MODULE II: CORRELATION AND REGRESSION ANALYSIS

Module Architectural Focus: This module delivers an exhaustive, mathematically rigorous treatment of Bivariate Data Analysis. It explores the conceptual, computational, and inferential dimensions of statistical association and functional dependence between commercial variables. Commencing with the philosophical and mathematical foundations of Correlation Analysis, the text explores Scatter Diagrams, Karl Pearson's Product-Moment Coefficient, Spearman's Non-Parametric Rank Correlation (including tied rank adjustments), and the Concurrent Deviation Method. It then advances to Linear Regression Analysis, detailing the Principle of Ordinary Least Squares (OLS), derivation and interpretation of bivariate regression equations (

Y on X

and

X on Y

), the fundamental properties of regression coefficients, angular divergence between regression lines, the Standard Error of Estimate, and real-world forecasting applications.

1. FOUNDATIONAL CONCEPTS OF BIVARIATE RELATIONSHIPS

In corporate and economic systems, variables rarely operate in isolation. Managerial decisions depend critically upon understanding how alterations in one market factor correspond with variations in another. When an investigation simultaneously examines two quantifiable characteristics across identical statistical entities, the distribution is characterized as a Bivariate Frequency Distribution.

Concept of Co-Variation

If two variables fluctuate simultaneously such that changes in the magnitude of one variable are accompanied by corresponding directional changes in the magnitude of the other, the variables are said to be correlated. Correlation measures the strength and direction of this mutual association.

Causation vs. Covariation

Correlation establishes co-variation, not causal dependency. A statistically significant correlation between (X) and (Y) may arise from: (a) (X) causing (Y), (b) (Y) causing (X), (c) both (X) and (Y) being influenced by an external lurking variable (Z), or (d) sheer coincidental chance (Spurious / Nonsense Correlation).

Taxonomy of Correlation

Statistical theory categorizes correlation across three distinct dimensions based on direction, linearity, and number of variables under examination:

Classification Basis	Sub-Categories	Mathematical Characterization & Commercial Illustration
1. Direction of Change	Positive (Direct) Negative (Inverse)	• Positive: Both variables move in the identical direction ((Delta X > 0 implies Delta Y > 0)). E.g., Advertising expenditure and product sales revenue; Disposable income and consumer expenditure. • Negative: Variables move in opposite directions ((Delta X > 0 implies Delta Y < 0)). E.g., Unit price and market quantity demanded; Production volume and per-unit fixed overhead cost.
2. Ratio of Variation	Linear Correlation Curvilinear (Non-linear)	• Linear: The amount of change in one variable bears a strictly constant ratio to the change in the other variable. Plotted on a Cartesian graph, points cluster along a straight line ((Y = a + bX)). • Curvilinear: The ratio of variation is not constant, altering at varying levels of the independent variable. Plotted points trace a curve (e.g., diminishing marginal returns curve, production cost curves).
3. Number of Variables	Simple Correlation Partial Correlation Multiple Correlation	• Simple: Studies mutual co-variation between exactly two variables ((X) and (Y)). • Partial: Measures relationship between two variables while holding one or more interrelated variables mathematically constant (denoted (r_{12.3})). • Multiple: Investigates the joint relationship between one dependent variable and the simultaneous combined effect of two or more independent variables (denoted (R_{1.23})).

+-----------------------------------------------------------------------------------+ | BIVARIATE STATISTICAL RELATIONSHIP ARCHITECTURE | +-----------------------------------------------------------------------------------+ | | | [BIVARIATE DATA (X, Y)] | | | | | +---> 1. CORRELATION ANALYSIS (Degree & Direction of Association) | | | ├── Qualitative / Graphic: Scatter Diagram | | | ├── Parametric: Karl Pearson's Product Moment (r) | | | ├── Non-Parametric / Ranked: Spearman's Rho (R) | | | └── Directional: Concurrent Deviation (rc) | | | | | +---> 2. REGRESSION ANALYSIS (Functional Dependence & Prediction) | | ├── OLS Best-Fit Regression Line: Y on X | | ├── OLS Best-Fit Regression Line: X on Y | | ├── Regression Coefficients: byx and bxy | | └── Precision Metric: Standard Error of Estimate (Syx, Sxy) | +-----------------------------------------------------------------------------------+

2. SCATTER DIAGRAM METHOD

The Scatter Diagram is a visual, diagrammatic device for inspecting bivariate data. Values of the independent or predictor variable ((X)) are plotted along the horizontal (X)-axis, while the corresponding values of the dependent or criterion variable ((Y)) are plotted along the vertical (Y)-axis. Each pair of observations ((X_i, Y_i)) is plotted as a discrete coordinate dot.

1. Perfect Positive Correlation ((r = +1.0))

All plotted coordinate points lie precisely upon a straight line rising upward from the lower-left corner to the upper-right corner. Every unit increment in (X) generates a constant proportional increment in (Y).

2. Perfect Negative Correlation ((r = -1.0))

All coordinate points fall strictly upon a straight line sloping downward from the upper-left corner to the lower-right corner. Every unit increment in (X) generates a constant proportional reduction in (Y).

3. High Degree Positive / Negative

The points do not fall exactly on a line but cluster closely around an upward-sloping straight line (

+ 0.75 \leq r < 1.0

) or downward-sloping straight line (

- 1.0 < r \leq - 0.75

4. Zero / Absence of Correlation ((r = 0))

The plotted dots are scattered chaotically across the Cartesian plane in a circular or shapeless cloud, or form a horizontal/vertical line, indicating zero systematic co-variation between (X) and (Y).

