Psychology 7

1
Psychology 7 Experimental Psychology
Quasi-Experiments
2
Quasi-Experimental Designs

• Quasi-Experiments Designs that lack all of the
  control features of true experiments
• A research design in which an experimental
  procedure is applied but all extraneous variables
  are not controlled.
• Definitions vary from person to person
• Inability to randomly assign participants to
  conditions is the most common reason for the use
  of quasi-experiments
• Quasi-experiments do not allow same confidence in
  causal inferences as true experiments (lower
  internal validity)

3
One Group Pretest-Posttest Design

• Same as pretest-posttest design but no random
  assignment to multiple conditions (no control
  condition)

Hot Dogs Increase IQ Scores (National Enquirer
Headline)
Take Mini-IQ test Eat a hot dog
Re-Take Mini-IQ test
Design is quasi-experimental because of the lack
of a control condition
4
Problems with One Group Pretest-Posttest

• History effects Any event that happens between
  pre and posttest is confounded with the
  manipulation
• Maturation Systematic changes in individuals
  over time confounded with the manipulation (major
  problem in research on children)
• Testing Any effects of pretest on posttest
  (e.g., practice)
• Instrument Decay Posttest measurement of
  dependent variable has changed since pretest
  measurement (e.g., observers changed how easily
  they observe certain behaviors)
• Regression toward the mean Problematic when
  subjects are chosen because they score high or
  low on some measure

5
Regression to the Mean

• Observed Score True Score Measurement Error
• e.g., SAT scores True Ability Chance
  (guessing, other)
• If score is extreme, more likely that
  guesses/other factors were in extreme direction
  (positive or negative)
• Next testing, on average, likely to reflect mean
  performance
• People often misperceive instances of statistical
  regression
• Belief that punishment improves behavior, praise
  hurts it

6
Sports Illustrated Jinx

• Team performance is imperfectly correlated from
  week to week.
• Excellent performance one week is likely to be
  associated with more average performance next
  week.
• Giants on cover of SI prior to Super Bowl in 2001

7
Sports Illustrated Jinx

• Team performance is imperfectly correlated from
  week to week.
• Excellent performance one week is likely to be
  associated with more average performance next
  week.

8
Sports Illustrated Jinx
May 28 Cover
July 2 Letter

• Well, youve done it. The Mariners have a chance
  to go all the way, and you have to put Suzuki on
  the cover. There goes the pennant, and probably
  there goes Ichiro. Just how much did George
  Steinbrenner pay you for this grand favor?
• Ed Griemsmann
• Bellinghamn, WA

9
Problems with One Group Pretest-Posttest

• History effects Any event that happens between
  pre and posttest is confounded with the
  manipulation
• Maturation Systematic changes in individuals
  over time confounded with the manipulation (major
  problem in research on children)
• Testing Any effects of pretest on posttest
  (e.g., practice)
• Instrument Decay Posttest measurement of
  dependent variable has changed since pretest
  measurement (e.g., observers changed how easily
  they observe certain behaviors)
• Regression toward the mean Problematic when
  subjects are chosen because they score high or
  low on some measure

10
Nonequivalent Control Group Design
Participants
Dependent Variable
Treatment
Participants
Dependent Variable
No Treatment
11
Brain Damage Patients Require Nonequivalent
Controls

• E.g., damage to the medial temporal lobes
  (hippocampus) appears to selectively impair the
  declarative memory system
• Anterograde amnesia Loss of ability to form new
  long-term memories after brain injury
• Retrograde amnesia Loss of long-term memories
  that occurred before the brain injury

12
Patient K.C.
- college graduate quality control for an
engineering plant.
- closed head injury due to a motorcycle accident
at age 30.
- no effect on his intelligence, but he had
severe retrograde and anterograde amnesia.
- only affected his episodic memory.
13
Experimental Design Problems in Testing
Brain-Damaged Patients
Humans with hippocampus damage display severe
spatial memory impairments in a virtual Morris
water task. Astur, Taylor, Mamelak, Philpott,
Sutherland Behavioral Brain Research (2002)
14
Experimental Design Problems in Testing
Brain-Damaged Patients
Nonequivalent control group
Subjects matched for age, education level, and
intelligence.
15
Experimental Design Problems in Testing
Brain-Damaged Patients
Nonequivalent control group
This is about as well as a patient study can be
controlled.
16
Quasi-Experiments

• One Group Pretest-Posttest Design
• IQ-hot dogs-IQ
• Problems History, maturation, testing,
  instrument decay, regression to the mean
• Nonequivalent control group design (posttest
  design)
• E.g., brain damage patients and matched control
  group
• Nonequivalent control group pretest-posttest
  design

17
Nonequivalent Control Group Pretest-Posttest
Participants Pretest Treatment Posttest
Participants Pretest Control Posttest
Participants are not randomly assigned
nonequivalent control group
18
Flu Shot Study

• Background Testosterone appears to regulate
  energy distribution
• Promotes fat breakdown and energy investment in
  muscle construction
• Decreases energy investment in immune function
• Evidence in nonhuman species that immune
  activation decreases testosterone production
  (makes more energy available for immune response)
• Hypothesis Flu shots will cause testosterone
  decreases (at post 2 weeks when antibody response
  is peaking)

19
Nonequivalent Control Group Pretest-Posttest
Participants Pretest Treatment Posttest
Participants Pretest Control Posttest
If assignment is random to flu shot vs. no flu
shot then true experiment, pretest/posttest
design If assignment is nonrandom, then have a
quasi-experiment, nonequivalent control group
pretest/posttest design
Which design has greater internal validity? Why?
20
Pretest Marital Satisfaction Share feeling
program Posttest
Pretest Marital Satisfaction No program
Posttest
21
Selection-Regression Effect?
22
Time Series Designs

• Similar to pretest-posttest designs, but take
  multiple measures before and after treatment
• Pretest measures Treatment Posttest measures
• Y1, Y2, Y3, Y4, Y5 Y6, Y7, Y8, Y9, Y10

12
10
8
Control Series
6
Marital Satisfaction
4
2
0
1
2
3
4
5
6
7
8
9
10
Pre vs. Post Feelings Program
23
Global Warming and Interrupted Time Series
24
Global Warming and Interrupted Time Series
25
Global Warming and Interrupted Time Series
26
Single Case Experimental Designs

• Different from case studies because investigator
  introduces some form of manipulation
• Employ time series designs
• ABA design
• baseline measures, treatment measures, baseline
  measures

27
Hypothetical ABA Design

• Contents of Aarons Lost Luggage
• Items Price
• 16 polo shirts 50 each
• 10 fancy shorts 60 each
• Hair Gel Priceless

28
A
B
A
29
Example of ABA Design
Walker Buckley, 1968
30
ABAB Design
Quattrochi-Tubin Jason, 1980
31
Limitations of Reversal Designs

• Some effects of treatments are difficult to
  reverse
• Reversal designs most useful for seeing
  short-term effects of treatments
• Multiple baseline designs
• Introduce treatment at different times for
  different subjects
• Introduce treatment at different times for
  different behaviors
• Introduce treatment at different times for
  different situations

32
Example of Multiple Baseline Design
Saigh, 1986