section 4_4

Report
Section 4.4
Creating
Randomization
Distributions
Randomization Distributions
How do we estimate P-values using randomization
distributions?
1. Simulate samples, assuming H0 is true
2. Calculate the statistic of interest for each sample
3. Find the p-value as the proportion of simulated
statistics as extreme as the observed statistic
Today we’ll discuss ways to simulate randomization
samples for a variety of situations.
Cocaine Addiction
• In a randomized experiment on treating
cocaine addiction, 48 people were randomly
assigned to take either Desipramine (a new
drug), or Lithium (an existing drug), and
then followed to see who relapsed
• Question of interest: Is Desipramine better
than Lithium at treating cocaine addiction?
Cocaine Addiction
• What are the null and alternative hypotheses?
• What are the possible conclusions?
Cocaine Addiction
• What are the null and alternative hypotheses?
Let pD, pL be the proportion of cocaine addicts
who relapse after taking Desipramine or Lithium,
respectively.
H0: pD = pL
Ha: pD < pL
• What are the possible conclusions?
Reject H0: Desipramine is better than Lithium
Do not reject H0: We cannot determine from
these data whether Desipramine is better than
Lithium
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
1. Randomly assign units to
treatment groups
Desipramine
R
R
R
R
Lithium
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
2. Conduct experiment
3. Observe relapse counts in each group
R = Relapse
N = No Relapse
1. Randomly assign units to
treatment groups
Desipramine
Lithium
R
R
R
R
R
R
pˆ D  pˆ L
R
R
R
R
R
R
R
R
R
R
N
R
N
R
R
R
R
R
R
R
N
R
N
R
N
N
N
N
R
R
R
R
R
R
N
N
N
N
N
N
10 18


24 24
 .333
N
N
N
N
N
N
10 relapse, 14 no relapse
18 relapse, 6 no relapse
Measuring Evidence against H0
To see if a statistic provides
evidence against H0, we need to
see what kind of sample
statistics we would observe,
just by random chance,
if H0 were true
Cocaine Addiction
• “by random chance” means by the random
assignment to the two treatment groups
• “if H0 were true” means if the two drugs were
equally effective at preventing relapses
(equivalently: whether a person relapses or not
does not depend on which drug is taken)
• Simulate what would happen just by random
chance, if H0 were true…
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
N
N
R
R
R
R
R
R
N
N
N
N
N
N
R
R
R
R
R
R
N
N
N
N
N
N
N
N
N
N
N
N
10 relapse, 14 no relapse
18 relapse, 6 no relapse
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
N
N
R
R
R
R
R
R
N
N
N
N
N
N
R
R
R
R
R
R
N
N
N
N
N
N
N
N
N
N
N
N
Simulate another
randomization
Desipramine
Lithium
R
N
R
N
R
R
R
R
R
R
R
N
R
R
R
N
R
N
N
N
R
R
16 relapse, 8 no relapse
pˆ D  pˆ L
16 12


24 24
 0.167
N
N
N
R
N
R
R
N
N
N
N
R
N
R
R
N
R
N
R
R
R
R
12 relapse, 12 no relapse
Simulate another
randomization
Desipramine
Lithium
R
R
R
R
R
R
R
N
R
R
N
N
R
R
N
R
N
R
R
N
R
N
R
R
17 relapse, 7 no relapse
pˆ D  pˆ L
17 11


