Quasi-experiment
differs from a true experiment in that the researchers do not have full experiments control
--still studying the effect of an IV on a DV
-often because people can not be assigned to specific levels of intended IV-gender, age (participant variables)
--why do people do them?
---convenience, ethics, external validity (at expense of internal)
-opportunity to observe-real world effect
Independent-Groups Quasi-experiment
different participants at each level of the independent variable
-nonequivalent control group design
-nonequivalent control group pretest/posttest
Nonequivalent control group design
a quasi experiment study that has at least one treatment group and one comparison group--but participants have no been randomly assigned to the two groups
--independent groups (between subjects)
-Head start study -match students based on school exams- one who qualifies financially for head start and one that doesn't- test both at later date
Nonequivalent control group pretest/posttest design
participants not randomly assigned and were tested before and after the intervention
-plastic surgery example-tested on life satisfaction before and after surgery (tested twice)
--group only exposed to 1 level-people either get the surgery or don't
-only measured twice
-you compare one group (non-surgery group) to the other (surgery group)
Repeated Measures Quasi Experiment
-within subjects
--same group of people exposed to different manipulations of the IV
-different from an experiment because researcher takes advantage of an already scheduled event/policy/chance occurrence to manipulate IV
--ex:interrupted time-series design, nonequivalent control-group interrupted time-series design
Interrupted time-series design
repeated measures quasi experiment (within subjects)
-a quasi experimental study that measured participants repeatedly on a DV before, during, and after an "interruption" caused by event
-measure the DV after each event (IV) (at least 3 times)
--ex:decision fatigue/judge granting parole: DV=proportion of criminals granted parole IV/event=snack/lunch
-television access and crime rate, crime rate measured before, during, and after introduction of television
-one continuous line on a graph-should show change over time (after introduction of IV)
-you compare measurement 1 of DV to measurement 2 of the DV-for the same group
Non-equivalent control group, interrupted time-series design
combines nonequivalent control group design (independent) with interrupted time series design (within-subjects)
-just like interrupted time series:measure DV before, during, and after introduction of IV (within subject) -measured crime rates before/during/after introduction of TV
-but also have a control group (nonequivalent control group) -also compared larceny rates in cities that had television with cities that did not
-two lines on a graph-experimental group should show change over time, and also show a greater change than the control group (to prevent against maturation threats)
Internal Validity problems in Quasi-experiments and how to rule out:
-selection effects, maturation, attrition, history, regression, testing/instrumentation threats
-design confounds
-observer bias, demand characteristics, and placebo effects
---good experimental designs, and pattern of results can help rule out internal validity threats
Selection Effects
threat to internal validity
-only relevant for independent groups design
-occurs when the groups at the various levels of the IV contain different types of participants
-->it's not clear whether the IV or the different types of people caused the change in the DV
-ex: head start -experimental group had impoverished children
--solved with pretest/posttest, matched groups, or wait-list design
Wait-list design
-solution to selection effects (threat to internal validity) with between groups (independent-groups) design
-all participants plan to receive treatment, but are assigned to do so at different times
--helps het same kind of people in both groups
-cosmetic surgery example
--can still be unethical depending on what the treatment is
-random assignment? hopefully
Design confounds
-internal validity threat
--some outside variable accidentally and systematically varies with the levels of the targeted IV
-can lead to an alternate explanation
--inspect data carefully to rule out design confounds
Maturation threat
-internal validity threat
--quasi experiment with pretest posttest design, when a treatment group shows an improvement over time, but it's unclear if the improvement was caused by the treatment of whether the group would have improved spontaneously without the treatment
--design (include a comparison group) + pattern of results (different between the two groups) help eliminate maturation threats
History Threat
-internal validity threat
-when an external/historical event happens for everyone in the study at the same time as the treatment variable
--can be especially relevant in quasi-experiments that rely on external factor to manipulate the key IV (ccc television broadcasting policies)
-recession possible history threat for TV introduction and crime rates
--comparison group--recession happened for all cities, but only cities with television had significant rises in larceny rates
--selection-history threat=only happens in one group
Regression to the Mean
-internal validity threat
--when an extreme finding is caused by a combination of random factors that are unlikely to happen in the same combination again
-head start-to create matched groups-paired high scoring head start kids with low scoring non-head start children--scores may have been caused by random factors--low scoring non-head start kids didn't get a good night sleep
--wouldn't happen again at the posttest-->regression to mean
-relevant for pretest/posttest-primatily when a group is selected for having extremely high or low scores (don't choose high/low scores=solution)
--not usually a problem in true experiments because of random assignment
Attrition Threat
-internal validity threat
--pretest/posttest when people drop out over time
-only a problem when people drop out systematically
--ex:least satisfied people stop reporting to questionnaires in cosmetic surgery exp.
-just drop pretest scores of people you don't get posttest scores from
Testing Threat
internal validity threat
-a problem when the DV is measured more than once (pretest/posttest or within [repeated measures])
--participant changes as a result of being tested before
-practice effect/fatigue effect
--comparison group helps
Instrumentation Threat
internal validity threat
-a problem when the DV is measured more than once
--when a measuring instrument changes over time
-comparison group helps
--instrumentation threat was the actual result in the parole hearing and lunch break exp.
