christopher dougherty ec220 - introduction to econometrics (review chapter) slideshow: one-sided t...
TRANSCRIPT
Christopher Dougherty
EC220 - Introduction to econometrics (review chapter)Slideshow: one-sided t tests
Original citation:
Dougherty, C. (2012) EC220 - Introduction to econometrics (review chapter). [Teaching Resource]
© 2012 The Author
This version available at: http://learningresources.lse.ac.uk/141/
Available in LSE Learning Resources Online: May 2012
This work is licensed under a Creative Commons Attribution-ShareAlike 3.0 License. This license allows the user to remix, tweak, and build upon the work even for commercial purposes, as long as the user credits the author and licenses their new creations under the identical terms. http://creativecommons.org/licenses/by-sa/3.0/
http://learningresources.lse.ac.uk/
1
probability densityfunction of X
10
This sequence explains the logic behind a one-sided t test.
ONE-SIDED t TESTS
0+sd0–sd
null hypothesis: H0 : = 0
alternative hypothesis: H1 : = 1
0–2sd 1+2sd
2.5% 2.5%
2
probability densityfunction of X
10
ONE-SIDED t TESTS
0+sd0–sd
null hypothesis: H0 : = 0
alternative hypothesis: H1 : = 1
0–2sd
We will start by considering the case where can take only two possible values: 0, as under the null hypothesis, and 1, the only possible alternative.
1+2sd
2.5% 2.5%
3
probability densityfunction of X
10
ONE-SIDED t TESTS
0+sd0–sd
null hypothesis: H0 : = 0
alternative hypothesis: H1 : = 1
0–2sd
An example of this situation is where there are two types of removable lap-top batteries: regular and long life.
1+2sd
2.5% 2.5%
4
probability densityfunction of X
10
ONE-SIDED t TESTS
0+sd0–sd
null hypothesis: H0 : = 0
alternative hypothesis: H1 : = 1
0–2sd
You have been sent an unmarked shipment and you take a sample and see how long they last. Your null hypothesis is that they are regular batteries.
1+2sd
2.5% 2.5%
5
probability densityfunction of X
10
ONE-SIDED t TESTS
0+sd0–sd
null hypothesis: H0 : = 0
alternative hypothesis: H1 : = 1
0–2sd
Suppose that the sample outcome is as shown. You would not reject the null hypothesis because the sample estimate lies within the acceptance region for H0.
1+2sd
2.5% 2.5%
6
probability densityfunction of X
10
ONE-SIDED t TESTS
0+sd0–sd
null hypothesis: H0 : = 0
alternative hypothesis: H1 : = 1
0–2sd
Here you would reject the null hypothesis and conclude that the shipment was of long life batteries.
1+2sd
2.5% 2.5%
7
probability densityfunction of X
10
ONE-SIDED t TESTS
0+sd0–sd
null hypothesis: H0 : = 0
alternative hypothesis: H1 : = 1
0–2sd
Here you would stay with the null hypothesis.
1+2sd
2.5% 2.5%
8
probability densityfunction of X
10
ONE-SIDED t TESTS
0+sd0–sd
null hypothesis: H0 : = 0
alternative hypothesis: H1 : = 1
0–2sd
A sample outcome like this one gives rise to a serious problem. It lies in the rejection region for H0, so our first impulse would be to reject H0.
1+2sd
2.5% 2.5%
9
probability densityfunction of X
10
ONE-SIDED t TESTS
0+sd0–sd
null hypothesis: H0 : = 0
alternative hypothesis: H1 : = 1
1+2sd0–2sd
But to reject H0 and go with H1 is nonsensical. Granted, the sample outcome seems to contradict H0, but it contradicts H1 even more strongly.
2.5% 2.5%
10
probability densityfunction of X
10
ONE-SIDED t TESTS
0+sd0–sd
null hypothesis: H0 : = 0
alternative hypothesis: H1 : = 1
1+2sd0–2sd
The probability of getting a sample outcome like this one is much smaller under H1 than it is under H0.
2.5% 2.5%
11
probability densityfunction of X
10
ONE-SIDED t TESTS
0+sd0–sd
null hypothesis: H0 : = 0
alternative hypothesis: H1 : = 1
1+2sd0–2sd
For this reason the left tail should be eliminated as a rejection region for H0. We should use only the right tail as a rejection region.
2.5%
12
probability densityfunction of X
10
ONE-SIDED t TESTS
0+sd0–sd
null hypothesis: H0 : = 0
alternative hypothesis: H1 : = 1
1+2sd0–2sd
The probability of making a Type I error, if H0 happens to be true, is now 2.5%, so the significance level of the test is now 2.5%.
2.5%
13
probability densityfunction of X
10
ONE-SIDED t TESTS
0+sd0–sd
null hypothesis: H0 : = 0
alternative hypothesis: H1 : = 1
1+2sd0–2sd
We can convert it back to a 5% significance test by building up the right tail until it contains 5% of the probability under the curve. It starts 1.645 standard deviations from the mean.
