Chapter Introduction
Measures of central tendency tell you where the centre of the data lies, but they do not tell you how spread out the data is. Two datasets can have the same mean but very different distributions — one could be tightly clustered, the other widely scattered. Measures of dispersion quantify this spread. They tell you how much the individual values differ from the average and from each other. This chapter covers five measures: range, quartile deviation, mean deviation, standard deviation, and the Lorenz curve.
Standard deviation is the most important measure — it forms the basis for almost all advanced statistical techniques including correlation and regression. This chapter is numerically intensive and high-weighted in the exam.
Key Definitions
| Term | Definition |
|---|---|
| Dispersion | The extent to which values in a dataset differ from the average and from each other; also called variability or spread |
| Range | The difference between the highest (H) and lowest (L) values: R = H - L |
| Coefficient of range | (H - L) / (H + L) x 100 — a relative measure for comparison |
| Quartile deviation | Half the difference between Q3 and Q1: QD = (Q3 - Q1) / 2 |
| Mean deviation | The arithmetic mean of absolute deviations from the mean or median |
| Standard deviation | The square root of the arithmetic mean of squared deviations from the mean |
| Variance | The square of the standard deviation: sigma-squared |
| Coefficient of variation | (SD / Mean) x 100 — a relative measure for comparing variability |
| Lorenz curve | A graphical method showing inequality by plotting cumulative percentages against the line of equal distribution |
| Absolute measure | Dispersion expressed in the same units as the original data |
| Relative measure | Dispersion expressed as a ratio or percentage, unit-free |
Key Formulas
Range and Coefficient of Range
| Measure | Formula |
|---|---|
| Range | R = H - L (Highest minus Lowest) |
| Coefficient of range | (H - L) / (H + L) x 100 |
Quartile Deviation
| Measure | Formula |
|---|---|
| Quartile deviation (QD) | QD = (Q3 - Q1) / 2 |
| Coefficient of QD | (Q3 - Q1) / (Q3 + Q1) x 100 |
Where Q1 and Q3 are found using the same method as median:
- Q1 class: where cumulative frequency first exceeds N/4
- Q3 class: where cumulative frequency first exceeds 3N/4
- Q1 = L1 + [(N/4 - cf1) / f1] x h
- Q3 = L3 + [(3N/4 - cf3) / f3] x h
Mean Deviation
| Series | About mean | About median | ||||
|---|---|---|---|---|---|---|
| Individual | MD = Sum of \ | x - x-bar\ | / n | MD = Sum of \ | x - Md\ | / n |
| Discrete | MD = Sum of f\ | x - x-bar\ | / N | MD = Sum of f\ | x - Md\ | / N |
| Continuous | MD = Sum of f\ | x - x-bar\ | / N (x = class mark) | MD = Sum of f\ | x - Md\ | / N |
Note: Use absolute values (ignore negative signs) when computing mean deviation.
Mean deviation about median is always less than or equal to mean deviation about mean.
