Topic 9.1 Introducing Statistics: Do Those Points Align?: explain why a sample regression slope varies from sample to sample, motivating inference about the true population slope of a linear model.

What this topic is asking

The College Board (Topic 9.1) opens Unit 9 with the idea behind slope inference: the slope $b$ of a least-squares line fitted to a sample is a statistic that varies from sample to sample, so it estimates, but rarely equals, the true population slope $\beta$. This motivates confidence intervals and tests about $\beta$.

The sample slope is a statistic

Units 2 fit least-squares lines descriptively; Unit 9 treats the slope as a quantity with sampling variability. Just as a sample mean $\bar{x}$ estimates $\mu$ and varies across samples, the sample slope $b$ estimates $\beta$ and varies across samples. Take a new random sample, refit the line, and you get a slightly different slope. The collection of all possible sample slopes forms a sampling distribution centered (under the model's conditions) at the true slope $\beta$.

Why a non-zero slope is not proof

This is the central caution of the unit, and the reason inference is needed. The question "do those points align?" is really "is the observed slope larger than sampling variability alone would typically produce if $\beta = 0$?" A small, easily-explained-by-chance slope is consistent with no relationship; a slope too large to be chance is evidence of a real linear association. Distinguishing the two requires the sampling distribution of $b$, not just its observed value.

The parameter and the tools, previewed

The parameter of interest is the population slope $\beta$. Unit 9 builds the two familiar tools on it, mirroring Units 6 and 7.

1. Confidence interval for $\beta$ (Topics 9.2 to 9.3): estimate the true slope with $b \pm t^{*}(\text{standard error of } b)$, and judge claims (including whether $0$ is plausible).
2. Significance test for $\beta$ (Topics 9.4 to 9.5): test $H_0: \beta = 0$ (no linear relationship) with a t-statistic and P-value.

Both use a $t$-distribution with $n - 2$ degrees of freedom (two are spent estimating the intercept and slope), and both rely on regression conditions about the residuals. Topic 9.1 plants the idea that $b$ is a variable estimate of a fixed $\beta$; the later topics supply the machinery.

The mindset for the unit

As in every inference unit, the key move is to see the observed slope as one draw from a distribution, not the truth. The fitted line summarizes one sample; the population line is fixed but unknown. Inference about $\beta$ is the disciplined way to ask whether an observed trend is real or could be a fluke of sampling, the precise meaning of "do those points align?"

Try this

Q1. What parameter does the sample slope $b$ estimate, and why does $b$ vary? [2 points]

• Cue. $b$ estimates the population slope $\beta$; it varies because each random sample yields a different fitted line (sampling variability).

Q2. Why is a non-zero sample slope not proof of a relationship? [1 point]

• Cue. Even if $\beta = 0$, chance variation across samples can produce a non-zero $b$; inference is needed to rule that out.