Confounding and randomisation
1.[3p] What makes a variable a confounder?
What makes a variable a confounder?
The answer is: It affects the outcome, is associated with the exposure, and is not on the causal path between them
The answer is: It affects the outcome, is associated with the exposure, and is not on the causal path between them
The answer is: It affects the outcome, is associated with the exposure, and is not on the causal path between them
2.[2p] Berkeley's six largest departments admitted 44.5 per cent of men and how many per cent of women, to one decimal place?
Berkeley's six largest departments admitted 44.5 per cent of men and how many per cent of women, to one decimal place?
3.[3p] What produced the Berkeley reversal?
What produced the Berkeley reversal?
The answer is: Women applied disproportionately to departments with low admission rates
The answer is: Women applied disproportionately to departments with low admission rates
The answer is: Women applied disproportionately to departments with low admission rates
4.[3p] Randomisation guarantees that the treatment and control groups are balanced on every covariate in any particular trial.
Randomisation guarantees that the treatment and control groups are balanced on every covariate in any particular trial.
The answer is: False
5.[2p] A trial randomises 50 per arm and ages have standard deviation 12 years. What is the standard error of the difference in mean age, in years?
A trial randomises 50 per arm and ages have standard deviation 12 years. What is the standard error of the difference in mean age, in years?
6.[3p] If 20 baseline covariates are each tested at the 5 per cent level, what is the probability that at least one comes out significant, to two decimal places?
If 20 baseline covariates are each tested at the 5 per cent level, what is the probability that at least one comes out significant, to two decimal places?
7.[3p] Why did observational studies of hormone replacement therapy disagree with the randomised trial?
Why did observational studies of hormone replacement therapy disagree with the randomised trial?
Select all that apply
The answer is: Women who chose HRT differed from those who did not in ways nobody had measured, Adjustment can only remove confounders that were measured and correctly identified, A clean residual plot does not rule out a large omitted variable
8.[3p] Match each design device to what it achieves.
Match each design device to what it achieves.
Randomisation
Blinding
Blocking
Stratified randomisation
removes block-to-block variation
hides the assignment from both sides
removes confounding, measured or not
forces balance on one variable
Show the answer
Randomisation: removes confounding, measured or not Blinding: hides the assignment from both sides Blocking: removes block-to-block variation Stratified randomisation: forces balance on one variable
9.[2p] Why is adjusting for a variable on the causal path a mistake?
Why is adjusting for a variable on the causal path a mistake?
The answer is: It removes the mechanism through which the exposure acts, erasing part of the real effect
The answer is: It removes the mechanism through which the exposure acts, erasing part of the real effect
The answer is: It removes the mechanism through which the exposure acts, erasing part of the real effect