Quick Answer: Point Biserial Correlation Result

The main worked example tests the association between sex and G3 final grade. The binary variable is sex, coded as F = 0 and M = 1. The continuous variable is G3. Since sex is a true two-category variable and G3 is a numeric grade score, the correct correlation is Point Biserial Correlation.

The verified Excel result is r_pb = -0.129077, with N = 649, t(647) = -3.310938, and p = 0.000982. The result is statistically significant at α = .05, but the effect size is small. The negative sign appears because the group coded 1, which is M, has a lower mean G3 score than the group coded 0, which is F.

What Is Point Biserial Correlation?

Point Biserial Correlation is a correlation coefficient used when one variable is naturally binary and the other variable is continuous. A binary variable has two categories, such as female/male, yes/no, pass/fail, treatment/control, urban/rural, public/private, or present/absent. A continuous variable has numeric values, such as final grade, test score, income, weight, time, age, or measurement score.

The point-biserial coefficient is written as r_pb. It is calculated like Pearson correlation after the binary variable is coded as 0 and 1. Even though the calculation is correlation-based, the interpretation is often easier through group means. A positive r_pb means the group coded 1 has a higher mean on the continuous variable. A negative r_pb means the group coded 1 has a lower mean.

In this guide, F is coded 0 and M is coded 1. The F group has mean G3 = 12.2533, while the M group has mean G3 = 11.4060. Since the group coded 1 has the lower mean, the point-biserial correlation is negative: r_pb = -0.129077.

Point Biserial Correlation connects naturally with Correlation in Python, Correlation in R, Correlation in SPSS, Correlation in Excel, t-Test in Python, t-Test in SPSS, Effect Size, p-value and Confidence Interval.

When Should You Use Point Biserial Correlation?

Use Point Biserial Correlation when the research question asks whether a two-category group variable is associated with a numeric outcome. The binary variable should be a real two-category variable, not an artificial split made from a continuous variable. The continuous variable should have meaningful numeric distances.

For the main worked example, the conditions are satisfied. Sex is binary, G3 is continuous/numeric, and the analysis asks whether the binary group coding is associated with final grade. The result is statistically significant, but the group mean difference is modest and the effect size is small.

Point Biserial vs Biserial vs Phi Coefficient

Students often mix up Point Biserial Correlation, Biserial Correlation, and Phi Coefficient because all three involve binary variables. The key difference is the scale of the second variable and whether the binary variable is a true binary variable or an artificial split.

Point Biserial Correlation Formula

Point Biserial Correlation can be calculated using Pearson correlation after coding the binary variable as 0 and 1. It can also be calculated from group means, sample standard deviation, and group sizes.

In this formula, M₁ is the mean of the continuous variable for the group coded 1, M₀ is the mean for the group coded 0, SD_Y is the sample standard deviation of the continuous variable, n₁ and n₀ are the group sizes, and n is the total sample size.

The sign is negative because the group coded 1 has a lower mean. If the coding were reversed, the sign would change, but the absolute strength would remain the same.

Significance Test Formula

The point-biserial significance test can be written using the same t-test logic used for Pearson correlation:

For the main example, this gives t = -3.310938, with df = 647 and p = 0.000982. The r-squared value is 0.016661, so the binary coding accounts for about 1.67% of the variance in G3.

Null and Alternative Hypotheses

The point-biserial test asks whether the population association between the binary variable and the continuous variable is zero.

Dataset and Variables Used

The analysis uses the student performance dataset with 649 valid observations. The main worked example uses sex as the true binary variable and G3 as the continuous outcome variable.

Before reporting the coefficient, it is useful to review Descriptive Statistics Explained, Standard Deviation, Standard Error, Confidence Interval, Effect Size, p-value and Null and Alternative Hypothesis.

Verified Point Biserial Results

The table below summarizes the verified main worked example from the Excel workbook. The values match the report-card logic used in the Python and R charts.

The result should be reported as significant and small. A large sample can make a small effect statistically significant, so interpretation must include both p-value and coefficient size.

Top Pairwise Point Biserial Relationships

The workbook also screens binary variables against continuous variables and ranks the strongest point-biserial relationships. This broader table helps show that the main sex by G3 example is not the strongest point-biserial relationship in the dataset.

The pairwise screening shows an important lesson: the largest point-biserial associations in this dataset are not always the most obvious pair. Higher-education intention has moderate positive associations with grades, while sex by G3 is statistically significant but small.

