Causatily VS Correlation in Bayesian Network

The engine.
Post Reply
Posts: 9
Joined: Sat Mar 27, 2010 2:53 pm

Causatily VS Correlation in Bayesian Network

Post by ben_mrad » Mon Mar 23, 2020 3:09 pm

Let’s suppose that we have two correlated variables (for example mark_in_math and mark_in_physics) but there is no causality between the two variables. For example, the level_of_the_student is a cause for these two variables.

How can we model the Bayesian network containing these two variables (mark in math and mark in physics but not including student_level variable) and of course other variables of the problem while ensuring the correlation between them?

marek [BayesFusion]
Site Admin
Posts: 279
Joined: Tue Dec 11, 2007 4:24 pm

Re: Causatily VS Correlation in Bayesian Network

Post by marek [BayesFusion] » Mon Mar 23, 2020 7:08 pm

Let the two variables be MM and MP. I would advise having a third variable, called LS and the following structure MM<-LS->MP. If you have no data, you will have to estimate the conditional probabilities P(MM|LS) and P(MP|LS). If you have data for MM and MP, you can learn the parameters without LS being measured using the EM algorithm (standard algorithm used in GeNIe/SMILE).
I hope this helps,


Post Reply