Incidence with aggregate data

[Josan Rueda]

Hello everyone,

I'm working on a dataset which contains the following variables: year sex agegroup numberofcases totalpopulation. I'm trying to compute the yearly incidence for each sex and agegroup with 95% CI.

The

proportion

or

ci proportions

commands don't work as they need a binary variable to work with (e.g. case(0/1)).

I thought the

expand

command would be of use, but since totalpopulation range is in the millions it is too computationally intensive. I have seached the Statalist but found no solution.

Anyone could help?

Thank you in advance!

[Ken Chui]

Maybe Poisson regression with offset? See this: https://www.stata.com/manuals/rpoisson.pdf, Example 2 starting on page 6 may be useful.

[Andrea Discacciati]

This seems exactly the task for which the command -xcipoibin- was thought (if you want binomial exact CIs) (disclaimer: I am the author of this user-contributed command).

See:
https://github.com/anddis/xcipoibin
https://anddis.github.io/xcipoibin/xcipoibin_ex.html

Otherwise, you can always use a -forvalues- loop and -cii proportions-, for example

use https://raw.githubusercontent.com/anddis/xcipoibin/master/ex_ir.dta, clear // example data


// loop
gen double p = .
gen double l = .
gen double u = .

qui forvalues i = 1/8 {
    cii proportions person_years[`i'] obs_pca_cases[`i'], exact
    replace p = r(proportion) in `i'
    replace l = r(lb) in `i'
    replace u = r(ub) in `i'
}

// xcipoibin
 xcipoibin  obs_pca_cases person_years, binomial
 
 list

[Josan Rueda]

Thank you so much, Andrea, your solution works fine!