0
$\begingroup$

I am doing a chi square test for categorical variables in a very large sample size. Is there any alternative to the Chi-Square Test (besides the cramer'S V test for associations )? Thanks,

$\endgroup$
6
  • 2
    $\begingroup$ Log-linear models. $\endgroup$
    – Nick Cox
    Feb 20, 2018 at 2:00
  • 2
    $\begingroup$ Could you explain why you seek an alternative? $\endgroup$
    – whuber
    Feb 20, 2018 at 2:19
  • $\begingroup$ The chi-square test you are using probably assumes a large sample size. If by very large you mean that it is so large that you get an error, you probably have no need for a statistical test. $\endgroup$
    – Tim
    Feb 20, 2018 at 3:27
  • 1
    $\begingroup$ @whuber Thanks for calling me out on my lazy language. I was assuming that the problem was that a computational error was occurring of some kind, as why else would one change test for a very large sample size? I have seen software that errors with large sample sizes (software that uses lookup tables transcribed from the back of old textbooks). $\endgroup$
    – Tim
    Feb 20, 2018 at 5:09
  • 4
    $\begingroup$ @Tim The most common situation on this site where somebody asks for an alternative test for large sample sizes occurs when they have discovered the test they are using achieves astronomically small p-values, perhaps for small effect sizes, and they are hoping to find some less powerful alternative! These questions arise from several basic misunderstandings of what tests are for and how to interpret them. That is why it is useful to ascertain the motivation behind the question before making assumptions (or, worse, actually answering the question). $\endgroup$
    – whuber
    Feb 20, 2018 at 5:21

1 Answer 1

2
$\begingroup$

Are you testing for the independence between variables? If so, except for the chi-square test, we may try (apart from log-linear, which is also suggested):

$\bullet$ Cochran–Mantel–Haenszel test: repeated tests of independence for nominal variables.

$\bullet$ Fisher's Exact Test: this test may give you more precise estimate for independence if the expected number is not large.

You may refer here, there are like G-test and more:

http://www.biostathandbook.com/cmh.html

$\endgroup$

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