3. KARL PEARSON'S COEFFICIENT OF CORRELATION ((r))

Introduced by the English biometrician Karl Pearson, the Product-Moment Correlation Coefficient ((r)) is the most widely utilized parametric measure of linear association between two quantitative variables. It is defined mathematically as the ratio of the covariance between (X) and (Y) to the product of their respective standard deviations.

Master Mathematical Formula:

r = \frac{Cov ( X , Y )}{σ _{X} \cdot σ _{Y}} = \frac{\frac{1}{N} \sum ( X - X ˉ ) ( Y - Y ˉ )}{\frac{1}{N} \sum ( X - X ˉ ) ^{2} \cdot \frac{1}{N} \sum ( Y - Y ˉ ) ^{2}} = \frac{\sum ( X - X ˉ ) ( Y - Y ˉ )}{\sum ( X - X ˉ ) ^{2} \sum ( Y - Y ˉ ) ^{2}}

Alternative Operational Formulas for Computation

In applied business statistics, direct computation from actual arithmetic means becomes cumbersome when $X ˉ$ or $Y ˉ$ results in fractional values. Consequently, statisticians employ specific computational transformations:

Formula 1: Direct Method from Actual Means
Letting deviations from actual means be (x =

X - X ˉ

) and (y =

Y - Y ˉ

r = \frac{\sum x y}{\sum x ^{2} \cdot \sum y ^{2}}

Formula 2: Shortcut Method (Using Assumed Means (A_x) and (A_y))
Where (dx = (X - A_x)) and (dy = (Y - A_y)):

r = rac{N sum dx dy - (sum dx)(sum dy)}{sqrt{[N sum dx^2 - (sum dx)^2][N sum dy^2 - (sum dy)^2]}}

· N ∑ dy^2 -

\sum d y

^2}} ]
Formula 3: Step-Deviation Method (Grouped or Scaled Series)
Where

d x^{'} = \frac{X - A _{x}}{c _{1}}

and

d y^{'} = \frac{Y - A _{y}}{c _{2}}

, where (c_1) and (c_2) represent class intervals:

r = rac{N sum dx' dy' - (sum dx')(sum dy')}{sqrt{[N sum dx'^2 - (sum dx')^2][N sum dy'^2 - (sum dy')^2]}}

· N ∑ dy'^2 -

\sum d y^{'}

^2}} ] Note: Scale factors (c_1) and (c_2) cancel out completely from numerator and denominator, proving that (r) is independent of scale.

Crucial Mathematical Properties of Pearson's (r)

Karl Pearson's correlation coefficient possesses vital mathematical characteristics that govern its analytical behavior:

Property 1: Strictly Bounded Limits

The value of Pearson's (r) always lies within the closed interval:

- 1.0 \leq r \leq + 1.0

A value of (+1) indicates perfect direct linear correlation; (-1) denotes perfect inverse linear correlation; (0) indicates complete absence of linear association.

Property 2: Invariance to Origin and Scale

The correlation coefficient is a pure, dimension-free relative number independent of the units of measurement. It is invariant to shifts in origin and changes of scale. If

u = \frac{X - a}{h}

and

v = \frac{Y - b}{k}

with (h > 0, k > 0), then (r_{uv} = r_{xy}).

Property 3: Symmetry of Relationship

The correlation between (X) and (Y) is identical to the correlation between (Y) and (X):

r_{x y} = r_{y x}] T h e d es i g na t i o n o f w hi c h v a r iab l e i sco n s i d er e d in d e p e n d e n t or d e p e n d e n t ha s n o ma t h e ma t i c a l e f f ec t o n (r) . < / d i v >< h 3 > P r o p er t y 4 : G eo m e t r i c M e an o f R e g r ess i o n < / h 3 >< d i v > P e a r so n^{'} s (r) i s t h e g eo m e t r i c m e an o f t h e tw o bi v a r ia t er e g r ess i o n coe f f i c i e n t s (b_{y x}) an d (b_{x y}) : [r = \pm b_{y x} \cdot b_{x y}

The sign of (r) must match the common sign of both regression coefficients.

4. STATISTICAL SIGNIFICANCE: PROBABLE ERROR & DETERMINATION

When a correlation coefficient (r) is calculated from a random sample of (N) pairs of observations drawn from a bivariate population, it represents a sample statistic subject to sampling fluctuations. Assessing whether the calculated (r) reflects a genuine population relationship requires statistical significance testing.

Standard Error and Probable Error of (r)

Standard Error of (r) ((S.E._r)):
The standard error of the correlation coefficient is given by:

S . E ._{r} = \frac{1 - r ^{2}}{N}

Probable Error of (r) ((P.E._r)):
Probable Error is a classical measure defining the limits within which the population correlation coefficient ( ho) is expected to lie with a 50% probability:

P . E ._{r} = 0.67449 \times S . E ._{r} = 0.6745 \cdot \frac{1 - r ^{2}}{N}

Rules for Interpreting Correlation via Probable Error

DECISION RULE

Condition 1 (Absence of Significance): If (|r| < P.E._r), the correlation is not significant. There is no empirical evidence of any association between the variables.
Condition 2 (Definite Significance): If $∣ r ∣ > 6 \times P . E ._{r}$ , the correlation is definitely significant. The sample statistic reliably demonstrates true population correlation.
Condition 3 (Inconclusive Zone): If $P . E ._{r} \leq ∣ r ∣ \leq 6 \times P . E ._{r}$ , the relationship is weak or inconclusive; no definitive decision can be drawn without increasing sample size.
Confidence Band: The limits for the true population correlation coefficient ( ho) are:

$Population Correlation Limits = r \pm P . E ._{r} (50% probability) or r \pm 3 (P . E ._{r}) (99.73% confidence)$

Coefficient of Determination ((r^2)) and Alienation

The correlation coefficient (r) measures the degree of linear association, but it does not represent a proportion. To evaluate the explanatory power of the relationship, statisticians use the Coefficient of Determination ((r^2)).

