# Should I use Pearson or Spearman correlation?

I have to know the correlation between 2 variables (bivariate distribution). None of them are normally distributed, so I assumed that I should run Spearman's correlation, which gave me a correlation coefficient of 0.392 (p<0.05, and Pearson's correlation was equally p<0.05).

My problem is that now I would like to test the correlation for both males and females, so the same data but split in 2. It turns out that the males data is normally distributed but the females data is not. What should I do? I don't think I can compare Pearson vs Spearman for the same data. Both in Pearson and Spearman, the correlation is much stronger in males compared to females, but I don't know which values I should use.

I don't know if what I wrote makes any sense, I hope someone can help! :)

• Your choice of statistic to test "correlation" ought to depend on what you mean by "correlated," as well as (or even more than) on the statistical characteristics of the sample. – whuber Jan 30 '17 at 23:36
• Are you sure you 'have to test the correlation between 2 variables', and not e.g. a linear model with binary variables? This may be a case of the XY problem. – N. Wouda Jan 31 '17 at 18:47
• Some discussion here may be relevant stats.stackexchange.com/questions/8071/… – Glen_b Jan 16 '18 at 23:13