Standard Deviation
| Series | Direct method | Shortcut method | Step-deviation method |
|---|---|---|---|
| Individual | sigma = sqrt(Sum of (x - x-bar) squared / n) | sigma = sqrt(Sum of d squared / n - (Sum of d / n) squared) where d = x - A | Not commonly used |
| Discrete | sigma = sqrt(Sum of f(x - x-bar) squared / N) | sigma = sqrt(Sum of fd squared / N - (Sum of fd / N) squared) | sigma = h x sqrt(Sum of fd' squared / N - (Sum of fd' / N) squared) |
| Continuous | sigma = sqrt(Sum of f(x - x-bar) squared / N) where x = class mark | sigma = sqrt(Sum of fd squared / N - (Sum of fd / N) squared) | sigma = h x sqrt(Sum of fd' squared / N - (Sum of fd' / N) squared) |
Key shortcut formula (variance):
sigma-squared = (Sum of fd' squared / N) - (Sum of fd' / N) squared x h squared
Or equivalently: sigma-squared = (Sum of fx squared / N) - (x-bar) squared
Coefficient of Variation
CV = (sigma / x-bar) x 100
Higher CV = more dispersion = less consistency Lower CV = less dispersion = more consistency
Real Income (using SD and CV context)
Real income = (Nominal income / CPI) x 100
Detailed Concepts
Absolute vs Relative Measures of Dispersion
| Feature | Absolute measures | Relative measures |
|---|---|---|
| Unit | Same as original data (e.g., rupees, kg) | Unit-free (ratio or percentage) |
| Purpose | Measure spread within one dataset | Compare spread across datasets with different units or scales |
| Examples | Range, QD, mean deviation, standard deviation | Coefficient of range, coefficient of QD, coefficient of variation |
| Comparison | Cannot compare datasets with different units | Can compare any two datasets |
| Use case | Describing one dataset | Comparing consistency or variability between groups |
When to use which: If you are asked "Which group is more consistent?" or "Which batsman is more reliable?", always use the coefficient of variation. Absolute measures cannot compare groups with different means or different units.
Comparison of All Measures of Dispersion
| Feature | Range | Quartile Deviation | Mean Deviation | Standard Deviation |
|---|---|---|---|---|
| Based on | Only two extreme values | Q1 and Q3 (middle 50%) | All values (absolute deviations) | All values (squared deviations) |
| Effect of extreme values | Highly affected | Not affected | Affected moderately | Affected (squaring amplifies) |
| Mathematical properties | Poor | Poor | Moderate (absolute values limit algebra) | Excellent (used in further analysis) |
| Rigidly defined | Yes, but too simple | Yes | Yes | Yes |
| Calculation difficulty | Easiest | Easy | Moderate | Moderate to complex |
| Uses all observations | No (only 2) | No (only 2 quartiles) | Yes | Yes |
| Affected by sampling | Highly | Moderately | Moderately | Least affected (best for samples) |
| Further statistical use | None | Limited | Limited | Extensive (correlation, regression, etc.) |
| Best for | Quick overview | Skewed distributions, open-ended classes | When outliers must not dominate | Normal distributions, further statistical work |
Standard Deviation: Detailed Method Comparison
Direct method: Compute each (x - x-bar), square it, sum, divide by N, take square root. Works but tedious for large datasets.
Shortcut method (assumed mean): Take assumed mean A, compute d = x - A. Then:
- sigma-squared = (Sum of fd squared / N) - (Sum of fd / N) squared
- sigma = square root of sigma-squared
This avoids computing (x - x-bar) for each value, which reduces rounding errors.
Step-deviation method: Best for continuous series with equal class width. d' = (x - A) / h. Then:
- sigma = h x sqrt[(Sum of fd' squared / N) - (Sum of fd' / N) squared]
This simplifies calculations the most because d' values are small integers.
Worked example (step-deviation):
| Class | f | x | d'=(x-25)/5 | fd' | fd' squared |
|---|---|---|---|---|---|
| 0-10 | 4 | 5 | -4 | -16 | 64 |
| 10-20 | 8 | 15 | -2 | -16 | 32 |
| 20-30 | 12 | 25 | 0 | 0 | 0 |
| 30-40 | 6 | 35 | 2 | 12 | 24 |
| 40-50 | 5 | 45 | 4 | 20 | 80 |
| Total | 35 | 0 | 200 |
A = 25, h = 5 sigma = h x sqrt[(Sum of fd' squared / N) - (Sum of fd' / N) squared] = 5 x sqrt[(200/35) - (0/35) squared] = 5 x sqrt[5.71 - 0] = 5 x 2.39 = 11.95
Lorenz Curve
The Lorenz curve is a graphical representation of dispersion (inequality) in a distribution.
