causal evaluation 2 (8) Flashcards
what is stratification
conducting same analysis in different strata, subgroup analysis
selection bias
selected participants don’t represent population of interest
not representative
problems with selection bias
data unlikely to be applicable to greater population
internal validity
estimated associations true in our study sample
external validity
estimated associations applicable to target population
colliders
common effects of exposure and outcome
eg frailty and severe covid both contribute to hospital admission
potential collider bias if recruiting hospital patients to study frailty relationship with covid
not adjusting for collider
makes it bias towards the null
ruling out selection bias
using gold standard random sample with high response rate (volunteer bias)
better to do before data collection
information bias
systematic error in measurement of exposure or outcome
non differential miss-classification
error in outcome/exposure lowers precision and bias towards null
when n is large outcome measurement error levels out
differential miss-classification
measurement accuracy doesn’t level out as don’t know bias direction
ruling out information bias
multiple sources
objective measurement
blinded 3rd party assessment
different recruitment method
primary data
collected specifically for addressing public health question
eg births deaths area based deprivation
secondary data
existing data collected for another purpose
eg health records cancer register prescriptions
problems with primary data
may need ethical approval
costly and slow
sometimes needs specialist collection
