One sample test of proportions

[Isha Chaudhry]

Dear Stata users,

I have a question regarding comparing 3 or more percentages within the same population.

The specific question, I have a continuous variable age split into 3 categories (age_cat), cat A, cat B and cat C and their reported percentages in the sample.
Cat A = 56%
Cat B = 20%
Cat C = 24%

I want to check whether percentages of cat A, cat B and cat C are significantly different from each other or not. and report the p-values.

I tried using this code but getting an error,

// chitest age_cat, sep(3)
error: option sep() not allowed

I believe I should be applying one-sample test of proportions to compare the percentages across categories, however, not sure how to execute it.

I have computed the 95% CI for each category and there is no overlap, however not sure about the exact p-values.

Kindly advise which test should be used; any help will be really appreciated. Thank you.

Regards,
Isha

[Joseph Coveney]

Your categories are ordered. Does -chitest- assume nominal categories?

I'm unclear about the motivation of your test, but with ordered categories, maybe you could consider something like this? (The illustration assumes that you have a sample of one hundred.)

input byte cat byte count
1 56
2 20
3 24
end

ologit cat [fweight=count], nolog
estimates store Model

// Linear, quadratic components of orthogonal polynomial contrasts (and overall)
quietly margins , post
contrast q._predict, noeffects

// Individually, pairwise
quietly estimates restore Model
margins , pwcompare(pveffects) mcompare(bonferroni)

You'd probably want to look at its frequency properties (-help simulate-) before making much of it.