Steps to construct a Lorenz curve:
1. Calculate cumulative frequencies and cumulative values (using class marks x frequencies) 2. Convert both to cumulative percentages 3. Plot cumulative percentage of observations (or frequency) on the X-axis 4. Plot cumulative percentage of the variable (e.g., income, wealth) on the Y-axis 5. Draw the line of equal distribution (45-degree diagonal from origin) 6. The Lorenz curve lies below this diagonal 7. The larger the gap between the diagonal and the Lorenz curve, the greater the dispersion (inequality)
Interpretation:
- If the Lorenz curve coincides with the 45-degree line, there is perfect equality (zero dispersion)
- The farther the curve sags below the diagonal, the greater the inequality
- Used commonly to show income inequality, land distribution, wealth concentration
- The Gini coefficient can be derived from the area between the diagonal and the Lorenz curve
Advantages of Lorenz curve:
1. Provides a visual representation of inequality 2. Easy to compare two distributions by overlaying curves 3. Shows the entire distribution, not just a single number
Limitations:
1. Does not give a precise numerical measure 2. Two different distributions can have the same Lorenz curve 3. Requires careful calculation of cumulative percentages
Worked Example: Coefficient of Variation Comparison
Group A (Maths scores): Mean = 60, SD = 12 → CV = (12/60) x 100 = 20% Group B (English scores): Mean = 50, SD = 15 → CV = (15/50) x 100 = 30%
Conclusion: Group A (CV = 20%) is more consistent than Group B (CV = 30%). Although Group B has a higher SD in absolute terms, Group A has less relative variability.
Note: You cannot compare SD directly because the means are different. CV is the correct tool for comparison.
Important Points for Exams
1. Dispersion measures how spread out the data is — high dispersion means values are far from the average. 2. Range = H - L; simplest but least reliable (based on only two values). 3. Quartile deviation = (Q3 - Q1) / 2; based on the middle 50% of data; not affected by extremes. 4. Mean deviation uses absolute deviations from the mean or median; ignores algebraic signs. 5. Standard deviation is the most important and widely used measure; it squares deviations, giving more weight to extreme values. 6. The shortcut formula for variance: sigma-squared = (Sum of fd' squared / N) - (Sum of fd' / N) squared. 7. Coefficient of variation = (SD / mean) x 100; used to compare variability across different datasets. 8. A lower CV means greater consistency; a higher CV means greater variability. 9. The Lorenz curve shows inequality graphically; the farther from the diagonal, the greater the dispersion. 10. SD is the basis for almost all advanced statistical techniques (correlation, regression, etc.). 11. Mean deviation about median is always less than or equal to mean deviation about mean.
Common Mistakes
| Mistake | Why it is wrong | Correct approach | ||
|---|---|---|---|---|
| Forgetting to take the square root when computing SD | Variance = sigma-squared; SD = sigma = square root of variance | Always take the square root of variance to get SD | ||
| Using signed deviations in mean deviation | Mean deviation uses absolute values — negative signs are ignored | Use \ | x - x-bar\ | , not (x - x-bar) |
| Confusing sigma and sigma-squared | sigma = SD (in original units); sigma-squared = variance (in squared units) | SD = sqrt(variance); variance = SD squared | ||
| Comparing absolute measures across datasets with different units | Absolute measures are in different units and cannot be compared directly | Use coefficient of variation for comparison | ||
| Not converting to step-deviation correctly | d' = (x - A) / h, not (x - A) x h | Divide deviation by class width, not multiply | ||
| Thinking higher CV always means worse data | Higher CV means more variability/less consistency, but whether that is "bad" depends on context | Compare CVs between groups to determine relative consistency | ||
| Forgetting to multiply by h in step-deviation SD | sigma = h x sqrt[...] — the h multiplication is essential | Always multiply the square root by h |
NCERT Exercise Focus
Focus on these NCERT exercise questions while preparing:
1. "Calculate SD using step-deviation method" — Practice the full calculation with d', fd', fd' squared columns. 2. "Calculate mean deviation about mean and about median" — Practice both and compare. 3. "Which group is more consistent? (CV comparison)" — Very common 3-mark question. 4. "Calculate range and coefficient of range" — Simple but frequently asked 1-2 mark question. 5. "Draw a Lorenz curve and interpret it" — Practice cumulative percentage calculations.
