Bias Detection CRAGE Practice Questions (Page 2)

Questions 6–10

1. 6application · medium

   A fairness audit of a model shows that demographic parity is satisfied (selection rates are equal across groups), but equalized odds is not satisfied (false positive rates differ). What does this indicate about the model?

   • AThe model is overfitting because demographic parity and equalized odds cannot both be satisfied.
   • BThe model has no bias because equalized odds is not required.
   • CThe model treats groups equally in terms of positive outcomes but not in terms of error rates.
   • DThe model is completely fair because demographic parity is satisfied.
   Reveal Solution Discussion Select an answer first
2. 7expert · hard

   A model predicts whether a patient will develop a chronic condition. The team finds that the model has a lower false-positive rate for one demographic group, meaning it misses more cases for that group. The team must decide how to address this. Which consideration is most important?

   • AThe demographic group is small, so the disparity is not significant
   • BThe cost of false negatives is high because missed diagnoses can lead to serious health outcomes
   • CThe model's overall accuracy is high, so the disparity is acceptable
   • DThe model should be deployed as-is because it is already approved
   Reveal Solution Discussion Select an answer first
4. 8expert · hard

   A hiring model is evaluated for fairness. The demographic parity gap is small, but the equalized odds gap is large. Specifically, the model has a much higher false-positive rate for one demographic group. Which conclusion is most appropriate?

   • AThe model treats qualified candidates similarly across groups, but may over-recommend unqualified candidates from one group
   • BThe model has no bias because the overall accuracy is high
   • CThe model is biased against the group with the lower false-positive rate
   • DThe model is fair because it selects the same proportion from each group
   Reveal Solution Discussion Select an answer first
5. 9application · medium

   A company is developing a model to screen resumes. The team is concerned about bias against certain universities. Which qualitative method would be most effective for detecting this bias?

   • AIncreasing the size of the training dataset
   • BComputing the overall accuracy of the model
   • CUsing a more complex model to improve performance
   • DReviewing the model's feature importance and analyzing predictions for resumes from different universities
   Reveal Solution Discussion Select an answer first
6. 10expert · hard

   A model is being developed to predict creditworthiness. The data includes a variable 'employment status' that is self-reported. The team discovers that self-reported employment status is often inaccurate for gig workers. The model performs well for salaried employees but poorly for gig workers. Which source of bias is most likely, and what is the best detection method?

   • AConfirmation bias because the team expected gig workers to be risky; detect it by interviewing the team.
   • BAlgorithmic bias because the model uses a complex algorithm; detect it by simplifying the model.
   • CMeasurement bias in the employment status variable; detect it by comparing self-reported status with payroll records for a sample of gig workers.
   • DSelection bias because gig workers are underrepresented in the data; detect it by comparing the proportion of gig workers in the data to the population.
   Reveal Solution Discussion Select an answer first