Quick Answer: Biserial Correlation Result

The main SPSS, Python and R report tested the association between sex and G3 final grade. The binary variable was coded F = 0 and M = 1. The continuous variable was G3. The sample contained N = 649 students, with 383 students in the F group and 266 students in the M group.

The group means showed that the F group had a higher mean G3 score (M = 12.25) than the M group (M = 11.41). Because M was coded as 1 and M had the lower mean, the coefficient was negative. The point-biserial result was r = -0.1291, with a 95% confidence interval from approximately -0.2040 to -0.0526. The test statistic was t(647) = -3.3109, with p = .000982. The normal-cut corrected biserial estimate was r_b = -0.1632.

What Is Biserial Correlation?

Biserial correlation is a correlation coefficient used when one variable is continuous and the other variable is binary because it has been cut from an underlying continuous or normally distributed trait. The word “biserial” means the analysis has one continuous series and one two-category series. The two-category series is treated as a thresholded version of something more continuous.

For example, suppose a researcher turns an exam score into a pass/fail variable using a cutoff of 10. The pass/fail variable has only two categories, but it was created from a continuous score. In that situation, the biserial correlation estimates what the relationship would look like before the continuous variable was split into two groups.

In contrast, if the binary variable is naturally binary, such as yes/no, present/absent, treatment/control or female/male coding, the point-biserial correlation is usually the cleaner coefficient. The point-biserial coefficient is simply the Pearson correlation between a 0/1 binary variable and a continuous variable. It is closely related to an independent-samples Student’s t-test, two-sample t-test and effect size reporting.

Biserial vs Point-Biserial vs Rank-Biserial Correlation

Many students search for biserial and point biserial correlation together because the names sound similar. They are connected, but they are not the same. The most important decision is whether the binary variable is naturally binary or artificially dichotomized.

Biserial Correlation Formula

The point-biserial formula uses the difference between two group means, the standard deviation of the continuous variable and the proportions in the two binary groups:

The biserial formula adds a correction for the normal cutoff that produced the binary groups:

The significance test is usually based on the observed point-biserial correlation because it is a standard Pearson correlation with a 0/1 variable. The common t formula is:

For the main report, this gives t(647) = -3.3109 and p = .000982. You can connect this interpretation with p-value, confidence interval, standard deviation and standard error guides.

Null and Alternative Hypotheses for Biserial Correlation

The hypothesis statement depends on whether the researcher is emphasizing the observed point-biserial association or the normal-cut corrected biserial estimate. In applied reports, the p-value is commonly tied to the point-biserial test.

Dataset and Variables Used

The main SPSS, Python and R charts use the student performance dataset with G3 final grade as the continuous variable and sex as the binary grouping variable. The report coded F = 0 and M = 1. This example is excellent for teaching the difference between point-biserial and biserial correlation because sex is a true binary group, while the Excel workbook also includes an artificial dichotomy example using G1 >= 10.

Before reporting the coefficient, it is useful to check descriptive statistics, frequency distributions, histograms, box plots, null and alternative hypotheses and parametric vs nonparametric tests.

SPSS Output Interpretation for Biserial Correlation

The SPSS output verifies the main binary-continuous result. The binary grouping variable has 383 F cases and 266 M cases, with no missing cases in the valid analysis. The continuous G3 score has N = 649, minimum 0, maximum 19, mean approximately 11.91 and standard deviation approximately 3.231.

Open the SPSS Biserial Correlation output PDF

SPSS Interpretation Summary

The SPSS result should be reported as a small but statistically significant binary-continuous association. The sign is a coding sign, not a value judgment: because M was coded 1 and the M mean was lower, the coefficient is negative. If the coding were reversed, the sign would become positive while the strength would remain the same.

Python Chart-by-Chart Interpretation

The Python charts show how the coefficient is built from group distributions, group means, coefficient comparison and a one-page report card. The charts are useful because they show why a statistically significant p-value can still represent a small practical association.

