Measures of Central Tendency

In any statistical distribution containing numerical data, observations rarely disperse haphazardly across the entire possible range. Instead, empirical distributions exhibit an inherent biological, economic, or physical propensity to cluster around a central or intermediate value. This statistical propensity of data to concentrate towards the center of a frequency distribution is known as Central Tendency. The single representative numerical magnitude around which the individual values concentrate is termed a Measure of Central Tendency, or colloquially, an Average.

"An average is an attempt to find one single figure to describe the whole of figures." — Sir Arthur Lyon Bowley

"A measure of central tendency is a typical value around which other figures aggregate." — Simpson and Kafka

1. Condensation of Mass Complex Data: Human cognitive capacity cannot absorb thousands of individual wage entries or crop outputs. An average condenses a sprawling dataset into a single, readily graspable figure (e.g., Per Capita Income of West Bengal).
2. Facilitating Comparative Analysis: Averages provide a standardized benchmark enabling valid comparisons between two or more disparate series (e.g., comparing the average monthly wage of jute mill workers with that of software engineers, or comparing wheat yields across districts).
3. Mathematical Foundation for Advanced Statistics: Statistical averages are indispensable mathematical inputs for calculating measures of dispersion (standard deviation), skewness, correlation, regression coefficients, and hypothesis tests.
4. Formulation of Economic and Social Policies: Governments and planning commissions utilize averages to define the official poverty line, calculate minimum statutory wage floors, determine food security allocations, and formulate fiscal budgets.

In classical statistical theory, George Udny Yule formulated six definitive criteria that an ideal measure of central tendency should fulfill:

Measures of central tendency are fundamentally bifurcated into two broad categories:

• Mathematical Averages: Measures derived strictly through arithmetic or algebraic operations involving every observation:
  • Arithmetic Mean (AM or $ar{X}$): Simple and Weighted.
  • Geometric Mean (GM): $n^{ ext{th}}$ root of the product of observations.
  • Harmonic Mean (HM): Reciprocal of the arithmetic mean of reciprocals.
• Positional Averages: Measures determined primarily by the specific location or relative ranking within an ordered array of data:
  • Median ($M$): The exact physical middle value dividing sorted data into two equal halves.
  • Partition Values: Quartiles ($Q_1, Q_2, Q_3$), Deciles ($D_1 \dots D_9$), and Percentiles ($P_1 \dots P_{99}$).
  • Mode ($Z$): The value corresponding to the point of maximum frequency concentration.

The Arithmetic Mean ($ar{X}$) is the most universally employed mathematical average. It is defined as the quotient obtained by dividing the algebraic sum of all observed values of a variable by the total number of observations.

Statisticians employ three primary computational techniques depending on the structure of the dataset:

The simple arithmetic mean assigns equal mathematical importance to every observation. However, in economic reality, commodities and items possess vastly different practical significance (e.g., in a household budget, food expenditure carries greater importance than entertainment). The Weighted Arithmetic Mean assigns explicit numerical weights ($w_i$) reflecting relative importance:

The arithmetic mean possesses five unique mathematical properties heavily tested in WBCHSE board examinations:

1. Sum of Deviations from Mean is Identically Zero: $$\sum_{i=1}^{N} (X_i - ar{X}) = 0 \quad ext{and for frequency data,} \quad \sum f(X - ar{X}) = 0$$ Proof: $\sum (X - ar{X}) = \sum X - \sum ar{X} = Nar{X} - Nar{X} = 0$.
2. Least Squares Property (Minimum Sum of Squared Deviations): $$\sum_{i=1}^{N} (X_i - ar{X})^2 < \sum_{i=1}^{N} (X_i - A)^2 \quad ext{for any arbitrary constant } A eq ar{X}$$ The sum of squared deviations from the arithmetic mean is strictly smaller than the sum of squared deviations from any other value.
3. Combined Mean Property ($ar{X}_{12}$): If a combined dataset consists of $k$ sub-groups with sample sizes $N_1, N_2, \dots, N_k$ and respective means $ar{X}_1, ar{X}_2, \dots, ar{X}_k$, the aggregate mean is: $$ar{X}_{12\dots k} = rac{N_1 ar{X}_1 + N_2 ar{X}_2 + \dots + N_k ar{X}_k}{N_1 + N_2 + \dots + N_k} = rac{\sum N_i ar{X}_i}{\sum N_i}$$
4. Linear Transformation (Change of Origin and Scale): If variable $X$ undergoes a linear transformation $Y = a + bX$, where $a$ is a change of origin and $b$ is a change of scale: $$ar{Y} = a + bar{X}$$
5. Correction of Erroneous Mean: If one or more observations were recorded incorrectly during data compilation: $$ ext{Corrected } \sum X = ext{Incorrect } \sum X - ext{Incorrect Value(s)} + ext{Correct Value(s)}$$ $$ ext{Corrected } ar{X} = rac{ ext{Corrected } \sum X}{N}$$

• Merits: Rigidly defined, utilizes all observations, intuitively simple, completely amenable to algebraic manipulation, and exhibits high sampling stability.
• Limitations: Highly sensitive to extreme outliers (unduly inflated by extreme values), cannot be determined graphically, cannot be calculated for open-ended classes without arbitrary boundary assumptions, and may yield an absurd theoretical result (e.g., average family size of 3.6 children).

