|
|
Re: Our first control group result is precisely how we want AML's
program to behave with control group data ....
Posted:
Apr 28, 2012 6:02 PM
|
|
I sent you the following results and question off-line, but am also
posting them here in case we want to refer to them in later posts
here.
Using a new three-predictor model:
{lnL,mv,lnLmv} (where mv = 0,1,2,3 instead of m1 = 0, 0, 1, 1)
I obtained these results for the a1 study group:
20619 cases have Y=0; 11029 cases have Y=1.
Overall Model Fit...
Chi Square= 3887.9731; df=3; p= 0.0000
Coefficients and Standard Errors...
Variable Coeff. StdErr p
1 2.0122 0.0765 0.0000
2 -0.7446 0.1593 0.0000
3 0.2368 0.0401 0.0000
Intercept -8.8669
Odds Ratios and 95% Confidence Intervals...
Variable O.R. Low -- High
1 7.4795 6.4380 8.6894
2 0.4749 0.3476 0.6489
3 1.2672 1.1713 1.3709
and these results for the a1 control group:
3765 cases have Y=0; 2333 cases have Y=1.
Overall Model Fit...
Chi Square= 1039.2102; df=3; p= 0.0000
Coefficients and Standard Errors...
Variable Coeff. StdErr p
1 3.3186 0.1799 0.0000
2 0.2694 0.3706 0.4673
3 -0.0639 0.0939 0.4962
Intercept -13.5375
Odds Ratios and 95% Confidence Intervals...
Variable O.R. Low -- High
1 27.6216 19.4123 39.3027
2 1.3092 0.6332 2.7069
3 0.9381 0.7805 1.1276
Even without comparing chi-squares after dividing by sample sizes, can
we not say "right-off" based solely on confidence intervals that the
variables mv and lnLmv are significant for the a1 study group but NOT
for the a1 control group, given the fact that the confidence intervals
for mv and lnLmv do not "cross 1" for the study group, but do "cross
1" for the control group?
While waiting for your reply, I will run the same model on the a3 and
b1 study and control groups ...
Thanks again for your patience .
|
|
