Skip to main content

Exam 3 Study Guide

Modules 10–12 · Simple regression, multiple regression, and model assumptions

This guide tells you what Exam 3 asks you to do, and where on the site each skill is taught. It doesn’t contain new examples. Every item below links to a Module section, a pre-study activity, or a lab step you’ve already worked through, so you can go straight to the material that matters and check that you can do it.

How to use it. Cover the Where links, read each Ready if you can statement, and try to explain or demonstrate the skill without looking anything up. If you get stuck, the links tell you exactly where to practice.

What Exam 3 looks like

Part A · written Part B · coding
When Monday, in class Wednesday, in class
Time Up to 90 minutes; most students need about 45 to 60 Up to 75 minutes; most students need about 45
Points 48 22
You may use One hand-written, letter-size (8½ × 11 inch) sheet of your own notes (both sides), a standard (basic) calculator, and a pencil. No other notes, and no phones, laptops, or other devices. The course website and your own hand-written notes. No other websites, and no AI assistance.
What you do Read printed R output from regression models, work short calculations by hand, and interpret coefficients, model comparisons, and diagnostics. There is no code to write, and every critical value you need is printed in the question. Run a short analysis from start to finish on one dataset, on an exam webR page like the pre-study pages: plot and fit a model, add a predictor, then check the model. Record your results on a paper answer sheet.
How you answer Write a number in a box, mark one option, or mark one box per row of a grid. One question asks for a short written explanation. Write numbers (and one name) in boxes, mark options and grid rows, and write one or two sentences interpreting a coefficient.

Part A’s points are spread across three sections: one predictor at a time (M10), several predictors at once (M11), and checking the model (M12). The three carry roughly equal weight, with M10 a little more.

What’s in scope. Everything on Exam 3 is taught in the main content of Modules 10–12, their pre-study pages, and their labs. The indigo Going further boxes aren’t tested; as How to Read This Site says, nothing in them is required. Part B also uses the M03 and M04 skills you’ve used all semester: a ggplot() scatterplot, filter(), and mutate().


Part A · what to be ready for

Because Part A is closed to the website, try each Ready if you can check with the website closed. If you can’t do it yet, follow the links.

One predictor at a time (M10)

1. What the slope and the intercept mean. Where: The intercept and the slope · Reading the coefficients · Centering · Pre-study Activities 1.2 and 1.3

  • Ready if you can: say what a slope means as a comparison between two people whose predictor values differ by one unit, and say what the intercept means, including when it describes a predictor value outside the data.

2. Predictions and residuals, by hand. Where: Predicted values and residuals, by hand · Pre-study Activity 2.1 · Lab Step 2

  • Ready if you can: compute a predicted value from printed coefficients, then the residual (observed minus predicted), with the right sign. Calculate with the coefficients as they are printed, and round only your final answer.

3. What \(R^2\) and \(\sigma\) tell you. Where: Model fit: \(R^2\) and \(\sigma\) · Pre-study Activity 2.1

  • Ready if you can: say what \(R^2\) means as a share of the outcome’s variation and what \(\sigma\) means in the outcome’s units, and sort true and false statements about them.

4. Inference for the slope, by hand. Where: Where the standard errors and CIs come from · CI on the slope · Hypothesis test on the slope · Pre-study Activity 3.2

  • Ready if you can: from the estimate, its standard error, and a printed \(t^*\), compute \(t_\text{obs}\) and both ends of the 95% CI, and say what an interval that excludes 0 does, and doesn’t, support.

5. Confidence interval versus prediction interval, for a prediction. Where: CI for the line vs PI for individuals · M11 pre-study Activity 2.4 · M10 Lab Step 4

  • Ready if you can: tell the two intervals apart in printed output, and choose the one that answers a question about the average of many people versus one new person.

6. Correlation and \(R^2\). Where: Correlation and standardized regression · Pre-study Activity 2.2

  • Ready if you can: in a simple regression, find \(r\) from \(R^2\) and the sign of the slope.

7. A binary predictor. Where: How a regression handles a categorical predictor · Recovering both group means · Connection to the two-sample t-test · Pre-study Activities 4.1 and 4.2

  • Ready if you can: read both group means off the intercept and the slope, and name the t-test that gives the same \(t_\text{obs}\) and p-value as the slope’s row.

Several predictors at once (M11)

8. Reading a multiple regression. Where: Centering for an interpretable intercept · Reading each coefficient as a partial effect · The two adjusted means · Pre-study Activity 2.1 · Lab Step 6

  • Ready if you can: say what each coefficient means while holding the other predictors constant, say what the intercept means when a predictor is centered, and compute a prediction for a particular kind of person by adding the coefficients that apply.

9. The precision gain. Where: The precision gain · Pre-study Activity 2.3

  • Ready if you can: compare a model with and without a strong baseline covariate and say what happened to the treatment estimate, its standard error, \(\sigma\), and \(R^2\); and explain why, in a randomized study, adding a strong baseline covariate barely changes the treatment estimate but makes it more precise, and what that means for the study’s conclusion.

10. Adding a multi-category predictor. Where: Fit the model with site · Why the gap is the same at every site · The nested-model F-test · Δ\(R^2\) · Pre-study Activities 3.1, 3.2, and 3.3

  • Ready if you can: compute a prediction for one group from the coefficients, explain why the model gives the same treatment gap in every group, read what a nested-model F-test shows, and compute Δ\(R^2\).

Checking the model (M12)

11. Which check goes with which assumption. Where: Two assumptions that come first · Five statistical assumptions · The remediation toolkit · Pre-study Video 1

  • Ready if you can: match the residual–fitted plot, the Q–Q plot, the scale–location plot, Cook’s distance, and the VIF to what each one checks, and name the two assumptions you judge from how a study was designed rather than from a plot.

