Omitted Variable Bias (OVB) occurs in statistical models when a relevant variable is left out, and this omission correlates with both the dependent variable and at least one included independent variable. This can lead to biased and inconsistent estimates. Below are some common examples to illustrate the concept:
Contents
1. Education and Earnings
- Model: Earnings=β0+β1Education+ϵ\text{Earnings} = \beta_0 + \beta_1 \text{Education} + \epsilon
- Omitted Variable: Innate ability or motivation
- Explanation: People with higher innate ability or motivation may achieve both higher education and earnings. If ability is omitted, the effect of education on earnings will likely be overstated.
2. Health and Income
- Model: Health=β0+β1Income+ϵ\text{Health} = \beta_0 + \beta_1 \text{Income} + \epsilon
- Omitted Variable: Access to healthcare or lifestyle choices
- Explanation: Access to better healthcare and healthier lifestyles, often associated with higher income, may drive better health outcomes. Omitting these factors may bias the income coefficient.
3. Housing Prices and School Quality
- Model: Housing Price=β0+β1School Quality+ϵ\text{Housing Price} = \beta_0 + \beta_1 \text{School Quality} + \epsilon
- Omitted Variable: Neighborhood amenities (e.g., parks, crime rates)
- Explanation: Neighborhood characteristics, which influence housing prices, are often correlated with school quality. Omitting them inflates the effect of school quality on housing prices.
4. Advertising and Sales
- Model: Sales=β0+β1Advertising+ϵ\text{Sales} = \beta_0 + \beta_1 \text{Advertising} + \epsilon
- Omitted Variable: Product quality
- Explanation: High-quality products may naturally sell better and receive more advertising. If quality is omitted, advertising may seem more effective than it is.
5. Crime Rates and Police Presence
- Model: Crime Rate=β0+β1Police Presence+ϵ\text{Crime Rate} = \beta_0 + \beta_1 \text{Police Presence} + \epsilon
- Omitted Variable: Economic conditions
- Explanation: Poor economic conditions may increase crime and lead to more police deployment. Ignoring these conditions can lead to biased estimates of the relationship between crime rates and police presence.
6. Weight Loss Programs and Weight Loss
- Model: Weight Loss=β0+β1Program Participation+ϵ\text{Weight Loss} = \beta_0 + \beta_1 \text{Program Participation} + \epsilon
- Omitted Variable: Participant motivation or initial weight
- Explanation: People who are more motivated to lose weight or have more weight to lose may be more likely to join programs. Omitting these factors inflates the program’s effectiveness.
Key Takeaways to Avoid OVB
- Include Relevant Variables: Use domain knowledge to identify and include important predictors.
- Use Instrumental Variables: If a key variable is hard to measure, find an instrument that is correlated with it but not with the error term.
- Conduct Robustness Checks: Experiment with different model specifications to ensure consistent results.
