Statistics 102H: Colinearity

Administrivia

• project due after spring break (also first midterm is after spring break)

Homework discussion

• Generally looked good
• Show a few pictures from student homeworks

Parcel Handling (page 148.)

• Log lives on percentile scale
  • log(1 + epsilon) = epsilon
  • Always use base e if you want to talk about percents!
  • compounding percents leads to exponental
• log-log models
  • log(Y) = alpha + beta log(X)
  • Y = alpha' X^beta
  • Intrinsically a power-law relationship
  • Nicely captures deminishing marginal returns
  • Called elasticities
• Converting elasticities to slopes and back
• Now look at book/data

Categorical variables

We will follow the book carefully on this one.

Homework

• Find the categorical variables in your project
• Do a cross tabs to see how colinear they are
• Look at a variety of relationships between the categorical variables and the continuous variables.
• Use a formula to convert a two-category variable into an indicator function. Now it can be summarized by a correlation with other variables. Why doesn't this work if there are three categories?