How is the author's application of the law of total expectation consistent with the definition?

88 Views Asked by At

I have the following solution:

Let $$Y_j = \begin{cases} 1, & \text{if $j$th toss is a head} \\ 0, & \text{if $j$th toss is a tail} \end{cases}$$

Let $X$ denote the waiting time to $HH$. First, we note that, if any toss is a tail, then the experiment regenerates. In particular, if $Y_1 = 0$, then the experiment regenerates. Therefore, we have that

$$E[X \vert Y_1 = 0] = E[X] + 1.$$

On the other hand, for the case where the first toss is a head, the law of total expectation -- with extra conditioning on $Y_1 = 1$ -- gives us that

$$\begin{align} E[X \vert Y_1 = 1] &= \sum_{y = 0}^1 E[X \vert Y_2 = y, Y_1 = 1] P(Y_2 = y \vert Y_1 = 1) \\ &= E[X \vert Y_2 = 0, Y_1 = 1] P(Y_2 = 0) + E[X \vert Y_2 = 1, Y_1 = 1] P(Y_2 = 1) \end{align}$$

Wikipedia gives the law of total expectation as follows:

If ${\displaystyle {\left\{A_{i}\right\}}_{i}}$ is a finite or countable partition of the sample space, then

$${\displaystyle \operatorname {E} (X)=\sum _{i}{\operatorname {E} (X\mid A_{i})\operatorname {P} (A_{i})}.}$$

How is the author's application of the law of total expectation consistent with the definition?

I would greatly appreciate it if people would please take the time to explain this.

1

There are 1 best solutions below

0
On

To answer "How is the author's application of the law of total expectation consistent with the definition?"

I think the Tower Property can explain it better. Tower property: For sub-σ-algebras $\mathcal H_1\subset \mathcal H_2\subset \mathcal F$ we have $$ E(E(X\mid \mathcal H_2) \mid \mathcal H_1)= E(E(X\mid \mathcal H_1)\mid \mathcal H_2)=E(X\mid \mathcal H_1)$$

Define $\mathcal H_1=\sigma(Y_1)$ and $\mathcal H_2=\sigma(Y_1,Y_2)$ so $\mathcal H_1 \subset \mathcal H_2\subset \mathcal F$ $$\begin{align} \mathbb E[X \vert Y_1 ] &\overset{\text{Tower property}}{=} \mathbb E\left( \mathbb E[X \vert Y_1 ] \mid Y_1,Y_2\right) \\ &\overset{\text{Tower property}}{=} \mathbb E\left( E[X \vert Y_1 ,Y_2 ] \mid Y_1\right) \\ &= \mathbb E\left( g(Y_1 ,Y_2 ) \mid Y_1\right) \\ &= \sum_{y} g(Y_1 ,Y_2=y )\mathbb P(Y_2 = y \vert Y_1 ) \\ &= \sum_{y } \mathbb E[X \vert Y_2 = y, Y_1] \mathbb P(Y_2 = y \vert Y_1 )= \cdots \end{align}$$