Dear All,
I have a series of proportions from different studies that I would like
to collapse into a single pooled proportion. The study sizes vary
greatly (from 6 to over 100) and I am not sure what would be the best
formula to use to construct the confidence interval for the pooled
proportion. I have seen that some people simply calculate the SE as
sqrt( p*(1-p)*(1/total n)) whereas I thought that I should be taking SE
= sqrt( p*(1-p)*(1/n1 + 1/n2 + ... )). Neither of these formulas,
however, seem very appropriate when some of the n values are small. Can
anyone suggest another approach?
Many thanks!
Kind regards,
Stephanie MacNeill
_________________________________________________
Stephanie MacNeill, MSc
Medical Statistician
Department of Occupational and Environmental Medicine
National Heart and Lung Institute
1B Manresa Rd
London SW3 6LR
Tel: 020 7351 8397
Fax: 020 7351 8336
|