Statistical Metric	Algebraic Formula	Managerial & Analytical Interpretation
Coefficient of Determination	$r^{2} = \frac{Explained Variance}{Total Variance}$	Represents the proportion (or percentage) of the total variation in the dependent variable that is directly explained by its linear relationship with the independent variable. For instance, if (r = 0.80), then (r^2 = 0.64), signifying that 64% of the variance in (Y) is explained by (X), while 36% remains unexplained.
Coefficient of Non-Determination	$k^{2} = 1 - r^{2} = \frac{Unexplained Variance}{Total Variance}$	Measures the proportion of variation in the dependent variable attributable to chance, omitted factors, or random disturbances.
Coefficient of Alienation	$k = 1 - r^{2}$	Measures the lack of association or the degree of departure from perfect linear relationship between the two variables.

5. SPEARMAN'S RANK CORRELATION COEFFICIENT ((R) OR ( ho))

Developed in 1904 by British psychologist Charles Edward Spearman, the Rank Correlation Coefficient is a non-parametric technique applied when variables cannot be measured on an exact quantitative scale, but can be ordered or ranked based on qualitative attributes (e.g., brand loyalty, employee leadership, aesthetics, consumer taste preferences).

Case I: When Ranks are Distinct (No Tied Ranks):
Let (R_1) and (R_2) be the ranks assigned to an individual on attributes (X) and (Y). Defining rank difference (D = R_1 - R_2):

R = 1 - \frac{6 \sum D ^{2}}{N ( N ^{2} - 1 )}

Where (N) represents the number of paired observations, and

\sum D^{2}

is the sum of squared differences of corresponding ranks.

Handling Tied / Repeated Ranks (Correction Factor)

When two or more items possess identical numerical values, they cannot be awarded distinct integer ranks. Under the bracket rank procedure, each tied observation is assigned the arithmetic mean of the ranks they would have occupied if they were distinct.

Case II: Formulation for Tied / Repeated Ranks:
When ties occur, the term ((m(m^2 - 1)) / (12)) is added to

\sum D^{2}

for each tied group, where (m) denotes the number of items sharing that particular rank:

R = 1 - rac{6 left[ sum D^2 + sum rac{m_i(m_i^2 - 1)}{12} ight]}{N(N^2 - 1)}

}{N(N^2 - 1)} ] Example: If a value repeats 3 times ((m = 3)), the correction term is ((3(9 - 1)) / (12) = (24) / (12) = 2.0). If another value repeats twice ((m = 2)), its correction term is ((2(4 - 1)) / (12) = (6) / (12) = 0.5).

6. CONCURRENT DEVIATION METHOD

The Method of Concurrent Deviations is a rapid, non-parametric approach focusing purely on the direction of change from one time period or observation to the next, ignoring the magnitude of changes. It is particularly useful for short-term financial time series and volatile economic data.

Concurrent Deviation Formula:

r_{c} = \pm \pm (\frac{2 c - n}{n})

Where:

(N) = Total number of paired observations.
(n = (N - 1)) = Total number of paired directional changes.
(c) = Number of concurrent deviations (i.e., pairs where both signs are (+), or both signs are (-)).
The algebraic sign $\pm$ is determined by the sign of ((2c - n)). If ((2c - n) > 0), the positive sign is retained both inside and outside the radical. If ((2c - n) < 0), the negative sign is retained outside, and the inner term is multiplied by (-1) to prevent imaginary numbers.

7. REGRESSION ANALYSIS: PRINCIPLES & LEAST SQUARES ARCHITECTURE

While correlation measures the degree and direction of association between two variables, it does not imply functional dependency or enable forecasting. Regression Analysis is the statistical methodology used to estimate or predict the unknown value of one variable (dependent or explained variable) from known values of another variable (independent or predictor variable).

Fundamental Distinction: Correlation vs. Regression

THEORETICAL COMPARISON

Analytical Dimension	Correlation Analysis	Regression Analysis
1. Objective	Measures the degree, strength, and direction of linear association.	Estimates the quantitative functional relationship to enable predictive forecasting.
2. Dependency Structure	Symmetrical: Variables (X) and (Y) are treated co-equally ((r_{xy} = r_{yx})).	Asymmetrical: Differentiates strictly between cause (independent) and effect (dependent).
3. Number of Equations	A single numerical coefficient ((r)) summarizes the relationship.	Yields two distinct regression lines: $Y on X$ and $X on Y$ .
4. Dimensionality	Relative, unit-free measure ( $- 1 \leq r \leq + 1$ ).	Absolute measures expressed in the original physical units of the dependent variable.
5. Impact of Origin/Scale	Invariant to changes of both origin and scale.	Invariant to change of origin, but sensitive to change of scale.
6. Error Terms	No assumption of stochastic error or disturbance term.	Explicitly minimizes the sum of squared residual errors ( $\sum e_{i}^{2}$ ).

