Quant quiz 2 Flashcards
Quasi – Experimental Designs
Used when control features of experimental
design cannot be achieved
*Independent variable cannot be manipulated, or
*Random assignment to groups cannot be achieved
*Internal validity may be affected
Quasi Example- One-Group Posttest-Only Design
- Lacks a control/comparison group
- Lack baseline data
- Difficulty making inferences about the effect of the IV on the
DV (low internal validity)
Quasi Example- One-Group pretest-posttest design
- Now there is baseline comparison data, but no control/comparison group
- Threats to the internal validity of these types of studies:
- History
- Maturation
- Testing
- Instrument Decay
- Regression towards the mean
Threats- history
Refers to any event that occurs between the 1st and 2nd
measurement (not a part of the experiment)
- Any confounding event that occurs at the same time as the
experimental manipulation - EXAMPLE:
- change in political leader
- terrorist attack
- natural disaster
- Global pandemic!
Threats- maturation
People change over time
- Short periods of time:
- Boredom
- Fatigue
- Hunger
- Wiser
Long periods of time:
* Increased coordination
* Increased analytical skills
* Change of priorities/core values
* Life events
Threats- testing
Testing may be acting as an
intervention on its own
- A pre-test may sensitize participant in
unanticipated ways and their
performance on the post-test may be
due to the pre-test, not to the
treatment, - Or, more likely, and interaction of the
pre-test and treatment
Threats- Instrument decay
- The characteristics of the
measurement change over time - Ex: observation or self-
monitoring - The rater changes their rating
behavior over time due to any
number of reasons - Scale needs calibration
Threats- Regression towards the mean
- When participants score really high or really low on a
measure, they are likely to score closer towards the mean
when retested at a different time - Statistically, outliers tend to be less extreme when tested again
Nonequivalent control group
design
- Separate control group, but the participants in the two conditions are not equivalent
- i.e., they were not randomly assigned so we cannot guarantee equivalence…
- Selection Difference/ bias
- The groups are not a result of random assignment but are from existing natural groups
- Thus, not matter what, are vulnerable to selection bias
- Example:
- Treatment group are people who volunteered for treatment and the control group are people who just meet criteria
- There might be group differences between the groups of volunteers
Nonequivalent Control Post-test
Group Design
*Participants are not randomly assigned to groups
*No pre-test
Nonequivalent Control Group
Pretest-Posttest Design
*Still not randomly assigned groups, BUT the model
is improved with a pretest
*Not randomly assigned but we can test for some
equivalence using the pretest
Propensity score matching of
nonequivalent treatment and control
groups
*Nonequivalent control group problem: groups can
differ in important ways
*Improve with:
* Score matching
* matches one participant from the experimental group with
one person from the control group on potential confounding
* E.g., Age
* Propensity score
* scores on multiple variables are combined to produce a
propensity score and then participants are matched using that
number
* E.g., Age, level of education, ethnicity
Why control/comparison groups,
equivalent or non-equivalent, are so
important
Can see if these threats to
internal validity affect the
control group AND the
treatment group!!
Single Case Experimental
Designs
Single Subjects
- Made popular from operant conditioning research (B.F. Skinner)
- Now used in ABA (applied behavior analysis)
- Does an experimental manipulation have an effect on a single research participant?
- You can use this with individual therapy clients to help foster insight into their own patterns. Have them do their own single case experiments to made behavioral changes (stop a bad habit)
Single Case Experimental
Designs
*Baseline (A):
*Observed behavior before manipulation
*Treatment (B):
*Introduce manipulation/treatment
*Many measurement timepoints
*Change in behavior from A -> B presumes a
treatment effect