Justifying a Claim About the Difference of Two Means Based on a Confidence Interval - AP Statistics

SECTION 1

Historical Context & Motivation

Comparing two groups is one of the most fundamental tasks in empirical research. A physician might ask whether a new drug lowers blood pressure more than an existing treatment; an educator might investigate whether a revised curriculum raises test scores relative to the traditional approach. In each scenario, the core statistical question is the same: does a meaningful difference exist between two population means, or could the observed gap be attributable to random sampling variability? The intellectual lineage behind the confidence-interval approach to this question stretches across more than a century of statistical innovation.

1908 • Student's t-Distribution

William Sealy Gosset, publishing under the pseudonym "Student," derived the t-distribution to handle small samples drawn from a normal population, replacing the need for large-sample z-approximations.

1935 • Fisher's Design of Experiments

Ronald A. Fisher formalized experimental design principles—randomization, replication, and blocking—that made rigorous two-sample comparisons possible in agricultural and biological research.

1937 • Welch's Approximate t-Test

Bernard L. Welch proposed an adjustment to the two-sample t-procedure that does not assume equal population variances, yielding the Welch–Satterthwaite degrees-of-freedom approximation now standard on AP Statistics.

1960s–1980s • Shift Toward Interval Estimation

Statisticians including John Tukey and the American Statistical Association increasingly advocated reporting confidence intervals alongside—or instead of—p-values, emphasizing the practical magnitude of effects.

2016–Present • ASA Statement on P-Values

The ASA's formal statement urged researchers to supplement significance tests with interval estimates, reinforcing the central role of confidence intervals in modern inference and AP curriculum design.

Against this historical backdrop, the AP Statistics curriculum asks a precise question: given a confidence interval for µ₁ − µ₂, how do we use that interval to justify a claim about whether the two population means are different, and in which direction? This lesson develops the conceptual framework, the mathematical machinery, and the interpretive language you need to answer confidently on exam day.

SECTION 2

Core Principles & Definitions

Before constructing or interpreting a two-sample confidence interval, it is essential to anchor your understanding in a few foundational ideas. These principles govern when the procedure is valid, what the resulting interval means, and how you can leverage it to support or refute a claim about two populations.

Parameter of Interest: µ₁ − µ₂

The target parameter is the true difference between two population means. The point estimate is x̄₁ − x̄₂, calculated from independent random samples drawn from each population.

Sampling Distribution & Standard Error

Under the appropriate conditions, the sampling distribution of x̄₁ − x̄₂ is approximately normal. The standard error combines the variability from both samples: SE = √(s₁²/n₁ + s₂²/n₂).

Confidence Level & Margin of Error

A C% confidence interval captures the parameter in C% of all possible samples. The margin of error equals t* × SE, where t* comes from the t-distribution with the appropriate degrees of freedom.

Interpreting the Interval

If the interval contains 0, we lack convincing evidence that the means differ. If the entire interval is positive (or entirely negative), we have evidence of a directional difference at that confidence level.

Conditions for Inference

Three conditions must be verified: (1) Random—data from random samples or random assignment; (2) Independent—the 10% condition or independent groups; (3) Normal—sample sizes are large (n ≥ 30 each) or populations are approximately normal.

✦

KEY TAKEAWAY

Think of a confidence interval for µ₁ − µ₂ as a net of plausible values cast over the unknown true difference. If your net catches zero, you cannot rule it out—there may be no real difference. If the net lands entirely on one side of zero, you have evidence that one population mean genuinely exceeds the other, much like a quality-control engineer who measures the gap between two machines' outputs and only declares one superior when the entire margin of uncertainty falls on one side.

SECTION 3

Visual Explanation

A number line diagram is the most direct way to see whether a confidence interval for µ₁ − µ₂ supports a claim of a difference. The following diagram illustrates three scenarios that arise in practice: an interval that contains zero (inconclusive), an interval entirely above zero (evidence µ₁ > µ₂), and an interval entirely below zero (evidence µ₁ < µ₂).

Scenario A straddles zero, so the data do not provide convincing evidence of a difference. Scenario B lies entirely to the right of zero, supporting the claim that µ₁ is greater than µ₂. Scenario C lies entirely to the left, supporting the claim that µ₁ is less than µ₂. The hollow dot on each interval marks the point estimate x̄₁ − x̄₂.

