Statistics

While measures of central tendency summarize a distribution by a single central value, measures of dispersion describe how individual values scatter around that center.

Consider three distinct data series, each containing 5 observations:

• Series A: $50, \; 50, \; 50, \; 50, \; 50 \implies \text{Mean } \bar{x} = 50$
• Series B: $48, \; 49, \; 50, \; 51, \; 52 \implies \text{Mean } \bar{x} = 50$
• Series C: $0, \; 25, \; 50, \; 75, \; 100 \implies \text{Mean } \bar{x} = 50$

All three series have identical means, medians, and modes equal to $50$. Yet Series A has zero spread, Series B has very low spread, and Series C is wildly dispersed. Thus, a measure of central tendency alone fails to characterize the internal distribution of data.

The simplest measure of dispersion is the Range ($R$), defined as the algebraic difference between the highest (maximum) and lowest (minimum) observations:

Merits & Demerits of Range:

• Merits: Extremely easy to compute and understand; useful in industrial statistical quality control charts (e.g., R-charts) and daily weather forecasts.
• Demerits: Highly sensitive to extreme outliers; utilizes only two extreme values and completely ignores the distribution of the remaining $n - 2$ intermediate observations; cannot be computed for open-ended frequency tables.

Mean Deviation measures dispersion by calculating the arithmetic average of the absolute differences between each observation and a chosen central value.

By the algebraic property of the arithmetic mean, the sum of signed deviations from the mean is identically zero:

To avoid this self-cancellation, we take the absolute value (modulus) of each deviation: $|x_i - \bar{x}| \ge 0$.

For a discrete frequency distribution $(x_i, f_i)$ or continuous frequency distribution with class mid-points $x_i$ and total frequency $N = \sum_{i=1}^k f_i$:

Introduced by Karl Pearson in 1893, the standard deviation overcomes the mathematical limitations of absolute values by squaring deviations, forming the gold standard of statistical dispersion.

Let $x_1, x_2, \dots, x_n$ have mean $\bar{x} = \frac{1}{n}\sum x_i$:

• Direct Definition Form: $$\mathbf{\sigma^2 = \frac{1}{n} \sum_{i=1}^n (x_i - \bar{x})^2}$$
• Expanded Computational Form: $$\sigma^2 = \frac{1}{n}\sum (x_i^2 - 2x_i\bar{x} + \bar{x}^2) = \frac{1}{n}\sum x_i^2 - 2\bar{x}\left(\frac{\sum x_i}{n}\right) + \bar{x}^2 = \frac{1}{n}\sum x_i^2 - \bar{x}^2$$
  $$\mathbf{\sigma^2 = \frac{\sum x_i^2}{n} - \left(\frac{\sum x_i}{n}\right)^2 \iff \sigma = \sqrt{\frac{\sum x_i^2}{n} - \bar{x}^2}}$$

Let $x_i$ denote the observations (or class mid-points in continuous intervals) with corresponding frequencies $f_i$, and total frequency $N = \sum f_i$:

When data values or class mid-points are large numbers with decimals, direct calculation of $x_i^2$ becomes tedious. Change of origin and scale radically streamlines arithmetic.

Let $A$ be an assumed mean. Define deviation $d_i = x_i - A$:

When the data is grouped into continuous classes of equal interval width $h$, define the step-deviation variable:

Understanding how linear transformations affect variance and standard deviation provides powerful theoretical shortcuts in competitive examinations.

Combining origin shift and scaling: if $y_i = a x_i + b$ for constants $a$ and $b$:

Let two groups have sizes $n_1$ and $n_2$, means $\bar{x}_1$ and $\bar{x}_2$, and variances $\sigma_1^2$ and $\sigma_2^2$:

1. Combined Mean: $\mathbf{\bar{x} = \frac{n_1 \bar{x}_1 + n_2 \bar{x}_2}{n_1 + n_2}}$
2. Let deviation of group means from combined mean be $d_1 = \bar{x}_1 - \bar{x}$ and $d_2 = \bar{x}_2 - \bar{x}$.
3. Combined Variance: $$\mathbf{\sigma^2 = \frac{n_1(\sigma_1^2 + d_1^2) + n_2(\sigma_2^2 + d_2^2)}{n_1 + n_2}}$$

When comparing dispersion between two distributions measured in different units (e.g., heights in cm vs weights in kg) or with widely different mean levels, standard deviation alone is inadequate.

• Quality Assurance & Manufacturing: In production lines, batches with lower C.V. of dimensions indicate superior machine calibration and fewer defective parts.
• Financial Risk Management: Investors evaluate the volatility per unit of expected return. Two stocks may offer 10% expected return, but the stock with lower C.V. delivers that return with much less risk.
• Sports Performance Analytics: In cricket, baseball, and athletics, selecting reliable tournament squads requires identifying players whose scoring distributions exhibit small C.V.