[Marvin Aliaga]

wbuchanan wbuchanan Thanks guys for all the comments. To be honest I am kind of lost since it is the first time I have to do a project like this. Let me explain you a little bit about this model. The model is not created yet since we are trying to collect new data that we think is import for the model. Also we don't know what would be the best data structure or statistical model. That's a different problem and post. I just wanted to get ahead and be prepare for when i have to data and the model.

My main goal is to create a model that predicts which staff member is more likely to use force (UOF) in a mental health center. The goal is to detect these staff and prevent them from using force. I envision this as staff members entering their ID (which will populate their demographic characteristics) as well as their schedule for the week (which will populate the characteristics of their schedule, role, location, shift, etc). Then a score will be generated for each day of the week (since they can work one day in a more dangerous facility and the other in a more quite an small facility). I will generate this model using historical data. I assume that my model will be a logistic regression with dichotomous DV - UOF (yes or not). i will post a new inquire about what should be the data structure and methodology for this type of analysis.

Thank you,
Marvin

[wbuchanan]

Marvin Aliaga you may want to consider looking at survival models. If there is reason to assume that some staff will never be in a position where there is a decision to use force (e.g., never in the risk pool) then the outcome would be unobserved, while it would be observed for staff who were in a position requiring that decision to be made. If you don't care about the estimation of the model parameters (e.g., only care about predicition) then there are lots of different approaches in the machine learning literature that may be a better fit for the particular application.