Principle of Ordinary Least Squares (OLS)

The standard criterion for fitting a line of best fit through scattered bivariate data is the Principle of Least Squares. The method determines parameter coefficients such that the sum of the squared vertical deviations of the observed data points from the fitted line is minimized:

Minimize i = 1 \sum N e_{i}^{2} = i = 1 \sum N (Y_{i} - Y^_{i})^{2} = i = 1 \sum N [Y_{i} - (a + b X_{i})

^2 ] Differentiating this sum of squares partially with respect to (a) and (b) and setting the partial derivatives to zero yields the classical Normal Equations:

Normal Equations for Line of (Y) on (X) ((Y = a + bX)):

\sum Y \sum X Y = N a + b \sum X = a \sum X + b \sum X^{2}

]
Normal Equations for Line of (X) on (Y) ((X = a' + b'Y)):
[

\sum X \sum X Y = N a^{'} + b^{'} \sum Y = a^{'} \sum Y + b^{'} \sum Y^{2}

]

8. THE TWO BIVARIATE REGRESSION EQUATIONS & COEFFICIENTS

In simple linear bivariate analysis, there are always two distinct regression lines:

1. Regression Equation of (Y) on (X)

Utilized to predict the most likely value of (Y) for a given known value of (X): [ Y - bar{Y} = b_{yx}(X - bar{X}) ] Where (b_{yx}) is the regression coefficient of (Y) on (X), representing the change in (Y) per unit change in (X).

2. Regression Equation of (X) on (Y)

Utilized to predict the most likely value of (X) for a given known value of (Y): [ X - bar{X} = b_{xy}(Y - bar{Y}) ] Where (b_{xy}) is the regression coefficient of (X) on (Y), representing the change in (X) per unit change in (Y).

Formulas for Regression Coefficients

The regression coefficients (b_{yx}) and (b_{xy}) can be computed using several equivalent formulations depending on the available data summary:

Method	Coefficient (b_{yx}) ((Y) on (X))	Coefficient (b_{xy}) ((X) on (Y))
From Correlation & Std Dev	(b_{yx} = r · (σ_y) / (σ_x))	(b_{xy} = r · (σ_x) / (σ_y))
From Actual Mean Deviations ((x = X - bar{X}, y = Y - bar{Y}))	(b_{yx} = (∑ xy) / (∑ x^2))	(b_{xy} = (∑ xy) / (∑ y^2))
From Assumed Means ((dx = X - A_x, dy = Y - A_y))	(b_{yx} = (N ∑ dx dy - (∑ dx)(∑ dy)) / (N ∑ dx^2 - (∑ dx)^2))	(b_{xy} = (N ∑ dx dy - (∑ dx)(∑ dy)) / (N ∑ dy^2 - (∑ dy)^2))
From Raw Scores	(b_{yx} = (N ∑ XY - (∑ X)(∑ Y)) / (N ∑ X^2 - (∑ X)^2))	(b_{xy} = (N ∑ XY - (∑ X)(∑ Y)) / (N ∑ Y^2 - (∑ Y)^2))

Fundamental Mathematical Properties of Regression Coefficients

The regression coefficients (b_{yx}) and (b_{xy}) exhibit critical mathematical relationships:

1. Common Sign Property

Both regression coefficients ((b_{yx}) and (b_{xy})) and the correlation coefficient ((r)) must have the identical algebraic sign. It is mathematically impossible for one regression coefficient to be positive and the other negative.

2. Geometric Mean Property

The correlation coefficient is the geometric mean between the two regression coefficients: [ r = ± √(b_{yx) · b_{xy}}

The sign of (r) is positive if both coefficients are positive, and negative if both coefficients are negative.

3. Complementary Magnitude

If one of the regression coefficients is strictly greater than unity ((|b_{yx}| > 1)), the other regression coefficient must be less than unity ((|b_{xy}| < 1)), since

b_{y x} \cdot b_{x y} = r^{2} \leq 1.0

. Both cannot exceed 1.

4. Arithmetic Mean Bound

The arithmetic mean of the two regression coefficients is always greater than or equal to the correlation coefficient:

\frac{b _{y x} + b _{x y}}{2} \geq r

Equality holds if and only if

b_{y x} = b_{x y} = r = \pm 1

5. Point of Intersection

The two regression lines of (Y) on (X) and (X) on (Y) always intersect precisely at the point of their arithmetic means:

(ar{X}, ar{Y}) ] Solving the two regression equations simultaneously yields the means of both variables.

6. Origin vs. Scale Invariance

Regression coefficients are independent of change of origin, but dependent on change of scale. If (u = (X - a) / (h)) and (v = (Y - b) / (k)), then (b_{yx} = (k) / (h) · b_{vu}) and (b_{xy} = (h) / (k) · b_{uv}).

9. ANGLE BETWEEN REGRESSION LINES & STANDARD ERROR OF ESTIMATE

The acute angle (theta) formed between the two regression lines reflects the degree of correlation between the variables. The formula for the tangent of this angle is:

Angular Divergence Formula:
[ tan theta = ≤ft| (1 - r^2) / (r) ≤ft( (σ_x · σ_y) / (σ_x^2 + σ_y^2) right) right|

Two Limiting Boundary Conditions:
• Case 1: When $r = \pm 1.0$ (Perfect Correlation):
( an heta = 0 implies heta = 0^circ). The two regression lines coincide completely into a single straight line.
• Case 2: When (r = 0) (Complete Absence of Linear Correlation):

tan θ = \infty ⟹ θ = 9 0^{\circ}

. The two regression lines are perpendicular to each other, running parallel to the Cartesian axes (

Y = Y ˉ

and

X = X ˉ

Standard Error of Estimate ((S_{yx}) and (S_{xy}))

The Standard Error of Estimate measures the dispersion of actual observed values around the fitted regression line, providing an index of forecasting accuracy:

Standard Error of (Y) on (X) ((S_{yx})):

S_{y x} = \frac{\sum ( Y - Y ^ ) ^{2}}{N} = σ_{y} 1 - r^{2}

Standard Error of (X) on (Y) ((S_{xy})):

