A probability problem comes up in policy that needs frequent solving:
Do we immediately send police to a residence of suspected activity based on probability the situation will escalate? Will a person likely skip bail? How likely is a benefit to be applicable after only a preliminary screening?
The challenge is that sometimes probability is involved. We can't immediately send police to every event all the time. We must sometimes be selective and our policy needs to allow for policy probability in some parts of our determinations.
The local department of criminal justice has provided the following statistics in determining the person’s probability of missing court (these statistics are made-up). The overall probably of missing a court appointment is 45%. Of those who miss court, 40% are in the local community, 80% had an outstanding warrant, and 10% had a job.
Allowing for uncertainty and collecting 1) whether a person is a member of the community, 2) has an outstanding warrant, and/or 3) has as job, write an OPA policy that implements the following:
As any FYI, my collection screen looks like this:
The solution to this problem is attached with a more "generic" structure for solving problems of "likelihood".
The solution is simple in that a one page word document sets up the math. A spreadsheet contains the list of question attributes used in screens to refine the final probability. Questions which refine the probability can be added into the spreadsheet. You can have 2-3 questions such as in this puzzle, or you can have several hundred.
A theoretical introduction to the solution can be found here: https://www.askiitians.com/iit-jee-algebra/probability/bayes-theorem.aspx
There are also many good youtube videos on Bayes reasoning: https://www.google.com/search?q=youtube+bayes+theorem&oq=youtube+bayes+theorem&aqs=chrome..69i57j69i64.3865j0j7&sourceid…
I will have a future blog post where I start to discuss intersection with AI and specifically with Machine Learning.