ERP Lecture 4

10 flashcards · Other · by Daniel F.

1. Violations of regression assumptions

   Multicollinearity Heteroskedasticity Reverse causality Omitted variable bias

2. Multicollinearity

   High correlation between at least two independent variables if to IVs correlate, we cant hold other variables constant the stronger the correlation the more problematic it is e.g. variables that capture similar things or lagged values of the same thing Result Hard to identify the individual impact of each x standard errors a larger, collinear IV can become significant Cannot trust coefficient results perfect correlation: dummy variable trap, x and 2x, m and cm of the same thing

3. How to detect a multicollinearity

   Use a correlation matrix of for all independent variables Correlation coefficients ≥ |0.70| (or 0.50) signal multicollinearity Variance inflation factor (VIF) shows the percentage the variance, inflated for each coefficient (1-inf) e.g., VIF of 1.7=> variance is 70% bigger in the data, compared to no multicollinearit VIF=1 no multicollinearity VIF between 1, 5 = not severe and not need to pay spenial attention VIF > 5 cautious or VIF > 10 liberal = Severe multicollinearity that affets estimates

4. Dealing with multicollinearity

   Increase sample size Drop 1 of problem variables use one for robustness check Transform the highly correlated IVs log transform combine IVs

5. Heteroscedasticity and homoscedasticity

   Homo Variance of the error term is constant over various values of the IVs Error has the same variance regardless of the values of IV Dispersion of the error remains the same over the range of observations Hetero The error term does not have a constant variance Variance changes in response to a change in the value of IVʼs Dispersion of the error changes over the range of observations e.g. low income people spend constant, but higher income people spend with more variation Result Biased standard…

6. Check for heteroschedasticity

   Breusch-Pagan test White test Plot residuals vs predicted values

7. Dealing with heteroscedasticity

   Transform the dependent - does not always work Use weighted regression - Observations with a higher variance gets lower weight and vv helps prevent high-variance observations from disproportionately influencing the regression coefficients Use ʻrobustʼ standard errors - White-corrected

8. Endogeneity

   one or more independent variables (IVs) are correlated with the error term (unobserved factors affecting the dependent variable). a. Omitted Variable Bias an excluded variable has some effect on your DV The omitted variable becomes part of the error term, causing the IV to correlate with the error b. Simultaneity (Reverse Causality) Y and an IV influence each other. For example: Price affects demand, and demand affects price. Here, price (an IV) correlates with the error term in the demand equa…

9. How to deal with Endogeneity

   Reverse Causality model grounded in theory Explained mechanism, reasoning Acknowledge possible endogeneity issues use lagged IVs, when suitable instrumental variables Omitted Variable Bias Include variables that are likely to be the most important theoretically in explaining the DV Simultaneity instrumental variables

10. Mean for dummy variables

    represents the proportion of observations with that characteristic