Deck 9 Flashcards
case control studies
observational cross-sectional design
comparison of affected cases to controls. looks at cases now and then asks them about their past
e.g. cancer patients and other cancer hospitalized patients, matched by age, sex, and hospital, asked units about smoking
limitations of case control studies
- retrospective: recall problems about past exposure (measurement validity and reliability)
- reverse causation: cases may change crucial behaviours or memories as a response to the disease
- the chance of cancer when smoking is not the same as the chance of having smoked while cancer
- comparability: units have to recall past behaviour good to be in the correct group
- selection bias: are your cases/controls representative for the whole population?
purpose of experimental research design
investigate a causal relationship between two (or more) variables by removing the influence of other variables, so the effect of the intervention can clearly be seen
with experimental design you…
- manipulate/intervene the independent variable
- allocate your units to treatments
- have control over conditions
conditions for an experiment
- researcher is able to manipulate the independent variable (x)
- ethical challenges are resolvable
the manipulation should:
- only wiggle x
- wiggle the whole part of x
- not run into plausible non-compliance
- be specified in detail
what/why a pre-test
measuring y before treatment has occurred, to check if there hasn’t been a change in y over time
what/why a control group
to filter out the influence of external events (extraneous variables). you create comparable groups to filter out the influences of difference
pre-test/conditioning effect
existing situation is changed by taking a pre-test (ask people about their attitudes and they become aware)
or there is a learning effect (bc of the information in the pre-test)
randomisation, matching, and homogenisation
to create comparable groups
randomisation: research units are assigned to groups by means of a lottery system. not always representative and may fail with a small number of units
matching: pair of sets are formed with specific, relevant similarities (you know what units you’re measuring so sets are similar on those specific units). then pairs are split between groups
homogenisation/restriction: restrict sample to a certain value or a limited range of values on important variables (bad for external validity)
interaction effect
when the effects of A and B together is bigger than the sum. adding sugar to coffee or stirring doesn’t make it sweeter, but combined they do
internal validity in experiments
- experimenter effect: researcher treats experimental and control group in a different way
- placebo/nocebo: belief of the participant that something does(not) work has an effect on measurement
- contagion/contamination/interference/diffusion/spillover: if experimental and control group have contact, they can exchange info on the experiment and influence result
solve by (double)blind study - differential compliance–> selective refusal –> confounding
- differential dropout –> selective refusal –>confounding
external validity in experiments
can my conclusions be generalised to:
- the target pop/other populations?
- referent/target context? (ecological validity)
ecological validity
the lab is not the real world
solutions:
- make the lab your world, simulate the real world environment in a lab (unrealistic, real world is very complex and you cannot model this in the lab)
- make the world your lab, select conditions in the real world (leads to better ecological validity but less control)
- replication in adverse circumstances, select important differences between the world and your lab. do your experiments in nature, where those differences are present and see if the results differ
experimentish design/quasi experimental design
a good balance between working in natural conditions and still applying some experimental design features that strengthen causal inference
