|
|
Re: Our first control group result is precisely how we want AML's
program to behave with control group data ....
Posted:
Apr 28, 2012 12:17 AM
|
|
1. You wrote:
"You can't use the 7-predictor model, because its coefficients are
indeterminate."
Ah - now I understand what you meant when you said offline that this
model had a "singularity". Thank you.
2. You wrote:
"The chi-square for the 'bigger' model (the one with more parameters
to estimate) will always be smaller;"
Will it be "smaller" regardless of the way the program is reporting
the chi-squarees, i.e. relative to the null model like John's program
or relative to the "saturated" model (like the one you use?) Or will
it be smaller in one case and bigger in the other?
3. You wrote:
"A significant chi-square is taken to mean that the bigger model is
better, that the extra parameters bought a better fit. However,
nonsignificance does NOT mean that the two models are equivalent; it
means only that you can not say that the bigger one is better."
After I have run all 12 cases (the study group and control group data
for all six folds) using
i) the 3-predictor model {lnL,x1,lnLx1} or the 3-predictor model
{lnL,x2,lnLx2}, whichever works better for each fold;
ii) the 5-predictor model obtained by adding m1,m2 to the better 3-
predictor model for each case;
I hope you will have time and energy to explain how you would "chi-
square" all 12 pairs of chi-aquares from all 12 "3 vs 5" cases in
order to reach the most informed conclusion about whether addition of
m1,m2 can be said to significantly increase chi-square (relative to
the null model.) I am assuming here that you can do more with 12 pairs
of chi-squares from 12 cases than you can with just one pair of chi-
squares from one case, but if I am wrong about this, i.e. if you still
have to evaluate each of the 12 cases independently of one another,
please correct me here.
And thanks very much for the continued dialog. (BTW, Jacques didn't
cc me on his note to you, so I didn't know till today that he had sent
it.)
|
|
