Measures of Dispersion

Measures of central tendency (Arithmetic Mean, Median, Mode) are designed to provide a single representative figure that summarizes an entire distribution. However, an average by itself fails to convey the complete picture of a statistical series. It indicates the location of the center, but reveals nothing about how individual observations are scattered or clustered around that center.

Consider three distinct factories employing 5 workers each, with the following daily wage distributions (in ₹):

• Factory A: 50, 50, 50, 50, 50 $\implies \text{Mean } (\bar{X}) = ₹50$
• Factory B: 45, 48, 50, 52, 55 $\implies \text{Mean } (\bar{X}) = ₹50$
• Factory C: 10, 20, 50, 80, 90 $\implies \text{Mean } (\bar{X}) = ₹50$

All three factories share the identical average wage of ₹50. Yet, in Factory A there is zero variation (perfect uniformity); in Factory B there is small variation (high consistency); and in Factory C there is extreme variation (severe wage inequality). Without measuring the spread of the data, the average alone is misleading. This spread or variation is termed Dispersion.

• Arthur L. Bowley: "Dispersion is the measure of the variations of the items."
• Brooks & Dick: "Dispersion or spread is the degree of the scatter or variation of the variables about a central value."
• Simpson & Kafka: "The measurement of the scatterness of the mass of figures in a series about an average is called measure of variation or dispersion."

Because measures of central tendency are called averages of the first order, measures of dispersion—which measure the average deviation from those central values—are logically designated as averages of the second order.

1. Assessing the Reliability of an Average: If dispersion is small, individual values cluster tightly around the average, making it highly representative and reliable. If dispersion is large, the average is unrepresentative.
2. Comparing Variability, Consistency, or Stability: Comparing two or more distributions (e.g., scoring consistency of two cricket batsmen, or production consistency of two machines).
3. Controlling Variability: In industrial manufacturing, keeping product dimensions within acceptable tolerance limits through Statistical Quality Control (SQC).
4. Facilitating Advanced Statistical Analysis: Dispersion serves as the foundational component for computing correlation coefficients, regression equations, testing hypotheses, and constructing economic index numbers.

According to statistician G. Udny Yule, a satisfactory measure of dispersion must satisfy six criteria:

1. It should be rigidly defined by a definitive mathematical formula.
2. It should be based on all observations in the distribution.
3. It should be readily comprehensible and simple to interpret.
4. It should be simple to compute without undue algebraic labor.
5. It should be capable of further algebraic manipulation (e.g., combined dispersion of multiple groups).
6. It should possess sampling stability, being least affected by random sampling fluctuations.

The Range is the simplest and crudest measure of dispersion. It is defined as the absolute difference between the largest (maximum) value and the smallest (minimum) value in a series.

Where $L$ is the largest value and $S$ is the smallest value.

• Individual & Discrete Series: Identify the maximum and minimum values directly from the observations $X$.
• Continuous Frequency Distribution: Computed either as:
  • Upper limit of the highest class interval minus Lower limit of the lowest class interval; or
  • Mid-point of the highest class minus Mid-point of the lowest class.

Evaluation:

• Merits: Exceptionally simple to compute; readily understood; widely used in daily weather reporting (maximum and minimum temperature) and industrial quality control charts (X-bar and R charts).
• Demerits: Highly unstable because it depends entirely on two extreme outlier values; completely ignores the intermediate 98% of observations; cannot be computed for open-ended frequency distributions.

To eliminate the extreme outlier sensitivity of the Range, statisticians developed positional measures based on partition values called Quartiles.

Quartiles divide an ordered distribution into four equal parts: $Q_1$ (First/Lower Quartile, 25th percentile), $Q_2$ (Median, 50th percentile), and $Q_3$ (Third/Upper Quartile, 75th percentile).

The Interquartile Range represents the absolute span of the central 50% of the distribution, while the Quartile Deviation (QD) is half of this span, hence termed the Semi-Interquartile Range.

Step-by-step procedure for Continuous Series:

1. Construct the cumulative frequency ($c.f.$) column.
2. Locate the $Q_1$ class corresponding to cumulative frequency containing $\frac{N}{4}$, and the $Q_3$ class containing $\frac{3N}{4}$.
3. Apply the interpolation formulas: $$Q_1 = L_1 + \frac{\frac{N}{4} - c.f.}{f_1} \times i_1$$ $$Q_3 = L_3 + \frac{\frac{3N}{4} - c.f.}{f_3} \times i_3$$ Where $L$ is the lower boundary of the quartile class, $c.f.$ is the cumulative frequency of the preceding class, $f$ is the simple frequency of the quartile class, and $i$ is the class width.

• Merits:
  • Superior to the Range because it excludes the extreme 25% lower and 25% upper observations, making it completely immune to extreme outliers.
  • Can be easily computed for open-ended distributions (e.g., "Below 10" or "Above 80") because $Q_1$ and $Q_3$ lie safely in the middle ranges.
• Limitations:
  • Ignores the actual values of 50% of the data (the bottom quarter and top quarter).
  • Not amenable to algebraic manipulation; combined quartile deviations of multiple distributions cannot be calculated.
  • More sensitive to sampling variations than standard deviation.

