Theory
Same average, different animals
Two batches sit the same test. Both average 60.
Batch P: 58, 59, 60, 61, 62. Batch Q: 20, 40, 60, 80, 100.
One is a tight, teachable group; the other spans failure to topper in one room. The mean, our star of last lesson, cannot tell them apart.
Every centre needs a second number: how far do values stray from it? That number is dispersion, and the exam wants three versions of it, by hand.
Theory
Archers around a bullseye
Two archers both average dead-centre. One groups every arrow within a coin's width; the other sprays the whole target face, misses cancelling misses.
Same average position, completely different archers. Dispersion measures the scatter of the arrows, not where their centre lies. A summary without it praises the sprayer and the sniper equally.
Theory
Range: the sixty-second answer
Range = maximum - minimum.
Batch P: 62 - 58 = 4. Batch Q: 100 - 20 = 80. Instant, and already tells the two batches apart.
Its weakness is structural: it consults only two values and ignores the other n-2. One eccentric outlier (a single 600-minute screen-timer) explodes the range while the crowd sits unchanged. Use it as a first glance, never as the verdict.
Theory
Variance: why we square
Better idea: measure every value's deviation from the mean and average them. One catch: raw deviations always sum to zero (the mean is their balance point): +2 and -2 cancel, reporting fake calm.
The fix: square each deviation first: negatives vanish, big misses count extra.
Population variance σ² = Σ(x - mean)² ÷ N.
Sample variance s² divides by (n - 1) instead (Bessel's correction: samples slightly understate spread; n-1 compensates: name it, use it whenever data is a sample).
Theory
Worked, start to finish
Marks: 4, 8, 6, 2 (a population, for simplicity).
1. Mean = 20 ÷ 4 = 5.
2. Deviations: -1, +3, +1, -3. (Sum = 0 ✓ the built-in error check.)
3. Squares: 1, 9, 1, 9. Sum = 20.
4. Variance σ² = 20 ÷ 4 = 5 (units: marks²).
5. SD σ = √5 ≈ 2.24 marks: back in real units, the reportable number.
As a sample: s² = 20 ÷ 3 ≈ 6.67, s ≈ 2.58. Same recipe, different divisor.
Think first
Your turn, full recipe
Screen-time hours of 5 students: 2, 4, 4, 5, 10. On paper: mean, deviations (check they sum to 0), squared deviations, POPULATION variance and SD. Then tap.
Show the answer
Mean = 25 ÷ 5 = 5.
Deviations: -3, -1, -1, 0, +5 (sum 0 ✓).
Squares: 9, 1, 1, 0, 25. Sum = 36.
Variance = 36 ÷ 5 = 7.2 hours². SD = √7.2 ≈ 2.68 hours.
Note how the single 10 contributed 25 of the 36: squaring makes outliers shout. If your deviations did not sum to zero, the mean was wrong: recompute before squaring anything.
Quiz
A student reports "the variance of marks is 5, so scores typically differ from the mean by 5 marks." What is wrong?
- Variance is in SQUARED units (marks²); the typical distance is the SD, √5 ≈ 2.24 marks
- Nothing: variance and SD are the same number
- Variance cannot be computed for marks
- The typical distance is the range, not the SD
Show the answer
Variance is in SQUARED units (marks²); the typical distance is the SD, √5 ≈ 2.24 marks
Variance lives in squared units, useful for algebra but unreadable as a distance: nobody strays "5 square marks". The standard deviation square-roots it back to real units, making √5 ≈ 2.24 the honest "typical distance from the mean". Confusing the two is the most common dispersion error in scripts; the range (option D) measures total width, not typical distance.
Watch out
The three dispersion traps
Skipping the squares: raw deviations sum to zero: every dataset would claim zero spread.
Divisor confusion: population ÷ N, sample ÷ (n-1). State which you used: R and pandas default to the SAMPLE formula, plain calculators often to population.
Reporting variance as a distance: convert to SD before interpreting. And remember the range consulted only two values: never call it robust.
Theory
Where SD quietly runs things
Result moderation scales marks using mean and SD. Quality control stops a production line when SD creeps up. Cricket commentators' "consistent batsman" is a low-SD claim. And two lessons ahead, the bell curve turns SD into a ruler: 68% of data within 1 SD, 95% within 2: today's hand-computed number becomes the unit the whole normal distribution is measured in. (Comparing spreads across units? Divide SD by mean: the coefficient of variation.)
Summary
Key takeaways
- Dispersion answers "how scattered?": the mean alone cannot separate tight from wild data.
- Range = max - min: instant, but two values only and outlier-fragile.
- Deviations from the mean sum to zero: squaring kills cancellation and weights big misses.
- Variance = average squared deviation (÷N population, ÷(n-1) sample: Bessel).
- SD = √variance: back in real units, the number you report and interpret.
- Deviation sum = 0 is the free arithmetic check before squaring.
- Memory hook: archers around a bullseye: same centre, different scatter.