How to Calculate Pooled Variance of Two Samples
Enter sample sizes and either standard deviations or variances. Get pooled variance, pooled standard deviation, degrees of freedom, and a visual comparison chart.
Input Settings
Formula used: pooled variance = [((n1 – 1) x s1 squared) + ((n2 – 1) x s2 squared)] / (n1 + n2 – 2).
Sample 1 and Sample 2
Expert Guide: How to Calculate Pooled Variance of Two Samples
Pooled variance is one of the most practical statistics in inferential analysis when you want to compare two groups and you can reasonably assume both groups come from populations with the same true variance. In plain language, pooled variance gives you a single, weighted estimate of spread by combining two sample variances. If you are doing a two-sample t-test with equal variances, calculating Cohen’s d using pooled standard deviation, or building confidence intervals based on equal-variance assumptions, this number is central.
Many people memorize the pooled variance formula but miss the intuition. The reason we pool is that each sample variance estimates the same population variance with noise. A sample with a larger size usually provides a more stable estimate, so we should weight it more. That is exactly why the formula uses each sample’s degrees of freedom, not just a simple average of two variances.
What Is Pooled Variance?
Pooled variance is a weighted average of two sample variances where weights are based on degrees of freedom. For two samples:
pooled variance = [((n1 – 1) x s1 squared) + ((n2 – 1) x s2 squared)] / (n1 + n2 – 2)
- n1, n2: sample sizes for group 1 and group 2
- s1 squared, s2 squared: sample variances for each group
- n1 – 1, n2 – 1: degrees of freedom attached to each variance estimate
If your input data are standard deviations, square each standard deviation first to obtain variance. After you calculate pooled variance, you can take the square root to get pooled standard deviation.
Why the Weighting Uses n Minus 1
Each sample variance is calculated from residuals around a sample mean, and estimating that mean consumes one degree of freedom. This is why the denominator in sample variance is n minus 1. When pooling variances, the correct weight is therefore the degrees of freedom for each estimate, not the sample size itself and definitely not equal 50-50 weighting unless both n values are exactly equal.
A quick practical takeaway: if one group is much larger, its variance estimate should influence pooled variance more strongly. Pooled variance naturally accomplishes that, while an unweighted average can distort downstream testing and effect sizes.
Step-by-Step Manual Calculation
- Collect n1, n2, and either variances or standard deviations for both groups.
- If you have SDs, square them to convert to variances.
- Multiply each variance by its degrees of freedom: (n1 – 1) and (n2 – 1).
- Add these two weighted sums.
- Add degrees of freedom: n1 + n2 – 2.
- Divide weighted sum by total degrees of freedom.
- Optionally square-root the result to get pooled standard deviation.
Example: suppose sample 1 has n1 = 25 and s1 = 4.2, sample 2 has n2 = 30 and s2 = 3.9.
- s1 squared = 17.64
- s2 squared = 15.21
- weighted sum = (24 x 17.64) + (29 x 15.21) = 423.36 + 441.09 = 864.45
- total df = 25 + 30 – 2 = 53
- pooled variance = 864.45 / 53 = 16.310
- pooled SD = square root(16.310) = 4.039
This gives one combined estimate of variability that can be used in equal-variance inferential procedures.
Comparison Table: Pooled Variance vs Simple Average of Variances
| Scenario | n1 | s1 squared | n2 | s2 squared | Simple average | Pooled variance |
|---|---|---|---|---|---|---|
| Balanced samples | 40 | 20.25 | 40 | 15.21 | 17.73 | 17.73 |
| Moderately unbalanced | 20 | 25.00 | 60 | 16.00 | 20.50 | 18.19 |
| Strongly unbalanced | 10 | 36.00 | 90 | 16.00 | 26.00 | 17.82 |
Notice how pooled variance stays close to the larger sample’s variance when sample sizes are highly unbalanced. That behavior is desirable because the larger sample provides a more stable estimate of spread.
When You Should Use Pooled Variance
- Two-sample t-tests under the equal variance assumption
- Confidence intervals for mean differences when equal variance is justified
- Effect size metrics such as Cohen’s d that often use pooled SD
- Educational, behavioral, and engineering settings where variance homogeneity is plausible
When You Should Not Use Pooled Variance
Do not pool if group variances are clearly different. If one group is much noisier than the other, equal-variance formulas can produce misleading p-values and confidence intervals. In that case, use methods that do not assume equal variances, such as Welch’s t-test.
- Visual spread differs a lot in boxplots or residual plots
- Variance ratio is large, especially with unbalanced sample sizes
- Domain knowledge suggests natural heteroscedasticity
Real-World Example: Comparing Exam Score Variability
Imagine two teaching formats were evaluated using the same standardized test. Group A has n = 52 students and SD = 11.4 points. Group B has n = 47 students and SD = 10.2 points. You want one variability estimate for an equal-variance t-test of mean scores.
- Convert to variances: 11.4 squared = 129.96, 10.2 squared = 104.04.
- Compute weighted sum: (51 x 129.96) + (46 x 104.04) = 6627.96 + 4785.84 = 11413.80.
- Total degrees of freedom: 52 + 47 – 2 = 97.
- Pooled variance: 11413.80 / 97 = 117.668.
- Pooled SD: square root(117.668) = 10.847.
This pooled SD can then be used in standardized mean difference formulas and equal-variance inferential calculations.
Reference Table: Common Inputs and Outputs
| Case | n1 | SD1 | n2 | SD2 | Pooled variance | Pooled SD |
|---|---|---|---|---|---|---|
| Clinical metric A | 35 | 5.6 | 33 | 5.1 | 28.718 | 5.359 |
| Manufacturing tolerance | 18 | 1.9 | 44 | 2.3 | 4.816 | 2.194 |
| Learning outcomes | 52 | 11.4 | 47 | 10.2 | 117.668 | 10.847 |
Common Mistakes and How to Avoid Them
- Mistake 1: Averaging SDs directly. You must pool variances, not standard deviations. Square first, pool, then square-root if needed.
- Mistake 2: Dividing by n1 + n2. The denominator is n1 + n2 – 2 because degrees of freedom are used.
- Mistake 3: Ignoring unequal variance evidence. If spreads differ strongly, prefer Welch methods.
- Mistake 4: Treating population and sample formulas as interchangeable. Pooled variance here uses sample-based df structure.
- Mistake 5: Using rounded intermediate values too early. Keep precision during calculations and round at the end.
Practical Decision Rule Before Pooling
Before you calculate pooled variance for hypothesis testing, do a quick check:
- Inspect group SDs and variances. Are they reasonably close?
- Check sample sizes. If highly unbalanced, be extra cautious with equal-variance assumptions.
- Review context. Does the measurement process suggest equal noise?
- If uncertain, compute both equal-variance and Welch results and compare conclusions.
In applied work, this protects against false confidence from a formula that is correct mathematically but wrong for the data-generating process.
Authoritative Resources for Deeper Study
- NIST Engineering Statistics Handbook (.gov)
- Penn State STAT 500 Applied Statistics (.edu)
- UC Berkeley Department of Statistics (.edu)
Final Takeaway
Calculating pooled variance of two samples is straightforward once you remember three principles: use variances, weight by degrees of freedom, and only apply pooling when equal variance is plausible. The calculator above automates these steps, reports pooled variance and pooled SD, and visualizes how each sample contributes. If you use this statistic carefully, it becomes a reliable foundation for robust two-group inference.