Abhijeet Mulgund's Personal Webpage

Search

Search IconIcon to open search

Conditional Probability

Last updated Nov 1, 2022

TODO - need to rope in Regular Conditional Distribution. This page can be redefined in terms of Regular Conditional Distribution of Identity Function.

# Definition 1

Let $(\Omega, \mathcal{B}, \mathbb{P})$ be a Probability Space and let $A \in \mathcal{B}$ be an Event. Let $\mathcal{G} \subset \mathcal{B}$ be a sub-Sigma Algebra. Then the Conditional Probability of $A$ given $\mathcal{G}$ is $$\mathbb{P}(A|\mathcal{G}) = \mathbb{E}(1_{A}|\mathcal{G})$$

Similarly to Conditional Expectation, if $Y$ is a Random Variable on $\Omega$ then $$\mathbb{P}(A|Y) = \mathbb{P}(A|\sigma(Y))$$ and if $B \in \mathcal{B}$ then $$\mathbb{P}(A|B) = \mathbb{P}(A|1_{B})(\omega)$$ for (any) $\omega \in B$.

# Definition 2

Let $(\Omega, \mathcal{B}, \mathbb{P})$ be a Probability Space and let $A, B \in \mathcal{B}$ be Events such that $\mathbb{P}(B) \neq 0$. Then the Conditional Probability of $A$ given $B$ is $$\mathbb{P}(A|B) = \frac{\mathbb{P}(A \cap B)}{\mathbb{P}(B)}$$

# Remarks

  1. Since Indicator Functions on Measureable Sets are in L+ we know $1_{A} \in L^{+}(\mathcal{B}) \subset \bar{L^{1}}(\mathcal{B})$, and using Conditional Expectation makes sense in our definition.

  2. We can also write the two auxiliary definitions in Definition 1 as $$\mathbb{P}(A|Y) = \mathbb{E}(1_{A}|Y)$$ and $$\mathbb{P}(A|B) = \mathbb{E}(1_{A}|B)$$

  3. We should check that Definition 1 naturally leads to Definition 2:

    Proof: Suppose $A, B \in \mathcal{B}$ so that $\mathbb{P}(B) \neq 0$. Then $$\mathbb{P}(A|B) = \mathbb{E}(1_{A}|B) = \frac{\mathbb{E}(1_{B}1_{A})}{\mathbb{P}(B)} = \frac{\mathbb{E}(1_{B \cap A})}{\mathbb{P}(B)} = \frac{\mathbb{P}(B \cap A)}{\mathbb{P}(B)}$$where the second equality follows because Conditional Expectation with respect to a Set is Integral divided by Probability. $\blacksquare$

    Thus, we can use whichever definition is more convenient.

  4. Suppose $B \in \mathcal{B}$ so that $\mathbb{P}(B) \neq 0$. Then $(\Omega, \mathcal{B}, \mathbb{P}(\cdot | B))$ is a Probability Space on $\Omega$.

    Proof: All we need do is check $\mathbb{P}(\cdot|B)$ is a valid Probability Measure. We will rely on the fact that $\mathbb{P}$ is a probability measure

    1. If $A \in \mathcal{B}$, $\mathbb{P}(A|B) = \frac{\mathbb{P}(A \cap B)}{\mathbb{P}(B)} \in [0,1]$ since $A \cap B \subset B$ so $0 \leq \mathbb{P}( A \cap B) \leq \mathbb{P}(B)$. $\checkmark$
    2. $\mathbb{P}(\Omega|B) = \frac{\mathbb{P}(\Omega \cap B)}{\mathbb{P}(B)} = \frac{\mathbb{P}(B)}{\mathbb{P}(B)} = 1$. $\checkmark$
    3. Suppose $(A_{n}){n=1}^{\infty} \subset \mathcal{B}$ are Mutually Disjoint Events. Then $$\mathbb{P}(\bigsqcup\limits{n \in \mathbb{N}}A_{n}|B) = \frac{\mathbb{P}\Big( \big(\bigsqcup\limits_{n \in \mathbb{N}}A_{n} \big) \cap B \Big)}{\mathbb{P}(B)} = \frac{\mathbb{P}\Big( \bigsqcup\limits_{n \in \mathbb{N}}(A_{n} \cap B) \Big)}{\mathbb{P}(B)} = \sum\limits_{n=1}^{\infty} \frac{\mathbb{P}(A_{n} \cap B)}{\mathbb{P}(B)} = \sum\limits_{n=1}^{\infty} \mathbb{P}(A|B)$$ $\checkmark$

    Thus $\mathbb{P}(\cdot | B)$ is a Probability Measure. $\blacksquare$

  5. I used to think about Conditional Expectation as just an Expectation taken with the Conditional Probabilitys. Here, however we’ve taken Conditional Expectation to be more fundamental. My old way of thinking about this is still valid, as I show in TODO

# Other Outlinks