How can I correlate 2 dichotomous variables?

Var1: Passed:1 Failed: 19

Var2: Passed:3 Failed:13

I tried the $\phi$ coefficient $X$: Var1 Var2

$Y$: Passed Failed

Everything went well but I realized that I'm correlating the failing and passing of each var1 and var2 separately.

What statistical tool is the best to correlate this?

  • 1
    $\begingroup$ Have you tried tetrachoric correlations? It's intended for use with binary information. jstor.org/stable/3701350 $\endgroup$
    – user78229
    Feb 8 at 2:19
  • $\begingroup$ Galen, OP noted they were using Excel only. So, I thought of applying the tag. $\endgroup$ Feb 8 at 2:49
  • 1
    $\begingroup$ How have you correlated $20$ observations with $16?$ $\endgroup$
    – Dave
    Feb 8 at 2:50

1 Answer 1


I am making an assumption that your two variables refer to two treatment states, say a control group and a treatment group.

In that case, your Var1 and Var2 are really one variable that indicates the state, if the subject is in the control group (typically coded with a zero) or the treatment group (typically coded as a one). Call this variable $X,$ and it has $36 = (1 + 19) + (3 + 13)$ observations.

Then your $Y$ variable is another indicator of passing or failing. It would be typical to code these as $0$s and $1$s, leaving you with another set of $36$ observations, paired with the $X$-values (for each treatment state, you know the pass/fail state).

You now have two equal-length variables that are paired. The usual Pearson correlation makes sense here (though the sign will depend on how you code the categorical states of exam pass/fail and control/treatment). This is the $\phi$ coefficient, which (discussed in the link) has a relationship to a $\chi^2$ test of the corresponding contingency table.

I have sample size concerns about your particular set of data.


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.