 |
 |
Hi Michiko, Please see below: On 2 June 2017 at 12:21, Michiko <[log in to unmask]> wrote: > Hi Anderson, > > Thanks very much again. However, I apologize but I didn't completely > understand your reply and would appreciate your kind clarification. > > 1. I understand in your earlier reply that you said mean centering is not > necessary if I am not testing the variables on their own. However, if I > were to be running only one model that includes patient-specific regressors > and would like to test the effect of patient-specific regressors (e.g. > duration of illness) on FA, would mean centering be necessary? How should I > mean center this - e.g. across HC, pt grp 1 and pt grp 2? Or only > across/within pt groups? > > Mean-centering is only about data that is in the model, not data that > doesn't go into it, so if you run a model *only* with group 1, mean > centering would be within group 1 only. However, you could forget about > mean-centering and simply include an intercept in the model. > > >>> This is in reference to running only one 3-group model that includes > both general (for all groups) and patient-specific (only available for 2 > groups) regressors. To be certain, do you mean that I would mean center > within patient group 1 and then mean center within patient group 2, for > patient-specific regressors? But what would I enter for the healthy control > group which would be in this model? > Sorry, the reply wasn't clear. Maybe the word "only" could have helped. I've just added it above (in bold). > > 2. If I employ the 2-model design, would I have to recreate the mean > template with only the pt groups? Might there be any issues with this? > > I don't think it's a problem to use the same template given that all > comparisons will be within-group in this second model. > > >>> This is in reference to running a separate model with only patient > groups after running a 3-group model. My question is if it is OK to use the > mean template generated from the 3-group model that initially included > healthy controls, to test for between-group differences between the 2 > patient groups? Or should I recalculate the mean template by > re-registeration with only patient groups this time? > I think it's not a problem to use the same template for all models if the template uses all subjects from all groups. The problem that could happen in group comparisons is that the template could favour one of the groups, but in your analysis, either all groups will be in the model (so, a template using all groups won't be biased), or the other models will not make group comparisons. So, in my view, it's ok to use the same template using subjects from all groups for all the proposed analyses. All the best, Anderson > > I apologize and thank you for your patience and time in offering advice! > > M. > >