1
$\begingroup$

I'm supposed to be writing R functions to calculate the Wald CI, Wilson CI, and Agresti-Coull CI for binomial proportion, I have a source that provides the equations, but I am completely lost. I am looking at the equations, but have no clue where I am supposed to be getting all the values I need.

Ward CI = $p̂ ± k n^{-1/2} (p̂q̂)^{1/2}$

Wilson CI = $(X+k^2/2)/(n+k^2)±(kn^{1/2})/(n+k^2)((p̂q̂+k^2)/(4n))^{1/2}$

Agresti-Coull CI = $p̂ ± k(p̂q̂)^{1/2}n̂^{1/2}$

I'm pretty sure I have most of it, but I am can't figure out what the value of k is. n=40 p = 0.2 alpha=0.05 q=0.8 These formulas are from http://projecteuclid.org/download/pdf_1/euclid.ss/1009213286 I was wondering if anyone could break these down and explain them to me?

| cite | improve this question | |
$\endgroup$
  • $\begingroup$ Where did you get the formulas? From something like this? en.wikipedia.org/wiki/Binomial_proportion_confidence_interval $\endgroup$ – Glen_b Oct 17 '16 at 1:55
  • $\begingroup$ The formulas were from formulas 1, 4 , and 5 in this article projecteuclid.org/download/pdf_1/euclid.ss/1009213286 $\endgroup$ – Jamie Leigh Oct 17 '16 at 1:58
  • $\begingroup$ The article defines the terms (like $\hat{p}$ and $n$) that it uses! $\endgroup$ – Glen_b Oct 17 '16 at 1:59
  • $\begingroup$ The one I am really struggling with is K, it says k=Z(alpha/2) but how do I get Z? $\endgroup$ – Jamie Leigh Oct 17 '16 at 2:04
  • $\begingroup$ The article defines $z$ also -- right before equation 1 it says $\kappa=z_{\alpha/2}=\Phi^{-1}(1-\alpha/2)$ $\endgroup$ – Glen_b Oct 17 '16 at 2:06
1
$\begingroup$

$X$ is the number of successes. $n$ is the number of trials. $\hat{p}=X/n$. $\hat{q}=1-\hat{p}$.

The article defines $\kappa$ in terms of $z$ and in turn defines $z$ right before equation 1:

$\kappa=z_{\alpha/2}=\Phi^{-1}(1-\alpha/2)$

$\Phi$ is the standard normal cdf.

$\Phi^{-1}$ is the inverse function of the standard normal cdf.

$\Phi^{-1}(1-\alpha/2)$ is the $1-\alpha/2$ upper tail quantile of the normal distribution.

For example for a 95% CI, $1-\alpha=0.95$ so $1-\alpha/2=0.975$ and $\Phi^{-1}(1-\alpha/2)\approx 1.96$.

In the article the Agresti-Coull interval is defined in terms of $\tilde{p}$ not $\hat{p}$. However, $\tilde{p}$ is of the same form as $\hat p$ but with an additional $\kappa^2/2$ successes and an additional $\kappa^2/2$ failures. For $\alpha=0.05$ that adds just under $2$ to $X$ and just under $4$ to $n$.

I think that covers all of the terms in equations 1, 4 and 5.

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ I'm still not sure what the Φ is or how to calculate that. Is it a standard from a table? $\endgroup$ – Jamie Leigh Oct 17 '16 at 2:14
  • $\begingroup$ @Jamie If you're working by hand you'd use normal tables. If you're on a computer you call a standard function to evaluate it. Suitable statistical/mathematical function libraries will either offer $Φ$ and $Φ^{-1}$ or the error function ($\text{erf}$) and its inverse from which $Φ$ and $Φ^{-1}$ are readily calculated. In R $Φ$ and $Φ^{-1}$ are given by pnorm and qnorm respectively. Try qnorm(0.975). See ?Distributions for a list of links to the various distributions in vanilla R for which the d,p,q and r functions work. $\endgroup$ – Glen_b Oct 17 '16 at 2:21

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.