S_{x y} = \frac{\sum ( X - X ^ ) ^{2}}{N} = σ_{x} 1 - r^{2}

Interpretation: If

r = \pm 1

, (S_{yx} = 0), meaning all observed values lie directly on the regression line with zero prediction error. As (|r| → 0),

S_{y x} \to σ_{y}

, meaning the regression equation provides no improvement in prediction over simply using the arithmetic mean

Y ˉ

10. STEP-BY-STEP SOLVED NUMERICAL ILLUSTRATIONS

ILLUSTRATION 1: Karl Pearson's Correlation Coefficient, Probable Error & Determination
A manufacturing firm tracked its advertising expenditure ((X), in lakhs of ₹) and resulting sales revenue ((Y), in crores of ₹) over 8 consecutive quarterly operating periods:

Data:
Advertising Expenditure ((X)): 12, 14, 15, 18, 20, 22, 24, 27
Sales Revenue ((Y)): 20, 22, 24, 28, 32, 35, 38, 41

Required:
1. Calculate Karl Pearson's Coefficient of Correlation ((r)) using the direct mean deviation method.
2. Compute the Standard Error ((S.E._r)) and Probable Error ((P.E._r)) of the correlation coefficient.
3. Test the statistical significance of the correlation coefficient.
4. Calculate the Coefficient of Determination ((r^2)) and interpret the findings.

Solution & Step-by-Step Working Notes:

Step 1: Compute Arithmetic Means:

X ˉ = \frac{\sum X}{N} = \frac{12 + 14 + 15 + 18 + 20 + 22 + 24 + 27}{8} = \frac{152}{8} = 19.0

Y ˉ = \frac{\sum Y}{N} = \frac{20 + 22 + 24 + 28 + 32 + 35 + 38 + 41}{8} = \frac{240}{8} = 30.0

Both means are exact integers, so the direct mean deviation method is suitable.

Quarter	(X)	(x = X - 19)	(x^2)	(Y)	(y = Y - 30)	(y^2)	(xy)
1	12	-7	49	20	-10	100	+70
2	14	-5	25	22	-8	64	+40
3	15	-4	16	24	-6	36	+24
4	18	-1	1	28	-2	4	+2
5	20	+1	1	32	+2	4	+2
6	22	+3	9	35	+5	25	+15
7	24	+5	25	38	+8	64	+40
8	27	+8	64	41	+11	121	+88
Total	152	0	190	240	0	418	281

Step 2: Calculate Karl Pearson's Correlation Coefficient ((r)):

r = \frac{\sum x y}{\sum x ^{2} \cdot \sum y ^{2}} = \frac{281}{190 \times 418} = \frac{281}{79420} = \frac{281}{281.8155} = + 0.9971

There exists an exceptionally strong positive linear relationship between advertising expenditure and sales revenue.

Step 3: Standard Error and Probable Error:

S . E ._{r} = \frac{1 - r ^{2}}{N} = \frac{1 - ( 0.9971 ) ^{2}}{8} = \frac{1 - 0.9942}{2.8284} = \frac{0.0058}{2.8284} = 0.00205

P . E ._{r} = 0.6745 \times S . E ._{r} = 0.6745 \times 0.00205 = 0.00138

Step 4: Significance Test:

6 \times P . E ._{r} = 6 \times 0.00138 = 0.00828

Since

r = 0.9971 > 6 \times P . E ._{r}

((0.9971 > 0.00828)), the correlation is definitely statistically significant beyond any reasonable doubt.

Step 5: Coefficient of Determination ((r^2)):

r^2 = (0.9971)^2 = 0.9942 implies 99.42%

] Managerial Interpretation: Approximately 99.42% of the variance in quarterly sales revenue is directly explained by changes in advertising expenditure, with only 0.58% attributable to unexplained residual factors.

ILLUSTRATION 2: Spearman's Rank Correlation with Tied Ranks
Ten corporate acquisition targets were evaluated independently by two investment analysts (Senior Analyst (X) and Risk Officer (Y)). Their evaluations resulted in the following performance scores:

Data:
Target Company: A, B, C, D, E, F, G, H, I, J
Analyst (X) Scores: 68, 75, 90, 75, 50, 62, 80, 75, 88, 92
Risk Officer (Y) Scores: 60, 68, 85, 70, 45, 55, 80, 68, 82, 90

Required: Compute Spearman's Rank Correlation Coefficient, applying the correction factor for repeated values.

Solution & Step-by-Step Working Notes:

1. Ranking for Analyst (X):
Sorting (X) in descending order: 92 (Rank 1), 90 (Rank 2), 88 (Rank 3), 80 (Rank 4).
The value 75 appears 3 times (positions 5, 6, 7). Average rank = ((5 + 6 + 7) / (3) = 6.0). Here (m_1 = 3).
Next values: 68 (Rank 8), 62 (Rank 9), 50 (Rank 10).

2. Ranking for Risk Officer (Y):
Sorting (Y) in descending order: 90 (Rank 1), 85 (Rank 2), 82 (Rank 3), 80 (Rank 4).
Next: 70 (Rank 5).
The value 68 appears twice (positions 6 and 7). Average rank = ((6 + 7) / (2) = 6.5). Here (m_2 = 2).
Next values: 60 (Rank 8), 55 (Rank 9), 45 (Rank 10).