5%
14
probability densityfunction of X
10
ONE-SIDED t TESTS
0+sd0–sd
null hypothesis: H0 : = 0
alternative hypothesis: H1 : = 1
1+2sd0–2sd
Why would we want to do this? For the answer, we go back to the trade-off between Type I and Type II errors.
5%
15
probability densityfunction of X
10
ONE-SIDED t TESTS
0+sd0–sd
null hypothesis: H0 : = 0
alternative hypothesis: H1 : = 1
1+2sd0–2sd
With a 5% test, there is a greater chance of making a Type I error if H0 happens to be true, but there is less risk of making a Type II error if it happens to be false.
5%
16
probability densityfunction of X
10
ONE-SIDED t TESTS
0+sd0–sd
null hypothesis: H0 : = 0
alternative hypothesis: H1 : = 1
1+2sd0–2sd
Note that the logic for dropping the left tail depended only on 1 being greater than 0. It did not depend on 1 being any specific value.
5%
17
probability densityfunction of X
10
ONE-SIDED t TESTS
0+sd0–sd
null hypothesis: H0 : = 0
alternative hypothesis: H1 : > 0
1+2sd0–2sd
Hence we can generalize the one-sided test to cover the case where the alternative hypothesis is simply that is greater than 0.
5%
18
probability densityfunction of X
10
ONE-SIDED t TESTS
0+sd0–sd
null hypothesis: H0 : = 0
alternative hypothesis: H1 : > 0
1+2sd0–2sd
To justify the use of a one-sided test, all we have to do is to rule out, on the basis of economic theory or previous empirical experience, the possibility that is less than 0.
5%
probability densityfunction of X
0 0 +sd0–sd
null hypothesis: H0 : = 0
alternative hypothesis: H1 : < 0
0+2sd0–2sd
Sometimes, given a null hypothesis H0: = 0, on the basis of economic theory or previous experience, you can rule out the possibility of being greater than 0.
19
ONE-SIDED t TESTS
1
5%
probability densityfunction of X
0 0 +sd0–sd
null hypothesis: H0 : = 0
alternative hypothesis: H1 : < 0
0+2sd0–2sd
20
ONE-SIDED t TESTS
1
In this situation you would also perform a one-sided test, now with the left tail being used as the rejection region. With this change, the logic is the same as before.
5%
We will next investigate how the use of a one-sided test improves the trade-off between the risks of making Type I and Type II errors.
21
ONE-SIDED t TESTS
probability densityfunction of X
10 0+sd0–sd
null hypothesis: H0 : = 0
alternative hypothesis: H1 : = 1
0–2sd 1+2sd
2.5% 2.5%
22
ONE-SIDED t TESTS
probability densityfunction of X
10 0+sd0–sd
null hypothesis: H0 : = 0
alternative hypothesis: H1 : = 1
0–2sd 1+2sd
We will start by returning to the case where can take only two possible values, 0 and 1.
2.5% 2.5%
23
ONE-SIDED t TESTS
probability densityfunction of X
10 0+sd0–sd
null hypothesis: H0 : = 0
alternative hypothesis: H1 : = 1
0–2sd 1+2sd
Suppose that we use a two-sided 5% significance test. If H0 is true, there is a 5% risk of making a Type I error.
2.5% 2.5%
24
ONE-SIDED t TESTS
probability densityfunction of X
10 0+sd0–sd
null hypothesis: H0 : = 0
alternative hypothesis: H1 : = 1
0–2sd 1+2sd
However, H0 may be false. In that case the probability of not rejecting it and making a Type II error is given by the blue shaded area.
2.5% 2.5%
25
ONE-SIDED t TESTS
probability densityfunction of X
10 0+sd0–sd
null hypothesis: H0 : = 0
alternative hypothesis: H1 : = 1
0–2sd 1+2sd
This area gives the probability of the estimate lying within the acceptance region for H0, if H1 is in fact true.
2.5% 2.5%
26
ONE-SIDED t TESTS
probability densityfunction of X
10 0+sd0–sd
null hypothesis: H0 : = 0
alternative hypothesis: H1 : = 1
0–2sd 1+2sd
Now suppose that you use a one-sided test, taking advantage of the fact that it is irrational to reject H0 if the estimate is in the left tail.
5%
27
ONE-SIDED t TESTS
probability densityfunction of X
10 0+sd0–sd
null hypothesis: H0 : = 0
alternative hypothesis: H1 : = 1
0–2sd 1+2sd
Having expanded the right tail to 5%, we are still performing a 5% significance test, and the risk of making a Type I error is still 5%, if H0 is true.
5%
28
ONE-SIDED t TESTS
probability densityfunction of X
10 0+sd0–sd
null hypothesis: H0 : = 0
alternative hypothesis: H1 : = 1
0–2sd 1+2sd
But if H0 is false, the risk of making a Type II error is smaller than before. The probability of an estimate lying in the acceptance region for H0 is now given by the green area.