The Mean Deviation (or Average Deviation) is defined as the arithmetic mean of the numerical deviations of observations from an average (Mean, Median, or Mode), with the critical mathematical stipulation that all algebraic signs (+ and -) are ignored (taking absolute values $|d|$).

If signs were not ignored, the sum of deviations taken from the arithmetic mean would be identically zero: $\sum (X - \bar{X}) = 0$. By taking absolute values $|X - \bar{X}|$, the measure quantifies the true average distance of observations from the center.

To convert absolute Mean Deviation into relative measures for comparative analysis:

• Merits:
  • Based on all observations in the dataset.
  • Simple to comprehend and intuitively meaningful as the average distance from the center.
  • Less distorted by extreme outliers compared to the standard deviation.
• Limitations:
  • Mathematically flawed: Artificially ignoring algebraic signs ($|-d| = +d$) violates basic algebraic principles and makes algebraic manipulation impossible.
  • Incapable of being combined algebraically across multiple groups.
  • Rarely used in advanced inferential statistics and hypothesis testing for this reason.

Introduced by the eminent British statistician Karl Pearson in 1893, the Standard Deviation (denoted by the lowercase Greek letter sigma, $\sigma$) is the master measure of dispersion in statistical science. It completely overcomes the mathematical flaw of Mean Deviation (dropping negative signs) by squaring the deviations before averaging them, and then taking the positive square root.

Formal Definition: Standard Deviation is the positive square root of the arithmetic mean of the squares of deviations of observations taken strictly from their arithmetic mean.

Because it computes the root of the mean of squared deviations, it is also classically termed the Root-Mean-Square Deviation.

The square of the standard deviation is called the Variance:

In WBCHSE examinations, candidates can compute $\sigma$ using three standard techniques:

Standard Deviation possesses unique algebraic properties that make it the bedrock of inferential statistics:

The sum of squares of deviations of observations is minimum when taken from the arithmetic mean:

This guarantees that the standard deviation is the unique minimum root-mean-square deviation of any dataset.

The Standard Deviation is independent of change of origin. If a constant $a$ is added to or subtracted from each observation ($Y = X \pm a$), the standard deviation remains completely unchanged:

The Standard Deviation is dependent on change of scale. If each observation is multiplied or divided by a constant $c$ ($Y = c \cdot X$), the standard deviation is multiplied by the absolute value of that constant:

If two independent samples have sizes $N_1, N_2$, means $\bar{X}_1, \bar{X}_2$, and standard deviations $\sigma_1, \sigma_2$, their combined standard deviation $\sigma_{12}$ is given by:

Where:

• Combined Mean: $\bar{X}_{12} = \frac{N_1\bar{X}_1 + N_2\bar{X}_2}{N_1 + N_2}$
• Mean deviations: $d_1 = \bar{X}_1 - \bar{X}_{12}$ and $d_2 = \bar{X}_2 - \bar{X}_{12}$

For a moderately symmetrical or bell-shaped normal distribution, well-established mathematical relationships exist among the three primary dispersion measures:

Area Spread of the Normal Curve:

• $\bar{X} \pm 1\sigma$ encompasses approximately 68.27% of all observations.
• $\bar{X} \pm 2\sigma$ encompasses approximately 95.45% of all observations.
• $\bar{X} \pm 3\sigma$ encompasses approximately 99.73% of all observations (virtually the entire population).
• $\bar{X} \pm 1 QD$ encompasses exactly 50.00% of all observations.

When comparing the variability, consistency, or stability of two or more distributions with different measurement units or widely differing arithmetic means, absolute standard deviation is inadequate. In 1895, Karl Pearson introduced the Coefficient of Variation (CV) as a standardized percentage metric of relative dispersion:

Formulated by American economic statistician Max O. Lorenz in 1905, the Lorenz Curve is a cumulative percentage graphical technique designed to measure and display economic inequality in the distribution of national income, wealth, profits, or landholdings.

Construction Principles:

1. Convert both the frequency (e.g., number of income earners) and the variable (e.g., total income earned) into cumulative percentages running from 0% to 100%.
2. Take the cumulative percentage of income earners on the horizontal axis ($X$-axis) from 0 to 100%.
3. Take the cumulative percentage of total income on the vertical axis ($Y$-axis) from 0 to 100%.
4. Draw the 45° diagonal line connecting $(0, 0)$ to $(100, 100)$. This is the Line of Equal Distribution (or Line of Absolute Equality). On this line, bottom 20% of earners receive 20% of income, bottom 50% receive 50% of income, etc.
5. Plot the actual cumulative percentage coordinates and connect them with a smooth curve. This is the Lorenz Curve.

• In any real-world unequal society, the Lorenz Curve always sags below the 45° Line of Equal Distribution.
• The Distance Principle: The greater the distance/curvature of the Lorenz Curve away from the 45° diagonal line, the greater the degree of economic inequality in the distribution of income.
• If Country A's Lorenz Curve lies closer to the diagonal than Country B's curve, Country A enjoys a more equitable distribution of wealth.
• Gini Coefficient ($G$): Defined as the ratio of the area between the 45° line and the Lorenz Curve to the total area under the 45° line. $G = 0$ signifies perfect equality, and $G = 1$ signifies absolute inequality (one person owns all national income).