Target	(X) Score	Rank (R_1)	(Y) Score	Rank (R_2)	(D = R_1 - R_2)	(D^2)
A	68	8.0	60	8.0	0.0	0.00
B	75	6.0	68	6.5	-0.5	0.25
C	90	2.0	85	2.0	0.0	0.00
D	75	6.0	70	5.0	+1.0	1.00
E	50	10.0	45	10.0	0.0	0.00
F	62	9.0	55	9.0	0.0	0.00
G	80	4.0	80	4.0	0.0	0.00
H	75	6.0	68	6.5	-0.5	0.25
I	88	3.0	82	3.0	0.0	0.00
J	92	1.0	90	1.0	0.0	0.00
Total	-	-	-	-	(∑ D = 0)	(∑ D^2 = 1.50)

3. Compute Correction Factors:
For (X) series: Value 75 repeated (m_1 = 3) times: [ CF_1 = (m_1(m_1^2 - 1)) / (12) = (3(3^2 - 1)) / (12) = (3 × 8) / (12) = 2.0

For (Y) series: Value 68 repeated (m_2 = 2) times:

C F_{2} = \frac{m _{2} ( m _{2}^{2} - 1 )}{12} = \frac{2 ( 2 ^{2} - 1 )}{12} = \frac{2 \times 3}{12} = 0.5

Total Correction Factor = (CF_1 + CF_2 = 2.0 + 0.5 = 2.5).

4. Calculate Corrected Spearman's Rank Correlation:

R = 1 - rac{6 left[ sum D^2 + sum ext{CF} ight]}{N(N^2 - 1)}

}{N(N^2 - 1)} = 1 - (6 [1.50 + 2.50]) / (10(100 - 1)) = 1 - (6 × 4.0) / (10 × 99) = 1 - (24) / (990) = 1 - 0.02424 = +0.9758 ] Conclusion: The rank correlation coefficient of +0.9758 demonstrates remarkable consensus and alignment between the Senior Analyst and the Risk Officer in rating candidate firms.

ILLUSTRATION 3: Comprehensive Two-Way Bivariate Regression & Forecasting
An industrial plant operations director examined the relationship between factory capacity utilization rate ((X), in %) and total factory monthly maintenance expenditure ((Y), in thousands of ₹) across 10 operating months:

Data:
Capacity Utilization ((X)): 60, 62, 65, 70, 72, 75, 78, 80, 82, 86
Maintenance Expenditure ((Y)): 25, 28, 30, 36, 38, 40, 42, 45, 48, 52

Required:
1. Calculate the arithmetic means

X ˉ

and

Y ˉ

, and standard deviations

σ_{x}

and

σ_{y}

.
2. Compute the two regression coefficients (b_{yx}) and (b_{xy}), and deduce the correlation coefficient (r).
3. Construct the regression equation of (Y) on (X) and forecast maintenance expenditure for a projected capacity utilization rate of 85%.
4. Construct the regression equation of (X) on (Y) and estimate capacity utilization if maintenance budget is restricted to ₹ 50,000 ((Y = 50)).
5. Compute the Standard Error of Estimate (S_{yx}).

Solution & Step-by-Step Working Notes:

Let Assumed Mean for (X) be (A_x = 72), and for (Y) be (A_y = 38).
Then (dx = X - 72) and (dy = Y - 38).

Month	(X)	(dx = X - 72)	(dx^2)	(Y)	(dy = Y - 38)	(dy^2)	$d x \cdot d y$
1	60	-12	144	25	-13	169	+156
2	62	-10	100	28	-10	100	+100
3	65	-7	49	30	-8	64	+56
4	70	-2	4	36	-2	4	+4
5	72	0	0	38	0	0	0
6	75	+3	9	40	+2	4	+6
7	78	+6	36	42	+4	16	+24
8	80	+8	64	45	+7	49	+56
9	82	+10	100	48	+10	100	+100
10	86	+14	196	52	+14	196	+196
Total	730	+10	702	384	+4	702	698

Step 1: Compute Means and Standard Deviations:

X ˉ = A_{x} + \frac{\sum d x}{N} = 72 + \frac{10}{10} = 73.0%

Y ˉ = A_{y} + \frac{\sum d y}{N} = 38 + \frac{4}{10} = 38.4 (Rs. 38,400)

σ_{x} = \frac{\sum d x ^{2}}{N} - (\frac{\sum d x}{N})^{2} = \frac{702}{10} - (1.0)^{2} = 70.2 - 1.0 = 69.2 = 8.3187

σ_{y} = \frac{\sum d y ^{2}}{N} - (\frac{\sum d y}{N})^{2} = \frac{702}{10} - (0.4)^{2} = 70.2 - 0.16 = 70.04 = 8.3690

Step 2: Compute Regression Coefficients (b_{yx}) and (b_{xy}):

b_{y x} = \frac{N \sum d x d y - ( \sum d x ) ( \sum d y )}{N \sum d x ^{2} - ( \sum d x ) ^{2}} = \frac{10 ( 698 ) - ( 10 ) ( 4 )}{10 ( 702 ) - ( 10 ) ^{2}} = \frac{6980 - 40}{7020 - 100} = \frac{6940}{6920} = + 1.00289 \approx + 1.003

b_{x y} = \frac{N \sum d x d y - ( \sum d x ) ( \sum d y )}{N \sum d y ^{2} - ( \sum d y ) ^{2}} = \frac{10 ( 698 ) - ( 10 ) ( 4 )}{10 ( 702 ) - ( 4 ) ^{2}} = \frac{6980 - 40}{7020 - 16} = \frac{6940}{7004} = + 0.99086 \approx + 0.991

Deduce Correlation Coefficient ((r)):

r = b_{y x} \cdot b_{x y} = 1.00289 \times 0.99086 = 0.99372 = + 0.9969

Validation: Both regression coefficients are positive, their product (r^2 = 0.9937 < 1.0), and (b_{yx} > 1.0 implies b_{xy} < 1.0), satisfying all fundamental mathematical properties!

