Scribd
Upload
70 views2 pages

Variance Ratio and Likelihood Ratio Tests

The document discusses two statistical hypothesis testing methods: the variance ratio test and the likelihood ratio test. The variance ratio test is used to test if two variances are equal. It involves taking the ratio of two sample variances and comparing it to a critical value from the F-distribution. The likelihood ratio test compares the likelihood of obtaining sample data under the null hypothesis versus an alternative hypothesis. It involves taking the ratio of the likelihoods and comparing the log likelihood ratio to a critical value from the chi-squared distribution. An example applying the likelihood ratio test to test if a Normal distribution has mean 0 or not is shown.

Uploaded by

bo
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
70 views2 pages

Variance Ratio and Likelihood Ratio Tests

The document discusses two statistical hypothesis testing methods: the variance ratio test and the likelihood ratio test. The variance ratio test is used to test if two variances are equal. It involves taking the ratio of two sample variances and comparing it to a critical value from the F-distribution. The likelihood ratio test compares the likelihood of obtaining sample data under the null hypothesis versus an alternative hypothesis. It involves taking the ratio of the likelihoods and comparing the log likelihood ratio to a critical value from the chi-squared distribution. An example applying the likelihood ratio test to test if a Normal distribution has mean 0 or not is shown.

Uploaded by

bo
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

A variance ratio test

Sometimes we want to test the equality of two variances.

 12  12
H0: 1 versus H1: 1
 22  22

As encountered in estimation, variances are associated with the 2 distribution. Here we have
the ratio of two variances however, so we use the F distribution.

X m
Theorem: If X ~ 2m and Y ~ 2n independently of X, then ~ Fm,n
Yn

Let X = (n1-1)s12/12 and Y = (n2 -1)s22/22. Each is 2. Hence

( n1  1) s12  12
n1  1
~ Fm ,n
( n 2  1) s 22  22
n2  1
Cancelling, we get s12/s22 ~ Fm,n The F distribution has two sets of degrees of freedom
(parameters), the first associated with the numerator, the second with the denominator. Both
are related to the size of each sample.

So if we have s12 = 20 and s22 = 15, n1 = 30, n2 = 50, then 20/15 = 1.33 is a single realisation
from an F29,19 distribution. The critical value, derived from tables, is F* = 1.74 (5%), hence
cannot reject H0 as the test statistic does not exceed the critical value.

Likelihood-ratio tests

A likelihood ratio test is a hypothesis test based upon the ML principle. The principle of the
LR test is as follows: to test the hypothesis

H0: 0
H1: 0

we compare the likelihood of obtaining the sample data (a) at the hypothesised value of  and
(b) at the MLE of . The likelihood in (b) must be greater than that in (a) but, if they are
fairly close we do not reject the null hypothesis.

We therefore form the likelihood ratio

L ( x:  0 )
LR  
L ( x: MLE )

where 0    1 and L(x: ) gives the likelihood of the sample x for a value of the parameter
.
The LR is a random variable with a distribution. If we can find it (it depends, in general on
the parent distribution) then we can find a critical region for the test. Small values of LR
reject H0 in favour of the alternative (because the likelihood under the null is much less than
under the MLE).

It is often difficult to find the distribution of , so we use the following theorem:

Theorem

The random variable -2ln has a distribution that approaches that of a 2 distribution as n 
, with degrees of freedom given by the number of parameters determined by H0.

Note that large values of –2ln now reject H0.

Example 1

Suppose X ~ N(and a sample of size n = 100 on X yields  x = 15. Use a LR test to test
H0: 0vs H1: 0.

Solution

First we need the MLE of . For a Normal pdf

 exp x 
100
 1 
L() =    
2
1
2 so
 2 
i

 1  1
ln L() = 100 ln   2   x   
2

 2 

 dlnL/d = (x-u) = 0 when   so   x = 0.15

Not surprisingly, the MLE is the sample mean.

Now

-2 ln = -2[ln L(0) - ln L(0.15)]


= -2[ -½ x2 + ½ (x-0.15)2]
= x2 - x2 + 0.15  2  x - 100  0.152 = 4.5 - 2.25 = 2.25

Since 21 = 3.84 (at 5%) we cannot reject H0.

You might also like