Python Chart-by-Chart Interpretation

The Python report includes five charts: distribution by binary group, group means with 95% confidence intervals, point-biserial report card, histogram overlay by binary group, and binary scatter with fitted line. Together, these visuals explain why the result is statistically significant but small.

Python Chart 1: Distribution by Binary Group Boxplot

The boxplot shows the distribution of G3 separately for the F-coded group and the M-coded group. The F group has a slightly higher center than the M group. This matches the verified means: F has mean G3 = 12.2533, while M has mean G3 = 11.4060.

The most important feature of the chart is overlap. The boxes and whiskers are not widely separated. Many students in both groups have similar final grades. This visual pattern explains why the coefficient is small even though the p-value is significant. A strong point-biserial relationship would show much clearer separation between the two groups.

The chart also helps prevent overstatement. The result does not mean one group always scores higher than the other. It only means that, on average, the group coded 1 has a lower G3 score in this dataset.

Python Chart 2: Group Means with 95% Confidence Intervals

The group means chart makes the direction of the result easy to see. The F-coded group has a higher mean G3 score, and the M-coded group has a lower mean G3 score. The mean difference is -0.847249 when calculated as M minus F.

This is the chart that most directly explains the negative sign. Since M is coded 1 and M has the lower mean, the coefficient is negative. If the coding were reversed, the line or bars would represent the same group difference but the sign of the correlation would reverse.

The confidence intervals help readers understand sampling uncertainty. Because N is large, the group means are estimated fairly precisely. Still, the actual mean gap is less than one grade point, so the practical effect remains small.

Python Chart 3: Point Biserial Report Card

The report card gives the final result in one place: binary variable sex, continuous variable G3, N = 649, F coded 0, M coded 1, r_pb = -0.129077, t(647) = -3.310938, p = 0.000982, and a significant decision at α = .05.

This is the best chart to use when writing the final results paragraph because it contains the coefficient and the test statistic. However, the report card should still be interpreted with the distribution and mean charts. The p-value tells us the association is statistically detectable, while the coefficient tells us the effect is small.

Python Chart 4: Histogram Overlay by Binary Group

The histogram overlay shows how the full G3 distribution differs across the two binary groups. The two distributions overlap heavily, which confirms that the association is not large. The F-coded distribution is slightly shifted toward higher final grades, while the M-coded distribution has relatively more lower-grade observations.

This chart is helpful because a boxplot summarizes distribution shape, but a histogram shows the frequency pattern across grade values. The overlap demonstrates why the practical effect should not be exaggerated. A statistically significant point-biserial correlation can still be small when group distributions largely overlap.

Python Chart 5: Binary Scatter with Fitted Line

The binary scatterplot places the coded binary variable on the x-axis and G3 on the y-axis. Because the x-axis has only two possible values, points stack at 0 and 1. Jittering or transparency helps show the density of observations.

The fitted line slopes downward because the group coded 1 has a lower mean G3 score. The slope is not steep, which visually matches the small value of r_pb. This chart is useful for connecting point-biserial correlation with ordinary Pearson correlation and simple linear regression. With a 0/1 predictor, the fitted-line difference corresponds to the group mean difference.

R Chart-by-Chart Interpretation

The R charts validate the Python results using a separate workflow and colorful chart versions. The same five visuals are repeated: boxplot, group means with confidence intervals, report card, histogram overlay, and binary scatter with fitted line.

R Chart 1: Colorful Distribution by Binary Group Boxplot

The R boxplot confirms the same distribution pattern as Python. The F group has a slightly higher central tendency than the M group, but the distributions overlap strongly. This independent software validation supports the same interpretation.

R Chart 2: Colorful Group Means with 95% Confidence Intervals

The R group means chart validates the same mean difference: the F group has a higher mean G3 score than the M group. The sign of r_pb is negative because M is coded 1 and the M group mean is lower.

R Chart 3: Colorful Point Biserial Report Card

The R report card confirms the same final statistics: N = 649, r_pb ≈ -0.129, t ≈ -3.311, p ≈ .001, and a significant decision. The chart is useful because it confirms that the result does not depend on one software environment.

R Chart 4: Colorful Histogram Overlay by Binary Group

The R histogram overlay again shows heavy overlap between the two G3 distributions. This supports the small-effect conclusion. The chart is especially useful for students who might otherwise interpret statistical significance as a large practical difference.

