Thank
you for catching this error in the answer key. Yes, in our example of
selection bias the E-D+ group is underrepresented which makes the risk
in the nonAD group underestimated.
Since the risk ratio = Risk in AC/Risk in nonAC, this bias will over-estimate
the numeric risk ratio. Under the null, this bias would make AC look
harmful (RR>1). If the AC were truly protective for PCSM (non-null),
this type of bias is towards the
null, making AC group less effective on preventing PCSM.
Grading for question 10 was not based on the directionality of the bias but on your ability to explain a scenario where sampling was differently in exposed/unexposed and high/low risk for the outcome. We made corrections in the answer key and we are happy to discuss any further question.