1
$\begingroup$

When using prop.test() in R, I noticed that changing my confidence interval does not impact the p value. I use correct = FALSE, conf.level = 0.95, alternative = "two-sided" as parameters in prop.test().

Here, I use 95% CI.

    2-sample test for equality of proportions without continuity correction

data:  c(332, 307) out of c(382, 333)
X-squared = 5.2235, df = 1, p-value = 0.02228
alternative hypothesis: two.sided
95 percent confidence interval:
 -0.097245583 -0.008378365
sample estimates:
   prop 1    prop 2 
0.8691099 0.9219219

Here, I use 99% CI, however, the p-value is the same:

     2-sample test for equality of proportions without continuity correction

data:  c(332, 307) out of c(382, 333)
X-squared = 5.2235, df = 1, p-value = 0.02228
alternative hypothesis: two.sided
99 percent confidence interval:
 -0.111207635  0.005583686 
sample estimates:
   prop 1    prop 2 
0.8691099 0.9219219

I would presume then the p-value is calculated from the X-squared test statistic, which doesn't incorporate confidence intervals in its calculation? I'd love clarification on this. The similar posts I found on StackExchange included a warning error in the output about the chi-square test possibly being incorrect, which my output did not include.

My second question would be that the 99% CI now includes 0. How, then, can we have a p value < 0.05?

$\endgroup$
1
$\begingroup$

The $p$-value is for the test that the two proportions are equal. You can then make that test at whatever significance level you like by seeing whether the $p$-value is less than your chosen criterion value.

The $p$-value is less than 0.05 but not less than 0.01 so it is consistent with the confidence intervals. The 95% one does not include zero, the 99% one does.

$\endgroup$
1
  • $\begingroup$ Thank you, that makes sense and is now obvious reading your response. Probably just being too deep in code/stats and needed to step back and look at it. Thanks! $\endgroup$ – grantog Dec 6 '17 at 17:33

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.