AP Statistics Unit 9 ReviewSlopes

AP Statistics Unit 9, Inference for Quantitative Data: Slopes, is the final inference unit of the course. It asks one question over and over. Your sample data shows a linear pattern, but is that pattern real in the population, or could a slope like yours happen just by random sampling variability? You answer it with two tools, a t-interval to estimate the true slope β and a t-test to decide whether β differs from zero. The unit makes up 2-5% of the AP exam.

What this unit covers

From sample slope to population slope

• The regression line you computed back in Unit 2, ŷ = a + bx, is built from a sample. The sample slope b is just an estimate of the true population slope β in the model μy = α + βx.
• Different random samples give different scatterplots and different slopes. Some of that variation is random noise around the true line. The whole unit is about deciding when the pattern in your sample is too strong to be noise.
• Residuals (y - ŷ) measure how far each point falls from the line. They estimate how far points fall from the population line, and they do double duty here as your tool for checking conditions.
• The key new number is the standard error of the slope, SEb. It tells you how much b typically varies from sample to sample. On the exam, you almost always read SEb straight from computer output rather than computing it by hand.

Confidence intervals for the slope

• The interval is point estimate plus or minus margin of error, so b ± t*(SEb), with the critical value t* coming from a t-distribution with n - 2 degrees of freedom. Note the n - 2, not n - 1. You lose two degrees of freedom because the line estimates two parameters, slope and intercept.
• Interpretation works like every other interval. "We are 95% confident that the interval from ___ to ___ captures the true slope of the population regression line relating x to y," stated in context with the variables named.
• The confidence level describes the method. In repeated random sampling with the same sample size, about C% of intervals built this way capture the true slope.
• A bigger sample size shrinks SEb, which shrinks the interval. Same logic as Units 6 and 7.
• The most common claim you'll justify with an interval is whether 0 is inside it. If the entire interval is above 0 (or entirely below 0), you have convincing evidence of a linear relationship. If the interval contains 0, you don't.

Significance tests for the slope

• The test is a t-test for slope. Hypotheses are written about β, never about b. Usually H₀: β = 0 (no linear relationship) against Hₐ: β ≠ 0, β > 0, or β < 0. The null can use a value other than 0 if a question hypothesizes a specific slope β₀.
• The test statistic is t = (b - β₀)/SEb with n - 2 degrees of freedom. With the usual β₀ = 0, that simplifies to t = b/SEb, which you can compute directly from a computer output table.
• The p-value is the probability of getting a sample slope at least as extreme as yours, assuming the true slope really is β₀. Small p-value means your data would be surprising if H₀ were true.
• Decision rule, same as always. If p ≤ α, reject H₀ and conclude there is convincing evidence of a linear relationship (or of whatever the alternative claims). If p > α, fail to reject. Never "accept" the null.

Conditions for inference on a slope

• Linearity. The true relationship between x and y must be linear. Check the residual plot for no curved pattern.
• Equal standard deviation. The spread of y around the line should be roughly the same at every x value. Check the residual plot for no fanning or funneling.
• Independence. Data should come from a random sample or randomized experiment, and when sampling without replacement, the sample should be less than 10% of the population.
• Normality. For each x, the y values should be roughly normal. With large samples this is less of a worry; with small samples, check that residuals show no strong skew or outliers.
• Many teachers remember these as LINE or LINER (Linear, Independent, Normal, Equal SD, Random). Whatever you call them, the residual plot is your main checking tool.

Choosing the right procedure

• Topic 9.6 zooms out to the whole inference toolkit. Given a scenario, you decide which test or interval fits. Categorical data with one or two proportions points to z-procedures (Unit 6). Counts in categories point to chi-square (Unit 8). Quantitative means point to t-procedures for means (Unit 7). A relationship between two quantitative variables points here, to slope inference.
• Useful sorting questions to ask. Is the variable categorical or quantitative? One sample, two samples, or paired? Am I estimating something (interval) or testing a claim (test)?
• This selection skill is heavily tested because the exam never labels a problem "this is a slope test." You have to recognize it from the setup.

Unit 9, Slopes at a glance

Why Unit 9, Slopes matters in AP Stats

Unit 9 is where the two halves of the course finally meet. The first half taught you to describe relationships with regression; the second half taught you to make inferences with intervals and tests. This unit fuses them, so you can move from "these variables look related in my sample" to "I have statistical evidence they're related in the population."

• It completes the inference family. After proportions, means, and chi-square, slope inference is the last procedure you learn, and it follows the exact same logic (statistic ± critical value × SE, and t = (statistic - parameter)/SE).
• It turns regression from description into decision-making. A slope of 0.8 in your sample means nothing until you know whether 0 is a plausible value for the true slope.
• It is the capstone for procedure selection, the skill of looking at any scenario and choosing the right tool, which is exactly what the exam demands.

