MaizePrepSTATS 250
Exam 3 · Lec 17

Inference for a regression slope

Move from a fitted sample trend to a population slope and its uncertainty.

25 minDraft

Key ideas

Population slope

The change in the population mean response associated with a one-unit increase in the predictor in the specified linear model.

Slope test

The coefficient row tests a population slope, usually against zero. A two-sided test asks about any nonzero linear association.

Slope interval

Estimate ± t multiplier × slope standard error gives plausible population slopes, in response units per predictor unit.

Regression conditions

Check linear mean structure, independent errors, approximately normal errors for small-sample inference, and constant error variance. Residual plots help assess shape and spread; study design addresses independence.

Formulas

Test and interval
t=β^1−β1,0SE(β^1),β^1±t∗SE(β^1)t=\frac{\hat\beta_1-\beta_{1,0}}{SE(\hat\beta_1)},\qquad \hat\beta_1\pm t^*SE(\hat\beta_1)
Simple regression uses n−2 residual degrees of freedom.
From simple to multiple
y^=β^0+β^1x1+β^2x2\hat y=\hat\beta_0+\hat\beta_1x_1+\hat\beta_2x_2
A multiple-regression slope holds other included predictors constant.

Which procedure?

Question asks whether a quantitative predictor has a linear association
Use the slope t-test.
Question asks how large the association could plausibly be
Use a confidence interval for the slope.
Question asks to account for another predictor
Fit multiple regression and interpret adjusted coefficients.
Worked example

Machine age and repair time

A random sample of 22 machines is used to model repair minutes by age in years. Give the slope test and an approximate 95% interval using t* = 2.086. This is an original example with composed R output.

summary(lm(minutes ~ age, data = machines))$coefficients
             Estimate Std. Error t value Pr(>|t|)
(Intercept)     18.00       4.00     4.5 0.000219
age              1.20       0.40     3.0 0.007076
# Residual df: 20
  1. Define beta_1 as the population mean repair-time change per additional year of machine age.

  2. For H0: beta_1 = 0 versus a nonzero slope, t = 1.20/0.40 = 3.

  3. The interval is 1.20 ± 2.086(0.40), approximately (0.37, 2.03).

  4. The test is two-sided, so the conclusion is two-sided: p = 0.007076 provides very strong evidence that the population slope is not 0 (a linear association between age and repair time). The sample slope is positive. Age was observed, not assigned.

Answer

We are 95% confident that the population mean repair time increases by between about 0.37 and 2.03 minutes per additional year of age under this model. This is an association, not a causal effect.

Three-question check

1
Lec 17

A slope estimate is −6 with SE 2. For a zero-slope null, t equals:

2
Lec 17

A 95% slope CI is (−1.8,−0.4). It estimates:

3
Lec 17

Residuals show a clear U-shaped pattern against x. What is the concern?

MaizePrep is an independent study resource. It is not affiliated with, sponsored by, or endorsed by the University of Michigan. Course names and numbers are used only to describe which class the material covers. Every practice question is original, written in the style of the course.