R devel - Oct 2006 - Confidence interval calculation in prop.test (PR#9325)

Full_Name: Richard Johnston
Version: 2.4.0
OS: OS X
Submission from: (NULL) (69.169.0.241)


The confidence interval calculation for prop.test appears incorrect when
alternative="greater" .  The upper limit is always set to 1.0000.  The
lower
limit appears to be correct.
> total=c(250,250)
> success=c(55,31)
> prop.test(success,total,alternative="greater",correct=TRUE)
	2-sample test for equality of proportions with continuity correction

data:  success out of total 
X-squared = 7.4289, df = 1, p-value = 0.003209
alternative hypothesis: greater 
95 percent confidence interval:
 0.03693065 1.00000000 
sample estimates:
prop 1 prop 2 
 0.220  0.124

On 10/30/2006 8:50 PM, rmjohnston at mac.com wrote:
> Full_Name: Richard Johnston
> Version: 2.4.0
> OS: OS X
> Submission from: (NULL) (69.169.0.241)
> 
> 
> The confidence interval calculation for prop.test appears incorrect when
> alternative="greater" .  The upper limit is always set to 1.0000.
The lower
> limit appears to be correct.
That looks like a one-sided confidence interval, i.e. an interval of the 
type Delta > 0.0369, where Delta is the difference in proportions.  It 
is then clipped to [-1, 1], since that's the largest possible range for 
the difference.

What were you expecting to see?

Duncan Murdoch

> 
>> total=c(250,250)
>> success=c(55,31)
>> prop.test(success,total,alternative="greater",correct=TRUE)
> 
> 	2-sample test for equality of proportions with continuity correction
> 
> data:  success out of total 
> X-squared = 7.4289, df = 1, p-value = 0.003209
> alternative hypothesis: greater 
> 95 percent confidence interval:
>  0.03693065 1.00000000 
> sample estimates:
> prop 1 prop 2 
>  0.220  0.124
> 
> ______________________________________________
> R-devel at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel