Chapter 14 Flashcards
randomized experiment
Assigning a group randomly to the control or experimental group
placebo
Control group receives
Inert or ineffective treatment
experimental subject/unit
Units randomized to control or experimental group
random assignment
Used to provide or withhold an intervention or treatment following a procedure to flipping a coin
Researcher must randomize assignment of the IV
Participants randomly assigned to treatment and control groups
Helps ensure all groups are identical on all z variables
Have random assignment but not random sample
statistical equivalence
Balances the treatment and control groups in terms of both measured and unmeasured variables with bounds of chance as differences
Experiments are the “gold standard” for getting close to causality
Note: we still can’t observe causation here. The book does a bad job of explaining this
But, out of all the research designs we have, it the design that generally gives us the most confidence in arguing for cause and effect
haphazard randomization
Less formal procedure that mimics true random assignment
why random assignment?
Statistical equivalence on all variables
randomized experiment
Seeks to isolate the causal relationship between one explanatory variable, the treatment (x) and one or more response variables or outcomes (y)
Treatment → outcome or x → y
exogenous
X, the treatment is known to independent from and uncorrelated with all the other variables causing y
experiments have what kind of internal and external validity
high internal low external
internal validity
Confidence that the effect of X on Y is real
external validity
Confidence that relationship between X and Y generalized beyond experiment
threats to internal validity
Internal - differences between participants in treatment and control groups (other than X)
selection
Individuals selected into treatment are different from those in control
experiment mortality
Ppl. turning out and not receiving treatment
Only threat if creates difference between treatment and control group so if don’t mess up random assignment should be no threat
threats to external validity
any differences between participants in study v. non participants and setting of lab v. real world
unrepresentative population
doesn’t represent real world
realism
setting we are in for experiment does not look anything like real world where might get this treatment
mundane realism
events look way they do in real world
experimental realism
subjects don’t take the experiment seriously
psychological realism
the events triggered the same thoughts and emotions and processes as real-world events
Hawthorne effect
subjects behave differently when they know they are being watched or monitored
lab experiments
Take place in lab or specialized location
Intervening to create exogenous change in one variable while holding other variables constant
field experiments
Takes place in real world setting not in artificial lab or simulation like school sor businesses → substantial time, organization
survey experiments
Method of incorporating an experiment in the design and structure of the survey instrument itself
Often used to test effects of variation in questionnaire or wording also referred to as split-ballot survey
split-ballot survey
print off copies of survey A and survey B and randomly assign
cluster randmoizibiliity
Groups or clusters rather than individuals are randomized
Helps prevent leakage of the treatment to the control group
Less statistical equivalence from randomization
arms experiment
Various possible treatments
Basic experiment has two arms - treatment and control
Randomized study can compare multiple possible treatment or control conditions
does-response relationship
How the effect of the experiment varies as dosage increases
Ex. large class size v. small, but how large?
linear trend
large the does larger the response
curvilinear trend
as when an increasing dose reaches a point of diminishing response
abrupt trend
effect kicks in all at once only when the dosage reaches a particular level
analysis of randomized experiments
Usually just a difference of means
sample size
Number of units in each group
Each group must be large enough to achieve balance and to get desired precision of treatment effect
Can use power analysis to determine sample sizes
drawbacks of randomized experiments
Not all IV’s are controllable
Ethical considerations prevent some experiments
Low external validity
Can resolve some of these problems with natural and quasi experiments
factorial experiments
Experiments that try to measure the effects of more than one independent variable called factors
main effects
Average of one factor across levels of the other factors in experiment
interaction effects
Effect of both factors together is more or less than the sum of the two main effects
between subjects experiments
When a group could end up i either one group or the other but not both
Comparison is between subjects or units in treatment or control group
within-subjects experiments
Experiment applies to two or more treatments to same people at different times
Experimental conditions are within each person or unit so they each are exposed to both treatment and control conditions
repeated - measures experiments
Repeated outcomes measures taken on each sub experimental and simply observe people repeatedly over time
unobtrusive measures
Measures that occur naturally or routinely or otherwise do not disturb the subjects in an experiment
blind experiments
Help correct the problem of the placebo effect by keeping subjects ignorant of which group they are in
double-blind experiments
Researcher and participant don’t know what groups are which
contamination
When something happens in the course of implementing the study that compromises the integrity and logic of the experiment
noncompliance
Treatment group fails to fully follow the treatment
attrition
Participants dropping out from experiment
minimal detectable effect
Smallest effect that would still have practical significance
heterogenous treatment effect
Different people respond differently to treatment
average treatment effects
Suggest the average of the various individual effects that the treatment ha on each person who participates in study