For the sample proportion $\hat{p}$, standard error is known as $\sqrt\frac{\hat{p}(1-\hat{p})}{n}$.

I wonder whether this is the unbiased estimator for the standard deviation of the sample proportion, which is $\sqrt\frac{p(1-p)}{n}$ where $p$ is the true proportion.

  • $\begingroup$ An estimate of the (large sample) standard error of $\hat p$ is $\sqrt\frac{\hat{p}(1-\hat{p})}{n}$, because standard error of $\hat p$ is just the s.d. of $\hat p$. $\endgroup$ – StubbornAtom Mar 16 at 15:57

Suppose that $X\sim \text{Bin}(n,p)$, so that $\hat{p}=X/n$. It follows that $$E\hat{p}=np/n=p; \quad {E\hat{p}^2}=n^{-2} EX^2=\frac{np(1-p)+n^2p^2}{n^2}=\frac{p(1-p)}{n}+p^2.$$ So $$ E\frac{\hat{p}(1-\hat{p})}{n}=\frac{1}{n}E\hat{p}-\frac{1}{n}E\hat{p}^2=\frac{p(1-p)}{n}-\frac{p(1-p)}{n^2} $$ By Jensen's inequality (square root function is concave), it follows that $$ E\sqrt{\frac{\hat{p}(1-\hat{p})}{n}}\leq\sqrt{E\frac{\hat{p}(1-\hat{p})}{n}}=\sqrt{\frac{p(1-p)}{n}-\frac{p(1-p)}{n^2}}<\sqrt{\frac{p(1-p)}{n}} $$ as long as $p\neq 0,1$.

It follows that the estimator is not unbiased.

  • $\begingroup$ Thanks! Then, what would be the unbiased estimator? Moreover, is there any reason to use the standard error after replacing $p$ in the s.d. with $\hat{p}$? $\endgroup$ – user3509199 Mar 16 at 15:59

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.