Justifying a Claim About the Slope of a Regression Model Based on a Confidence Interval - AP Statistics

SECTION 1

Historical Context & Motivation

The question of whether two quantitative variables are genuinely related—or whether an apparent linear trend is merely an artifact of sampling variability—has been central to statistical thinking since the late nineteenth century. Early pioneers of regression analysis recognized that a sample slope alone is insufficient evidence; one must also quantify the uncertainty surrounding that estimate before making claims about the population. The development of confidence intervals for regression slopes wove together contributions from several statistical traditions, culminating in the inferential framework that AP Statistics students use today.

1885 • Galton's Regression to the Mean

Francis Galton introduced the concept of regression while studying hereditary traits, observing that extreme parental measurements tended to produce less extreme offspring measurements—a phenomenon he called "regression toward mediocrity."

1908 • Student's t-Distribution

William Sealy Gosset, publishing under the pseudonym "Student," derived the t-distribution, which accounts for the additional uncertainty that arises when sample sizes are small and the population standard deviation is estimated from the data.

1922 • Fisher's Maximum Likelihood Framework

Ronald A. Fisher formalized the least-squares estimation of regression coefficients and derived the sampling distributions needed to construct confidence intervals and perform hypothesis tests for regression parameters.

1937 • Neyman's Confidence Interval Theory

Jerzy Neyman rigorously defined the confidence interval as a random interval that captures the true parameter value at a specified long-run rate, providing the theoretical backbone for the interval estimation of regression slopes.

These historical threads converge in a single practical question: given a sample least-squares regression line with slope b, how can we construct an interval of plausible values for the true population slope β, and how do we use that interval to justify or refute a claim about the nature of the linear relationship? This section of the course brings together your knowledge of confidence intervals and regression into a single, powerful inferential tool.

SECTION 2

Core Principles & Definitions

Before constructing or interpreting a confidence interval for the slope, you must internalize several foundational ideas. The population regression model assumes that for every value of the explanatory variable x, the response variable y follows a normal distribution whose mean lies on the true regression line μ_y = α + βx. The sample slope b is our best point estimate of β, but a confidence interval communicates the range of β values that are consistent with the observed data at a chosen confidence level.

Population Slope (β)

The true rate of change in the mean response for each one-unit increase in the explanatory variable. This is the fixed but unknown parameter we wish to estimate.

Sample Slope (b)

The least-squares estimate of β computed from a particular sample. Because it varies from sample to sample, b is a random variable with its own sampling distribution.

Standard Error of b ( $SE_b$ )

An estimate of the standard deviation of the sampling distribution of b. Smaller SE_b values indicate more precise slope estimates and produce narrower confidence intervals.

Confidence Level (C)

The long-run proportion of intervals—computed from repeated sampling—that would capture the true β. A 95% confidence level means roughly 95 out of 100 such intervals contain β.

Degrees of Freedom (df = n − 2)

The number of independent pieces of information remaining after estimating both the intercept and slope. The t-distribution with df = n − 2 governs the confidence interval's critical value.

✦

KEY TAKEAWAY

Think of a confidence interval for the slope as a net cast around your sample slope estimate. A narrow net (small SE_b) indicates that repeated experiments would yield similar slope estimates, so you can speak with precision. A wide net signals that sampling variability is large, and the true slope could plausibly be quite different from what you observed. If that net fails to capture zero, you have convincing evidence that the slope is not zero—meaning a genuine linear association exists in the population.

SECTION 3

Visual Explanation: Anatomy of a Slope Confidence Interval

The number line shows a 95% confidence interval for β running from 0.25 to 1.25, centered on the sample slope b = 0.75 (violet dot). The red vertical line marks β = 0, the value corresponding to "no linear relationship." Because zero falls outside the interval, we conclude that the data provide convincing evidence of a nonzero slope at the 95% confidence level.

