moving from Simple Linear Regression Models to Multiple Linear Regression Models opens an almost unlimited opportunity for us to improve prediction by adding more and more predictor variables 𝑿 into our model. On the other hand, overfitting a model leads to a low prediction power. Moreover, it will often result in large variances 𝜎²(𝑏_𝑖) and therefore, unstable regression estimates.

Then, how can we build a model with the right, optimal set of predictors ⊆𝑿 that will give us a good, accurate fit?

Methods

• LR - Comparing 2 Models (Extra Sum of Squares & Partial F-Test Statistic)
• LR - Predictor Variable Selection (Stepwise Selection & Backward Elimination)
• LR - Ramsey RESET Test for Functional Misspecification