Multicollinearity and others

Summaries

Multicollinearity: some independent variable are correlated

This increases type II error

Usually, F value will be high but individual t-values are low. This is because while individual has not much effect on dependent variable, the combined (within which are correlated) has strong effect.

Systematically removing dependent variables can help to solve the multicollinearity issue

Regression model specification is the selection of explanatory variables to be included in the regression

Model misspecification => cannot have unbiased (expected value of the estimated equals the real value) and consistent estimators (error reduces as sample size increases) => unreliable tests

1. Important variables omitted
2. Variables not transformed
3. Data improperly pooled
4. Lagged dependent variable used as independent variable
5. forecasting the past
6. independent variables with errors

Qualitative Dependent Variables

Probit and Logit models

Discriminant Models