|
|
Back to UserFriendly Strip Comments Index
| Statistics question. |
by Esteis |
2007-05-06 16:32:01 |
A friend of mine wants to see whether morning mood (scale of
1-10) predicts evening mood (also 1-10). She has 18 participants,
who each participated for a week, providing morning and evening
mood every day --- that's 18*7 = 126 morning-evening pairs.
Now, she wants to know something very general: whether a higher
morning mood is correlated to a higher evening mood. However, she
can't just do a regression and take the R^2, as the 'independence
of observations' assumption is violated: measurements are clustered
within individual participants.
My advice to her: add a 'participant' term to the linear regression.
(Well, a set of terms. It's a categorical variable, after all, so "helloooo
dummies!") So you have
moodPM = a + b1 * moodAM + [ b2-19 * participant2-19 ]
(Just imagine the bit between brackets is a load of dummies.) I think
that ought to catch the between-participant variance. Then, a large
R^2 and a b1 significantly larger than zero indicates a positive correlation
between morning and evening mood. My question: does this hold water? |
|
[ Reply ] |
|
if you run a regression on that | by subbywan | 2007-05-06 17:01:50 |
|
No, I don't mean whether her theory holds water. | by Esteis | 2007-05-06 17:06:41 |
|
No, i don't think it'll work then | by subbywan | 2007-05-06 17:13:17 |
|
Oh, there's nothing wrong with making a subject | by Esteis | 2007-05-06 17:38:16 |
|
ehhh ... you know something about dummie vars | by subbywan | 2007-05-06 17:44:16 |
|
|
[Todays Cartoon Discussion]
[News Index]
|
|
