Does a Degree of Conditional Independence Make Sense?

42 Views Asked by At

Conditional independence implies

$$p(A, B \, | \, C) = p(A \, | \, C)\:p(B \, | \, C).$$

If the two sides are equal, then $A$, and $B$ are independent conditional on $C$ and there's nothing more to it.

But what if they are roughly equal? Does that suggest anything? If you were to compute conditional independence empirically from a dataset, how could you then decide how close to equality is close enough?

1

There are 1 best solutions below

0
On

Since independent and uncorrelated are the same for two Bernoulli variables, and since indicator functions for random events are Bernoulli, the simplest answer is to get a confidence interval (or a credible interval, if you're a Bayesian) for the correlation of $I_A$ with $I_B$ and see whether it includes $0$.