R Chart 5: Colorful Binary Scatter with Fitted Line

The R fitted-line chart confirms the negative direction. The line decreases from code 0 to code 1 because the M-coded group has the lower mean. The decline is visible but not steep, which matches the small magnitude of the point-biserial coefficient.

SPSS Output Interpretation

The SPSS output is included for formal reporting. In SPSS, Point Biserial Correlation can be obtained by recoding the binary variable into 0 and 1 and then running a Pearson correlation between the binary code and the continuous variable. It can also be interpreted alongside an independent-samples t-test because both methods describe the same two-group mean difference from different perspectives.

Open the SPSS Point Biserial Correlation Output PDF

SPSS users should avoid reporting only the correlation table. A clear report should also mention the group means. Without the means and coding direction, readers may misunderstand why the coefficient is negative.

Excel Worked File Explanation

The Excel workbook gives a fully worked Point Biserial Correlation example. It includes raw data, encoded variables, the main worked example, pairwise point-biserial screening, Excel method notes and a dashboard. The workbook is useful for students because it shows both the direct Excel correlation method and the manual formula using group means.

Download the Point Biserial Correlation Fully Worked Excel File

Excel Main Worked Example

The workbook uses G3 as the continuous variable and sex as the binary variable. The coding rule is F = 0 and M = 1. The direct Excel method is:

The formula-driven Excel result is -0.1290774866. The manual group-means formula gives the same value, confirming the calculation. The t statistic is -3.3109376930, df is 647, and the two-tailed p-value is 0.0009815287.

Excel Interpretation Rule

Excel makes the coding direction very clear. Since the binary code is F=0 and M=1, a negative coefficient means the M group has a lower mean. It does not mean the variable is “bad,” and it does not automatically imply a large effect. It only describes the direction of the mean difference under the chosen coding.

Python, R, SPSS and Excel Workflows

The same Point Biserial Correlation analysis can be reproduced in Python, R, SPSS and Excel. The steps are similar in every tool: code the binary variable, choose the continuous variable, compute the correlation, check group means, run the significance test and interpret the effect size.

Code Blocks and Excel Formulas

Python Code for Point Biserial Correlation

import pandas as pd
import numpy as np
from scipy import stats

# Load data
df = pd.read_csv("dataset.csv")

# Main worked example
continuous_var = "G3"
binary_var = "sex"

work = df[[binary_var, continuous_var]].dropna().copy()
work[continuous_var] = pd.to_numeric(work[continuous_var], errors="coerce")

# Code F = 0 and M = 1
work["binary_code"] = work[binary_var].map({"F": 0, "M": 1})
work = work.dropna()

x = work["binary_code"].to_numpy()
y = work[continuous_var].to_numpy()
n = len(work)

# Point-biserial correlation is Pearson correlation with a 0/1 variable
r_pb, p_value = stats.pearsonr(x, y)

# Group means
n0 = int((x == 0).sum())
n1 = int((x == 1).sum())
mean0 = y[x == 0].mean()
mean1 = y[x == 1].mean()
sd_y = y.std(ddof=1)

# Manual formula
r_manual = ((mean1 - mean0) / sd_y) * np.sqrt((n0 * n1) / (n * (n - 1)))

# t statistic and r squared
t_value = r_pb * np.sqrt((n - 2) / (1 - r_pb**2))
df_t = n - 2
r_squared = r_pb**2

print("N:", n)
print("n0:", n0, "n1:", n1)
print("Mean code 0:", mean0)
print("Mean code 1:", mean1)
print("Mean difference 1-0:", mean1 - mean0)
print("Point-biserial r:", r_pb)
print("Manual r:", r_manual)
print("r squared:", r_squared)
print("t:", t_value, "df:", df_t, "p:", p_value)

R Code for Point Biserial Correlation

# Load data
df <- read.csv("dataset.csv", stringsAsFactors = FALSE)

# Main worked example
work <- na.omit(df[, c("sex", "G3")])
work$G3 <- as.numeric(work$G3)

# Code F = 0 and M = 1
work$sex_code <- ifelse(work$sex == "M", 1, 0)

# Point-biserial correlation is Pearson correlation with a 0/1 variable
cor_result <- cor.test(work$sex_code, work$G3, method = "pearson")
print(cor_result)

