 |
 |
Hi Elisa, Please see below: On 13 April 2017 at 17:08, Elisa <[log in to unmask]> wrote: > Dear All, > > I’m setting up a GLM model for my ICA dual regression analysis and I have > two questions about my model. > > To summarize the design, I have 4 groups: 1) Patients age1, 2) Controls > age1, 3) Patients age2, 4) Controls age2. Patients and controls within an > age group are matched for age and gender. There is not the same number of > participants in age 1 and age 2 and there is not the same number of male > and female in age1. I am interested by the interaction between age 1 and > age 2 and by the effect in each age group (patients vs controls). > > When I run a GLM with only one age group in the model (Patients age1 vs > controls age1 – contrast 1 -1 and -1 1) and an unpaired t-test, I have not > the same results than when I run the GLM with the 4 groups in the same > model and assess the result of the same exact contrast (Patients age1 vs > controls age1: contrast 1 -1 and -1 1). I have read on the forum that when > you put all groups together in the same model, other groups contribute to > the final result through error term that is pooled across all groups. I am > then wondering what is the best solution in my case? > > Given that you are interested in the interaction, I would say it is better to keep all subjects in a single model, not separating into one model for age1 and another model for age2. ] > I have a second question regarding demeaning. I have read a lot of > comments on the forum about this question but I have not found an answer to > my particular question yet. I would like to put gender in nuisance factor > in my model. As I have set up 4 different groups, I have defined 4 EV for > gender and I have demeaned gender within groups (with males =1 and females > = 0). Is that correct or should I rather demean across groups first and > separate the 4 EV for gender after? > You'd mean-center across all subjects, but in this case, there is no need for any mean-centering, because the intercept is already present in the model (jointly coded among EVs A, B, C, and D). Hope this helps! All the best, Anderson > > If I have 3 participants in each group and all groups in the same model, > my model is like this: > > A B C D E F G H I > 1 1 0 0 0 -0.3 0 0 0 > 1 1 0 0 0 0.7 0 0 0 > 1 1 0 0 0 0.7 0 0 0 > 2 0 1 0 0 0 0.5 0 0 > 2 0 1 0 0 0 -0.5 0 0 > 2 0 1 0 0 0 0.5 0 0 > 3 0 0 1 0 0 0 -0.3 0 > 3 0 0 1 0 0 0 0.7 0 > 3 0 0 1 0 0 0 0.7 0 > 4 0 0 0 1 0 0 0 0.5 > 4 0 0 0 1 0 0 0 -0.5 > 4 0 0 0 1 0 0 0 0.5 > > (A=group, B=patients age1, C=patients age2, D=controls age1, E=controls > age2, F,G,H,I = the 4 EVs for gender > > And my contrasts > > A B C D E F G H > 1 -1 -1 1 0 0 0 0 > -1 1 1 -1 0 0 0 0 > 1 1 -1 -1 0 0 0 0 > -1 -1 1 1 0 0 0 0 > 1 0 -1 0 0 0 0 0 > -1 0 1 0 0 0 0 0 > 0 1 0 -1 0 0 0 0 > 0 -1 0 1 0 0 0 0 > > > Many thanks in advance! > Best regards, > Elisa >