12. Reading the diagnostic plots. Where: The residual–fitted plot · The Q–Q plot · The scale–location plot · Pre-study Activities 1.1, 2.1, and 2.2

  • Ready if you can: say whether a residual–fitted smoother shows a curve the straight line misses, whether a scale–location smoother shows the spread changing, and whether the points of a Q–Q plot follow its line.

13. Influence and leverage. Where: Influence and leverage · Identifying influential observations · What to do about them · Pre-study: every column augment() adds · Pre-study Activities 3.1 and 3.2 · M12 Lab Step 9

  • Ready if you can: compare Cook’s distances with the classic threshold of 1 and with the more sensitive 4/n screen (and compute 4/n), say what leverage measures, and say what to do about flagged observations.

14. Multicollinearity. Where: The variance inflation factor · Pre-study Activity 3.3 · M12 Lab: the VIF

  • Ready if you can: read a set of VIFs, say what values near 1 mean, and say why random assignment keeps a treatment’s VIF near 1.

Part B · what to be ready for

Part B is a short analysis of one dataset, from start to finish, and each task builds on the one before. The website is open, so the skill being tested is knowing which recipe to use and where it lives, then running it carefully. For each task type below, find the linked code now and make sure you can run it from a blank code box.

1. Plot first, then fit and read a simple regression. Where: M10 pre-study Activities 1.1, 1.2, and 2.1 · M10 · looking at the data · M10 · lm() · M10 Lab Step 2

  • Ready if you can: make a scatterplot with a straight-line fit and describe what it shows (its direction and form, and how the points spread around the line); fit the model with lm(); read the intercept, the slope, \(R^2\), and \(\sigma\) with tidy() and glance(); and use augment() with data =, so the data’s own columns (such as a name) come along, then filter() to one row and read its fitted value and residual.

2. Add a categorical predictor. Where: M10 · letting lm() build the dummies · M10 pre-study Activity 4.3 · M11 pre-study Activities 3.1 and 3.2 · M11 · fit the model with site · M11 · the nested-model F-test · M11 · reading each coefficient · M11 · prediction vs causation

  • Ready if you can: set a factor’s reference level with fct_relevel() and save the result back into the data; fit a model with a numeric predictor and a categorical one; read each coefficient and its 95% CI from tidy(); compare the two models with anova() and read F and both degrees of freedom; and write one or two sentences saying what a coefficient means, naming the two groups it compares and that the comparison is at the same value of the other predictor, with its size and units. Use causal language only when the predictor was randomly assigned.

3. Check the model, and decide what it means. Where: M12 pre-study Activities 1.1, 2.1, 2.2, and 3.2 · M12 Lab Steps 8 and 9 · What to do when equal variance fails · When violations matter · The remediation toolkit

  • Ready if you can: build the residual–fitted plot (with a LOESS smoother), the Q–Q plot, and the scale–location plot from augment(); sort the rows by Cook’s distance and name the most influential one (keep an identifying column beside the diagnostics, as the M12 lab’s Step 9 does) and compare its Cook’s distance with 1; say what each plot shows, reading a smoother where most of the data sit and expecting it to wobble at the sparse ends; and match what you see to the Module’s remedies, knowing what each one fixes and which goal it serves.

Part B habits that save points

  • Do the tasks in order. Later tasks use the models earlier ones create.
  • Check before you record. Glance at each result and ask whether it makes sense: is the sign what you expect, and is the reference level the one the task asked for?
  • Record numbers the way the sheet asks, usually two decimals.
  • If a task won’t run, record any results you already have, leave the remaining boxes blank, and move on. Return to it if time permits, and check the exam page for help to keep going when a later task needs something you couldn’t create.

Making your one-page sheet for Part A

You may bring one hand-written, letter-size (8½ × 11 inch) sheet of your own notes, both sides. Building it is good studying in itself: deciding what goes on it makes you sort out what you understand from what you’d need to look up. Ideas worth including:

  • A prediction from a fitted line, and a residual (observed minus predicted).
  • \(t_\text{obs}\) for a coefficient, and its 95% CI (estimate ± \(t^*\) × SE).
  • In a simple regression, how \(r\) relates to \(R^2\) and the slope’s sign.
  • What the intercept and the slope are when the predictor is binary.
  • How to build a prediction from a multiple regression by adding the coefficients that apply, and what a centered predictor contributes at its mean.
  • Δ\(R^2\), and what a nested-model F-test asks.
  • Which diagnostic checks which assumption, the two assumptions judged from the study’s design, and the cutoffs for Cook’s distance (1 and 4/n) and the VIF.
  • What leverage measures, and how it differs from Cook’s distance.
  • Why adding a strong baseline covariate to a randomized study changes the treatment estimate’s precision and not its size.
  • One-line interpretations you trust: a slope, a coefficient in a multiple regression, \(R^2\), \(\sigma\), and a CI versus a PI.

You don’t need a table of critical values; every one you need is printed in the question.


The week before the exam

  • A few days out: work through the Part A checks above with the website closed, and mark any you can’t do yet. Follow their links and try again later the same day.
  • Redo pre-study activities from the ✍️ tab only. Don’t open 💡 Hint or 👀 Spoiler until you’ve written your own answer.
  • For Part B, open a pre-study page and rerun each recipe above from a blank code box, with your lab notebook beside you. Then do it once more with only the website and your hand-written notes, which is all you’ll have on exam day. Practice finding the right section of the site quickly.
  • The day before: finish your notes sheet, and reread the Summary of M10, M11, and M12.
  • Questions? Ask before the exam. A question asked is always better than a guess.

Bring: on Monday, a pencil, a standard (basic) calculator, and your notes sheet; on Wednesday, your laptop (charged) and a pencil for the answer sheet. The syllabus has the dates.