This preview shows pages 1–4. Sign up to view the full content.
This preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full Document
Unformatted text preview: Further Issues in Regression Analysis ECON 399 Neil Hepburn Contents 1 Introduction 1 2 Goodness of Fit in the kVariable Model 1 3 Selecting Between Alternate Models 2 3.1 Log vs Linear Specifications . . . . . . . . . . . . . . . . . . . . . 3 4 Forecasts and Forecast Errors 5 1 Introduction Introduction • We now clean up our initial look at linear regression by touching on a couple of final topics • First we look at the R 2 in the multiple regression model and examine a method for preventing cheating by adding irrelevant variables • We conclude with the issue of forecasting, prediction, and confidence in tervals for predictions. 2 Goodness of Fit in the kVariable Model Goodness of Fit and Degrees of Freedom • In theory it is possible to make the unadjusted R 2 equal to one just by adding more regressors, even if they are meaningless • Doing so will certainly not make R 2 go down • We need to correct for this by taking into account the degrees of freedom. 1 Goodness of Fit and the Adjusted R 2 • We can compute an adjusted R 2 that prevents this little fudge from oc curring • The adjusted ¯ R 2 is ¯ R 2 = 1 SSR / k SST /( n 1) (1) • We can easily turn an unadjusted R 2 into an adjusted R 2 ¯ R 2 = 1 ( 1 R 2 ) ( n 1) ( n k 1) (2) Goodness of Fit and the Adjusted R 2 • Imagine that we add several nonsense variables to our model. • There will be some change in R 2 , but it will be slight • If we look at Equation (2) we can see that ( 1 R 1 ) will become a bit smaller • However, ( n k 1) will become much smaller. The overall result in Equation (2) is that ¯ R 2 will become smaller. 3 Selecting Between Alternate Models Comparisons When y i is the Same in Both Models • When the dependent variable is the same in both models, we can compare the ¯ R 2 for both models • Since the dependent variable is the same in both models, the ¯ R 2 are telling us the same thing  the variation in y i that is explained by the model • If the dependent variables are different in the two models, then we cannot make a comparison this way Comparisons When y i is Not the Same • One case where this frequently comes up is when we have one model with y i and the other with ln( y i ) • Since the dependent variables are different we cannot make a comparison • There is a test that we will see later on that we can use to test for a linear vs logarithmic specification 2 3.1 Log vs Linear Specifications Comparing Log and Linear Models • There is a way that we can compare two specifications of the same under lying model • Suppose that we are trying to choose between a model that uses y as the dependent variable and one that uses log ( y ) • Our two competing models are wage = β + β 1 belavg + β 2 abvavg + β 3 exper + β 4 expersq + β 5 service + β 6 educ + β 7 female + u log (wage) = β + β 1 belavg + β 2 abvavg + β 3 exper + β 4 expersq + β 5 service + β 6 educ + β 7 female + u Comparing Log and Linear Specifications...
View
Full
Document
This note was uploaded on 10/04/2011 for the course ECONOMICS 399 taught by Professor Neil.h during the Spring '11 term at University of Alberta.
 Spring '11
 Neil.H
 Econometrics

Click to edit the document details