the reference variable. It's completely arbitrary and equivalent, and only something you have to bear in mind when analysing how the groups compare, which I'm not.
(Example: I want to compare weight among tortoises, plumbers, and gorillas. I can choose any one as my reference group: if I choose the heaviest, the other two will get negative coefficients; if I choose the lightest, the other two will get positive coefficients; if I choose the middle one, etc.)
No, the reason I'm introducing the subject variable is to catch any between-subject variance. Say, for instance, that evening mood is only determined by character, and has nothing to do with morning mood. Any one subject will then have a circular cloud of morning-evening points, with no correlation. However, add a second subject with naturally higher mood, and suddenly the two together will look like an elongated cloud, and create the impression that there is a correlation between morning and evening mood. When in fact all the evening mood variance is due to the subject, and none to the morning mood. I've drawn a crude diagram below.
3
3 3 3
2 3
2 2 2
1 2
1 1 1
1
So I'm wondering whether having the subject dummies in there will correctly explain whatever variance there is to be explained by the subject, and then will correctly leave the morning mood explaining no variance. (Or having a slope of 0. However you care to call it.) |
