95% CI for Percent Agreement?

[Chris Mallow]

Hello,

Im doing stats for a paper right now, and the first author would like me to mimic what was done in a previous article that is somewhat related. He is asking for concordance with a 95%CI. When I plug in the numbers from the prior paper, they are using percent agreement (and he was specific that he didnt want kappa) but with a 95% CI? Im not quite sure how to calculate a 95%CI for a percent agreement, and wanted to reach out and see if anyone could help?

Thanks!

[Clyde Schechter]

So suppose your two variables whose agreement you are checking are x and y

gen byte agree = (x == y)
ci proportions agree

By the way, there are options to -ci proportions- that let you pick your preferred "brand" of confidence interval.

[Bruce Weaver]

Hi Chris. It sounds like you want the CI for a binomial proportion multiplied by 100. You can compute various kinds of CIs for binomial proportions using the ci command. Here is an example showing how to compute the Wilson score CI for a proportion:

* Create a small dataset to illustrate.
clear
set obs 100
generate byte a = rbinomial(1,.3)
generate byte b = rbinomial(1,.5)
generate byte c = rbinomial(1,.7)
ci a b c, binomial wilson

                                                         ------ Wilson ------
    Variable |        Obs        Mean    Std. Err.       [95% Conf. Interval]
-------------+---------------------------------------------------------------
           a |        100         .26    .0438634         .184047    .3537099
           b |        100         .56    .0496387         .462281    .6532797
           c |        100         .68    .0466476        .5833738    .7633085

Note that variables a-c are all 1-0 variables, and the Mean column above shows the proportion of 1s.

Type help ci in the command window for more info and examples. If you have only summary data, you can use the so-called immediate form of ci. E.g.,

cii 100 26, wilson
cii 100 56, wilson
cii 100 68, wilson

The general format there is

cii Denominator Numerator, options

where the proportion of interest = Numerator/Denominator.

HTH.