Answer-first
Multicollinearity inflates variance because highly correlated regressors make it difficult for OLS to isolate each variable’s unique contribution. When predictors move together, the model cannot distinguish their separate effects, causing the slope estimates to become unstable and their standard errors to explode.
Warm intro (and where to find the “why” pages)
If you’re staring at a multicollinearity question at 11:47pm, feeling stuck, behind, or low‑key panicking because the “correlated regressors” explanation feels vague, you’re not alone. These questions look simple—“VIF > 10”—but under exam pressure, students freeze when asked to explain why multicollinearity inflates variance and what actually breaks inside the OLS formula.
If you’re rebuilding your foundation across topics, start at our Why Hub. If you need the full econometrics & time series roadmap for last-minute studying or troubleshooting, see Econometrics & Time Series.
Problem setup
Consider the multiple regression model:
\[ y_i = \beta_0 + \beta_1 x_{1i} + \beta_2 x_{2i} + u_i. \]
Multicollinearity occurs when:
\[ \text{Corr}(x_1, x_2) \approx \pm 1. \]
This makes the regressors nearly linearly dependent.
Step-by-step solution (WordPress-safe MathJax)
Step 1: Look at the OLS variance formula
\[ \text{Var}(\hat{\beta}) = \sigma^2 (X’X)^{-1}. \]
When regressors are highly correlated, the matrix \(X’X\) becomes nearly singular.
Step 2: Near-singularity makes the inverse explode
If \(X’X\) is close to non-invertible, its inverse contains very large values. These large values inflate the variance of the OLS estimates.
Step 3: Standard errors become huge
Large variance → large standard errors → insignificant t‑statistics.
Step 4: Coefficients become unstable
Small changes in the data produce large swings in the estimated coefficients.
Step 5: State the identifying assumption clearly
OLS requires that regressors are not perfectly collinear:
\[ \text{Rank}(X) = k+1. \]
Perfect collinearity makes OLS impossible; near-collinearity makes OLS unstable.
Intuition
Multicollinearity is like trying to separate the effect of height and arm length on basketball performance. They move together, so the model struggles to assign credit. The result: noisy, unstable estimates with huge standard errors.
Common exam mistakes
- Claiming multicollinearity biases OLS. It does not.
- Ignoring VIF diagnostics.
- Dropping variables without theoretical justification.
- Confusing multicollinearity with heteroskedasticity.
- Misinterpreting insignificant coefficients in the presence of high correlation.
Why this matters
Multicollinearity is common in real data—education & income, advertising channels, macroeconomic indicators. If you ignore it, your model becomes unstable, your inference collapses, and your coefficients become impossible to interpret.
Final summary
- Multicollinearity inflates variance, not bias.
- It destabilizes coefficients and standard errors.
- VIF and correlation matrices help diagnose it.
- Solutions include dropping variables, combining them, or using PCA/Ridge.
Talk Directly to a Tutor, Not a Marketer
Call/Text: 510-398-0006
Email: tutor@californiagraduatetutor.com