How this unit connects across the course

• Everything about ŷ = a + bx, residuals, residual plots, and interpreting slope in context comes straight from exploring two-variable data (Unit 2). Unit 9 adds the inference layer on top of that foundation.
• The random sampling and randomized experiment requirements in the conditions trace back to data collection design (Unit 3). Bad data collection invalidates slope inference just like any other procedure.
• The idea that b varies from sample to sample and has its own sampling distribution is the same logic you built for sample proportions and sample means (Unit 5).
• The four-step inference structure (hypotheses, conditions, calculations, conclusion) and the t-distribution itself carry over directly from inference for means (Unit 7). A t-test for slope feels like a one-sample t-test with df = n - 2 instead of n - 1.
• Procedure selection in Topic 9.6 requires you to distinguish slope inference from proportion z-procedures (Unit 6) and chi-square tests (Unit 8), so keep those decision rules fresh.

Key formulas and procedures

• Confidence interval for slope: b ± t*(SEb). Estimates the true population slope β; use df = n - 2 to find t*.
• Test statistic: t = (b - β₀)/SEb with df = n - 2. Measures how many standard errors your sample slope sits from the hypothesized slope. With H₀: β = 0, it's just b/SEb.
• Standard error of the slope: SEb = s/(sx√(n-1)), where s estimates the spread of residuals and sx is the standard deviation of the x values. In practice, read SEb from the "SE Coef" column of computer output.
• Hypotheses: H₀: β = β₀ versus Hₐ: β < β₀, β > β₀, or β ≠ β₀. Always define β in context (the true slope relating y to x).
• Conditions check: linear relationship (residual plot has no curve), equal SD of y across x (no fanning in residuals), independence (random sample or randomized experiment, 10% condition when sampling without replacement), approximately normal y values for each x.
• Decision rule: compare p-value to α. If p ≤ α, reject H₀; if p > α, fail to reject H₀. Then write the conclusion in context.
• Reading computer output: the slope row gives you b, SEb, the t statistic, and a p-value. Watch out, software usually reports a two-sided p-value, so halve it for a one-sided test (when the sample slope is in the direction of Hₐ).

Unit 9, Slopes on the AP exam

Slope inference is 2-5% of the exam, the smallest unit weight in the course, but it shows up reliably and in predictable ways. Multiple-choice questions hand you a regression computer output table and ask you to identify the test statistic, build the confidence interval, interpret the slope or the p-value, or check which condition a residual plot does or doesn't satisfy. Free-response questions often fold slope inference into a larger regression problem, where part (a) asks you to interpret the slope or a residual plot and a later part asks you to carry out the full test or interval with all four steps shown. Topic 9.6 is also fair game in disguise. A free-response prompt may simply describe a study and ask you to "name the appropriate inference procedure and check its conditions," and you have to recognize that two quantitative variables and a question about their relationship means a t-test or t-interval for slope. Expect to read output rather than compute SEb by hand, remember df = n - 2, and always write conclusions in context with β defined.

Essential questions

• How can I tell whether a linear pattern in a scatterplot reflects a real relationship in the population or just random sampling variability?
• What does the slope of a sample regression line actually estimate, and how precise is that estimate?
• What has to be true about my data before slope inference gives trustworthy answers?
• Given any statistical scenario, how do I choose the right inference procedure?

Key terms to know

• Population regression line: the true linear model μy = α + βx that describes the average y value at each x in the whole population.
• Sample regression line: the line ŷ = a + bx fit from sample data, used to estimate the population line.
• β (beta): the true slope of the population regression line, the parameter all inference in this unit targets.
• b: the sample slope, your point estimate for β.
• Standard error of the slope (SEb): an estimate of how much b varies from sample to sample.
• Residual: the difference y - ŷ between an observed value and the value the line predicts; residual plots verify the linearity and equal-SD conditions.
• t-distribution with n - 2 degrees of freedom: the null distribution of the slope test statistic; you lose two df because the line estimates both a slope and an intercept.
• Margin of error: t* times SEb, the half-width of the confidence interval for the slope.
• t-test for slope: the significance test that checks whether the true slope equals a hypothesized value, usually 0.
• p-value: the probability, computed assuming H₀ is true, of getting a sample slope at least as extreme as the one observed.
• Equal SD condition: the requirement that the spread of y around the line is roughly the same at every x value.
• 10% condition: when sampling without replacement, the sample should be less than 10% of the population so observations are approximately independent.

Common mix-ups

• b vs. β: hypotheses are always about the parameter β, never the statistic b. Writing H₀: b = 0 loses credit because you already know b from your sample; the question is about the population.
• Degrees of freedom: slope inference uses n - 2, not the n - 1 you used for means in Unit 7. Mixing these up gives the wrong t* and the wrong p-value.
• Interval contains 0 vs. doesn't: an interval entirely above or below 0 is evidence of a linear relationship. An interval containing 0 means you lack convincing evidence, not that you've proven there's no relationship.
• Two-sided output, one-sided test: computer output usually reports a two-sided p-value. For a one-sided alternative, cut it in half (as long as the sample slope points in the direction of Hₐ).