Hi again,
>
> Just making sure I got it:
> If have several subjects with missing values for the same independent
> variable, do I need to add one extra variable for each subject? or
> can I add
> just one for each independent variable, coded "1"' for all subjects
> with
> missing values?
> To remain within your example: if I have missing values for subject
> 10, 15
> and 20 for disease duration, and missing values for subject 3, 4 and
> 5 for
> age, do I really need to add 6 extra variables? Or just 2: one for
> each
> indep var?
neither. If you add 6 extra columns that is exactly equivalent to
throwing away all data for which there is missing data (which I
thought we wanted to avoid).
My idea was to only "throw away" the data when there were missing
values for the specific variable that we wanted to test for. So if in
the first analysis you want to test for "age" and that you are missing
data on age for subjects 10, 15 and 20 you should then add three extra
columns. On for each subject that has missing age data. In the design
you include both the age covariate and the "disease duration"
covariate, with missing values in disease duration filled in in one of
the ways I suggested. For this analysis you only test for age.
You then do the same thing for disease duration, this time adding as
many extra columns as there are subjects with missing data for disease
duration.
N.B. that as I indicated this is not strictly kosher. But rather a
kind of hand-wavy compromise that I think would be reasonable.
Jesper
|