BUSS1020 Chap.10 Correlation and Simple Linear Regression
Correlation and Simple Linear Regression
Simple linear regression describes how a quantitative response changes with one quantitative predictor. This chapter begins with scatterplots, covariance and correlation, then interprets the fitted intercept and slope in units. Residuals connect observed values with predictions, while total, regression and error sums of squares build the coefficient of determination and standard error of regression.
Every interpretation separates association from causation and interpolation from unsupported extrapolation. The warehouse example turns an equation into a prediction, a slope sentence and an observed-minus-predicted residual. Variation decomposition is checked through SST equals SSR plus SSE before R-squared is interpreted.
Residual plots challenge linearity and constant spread, while regression to the mean warns against crediting an intervention merely because an extreme period was followed by a less extreme one. Prediction error remains in the response unit and is compared with a decision tolerance, complementing the unit-free R-squared summary. Omitted variables and range limits appear in every final conclusion.
A fitted equation earns trust through residual behaviour and design evidence, not through a coefficient printed by software alone.
What this chapter covers
- 01
Scatterplots and linear pattern
- 02
Covariance and correlation
- 03
Least-squares fitted equation
- 04
Slope and intercept interpretation
- 05
Residuals and diagnostic plots
- 06
SST, SSR, SSE and coefficient of determination
Worked example · free
Predict warehouse labour hours
- 1Substitute 350 to obtain predicted labour of 40 hours.
- 1Interpret the slope as 0.08 additional predicted labour hours per extra order on average.
- 1Treat the intercept cautiously because zero orders lies outside the observed range.
- 1Compute residual as 43 minus 40, giving positive 3 hours.
- 1Interpret that the observed day used three more hours than the model predicted.
Key terms
- Covariance
- A unit-dependent measure of whether two quantitative variables tend to move together.
- Correlation
- A unit-free measure from negative one to one describing linear direction and strength.
- Fitted value
- The response predicted by the estimated regression equation at a stated predictor value.
- Residual
- Observed response minus fitted response for one observation.
- Coefficient of determination
- The proportion of sample response variation accounted for by the fitted linear model.
- Extrapolation
- Using a fitted relationship beyond the predictor range represented by the data.
Correlation and Simple Linear Regression FAQ
Does strong correlation prove causation?
No. Correlation can reflect omitted variables, reverse direction, selection or shared trends. A causal conclusion needs a design and assumptions that isolate the effect of changing the predictor. Regression from observational data usually supports association and prediction.
How should I interpret the slope?
State the average predicted change in the response for a one-unit increase in the predictor, keeping both units visible and limiting the statement to the relevant data range. Do not describe the slope as a percentage unless the variables support that scale.
What does R-squared tell me?
It is the proportion of observed response variation accounted for by the fitted model in the sample. It is not the percentage of cases predicted correctly, the probability the model is true or evidence that the predictor causes the response.
Exam move
Begin every regression task with a scatterplot description: direction, form, strength and unusual observations. Practise slope and intercept interpretations using both variable units. Compute residuals as observed minus predicted and verify their signs against the graph. Rebuild SST equals SSR plus SSE and calculate R-squared in both equivalent ways. Inspect residual plots for curvature, changing spread and outliers.
For every prediction, mark whether it is interpolation or extrapolation. End by naming a plausible omitted variable and using association language unless the study design supports causation. Use a fresh equation to write three separate sentences for intercept, slope and one prediction. Reject any slope sentence that omits the one-unit increase in X, average predicted change in Y or either measurement unit.
Plot several observed points and draw vertical residuals so observed minus fitted becomes a geometric rule rather than a memorised sign. Calculate SST, SSR and SSE from a small dataset and verify the identity before computing R-squared. Then explain why a large R-squared does not certify causation, assumption validity or accurate extrapolation.
In a residual plot, practise naming curvature, funnel shape and an isolated point, plus the modelling concern each pattern raises. Mark the observed X range beside every prediction and label values outside it as extrapolation. For regression to the mean, imagine selecting units after an extreme period and explain why later moderation can occur without an intervention effect.
Finish with a decision-focused model review that includes prediction error in Y units, not only correlation or R-squared.
Working through Correlation and Simple Linear Regression in BUSS1020? Sia is AskSia’s AI Statistics tutor — ask any BUSS1020 Correlation and Simple Linear Regression question and get a clear, step-by-step explanation grounded in how BUSS1020 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.