Point-Biserial Correlation Calculator

Use point-biserial correlation when you want to measure how strongly a yes/no grouping relates to a numerical measurement. For instance, does gender relate to test scores? Does treatment group membership predict recovery time? This gives you a correlation coefficient that quantifies the strength and direction of the relationship between your two-category variable and your continuous variable.

What Is Point-Biserial Correlation?

Point-biserial correlation ($r_{pb}$) measures the relationship between a binary (dichotomous) variable and a continuous variable.

Question answered: "Is there a relationship between group membership and a continuous outcome?"

Range: -1 to +1

• Positive: Higher values associated with coded-1 group
• Negative: Higher values associated with coded-0 group
• Zero: No relationship

When to Use

Common Scenarios

• Male/female and test scores
• Treatment/control and blood pressure
• Pass/fail and study hours
• Yes/no and continuous measurements

Relationship to Other Tests

Point-biserial correlation is mathematically equivalent to:

• Independent samples t-test (same p-value)
• Pearson correlation with binary variable coded 0/1

The Formula

$r_{pb} = \frac{\bar{Y}_1 - \bar{Y}_0}{S_Y} \times \sqrt{\frac{n_1 n_0}{n^2}}$

Where:

• $\bar{Y}_1$ = mean of continuous variable for group 1
• $\bar{Y}_0$ = mean of continuous variable for group 0
• $S_Y$ = standard deviation of the continuous variable (all data)
• $n_1, n_0$ = sample sizes for each group
• $n$ = total sample size

Equivalent Formula

$r_{pb} = \frac{\bar{Y}_1 - \bar{Y}_0}{S_Y} \times \sqrt{p(1-p)}$

Where p = proportion in group 1.

Example Calculation

Data: Treatment (1) vs Control (0) and pain reduction scores

Step 1: Calculate group means

• $\bar{Y}_1 = (15+18+22+20)/4 = 18.75$
• $\bar{Y}_0 = (8+12+10+11)/4 = 10.25$

Step 2: Calculate overall SD

All values: 15, 18, 22, 20, 8, 12, 10, 11 Mean = 14.5, SD = 4.96

Step 3: Calculate correlation

$r_{pb} = \frac{18.75 - 10.25}{4.96} \times \sqrt{\frac{4 \times 4}{64}} = 1.71 \times 0.5 = 0.86$

Interpretation: Strong positive correlation—treatment group has much higher pain reduction.

Significance Testing

T-Test Approach

$t = r_{pb} \sqrt{\frac{n-2}{1-r_{pb}^2}}$

With df = n - 2.

Relationship to Independent T-Test

The p-value from point-biserial equals the p-value from independent t-test on the same data.

$t = \frac{\bar{Y}_1 - \bar{Y}_0}{SE_{\text{diff}}}$

Effect Size Interpretation

| |$r_{pb}$| | Interpretation | |------------|----------------| | 0.10 | Small | | 0.30 | Medium | | 0.50 | Large |

These align with Cohen's conventions for correlation coefficients.

Relationship to Cohen's d

$r_{pb} = \frac{d}{\sqrt{d^2 + 4}}$

And conversely:

$d = \frac{2r_{pb}}{\sqrt{1-r_{pb}^2}}$

For our example: $r_{pb} = 0.86$ → $d = 3.36$ (very large effect)

Point-Biserial vs. Biserial Correlation

Use point-biserial for naturally dichotomous variables.

Coefficient of Determination

$r_{pb}^2 = \text{proportion of variance explained}$

From example: $0.86^2 = 0.74$

Group membership explains 74% of variance in pain reduction.

Assumptions

Dichotomous Variable

One variable must be binary (two categories only).

Continuous Variable

The other variable should be approximately continuous.

Normality

For significance testing, the continuous variable should be approximately normal within each group.

Homogeneity of Variance

Equal variances in both groups (for valid significance testing).

Independence

Observations are independent.

Applications

Item Analysis

Correlate item scores (right/wrong) with total test scores to assess item discrimination.

Clinical Research

Relationship between treatment group and outcomes.

Psychology

Association between binary categories (e.g., diagnosis) and continuous measures.

Education

Gender differences in test performance.

Confidence Interval

Use Fisher's z transformation:

$z_r = 0.5 \ln\left(\frac{1+r}{1-r}\right)$

$SE_z = \frac{1}{\sqrt{n-3}}$

$95\% \text{ CI for } z_r = z_r \pm 1.96 \times SE_z$

Transform back to correlation scale.

Reporting Results

APA format:

"There was a strong positive point-biserial correlation between treatment condition and pain reduction, $r_{pb}$ = .86, 95% CI [.54, .96], p < .001."

With effect size:

"The point-biserial correlation between group membership (treatment vs. control) and pain reduction scores was $r_{pb}$ = .86 (95% CI: .54-.96), indicating a large effect (Cohen's d = 3.36). The treatment group (M = 18.75, SD = 2.99) showed significantly greater pain reduction than the control group (M = 10.25, SD = 1.71), t(6) = 4.72, p = .003."

Common Mistakes

Using with Polytomous Variables

Point-biserial requires exactly two groups. For multiple categories, use eta or other measures.

Ignoring Unequal Group Sizes

Maximum possible $r_{pb}$ decreases with unequal groups. With 90/10 split, max $|r_{pb}|$ ≈ 0.6.

Confusing with Biserial

Point-biserial is for naturally dichotomous variables; biserial assumes underlying continuity.

Not Checking Assumptions

Violation of normality or homogeneity affects significance tests.

Maximum Possible Value

With unequal group sizes:

$r_{pb,max} = \sqrt{p(1-p)} \times \frac{1}{\phi(z_p)}$

Where p = proportion in smaller group and φ is the standard normal density.

Implication: With very unequal groups, even perfect separation gives $r_{pb}$ < 1.

Relationship to Other Statistics

To F-statistic

$F = \frac{r_{pb}^2(n-2)}{1-r_{pb}^2}$

To Eta-squared

For two groups: $\eta^2 = r_{pb}^2$

To Omega-squared

$\omega^2 = \frac{r_{pb}^2 - (1-r_{pb}^2)/(n-2)}{1 + (1-r_{pb}^2)/(n-2)}$

Historical Context

Point-biserial correlation has been used since the early 20th century, particularly in educational and psychological measurement. It became standard in item analysis for tests, where it indicates how well an item discriminates between high and low performers.

Related Calculations

• Pearson Correlation — Two continuous variables
• Spearman Correlation — Rank-based correlation
• Two-Sample T-Test — Equivalent hypothesis test
• Cohen's d Calculator — Related effect size

Summary

Point-biserial correlation measures the relationship between a binary variable and a continuous variable. It ranges from -1 to +1 and is mathematically equivalent to Pearson correlation when the binary variable is coded 0/1. The significance test gives the same p-value as an independent t-test. Use interpretation guidelines of 0.1/0.3/0.5 for small/medium/large effects. Applications include item analysis in testing, treatment effect assessment, and group comparison studies. Report the correlation, confidence interval, and p-value.