Exam Tips
1. SD numerical problems carry the most marks — Practice computing SD using all three methods (direct, shortcut, step-deviation) for continuous series. 2. Step-deviation method is the most efficient — Master it. It is the fastest and least error-prone for large datasets. 3. Coefficient of variation comparison — You may be asked: "Which group is more consistent?" Compute CV for both; lower CV = more consistent. 4. Mean deviation — Remember to use absolute values. Practice both about mean and about median. 5. Lorenz curve — Describe construction steps and interpretation. The key point: larger gap from diagonal = more inequality. 6. Always show formulas before substitution — This earns step marks even if the final answer is wrong. 7. Shortcut variance formula — Memorise: sigma-squared = (Sum of fd' squared / N) - (Sum of fd' / N) squared. It saves enormous calculation time. 8. Range and QD — Quick 1-2 mark questions. Simple formulas but easy to confuse with each other.
Practice Questions for Self-Test
1. Calculate standard deviation using step-deviation method from given data. (5 marks) 2. Distinguish between absolute and relative measures of dispersion. (3 marks) 3. What is the coefficient of variation? When is it used? (2 marks) 4. Which group is more consistent: Group A (Mean=50, SD=10) or Group B (Mean=100, SD=15)? (2 marks) 5. Calculate mean deviation about median for the given data. (3 marks) 6. What is a Lorenz curve? How is it constructed and interpreted? (3 marks) 7. Compare range, quartile deviation, mean deviation, and standard deviation. (5 marks) 8. Calculate range and coefficient of range from the data: 12, 18, 25, 8, 30, 15. (2 marks) 9. What is variance? How is it related to standard deviation? (2 marks) 10. Explain the shortcut method for calculating standard deviation. (3 marks)
Key Statistics and Facts for Quick Reference
| Fact | Detail |
|---|---|
| Range formula | R = H - L |
| Coefficient of range | (H - L)/(H + L) x 100 |
| QD formula | (Q3 - Q1) / 2 |
| Mean deviation uses | Absolute values (ignore negative signs) |
| SD shortcut formula | sigma-squared = (Sum of fd' squared / N) - (Sum of fd' / N) squared |
| CV formula | (SD / Mean) x 100 |
| Lower CV means | More consistency |
| Higher CV means | More variability |
| Lorenz curve interpretation | Gap from 45-degree line shows inequality |
| Best measure for further analysis | Standard deviation |
| Least reliable measure | Range |
Quick Revision Notes
- Range: R = H - L. Coefficient of range = (H-L)/(H+L) x 100. Simplest, least reliable.
- Quartile deviation: QD = (Q3-Q1)/2. Coefficient of QD = (Q3-Q1)/(Q3+Q1) x 100. Not affected by extremes.
- Mean deviation: average of absolute deviations. About mean or about median. Uses \|x - x-bar\| (ignore signs).
- Standard deviation: sigma = sqrt(Sum of (x-x-bar) squared / N). Most important measure.
- Shortcut variance: sigma-squared = (Sum of fd' squared / N) - (Sum of fd' / N) squared. SD = sqrt of this.
- Step-deviation: sigma = h x sqrt[(Sum of fd' squared / N) - (Sum of fd' / N) squared].
- Coefficient of variation: CV = (sigma / x-bar) x 100. Lower CV = more consistent.
- Lorenz curve: plots cumulative percentages; gap from 45-degree line shows inequality.
- Absolute vs relative: absolute = same units as data; relative = unit-free (percentage).
- SD is best for further analysis; range is quickest; QD for open-ended data; MD for moderate outliers.