Lecture 11_multregII-2012

Lecture 11_multregII-2012 - Lecture 11 Multiple Regression...

This preview shows pages 1–7. Sign up to view the full content.

1 Lecture 11: Multiple Regression II Stat 102 2012 R 2 as a Measure of Model Fit. Read Chapter 4.3 Sums of Squares and Effect Tests. Read Chapter 4.4 & Sums of Squares SST, SSE and SSR & ANOVA Table & R 2 & F-statistic and the F-test for ALL coefficients together & Effect Test Table and Tests for Individual Coefficients Example : Gasoline Mileage for New Cars [cont]

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

View Full Document
2 SST and SSE In multiple regression, this works the same general way as in ordinary regression. The Total Sum of Squares has the same formula: ( ) 2 1 n i i SST Y Y = = . The Sum of Squares for Error has the same sort of formula: ( ) 2 1 ˆ n ii i SSE Y Y = = . & The only difference here is that the formula for the prediction of the conditional mean involves all the x-variables – ˆ Y i ! b 0 + b 1 i x 1 i + .. + b Ki x Ki .
3 Sum of Squares for Regression The decomposition of the sums of squares also works just like in ordinary regression. The S um of S quares for R egression is called SSR in Dielman , and S um of S quares for the M odel in JMP. The basic decomposition is SST = SSR+SSE . Hence, from SST and SSE you can find SSR as SSR = SST-SSE These sums of squares appear in the “Analysis of Variance Table” (aka ANOVA table)

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

View Full Document
4 ANOVA Table Here is the JMP ANOVA table for our data Analysis of Variance Source DF Sum of Squares Mean Square F Ratio Model K 4 SSR 2291.2 SSR/DFR 573 MSR / MSE 210 Error n - K -1 217 SSE 589.9 SSE/DFE 2.72 Prob > F C. Total n- 1 221 SST 2881.1 <.0001 Note the D egrees of F reedom: DFR = K , DFE = n-K- 1, DFT otal = n- 1. Note the M ean S quares: “Mean Square for Model” = MSR = SSR / DFR 2 e s = MSE = SSE / DFE. Finally NOTE: F = MSR / MSE.
5 The “Coefficient of Determination”, R 2 Relative to the terms in the ANOVA table this also has the same definition and interpretation as in ordinary regression: 2 SSR SST SSE R SST SST == . THUS, R 2 is the proportion of squared variability accounted for by the linear regression using all the x- variables relative to squared variability of the best fit that does not use any of the x-variables. One difference between the situation here and in ordinary regression: In ordinary regression, R 2 is the square of a sample regression coefficient – There’s no such interpretation here !

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

View Full Document
6 R 2 appears in the JMP Summary of Fit Table: Summary of Fit RSquare R 2 0.795 RSquare Adj 0.791 Root Mean Square Error s e 1.65 Mean of Response Y 19.40 Observations n 222 See Dielman for definition of “Adjusted R 2 ”. We don’t find this concept particularly useful, and so will always ignore it. Always 2 01 R ≤≤ . & R 2 = 0 means the linear regression equation is useless as a means of predicting the Y i .
This is the end of the preview. Sign up to access the rest of the document.

{[ snackBarMessage ]}

Page1 / 21

Lecture 11_multregII-2012 - Lecture 11 Multiple Regression...

This preview shows document pages 1 - 7. Sign up to view the full document.

View Full Document
Ask a homework question - tutors are online