The critical question is always whether the value zero is inside or outside the interval. Zero represents the null claim that the two populations share the same mean. When the interval excludes zero, every plausible value of µ₁ − µ₂ is either strictly positive or strictly negative, providing convincing evidence at the given confidence level that a real difference exists. When zero is captured, we cannot rule out the possibility that any observed sample difference is simply the product of chance.

SECTION 4

Mathematical Framework

The two-sample t-interval for µ₁ − µ₂ has the familiar structure of "point estimate ± margin of error." The formulas below show how the interval is constructed and how each component contributes to the final result.

POINT ESTIMATE

x̄₁ − x̄₂

x̄₁ and x̄₂ are the sample means from populations 1 and 2, respectively. This quantity estimates the true difference µ₁ − µ₂.

STANDARD ERROR

SE = √(s₁² / n₁ + s₂² / n₂)

s₁ and s₂ are the sample standard deviations; n₁ and n₂ are the sample sizes. Because the samples are independent, the variances add under the square root.

CONFIDENCE INTERVAL

(x̄₁ − x̄₂) ± t* × √(s₁² / n₁ + s₂² / n₂)

t* is the critical value from the t-distribution for the chosen confidence level C. Degrees of freedom are computed using the Welch–Satterthwaite approximation (or by calculator). The interval gives all plausible values for µ₁ − µ₂ at confidence level C.

WELCH–SATTERTHWAITE DEGREES OF FREEDOM

df = (s₁²/n₁ + s₂²/n₂)² / [ (s₁²/n₁)²/(n₁−1) + (s₂²/n₂)²/(n₂−1) ]

This approximation yields a non-integer df, which is typically truncated or handled by technology. On the AP exam, a calculator or table provides the critical value; you do not need to memorize this formula but should know it exists.

💡

Exam Tip

On the AP Statistics exam, you are expected to state, check, and verify the Random, Independent (10%), and Normal/Large Sample conditions before constructing the interval. Failure to address conditions is one of the most common point deductions on FRQs.

SECTION 5

Decision Rules & Interpretation

Once the confidence interval has been constructed, the interpretive step is what earns full credit on the AP exam. The decision rule is straightforward, but the language of justification must be precise. The following table codifies the three cases and the corresponding claim each supports.

Position of Interval Relative to 0	Conclusion	Justification Template
Entire interval is positive (both endpoints > 0)	Convincing evidence that µ₁ > µ₂	"Because the interval (L, U) lies entirely above 0, we are C% confident that µ₁ − µ₂ > 0, meaning µ₁ is greater than µ₂."
Entire interval is negative (both endpoints < 0)	Convincing evidence that µ₁ < µ₂	"Because the interval (L, U) lies entirely below 0, we are C% confident that µ₁ − µ₂ < 0, meaning µ₁ is less than µ₂."
Interval contains 0 (one endpoint negative, one positive)	No convincing evidence of a difference	"Because the interval (L, U) contains 0, it is plausible that µ₁ − µ₂ = 0, so there is not convincing evidence that the means differ."

Decision rules for a confidence interval for µ₁ − µ₂

This flowchart summarizes the decision process. After computing the confidence interval (L, U), check whether zero lies inside. If it does not, determine the sign of the interval to identify which mean is larger.

⚠

Common Mistake

Students sometimes say "we accept that the means are equal" when the interval contains zero. This is incorrect. Containing zero means we fail to find convincing evidence of a difference—absence of evidence is not evidence of absence. Always frame the conclusion in terms of lacking evidence, not accepting equality.

SECTION 6

Worked Example

A nutritionist wants to compare the mean daily caloric intake of adults in City A versus City B. She selects independent random samples of 35 adults from each city and records their daily caloric intake. The sample data are as follows: City A has x̄₁ = 2180 kcal, s₁ = 320 kcal, n₁ = 35. City B has x̄₂ = 2040 kcal, s₂ = 280 kcal, n₂ = 35. Construct a 95% confidence interval for µ₁ − µ₂ and determine whether there is convincing evidence that the mean caloric intakes differ.

