I'm having a little trouble with the proof that the expected value of $x_i$ is $ \bar{X} $.

What I have is

$E[x_i]=\sum_{j=1}^{N}X_j Pr(x_i=X_j) $


$Pr(x_i=X_j) = 1/N $

This is the bit I can't understand, how does that probability evaluate to that value.

I know the denominator is how many ways you can choose n out N. I think that the numerator should be how many ways you can choose (n-1) out of (N-1). But I seem to have an extra n.

Any advice ?

  • $\begingroup$ $x_i$ is the random variable and the $X_j$'s are the possible values that it can take? $\endgroup$ Commented Apr 5, 2014 at 18:17
  • $\begingroup$ $x_i$ is a unit in a random sample of size n, from a population $X_j$ where j=1..N $\endgroup$ Commented Apr 5, 2014 at 18:23
  • $\begingroup$ Then, the RHS of your first expression is a weighted sum of random variables and hence a random variable itself -it cannot express an unconditional expected value (the LHS), which is a number. $\endgroup$ Commented Apr 5, 2014 at 18:36
  • $\begingroup$ @AlecosPapadopoulos I think the $X_i$ represent potentially values from a finite population - they are not random - and $x_i$ are realizations from a simple random sample of that population. Hence $\bar X$ is a number. $\endgroup$
    – guy
    Commented Apr 5, 2014 at 20:42
  • 1
    $\begingroup$ @guy Yes, that could be - and then $\bar X$ would be the expected value. But then, using the sample-mean notation to denote $E(X)$ is just confusing practice. $\endgroup$ Commented Apr 5, 2014 at 21:08

2 Answers 2


Every $X_j$ is equally likely to be selected (this isn't true for every population, but it is a common situation [for example, a die, where 1, 2, 3, 4, 5, 6 are all equally likely to be thrown]). So since there are $N$ possible values, the probability of each is ${1 \over N}.$

So you have \begin{align}\mathrm E[x_i]&=\sum_{j=1}^N X_j\,\mathrm P[x_i=X_j]\\ &={1 \over N} \sum_{j=1}^N X_j\\ &={1 \over N} (N \bar{X})\\&= \bar{X}\;.\end{align}


If, as some of the comments have suggested, that $x_i$ is a single (discrete) random variable, and the $X_j$ are all the potential values that $x_i$ can take, then basically the definition of expectation of $x_i$ is $\mathbb{E}(x_i) := \sum_{j=1}^n X_j \mathbb{P}(x_i = X_j)$. However, in no way does this imply anything about the values that $\mathbb{P}(x_i = X_j)$ can take (other than that they are non-negative and must sum to unity over $j$)!! The probabilities $\mathbb{P}(x_i = X_j)$ define the distribution of the discrete random variable $x_i$, and since the expectation of $x_i$ is just a quantity associated with its distribution, then $\mathbb{E}(x_i)$ changes as the probabilities $\mathbb{P}(x_i = X_j)$ change.

Perhaps you're missing something from your question - maybe you've been given a question where you know that $\mathbb{E}(x_i)$ is fixed to be equal to a given value, and you're also given the values $X_j$, then it may be that you're asked to solve for all possible values of $\mathbb{P}(x_i = X_j)$?

Several comments:

1) Your use of capitals to represents values and lower-case to represent random variables is very confusing - mathematicians use the opposite convention.

2) Your use of the index $i$ is very confusing - it is not used anywhere (usefully) in your question. Come to think of it, your question mentions some combinatorial statement - perhaps you have written your question badly, in which case I urge you to rephrase your question. If this question was prompted by some other question that you're trying to solve, I would ask that original question.


EDIT in response to comment:

Ah, if you're trying to prove that the sample mean is unbiased estimator of the population mean, then let's start with only two random variables $x_1$ and $x_2$ (using your notation). We need to prove that the expectation operator is linear i.e. $\mathbb{E}(x_1 + x_2) = \mathbb{E}(x_1) + \mathbb{E}(x_2)$ (note that if this is the case, then it's trivial to extend this to the case of there being $n$ random variables - use induction). So to prove linearity (and letting the $X_1$ and $X_2$ are the possibles values that $x_1$ and $x_2$ can take, respectively):

$\begin{align} \mathbb{E}(x_1 + x_2) &= \sum_{X_1, X_2} (X_1+X_2) \mathbb{P}(x_1 = X_1, x_2 = X_2) \\ &= \sum_{X_1} \sum_{X_2} \left[X_1 \mathbb{P}(x_1 = X_1, x_2 = X_2) + X_2 \mathbb{P}(x_1 = X_1, x_2 = X_2)\right] \\ &= \sum_{X_1} \sum_{X_2} X_1 \mathbb{P}(x_1 = X_1, x_2 = X_2) + \sum_{X_2} \sum_{X_1} X_2 \mathbb{P}(x_1 = X_1, x_2 = X_2) \\ &= \sum_{X_1} \left[ X_1 \sum_{X_2} \mathbb{P}(x_1 = X_1, x_2 = X_2)\right] + \sum_{X_2} \left[ X_2\sum_{X_1} \mathbb{P}(x_1 = X_1, x_2 = X_2)\right] \\ &= \sum_{X_1}X_1\mathbb{P}(x_1 = X_1) + \sum_{X_2}X_2\mathbb{P}(x_2 = X_2)\\ &= \mathbb{E}(x_1) + \mathbb{E}(x_2) \,\,\,\,\,\text{QED} \end{align}$

From this, it should be easy to see that $\mathbb{E}(\bar{x})$ is equal to the population mean.

  • $\begingroup$ I appreciate all the comments on notation, but it is taken exactly from Sampling Techniques, Cochran(1977) which is the 'bible' for sampling theory. Also, all I am trying to do is work through the proof that the sample mean $\bar{x}$ is an unbiased estimator of the population mean $\bar{X}$. $\endgroup$ Commented Apr 6, 2014 at 16:35
  • $\begingroup$ Please see my edit. $\endgroup$
    – queenbee
    Commented Apr 6, 2014 at 16:59

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.