Grade 12 Statistics Exam Tips: Regression & Correlation
The correlation coefficient \(r\) looks like a single easy number to interpret — but most learners lose marks by guessing its strength instead of checking it against the actual scale, or by predicting values the data never supports.
Grade 12 Statistics builds on measures of central tendency, spread, and grouped-data techniques from earlier grades, then adds bivariate data: representing two-variable data as a scatter plot, using a calculator to find the least-squares regression line and correlation coefficient, and interpreting what both actually mean.
The correlation coefficient scale
CAPS uses named strength bands for \(|r|\), and knowing the exact ranges avoids vague guesses:
- Very weak: \(0\) to \(0{,}25\)
- Weak: \(0{,}25\) to \(0{,}5\)
- Moderate: \(0{,}5\) to \(0{,}75\)
- Strong: \(0{,}75\) to \(0{,}9\)
- Very strong: \(0{,}9\) to \(1\)
The sign of \(r\) tells you direction (positive means the variables increase together; negative means one increases as the other decreases), and its size tells you strength — these are two separate pieces of information that both need stating in a full answer.
Mistake 1: Guessing correlation strength instead of using the scale
An \(r\) value of \(0{,}6\) is "moderate," not "strong" — a common error is treating any \(r\) reasonably close to 1 as automatically "strong." Always check the calculated value against the actual named bands rather than describing it from a general impression of the scatter plot.
Mistake 2: Confusing correlation with causation
A strong correlation between two variables does not mean one causes the other — both could be driven by a third factor, or the relationship could be coincidental. CAPS-style questions sometimes explicitly test whether learners understand this distinction, so never write that one variable "causes" the other just because \(r\) is high.
Mistake 3: Extrapolating far beyond the data's actual range
Using the regression equation to predict a value within the range of your original data (interpolation) is generally reliable. Predicting a value outside that range (extrapolation) is far less reliable, since there's no evidence the same linear relationship continues beyond what was actually measured. Always state this limitation explicitly when a question asks you to predict outside the data's range.
Mistake 4: Forcing a straight line onto clearly curved data
Before calculating a regression line, look at the scatter plot's actual shape — if the points clearly curve (suggesting a quadratic or exponential relationship instead), a straight-line regression equation will systematically misrepresent the pattern, even though the calculator will still produce a numerical answer for it.
Mistake 5: Mixing up which variable is independent and which is dependent
The regression equation \(\hat{y}=a+bx\) predicts \(y\) from \(x\) — not the other way around. Substituting a \(y\)-value into the equation to "solve for \(x\)" using ordinary algebra gives a technically different (and usually wrong) answer from the actual regression of \(x\) on \(y\), which is a separate calculation entirely.
Skewness: two ways to check it, not just one
You can determine whether data is symmetric, positively skewed, or negatively skewed by comparing the mean and median (mean greater than median suggests positive skew), or by comparing how far the median sits from \(Q_1\) versus \(Q_3\) in the five-number summary. These two tests usually agree, but on some data sets they can give a weaker or contradictory signal — when that happens, describe the data as "roughly symmetric" rather than forcing a skew label either test only weakly supports.
A revision checklist
- Can you correctly classify an \(r\) value using the named strength bands, not a general impression?
- Do you know why correlation never proves causation?
- Can you tell interpolation from extrapolation, and explain why extrapolation is less reliable?
- Do you check a scatter plot's shape before assuming a linear regression line is appropriate?
- Can you use two independent methods to check whether data is skewed?
Final thought
Statistics rewards careful, honest interpretation more than complex calculation — your calculator does the regression arithmetic for you, but only you can judge whether a linear model actually fits, whether a prediction is trustworthy, and whether a correlation genuinely means what it appears to. Practising the interpretation, not just the button presses, is what turns this into a reliable Paper 2 section.