The Median ($M$) is the positional average defined as the value of that item which divides a distribution into exactly two equal halves when the observations are arranged in an ordered sequence of magnitude (ascending or descending). Exactly 50% of the observations lie below the median, and 50% lie above it.

1. Individual Series:
   • Step 1: Arrange data in ascending array: $X_{(1)} \le X_{(2)} \le \dots \le X_{(N)}$.
   • If $N$ is odd: $ ext{Median} = \left(rac{N + 1}{2} ight)^{ ext{th}} ext{ item}$.
   • If $N$ is even: $ ext{Median} = rac{\left(rac{N}{2} ight)^{ ext{th}} ext{ item} + \left(rac{N}{2} + 1 ight)^{ ext{th}} ext{ item}}{2}$.
2. Discrete Frequency Distribution:
   • Step 1: Construct cumulative frequency ($cf$) column.
   • Step 2: Determine position index $rac{N + 1}{2}$ where $N = \sum f$.
   • Step 3: Locate the value of $X$ corresponding to the cumulative frequency immediately greater than or equal to $rac{N + 1}{2}$.
3. Continuous Frequency Distribution:
   • Step 1: Construct cumulative frequency ($cf$) distribution.
   • Step 2: Locate the Median Class containing the $rac{N}{2}^{ ext{th}}$ observation ($N = \sum f$).
   • Step 3: Apply the linear interpolation formula:
     $$M = L + \left[rac{rac{N}{2} - cf_0}{f_m} ight] imes c$$

     • $L$ = True lower limit/boundary of the median class
     • $N = \sum f$ = Total number of observations
     • $cf_0$ = Cumulative frequency of the pre-median class (class preceding median class)
     • $f_m$ = Simple frequency of the median class
     • $c$ = Class interval width of the median class

The median can be determined geometrically without mathematical calculation using Cumulative Frequency Curves (Ogives):

• Intersection Method: Plot both the "Less-Than" Ogive and the "More-Than" Ogive on the same graph with common axes. From their intersection point $P$, drop a perpendicular to the horizontal X-axis. The abscissa (value on the X-axis) is the exact Median ($M$).
• Single Ogive Method: Construct a "Less-Than" Ogive. On the vertical Y-axis, locate the value $rac{N}{2}$. Draw a horizontal line parallel to the X-axis until it intersects the curve. Drop a perpendicular from that point of intersection to the X-axis to read the Median.

Partition values are positional measures that divide a sorted frequency distribution into multiple equal portions:

• Minimal Absolute Deviation Property: The sum of absolute deviations of observations from the median is strictly minimal: $$\sum_{i=1}^{N} |X_i - M| \le \sum_{i=1}^{N} |X_i - A| \quad ext{for any arbitrary constant } A$$
• Key Merits: Unaffected by extreme outliers, readily computable for open-ended classes (since only the middle class interval requires boundaries), determinable graphically via Ogives, and ideal for qualitative phenomena (intelligence, poverty, health).
• Key Limitations: Not rigidly algebraic (cannot calculate combined median of two merged datasets), ignores the magnitude of non-central items, and has lower sampling stability than the mean.

The Mode ($Z$ or $M_o$) is derived from the French word la mode (meaning 'fashion'). In statistics, the mode is defined as the value of the variable which occurs with the greatest frequency in a distribution, representing the point of maximum density or concentration.

• Unimodal Distribution: Contains a single clear peak or dominant frequency.
• Bimodal Distribution: Contains two distinct values sharing identically or nearly identical high frequencies.
• Multimodal Distribution: Exhibits multiple local frequency peaks.
• Ill-Defined Mode: When all items occur with equal frequency, no mode exists.

