Correlation analysis measures the degree and direction of linear association between two or more economic variables, serving as a foundational quantitative tool in econometric modeling. Mastery of Pearson's, Spearman's, and regression relationships is vital for cracking competitive quantitative economics questions in UP PGT and UGC-NET examinations.
Introduced the foundational concept of co-relation and regression in biometrics, which was later adapted for economic data analysis.
Developed the product-moment correlation coefficient formula (r), establishing the parametric standard for linear association.
Formulated the non-parametric Rank Correlation Coefficient for ordinal data and psychological/economic preference rankings.
Proposed early non-parametric measures of co-variation based on signs of deviations, precursor to concurrent deviation methods.
Extended correlation theory to multiple and partial correlation, establishing the multivariate framework for economic forecasting.
Correlation quantifies the interdependence between economic variables. It is classified by direction into positive (direct movement, e.g., income and consumption) and negative (inverse movement, e.g., price and quantity demanded). The degree ranges from perfect (+1 or -1) to zero correlation (absence of linear relation), visualized through scatter diagrams.
The most widely used parametric measure of linear association, denoted by r. It is defined as the covariance of the two variables divided by the product of their standard deviations. It requires interval or ratio data and assumes underlying bivariate normality.
A non-parametric measure developed for ordinal data where actual numerical values are replaced by ranks. It is particularly useful when quantitative measurement is difficult (e.g., measuring utility, consumer preferences, or interview scores) and when data contains extreme outliers.
Correlation does not imply causation (spurious correlation). Lurking variables or common trends over time can create high mathematical correlation between entirely unrelated economic series (e.g., aggregate rainfall and stock market indices). Furthermore, non-linear relationships can yield a correlation coefficient near zero.
Measures the standardized covariance between two variables. Ranges from -1 to +1.
Non-parametric measure based on rank differences (d_i). Robust against outliers and non-normal distributions.
Proves that the sign of r must match the sign of both regression coefficients. The geometric mean of slopes.
Used for testing significance. If r < 6 \times P.E.(r), the correlation is considered statistically insignificant.
Indicates the proportion of total variation in Y that is accounted for by the linear regression on X.
Exam Application: If b_xy = 0.8 and b_yx = 0.2, instantly calculate r = \sqrt{0.8 \times 0.2} = \sqrt{0.16} = 0.4.
Exam Application: If exam options give regression coefficients with conflicting signs (e.g., +1.2 and -0.5), immediately eliminate that option.
Exam Application: Do not recalculate correlation if variables are transformed (e.g., X' = X - A / h). The correlation coefficient remains unchanged.
Exam Application: Quickly evaluate hypothesis testing questions in UGC-NET by checking if r exceeds 6 times the calculated P.E.
Conditions: Two completely unrelated time-series variables exhibit high correlation due to a shared common trend (e.g., time variable or inflation).
Diagrammatic Behavior: Scatter plot shows an apparent tight linear fit, but economic theory dictates no causal mechanism whatsoever.
Conditions: A trend appears in different groups of data but disappears or reverses when these groups are combined.
Diagrammatic Behavior: Sub-groups show positive slopes, but the aggregate pooled scatter plot displays a negative slope.
Conditions: Variables share a quadratic or U-shaped relationship (e.g., Phillips curve or Laffer curve over certain domains).
Diagrammatic Behavior: Pearson's linear correlation coefficient computes close to zero, completely missing the strong deterministic non-linear association.
| Parameter | Karl Pearson's Correlation (r) | Spearman's Rank Correlation (R_s) |
|---|---|---|
| Data Requirement | Interval or Ratio scale data | Ordinal or Ranked data |
| Underlying Distribution | Assumes bivariate normal distribution | Distribution-free (Non-parametric) |
| Sensitivity to Outliers | Highly sensitive; outliers distort value | Robust; uses relative ranks instead of absolute values |
| Computational Complexity | Computationally intensive for large datasets | Simpler computation based on rank differences |
Pearson requires quantitative metric data and assumes bivariate normality with linear relationships, whereas Spearman is non-parametric, relies on ordinal data (ranks), and can capture monotonic non-linear relationships.
Correlation measures the degree of linear association between two variables without implying cause-and-effect; causation proves that a change in one variable directly produces a change in the other.
Zero linear correlation means there is no linear association between variables, but a strong non-linear (quadratic, cubic) relationship may still exist. Statistical independence implies zero correlation, but zero correlation does not necessarily imply complete independence.
'r' indicates both the direction and strength of linear association ranging from -1 to +1, whereas '$r^2$' indicates the proportion of total variance explained by the regression model, ranging from 0 to 1.
Every fundamental term, curve, and mechanism covered in this chapter.
Select an answer to see instant feedback and complete analytical explanation.
Exam patterns from UP PGT Economics, UP TGT Social Science, and UGC NET.