1

Let $X_{1}, X_{2}, ..., X_{n}$ be independent and identically distributed random variables with finite expected value. Find $\mathbb{E}[X_{1}|(X_{1}+X_{2}+ ... + X_{n})=x]$. Assuming that $f_{(X_{1}+X_{2}+ ... + X_{n})}(x)\neq 0$, I must have the expected value is $\frac{x}{n}$. ¿Could you give me a hint to do it, please?

1 Answers1

3

By symmetry, $$\mathbb E[X_1\mid X_1+\dots+X_n]=\mathbb E[X_2\mid X_1+\dots+X_n]=\dots=\mathbb E[X_n\mid X_1+\dots+X_n].$$ But by linearity $$\sum_{i=1}^n\mathbb E\left[X_i\,\middle|\,\sum_{j=1}^n X_j=x\right]=\mathbb E\left[\sum_{i=1}^nX_i\,\middle|\,\sum_{j=1}^nX_j=x\right]=x,$$ so we have $\mathbb E[X_i\mid X_1+\dots+X_n=x]=\frac{x}{n}$.

jlammy
  • 9,164
  • 1
    Maybe worth pointing out where/if the independence assumption is used ("by symmetry" can be a bit of a magic wand). – Clement C. Jan 19 '21 at 03:32
  • Jlammy and @ClementC. could you please look at my recent question? – Daman Jan 19 '21 at 03:59
  • I'm also curious where/if the independence assumption is used. – Pedro Amaral Jan 19 '21 at 04:10
  • 1
    @PedroAmaral It has to be: if not, the first fact stated doesn't hold. For instance, take $X_1,X_2,X_3$ all (marginally) uniform on ${-1,1}$, with $X_3$ independent of $(X_1,X_2)$ but $X_1=-X_2$ always. Then $X_1+X_2+X_3=X_3$ (the first two cancel), so $$\mathbb{E}[X_1 \mid X_1+X_2+X_3] = \mathbb{E}[X_1 \mid X_3] = 0$$ and $$\mathbb{E}[X_2 \mid X_1+X_2+X_3] = \mathbb{E}[X_2 \mid X_3] = 0$$ but $$\mathbb{E}[X_3 \mid X_1+X_2+X_3] = \mathbb{E}[X_3 \mid X_3] = X_3 \in {-1,1}$$ – Clement C. Jan 19 '21 at 04:15
  • @ClementC. Ah very tricky! Is there a simple way to prove that the expectations are equal when the $X_i$'s are all independent? I came up with the following but it is somewhat messy:

    $$ f_{X_1,\sum X_i}(x_1, x) $$ $$ = \idotsint f_{X_1, \sum X_i,X_2, ...,X_n}(x_1, x, x_2, .., x_n)dx_2...dx_n $$

    $$ = \idotsint \delta_{x,\sum x_i} \cdot f_{X_1,X_2, ...,X_n}(x_1, x_2, .., x_n)dx_2...dx_n $$

    $$ = \idotsint \delta_{x,\sum x_i} \cdot f_{X_1,X_2, ...,X_n}(x_1, x_2, .., x_n)dx_2...dx_n $$

    $$ = \idotsint \delta_{x,\sum x_i} \cdot f(x_1)f(x_2)...f(x_n)dx_2...dx_n $$

    – Pedro Amaral Jan 19 '21 at 05:14
  • @PedroAmaral I think the notion needed here is exchangeability, which is implied by i.i.d. Once you have that, the symmetry argument is basically by definition. – Clement C. Jan 19 '21 at 06:35