Two-Sample t-Interval for µ₁ − µ₂

1
Step 1 — Define the Parameter

Let µ₁ = the true mean daily caloric intake for all adults in City A, and µ₂ = the true mean daily caloric intake for all adults in City B. We wish to estimate µ₁ − µ₂.
2
Step 2 — Check Conditions

Random: Both samples are independent random samples from their respective cities. Independent (10% condition): It is reasonable to assume there are more than 350 adults in each city, so each sample is less than 10% of its population. The two samples are also independent of each other. Normal/Large Sample: Both n₁ = 35 ≥ 30 and n₂ = 35 ≥ 30, so by the Central Limit Theorem, the sampling distribution of x̄₁ − x̄₂ is approximately normal.
3
Step 3 — Calculate the Point Estimate

x̄₁ − x̄₂ = 2180 − 2040 = 140 kcal.

Point estimate = 140 kcal
4
Step 4 — Compute the Standard Error

SE = √(s₁²/n₁ + s₂²/n₂) = √(320²/35 + 280²/35) = √(102400/35 + 78400/35) = √(2925.71 + 2240.00) = √5165.71 ≈ 71.87 kcal.

SE ≈ 71.87 kcal
5
Step 5 — Find t* and Compute the Interval

Using technology (or the Welch–Satterthwaite formula), df ≈ 66.1. For a 95% confidence interval, t* ≈ 1.997. The margin of error is ME = 1.997 × 71.87 ≈ 143.5 kcal. The confidence interval is 140 ± 143.5, which gives (−3.5, 283.5).

95% CI: (−3.5, 283.5)
6
Step 6 — Interpret and Justify

We are 95% confident that the true difference in mean daily caloric intake (City A − City B) is between −3.5 kcal and 283.5 kcal. Because this interval contains 0, there is not convincing evidence at the 95% confidence level that the true mean daily caloric intake differs between adults in City A and adults in City B. It is plausible that the two population means are equal.

SECTION 7

Confidence Intervals vs. Hypothesis Tests

Students often wonder about the relationship between a confidence interval for µ₁ − µ₂ and a two-sample t-test. Both procedures use the same underlying sampling distribution and conditions, but they answer slightly different questions. Understanding their connection deepens your ability to justify claims on the AP exam, where you may be asked to draw conclusions from either output.

Feature	Confidence Interval	Hypothesis Test
Output	A range of plausible values for µ₁ − µ₂	A p-value measuring evidence against H₀: µ₁ − µ₂ = 0
Decision criterion	Does the interval contain 0?	Is the p-value ≤ α?
Effect size information	Yes — the interval shows plausible magnitudes	Not directly — the p-value does not indicate magnitude
Consistency	A two-sided C% CI excludes 0 ⟺ the two-sided test rejects H₀ at α = 1 − C	A two-sided test at α rejects ⟺ the (1 − α)×100% CI excludes 0
Advantage	Communicates both direction and practical significance	Familiar reject/fail-to-reject framework; useful for one-sided alternatives

Comparing confidence intervals and hypothesis tests for two means

✦

KEY TAKEAWAY

A confidence interval is like reading a thermometer—it tells you the range and magnitude of the difference, not just whether the temperature crossed a single threshold. A hypothesis test only tells you whether the reading exceeded a critical mark (α). In research and on the AP exam, the interval provides richer information because it simultaneously addresses both statistical significance (does the interval exclude zero?) and practical significance (how large could the difference plausibly be?).

SECTION 8

Connection to Advanced Topics

The two-sample t-interval is a gateway to several more sophisticated techniques encountered in college-level and graduate statistics courses. Understanding the logic of "does the interval capture the null value?" transfers directly to each of these extensions.

AP Statistics (This Lesson)	Advanced Extension
Two independent samples, each with one quantitative variable	Paired t-interval (matched pairs), where the difference is computed within each pair before building a one-sample interval on the differences
Compare two groups' means	ANOVA and simultaneous confidence intervals (e.g., Tukey HSD) compare three or more group means while controlling the family-wise error rate
Welch t-interval with approximate df	Bootstrap confidence intervals that make no distributional assumption; resample with replacement to estimate the sampling distribution directly
Check whether 0 is in the interval (binary decision)	Equivalence testing (TOST procedure): reverse the logic and show the difference is small enough to be practically negligible

How the two-sample CI concept extends beyond AP Statistics

The interpretive skill you build here—checking whether a null value falls inside or outside an interval and articulating what that implies—remains the universal reasoning pattern across all of these methods. Whether you are comparing means, proportions, regression slopes, or any other parameter, the question is always: does the interval for the relevant difference include the value specified by the null claim? Mastering this logic now positions you well for future coursework in mathematical statistics, experimental design, and applied data science.

