Non-parametric correlation test between dichotomous variable and continuous one

[Ibai Ostolozaga Falcon]

Dear Statalist.

I have a small sample of countries. I have two variables, X which is a continuous variable, and Y which is a dichotomous one. The problem is that my variable X is not normally distributed, so I cannot use point-biserial test, then I am looking for a non-parametric test. Reading the literature I have read about "Eta correlation test". However, I do not Know how to compute it in Stata. Any advise?

Thanks in advanced,
Ibai

[Nick Cox]

I'd recommend rephrasing this as a logit regression if the dichotomous variable is an outcome and a t or other test if it is a predictor.

Any kind of correlation here is just a diversion: the more interesting question is to quantify the relationship, which doesn't rule out testing for it.

[Ibai Ostolozaga Falcon]

I'd recommend rephrasing this as a logit regression if the dichotomous variable is an outcome and a t or other test if it is a predictor.

Any kind of correlation here is just a diversion: the more interesting question is to quantify the relationship, which doesn't rule out testing for it.

Thank you for your answer Nick.

I have been thinking about run a Logit regression where my outcome would be the dichotomous variable Y and the input the continuous variable X. However, I think that I could be omitting relevant variables, which if I am not mistaken, would give me a biased estimate.

[Nick Cox]

It's the other way round: logit regression lets you look at the effects of other variables, so long as you have the data, whereas correlations focus on pairwise relationships. (I've never found partial correlations easier to think about than a model with important predictors.)