\(\beta_2 \neq 0\) (the omitted variable matters), and
\(\text{Cov}(X_1, X_2) \neq 0\) (the omitted variable is correlated with the included regressor).
Direction of Bias
Effect of omitted variable on \(Y\)
Correlation with included regressor
Bias direction
Positive
Positive
Upward
Positive
Negative
Downward
Negative
Positive
Downward
Negative
Negative
Upward
Always think in two signs: the omitted effect and the correlation sign.
Short and Long Regressions in Python
Interpretation of the Comparison
The short regression mixes the effect of income with the effect of credit score if credit score is related to income.
The long regression separates these channels more clearly.
A big change in the income coefficient after adding credit score is a warning sign that OVB was important.
Proxy Variables
Sometimes the true omitted variable is not observed.
A proxy variable is an imperfect but related measure.
Example: if repayment discipline is unobserved, branch inspection score may capture part of it.
A proxy rarely removes all bias, but it can reduce it.
Using a Proxy in Python
Kerala Credit Score Example
Suppose we estimate loan default using only borrower income.
But if credit score is omitted and higher-income borrowers also tend to have stronger credit history, then the income coefficient partly captures the effect of credit quality.
That is why econometrics always asks: what important variable may still be missing?
A Simple Sensitivity Check
Practical Checklist
Start from economic theory, not only data availability.
Ask which omitted factors could affect both the dependent variable and a key regressor.
Compare short and long regressions.
Use better data, fixed effects, instruments, or proxies when possible.
Remember: OVB is about causal credibility, not only statistical significance.
Exercise
Using the starter code, compare the income coefficient across the short, proxy, and full regressions. Which coefficient would you report to a cooperative bank manager, and why?
Summary
✅ OVB appears when a relevant omitted variable both affects Y and is correlated with an included regressor.
✅ The OVB formula helps determine whether the bias is upward or downward.
✅ Comparing short and long regressions is a useful diagnostic step.
✅ Proxy variables may reduce bias, but theory and data quality remain crucial.
Next Lecture
We test model specification with the Ramsey RESET test.
You will see how fitted values squared and cubed can reveal functional-form problems.
Today’s message about misspecification continues in a formal testing framework.