compas-equality-of-opportunity

COMPAS and Equality of Opportunity

Recently, we replicated ProPublica’s investigative reporting on the COMPAS algorithm and studied several competing statistical definitions of fairness.

Imagine that this article came out two weeks ago. What should we do about the situation reported by ProPublica? Please address this question from three distinct perspectives:

The perspective of someone who adheres to the narrow view of equality of opportunity.
The perspective of someone who adheres to the middle view of equality of opportunity.
The perspective of someone who adheres to the broad view of equality of opportunity.

Write at least three sentences from each of these perspectives. Your response should, at minimum, address the following two questions:

You may find it useful to remember from the readings that calibration, as discussed in BHN Ch. 4 is related to sufficiency, which is discussed in BHN Ch. 3 and also in our notes on statistical conceptions of fairness.

Is the [narrow/middle/broad] view of equality of opportunity in fact violated, according to our previous data analysis?
If so, what change is obligated?
Whose obligation is it to enact that change? Whose obligation is it to incur the costs of that change?

Please ground your sentences in our quantitative findings and our recent readings on statistical and normative notions of fairness. If you think the answer isn’t cut-and-dried, please do your best analysis and make a note about what’s complicated and why.