SECTION 9

Practice Problems

Question 1: CONCEPTUAL

A 95% confidence interval for µ₁ − µ₂ is (−4.2, 12.8). Which of the following is the best conclusion?

A. There is convincing evidence that µ₁ > µ₂ because the point estimate is positive.
B. There is convincing evidence that µ₁ < µ₂ because the lower bound is negative.
C. There is not convincing evidence that the two population means are different because the interval contains 0.
D. The two population means are equal because 0 is inside the interval.
E. We are 95% confident that every individual in population 1 scores between −4.2 and 12.8 higher than every individual in population 2.

Show answer (C)

C is correct. Because the interval (−4.2, 12.8) contains 0, it is plausible that µ₁ − µ₂ = 0, so there is not convincing evidence that the means differ. A is wrong because the positive point estimate alone is not sufficient when the interval spans zero; the margin of error accounts for sampling variability. B is wrong because having a negative lower bound does not by itself imply µ₁ < µ₂; you need the entire interval below 0 for that conclusion. D is wrong because we never conclude that two means are equal; we only say the data do not provide convincing evidence they differ. E is wrong because the interval estimates the difference of population means, not the difference between individual observations.

Question 2: BASIC CALCULATION

Two independent random samples yield: x̄₁ = 50, s₁ = 8, n₁ = 40, and x̄₂ = 45, s₂ = 10, n₂ = 50. The standard error of x̄₁ − x̄₂ is closest to which value?

A. 1.26
B. 1.89
C. 2.05
D. 3.60
E. 4.21

Show answer (B)

B is correct. The standard error of the difference of two independent sample means is SE = √(s₁²/n₁ + s₂²/n₂). Substituting the given values: SE = √(8²/40 + 10²/50) = √(64/40 + 100/50) = √(1.60 + 2.00) = √3.60 ≈ 1.897, which rounds to approximately 1.90. The closest listed value is 1.89. A (1.26) is the standard error of x̄₁ alone, since 8/√40 ≈ 1.265. This ignores the second sample entirely. C (2.05) does not result from the correct application of the formula with the given values. D (3.60) is the value of s₁²/n₁ + s₂²/n₂ before taking the square root. This is a common error—forgetting to take the final square root. E (4.21) results from incorrectly adding the standard deviations rather than working with variances, such as computing √((8+10)²/(40+50)) or a similar misapplication of the formula.

Question 3: INTERMEDIATE

A researcher obtains a 90% confidence interval for µ₁ − µ₂ of (2.3, 8.7). She then constructs a 99% confidence interval using the same data. Which of the following best describes the 99% interval and its implications?

A. The 99% interval is narrower and still excludes 0, providing stronger evidence of a difference.
B. The 99% interval is wider and must still exclude 0, providing stronger evidence of a difference.
C. The 99% interval is wider and may or may not contain 0, so the evidence for a difference may change.
D. The 99% interval is wider but the point estimate changes, so no comparison is possible.
E. The 99% interval is the same width but centered at a different point estimate.

Show answer (C)

C is correct. Increasing the confidence level from 90% to 99% increases t* and therefore increases the margin of error, producing a wider interval. The wider interval extends further in both directions from the same point estimate. Because the 90% interval's lower bound (2.3) is relatively close to 0, the wider 99% interval might extend below 0, which would change the conclusion. A is wrong because increasing confidence widens, not narrows, the interval. B is wrong because the wider interval is not guaranteed to exclude 0. D is wrong because the point estimate x̄₁ − x̄₂ does not change when you change the confidence level. E is wrong because the width changes with confidence level, and the center remains the same.

Question 4: APPLIED

5 pts

An agricultural scientist randomly assigns 25 plots to Fertilizer A and 25 plots to Fertilizer B, measuring crop yield (bushels per acre). She obtains: x̄_A = 142,

s_A

= 18, x̄_B = 128,

s_B

= 22. Construct a 95% confidence interval for µ_A − µ_B and use it to determine whether there is convincing evidence that Fertilizer A produces a higher mean yield than Fertilizer B. State the conditions, show your work, and write a conclusion in context.

