2
$\begingroup$

Let's say I want to do a multiple-sample z-test for proportions. Is there any way to rank them?

How would I compare more than two samples at once?

$\endgroup$
1
  • $\begingroup$ What do you mean by 'rank them'? $\endgroup$
    – Glen_b
    Aug 4, 2014 at 9:11

1 Answer 1

3
$\begingroup$

The most commonly used - and fairly basic - multisample extension of a two-sample proportions test (akin to the relationship between two-sample t-tests and ANOVA) is the test of homogeneity of proportions.

When you condition on both margins, this is the same as the chi-square test of independence - same test statistic and degrees of freedom.

Also see here.

Any stats package worth the name will do one of these for you easily.

You don't have to rely on the chi-square approximation; it's possible to do an exact test (either akin to a Fisher test in the 2x2 table, or based on the chi-square statistic itself).

Many packages also make such exact tests straightforward. There's also approaches that don't require you to condition on both margins.

A search here on CV will turn up some useful discussion; a wider internet search has many useful resources.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.