Hi Jeremy,

Most of researchers doesn't test assumptions of point biserial correlations. As you mentioned there are several assumptions. I can say you another assumption. There should be no outliers for the continuous variable for each category of the dichotomous variable. You can test for outliers using boxplots. It can be related with other assumptions but i read something like this.

Jeremy, i agree with Hüseyin, even though you will find in the most articles that your data dit not pass (failed) any of these assumptions: normality, no outliers equal variances, your Point Biserial Correlation will be not correct.

Hello Jeremy,

A point biserial correlation is merely a "simplified" formula for a Pearson correlation that may be applied when one of the variables is dichotomous. You'll get exactly the same value plugging in the numbers for an ordinary Pearson product-moment correlation. It is true that if two variables have different marginal distributions, the maximum possible correlation is constrained to be less than |1.0|.

For item analysis, most people behave as if the item-total correlations are either descriptive values or population values (note that in a lot of textbooks for psychometrics or measurement, the divisors for variance are "N", not "N - 1"). Either way, there's no need to check assumptions in these cases. For one thing, there's no way a dichotomously scored variable could ever have a distribution that one would call "normal" in form!

Good luck with your work.