Hi,

I have an experiment with 3 factors: cell_type (`cell1`

, `cell2`

, ... `cell10`

), time (`t1`

, `t2`

, ... `t7`

) and treatment (`control`

, `treated`

). I'm modeling it as `~0 + time*treatment + cell_type`

, and my resulting design matrix has columns: `control.t1`

, `treated.t1`

, `control.t2`

, ..., `cell1`

, `cell2`

. Note how I do not have independent columns for `control`

or `treated`

only, This is to make it easier to handle the contrasts, in which case I am combining multiple columns if I just want to compare `control`

vs `treated`

. But I digress.

My question is the following: I want to analyze the coefficients at different times or treatments. For this I can extract my `fit$coefficients`

object resulting from `lmFit`

, but then I get coefficients with the same columns as my design matrix, i.e. coefficients for columns `control.t1`

, `treated.t1`

, `control.t2`

, ..... But I want to extract a representative coefficient of `t1`

, `t2`

, etc irrespective of `treatment`

. For instance, I want to take the coefficients at time 1 so both `control.t1`

and `treated.t1`

, but I'm not sure how I can combine the coefficients. I.e. I want to analyze the `t1`

coefficient irrespective of treatment.

I'm guessing that it should be possible to alter the design matrix to represent the individual effects but it would fail since the column for `t1`

would basically be given by `control.t1 + treatment.t1`

(and the same for all time points), in which case the model would not be fittable. I'm also thinking that alternatively I could compute the mean of the coefficients `control.t1`

and `treatment.t1`

, but not sure how accurate this would be. Any suggestions?

I've posted a related question, though it is a different question here.