Professional Documents
Culture Documents
Hypothesis testing #2
Class 15 Tuesday 3/26/2013
From Cornelia Grunwald-Kennedy: On April 4, mechanical engineering alumnus, Manuel Pablo Zuniga-Pflucker (Manolo Zuniga) will be giving a special presentation entitled Leadership Evolution Managing an International Publicly Traded Energy Company in room 1202 Martin Hall beginning at 5:00 p.m. We will have a brief reception before the presentation. This presentation may be of particular interest to students because he grew a small company that was founded by his father, and took it to the New York Stock Exchange in 2004. He will talk about what it takes to be a leader in an uncertain industry, and the new technology that they are using in oil exploration in a variety of environments. From www.enme.umd.edu: University of Maryland, College Park, graduates earn the second-highest starting salaries among students from all 50 U.S. state flagship schools, according to a recent study.
sample
Exercise 10.1 An allergist wishes to test the (null) hypothesis that at least 30% of the public is allergic to cheese. How can the allergist commit (a) a type I error? (b) a type II error? (a) Null hypothesis: > 30% are allergic Conclude that <30% of the public are allergic, when > 30% are allergic. (b) Conclude that > 30% of the public are allergic, when < 30% are allergic.
Example
Exercise 10.6 The proportion of adults in a small town who are college graduates is estimated to be p = 0.6. In a sample of 15 adults, we shall not reject the null hypothesis if the number of college graduates is from 6 to 12. (a) Find . (b) Evaluate for the alternative p = 0.5. (c) Is this a good test procedure?
Example
(b) If p = 0.5, then what are we looking for? Make a sketch. H1: p = 0.5 H0: p = 0.6 This is a conceptual sketch, not an accurate one for this problem
x = 12
P (x 12) =
x=0
12 X
(c) No. There is a large probability of error! The test cannot detect differences of 0.1 in p.
Example
Exercise 10.7 The proportion of adults in a small town who are college graduates is estimated to be p = 0.6. In a sample of 200 adults, we shall not reject the null hypothesis if the number of college graduates is from 110 to 130. (a) Find . (b) Evaluate for the alternative p = 0.5. (c) Is this a good test procedure? (a) Use the normal approximation to the binomial distribution.
Z=
Z=
= np
= npq
Z=
The probability of rejecting the null hypothesis even if its true (Type 1 error) is = 2 * P(Z < -1.515) = 2 * 0.0649 = 0.130
Example
(b) If p = 0.5, what is ? Make a sketch!
= 100
=
120
H0: p = 0.6
H1: p = 0.5
Z=
= np
= npq
n = 200; p = 0.5; q = 0.5 = 100; = 7.07 = 1 - P(Z < 1.344) = 1 - 0.91 = 0.09
Z=
(c) Not great. The probability of a type I error is actually higher, (0.13 compared to 0.061), and the probability of a type II error, while much lower (0.09 instead of 0.85), is sDll rather high.
Z=
x p = 1.6 / n
P = 2 CDF( Z )
1. hypotheses: H0: = 0; H1: 0 2. two-sided test, /2 = 0.025 3. test statistic? We are told to assume that the populations are normal, so we can use the t-test.
T=
D D Sd / n
v = n 1 = 5 1 = 4 d.o.f.
t0.025 = 2.776 the two critical regions are t < -2.776 and t > 2.776
T=
5. decision: Since t = -1.58 is larger than t0.025 = -2.776, we cannot reject H0 The two methods give the same results (when tested at =0.05 level)
Test on Proportions
p = x/n.
x np0 z= np0 ( 1 p0 )
or
z=
p p0 p0 q0 / n
x is the number of good items if you are comparing proportions of yield, or the number of defects if you are comparing proportions of defectives. n is the sample size p^ is the sample proportion p0 is the target proportion
Example
Example A semiconductor manufacturer produces controllers used for automobile engines. The customer requires that the fraction defective (process fallout) at a critical manufacturing step not exceed 0.07 and that the manufacturer demonstrate process capability at this level of quality using = 0.05. The semiconductor manufacturer takes a random sample of 300 devices and finds that seven of them are defective. Can the manufacturer demonstrate adequate process capability for the customer? The parameter of interest is the fraction defective p0. 1. H0: p0 = 0.07, H1: p0 < 0.07 2. One-sided test, = 0.05 3. Since n is large, the test statistic is
z=
p p0 p0 q0 / n
Example
4. calculate: n = 300, p^ = 7/300 = 0.023, p0 = 0.07, q0 = 0.93
z= =
p p0
0.023 0.07 0.047 0.047 = = = p0 q0 / n 0.07 * 0.93 / 300 0.065 / 300 0.00022
5. decision: The P-value for this is small, 0.0007, much less than = 0.05, so we can reject H0 and conclude that our process meets the customer requirements.