WEEK 5- 10.1 experiments Flashcards
key components of experiments
3
- randomized intervention
- random assignment
- design- between or within participants
between participants
participants assiged to different groups, the different groups are compared to eachother. (each gets 1 treatment)
within participants
participants all receive the same intervention, but have different sequential orders of the treatment.
types of experiments by setting
4
- labratory
- field
- natural
- survey
labratory experiment
artificial setting, standardized, full control
field experiment
embedded in the real world, less control
treatment is still randomized
control less complete as people are exposed to other factors
natural experiment
technically not an experiment due to lack of random assignment
merely a comparison between groups which cannot be seen as equivalent due to non-random assignments
survey experiment
embedded in population survey
you get advantages of a representative population survey while testing causal mechanisms using randomized treatments
basic experimental designs
- post test only
- pretest- post test
post test only design
random assign ent into two groups
g1 has a manipulation of the treatment variable leading to y1
control group leading to y2
you don’t know whether the groups are equivalent
pretest postest design
random assignment to two groups
measuring the dv before and after the treatment effect
allowing for analysis on the changes between pre-test and posttest
=> allows you to precisely see if the outcome measure changes in resopnse to the treatment, controlling for the baseline measurement in the begining
other experimental designs
3
- solomon four group
- delayed effects
- factorial
prospect theory
when people see a positively framed scenario, they become risk averse, and go with the safe option
when people see a negatively framed scenario, they become risk taking, and go with the risky option
counterfactual assumption
assuming that if the cause or treatment were different, the outcome or decision would be different too.
- fundamental problem of causal inference\
unit homogeneity
for testing our causal r.ship, we can resolve this if we have an equivalent case or unit that would be exposed to another cause, and we hope wo will take decision b
=> the cause has a consistent effect
=> homogeneity is when they react the same
conditional independence
IV is assigned independent of DV
1. no endogeneity (reciprocal causation)
2. no selection bias
3. no ommited variable bias (confound)
assumptions of experiments
2
- same unit, different treatment
- conditional independence (iv is independent of dv)
matching (quasi)
used when participant number is low, and the researcher wants to make sure that the experimental groups are as equivalent as possible.
solomon 4 group design
participants are randomly assigned to four groups (2 treatment, 2 control)
=> you can determine if the pretest sensetizes the participants
delayed effects design
in one treatment and control group the outcome is measured immediately, and the other treatment and control group is measured only later.
why is factorial design used the most?
used most often, bc there’s more than one variable that needs consideration
factorial design
- CONTROL: control group w/ pre test and post test measurement of y
- MAIN EFFECT 1: a treatment group where only the first experimental treatment is present
- MAIN EFFECT 2: a second treatment group where the second experimental treatment is present
- INTERACTION EFFECT: third treatment group where both treatments are present
factorial design
- CONTROL: control group w/ pre test and post test measurement of y
- MAIN EFFECT 1: a treatment group where only the first experimental treatment is present
- MAIN EFFECT 2: a second treatment group where the second experimental treatment is present
- INTERACTION EFFECT: third treatment group where both treatments are present
zimbardo and milgram studies are examples of…
post test only measurement
=> but is milgram really an experiment? there’s no randomized treatment
=> stanford prison is. a real experiment
when are experiments useful?
- communication, language, public opinion
(priming, framing, persuasion) - campaigning- classic domain of field experiments
- policy making- effectiveness of different policies (natural)
- decision making of citizens and political elites
- proces tracing
validity problems with experiments
- internal validity tends to be high becuase they make causal inferences possible
- ecological validity tensd to be low because the extent to whcih the research reflects real-life situations is debatable
- problems about reactivity- placebo effects may lead to participants mis-estimating the effects, and demand characteristics (giving the researcher what they want)
- external validity: weirc
experiments ethics
treatment usually requires concealment or deceptino- the use has to be well justified, and debriefing is required
in field studies, debriefing is often not possible