Which test to use to calculate correlations between them?
- Cramer's V,
- Kruskal-Wallis,
- or something else?
Asked
Active
Viewed 2,997 times
1
gung - Reinstate Monica
- 132,789
- 81
- 357
- 650
user46343
- 11
- 1
- 2
-
1Are you wondering how such correlations are calculated (ie, how to arrive at the number), how they are tested (ie, if they could have come from a population w/o any correlation), or what they mean? – gung - Reinstate Monica May 28 '14 at 23:33
-
_italic_ Yes! @gung – user46343 May 28 '14 at 23:46
1 Answers
3
Cramér's V and the Kruskal–Wallis test are for nominal data; the latter is a null hypothesis test, not a correlation. If you want to calculate the correlation between a dichotomous variable and an ordinal variable, you could use Kendall's $\tau$, the Goodman–Kruskal $\gamma$, or Spearman's $\rho$ (listed in the order in which I'd recommend them, I suppose). If you're estimating population parameters, you can also generate confidence intervals around these statistics or perform null hypothesis tests if you wish.
For some comparisons and a conversion method, see:
Nick Stauner
- 11,558
- 5
- 47
- 105
-
@user46343: was there anything else you wanted to know, or is my answer [acceptable](http://meta.stackexchange.com/a/5235/246931)? – Nick Stauner May 30 '14 at 20:55