Nature & Detection of Multicollinearity

ECON 3209 · Week 11, Lecture 1 · Kerala Agricultural University

Department of Development Economics, KAU

Autumn 2026

Learning Outcomes

By the end of this lecture, you will be able to:

1. define multicollinearity and distinguish perfect from imperfect collinearity
2. explain why highly correlated regressors are common in applied economics
3. use correlation matrices and scatter plots to diagnose collinearity
4. recognize singular design matrices created by exact linear relationships
5. interpret detection evidence in a Kerala agricultural context

What Is Multicollinearity?

Multicollinearity means two or more regressors contain overlapping information.

Perfect multicollinearity: one regressor is an exact linear combination of others.

Imperfect multicollinearity: regressors are highly, but not perfectly, correlated.

OLS can tolerate imperfect multicollinearity, but not perfect multicollinearity.

Perfect vs Imperfect Collinearity

Type	Example	Consequence
Perfect	total_cost = fixed_cost + variable_cost and all three are included	\(X'X\) is singular; OLS cannot be computed
Perfect	all crop dummies plus an intercept	dummy variable trap
Imperfect	fertilizer and irrigation move together	OLS runs, but precision worsens

Imperfect multicollinearity is a data problem of overlap, not a coding error.

Why Does It Arise in Kerala Data?

• Farmers with better irrigation often also apply more fertilizer.
• Larger landholders may have both higher income and larger loan sizes.
• Cooperative members in richer panchayats may have similar access to credit, extension, and infrastructure.

In observational data, regressors often move together because real-world decisions are interconnected.

First Detection Tools

1. Economic intuition: do the regressors measure similar concepts?
2. Correlation matrix: are pairwise correlations very high?
3. Scatter plots: do points lie close to a line?
4. Regression output: do signs or standard errors look unstable?

Pairwise correlation is useful, but multicollinearity can also involve more than two variables at once.

Python Demo: Simulate Correlated Regressors

Python Demo: Scatter Plot for Detection

Python Demo: Perfect Multicollinearity Example

If rank is smaller than the number of columns, the design matrix is singular.

Correlation Matrix: Useful but Not Sufficient

• A very high pairwise correlation is a strong warning sign.
• But low pairwise correlations do not guarantee safety.
• One regressor may be explained well by a combination of several other regressors.

This is why later we use the Variance Inflation Factor: \[\text{VIF}_j = \frac{1}{1-R_j^2}\]

Python Demo: Condition of the Design Matrix

A very large condition number suggests severe numerical dependence among regressors.

Detection Checklist for Practice

• Start with subject-matter logic.
• Inspect descriptive statistics and pair plots.
• Check for exact duplicate information or dummy traps.
• Review software warnings about singular matrices.
• Use VIFs for a more formal diagnostic.

Detection is about combining theory + graphics + statistics, not relying on a single number.

🏋️ Exercise

1. Create three regressors where two are strongly correlated.
2. Compute the correlation matrix and make a scatter plot.
3. Add a perfectly collinear variable such as 2 * fertilizer.
4. Check the rank of the design matrix and explain the result.

Summary

✅ Multicollinearity means regressors overlap in the information they carry.

✅ Perfect collinearity makes OLS impossible because \(X'X\) is not invertible.

✅ Imperfect collinearity is common in real economic data and mainly hurts precision.

✅ Correlation matrices, scatter plots, matrix rank, and condition numbers are practical early diagnostics.

✅ Theory should guide diagnosis: related economic variables often move together for real reasons.

Next Lecture

Lecture 2 — Consequences of Multicollinearity

We will cover: - inflated standard errors - unstable coefficients - high R² with weak t-statistics - why prediction can still remain acceptable