Crash Testing Data

Crash Testing Data

1st Issue: Multicollinearity

High Correlation between Explanatory Variables

2nd Issue: Heteroskedasticity

NFL Quarterback Salary Example

Data can crash if assumptions are violated

Brief Regression Review of the b's

Do the coefficient signs meet intuition?

Are variables significant? (2-t Rule of Thumb)

Explore Potential Multicollinearity

Correlation Matrix (High if greater than 0.7)

Scatterplot

Decision Tree for Checking

High Collinearity Leads to High Variance

Test for High Variance with VIF

Consequences of High Variance in an Explanatory Variable

Remedy for Multicollinearity: Ridge Regression

Avoid Dropping Variable

Ridge does not Use OLS (OLS uses B.L.U.E.)

Ridge cares more about reducing Variance

What High Variance of Coefficients looks like
(unbiased but not-precise/"wild")

Low Variance of Coefficients with Ridge (more precise, but some bias)

Quarterback Data Example (Better Coefficients, Lower Variance)

All Rights Reserved by Brent Marinan & University of Arizona

eller icon

Correlation

regression review

signs meet intuition.

Variable significance

Violation of Assumptions

scatter plot & matrix

high variance

VIF

Omitted Variable Bias

BLUE

Ridges concern

look of high variance

Ridge precision

Tableau Video

ridge example

What is a Residual?

Homoskedasticity (Constant Variance of the Residuals)

Heteroskedasticity (Non-Constant Variance of the Residuals)

vs.

Common Example: Spending & Income

Hypothesis Test for Heteroskedasticity: Breusch-Pagan

Remedy for Heteroskedasticity:

  1. Transform Response Variable OR
  2. Weight the Residuals

Check if Residuals are Normal

Example & Remedy of Heteroskedasticity (Spending & Income)

Visual of Residuals on Income

Regress Spending on Income (is Variance of observations increasing?)

Breusch-Pagan Test for Heterskedasticity

Attempted Remedies:

error

normality of residuals

homosked

heterosked

breusch-pagan

remedy for heterosked

hetero example

hetero example test

hetero example breusch

  1. Transform Spending to Log Form
  1. Weighted Least Squares

hetero example log

weight residuals eg

1.

2.

3.

4.

5.

6.

7.

8.

9.

10.

11.

12.

13.

14.

15.

16.

17.

18.

19.

20.

decision tree multicollin