4
$\begingroup$

I'm trying to determine n, the sample size.

We have $N=580$ and we want to estimate population proportion, $p$, at a 95%-CI and a margin of error of $0.10$, then I get

$$n=Npq / (N-1)D + pq$$

where $D= B^2 / 4$ to be about $n=86$.

But what if instead we keep that same margin of error at $0.10$ and now only want a 90%-CI, if we use the same formula don't we get the same $n$?

Am I using the formula wrong here?

$\endgroup$
  • 1
    $\begingroup$ Proportion of what? Is this binomial data? What's B? Why do you want a smaller N than 580? Are you sampling from the original 580 or is it a new random sample? What are you trying to find out from the sample? $\endgroup$ – John Oct 3 '13 at 6:51
3
$\begingroup$

The formulas concerning the calculation of the sample size to estimate a proportion $p$ in for finite populations are provided on this website. It also contains the derivations.

In short, the sample size necessary for estimating a population proportion $p$ of a finite population with $(1-\alpha)100\%$ confidence and a margin of error no larger than $\epsilon$ is:

$$ n = \frac{m}{1+\frac{m-1}{N}} $$ where $$ m=\frac{z_{1-\alpha/2}^{2}\hat{p}(1-\hat{p})}{\epsilon^{2}}. $$ $N$ denotes the population size, $z_{1-\alpha/2}$ the $(1-\alpha/2)$-quantile of the standard normal distribution and $\hat{p}$ the estimated proportion.

For $N=580, \alpha=0.05, \epsilon=0.1, \hat{p}=0.5$ (i.e. 95% confidence), we get $n\approx 83$. If we take $\alpha = 0.1$ which corresponds to 90% confidence, we get $n\approx 61$.

The more precise we want our estimate of the popultion proportion to be, the higher our sample size needs to be.

This means that the lower $\alpha$, the higher the necessary sample size will be. The following graph illustrates this (for $N=580, \epsilon=0.1, \hat{p}=0.5$): Sample size alpha

The necessary sample size will also increase with decreasing margin of error $\epsilon$ (note the reversed $x$-axis; graph for $N=580, \alpha=0.05, \hat{p}=0.5$):

Sample size margin of error

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ Why, mathematically, do I need to know the estimated population proportion ex-ante to determine sample size? I have been looking for the reasoning, but also the site you linked does not explan that and just assumes a p-hat of 0.5. Do we always assume p-hat of 0.5 when we have not taken a pre-sample? Does our estimated required sample size change during the sampling process the closer we get to the population p? So if we have no clue about p-hat and after some sample n we have not yet found a success, we just stop and conclude p is between 0 and the error we initially set? $\endgroup$ – iraserd Mar 2 '17 at 10:14
  • $\begingroup$ And, should I post the comment above in a separate question? $\endgroup$ – iraserd Mar 2 '17 at 10:15
  • $\begingroup$ Also, is this true for small p? And what is a small p? 10%? 3%? 0.3%? $\endgroup$ – iraserd Mar 2 '17 at 10:19
  • $\begingroup$ @iraserd $\hat{p}=0.5$ is usually taken because it's the worst-case scenario. The standard error is largest when $p=0.5$. So by taking $\hat{p}=0.5$ we're on the safe side. If you have an a-priori estimate of the population proportion, you could take that instead. You certainly could design a sequential approach, where you update your estimate on the fly. I assume that Bayesian methods would be suited for that. This approach is also valid for small $p$, as far as I'm aware. What a small $p$ is is subjective. $\endgroup$ – COOLSerdash Mar 2 '17 at 11:03

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.