# Constructing a Confidence Interval for the Difference Between Two Population Means

> AP Statistics · AP 2024 Statistics
> Source: https://www.owlsprep.com/study/ap-statistics-u13-constructing-a-confidence-interval-for/

We cover required condition checks, step-by-step interval calculation, proper context-specific interpretation, and AP rubric-aligned scoring rules for estimating the difference between two independent population means.

**Prerequisites:** [One-sample t confidence interval construction](https://www.owlsprep.com/study/ap-statistics-u12-one-sample-t-confidence-intervals/); [Sampling distributions for difference in sample means](https://www.owlsprep.com/study/ap-statistics-u13-sampling-distribution-difference-means/)

## Learning objectives

- Identify all required conditions to construct a valid two-sample t confidence interval for the difference in independent population means
- Calculate the point estimate, standard error, critical t-value, and final interval bounds using the unpooled formula
- Write a full-credit AP rubric-aligned interpretation of the completed interval in problem context
- Distinguish between appropriate use cases for unpooled and pooled variance two-sample t intervals

## Pre-Construction Condition Verification

Before you can calculate a valid two-sample t confidence interval, you must explicitly verify three non-negotiable conditions, which are worth 1-2 points on every AP free response question testing this skill.

**Independence Between Groups** — The two samples must be drawn from separate, independent populations, with no paired or matched observations across groups.

*Example:* Comparing test scores of two separate classes counts as independent, while comparing pre-test and post-test scores for the same student does not.

1. Random: Both samples must be selected via random sampling from their respective populations, or groups must be randomly assigned in an experiment.
2. 10% Condition: When sampling without replacement, each sample size must be less than 10% of its corresponding population.
3. Normality/Large Sample: At least one of the following holds: both underlying populations are normally distributed, or each sample size n ≥ 30.

> **warning**
>
> Failing to state all three conditions explicitly will cost you full credit on AP free response, even if your final interval calculation is perfect.

**Check your understanding**

Identify which scenario meets all conditions for a two-sample t interval:

1. Scenario A: Paired pre-test and post-test scores for 25 students

   - Valid
   - Invalid

   *Why:* Paired data requires a one-sample t interval for mean differences, not a two-sample interval.

2. Scenario B: Random samples of 40 male and 40 female AP Stats students, drawn from a school of 1200 total students

   - Valid
   - Invalid

   *Why:* Both samples are random, under 10% of their respective populations, and n ≥ 30 so normality condition is satisfied.

## Core Two-Sample t Interval Calculation

$$(\bar{x}_1 - \bar{x}_2) \pm t^* \cdot \sqrt{\frac{s_1^2}{n_1} + \frac{s_2^2}{n_2}}$$

This unpooled formula is the default accepted by the AP College Board for all two-sample t interval questions, even if variances are equal. You do not need to use the pooled variance method unless explicitly instructed.

**Welch-Satterthwaite Degrees of Freedom** — The adjusted degrees of freedom calculated to avoid overestimating interval width for unequal sample variances, given by the formula $df = \frac{\left(\frac{s_1^2}{n_1} + \frac{s_2^2}{n_2}\right)^2}{\frac{(s_1^2/n_1)^2}{n_1-1} + \frac{(s_2^2/n_2)^2}{n_2-1}}$

**Worked example:** A researcher collects a random sample of 32 coffee drinkers who consume 1 cup daily, finding mean daily sleep of 7.2 hours with standard deviation 0.8 hours. A second random sample of 36 non-coffee drinkers finds mean sleep of 7.8 hours with standard deviation 0.7 hours. Construct a 95% confidence interval for the difference in population mean sleep (non-coffee minus coffee).

1. Calculate the point estimate:
2. $$\bar{x}_1 - \bar{x}_2 = 7.8 - 7.2 = 0.6 \text{ hours}$$
3. Calculate the standard error:
4. $$SE = \sqrt{\frac{0.7^2}{36} + \frac{0.8^2}{32}} \approx 0.175$$
5. Find the t* critical value: Using Welch-Satterthwaite df ≈ 62, 95% confidence gives t* ≈ 2.0
6. Compute the margin of error and final interval:
7. $$0.6 \pm (2.0 \times 0.175) = (0.25, 0.95) \text{ hours}$$

> **Exam tip:** You can safely use the conservative df of the smaller of n₁-1 and n₂-1 on the AP exam for full credit, no need to calculate the full Welch-Satterthwaite value manually.

