lec34v1_1up

lec34v1_1up - Stat 104 Quantitative Methods for Economists...

This preview shows pages 1–17. Sign up to view the full content.

Stat 104: Quantitative Methods for Economists Class 34: Heteroskedasticity and Multicollinearity 1

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
The Model and Data Y X = + + β ε 0 1 Given X ’s X X X m 1 2 , , K We assume the following model holds: 2 ~ (0, ) i N independent σ Y X i i i = + + 0 1 2
Non-Constant Variance or Heteroskedasticity Another of our basic assumptions is that the all have the same distribution and in particular, the same variance. What does a violation of this assumption look like ? i ε heteroskedasticity means the variance of the errors changes. our model assumes “homoskedasticity” 3

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
Example: We have data on manufacturing plants for a Fortune 500 company. The data consists of the number of supervisors (Y) and the associated number of supervised workers (X), 200 4 0 super 0 500 1000 1500 2000 worker
Regression Output 5

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
Residual Diagnostics -20 0 20 40 Residuals If you have heteroskedasaticity, your estimates are ok, but your standard errors are incorrect. That’s not good. 6 -60 -40 50 100 150 200 Fitted values
Good basic solution for heteroskedasticity s One way to think of heteroskedasticity is that the noise in the model increases as the value of |X| increases ssentially there is too much variation in the s Essentially there is too much variation in the model.That is, there is excess variation in the Y variable. s An easy way to reduce the variation in Y is to take the log of it. 7

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
The log s By “logging” your data, you are transforming it to a different scale. s The log scale squeezes numbers together, so ere is less variation. However, the model there is less variation. However, the model becomes slightly different to interpret since the scale of the Y variables changes (instead of dollars we are modeling “log dollars”; what exactly are those ?) 8
The log again- shrinks the scale Y log(Y) 1 0 5 1.609438 10 2.302585 50 3.912023 100 4.60517 1000 6.907755 15000 9.615805 Variance 31446401 10.80543 9

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
Plot of Log Y versus X 5 5.5 10 3.5 4 4.5 lsuper 0 500 1000 1500 2000 worker
New Regression Output WARNING : now can’t compare s or R-sq 11

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
New Residual Plot -.2 0 .2 .4 Residuals Are we done or is there anything else wrong ? 12 -.6 -.4 4 4.5 5 5.5 Fitted values
What transformation should we try ? o back to our Go back to our transformation guide 13

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
We will try log(X) 5 5.5 14 3.5 4 4.5 lsuper 5.5 6 6.5 7 7.5 lworker
New Output 15

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
0 .2
This is the end of the preview. Sign up to access the rest of the document.

This note was uploaded on 03/27/2012 for the course STATS 104 taught by Professor Michaelparzen during the Fall '11 term at Harvard.

Page1 / 42

lec34v1_1up - Stat 104 Quantitative Methods for Economists...

This preview shows document pages 1 - 17. Sign up to view the full document.

View Full Document
Ask a homework question - tutors are online