in my lecture our professor said the following: Given a population of N, where each unit has the same probability to get chosen, and a sampling method where the population is divided to samples of size n for all possible samples, the sampling probability for each unit in the population will be n/N.

It sounds rather counter intuitive - I know it's true for a simple random sample, but how can I derive the general case? Thanks.


1 Answer 1


I'm not sure this is the best way to prove this, but I think it does the job.

Suppose $\{M\}_i \subset \{1,\ldots, N\}, i \in 1,\ldots,K, |\{M\}| = n$ is a collection of $K$ possible samples of size $n$ from the population, not necessarily containing all possible subsets of size $n$. Let $\{m\}$ be sampled from $\{M\}_i$ with probabilities $p_1, \ldots, p_K$, not necessarily equal. But assume that this entire construction is such that the probability that individual $j$ is in $\{m\}$, say $q_j$, is equal for all $j$, $q_j = q, j \in \{1, \ldots, N\}$

Define $I_j, j \in \{1,\ldots,N\}$ to be an indicator function, taking value 1 if $j$ is in $\{m\}$ and 0 otherwise. Then the expected value of $I_j$ is equal to $q_j = q$.

Sum the $N$ indicator functions, $\sum_{j=1}^{N} I_j$. The sum now counts how many individuals were in $\{m\}$. But $|\{m\}| = n$ for all possible samples. Therefore $\sum_{j=1}^{N} I_j = n$. Take expectations of both sides. As $n$ is constant, $E[n] = n$.

$E[\sum_{j=1}^{N} I_j] = n$

By linearity of expectation,

$\sum_{j=1}^{N} E[I_j] = n$

Since $E[I_j] = q$,

$\sum_{j=1}^{N} q = n$

$ q = n/N$

  • $\begingroup$ Great proof. super clear and understandable. thanks! $\endgroup$
    – Omri. B
    Jul 23, 2021 at 17:36

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.