When the distribution is irregular, or when two adjacent values exhibit nearly equal maximum frequencies, visual inspection can be deceptive. Statisticians utilize the Grouping Method consisting of two complementary tables:

1. Grouping Table: Contains 6 columns of frequencies:
   • Col 1: Original frequencies.
   • Col 2: Frequencies combined in pairs $(1+2), (3+4), (5+6) \dots$
   • Col 3: Frequencies combined in pairs starting from the 2nd item $(2+3), (4+5) \dots$
   • Col 4: Frequencies combined in threes $(1+2+3), (4+5+6) \dots$
   • Col 5: Frequencies combined in threes starting from the 2nd item $(2+3+4), (5+6+7) \dots$
   • Col 6: Frequencies combined in threes starting from the 3rd item $(3+4+5), (6+7+8) \dots$
2. Analysis Table: Tallies how many times each class interval or value is associated with the maximum frequency across all 6 columns. The class with the highest tally count is the true Modal Class.

Once the modal class is identified, the exact modal value is calculated via linear interpolation:

Alternative Formula: When $(2f_1 - f_0 - f_2) \le 0$ or when the formula yields a value outside the modal class, the absolute difference formula is applied:

The mode of a continuous frequency distribution can be directly determined from a Histogram:

1. Construct a histogram with class intervals on the X-axis and frequencies on the Y-axis.
2. Identify the highest rectangular column corresponding to the modal class.
3. From the top-left corner of the modal column, draw a diagonal straight line to the top-left corner of the immediately following (succeeding) column.
4. From the top-right corner of the modal column, draw a diagonal straight line to the top-right corner of the immediately preceding column.
5. From the point of intersection of these two diagonal lines, drop a perpendicular line straight down to the horizontal X-axis.
6. The coordinate point where the perpendicular cuts the X-axis gives the exact graphical value of the Mode ($Z$).

• Merits: Highly intuitive, easy to identify by inspection in discrete data, completely immune to extreme outlier distortion, computable in open-ended classes, and determinable graphically from histograms.
• Limitations: Ill-defined in bimodal or multimodal series, not based on all observations, incapable of further algebraic operations, and possesses poor sampling stability.
• Commercial & Economic Applications: Indispensable in commercial forecasting: ready-made clothing sizing (determining the most common shirt collar or chest size), shoe manufacturing (e.g., Bata stocking more size 8 and 9 shoes), fast-food inventory planning, and identifying peak traffic transit hours.

The Geometric Mean ($GM$) of a series of $n$ positive numerical observations is defined as the $n^{ ext{th}}$ root of their continued product:

Because multiplying $n$ large numbers directly is computationally intractable, logarithms are applied:

For discrete and continuous grouped frequency distributions:

• Averaging Rates, Ratios, and Percentages: The arithmetic mean severely overestimates compound growth. GM is the theoretically optimal average for compounding economic growth rates, population expansion rates, interest rates, and financial asset returns.
• Index Number Construction: As proven by Irving Fisher, GM is the only average that satisfies the rigorous Time Reversal Test and Factor Reversal Test in index number theory.
• Critical Limitation: If any single observation in the dataset is zero ($X_i = 0$), the entire product becomes zero ($GM = 0$). If any observation is negative, the $n^{ ext{th}}$ root may yield an imaginary number. Hence, GM is applicable strictly to non-zero, positive numbers.

The Harmonic Mean ($HM$) of a series of positive observations is defined as the reciprocal of the arithmetic mean of the reciprocals of the individual observations:

For grouped frequency distributions:

The Harmonic Mean is the mathematically correct average whenever the variable involves rates, speeds, prices per unit, or productivity per hour where the numerator of the rate is held constant across journeys or purchases:

• Average Speed over Equal Distances: If an automobile travels distance $d$ at speed $v_1$ and returns the same distance $d$ at speed $v_2$, the average speed is given by: $$v_{ ext{avg}} = rac{2}{rac{1}{v_1} + rac{1}{v_2}} = rac{2 v_1 v_2}{v_1 + v_2}$$ (Using simple arithmetic mean yields an incorrect, inflated average speed).
• Average Price under Constant Expenditure: When a consumer spends a fixed monetary budget (e.g., Rs. 500 per week) purchasing a commodity across different market days at varying prices per kg.
• Limitation: Cannot be calculated if any observation is zero (division by zero is undefined), and assigns disproportionately excessive weight to tiny observations.

For any set of positive real numbers $X_1, X_2, \dots, X_n$:

Furthermore, for any two positive real numbers $a$ and $b$:

In a perfectly symmetrical distribution, the Arithmetic Mean, Median, and Mode coincide at the exact identical point ($Mean = Median = Mode$). However, real-world economic distributions are typically moderately asymmetrical (skewed). Renowned statistician Karl Pearson established an empirical rule governing moderately skewed unimodal distributions:

Choosing the correct statistical average in economic analysis depends strictly on the nature of the data and the purpose of the study: