20 Psychology Questions
Zambi
Ch13-2020_Student.pptxChapter 13
Alternative Research Designs
Learning Objectives
13.1 Review the importance of the internal validity to establish the cause-and-effect relationship in any experiment
13.2 Describe the single-case experimental design
13.3 Examine the role of quasi-experimental designs
2
Protecting Internal Validity Revisited
Objective 13.1 Review the importance of the internal validity to establish the cause-and-effect relationship in any experiment
3
13.1 Protecting Internal Validity Revisited
Examining Your Experiment From the Inside
- Extraneous variables
Control to identify relationships
Precautions
Evaluation
Watch for confounders
Nine threats
4
Protecting Internal Validity Revisited
Good experimental control leads to internally valid experiments.
Psychological Detective
Imagine you have been given responsibility for conducting the famous Crest test – you are supposed to determine whether brushing with Crest actually does reduce cavities. Your boss wants you to use an experimental group (Crest) and a control group (Brand X) in the experiment. What are at least five potential extraneous variables for this experiment?
13.1 Protecting Internal Validity Revisited
Protecting Internal Validity With Research Designs
- Random assignment
Create equal groups
Cannot guarantee equality
May have to use matched groups
Not the same as random selection
- Experimental design
7
Protecting Internal Validity with Research Designs
Campbell and Stanley (1966) recommended three experimental designs as being able to control the threats to internal validity
These designs are:
The Pretest-Posttest Control Group Design
The Solomon Four-Group Design
The Posttest-Only Control-Group Design
The Pretest-Posttest Control Group Design
The pretest-posttest control group design appears below:
(Control Group)
(Experimental Group)
Key:
R = Random Assignment
O = Pretest or posttest observation or measurement
X = Experimental variable or event
Each row represents a different group of participants.
Left-to-right dimension represents passage of time.
Any letters vertical to each other simultaneously.
The Solomon Four-Group Design
First proposed by Solomon (1949), this design is identical to the pretest-posttest control-group design with the first two groups but adds an additional two groups.
Because the Solomon four-group design has the same two groups as the pretest-posttest control-group design, it has the same protection against the threats to internal validity.
The main advantage gained by adding the two additional groups relates to external validity.
One problem with the Solomon design is that there is no statistical test that can treat all six sets of data at the same time.
The Posttest-Only Control-Group Design
The posttest-only control-group design can be extended by adding additional treatment groups as shown in this figure.
The posttest-only control-group design is a copy of the pretest-posttest control-group design but without the pretests included and is a copy of the two added groups in the Solomon four-group design.
Random assignment to groups equates the two groups and withholding the IV from one group to make it a control group make it a powerful experimental design that covers the threats to internal validity.
The Pretest-Posttest Control Group Design
This design consists of two randomly assigned groups of participants, both of which are pretested, with one group receiving the IV.
The random assignment of participants to groups allows us to assume that the two groups are equated before the experiment thus ruling out selection as a problem.
Factorial Treatment of Solomon Four-Group Design Posttest Scores
Finally, we could create a factorial design from the posttest-only control group design by combining two of these designs simultaneously so that we ended up with a block diagram similar to those from Chapter 11.
13
An Extension of the Posttest-Only Control-Group Design
This design allows for testing multiple treatment groups.
13.1 Protecting Internal Validity Revisited
A Final Note on Internal Validity
- Importance
Most important part of a study
Experiments intended to identify cause and effect
Cannot count on statistics to provide control
15
Single-Case Experimental Designs
Objective 13.2 Describe the single-case experimental design
16
Single-Case Experimental Designs
Single-case experimental design
An experiment that consists of one participant (also known as N = 1 designs).
Includes controls just as in a typical experiment.
Precautions are taken to insure internal validity.
History of Single-Case Experimental Designs: Examples
1860’s Gustav Fechner explored sensory processes on an in-depth basis with a series of individuals.
Wilhelm Wundt (founder of the first psychology laboratory) conducted his pioneering work on introspection with highly trained individual participants.