The diagram above encapsulates the core reasoning used on the AP Statistics exam. When you are asked to justify a claim about the slope using a confidence interval, the argument follows a clear logical chain: (1) state the confidence interval, (2) note whether the claimed value of β—most commonly zero—falls inside or outside the interval, and (3) draw a conclusion. If zero is not contained in the interval, the data provide convincing evidence at that confidence level that a linear relationship exists. If zero is contained in the interval, you cannot rule out the possibility that β = 0, and thus you lack convincing evidence of a linear association.

SECTION 4

Mathematical Framework

The confidence interval for the population slope β is constructed using the sample slope, its standard error, and a critical value from the t-distribution. Understanding each component and the conditions required for the interval to be valid is essential.

CONFIDENCE INTERVAL FOR β

b ± t* × SE_b

where b is the sample slope, t* is the critical value from the t-distribution with df = n − 2 for the chosen confidence level, and SE_b is the standard error of the slope estimate.

STANDARD ERROR OF THE SLOPE

SE_b = s / √(Σ(xᵢ − x̄)²)

where s is the residual standard deviation (estimated σ from √(Σresiduals² / (n − 2))), and the denominator is the square root of the sum of squared deviations of x-values from their mean, which measures the spread of the explanatory variable.

Conditions for Inference

The validity of the confidence interval rests on four conditions, often remembered by the acronym LINE. First, the relationship between x and y must be Linear, which we verify by examining a scatterplot or residual plot for a random pattern with no curvature. Second, the observations must be Independent; this is satisfied by random sampling or random assignment, or by confirming that the sample is less than 10% of the population. Third, the residuals must be Normal—for each value of x, the distribution of y values (or equivalently, the residuals) should be approximately normal, checked via a histogram or normal probability plot of the residuals, though this condition becomes less critical as n grows large. Fourth, the variability of the residuals must be Equal across all x-values (homoscedasticity), verified by checking that the residual plot shows a roughly constant vertical spread.

MARGIN OF ERROR

ME = t* × SE_b

The margin of error determines the half-width of the confidence interval. A larger t* (from a higher confidence level or smaller df) or a larger SE_b widens the interval, reducing precision.

SECTION 5

Decision Logic: Does the Interval Capture Zero?

The critical reasoning step on the AP exam is linking the confidence interval to a claim about the slope. Virtually every free-response prompt in this unit reduces to one question: does the interval contain the value zero? The diagram below contrasts three scenarios, each leading to a different justified conclusion.

Scenario A shows an interval entirely to the right of zero (evidence of a positive linear relationship). Scenario B shows an interval entirely to the left of zero (evidence of a negative relationship). Scenario C shows an interval that straddles zero, so we cannot conclude that the slope differs from zero at that confidence level.

💡

AP Exam Tip

When justifying your claim, always cite the specific interval values and explicitly state whether the claimed value (typically 0) falls inside or outside the interval. A response like "Because the 95% CI is (0.25, 1.25) and this interval does not contain 0, there is convincing evidence at the 95% confidence level that the true slope is positive" earns full credit. Simply saying "the slope is significant" without referencing the interval values will typically cost you points.

Notice the deep connection between confidence intervals and two-sided hypothesis tests. A 95% confidence interval that does not contain zero corresponds exactly to rejecting H₀: β = 0 at the α = 0.05 significance level. This duality means you can use either a confidence interval or a hypothesis test to make the same conclusion, but on the AP exam, the prompt will specify which approach you should use. When asked to use a confidence interval, do not compute a p-value—use the interval directly.

SECTION 6

Worked Example

A researcher collects data on 22 randomly selected homes to investigate whether the size of a home (in hundreds of square feet) is linearly associated with its selling price (in thousands of dollars). A least-squares regression yields ŷ = 45.3 + 12.8x, with SE_b = 3.65. Construct a 95% confidence interval for the slope and use it to determine whether there is convincing evidence of a linear association between home size and selling price.