Python Chart 1: Continuous Scores by Binary Group

This chart shows the G3 distribution separately for F and M. The F group has a slightly higher center than the M group, with the F median around 12 and the M median around 11. The boxes still overlap strongly, which means the association is not large even though the mean difference is statistically significant. The F group has one very low outlier at 0, while the M group shows several low outliers near 0 or 1 and a few high values near 18 to 19.

The chart explains why the point-biserial coefficient is small. The two groups do not form clearly separated distributions. Instead, most scores occupy the same broad grade range, with only a modest downward shift for the M-coded group. This is exactly the pattern that produces a small negative correlation such as r = -0.129.

Python Chart 2: Group Mean Difference

The mean chart gives the clearest numerical story. The F group mean is labeled about 12.25, while the M group mean is about 11.41. The difference is roughly 0.85 grade points. The error bars are small because the sample is large, so even this modest group gap becomes statistically significant.

This chart should be read together with the boxplot. The mean gap is real, but the distributions overlap. Therefore, the best conclusion is not “the groups are very different.” The better conclusion is that the F-coded group scored slightly higher on average, producing a statistically significant but small point-biserial association.

Python Chart 3: Point-Biserial vs Biserial Coefficient

This chart shows two coefficient bars below zero. The point-biserial coefficient is labeled -0.129, and the biserial coefficient is labeled -0.163. Both signs are negative because the group coded 1 has the lower G3 mean. The biserial estimate is larger in magnitude because it adjusts the observed 0/1 split as though it came from an underlying normal threshold.

The chart is especially useful for explaining the difference between the two coefficients. Point-biserial r is the direct observed correlation with the coded binary variable. Biserial r is an adjusted estimate. Since sex is a true binary variable, the point-biserial value is the cleaner main report statistic, while the biserial value is useful as a demonstration of the correction.

Python Chart 4: Jittered Binary Scatter

The jittered scatterplot displays the individual student scores instead of only summary boxes or means. The horizontal separation represents the binary groups, while the vertical position represents G3. The horizontal mean line is higher for F than for M, matching the mean chart. The individual dots show substantial overlap between groups, especially from about 8 to 16.

This chart prevents overinterpretation. The significant result does not mean every F-coded student scored higher than every M-coded student. Many students in both groups have similar grades. The association is driven by a small average shift and the presence of slightly more lower-end values in the M-coded group.

Python Chart 5: Statistical Summary Report Card

The report card gives the final output in one place: continuous variable G3, binary variable sex, N = 649, n0 = 383, n1 = 266, point-biserial r = -0.1291, 95% CI [-0.2040, -0.0526], t(647) = -3.3109, p = .000982, and biserial r = -0.1632. The decision line states that the association is significant at alpha .05.

This is the best chart to use when writing the final result paragraph. It contains the sample size, coding, coefficient, confidence interval, test statistic, p-value and decision. The interpretation should still mention that the effect is small, because statistical significance is strengthened by the large sample size.

R Chart-by-Chart Validation

The R charts validate the same result using a separate workflow. They repeat the distribution, mean, coefficient, scatter and report-card views with a colorful style. This software-to-software agreement supports the reliability of the reported values.

R Chart 1: Colorful Scores by Binary Group

The R boxplot confirms the Python pattern. The F group has a slightly higher central line and a wider upper spread, while the M group has a lower center and multiple low outlying values. Both groups cover much of the same G3 range, so the practical separation is limited.

This visual agreement matters because it confirms that the small negative correlation is not a Python plotting artifact. The same group pattern appears when the analysis is recreated in R.

R Chart 2: Colorful Group Mean Difference

The R mean chart labels the same average values: approximately 12.25 for F and 11.41 for M. The bars make the direction easy to see, and the error bars show that the group means are estimated precisely in this large sample.

The correct report sentence should connect direction and coding. Since M is coded 1, the lower M mean produces a negative point-biserial coefficient. Reversing the coding would reverse the sign but not the strength.

