I am running the following simple Z test for proportions in R:

prop.test(x = c(10, 20), n = c(100, 500))

The results are as follows:

2-sample test for equality of proportions with continuity correction

data:  c(10, 20) out of c(100, 500)
X-squared = 5.1158, df = 1, p-value = 0.02371
alternative hypothesis: two.sided
95 percent confidence interval:
 -0.007256322  0.127256322
sample estimates:
prop 1 prop 2 
  0.10   0.04 

How is it possible for the p value of the Z test to be below 0.05, but for the 95% confidence interval to straddle zero? I've found this question on the behaviour of the prop.test function itself and this one on the alternative ways of constructing the CI but I don't think either explains what's going on in this case.

The top answer to this question suggests I'm right in thinking this output should be impossible!


I think one difficulty may be in the normal (or chi-squared) approximation for relatively small sample sizes. Also, the 'test' part of the output uses the null hypothesis $p_1 = p_2$ to get a pooled estimate for the standard error, whereas the '95% CI' does not assume $p_1 = p_2.$

Fisher's exact test fails to reject the null hypothesis at the 5% level.

TAB = rbind(c(10,20), c(90,80))
     [,1] [,2]
[1,]   10   20
[2,]   90   80


        Fisher's Exact Test for Count Data

data:  TAB
p-value = 0.07343
alternative hypothesis: 
 true odds ratio is not equal to 1
95 percent confidence interval:
 0.1754757 1.0691978
 sample estimates:
 odds ratio 

Also, prop.test in R is roughly equivalent to using a chi-squared test on the $2\times 2$ table. With Yates' continuity correction, this procedure does not reject.


        Pearson's Chi-squared test 
        with Yates' continuity correction

data:  TAB
X-squared = 3.1765, df = 1, p-value = 0.07471

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.