Take the following random variables $Y,\epsilon_0,\epsilon_1,V_0,V_1$, with finite supports $\mathcal{Y}\equiv \{0,1\},\mathcal{E}_0, \mathcal{E}_1,\mathcal{V}_0,\mathcal{V}_1$, respectively.

Suppose $$P_Y(1|e_0,e_1,v_0,v_1)=\mathbb{1}\{e_0+v_0\leq e_1+v_1\}$$ $\forall e_0\in \mathcal{E}_0$,$\forall e_1\in \mathcal{E}_1$,$\forall v_0\in \mathcal{V}_0$,$\forall v_1\in \mathcal{V}_1$, where $\mathbb{1}\{a\leq 0\}$ is $1$ if $a\leq 0$ and $0$ otherwise.

I want to derive the unconditional probability distribution of $Y$.

I can use the law of total probability, that is $$ (\diamond) \text{ }\text{ }\text{ }P_Y(1)=\sum_{e_0,e_1,v_1,v_0} \mathbb{1}\{e_0+v_0\leq e_1+v_1\} \times P_{\epsilon_0,\epsilon_1,V_0,V_1} $$

Question: Let $W_0\equiv \epsilon_0+V_0$ and $W_1\equiv \epsilon_1+V_1$. My intuition without going through the law of total probability is that $$ (\diamond \diamond) \text{ }\text{ }\text{ }P_Y(1)= \sum_{w_0,w_1} \mathbb{1}\{w_0\leq w_1\}\times P_{W_0, W_1}(w_0, w_1) $$ In other words, knowing the joint distribution of the sums, $P_{W_0, W_1}$ (rather than the joint distribution of each term, $P_{\epsilon_0,\epsilon_1,V_0,V_1}$) is sufficient. Is this correct? If yes, how do we show that $(\diamond)=(\diamond \diamond)$?


1 Answer 1


Yes, it is sufficient. Letting $w_0=e_0+v_0$ and $w_1=e_1+v_1$, you can safely add functions of your RVs to the given side of the probability formulas, and then rename them: $$\begin{align}P_Y(1|e_0,e_1,v_0,v_1)&=P_Y(1|e_0,e_1,v_0,v_1,e_0+v_0,e_1+v_1)\\&=P_Y(1|e_0,v_0,e_1,v_1,w_0,w_1)\\&=\mathbb{1}\{w_0\leq w_1\}\end{align}$$ Since, the RHS probability doesn't depend further on $e_i,v_i$, we can safely omit them in LHS: $$P_Y(1|w_0,w_1)=\mathbb{1}\{w_0\leq w_1\}$$ Now, you can rewrite your second formula, i.e. $\diamond\diamond$, as follows by changing the first multiplicand inside the summation expression: $$P_Y(1)=\sum_{w_0,w_1}P_Y(1|w_0,w_1)\times P_{W_0,W_1}(w_0,w_1)$$ this is always true due to Total Probability Law, which also applies to $(\diamond)$ saying also that these two are equivalent because they are both different versions of $P_Y(1)$.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.