Dear all,

My name is Lawson. I would be indebted to anyone able to offer advice on post-estimation techniques following logistic regression.

I am interested in validating a prediction algorithm which consists of 3 rules (each binary yes-no predictors) on an outcome of interest.

Please see here for my Stata output.

. logit outcome rule1 rule2 rule3

Iteration 0: log likelihood = -360.83785
Iteration 1: log likelihood = -275.71313
Iteration 2: log likelihood = -270.92627
Iteration 3: log likelihood = -270.88658
Iteration 4: log likelihood = -270.88658

Logistic regression Number of obs = 578
LR chi2(3) = 179.90
Prob > chi2 = 0.0000
Log likelihood = -270.88658 Pseudo R2 = 0.2493

------------------------------------------------------------------------------
outcome | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
rule1 | .7363579 .2172793 3.39 0.001 .3104984 1.162217
rule2 | 2.058142 .2560151 8.04 0.000 1.556362 2.559922
rule3 | 1.03453 .2153743 4.80 0.000 .6124041 1.456656
_cons | -4.690449 .4938394 -9.50 0.000 -5.658356 -3.722542
------------------------------------------------------------------------------

As we can see from the output, all three rules in this diagnostic algorithm are statistically significant predictors of my outcome.

After creating the predicted probability of developing this outcome using the predict function, each patient in my database has been assigned their probability of developing my outcome of interest based on their satisfaction (and non-satisfaction) of these three rules.

I am interested in a way of graphically representing this data, and unfortunately I haven't quite been able to grasp the marginsplot function. Does anyone have any ideas? There are 8 possible combinations of the 8 rules, maybe plotting the probability of developing the outcome for each combination and the associated confidence intervals? Thoughts?

Thank you so much for your help.
Lawson