## Full Credit Interval Interpretation

A full credit interpretation of a two-sample confidence interval must include four mandatory components: confidence level, reference to the true difference in population parameters, context of the problem, and explicit bounds of the interval.

**Exam command terms**

AP exam questions will often ask you to 'interpret the interval' which requires all four components, not just a partial statement.

- **We are 95% confident** — The correct opening that references the confidence level, not a probability statement about the specific interval.

- **the true difference between the population mean of [group 1] and [group 2]** — You must explicitly reference population parameters, not sample statistics, to earn full credit.

> **Interpretation Mnemonic**
>
> Use the mnemonic C-P-C-B to remember: Confidence level, Population parameter, Context, Bounds.

## Pooled Variance Special Case

The pooled variance method is only appropriate if you are explicitly told that the two population variances are equal, a scenario that almost never appears on the AP exam.

$$s_p^2 = \frac{(n_1 - 1)s_1^2 + (n_2 - 1)s_2^2}{n_1 + n_2 - 2}$$

- Pooled degrees of freedom = n₁ + n₂ - 2
- Pooled standard error replaces the unpooled SE term in the core interval formula

## Common pitfalls

- **Wrong:** Using the two-sample independent interval formula for paired matched data
  - Why it fails: Paired data reduces variability and uses a different sampling distribution, leading to an overly wide, incorrect interval.
  - Correct: First confirm independence between groups, use a paired t interval if observations are matched across groups.
- **Wrong:** Saying 'there is a 95% probability that the true difference lies inside this specific interval'
  - Why it fails: The interval and parameter are both fixed values; probability refers to the long-run proportion of intervals that capture the true value.
  - Correct: Frame the interpretation as confidence in the method, not probability for the single calculated interval.
- **Wrong:** Swapping the order of subtraction for the two sample means without documenting the direction
  - Why it fails: You will misinterpret which group has a higher/lower mean, losing all interpretation points.
  - Correct: Explicitly state your subtraction order (group 1 minus group 2) before calculating the point estimate.
- **Wrong:** Using a z critical value instead of a t critical value for unknown population standard deviations
  - Why it fails: Population standard deviations are almost never known for real two-sample inference scenarios, making z invalid.
  - Correct: Always use t* for two-sample intervals for means, unless population σ values are explicitly provided.
- **Wrong:** Forgetting to verify the 10% condition for both samples when sampling without replacement
  - Why it fails: This violates the independence requirement for the sampling distribution, making the standard error calculation unreliable.
  - Correct: Explicitly confirm both samples are less than 10% of their respective populations.

## Cheatsheet

| Component | Formula / Rule | AP Scoring Note |
| --- | --- | --- |
| Point Estimate | $\bar{x}_1 - \bar{x}_2$ | Must explicitly state subtraction order |
| Standard Error | $\sqrt{s_1^2/n_1 + s_2^2/n_2}$ | Unpooled is default, no pooled unless told equal variance |
| Degrees of Freedom | min(n₁-1, n₂-1) or Welch-Satterthwaite | Conservative min method earns full AP credit |
| Interpretation | C-P-C-B structure | Missing any component loses 1 point |
| Conditions | Random, 10%, Normality | Must state all three explicitly for full credit |

## What's next

Mastering two-sample t confidence intervals is the foundation for the corresponding hypothesis test for the difference in two population means, which is one of the most frequently tested free response topics on the AP Stats exam. You will also build on this skill to compare multiple group means using ANOVA in later units, and learn to distinguish between paired and independent two-sample inference scenarios to avoid costly method selection errors. This topic accounts for roughly 10-15% of the total exam score, so consistent practice with condition checks, calculation, and rubric-aligned interpretation is critical to earning a 5 on your exam.

---

From [OwlsPrep](https://www.owlsprep.com) — free study guides for A-Level, IB, AP and IGCSE, written against the official syllabus. Canonical page: https://www.owlsprep.com/study/ap-statistics-u13-constructing-a-confidence-interval-for/