Scribd
Upload
6 views2 pages

SRM 2

Uploaded by

chowshichun
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
6 views2 pages

SRM 2

Uploaded by

chowshichun
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

Note that = (only for simple regression and multiple regression when only one variable is tested).

Suppose we go from a full model to a reduced model. The statistic is defined to be the following.
∆ /∆
=
/ − −1

This is analogous to percentage change in , the portion not explained by regression.

For simple regression, = and = − .

/
⇒ = = −2 = −2
/ −2 / 1−

Statistical inferencing

For confidence interval, standard error is just the above.

For prediction interval, since we need Var , we need to further add to the variance above.

Partial correlation

For and , the partial correlation is calculated by the following.

Step 1: Estimate the full linear regression model without .

! = "# + % "& '&


&(),&+

Then calculate the residuals for each observation.

Step 2: Estimate the full linear regression model using as the dependent variable.

' = "# + % "& '&


&(),&+

Also calculate the residuals for each observation.

Step 3: Calculate the correlation of the two sets of residuals.

If value is provided, the partial correlation is the following.

",
r , =
-. ", / + − +1
Outlier identification

Let 0 = 1 2) 1
be the hat matrix. Then we have the following.

Var 3 = 1−ℎ
Here, ℎ is the leverage of the 5th observation.

Then standardized residual is the following.


3
3 67 =
8 1−ℎ

Another method of calculating standardized residual is Student’s residual.


3
3 67 9
=
- 1−ℎ

Here, ℎ is still from the original definition, but is from the regression without the 5th observation.
; <=)
Rule of thumb: Leverage is big if ℎ > . Here, note that ∑ () ℎ = + 1 (the number of variables
to be estimated in the regression model).

Influential point: Cook’s distance

!?& : Fitted value of ! using the model with @th observation included

!?& : Fitted value of ! using the model with @th observation excluded

: Squared standard error using the model with 5th observation included

1
A = % !?& − !?&
+1
&(),&+

)
Rule of thumb: If A > , 5th observation is influential.

Simpler alternative (just memorize, proof is too difficult):


1 ℎ
A = 3 67 B C
+1 1−ℎ
 Standardized residual, not studentized one.

You might also like