st: Confidence interval of difference between two proportions and -csi-

[Garry Anderson <g.anderson@unimelb.edu.au>]

Hi,

I am enquiring if a more appropriate method could be used please to
calculate the 95% CI of the difference between two proportions in the
-csi- command? 

At the moment it is possible for the upper bound of the confidence
interval of the difference between two proportions to be greater than
1.0. I realize that the approximation that is used is not appropriate for
small sample sizes, however I think that reporting of results that are
impossible should be avoided.

For example
-csi 5 1 0 4,e-


                
|   Exposed   Unexposed 
|     Total
-----------------+------------------------+----------
           Cases
|        
5           1 
|         6
        Noncases
|        
0           4 
|         4
-----------------+------------------------+----------
           Total
|        
5           5 
|        10
                
|                       
|
            Risk
|        
1          .2 
|        .6
                
|                       
|
                
|      Point estimate    | 
[95% Conf. Interval]
                
|------------------------+----------------------
 Risk difference
|              
.8       |  
.449391    1.150609  
      Risk ratio
|               
5       |  
.866228    28.86076  
 Attr. frac. ex.
|              
.8       | -.1544305   
.9653509  
 Attr. frac. pop |        
.6666667       |
                
+-----------------------------------------------
                               
1-sided Fisher's exact P = 0.0238
                               
2-sided Fisher's exact P = 0.0476



The value of 1.15 is not desirable.

One potential solution to this is to use a method that is described
in
Newcombe RG (1998) Interval estimation for the difference between
independent proportions: comparison of eleven methods. Statistics in
Medicine 17: 873-890

The -ci- command gives an exact 95% CI for a proportion, therefore I
think that it is desirable that the difference between two proportions
have sensible 95% confidence intervals.

Kind regards, Garry