0
$\begingroup$

I am writing my thesis at the moment and I am pretty new with statistics. I tired to google around but I don’t know what to do anymore, since I don’t really find an answer in this case. I hope you can help me…

I want to calculate a bivariate correlation, but as I read everywhere the variables I use for this have to be normally distributed and one should be able to see a linear connection between them in a scatterplot. They are normally distributed but one of them is ordinal though and my graph looks like this:

First graph

It gets even worse when I have two ordinal variables. Then it looks like this:

Second graph

I get significant (but weak) r values in both cases you see above, when I calculate a correlation, but I am not sure if I am allowed to use my results, since I cannot see any linearity in my graphs in the first place. And as I read it this should be a pre-requirement for a correlation. But the fact that mz avriables are ordinal make it hard to see if there might be some linearity, alsothough I am not able to see it. So, basically I don't know if I have a correlation now or not...

I also read that I could use Spearman's correlation instead, but isn't it that I need to have some kind of connection in the graph for this one as well? I my case I don’t see anything in my scatter plots.

Thank you in advance!

$\endgroup$
  • $\begingroup$ I've been having the exact same issue. Have you come with any ideas!? Thanks $\endgroup$ – user123879 Jul 20 '16 at 11:07
  • $\begingroup$ I'm not going to put this as an answer, because I'm not sure about the linearity assumption underlying these statistics, but for the first analysis (ordinal vs. interval) you may want to investigate polyserial correlation. For the second analysis (ordinal vs. ordinal) polychoric correlation may be appropriate. $\endgroup$ – Ian_Fin Jul 20 '16 at 11:10
1
$\begingroup$

You cannot use Pearson's correlation with ordinal variables. You can use Spearman's rank correlation and there are other measures of association as well.

But correlation does not assume the variables are normally distributed.

| cite | improve this answer | |
$\endgroup$
  • 1
    $\begingroup$ Since the question asker seems more concerned with linearity than normality, it may be worth pointing out that Spearman's rank correlation does not assume linearity (although it does assess monotonicity) $\endgroup$ – Ian_Fin Jul 20 '16 at 11:23

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.