Hermann Ebbinghaus pioneered research in verbal learning and memory using himself as subject in a single-case design.
Single-case designs declined in popularity as statistical tests for group experiments were developed in the 1920’s.
History of Single-Case Experimental Designs
Less common in current literature
Inferential statistics
ANOVA
Uses of Single-Case Experimental Designs
There are still researchers who use single-case designs.
Founded by B.F. Skinner, the experimental analysis of behavior approach continues to employ this technique.
Skinner (1966) summarized his philosophy in this manner:
The Society for the Experimental Analysis of Behavior was formed and began publishing its own journals
Uses of Single-Case Experimental Designs
Reasons to use single-case experimental designs
Unique individual
If you can, assume perfect generalizability
When a single negative instance would refute a theory
Limitations on opportunities to observe a behavior
When research is very expensive, time-consuming, or needs extensive training
21
General Procedures of Single-Case Designs
Hersen (1982) listed three procedures that are characteristic of single-case designs:
Repeated measures
Baseline measurement
Changing one variable at a time
General Procedures of Single-Case Experimental Designs
Repeated Measures
When you are dealing with only one participant, it is important to make sure the behavior you are measuring is consistent. Therefore, you would repeatedly measure the participant’s behavior.
Hersen and Barlow (1976) noted that the procedures for measurement “must be clearly specified, observable, public, and replicable in all respects (p. 71).
Repeated measurements “must be done under exacting and totally standardized conditions with respect to measurement devices used, personnel involved, time or times of day measurements are recorded, instructions to the subject, and the specific environmental conditions” (p. 71).
General Procedures of Single-Case Experimental Designs
Baseline Measurement
A measurement of behavior that is made under normal conditions (e.g., no IV is present); a control condition.
Baseline measurement serves as the control condition against which to compare the behavior as affected by the IV.
Barlow and Hersen (1973) recommend that you collect at least three observations during the baseline period in order to establish a trend in the data.
General Procedures of Single-Case Experimental Designs
Changing one variable at a time
If you allow variables to change simultaneously, then you have a confounded experiment and cannot tell which variable has caused the change in behavior that you observe.
If you record your baseline measurement, change several aspects of the participant’s environment, and then observe the behavior again, you have no way of knowing which changed aspect affect the behavior.
Statistics and Single-Case Experimental Designs
The case against statistical analysis
The case for statistical analysis
26
Representative Single-Case Experimental Designs
Researchers use standard notation for single-case designs that makes the information easier to present and conceptualize.
Standard Notation:
A
B
13.2 Single-Case Experimental Designs
Representative Single-Case Experimental Designs
- A-B design
- A-B-A design
- A-B-A-B design
- Design and the real world
- Additional single-case designs
28
Representative Single-Case Experimental Designs
The A-B design
The A-B design is the simplest of the single-case designs.
We make baseline measurements, apply a treatment, and then take a second set of measurements.
The A-B design is poor for determining causality because of many of the threats to internal validity (e.g., history, maturation, and instrumentation).
Representative Single-Case Experimental Designs
The A-B-A design
In the A-B-A design
If a change in behavior during B is actually due to the experimental treatment, the change should disappear when B is removed and you return to the baseline condition.
If, on the other hand, a change in B was due to some extraneous variable, the change will not disappear when B is removed.
Thus, the A-B-A design allows a causal relation to be drawn.
If you end your experiment on an A phase, this leaves the participant “hanging” without the treatment.
ABA design
Plot of data from the hypothetical “talking to plants” ABA study
Representative Single-Case Experimental Designs
The A-B-A-B design
The A-B-A-B design
This design adds a final treatment period to the A-B-A design, thereby completing the experimental cycle with the participant in a treatment phase.
Hersen and Barlow (1976) point out that this design gives two transitions (B to A and A to B) that can demonstrate the effect of the treatment variable.
Thus, our ability to draw a cause-and-effect conclusion is further strengthened.
A-B-A-B Design
This graph represents talking-out behavior in a mentally retarded student.
