I'm working on research into the vitamin D levels of professional and amateur soccer players, and relationship with factors such as skin color, vitamin D-intake etc.

I don't know which test to use. I'll give you an idea of the situation:

  • I have a continuous variable: the measured vitamin D levels
  • A categorial variable: the skin color (either black or white )

I've read about the One-way ANOVA, but it requires three options for the categorical variable. I used the ANOVA test, the p-value was 0.005. Is it possible to use this test with just two options for the categorical variable? Or is there another test I could use?

And if it's possible to use the ANOVA, does the value F mean anything, or just the p-value?

  • 1
    $\begingroup$ Hi @Joost, welcome to the site. If you have only two groups (black and white), you could do a simple two-sample t-test to compare the mean vitamin D levels between black and white skin color. By the way: what software do you use? Stata? SPSS? R? $\endgroup$ Jun 8, 2013 at 19:55
  • $\begingroup$ I am actually looking for the correlation/association between the vitmamin D status and the skin color, sorry if my question wasnt clear enough. Any ideas on which test to use in this case? $\endgroup$
    – Joost
    Jun 8, 2013 at 20:16
  • 1
    $\begingroup$ Okay, then you could run a regression model with vitamin D as dependent and the skin color as independent categorical variable. But if you have only those two variables, that's basically the same as a two-sample t-test (because you have only 2 categories). $\endgroup$ Jun 8, 2013 at 21:00
  • $\begingroup$ If you want a correlation coefficient, you could calculate the Point-biserial correlation coefficient which is just the normal Pearson correlation coefficient where one variable is dichotomous (skin color). So just calculate the normal correlation coefficient between vit. D and skin color. $\endgroup$ Jun 8, 2013 at 21:15

1 Answer 1


Anova with two categories is the same as the two-sample t-test (apart from the standard presentation of the results). You can check yourself that the square of the t-value equals the F-value. The same model can also be presented as a linear regression with skin-color as a categorical predictor, coded via dummys.

You should also visualize your data, in your case a parallel boxplot could be a good start.


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.