Background

You will be shown a total of ten scenarios. Each scenario displays the effect of one parameter (for example 'age') on the algorithm's risk calculation (for example 'risk of crashing a car'). In each scenario you will be able to explore this risk calculation for a different set of groups (for example between males and females).

Based on historical data of many people, algorithms determine which parameters (age, number of previous accidents, etc.) are useful in distinguishing between low- and high-risk individuals.

In this task, you are asked to assess whether or not a parameter (for example 'age') should be included in an algorithm predicting a specified risk, as well as whether you believe it is fair to include this parameter. You will be asked to motivate your choices. Finally, we ask you to state whether you think the parameter would negatively effect on one (or more) of the shown groups as compared to the other group(s). For example - an algorithm that determines young drivers are more at risk may have a more negative effect on male drivers than on female drivers.

Question 1.
Algorithms calculate the risk of an individual by;


The algorithms presented in this study are based on two real datasets: one contains information on (the payback of) loans, the other contains information on reoffending criminal behaviour.

Customers applying for a loan will either pay back the loan or fail to pay back their loan. Predicting whether someone will pay back their loan can be used to reject or approve a person's loan application or help determine the loan's interest rate.

Individuals which are caught for a crime will either become reoffenders (this is known as recidivism) or non-reoffenders. Predicting whether someone becomes a reoffender can be used to adjust their sentence or the support receive during or after their sentence.

Question 2.
A recidivist is someone who;



Open first scenario