Unit 5 : Multiple Linear Regression Model Flashcards Preview

Econometrics 1 > Unit 5 : Multiple Linear Regression Model > Flashcards

Flashcards in Unit 5 : Multiple Linear Regression Model Deck (7)
Loading flashcards...

What is a multiple linear regression model?

This is where there is more than one explanatory variable. Therefore, we have to account for numerous independent variables.


What is adjusted R squared?

When adding other regressors to the equation, we need to account for their influence on Y. Some regressors will be useless, which decreases the R2 value whereas the useful regressors will increase the R2 value.


What is omitted variable bias?

This is when you do not account for a relevant variable that affects an outcome, leading to a bias in the OLS estimators.

This is true when the omitted variable :

- directly affects the outcome Y
- is correlated with another regressor X


What is asymptomatic normality?

This is a property that an estimator may possess if it produces a normal distribution as the sample size gets larger.


What does the direction of bias mean?

When there is OMV, the direction of bias will tell us if there is an overestimation (upwards bias) or underestimation (downwards bias) of results when that specific variable was left out of the regression model.

If B1< B1*, then we say that there has been a downwards bias, with B1* being the new regressor with the additional variable added.

If the new variable in the model is found to be less than 0, we say that there is a negative relationship between that variable and Y the outcome.


What is multicollinearity?

This is when the regressors in the model portray correlation - this is not ideal as these are supposed to be independent variables.

If multicollinearity exists, the coefficients often vary significantly in size and also reduces the precision of estimates.

Perfect collinearity exists when one regressor affects another by a constant amount. This leads to an oddity in the results.


What is a dummy variable and how can we get trapped using it?

A dummy variable (indicator variables) are qualitative - they just represent the presence (or no presence) of something, i.e. 0 = 5 teachers, 1 = 10 teachers.

If we make the dummy variables dependent on each other, then that could lead to perfect collinearity, leading to a 'dummy variable trap'.
The solution to this would be to drop one of the variables and indicate the next dummy variable as (t-10.