Point-Biserial Correlation — continuous vs. dichotomous
VerifiedAdvanced & specialized
Run this test straight away on a free built-in teaching dataset — no data of your own needed — or bring your own. Either opens the guided workspace: variable setup, assumption diagnostics, results with effect sizes and confidence intervals, figures, and APA-ready reporting.
Loading teaching datasets…
Or use your own dataset
Loading your datasets…
Pearson correlation between a CONTINUOUS outcome y and a DICHOTOMOUS predictor x coded 0/1.
Mathematically identical to the independent-samples t-test re-expressed as a correlation: r_pb and Cohen's d encode the same effect. Reports r_pb, Fisher-z 95% CI, equivalent t and p, per-group descriptives, Cohen's d, AND the theoretical r_pb ceiling imposed by the observed binary split (max |r_pb| = 2·sqrt(b)/(1+b) where b is the balance ratio min(n0,n1)/max(n0,n1)).
Worked example
Does passing a course relate to weekly study hours?
For 70 students, pass/fail (binary) and weekly study hours (continuous) were recorded; the point-biserial correlation measures the link between a binary and a continuous variable.
Study hours were associated with passing, r_pb = .65, p < .001 — those who passed had studied more.
A point-biserial correlation showed study hours were significantly associated with passing, r_pb = .65, p < .001.
Try it yourself: Load this ready-made sample and follow the run above.
When to use it
- Treatment vs control on a continuous outcomeA trial randomises 80 patients to drug vs placebo and measures systolic blood pressure at 12 weeks.
- Naturally-binary demographic predictorAn education researcher measures reading speed in 200 schoolchildren and computes r_pb between sex and reading speed.
- Item-total correlation in psychometricsA 40-item knowledge test is administered to 500 students.
When NOT to — use instead
- Three or more groups on a continuous outcomeUse one-way ANOVA; r_pb is binary-by-construction. → One-Way ANOVA (3+ groups)
- Continuous predictorIf the predictor is genuinely continuous, use Pearson r — don't dichotomise. → Pearson Correlation (linear association)
- Non-normal continuous outcome with severe heteroscedasticityMann-Whitney U gives a robust rank-based alternative when the within-group distribution violates t-test assumptions. → Mann-Whitney U
- Need to control for confounding covariatesPoint-biserial is bivariate. → Binomial Logistic Regression \u2014 odds ratios + ROC AUC + classification
Hypotheses
Parameter tested: point-biserial correlation ρ_pb
Assumptions (and what to do if they fail)
Check: Levene's p ≥ .05 and SD ratio ≤ 2.
If violated: Unequal variances inflate the Type I error of the equivalent t-test.
Check: Per-group |z| > 3 count.
If violated: A single outlier can drive r_pb up or down.
Check: smaller_n / larger_n ≥ 0.3.
If violated: Severe imbalance caps the maximum achievable r_pb.
Ready to run a Point-Biserial Correlation — continuous vs. dichotomous on your own data?
Guided setup, automatic assumption checks, effect sizes, figures and an APA write-up.
Run this test →