24 24
 0.250
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
N
N
N
N
N
N
11 relapse, 13 no relapse
Simulate Your Own Sample
 In the experiment, 28 people relapsed and 20
people did not relapse. Create cards or slips of
paper with 28 “R” values and 20 “N” values.
 Pool these response values together, and
randomly divide them into two groups
(representing Desipramine and Lithium)
 Calculate your difference in proportions
 Plot your statistic on the class dotplot
 To create an entire randomization distribution,
we simulate this process many more times with
technology: StatKey
www.lock5stat.com/statkey
p-value
Randomization Distribution Center
A randomization distribution simulates samples
assuming the null hypothesis is true, so
A randomization distribution is
centered at the value of the parameter
given in the null hypothesis.
Randomization Distribution
In a hypothesis test for H0:  = 12 vs Ha:  < 12,
we have a sample with n = 45 and  = 10.2.
What do we require about the method to
produce randomization samples?
a)  = 12
b)  < 12
c)  = 10.2
We need to generate
randomization
samples assuming the
null hypothesis is true.
Randomization Distribution
In a hypothesis test for H0:  = 12 vs Ha:  < 12,
we have a sample with n = 45 and  = 10.2.
Where will the randomization distribution be
centered?
a) 10.2
b) 12
c) 45
d) 1.8
Randomization
distributions are
always centered
around the null
hypothesized value.
Randomization Distribution
In a hypothesis test for H0:  = 12 vs Ha:  < 12,
we have a sample with n = 45 and  = 10.2.
What will we look for on the randomization
distribution?
a)
b)
c)
d)
e)
We want to see how
extreme the observed
statistic is.
How extreme 10.2 is
How extreme 12 is
How extreme 45 is
What the standard error is
How many randomization samples we collected
Randomization Distribution
In a hypothesis test for H0: 1 = 2 , Ha: 1 > 2
sample mean #1 = 26 and sample mean #2 = 21.
What do we require about the method to
produce the randomization samples?
a) 1 = 2
b) 1 > 2
c) 1 =26, 2 =21
d) 1 − 2 = 5
We need to generate
randomization
samples assuming the
null hypothesis is true.
Randomization Distribution
In a hypothesis test for H0: 1 = 2 , Ha: 1 > 2
sample mean #1 = 26 and sample mean #2 = 21.
Where will the randomization distribution be
centered?
0
b) 1
c) 21
d) 26
e) 5
a)
The randomization
distribution is centered
around the null
hypothesized value,
1 - 2 = 0
Randomization Distribution
In a hypothesis test for H0: 1 = 2 , Ha: 1 > 2
sample mean #1 = 26 and sample mean #2 = 21.
What do we look for in the randomization
distribution?
a) The
standard error
b) The center point
c) How extreme 26 is
d) How extreme 21 is
e) How extreme 5 is
We want to see how
extreme the observed
difference in means is.
Randomization Distribution
For a randomization distribution,
each simulated sample should…
• be consistent with the null hypothesis
• use the data in the observed sample
• reflect the way the data were collected
Randomized Experiments
In randomized experiments the “randomness” is the
random allocation to treatment groups
• If the null hypothesis is true, the response values
would be the same, regardless of treatment group
assignment
• To simulate what would happen just by random
chance, if H0 were true:
Reallocate cases to treatment groups, keeping
the response values the same
Observational Studies
In observational studies, the “randomness” is
random sampling from the population
To simulate what would happen, just by random
chance, if H0 were true:
Simulate drawing samples from a population in
which H0 is true
How do we simulate sampling from a population in
which H0 is true when we only have sample data?
Adjust the sample to make H0 true, then bootstrap!
Body Temperatures
Let   the average human body temperature
H0:  = 98.6
Ha:  ≠ 98.6
sample mean = 98.26
• Adjust the sample by adding 98.6 – 98.26 = 0.34
to each value. The sample mean becomes 98.6,
exactly the value given by the null hypothesis.
• Bootstrapping the adjusted sample allows us to
simulate drawing samples as if the null is true!
Body Temperatures
In StatKey, when we enter the null hypothesis,
this shifting is automatically done for us
StatKey
p-value
= 0.002
Exercise and Gender
Do males exercise more hours per week than females?
sample mean difference
xm– xf = 3
1. State null and alternative hypotheses
2. Devise a way to generate a randomization
sample that
• Uses the observed sample data
• Makes the null hypothesis true
• Reflects the way the data were collected
Exercise and Gender
1. H0: m = f
Ha: m > f
2. Generating a randomization distribution can
be done with the “shift groups” method:
• To make H0 true set the sample means
equal by adding 3 to every female value.
Now bootstrap from this modified sample
Note: There are other ways. In StatKey, the
default randomization method is “Reallocate
Groups”, but “Shift Groups” is also an option.
Exercise and Gender
p-value =
0.095
Exercise and Gender
The p-value is 0.095. Using α = 0.05, we
conclude….
a) Males exercise more than females, on
average
b) Males do not exercise more than females,
on average
Do not reject the
c) Nothing
null… we can’t
conclude anything.
Blood Pressure and Heart Rate
Is blood pressure negatively correlated with heart rate?
sample correlation
r = -0.037
1. State null and alternative hypotheses
2. Devise a way to generate a randomization sample
that
• Uses the observed sample data
• Makes the null hypothesis true
• Reflects the way the data were collected
Blood Pressure and Heart Rate
1. H0:  = 0
Ha:  < 0
2. Generating a randomization distribution:
Two variables have correlation 0 if they are
not associated (null hypothesis). We can
“break the association” by randomly shuffling
one of the variables.
Each time we do this, we get a sample we
might observe just by random chance, if there
really is no correlation
Blood Pressure and Heart Rate
p-value =
0.219
Even if blood pressure and
heart rate are not correlated,
we would see correlations this
extreme about 22% of the
time, just by random chance.
Randomization Distributions:
Cocaine Addiction (randomized experiment)
 Rerandomize cases
to treatment groups, keeping
response values fixed
Body Temperature (single mean)
 Shift to make H0
true, then bootstrap
Exercise and Gender (observational study)
 Shift to make H0
true, then bootstrap
Blood Pressure and Heart Rate (correlation)
 Randomly
shuffle one variable
Generating Randomization Samples
• As long as the original data is used and the
null hypothesis is true for the randomization
samples, most methods usually give similar
p-values
• StatKey generates the randomizations for us.
We will not be concerned with the details of
the process. It is enough to understand the
general principles.
Summary
 Randomization samples should be generated
•
•
•
Consistent with the null hypothesis
Using the observed data
Reflecting the way the data were collected
 The specific method varies with the
situation, but the general idea is always the
same

similar documents