Suppose $X_1$,$X_2$,$X_3$,.....,$X_n$ are i.i.d. random variables with a common pmf poisson(λ)

(t = a value)

How would you calculate the below without using intuition (I would appreciate if you could show me with steps )

E $[$$X_3$ |$\sum_{i=1}^n X_i=t$]

  • 1
    $\begingroup$ Hints: (1) You don't even need to know the common distribution of the $X_i$: it suffices that they be exchangeable (and have finite expectations). (2) How would $E\left[X_j\mid \sum_{i=1}^n X_i=1\right]$ compare to the answer when $j\ne 3$? $\endgroup$
    – whuber
    Apr 29, 2019 at 19:12
  • $\begingroup$ is it equal to 1/n ? $\endgroup$
    – GAGA
    Apr 29, 2019 at 20:11
  • $\begingroup$ stats.stackexchange.com/q/374989/119261 $\endgroup$ Apr 27, 2021 at 16:07

1 Answer 1


IID (independent and identically distributed) implies exchangeability: this means that no reordering of the indices in the vector $(X_1, X_2, \ldots, X_n)$ changes its distribution.

Writing $S=X_1+X_2+\cdots+X_n$, notice that $S$ is unchanged whenever the indices are reordered. Therefore, because they can be reordered to place any one of the components first in the vector,

$$E[X_1\mid S=t] = E[X_2\mid S=t] = \cdots = E[X_n\mid S=t].$$

Add these $n$ values--let's temporarily call their common value $\mu$--and use linearity of expectation to express that as

$$\eqalign{ n\mu &= \mu + \mu + \cdots + \mu \\ &= E[X_1\mid S=t] + E[X_2\mid S=t] + \cdots + E[X_n\mid S=t] \\ &= E[X_1+X_2+\cdots+X_n\mid S=t] \\ &= E[S\mid S=t] \\ &= t. }$$

The last step is called "taking out what is known" in conditional expectations.

Now you can solve for $\mu,$ producing (among other things) the result

$$E[X_3 \mid S=t] = \frac{t}{n}.$$

Intuitively, when the $X_i$ are exchangeable they must contribute equally to the sum, so once you assume the sum equals $t,$ each $X_i$ must contribute $1/n$ of that total on average.


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.