Correlation

The full picture

When you look at economic data, you often wonder: do two variables move together? Does more rainfall mean more rice output? Does higher household income lead to more spending? Correlation is the statistical tool that answers this question. It measures the strength and direction of a relationship between two variables. Notice that correlation only describes a pattern — it does not tell you that one variable causes the other. For example, both umbrella sales and cold and flu cases rise in the monsoon season, but umbrellas do not cause illness. Keeping the correlation-versus-causation distinction clear is the single most important habit in statistics.

The first tool you will use is the scatter diagram. Plot one variable on the X-axis and the other on the Y-axis, one point per observation. The shape that emerges tells you everything before you calculate a single number. When points slope upward from left to right, you have positive correlation — both variables rise together. When points slope downward, you have negative correlation — one rises as the other falls. When points are scattered randomly with no pattern, there is no (or very weak) correlation. Always draw a scatter diagram first; it catches non-linear curves and stray outliers that any single number can hide.

Karl Pearson's correlation coefficient, written as r, gives you a precise number. It ranges from −1 to +1. A value of +1 means a perfect positive linear relationship; −1 means a perfect negative linear relationship; 0 means no linear relationship. The formula divides the covariance (how much the two variables vary together) by the product of their standard deviations. In practice, r above 0.7 (or below −0.7) is considered strong, between 0.3 and 0.7 (or −0.3 and −0.7) is moderate, and below 0.3 is weak. Pearson's method requires both variables to be measured on a continuous scale — such as income in rupees or weight in kilograms — and the relationship should be roughly linear, which is exactly why you draw the scatter diagram first.

Spearman's rank correlation coefficient solves a real problem. When your data are ordinal — ranked positions rather than precise measurements — or when a few extreme outliers are pulling Pearson's r away from the true picture, Spearman is the better choice. You convert each variable's raw values into ranks (1st, 2nd, 3rd…) and then apply a simple formula using the difference in ranks (written as D). Because Spearman works with positions rather than actual values, an extreme outlier occupies the same rank boundary (first or last) no matter how extreme its measured value is, leaving the overall coefficient stable. This makes Spearman ideal for comparing students by exam rank and attendance rank, or judging cities by quality-of-life score.

Here is how both methods work together in practice. Suppose a school teacher in Kochi records the height in centimetres and weight in kilograms of 10 students. She first plots a scatter diagram — the points slope upward, suggesting positive correlation. She calculates Pearson's r = 0.88, confirming strong positive linear correlation: taller students tend to weigh more. Now she also ranks students by height and by weight separately, then applies Spearman's formula. She gets r_s = 0.91, slightly higher because one very short but heavy student (an outlier) was dragging Pearson's r down slightly. Both coefficients together paint a reliable picture. She presents the scatter diagram alongside both numbers in her school project — this is exactly the approach NCERT expects.

An Indian example

Priya runs a small kirana store in Coimbatore. She notices her monthly cold-drink sales (in ₹ thousands) seem to rise and fall with local temperature. She records data for 8 months: in April (42°C) she sold ₹38,000; in June (28°C, monsoon) she sold ₹18,000; in December (20°C) she sold ₹12,000. She plots these on a scatter diagram — the points slope clearly upward from left to right, so she expects positive correlation. She calculates Pearson's r and gets 0.91 — strong positive correlation. Her supplier then offers a bulk discount if she orders in advance. Because of her correlation analysis, Priya confidently orders extra stock before March each year, cutting her per-unit cost by 8%. She does not assume temperature causes her sales to rise; she knows both are driven by the season. But the pattern is consistent enough to plan inventory around — and that is exactly what correlation is for.

Key concepts covered

• Karl Pearson
• Spearman's rank
• Scatter diagrams

Common misconceptions to watch for

• Misconception: 'If r is high, one variable must be causing the other to change.' Correction: Correlation only shows that two variables move together — it never proves causation. In India, both ice cream sales and heatstroke cases rise together every summer, but ice cream does not cause heatstroke; a common third factor (hot weather) drives both. Always ask whether a third variable could explain the pattern before drawing any causal conclusion.
• Misconception: 'Pearson's r works for any type of data, including ranked lists.' Correction: Pearson's r requires continuous, interval-or-ratio-scale data (like income in ₹ or height in cm) and a linear relationship. If you are comparing students' rank in Maths with their rank in Science — ordinal data — you should use Spearman's rank correlation. Using Pearson on rank numbers treats a difference from rank 1 to rank 2 as the same as a difference from rank 9 to rank 10, which is not meaningful in a ranked list.
• Misconception: 'An r value near zero means the two variables are completely unrelated.' Correction: Pearson's r only measures linear (straight-line) relationships. Two variables can have r ≈ 0 and still be strongly related through a curve. For example, electricity demand is high in both extreme summer (air conditioners) and extreme winter (heaters) but low in mild weather — a U-shape on a scatter diagram — yet Pearson's r for this pattern comes out near zero. Always draw the scatter diagram first to check the shape before trusting any single coefficient.