Confidence Interval for the Slope of Home Price vs. Size

1
Step 1 — State the Parameter

We wish to estimate β, the true slope of the population regression line relating home selling price (in thousands of dollars) to home size (in hundreds of square feet). The parameter β represents the average change in selling price for each additional 100 square feet of home size.
2
Step 2 — Check Conditions (LINE)

Linear: We assume a residual plot was examined and shows no clear pattern. Independent: The homes were randomly selected, and 22 is certainly less than 10% of all homes in the region. Normal: We assume a histogram or normal probability plot of the residuals shows no strong skewness or outliers. Equal variance: We assume the residual plot shows roughly constant spread across all x-values. With conditions satisfied, a t-interval for the slope is appropriate.
3
Step 3 — Identify Values and Find t*

Sample slope: b = 12.8. Standard error: SE_b = 3.65. Degrees of freedom: df = n − 2 = 22 − 2 = 20. For a 95% confidence level with df = 20, the critical value from the t-table is t* = 2.086.

t* = 2.086, df = 20
4
Step 4 — Compute the Interval

CI = b ± t* × SE_b = 12.8 ± 2.086 × 3.65 = 12.8 ± 7.614. Lower bound: 12.8 − 7.614 = 5.186. Upper bound: 12.8 + 7.614 = 20.414.

95% CI: (5.186, 20.414)
5
Step 5 — Interpret and Justify the Claim

We are 95% confident that the true slope of the population regression line relating selling price to home size is between 5.186 and 20.414 thousand dollars per hundred square feet. Because this interval does not contain 0, there is convincing evidence at the 95% confidence level that a positive linear association exists between home size and selling price in the population. Furthermore, since the entire interval is positive, each additional 100 square feet is associated with an increase in expected selling price.

Conclusion: Convincing evidence of a positive linear relationship (β > 0).

SECTION 7

Confidence Intervals vs. Hypothesis Tests for Slope

On the AP exam, you may be asked to use either a confidence interval or a hypothesis test to justify a claim about the slope. While the two approaches are mathematically equivalent for two-sided tests, they communicate different kinds of information, and the exam will specify which tool to use. Understanding the strengths and limitations of each approach ensures that you select the right strategy and justify your answer properly.

Feature	Confidence Interval	Hypothesis Test
Output	Range of plausible values for β	A p-value and decision (reject or fail to reject)
Answers the question	What values of β are consistent with the data?	Is there sufficient evidence that β ≠ 0?
Estimates effect size	Yes—bounds tell how steep the slope might be	No—only whether the effect is statistically distinguishable from 0
One-sided claims	Can address direction by inspecting sign of interval bounds	Can use a one-sided alternative hypothesis directly
Exam tip	Use when prompt says "construct/use a CI" or "estimate the slope"	Use when prompt says "test whether" or "significance test"

Comparison of two approaches to inference about the regression slope

✦

KEY TAKEAWAY

A confidence interval is like a fishing expedition that tells you how big the fish might be, whereas a hypothesis test only tells you whether or not you caught something. The CI for β gives you both the direction and the plausible magnitude of the relationship, which is often more informative than a simple yes-or-no p-value decision. In research contexts, confidence intervals are increasingly preferred because they convey practical significance alongside statistical significance.

SECTION 8

Connections to Advanced Inference

The single-slope confidence interval you study in AP Statistics is a gateway to a much broader set of inferential tools used in college-level regression courses. The table below previews how the concepts you have mastered here extend into multiple regression, where models include several explanatory variables simultaneously, and into non-parametric approaches that relax the LINE conditions.

Feature	AP Statistics (Simple Linear Regression)	Advanced (Multiple Regression & Beyond)
Number of predictors	One explanatory variable x	Multiple predictors x₁, x₂, …, xₖ
Degrees of freedom	df = n − 2	df = n − k − 1
Interval interpretation	Marginal relationship between x and y	Partial effect of xⱼ holding other predictors constant
Conditions	LINE (parametric assumptions)	Extended diagnostics; bootstrap CIs if conditions fail
Software output	Read b and SE_b from a regression table	Software provides full coefficient tables with CIs built in

From AP Statistics to advanced regression inference

In multiple regression, you will encounter confidence intervals for each slope coefficient β_j that work in the same conceptual way: compute the interval using b_j ± t* × SE_bⱼ, and check whether zero is included. The conceptual logic—does the interval contain the hypothesized value?—remains identical regardless of the number of predictors, the estimation method, or the statistical software used. Mastering this reasoning now equips you for virtually every parametric inference problem you will encounter in future statistics courses.