Baseline1: before experimental conditions
Contingent attention1: systematic ignoring of talking out and increased teacher attention to appropriate behavior
Baseline2: reinstatement of teacher attention to talking-out behavior.
Contingent Attention2: return to systematic ignoring of talking out and increased attention to appropriate behavior.
Miller & Kelley’s (1994) ABAB homework study
Representative Single-Case Experimental Designs
Design and the real world
From the preceding sections it should be clear that the A-B-A-B design is the preferred design for single-case research.
However, we must ask whether typical practice actually follows the recommended path.
Hersen and Barlow (1976) acknowledged that researchers often use the A-B design despite its shortcomings in terms of demonstrating causality.
The main reason the A-B design is used concerns either the inability, undesirability, or unethical to return to baseline in the third stage.
Hall, Alley, & Cox’s (1971) “newlywed husband” ABABA study
Hypothetical Multiple Baseline Design
Representative Single-Case Experimental Designs
This picture is of a 9-month-old boy who was hospitalized for frequent vomiting (left picture) and after treatment (right picture; 13 days later).
Lang and Melamed instituted a treatment consisting of brief and repeated shocks applied to the boy’s leg at the first signs of vomiting and ended when the vomiting ceased.
By the third treatment session, on or two brief shocks were enough to stop the vomiting.
By the fourth day of treatment, vomiting stopped, so treatment was discontinued.
Two days later, some vomiting occurred, so the procedure was reinstated for three sessions. Five days later, the child was dismissed from the hospital.
Quasi-Experimental Designs
Objective 13.3 Examine the role of quasi-experimental designs
39
Quasi-Experimental Designs
When we are able to manipulate an IV and measure a DV but cannot randomly assign our participants to groups, we must use a quasi-experimental design.
History of Quasi-Experimental Designs
Campbell and Stanley (1966)
Researchers used it before
Term originated later
41
Uses of Quasi-Experimental Design
Hendrick et al. (1993) listed several specific situations that require quasi-experimental designs:
Variables that make random assignment impossible
Evaluate an ongoing program or retrospective study
Studies of social conditions
Random assignment impossible due to expense, time, etc.
Ethics
Representative Quasi-Experimental Designs
Because quasi-experimental designs resemble true experiments, the use of statistics is not an issue
The traditional statistical tests used with true experiments are also appropriate for quasi-experiments.
Nonequivalent group design
Interrupted time-series design
Nonequivalent Group Design
You will see that the nonequivalent group design bears a distinct resemblance to the pretest-posttest design.
However, the nonequivalent group design is missing the Rs in front of the two groups; random assignment is not used to create the groups.
The lack of random assignment means that the groups may differ before the experiment.
Nonequivalent Group Design
You also will notice that the two groups are labeled as the comparison group (rather than control) and treatment group (rather than experimental).
Changing treatment to experimental is not particularly important, those terms can be used interchangeably.
Changing the name from control to comparison is important and meaningful.
Nonequivalent Group Design
It is possible to extend the nonequivalent group design to include more than one treatment group if you wish to contrast two or more treatment groups with your comparison groups.
The key to nonequivalent group design is creating a good comparison group.
We attempt to create an equal group through our selection rather than through random assignment.
Interrupted Time-Series Design
This design involves measuring a group of participants repeatedly over time (the time series), introducing a treatment (the interruption), and measuring the participants repeatedly again (more of the time series).
Potential Changes in Trend in a Time Series Design
Change in level, no chance in rate (part A)
No change in level, change in rate (part B)
Change in level, change in rate (part C)
Interrupted Time-Series Design with Control Group
Although the interrupted time-series design can control for some of the internal validity threats, we still face the potential threat of history.
This threat is typically handled in one of the three manners (Cook & Campbell, 1979):
Frequent testing intervals
Include a comparison (control) group that does not receive the treatment.
The third solution, although probably the best, is not always possible. In essence, the solution involves using an A-B-A format within the interrupted time series design.
Upcoming
Chapter 9: Using Statistics to Answer Questions
50
End of Chapter 13
Have a nice day!
