect that that the endogenous factor will increase linearly from P1 to P5, though the values 1 to 5 I have chosen are arbitrary.

When I then run the GLM looking for correlation between dependent measure and this covariate, I see significant results in a similar (though more widespread) location compared to when I simply look for a difference between all controls and all patients (C v combined P1 to P4).  Thus, I think (hope) the model has shown that

Hopefully this is one valid way in which to show that it is the changes in the level of the endogenous factor contribute to the overall group differences (C v combined P1 to P4). Do you agree ?

I will however try Donald's method as I also want to show that the interaction between age-thickness (that I see in C v combined P1 to P4) shows a linear effect across P1 to P4. Any suggestions for how I do that ?

Thanks,