Step 3: Regression Equation of (Y) on (X) (Predicting Maintenance Cost):

Y - Y ˉ = b_{y x} (X - X ˉ) ⟹ Y - 38.4 = 1.00289 (X - 73.0)] [Y = 1.00289 X - (1.00289 \times 73.0) + 38.4 = 1.00289 X - 73.211 + 38.4

Y = - 34.811 + 1.00289X] < s t r o n g > F or ec a s t in g f or (X = 85%) C a p a c i t y U t i l i z a t i o n :< / s t r o n g > [Y = - 34.811 + 1.00289 (85) = - 34.811 + 85.246 = 50.435] E s t ima t e d m o n t h l y main t e nan cecos t a t 85% c a p a c i t y u t i l i z a t i o ni s < s t r o n g > Rs. 50, 435 < / s t r o n g > . < / p >< p >< s t r o n g > S t e p 4 : R e g r ess i o n E q u a t i o n o f (X) o n (Y) (E s t ima t in g P er mi ss ib l e C a p a c i t y) :< / s t r o n g > [X - X ˉ = b_{x y} (Y - Y ˉ) ⟹ X - 73.0 = 0.99086 (Y - 38.4)] [X = 0.99086 Y - (0.99086 \times 38.4) + 73.0 = 0.99086 Y - 38.049 + 73.0

X = 34.951 + 0.99086Y] < s t r o n g > F or ec a s t in g C a p a c i t y f or B u d g e t L imi t (Y = 50) (Rs. 50, 000) :< / s t r o n g > [X = 34.951 + 0.99086 (50) = 34.951 + 49.543 = 84.494% \approx 84.5%

The plant can operate at approximately 84.5% capacity without exceeding the ₹ 50,000 maintenance budget.

Step 5: Standard Error of Estimate ((S_{yx})):

S_{y x} = σ_{y} 1 - r^{2} = 8.3690 \times 1 - 0.99372 = 8.3690 \times 0.00628 = 8.3690 \times 0.07925 = 0.663 (Rs. 663)

The standard error of estimate is merely ₹ 663, verifying exceptional predictive precision.

11. COMPREHENSIVE SELF-ASSESSMENT MODULE

SECTION A: CONCEPTUAL & QUANTITATIVE MCQS

Select the most appropriate option. Complete solution key with mathematical rationale follows.

1. If Karl Pearson's correlation coefficient (r = +0.80), the percentage of variation in the dependent variable left unexplained by the independent variable is:
(a) 80% (b) 20% (c) 64% (d) 36%

2. If the two regression coefficients are (b_{yx} = -1.6) and (b_{xy} = -0.4), the value of the correlation coefficient (r) is:
(a) +0.80 (b) -0.80 (c) -0.64 (d) +0.64

3. Which of the following statements regarding regression coefficients is mathematically FALSE?
(a) Both regression coefficients must share the same algebraic sign.
(b) If (b_{yx} > 1), then (b_{xy}) must be less than 1.
(c) Regression coefficients are independent of change of scale.
(d) The arithmetic mean of (b_{yx}) and (b_{xy}) is greater than or equal to (r).

4. If the acute angle between the two regression lines is ( heta = 0^circ), the coefficient of correlation (r) is equal to:
(a) 0 (b) $\pm 0.50$ (c) $\pm 1.0$ (d) $\infty$

5. The two lines of regression intersect at the point defined by:
(a) ((0, 0)) (b) ( $σ_{x}, σ_{y}$ ) (c) ( $X ˉ, Y ˉ$ ) (d) ((b_{yx}, b_{xy}))

7. If in a tied rank correlation problem, a score repeats 4 times, the correction factor ((m(m^2 - 1)) / (12)) added to $\sum D^{2}$ is:
(a) 4 (b) 5 (c) 16 (d) 2.5

8. If (u = 2X - 5) and (v = -3Y + 4), and the correlation between (X) and (Y) is (r_{xy} = 0.75), the correlation between (u) and (v) is:
(a) +0.75 (b) -0.75 (c) +1.50 (d) -0.50

9. The Standard Error of Estimate (S_{yx}) becomes equal to the standard deviation $σ_{y}$ when:
(a) (r = 1.0) (b) (r = -1.0) (c) (r = 0) (d) (b_{yx} = 1.0)

10. In the Method of Concurrent Deviations, if (N = 11) and the number of concurrent deviations (c = 2), then ((2c - n)) evaluates to:
(a) -6 (b) +6 (c) -7 (d) +4

Answers & Conceptual Rationale for Section A:
1. (d) 36%: (r^2 = (0.80)^2 = 0.64) (explained variance). Unexplained variance is (1 - r^2 = 1 - 0.64 = 0.36 = 36%).
2. (b) -0.80: (r = -√((-1.6) × (-0.4)) = -√(0.64) = -0.80). The sign of (r) must match the signs of the coefficients.
3. (c): Regression coefficients are dependent on change of scale (

b_{y x} = \frac{k}{h} b_{v u}

); only correlation is independent of scale.
4. (c) $\pm 1.0$ : When

r = \pm 1

, ( an heta = 0 implies heta = 0^circ), meaning both lines coincide.
5. (c) ( $X ˉ, Y ˉ$ ): The lines (Y - Ȳ = b_{yx}

X - X ˉ

) and (X - X̄ = b_{xy}

Y - Y ˉ

) simultaneously satisfy

X = X ˉ, Y = Y ˉ

.
6. (c) $r > 6 \times P . E ._{r}$ : By Horace Secrist and classical statistical doctrine, significance is assured when (r) exceeds 6 times its probable error.
7. (b) 5: ((4(4^2 - 1)) / (12) = (4(15)) / (12) = (60) / (12) = 5.0).
8. (b) -0.75: The scale transformation on (Y) involves multiplying by (-3) (a negative scalar), while (X) is multiplied by (+2). An odd number of negative scale changes flips the sign of (r).
9. (c) (r = 0):

S_{y x} = σ_{y} 1 - 0^{2} = σ_{y}

. When no correlation exists, regression line prediction error equals total standard deviation.
10. (a) -6: (n = N - 1 = 11 - 1 = 10). Thus (2c - n = 2(2) - 10 = 4 - 10 = -6).