# Group summaries
n <- nrow(work)
n0 <- sum(work$sex_code == 0)
n1 <- sum(work$sex_code == 1)
mean0 <- mean(work$G3[work$sex_code == 0])
mean1 <- mean(work$G3[work$sex_code == 1])
sd_y <- sd(work$G3)

r_manual <- ((mean1 - mean0) / sd_y) * sqrt((n0 * n1) / (n * (n - 1)))
t_value <- unname(cor_result$estimate) * sqrt((n - 2) / (1 - unname(cor_result$estimate)^2))
r_squared <- unname(cor_result$estimate)^2

cat("N =", n, "\n")
cat("n0 =", n0, "n1 =", n1, "\n")
cat("Mean code 0 =", mean0, "\n")
cat("Mean code 1 =", mean1, "\n")
cat("Manual point-biserial r =", r_manual, "\n")
cat("t =", t_value, "\n")
cat("r squared =", r_squared, "\n")

# Mean difference context
t.test(G3 ~ sex_code, data = work, var.equal = TRUE)

SPSS Syntax for Point Biserial Correlation

* Point Biserial Correlation in SPSS.
* Binary variable: sex coded F=0 and M=1.
* Continuous variable: G3.

OUTPUT CLOSE ALL.
OUTPUT NEW NAME=Point_Biserial_Output.

RECODE sex ('F'=0) ('M'=1) INTO sex_code.
VARIABLE LABELS sex_code 'Sex coded for point-biserial correlation: F=0, M=1'.
VALUE LABELS sex_code 0 'F' 1 'M'.
EXECUTE.

FREQUENCIES VARIABLES=sex sex_code.
DESCRIPTIVES VARIABLES=G3 sex_code
  /STATISTICS=MEAN STDDEV MIN MAX.

CORRELATIONS
  /VARIABLES=sex_code G3
  /PRINT=TWOTAIL
  /MISSING=PAIRWISE.

T-TEST GROUPS=sex_code(0 1)
  /VARIABLES=G3
  /CRITERIA=CI(.95).

EXAMINE VARIABLES=G3 BY sex_code
  /PLOT BOXPLOT HISTOGRAM NPPLOT
  /STATISTICS DESCRIPTIVES
  /CINTERVAL 95
  /MISSING LISTWISE.

OUTPUT EXPORT
  /CONTENTS EXPORT=VISIBLE
  /PDF DOCUMENTFILE='Point-Biserial-Correlation-SPSS-Output.pdf'.

Excel Formulas for Point Biserial Correlation

Assume:
continuous_range = G3 values
binary_range = 0/1 coding for sex

N:
=COUNT(continuous_range)

n0:
=COUNTIF(binary_range,0)

n1:
=COUNTIF(binary_range,1)

Mean for code 0:
=AVERAGEIF(binary_range,0,continuous_range)

Mean for code 1:
=AVERAGEIF(binary_range,1,continuous_range)

Sample SD of continuous variable:
=STDEV.S(continuous_range)

Direct point-biserial correlation:
=CORREL(continuous_range,binary_range)

Manual point-biserial formula:
=((Mean1-Mean0)/SD_Y)*SQRT((n0*n1)/(N*(N-1)))

r squared:
=r_pb^2

t statistic:
=r_pb*SQRT((N-2)/(1-r_pb^2))

Degrees of freedom:
=N-2

Two-tailed p-value:
=T.DIST.2T(ABS(t),df)

Decision:
=IF(p_value<0.05,"Significant at alpha = 0.05","Not significant at alpha = 0.05")

Assumptions and Data Checks

Point Biserial Correlation is closely related to Pearson correlation and the independent-samples t-test. It does not require both variables to be continuous, but it does require one true binary variable and one continuous outcome. The observations should be independent, and the continuous outcome should be interpreted sensibly within each group.

The main caution is effect-size interpretation. Since r_pb is only -0.129, the result is small even though it is statistically significant. This is a common situation in large datasets.

How to Report Point Biserial Correlation

A complete report should include the binary coding, continuous variable, group means, sample size, coefficient, t statistic, degrees of freedom, p-value, effect-size strength and interpretation of direction.

APA-Style Full Report

Short APA-Style Version

Plain-Language Version

Common Mistakes in Point Biserial Correlation Interpretation

Downloads and Resources

External References

For additional learning, review documentation and references on Pearson correlation, point-biserial correlation, independent-samples t-test, binary coding, effect size interpretation and confidence intervals. These topics are commonly taught together because point-biserial correlation is the correlation form of a two-group mean comparison.

Related Statistical Guides