Show answer

Rubric (5 points): Point 1 — Parameter and procedure (1 pt): Define µ_A = true mean yield with Fertilizer A and µ_B = true mean yield with Fertilizer B. State that a two-sample t-interval for µ_A − µ_B will be constructed. Point 2 — Conditions (1 pt): Random: Plots were randomly assigned to treatments. Independent: The two groups are independent due to random assignment, and the 10% condition is not required in an experiment with random assignment. Normal/Large Sample:

n_A

= 25 and

n_B

= 25 are close to 30; if the distributions of yield are not strongly skewed (no extreme outliers stated), the t-procedure is approximately valid. Point 3 — Calculations (1 pt): Point estimate: 142 − 128 = 14. SE = √(18²/25 + 22²/25) = √(324/25 + 484/25) = √(12.96 + 19.36) = √32.32 ≈ 5.686. Using technology, df ≈ 45.7, so t* ≈ 2.014 for 95% confidence. ME = 2.014 × 5.686 ≈ 11.45. CI: 14 ± 11.45 = (2.55, 25.45). Point 4 — Interpretation (1 pt): We are 95% confident that the true difference in mean crop yield (Fertilizer A − Fertilizer B) is between 2.55 and 25.45 bushels per acre. Point 5 — Conclusion with justification (1 pt): Because the entire 95% confidence interval lies above 0, there is convincing evidence that the true mean yield with Fertilizer A is higher than the true mean yield with Fertilizer B. Because plots were randomly assigned to treatments, this difference can be attributed to the fertilizer used.

Question 5: CRITICAL THINKING

6 pts

A 95% confidence interval for µ₁ − µ₂ is (0.5, 6.3), based on independent random samples from two populations. (a) A classmate claims, "Since the interval does not contain 0, we have proven that µ₁ ≠ µ₂." Explain what is wrong with this statement. (b) Another classmate says, "A 99% confidence interval from the same data would also exclude 0, so we'd have even stronger evidence." Evaluate this claim. (c) Suppose both sample sizes were quadrupled while the sample means and standard deviations remained the same. Describe how the 95% confidence interval would change and what effect this would have on the strength of evidence.

Show answer

Rubric (6 points): Part (a) — 2 points: (1 pt) The student should explain that statistical inference does not "prove" anything. A 95% confidence interval means that if we repeated the sampling process many times, approximately 95% of the intervals would capture the true parameter. There is still a 5% chance that this particular interval does not contain the true value of µ₁ − µ₂. (1 pt) The correct language is "there is convincing evidence at the 95% confidence level that µ₁ ≠ µ₂," not that it has been proven. Proof requires certainty; inference operates under uncertainty. Part (b) — 2 points: (1 pt) The claim is not necessarily correct. A 99% CI uses a larger t* value, producing a wider interval. The 95% interval is (0.5, 6.3) with center 3.4 and half-width 2.9. Increasing to 99% confidence increases the half-width, so the lower bound moves left. If the lower bound drops below 0, the 99% interval would contain 0. (1 pt) Furthermore, even if a 99% interval still excluded 0, the phrase "stronger evidence" is misleading. A wider interval trades precision for coverage probability. The strength of evidence is better assessed by examining the p-value; a CI that barely excludes 0 at 99% confidence actually indicates very strong evidence, but the student's reasoning (that higher confidence automatically means stronger evidence) is flawed. Part (c) — 2 points: (1 pt) Quadrupling both sample sizes would halve the standard error because SE is proportional to 1/√n, and √4 = 2. The margin of error would be approximately halved (slightly different due to a change in df and therefore t*, but the dominant effect is the halving of SE). (1 pt) The interval would become much narrower while remaining centered at 3.4. A narrower interval that still excludes 0 provides stronger evidence of a difference because it shows the difference is not only statistically significant but also estimated with greater precision. For example, the interval might become approximately (2.0, 4.8), reinforcing the conclusion that µ₁ > µ₂.

SUMMARY

Lesson Summary

Summary

A two-sample t-interval for µ₁ − µ₂ is constructed as (x̄₁ − x̄₂) ± t* × SE, where SE = √(s₁²/n₁ + s₂²/n₂). Before computing this interval, you must verify the Random, Independent, and Normal/Large Sample conditions. The degrees of freedom are computed via the Welch–Satterthwaite approximation and used to find the critical value t*.

To justify a claim, check whether zero lies inside or outside the interval. If the entire interval is above zero, there is convincing evidence that µ₁ > µ₂. If entirely below zero, there is convincing evidence that µ₁ < µ₂. If the interval contains zero, there is not convincing evidence of a difference—but this does not prove the means are equal. Always frame conclusions in context, reference the confidence level, and connect the interval's position to your inferential claim.