R Chart 3: Colorful Coefficient Comparison

The R coefficient chart again shows point-biserial r = -0.129 and biserial r = -0.163. The larger absolute biserial estimate reflects the normal-cut adjustment. The values are both small in magnitude, so the result is statistically detectable but not practically large.

This chart is useful for teaching because it visually separates the observed coefficient from the corrected estimate. It also makes clear that the biserial correction does not change the direction of the association.

R Chart 4: Colorful Binary Scatter

The R scatterplot shows two vertical clouds of observations. The F group and M group overlap heavily across the main score range, while the mean marker for F is above the mean marker for M. This is a small-shift pattern, not a strong-separation pattern.

The scatterplot is the most transparent visual because every observation is shown. It helps readers understand why a small coefficient can still be significant when the dataset is large.

R Chart 5: Colorful Statistical Summary

The R report card matches the Python report card: N = 649, n0 = 383, n1 = 266, point-biserial r = -0.1291, 95% CI [-0.2040, -0.0526], t(647) = -3.3109, p = .000982, and biserial r = -0.1632. This confirms the same final decision in R.

The final interpretation remains the same: the association is statistically significant, but the magnitude is small. The result should be reported with direction, coding, confidence interval and practical effect size language.

Excel Results Explained

The uploaded Excel workbook adds an important extra result. It does not only repeat the sex example. It also gives a fully worked formula-based example where G1 is artificially dichotomized using the rule G1 >= 10 and then compared with continuous G3. This is a stronger demonstration of true biserial correlation because G1 was originally a numeric grade variable before it was converted into two groups.

Excel Worked Example: G1 Dichotomized at 10 Against G3

This Excel result is much stronger than the sex example because the split variable is directly related to school performance. When G1 is converted into a high/pass versus low/fail group, the group means on G3 are far apart: 13.02 versus 8.41. The point-biserial coefficient of 0.6117 already shows a strong observed relationship, and the biserial correction increases the estimate to 0.8390 because the binary split is treated as a cut of an underlying continuous grade variable.

The workbook also calculates median-split biserial relationships for several numeric variables against G3. The strongest corrected biserial result is G2 with r_b = 0.8841, followed by G1 with r_b = 0.8281. Studytime is positive but smaller (r_b = 0.2976). Mother’s education and father’s education are also positive and significant. Some lifestyle variables are negative, including weekend alcohol use (Walc, r_b = -0.1405) and free time (freetime, r_b = -0.1217). Age, absences, health and goout are not statistically significant in the workbook table.

Excel Formula Steps Used in the Workbook

SPSS, R, Python and Excel Workflows for Biserial Correlation

The same analysis can be reproduced in all four tools. The exact method depends on whether the binary variable is true binary or artificially dichotomized.

Code Blocks for Biserial Correlation

SPSS Syntax for Biserial Correlation

* Biserial / point-biserial correlation in SPSS.
* Continuous variable: G3.
* Binary variable: sex coded F = 0 and M = 1.

OUTPUT CLOSE ALL.
OUTPUT NEW NAME=Biserial_Output.

* Example assumes data are already open in SPSS.
RECODE sex ('F'=0) ('M'=1) INTO sex_binary.
VARIABLE LABELS sex_binary 'Binary variable coded 0/1 for point-biserial and biserial correlation'.
VALUE LABELS sex_binary 0 'F / group 0' 1 'M / group 1'.
EXECUTE.

