power of a test
DESCRIPTION
Power of a test. The power of a test (against a specific alternative value). Is the probability that the test will reject the null hypothesis when the alternative is true In practice, we carry out the test in hope of showing that the null hypothesis is false, so high power is important. - PowerPoint PPT PresentationTRANSCRIPT
Power of a testPower of a test
The powerpower of a test (against a specific alternative value)• Is the probability that the test
will reject the null hypothesis when the alternative is true
• In practice, we carry out the test in hope of showing that the null hypothesis is false, so high power is important
H0 True
H0
FalseReject
Fail to reject
Type I Correct
Correct
Type II
Power
Suppose H0 is true – what if we decide to
fail to reject it?
Suppose H0 is false – what if we decide to
reject it?
Suppose H0 is true – what if we decide to reject
it?
Suppose H0 is false – what if
we decide to fail to reject it?
We correctly We correctly reject a false Hreject a false H00!!
A new flu vaccine claims to prevent a certain type of flu in 70% of the people who are vaccinated. A random sample of 100 people are vaccinated and then exposed to the flu. Is this claim too high? Use = .05.
What are the hypotheses?H0: p = .7
Ha: p < .7
A new flu vaccine claims to prevent a certain type of flu in 70% of the people who are vaccinated. A random sample of 100 people are vaccinated and then exposed to the flu. Is this claim too high? Use = .05.
Find p and p.p = .7p = .0458
A new flu vaccine claims to prevent a certain type of flu in 70% of the people who are vaccinated. A random sample of 100 people are vaccinated and then exposed to the flu. Is this claim too high? Use = .05.
What is the probability of committing a Type I error? = .05
.7
H0: p = .7Ha: p < .7 = .05For what values of the sample proportion would you reject the null hypothesis?
Invnorm(.05,.7,.0458) =.625
= .05
p?
So if we get p-
hat=.625 or less, we
would reject H0.
H0: p = .7Ha: p < .7
We reject H0 and decide that p<.7.
Suppose that pa is 0.6.
What is the probability of committing a Type II error?
Where did this
number come from?
I selected a number that
was less than .7
.7
.6
= .05Reject
What is a type II error?
= ?
How can we find this area?What is the standard deviation of this curve?
Normalcdf(.625,∞,.6,.0458) =.293
failing to reject H0 when the alternative is true
What is the power of the test?
Power = 1 - .293= .707
.7
.6
= .05
=.293
What is the definition of
power?The probability that the test correctly rejects H0
Power = ?
Power - the probability
that the test correctly
rejects H0, if p = .6, is .707
Is power a conditional probability?
Suppose we select .55 as the alternative proportion (p).
a)What is the probability of the type II error?
b) What is the power of the test?
= normalcdf(.625,∞, .55,.0458) = .051
.7
.6
.55
= .05
Power = 1 - .051= .949
What happened to the power of the
test when the difference |p0 – pa| is
increased?
Suppose we select .65 as the alternative proportion (p).
a) What is the probability of the type II error?
b) What is the power of the test?
Power = 1 - .707= .293
.7
.6
.65
= .05
= normalcdf(.625,∞, .65,.0458) = .707
What happened to the power
when the difference |p0-pa|
is decreased?
Power
Suppose that we change alpha to 10%.
Using pa = .6, what would happen to the probability of a type II error and the power of the test?
.7
.6
= .05 = .1
Power
The probability of the type II error () decrease and power increased, BUT BUT the probability of a type I error also also increasedincreased.
= .1836
Power = .8164
What happens to What happens to , , , & , & power when the sample size power when the sample size
is increased?is increased?Reject H0Fail to Reject H0
p0
pa
P(type II) decreases
when n increases
Power
Power increases when n
increases
p0
pa
Reject H0Fail to Reject H0
Power = 1 -
Recap:What affects the power of a test?
As |p0 – pa| increases, power increases
As increases, power increases
As n increases, power increases
Facts:Facts:• The researcher is free to determine the
value of .• The experimenter cannot control , since it
is dependent on the alternate value.• The ideal situation is to have as small as
possible and power close to 1. (Power > .8)• As increases, powerpower increases. (But also
the chance of a type I error has increased!)• Best way to increase power, without
increasing , is to increase the sample sizesample size
A new flu vaccine claims to prevent a certain type of flu in 70% of the people who are vaccinated. In a test, vaccinated people were exposed to the flu. Is this claim too high?
Identify the decision:
a) You decide that the proportion of vaccinated people who do not get the flu is less than 70% when it really is not.
Type I Error
A new flu vaccine claims to prevent a certain type of flu in 70% of the people who are vaccinated. In a test, vaccinated people were exposed to the flu. Is this claim too high?
Identify the decision:
b) You decide that the proportion of vaccinated people who do not get the flu is less than 70% when it really is.
Correct –
Power!!