SECTION 9

Practice Problems

Question 1: CONCEPTUAL

A 90% confidence interval for the slope of a regression line relating hours studied to exam score is (1.2, 4.8). Which of the following is a correct interpretation of this interval?

A. There is a 90% probability that the true slope is between 1.2 and 4.8.
B. We are 90% confident that the true slope of the population regression line is between 1.2 and 4.8.
C. 90% of all sample slopes will fall between 1.2 and 4.8.
D. We are 90% confident that the sample slope is between 1.2 and 4.8.
E. If we repeated the study, exactly 90% of the slopes would be positive.

Show answer (B)

B is correct. A confidence interval estimates a population parameter (β), not a probability statement about where β lies after the interval is computed (which eliminates A). Choice C confuses the interval with the sampling distribution of the slope; the interval estimates the parameter, not the distribution of statistics. Choice D is wrong because the sample slope b is a known quantity (3.0 in this case)—we do not need to estimate it. Choice E incorrectly characterizes what the confidence level means. The correct interpretation is: we are 90% confident that the method used to construct this interval has captured the true population slope, meaning β is between 1.2 and 4.8.

Question 2: BASIC CALCULATION

A regression analysis with n = 27 observations yields b = 3.4 and SE_b = 1.2. The critical value t* for a 95% confidence interval with df = 25 is 2.060. What is the 95% confidence interval for β?

A. (0.928, 5.872)
B. (1.340, 5.460)
C. (1.928, 4.872)
D. (0.472, 6.328)
E. (2.200, 4.600)

Show answer (A)

A is correct. The margin of error is t* × SE_b = 2.060 × 1.2 = 2.472. So the interval is 3.4 ± 2.472 = (0.928, 5.872). Choice B uses t* = 1.708 (the value for a 90% interval, df = 25). Choice C incorrectly uses ME = 1.2 × 1.228. Choice D uses t* = 2.060 but then multiplies by 1.2 twice or uses the wrong df. Choice E incorrectly uses a z-value of 1.0. Since the interval does not contain 0, there is convincing evidence of a positive linear association.

Question 3: INTERMEDIATE

A researcher claims that for every additional year of experience, a worker's annual salary increases by more than $2,000. A 95% confidence interval for the slope (in thousands of dollars per year of experience) is (1.4, 3.8). Does this confidence interval support the researcher's claim?

A. Yes, because the entire interval is positive.
B. Yes, because the sample slope of 2.6 exceeds 2.
C. No, because the interval contains values less than 2, so it is plausible that the true slope is less than $2,000 per year.
D. No, because the interval does not contain 0.
E. Yes, because 2.0 is inside the interval and 2.0 represents the claim.

Show answer (C)

C is correct. The researcher claims β > 2. To support this claim, the entire interval would need to be above 2 (i.e., the lower bound exceeds 2). Since the lower bound is 1.4, the interval includes values between 1.4 and 2.0 where the slope would be less than $2,000 per year—so the claim is not fully supported. Choice A addresses whether the slope is positive (β > 0) but not whether it exceeds 2. Choice B relies on the point estimate alone without accounting for uncertainty. Choice D is irrelevant to the specific claim about β > 2. Choice E reverses the logic: the fact that 2.0 is inside the interval means we cannot rule out β = 2, which actually undermines the "more than 2" claim.

Question 4: APPLIED

4 pts

A marine biologist studies the relationship between ocean water temperature (°C) and coral growth rate (mm/year) by randomly selecting 30 reef sites. The computer output from a least-squares regression analysis is shown below: Predictor Coef SE Coef T P Constant −2.15 1.08 −1.99 0.056 Temp 0.42 0.14 3.00 0.006 s = 1.72 R-Sq = 24.3% R-Sq(adj) = 21.6% (a) Construct a 95% confidence interval for the slope of the population regression line. Use df = 28 and t* = 2.048. (b) Based on your interval, is there convincing evidence that water temperature is linearly associated with coral growth rate? Justify your answer. (c) Interpret the confidence interval in context.

