← All teaching guides
For the tutor · not Josh-facing

📈 Regression + Residuals

Unit 3 Topic 1 (bivariate data, second half) · week 6 · formula book writes the line as y = mx + c with m = r·sᵣ/sₓ and c = ȳ − m·x̄
The concept in one sentence: the least-squares line is the best straight-line summary of the cloud, the gradient is the story it tells ("each extra x buys m more y"), and residuals measure how wrong the line is, one point at a time.

The thing that makes it click: "real minus robot." The line is a robot making predictions; the residual is how much the real world beat or undershot the robot: actual − predicted.

1 · Prerequisite check (30 seconds)

"In y = 3x + 2, what is the gradient? What is y when x = 4?" And from last week: explanatory on which axis?

If substitution wobbles: foundations.html's substituting-into-a-formula recipe. If gradient-reading wobbles, spend five minutes on "m is the per-one step" with a simple line before least-squares appears.

2 · The teaching path

  1. Concrete: a plot from last week's chant work with a line ruled through it by eye. "The line is a robot that predicts. Feed it x = 8, it predicts y." Predictions before formulas
  2. Interpret before compute: given ŷ = 2.5x + 10 (study hours vs score), ask what the 2.5 MEANS and what the 10 MEANS in words. The interpretation sentences are the actual exam skill:
Each extra hour of study is associated with 2.5 more marks. Zero hours predicts 10 marks.
  1. Note "associated with", not "causes", the causation discipline from last week carries straight into the interpretation sentence
  2. Then the formula-book build: given r, sₓ, sᵣ, x̄, ȳ, construct the line. One worked run: r = 0.8, sₓ = 2, sᵣ = 6, x̄ = 10, ȳ = 50 → m = 2.4, c = 26 → ŷ = 2.4x + 26
  3. Residuals last: real minus robot. Predicted 38, scored 42, residual +4, the dot floats ABOVE the line. Then the residual PLOT: random scatter = the line was a good idea, a curve = it wasn't
m = r · sᵣ ÷ sₓ  ·  c = ȳ − m·x̄  ·  residual = actual − predicted

3 · Misconception catalogue

Looks likeWhy brains do itThe fix script
Residual computed as predicted − actualThe line feels like the "truth" to measure from"Real minus robot, reality goes first." Then check the sign against the picture: above the line must come out positive
Interprets the intercept when it's meaningless (0 kg newborns, 0 cm humans)Every number must mean something, right?"Interpret the intercept, THEN rate it: sensible or silly? Saying 'not meaningful here because x = 0 is outside the data' earns the mark"
Predicts far outside the data happilyThe formula works for any x, so why not"The robot only trained on 1 to 12 hours. Asking it about 40 is asking a learner driver to fly a plane." Name it: extrapolation, flag it every time
Textbook letters (ŷ = a + bx) collide with the book's mx + cDifferent sources, different alphabets"The QCAA book says m and c, same as junior maths. If a source says a + bx, the letter NEXT TO x is always the gradient"
Uses m = r·sₓ/sᵣ (flipped spreads)Which s goes on top doesn't feel like it matters"y over x, response over explainer, same order as the line reads." Then sanity-check: steep data should give a big m
Reads a curved residual plot as "good, they're all close"Small = good instinct"The residual plot isn't about SIZE, it's about PATTERN. Random confetti = good. Any shape = the straight line was the wrong tool"

4 · Questioning ladder

RungAskListen for
Recall"Residual = ?""Actual minus predicted", or "real minus robot"
Do"ŷ = 2.5x + 10. Predict for 8 hours"30, by substitution, no drama
Explain back"Interpret the 2.5 like the exam wants"The sentence, in context, with "associated with" not "causes"
Transfer"r = 0.8, sₓ = 2, sᵣ = 6, x̄ = 10, ȳ = 50. Build the line"m = 2.4, c = 26, writes ŷ = 2.4x + 26, formula book open and used

5 · How QCAA examines it

6 · Stuck scripts

Real minus robot.
Is that x inside the data the robot trained on?
Say the gradient sentence. Start with "each extra..."
y over x, response over explainer.

7 · ADHD delivery notes