Scribd
Upload
27 views1 page

Exact Confidence Regions For Regression Parameters

This document discusses constructing exact confidence regions for regression parameters in the classical normal linear model. It explains that an F-statistic can be used to test hypotheses about the vector of parameters β2. This F-statistic follows an F-distribution under the null hypothesis, allowing a confidence region to be constructed using the quantiles of the F-distribution. Specifically, the 1-α confidence region consists of all values of β20 where the F-statistic is less than the 1-α quantile of the F-distribution, resulting in an elliptical or ellipsoidal confidence shape depending on the dimension of β2.

Uploaded by

Leonard Gonzalo Saavedra Astopilco
Copyright
© Attribution Non-Commercial (BY-NC)
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
27 views1 page

Exact Confidence Regions For Regression Parameters

This document discusses constructing exact confidence regions for regression parameters in the classical normal linear model. It explains that an F-statistic can be used to test hypotheses about the vector of parameters β2. This F-statistic follows an F-distribution under the null hypothesis, allowing a confidence region to be constructed using the quantiles of the F-distribution. Specifically, the 1-α confidence region consists of all values of β20 where the F-statistic is less than the 1-α quantile of the F-distribution, resulting in an elliptical or ellipsoidal confidence shape depending on the dimension of β2.

Uploaded by

Leonard Gonzalo Saavedra Astopilco
Copyright
© Attribution Non-Commercial (BY-NC)
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/ 1

192

Condence Intervals

Exact Condence Regions for Regression Parameters


Suppose that we want to construct a condence region for the elements of the vector 2 in the classical normal linear model (4.28), which we rewrite here for ease of exposition: y = X1 1 + X2 2 + u, u N(0, 2 I), (5.24)

where 1 and 2 are a k1 --vector and a k2 --vector, respectively. The F statistic that can be used to test the hypothesis that 2 = 0 is given in (4.33). If we wish instead to test 2 = 20 , then we can write (5.24) as y X2 20 = X1 1 + X2 + u, u N(0, 2 I), (5.25)

where = 2 20 . We wish to test the hypothesis that = 0. It is not hard to show that the F statistic for this hypothesis takes the form 2 20 ) X2 M1 X2 ( 2 20 )/k2 ( , y MX y /(n k ) (5.26)

where k = k1 + k2 ; see Exercise 5.10. When multiplied by k2 , this F statistic is in the form of the Wald statistic (5.18). For the purposes of inference on 2 , regression (5.24) is, by the FWL Theorem, equivalent to the regression M1 y = M1 X2 2 + M1 u. 2 ) is equal to 2 (X2 M1 X2 )1. Since the denominator of (5.26) is Thus Var( 2 just s , the OLS estimate of the error variance from running regression (5.24), k2 times the F statistic (5.26) can be written in the form of (5.18), with 2 = s2 X2 M1 X2 1 Var 2 ; compare (3.50). providing a consistent estimator of the variance of Under the assumptions of the classical normal linear model, the F statistic (5.26) follows the F (k2 , n k ) distribution when the null hypothesis is true. Therefore, we can use it to construct an exact condence region. If c denotes the 1 quantile of the F (k2 , n k ) distribution, then the 1 condence region is the set of all 20 for which 2 20 ) X2 M1 X2 ( 2 20 ) c k2 s2. ( (5.27)

Since the left-hand side of this inequality is quadratic in 20 , the condence region is, for k2 = 2, the interior of an ellipse and, for k2 > 2, the interior of a k2 --dimensional ellipsoid.

You might also like