Show answer

Scoring rubric (4 points): Point 1 — Correct computation of the confidence interval. CI = b ± t* × SE_b = 0.42 ± 2.048 × 0.14 = 0.42 ± 0.2867 = (0.1333, 0.7067). Award the point for correct substitution and arithmetic. Minor rounding differences are acceptable. Point 2 — Correct identification of whether 0 is in the interval. The student must state that 0 is NOT contained in the interval (0.1333, 0.7067). Point 3 — Correct conclusion linked to the interval. Because 0 is not in the 95% CI, there is convincing evidence at the 95% confidence level that the true slope is not zero—meaning there is a positive linear association between water temperature and coral growth rate. The student must connect the conclusion to the interval (not just the p-value). Point 4 — Correct interpretation in context. We are 95% confident that for each 1°C increase in water temperature, the average coral growth rate increases by between approximately 0.133 mm/year and 0.707 mm/year. The interpretation must reference the population parameter β (not the sample slope), include the confidence level, specify both the explanatory and response variables in context, and include units.

Question 5: CRITICAL THINKING

4 pts

A statistics student constructs a 99% confidence interval for the slope of a regression model and obtains (−0.8, 3.2). A classmate constructs a 95% confidence interval from the same data and obtains (−0.2, 2.6). (a) Explain why the 99% interval is wider than the 95% interval. (b) Based on the 95% interval, can you conclude that there is convincing evidence of a linear association? Explain. (c) Based on the 99% interval, can you conclude that there is convincing evidence of a linear association? Explain. (d) A third student claims: "Since one interval says there is evidence and the other says there is not, the two intervals contradict each other." Is this student correct? Explain the relationship between confidence level and the strength of evidence required to reject β = 0.

Show answer

Scoring rubric (4 points): Point 1 — Part (a): Explanation of interval width. A 99% confidence interval uses a larger critical value t* than a 95% interval (because we need more confidence that we have captured the true parameter). A larger t* produces a larger margin of error (ME = t* × SE_b), so the interval is wider. Both intervals are centered on the same sample slope, so the wider 99% interval extends farther in both directions. Point 2 — Part (b): 95% CI conclusion. The 95% CI is (−0.2, 2.6). Because this interval contains 0, we do NOT have convincing evidence at the 95% confidence level that a linear association exists between the two variables. Point 3 — Part (c): 99% CI conclusion. The 99% CI is (−0.8, 3.2). Because this interval also contains 0, we do NOT have convincing evidence at the 99% confidence level either. In fact, since the 95% interval already fails to exclude 0, the wider 99% interval will necessarily also contain 0. Point 4 — Part (d): Reconciliation of results. The third student is incorrect. The two intervals do not contradict each other—they are consistent. Both intervals contain 0, so neither provides convincing evidence of a linear association. More broadly, if a 95% CI contains 0, the 99% CI from the same data will always also contain 0 (since the 99% CI is wider and centered at the same point). The two intervals would only yield different conclusions if the 95% CI excluded 0 but the wider 99% CI included it—in which case you would have evidence at the 5% significance level but not at the 1% level. This reflects the principle that a higher confidence level demands stronger evidence (equivalent to a lower α) to exclude a particular value.

SUMMARY

Summary

A confidence interval for the slope of a regression model takes the form b ± t* × SE_b, where b is the sample slope, t* is the critical value from the t-distribution with df = n − 2, and SE_b is the standard error of the slope. Before constructing the interval, verify the LINE conditions: Linearity, Independence, Normality of residuals, and Equal variance.

To justify a claim about the slope, state the interval and check whether the claimed value of β (typically 0) lies inside or outside it. If zero is not contained in the interval, you have convincing evidence at the given confidence level that a linear association exists in the population. If zero is contained in the interval, you lack convincing evidence of such an association. Always interpret the interval in context—referencing the variables, units, and the population parameter β—to earn full credit on the AP exam.