Observer Bias
-internal validity threat
--also threatens construct validity (observer's interpretations don't reflect real levels of DV)
--expectations of the observer influence how they interpret the results
-especially relevant for behavioral studies
-double blind can help
-carefully train coders+manual+multiple codes
Demand Characteristics
internal validity threat
-participants guess what the study is about and change their behavior in the expected direction
--double blind study
Placebo effect
internal validity threat
-when participants experience actual change, but not as a result of the IV-as a result of their belief that they are receiving a legitimate treatment
-double blind+comparison groups
External Validity in quasi experiments
real world setting increase external validity
-still need to ask if experiments would generalize to other people/situations
--dont have to ask if it would generalize to real world situations
Ethics of Quasi Experiments
-isn't ethical to withhold things like head start with random assignment-cant to a true experiment
-even wait list designs still questionable
Construct Validity in Quasi Experiments
usually have successful manipulation of the IV (how well was the IV manipulated)
--how well was the DV measured is more important
-can be questionable in behavioral DV-with observer bias
Statistical Validity in Quasi Experiments
-need to address how large the group differences were (effect size)
-were the results statistically significant
Quasi-Experiments vs Correlational Studies
-independent groups design are similar to correlational studies
--quasi=people not randomly assigned
--correlational=two measured variables
-have similar internal validity questions
--quasi=selection effects/design confounds
--correlational=lurking variables
-quasi-experiments attempt to have internal validity-matched groups, wait-list designs, or a natural comparison group
Small-N designs
-experiments that study only a few individuals
--external validity=how a sample is selected (more important that number of people in sample)
-instead of gathering a little information from a large sample, they obtain a lot of information from just a few cases
-researchers often study special cases, taking advantage of special medical cases (similar to quasi experiment)-ie split brain
--almost always repeated measures
Single-N design
sample is restricted to one animal/person
Differences between large/small-N groups
-Large N: participants are grouped and data is represented as a group average
-small N: each participant is treated as a separate experiment (almost always repeated measures) and individuals data is presented
Research on Split Brains
corpus callosum cut in severe seizure patients
--people with condition were studied to see how the different hemispheres of the brain act independently
-speech=left hemisphere/right eye
-left hemisphere responsible for speech, storytelling/cause and effect, and sense of self
--how much can you conclude from such a small N sample
---experimental control, strong manipulation, and replication help you conclude more
How much can you conclude from a small N?
experimental control, strong manipulation, and replication help you conclude more
Experimental control
-how much can you conclude from such a small N sample
control possible third variables?
-controlled the movement of participants eyes to control which hemisphere saw what
Strong manipulations
--how much can you conclude from such a small N sample
-making sure you manipulate the IV enough to cause a significant change in the DV (making your levels different enough from one another)
-tasks that would be easy for normal brain people and impossible for split brain people were used (showed the change in DV)
Replication
-how much can you conclude from such a small N sample
--are the results among your small number of participants similar?
-results were replicated among 5 split-brain patients
Disadvantages of Small-N designs
-a few people may not represent the whole population well (data was consistent among 5split-brain patients, but they were all severely epileptic=systematically different brains that the general population)
- it is often unethical to create the necessary comparison group
Triangulation
similarities in animal studies, brain imaging, and weight of evidence can support a claim and lead to a parsimonious theory (about brain organization)
-external validity
Behavior Change studies
small N designs that attempt to modify behavior
-alzheimer's new memory strategy, special ed teacher discourages face touching,
--can have many internal validity questions
Stable-baseline design
Behavior Change studies (small n)
-a researcher observes behavior for an extended baseline period before beginning a treatment
--if baseline is stable-behavioral change during the treatment period was more likely a result of the IV
-better than single pretest (in pretest-posttest) extended pretest period lowers the chance that a single spontaneous event occurred at the same time as the pretest and changed the behavior
(rule our maturation/history/and regression threats)
Multiple baseline design
Behavior Change studies (small n)
-researchers stagger their introduction of an intervention across a variety of contexts, times, and situations
--special ed girl touched her face/objects/hair- introduced overcorrections for each behavior separately -each decreased after overcorrection was applied (overcorrection works)
-if a single event was responsible for the change, all of the behaviors would have changed at once instead of staggering with overcorrection
-baseline can also be different situations-correct a behavior at work/home/school
---can be different people, behavior modification for 3 children in the same class, one at a time
-provide automatic comparison condition
Reversal Designs
Behavior Change studies (small n)
-researcher observes subject with and without treatment but takes the treatment away for a while (reversal period) to see whether the problem returns (reverses)
--test internal validity-if treatment is causing change, problem should return when it is removed
-only really works when treatment doesn't cause lasting change
--ex:doesn't work for education-cant relearn the alphabet
-questionable ethics of removing a treatment that works (depression)
Validities in Small-N designs
-internal validity because of repeated measures/within group, replication, experimental control, strong manipulations, pretest/posttest, baseline designs and reversals
-external validity-triangulation + weight of evidence + specific population to generalize results to, and can still be useful without generalizing
-construct-multiple observers + interrater reliability for behavior
-statistical-graphs, effect size (by what margin did client's behavior improve) not traditional statistics