One-sided chi-square test

[Tim Vodegel]

Hi all,

I am using the two-tailed Pearson's chi-square test in my bachelor thesis. My supervisor says that it would be better if I used the one-tailed test. However, I cannot find a command for that in Stata. Is this even possible with the chi-square test?

Many thanks!

Tim

[Rich Goldstein]

because of the squaring you are using only one of the tails for any particular test; so, no it is not possible anywhere

[Tim Vodegel]

Ok, thanks!
Now I know that it is possible to do a one-tailed Wilcoxon rank sum test (MWU-test), but Stata only returns the two-sided p-value. Is it possible to calculate the one-tailed p-value as well using the ranksum command?

[daniel klein]

The one-tailed p-value should just the two-tailed p-value devided by 2.

Hope this helps.
Daniel

[Rich Goldstein]

since (quoting from the help file for -ranksum-), " ranksum tests the hypothesis that two independent samples (that is,
unmatched data) are from populations with the same distribution", it is not clear what a one-sided (not tailed) test is testing; please explain what your null hypothesis is

note regarding one-sided chi-squared test: if there is only 1 df for the test, then chi-squared = z-squared and a one-sided (and now one-tailed also) test is possible

[Tim Vodegel]

Thanks for the advice! In case of the ranksum test, I have a variable of interest that ranges from 0 (low) to 6 (high). Based on certain characteristics, the subject pool was divided into two groups. My null hypothesis is thus that those two groups do not have different scores in terms of this variable of interest.

However, a priori it can be expected that group 1 has a higher score, so I want to test this by using a one-sided test. As there are two independent samples and the variable of interest is ordinal, I thought the Wilcoxon rank sum test (Mann Whitney U test) would be most appropriate.

[Joe Canner]

Tim,

Is your variable continuous or categorical? If the latter, and you really want to do a one-sided test, Fisher's Exact Test has one sided version. Do tab group score, exact and that will show both one-sided and two-sided p-values. This does not account for the ordinal nature of your variable, however. For that you might need the Cochran-Mantel-Hazenzsel test.

Regards,
Joe