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

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

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

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

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

Python Code for Biserial Correlation

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

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

# Continuous variable and binary coding
work = df[["sex", "G3"]].dropna().copy()
work["sex_binary"] = work["sex"].map({"F": 0, "M": 1})
work["G3"] = pd.to_numeric(work["G3"], errors="coerce")
work = work.dropna()

x = work["sex_binary"].to_numpy()
y = work["G3"].to_numpy()
n = len(work)

# Point-biserial correlation
r_pb, p_value = stats.pearsonr(x, y)
t_value = r_pb * np.sqrt((n - 2) / (1 - r_pb**2))
df_t = n - 2

# Group summaries
m0 = y[x == 0].mean()
m1 = y[x == 1].mean()
sd_y = y.std(ddof=1)
p = (x == 1).mean()
q = (x == 0).mean()
z_cutoff = stats.norm.ppf(q)
y_ordinate = stats.norm.pdf(z_cutoff)

# Biserial correlation
r_b = ((m1 - m0) / sd_y) * ((p * q) / y_ordinate)

# Fisher confidence interval for point-biserial r
z = np.arctanh(r_pb)
se = 1 / np.sqrt(n - 3)
ci_low, ci_high = np.tanh([z - 1.96 * se, z + 1.96 * se])

print("N:", n)
print("Mean group 0:", m0, "Mean group 1:", m1)
print("Point-biserial r:", r_pb)
print("95% CI:", ci_low, ci_high)
print("t:", t_value, "df:", df_t, "p:", p_value)
print("Biserial r:", r_b)

R Code for Biserial Correlation

# Biserial / point-biserial correlation in R

df <- read.csv("dataset.csv")
work <- na.omit(df[, c("sex", "G3")])
work$sex_binary <- ifelse(work$sex == "M", 1, 0)
work$G3 <- as.numeric(work$G3)

# Point-biserial correlation is Pearson correlation with 0/1 coding
cor_test <- cor.test(work$sex_binary, work$G3)
print(cor_test)

n <- nrow(work)
m0 <- mean(work$G3[work$sex_binary == 0])
m1 <- mean(work$G3[work$sex_binary == 1])
sd_y <- sd(work$G3)
p <- mean(work$sex_binary == 1)
q <- mean(work$sex_binary == 0)
z_cutoff <- qnorm(q)
y_ordinate <- dnorm(z_cutoff)

r_b <- ((m1 - m0) / sd_y) * ((p * q) / y_ordinate)
cat("Biserial r =", r_b, "\n")

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

Excel Formulas for Biserial Correlation

Assume:
Y_range = continuous G3 values
binary_range = 0/1 group coding

N:
=COUNT(Y_range)

n1 and n0:
=COUNTIF(binary_range,1)
=COUNTIF(binary_range,0)

p and q:
=n1/N
=n0/N

Group means:
=AVERAGEIF(binary_range,1,Y_range)
=AVERAGEIF(binary_range,0,Y_range)

Standard deviation of Y:
=STDEV.S(Y_range)

Point-biserial r:
=((M1-M0)/SD_Y)*SQRT(p*q)

Normal cutoff and ordinate:
=NORM.S.INV(q)
=NORM.S.DIST(z_cutoff,FALSE)

Biserial r:
=((M1-M0)/SD_Y)*(p*q/y_ordinate)

Significance test based on point-biserial r:
=r_pb*SQRT((N-2)/(1-r_pb^2))
=T.DIST.2T(ABS(t),N-2)

APA Reporting Wording for Biserial Correlation

When reporting a Biserial Correlation or point-biserial correlation, describe the continuous variable, the binary coding, group counts, group means, coefficient, confidence interval, p-value, and whether the coefficient is an observed point-biserial result or a biserial normal-cut correction.

APA-Style Full Report

Short APA-Style Version

Excel Worked-Example Wording

Common Mistakes in Biserial Correlation Interpretation

When to Use Biserial Correlation

Use biserial correlation when one variable is continuous and the other variable is binary because it was created by cutting a continuous or latent variable into two groups. Examples include pass/fail from an exam score, high/low anxiety from a scale score, or high/low performance from a continuous grade.

Use point-biserial correlation when the binary variable is naturally binary. Examples include treatment/control, yes/no, present/absent, group 0/group 1 or sex coded as 0/1. Use rank-biserial correlation when the analysis is rank-based or tied to a Mann-Whitney style comparison. If both variables are continuous, use ordinary Pearson correlation instead.

Downloads and Resources for Biserial Correlation

Related Statistical Guides