Lecture 32: Solitary Confinement Experiment

advertisement
Lecture 32: Solitary Confinement Experiment
Solitary Confinement
Experiment
Solitary Confinement
Experiment
Response: Frequency of brain
waves.
Conditions:
12 to 36 Hz – alert.
8 to 12 Hz – relaxed inactivity.
Below 8 Hz – abnormal for
someone who is awake.
Solitary Confinement – Yes,
No.
Time – 1 day, 4 days, 7 days.
1
Solitary Confinement
Experiment
2
Solitary Confinement
Experiment
Treatment Combinations;
Experimental Material
There are 6 treatment
combinations: Yes,1day;
Yes,4days; Yes,7days;
No,1day; No,4days; and
No,7days.
20 prisoners at a maximum
security penitentiary.
3
4
Completely Randomized
Completely Randomized
If we were to use a completely
randomized design we could
use only 18 prisoners (has to
be a multiple of the number of
treatment combinations, 6).
Each of the 6 treatment
combinations would be
assigned, at random, to 3
prisoners.
5
6
1
Lecture 32: Solitary Confinement Experiment
Analysis of Variance
Source
Solitary
Time
Solitary*Time
Error
C. Total
Random Error
df
1
2
2
12
17
Differences among prisoners
treated the same will
contribute to random error.
For example, different
prisoners may have different
natural brain wave
frequencies.
7
Block Design
8
Comment
We could sort on natural brain wave
frequencies to make blocks of size 6.
Prisoners within a block are
more similar in terms of
natural brain wave
frequencies but will not be
exactly the same.
 Block 1: 6 prisoners with the highest
frequencies.
 Block 2: 6 prisoners with the next
highest frequencies.
 Block 3: 6 prisoners with the lowest
frequencies.
9
Randomization
10
Analysis of Variance
Randomly assign the 6
treatment combinations to
the 6 prisoners within each
block.
11
Source
Solitary
Time
Solitary*Time
Block
Error
C. Total
df
1
2
2
2
10
17
12
2
Lecture 32: Solitary Confinement Experiment
Error
Alternative Blocking
The sum of squares error in a
block design is the interaction
between treatments and blocks
(df = 5*2 = 10).
This is the inconsistency of
treatment effect across blocks.
Reuse prisoners so that each
prisoner experiences all 6
treatments.
The order of the treatments is
randomized for each prisoner.
13
Alternative Blocking
14
Analysis of Variance
Advantages: Can use all 20
prisoners. Reusing forms the
most uniform (homogeneous)
blocks.
Disadvantages: More time for
the experiment (24 days).
Difficult to run. What to do with
treatment carry over effects?
15
Source
Solitary
Time
Solitary*Time
Prisoner
Error
C. Total
df
1
2
2
19
95
119
Error
Repeated Measures
The sum of squares error in a
block design is the interaction
between treatments and
prisoners (df = 5*19 = 95).
This is the inconsistency of
treatment effect across
prisoners.
Whole plot/Between subject
factor: Solitary – Yes, No.
Sub plot/Within subject
factor: Time – 1, 4 and 7
days.
17
16
18
3
Lecture 32: Solitary Confinement Experiment
Completely Randomized
Comment
Randomly assign solitary
confinement to 10 prisoners
and no solitary confinement
to the other 10 prisoners.
If we didn’t have the second
factor of time, the analysis
would have two sources of
variation; solitary and error.
19
Random Error
20
Analysis of Variance
The sum of squares for error
comes from differences among
prisoners treated the same.
Solitary – 10 prisoners – 9 df
No solitary – 10 prisoners – 9 df
Source
Solitary
Error
C. Total
df
1
18
19
21
Analysis of Variance
Source
Solitary
Prisoners[Solitary]
C. Total
22
Block Design
df
1
18
19
23
Reuse the prisoners by
measuring their brain waves
on days 1, 4 and 7.
Repeatedly measure their
brain waves.
24
4
Lecture 32: Solitary Confinement Experiment
Analysis of Variance
Source
Solitary
Prisoners[Solitary]
Time
Solitary*Time
Error
C. Total
df
1
18
2
2
36
59
Error
The sum of squares error in a
block design is the interaction
between time variable and
prisoners (df = 2*18 = 36).
This is the inconsistency of the
effect of time across prisoners.
25
Comment
26
Comment
Variation among prisoners in
the same confinement does not
come into play when
evaluating the effects of time
or the interaction effects of
Solitary*Time.
Variation among prisoners
in the same confinement
(solitary or no solitary) is
accounted for separately in
the analysis of variance.
27
28
Comment
Comment
When evaluating the effects of
solitary confinement (Yes or
No) the variation among
prisoners in the same
confinement is the error term
and will probably be fairly
large.
It will be harder to see
differences in average brain
wave frequencies due to
confinement than it will to see
differences due to time or the
interaction between Solitary
and Time.
29
30
5
Download