Why do my binomial significance test and CI test results do not match?

30 Views Asked by At

Could anyone please explain in a simple manner why do I get different results or point to the mistake I make? Do these calculations even make sense given the sample sizes and binomial outcomes?

I have two independent equal binomial samples and I want to test whether the observed difference between them is significant.

$n_1$ = $n_2$ = 112

"Success rate" in the first sample: $\hat p_1$ = 55.4

"Success rate" in the second sample:t: $\hat p_2$ = 46.4

Hence, I test the following hypothesis: $H_0:p_1=p_2$ versus $H_A:p_1\neq p_2$ by using the following formula:

$$z=\frac{\hat p_1-\hat p_2}{\sqrt{\hat p(1-\hat p)\left(\frac{1}{n_1}+\frac{1}{n_2}\right)}}$$

where $\hat p=\frac{n_1\hat p_1+n_2\hat p_2}{n_1+n_2}$.

I get that the $z$ = 2.69 which is larger than $z_{\alpha/2}=1.96$. Hence, I can reject the null hypothesis with 95% confidence.

However, if I compute CI for a difference in proportions using this formula:

$$\hat p_1 - \hat p_2±z\sqrt{\frac{\hat p_1(1-\hat p_1)}{n_1}+\frac{\hat p_2(1-\hat p_2)}{n_2}}$$

I get that the difference falls into CI of $(-0.04,0.22)$ which suggests that 9 p.p. difference between $\hat p_1$ and $\hat p_2$ is actually insignificant given that $0$ is in the CI.

Thanks for your insights.

1

There are 1 best solutions below

1
On BEST ANSWER

Looks like you made a mistake in calculating $z$ in the first part. With $\hat p=0.509$, I get

$$ z=\frac{0.554-0.464}{\sqrt{0.509(1-0.509)\left(\frac1{112}+\frac1{112}\right)}}\approx1.35\;. $$