I have data at event A to predict a later event B. Fit using a multiple logistic regression model.
In order to improve the positive outcome rate for event B, a process was implemented just after event A using the predicted risk from the model.
Measuring the overall significance of the outcome was a straightforward test--even adjusting for the population variances.
Now I want to use the new data to make the model more robust. However the newly calculated risk scores are diminished due to the implemented process. AFAIK, I can do my best to model the process in the regression and just use those as adjustor and get a predicted risk score without the intervention.
This of course hinges on how the process is modeled. How do people typically deal with these iterative process questions? I want my risk scores to be meaningful, and not based on the list of previous processes.
PS. I wouldn't mind some direction to another post or some literature on this issue.
Advertise on Talk Stats