|
|
Re: Our first control group result is precisely how we want AML's
program to behave with control group data ....
Posted:
Apr 27, 2012 9:35 AM
|
|
Here is another example from the study group data for the b47 fold
(our fourth fold) of how adding m1,m2 to the predictor set seems to
improve an already reasonable model.
First, note that our "starter" {lnL,x1,lnLx1} model (which worked well
for the a1/a3/b1 folds) does NOT work at all for the b47 fold:
Overall Model Fit...
Chi Square= 6978.5794; df=3; p= 0.0000
Coefficients and Standard Errors...
Variable Coeff. StdErr p
1 2.9731 0.0558 0.0000
2 -0.1482 0.3234 0.6468
3 0.0985 0.0791 0.2131
Intercept -13.1380
Odds Ratios and 95% Confidence Intervals...
Variable O.R. Low -- High
1 19.5529 17.5265 21.8135
2 0.8623 0.4575 1.6252
3 1.1035 0.9450 1.2885
whereas if we switch to the alternative model {lnL,x2,lnLx2}, we get a
much better result:
Overall Model Fit...
Chi Square= 6875.1886; df=3; p= 0.0000
Coefficients and Standard Errors...
Variable Coeff. StdErr p
1 2.9442 0.0560 0.0000
2 -0.6323 0.3233 0.0505
3 0.1623 0.0791 0.0401
Intercept -12.9139
Odds Ratios and 95% Confidence Intervals...
Variable O.R. Low -- High
1 18.9955 17.0209 21.1991
2 0.5314 0.2820 1.0014
3 1.1762 1.0073 1.3733
Even though both of the confidence intervals for x2 and lnLx2 do span
1.00, they do so to much less an extent than x1 and lnLx1 do in the
{lnL,x1,lnLx1} model for b47:
{lnL,x1,lnLx1}:
Odds Ratios and 95% Confidence Intervals...
Variable O.R. Low -- High
1 19.5529 17.5265 21.8135
2 0.8623 0.4575 1.6252
3 1.1035 0.9450 1.2885
{lnL,x2,lnLx2}:
Odds Ratios and 95% Confidence Intervals...
Variable O.R. Low -- High
1 18.9955 17.0209 21.1991
2 0.5314 0.2820 1.0014
3 1.1762 1.0073 1.3733
And, if we now add m1,m2 to the {lnL,x2,lnLx2} model for the b47 fold,
we again get the same improvement in chi-square (relative to John's
null model) as we got for the a1/a3/b1 folds when we added m1,m2 to
the {lnL,x1,lnLx1} model:
Overall Model Fit...
Chi Square= 7000.9269; df=5; p= 0.0000
Coefficients and Standard Errors...
Variable Coeff. StdErr p
1 2.9419 0.0559 0.0000
2 -0.9085 0.3255 0.0053
3 0.1611 0.0792 0.0421
4 0.3638 0.0350 0.0000
5 -0.1433 0.0348 0.0000
Intercept -12.8209
Odds Ratios and 95% Confidence Intervals...
Variable O.R. Low -- High
1 18.9524 16.9855 21.1471
2 0.4031 0.2130 0.7630
3 1.1748 1.0058 1.3722
4 1.4388 1.3434 1.5410
5 0.8665 0.8094 0.9277
|
|