SECTION B: ANALYTICAL & THEORETICAL ESSAY QUESTIONS

Why are there two regression lines in bivariate analysis? Under what operational circumstances will the two regression lines coincide, and under what circumstances will they be mutually perpendicular? Explain with mathematical proofs.
Evaluate the properties of Regression Coefficients. Prove mathematically that the arithmetic mean of the two regression coefficients is always greater than or equal to the geometric mean (correlation coefficient).
Contrast Karl Pearson's Correlation with Spearman's Rank Correlation. Delineate specific commercial and behavioral circumstances where Spearman's rank method is statistically superior to product-moment correlation.
Define the Coefficient of Determination ((r^2)). How does it distinguish genuine explanatory power from simple co-variation? Explain its role in multiple regression analysis.
Explain the role of Probable Error in Statistical Inference. State the precise conditions under which Probable Error can be legitimately employed to assess the reliability of a sample correlation coefficient.

SECTION C: ADVANCED QUANTITATIVE PROBLEMS

Problem 1: Bivariate Sales & Production Regression Analysis
A consumer goods corporation records the following summary statistics from 100 operating branch stores across South India:
• Mean Annual Branch Promotion Outlay ((X)) = ₹ 50,000, Standard Deviation ( $σ_{x}$ ) = ₹ 10,000
• Mean Annual Branch Sales Revenue ((Y)) = ₹ 400,000, Standard Deviation ( $σ_{y}$ ) = ₹ 50,000
• Karl Pearson's Correlation Coefficient between Promotion and Sales ((r)) = +0.80

Required:
1. Compute the two regression coefficients (b_{yx}) and (b_{xy}).
2. Construct the regression equation of Sales ((Y)) on Promotion Outlay ((X)).
3. Forecast the expected sales revenue of a branch that spends ₹ 70,000 on promotion.
4. Construct the regression equation of Promotion Outlay ((X)) on Sales ((Y)) and determine the promotion expenditure recommended to achieve a target branch sales revenue of ₹ 550,000.
5. Calculate the Standard Error of Estimate (S_{yx}) and construct a 95% confidence interval ( $\pm 1.96 S_{y x}$ ) for the sales forecast in (3).

Problem 2: Spearman's Rank Correlation with Severe Ties
Two market research firms ranked 8 competing mobile smartphone brands based on customer satisfaction metrics. The assigned scores were as follows:
Brand: A, B, C, D, E, F, G, H
Agency 1: 85, 70, 70, 90, 60, 70, 85, 55
Agency 2: 78, 65, 72, 85, 50, 65, 78, 48

Required:
1. Assign average fractional ranks to both agencies' scores.
2. Identify the tied groups and compute the individual and aggregate correction factors.
3. Calculate the corrected Spearman's Rank Correlation Coefficient ((R)) and comment on the degree of agreement between the two rating agencies.

MODULE II: CORRELATION AND REGRESSION ANALYSIS

1. FOUNDATIONAL CONCEPTS OF BIVARIATE RELATIONSHIPS

Concept of Co-Variation

Causation vs. Covariation

Taxonomy of Correlation

2. SCATTER DIAGRAM METHOD

1. Perfect Positive Correlation ((r = +1.0))

2. Perfect Negative Correlation ((r = -1.0))

3. High Degree Positive / Negative

4. Zero / Absence of Correlation ((r = 0))

3. KARL PEARSON'S COEFFICIENT OF CORRELATION ((r))

Alternative Operational Formulas for Computation

Crucial Mathematical Properties of Pearson's (r)

Property 1: Strictly Bounded Limits

Property 2: Invariance to Origin and Scale

Property 3: Symmetry of Relationship

4. STATISTICAL SIGNIFICANCE: PROBABLE ERROR & DETERMINATION

Standard Error and Probable Error of (r)

Rules for Interpreting Correlation via Probable Error

Coefficient of Determination ((r^2)) and Alienation

5. SPEARMAN'S RANK CORRELATION COEFFICIENT ((R) OR ( ho))

Handling Tied / Repeated Ranks (Correction Factor)

6. CONCURRENT DEVIATION METHOD

7. REGRESSION ANALYSIS: PRINCIPLES & LEAST SQUARES ARCHITECTURE

Fundamental Distinction: Correlation vs. Regression

Principle of Ordinary Least Squares (OLS)

8. THE TWO BIVARIATE REGRESSION EQUATIONS & COEFFICIENTS

1. Regression Equation of (Y) on (X)

2. Regression Equation of (X) on (Y)

Formulas for Regression Coefficients

Fundamental Mathematical Properties of Regression Coefficients

1. Common Sign Property

2. Geometric Mean Property

3. Complementary Magnitude

4. Arithmetic Mean Bound

5. Point of Intersection

6. Origin vs. Scale Invariance

9. ANGLE BETWEEN REGRESSION LINES & STANDARD ERROR OF ESTIMATE

Standard Error of Estimate ((S_{yx}) and (S_{xy}))

10. STEP-BY-STEP SOLVED NUMERICAL ILLUSTRATIONS

11. COMPREHENSIVE SELF-ASSESSMENT MODULE

Finished reading Module 2?