Confidence Interval for Difference of Two Means (Unequal Variances - Welch’s Method)

Use this calculator for confidence intervals when comparing two independent means where sample standard deviations are unknown and unequal (Welch’s method is more robust).

When to Use This Calculator

  • Two independent samples with different variances
  • Unequal population variances - don’t assume σ₁ = σ₂
  • Sample SDs unknown - estimated from samples
  • Most robust choice - use this unless you’ve verified equal variances with Levene’s test
  • Default recommendation when unsure whether variances are equal

How to Use

Step 1: Enter both sample means

Step 2: Enter both sample sizes

Step 3: Enter both sample standard deviations

Step 4: Select confidence level (typically 95%)

Step 5: Click “Calculate”

Confidence Interval Calculator for Difference between means
  Sample 1 Sample 2
Sample Mean
Sample Size
Standard Deviation
Confidence Level ($1-\alpha$)
Results
Standard Error of Diff. of Means:
Degrees of Freedom:
t-critical value: ($t_{\alpha/2,df}$)
Margin of Error: ($E$)
Lower Confidence Limits:
Upper Confidence Limits:

Welch’s Method (Unequal Variances)

$$CI = (\overline{x}_1 - \overline{x}2) \pm t{\alpha/2, df} \times \sqrt{\frac{s_1^2}{n_1} + \frac{s_2^2}{n_2}}$$

Welch-Satterthwaite degrees of freedom: $$df = \frac{\left(\frac{s_1^2}{n_1} + \frac{s_2^2}{n_2}\right)^2}{\frac{(s_1^2/n_1)^2}{n_1-1} + \frac{(s_2^2/n_2)^2}{n_2-1}}$$


Advantages of Welch’s Method

  • Doesn’t assume equal variances - more conservative
  • Better Type I error control - maintains correct rejection rate
  • Recommended by statisticians - robust choice when unsure
  • Use by default - when you haven’t tested for equal variances

Worked Example

Scenario: Testing if Brand A (s₁=15, n₁=30, x̄₁=85) differs from Brand B (s₂=25, n₂=25, x̄₂=75).

Solution:

  • $SE = \sqrt{\frac{225}{30} + \frac{625}{25}} = 4.15$
  • Welch df ≈ 40
  • $t_{0.025,40} = 2.021$
  • E = 2.021 × 4.15 = 8.39
  • CI = 10 ± 8.39 = [1.61, 18.39]

Interpretation: We’re 95% confident Brand A differs from Brand B by 1.61 to 18.39 points.


When variances are equal: Use equal variances method

Related: Tutorial