5%
29
ONE-SIDED t TESTS
probability densityfunction of X
10 0+sd0–sd
null hypothesis: H0 : = 0
alternative hypothesis: H1 : = 1
0–2sd 1+2sd
This is smaller than the blue area for the two-sided test.
5%
30
ONE-SIDED t TESTS
probability densityfunction of X
10 0+sd0–sd
null hypothesis: H0 : = 0
alternative hypothesis: H1 : = 1
0–2sd 1+2sd
Thus, with no increase in the probability of making a Type I error (if H0 is true), we have reduced the probability of making a Type II error (if H0 is false).
5%
31
ONE-SIDED t TESTS
probability densityfunction of X
10 0+sd0–sd
null hypothesis: H0 : = 0
alternative hypothesis: H1 : = 1
0–2sd 1+2sd
When the alternative hypothesis is H1: > 0 or H1: < 0, the more general (and more typical) case, we cannot draw this diagram.
5%
32
ONE-SIDED t TESTS
probability densityfunction of X
10 0+sd0–sd
null hypothesis: H0 : = 0
alternative hypothesis: H1 : = 1
0–2sd 1+2sd
Nevertheless we can be sure that, by using a one-sided test, we are reducing the probability of making a Type II error, if H0 happens to be false.
5%
33
One-sided tests are often particularly useful where the analysis relates to the evaluation of treatment and effect. Suppose that a number of units of observation receive some type of treatment and Xi is a measure of the effect of the treatment for observation i.
null hypothesis: H0 : = 0
ONE-SIDED t TESTS
34
To demonstrate that the treatment did have an effect, we set up the null hypothesis H0: = 0 and see if we can reject it, given the sample average X.
null hypothesis: H0 : = 0
ONE-SIDED t TESTS
35
probability densityfunction of X
0
If you use a two-sided 5% significance test, X must be 1.96 standard deviations above or below 0 if you are to reject H0.
null hypothesis: H0 : = 0
alternative hypothesis: H1 : = 0
reject H0reject H0 do not reject H0
1.96 sd-1.96 sd
ONE-SIDED t TESTS
2.5% 2.5%
36
probability densityfunction of X
0
However, if you can justify the use of a one-sided test, for example with H1: > 0, your estimate only has to be 1.65 standard deviations above 0.
reject H0do not reject H0
1.65 sd
ONE-SIDED t TESTS
null hypothesis: H0 : = 0
alternative hypothesis: H1 : > 0
5%
37
probability densityfunction of X
0
reject H0do not reject H0
1.65 sd
ONE-SIDED t TESTS
null hypothesis: H0 : = 0
alternative hypothesis: H1 : > 0
This makes it easier to reject H0 and thereby demonstrate that the treatment has had a significant effect.
5%
38
probability densityfunction of X
0
reject H0do not reject H0
1.65 sd
ONE-SIDED t TESTS
null hypothesis: H0 : = 0
alternative hypothesis: H1 : > 0
Throughout this sequence, it has been assumed that the standard deviation of the distribution of b2 is known, and the normal distribution has been used in the diagrams.
5%
39
probability densityfunction of X
0
reject H0do not reject H0
1.65 sd
ONE-SIDED t TESTS
null hypothesis: H0 : = 0
alternative hypothesis: H1 : > 0
In practice, of course, the standard deviation has to be estimated as the standard error, and the t distribution is the relevant distribution. However, the logic is exactly the same.
5%
40
probability densityfunction of X
0
reject H0do not reject H0
1.65 sd
ONE-SIDED t TESTS
null hypothesis: H0 : = 0
alternative hypothesis: H1 : > 0
At any given significance level, the critical value of t for a one-sided test is lower than that for a two-sided test.
5%
41
probability densityfunction of X
0
reject H0do not reject H0
1.65 sd
ONE-SIDED t TESTS
null hypothesis: H0 : = 0
alternative hypothesis: H1 : > 0
Hence, if H0 is false, the risk of not rejecting it, thereby making a Type II error, is smaller.
5%
Copyright Christopher Dougherty 2011.
These slideshows may be downloaded by anyone, anywhere for personal use.
Subject to respect for copyright and, where appropriate, attribution, they may be
used as a resource for teaching an econometrics course. There is no need to
refer to the author.
The content of this slideshow comes from Section R.13 of C. Dougherty,
Introduction to Econometrics, fourth edition 2011, Oxford University Press.
Additional (free) resources for both students and instructors may be
downloaded from the OUP Online Resource Centre
http://www.oup.com/uk/orc/bin/9780199567089/.
Individuals studying econometrics on their own and who feel that they might
benefit from participation in a formal course should consider the London School
of Economics summer school course
EC212 Introduction to Econometrics
http://www2.lse.ac.uk/study/summerSchools/summerSchool/Home.aspx
or the University of London International Programmes distance learning course
20 Elements of Econometrics
www.londoninternational.ac.uk/lse.
11.10.13