Feature fairness

Hi there!
Thanks for taking the time to give us your valuable feedback!
We are a team of researchers from the Social Computing Research Group at the Max Planck Institute for Software Systems and the Machine Learning Group at the University of Cambridge.

In this survey, you will be shown 3 sets of questions about 9 features that can be used to predict the risk of recidivism of offenders. Features are measurable properties of a phenomenon that is being observed. For offenders, those features can be their race, gender and data about their previous offenses.

Based on the values of these features, the offenders are subsequently classified as having a high risk, medium risk or low risk of recidivism.

For each of the 9 features, you will have to answer whether you believe it is fair or unfair to assess offenders' risk of recidivism based on the values of these features, in three different scenarios.
You need to give your judgment for all 3 scenarios, for all 9 features in order to complete the task.





Do you believe it is fair or unfair to use these features in the process of estimating offenders' risk of recidivism?

Do you believe it is fair or unfair to use information about the offender's age when estimating his/her risk of recidivism?

Do you believe it is fair or unfair to use information about the offender's gender when estimating his/her risk of recidivism?

Do you believe it is fair or unfair to use information about the offender's race when estimating his/her risk of recidivism?

Do you believe it is fair or unfair to use information about the degree of the offender's current arrest charge (i.e., misdemeanor or felony) when estimating his/her risk of recidivism?

Do you believe it is fair or unfair to use the precise description of the offender's current arrest charge (e.g., grand theft) when estimating his/her risk of recidivism?

Do you believe it is fair or unfair to use the number of the offender's prior offenses when estimating his/her risk of recidivism?

Do you believe it is fair or unfair to use the number of the offender's juvenile felony offenses when estimating his/her risk of recidivism?

Do you believe it is fair or unfair to use the number of the offender's juvenile misdemeanor offenses when estimating his/her risk of recidivism?

Do you believe it is fair or unfair to use the number of the offender's other juvenile offenses when estimating his/her risk of recidivism?




When these features are used in the process of estimating offenders' risk of recidivism, black people are more likely to be assessed as having a higher risk of recidivism than white people. Do you believe it is fair or unfair to use these features in the process of estimating offenders' risk of recidivism?

Do you believe it is fair or unfair to use information about the offender's age when estimating his/her risk of recidivism, if it makes black people more likely to be assessed as having a higher risk of recidivism than white people?

Do you believe it is fair or unfair to use information about the offender's gender when estimating his/her risk of recidivism, if it makes black people more likely to be assessed as having a higher risk of recidivism than white people?

Do you believe it is fair or unfair to use information about the offender's race when estimating his/her risk of recidivism, if it makes black people more likely to be assessed as having a higher risk of recidivism than white people?

Do you believe it is fair or unfair to use information about the degree of the offender's current arrest charge (i.e., misdemeanor or felony) when estimating his/her risk of recidivism, if it makes black people more likely to be assessed as having a higher risk of recidivism than white people?

Do you believe it is fair or unfair to use the precise description of the offender's current arrest charge (e.g., grand theft) when estimating his/her risk of recidivism, if it makes black people more likely to be assessed as having a higher risk of recidivism than white people?

Do you believe it is fair or unfair to use the number of the offender's prior offenses when estimating his/her risk of recidivism, if it makes black people more likely to be assessed as having a higher risk of recidivism than white people?

Do you believe it is fair or unfair to use the number of the offender's juvenile felony offenses when estimating his/her risk of recidivism, if it makes black people more likely to be assessed as having a higher risk of recidivism than white people?

Do you believe it is fair or unfair to use the number of the offender's juvenile misdemeanor offenses when estimating his/her risk of recidivism, if it makes black people more likely to be assessed as having a higher risk of recidivism than white people?

Do you believe it is fair or unfair to use the number of the offender's other juvenile offenses when estimating his/her risk of recidivism, if it makes black people more likely to be assessed as having a higher risk of recidivism than white people?




Using these features makes the risk estimation process more accurate. Do you believe it is fair or unfair to use these features in the process of estimating offenders' risk of recidivism?

Do you believe it is fair or unfair to use information about the offender's age when estimating his/her risk of recidivism, if it makes the estimation more accurate?

Do you believe it is fair or unfair to use information about the offender's gender when estimating his/her risk of recidivism, if it makes the estimation more accurate?

Do you believe it is fair or unfair to use information about the offender's race when estimating his/her risk of recidivism, if it makes the estimation more accurate?

Do you believe it is fair or unfair to use information about the degree of the offender's current arrest charge (i.e., misdemeanor or felony) when estimating his/her risk of recidivism, if it makes the estimation more accurate?

Do you believe it is fair or unfair to use the precise description of the offender's current arrest charge (e.g., grand theft) when estimating his/her risk of recidivism, if it makes the estimation more accurate?

Do you believe it is fair or unfair to use the number of the offender's prior offenses when estimating his/her risk of recidivism, if it makes the estimation more accurate?

Do you believe it is fair or unfair to use the number of the offender's juvenile felony offenses when estimating his/her risk of recidivism, if it makes the estimation more accurate?

Do you believe it is fair or unfair to use the number of the offender's juvenile misdemeanor offenses when estimating his/her risk of recidivism, if it makes the estimation more accurate?

Do you believe it is fair or unfair to use the number of the offender's other juvenile offenses when estimating his/her risk of recidivism, if it makes the estimation more accurate?

Imprint / Data Protection