I know that I recently post the same message, but I'm still wandering in the desert and I really hope to get some feedback.
Dear all,
I am looking at correlation between PET images and age in a sample of healthy control.
The matrix is pretty easy, as age is the only covariate entered in the model. I get some nice cluster of significant correlation.
Then, when I look at the plot (effect of age on the signal, fitted response), I see that the predicted PET signal is sometimes negative, as is the age. While I understand why the age has negative value (due to the centering of the covariate), I don't understand why the PET value become negative for some subjects. Is there a "silent" centering of the PET data during the estimation ? If it is true, is there a way in which I can get rid of this step ?
Basically, I would like to obtain "realistic" prediction for the PET values, using age as a parameter (i.e. PET value=beta from the regression*age, where PET value is a realistic value).
I hope I made myself clear, thank you in advance for any tips or advice
Best
F.
|