**fairness**1602

American Economic Association

3 days ago by MarcK

"We analyze how separations responded to arbitrary differences in own and peer wages at a large US retailer. Regression-discontinuity estimates imply large causal effects of own-wages on separations, and on quits in particular. However, this own-wage response could reflect comparisons either to market wages or to peer wages. Estimates using peer-wage discontinuities show large peer-wage effects and imply the own-wage separation response mostly reflects peer comparisons. The peer effect is driven by comparisons with higher-paid peers—suggesting concerns about fairness. Separations appear fairly insensitive when raises are similar across peers—suggesting search frictions and monopsony are relevant in this low-wage sector. "

fairness
wages
AER
3 days ago by MarcK

Putting the J(ustice) in FAT – Berkman Klein Center Collection – Medium

17 days ago by tarakc02

> ... because much of what we intuitively conceive of as “unfair” is actually “unjust,” FAT researchers need to more deeply consider social justice as an essential component of their work. Put another way, many of the concerns about biased algorithms are actually concerns about the impacts of predictions and the systems that those algorithms enhance. A predictive policing algorithm might perfectly forecast where crime will occur (hence satisfying notions of fairness), but if those predictions are used by police departments to harass and oppress communities of color, then the algorithm is unjust.

FATML
fairness
justice
fairml
17 days ago by tarakc02

Risk Assessment: Explained - The Appeal

23 days ago by tarakc02

overview of pretrial risk assessments -- debates, criticisms, controversies, etc.

pretrial-detention
risk-assessment
algorithmic-bias
fairness
23 days ago by tarakc02

[1705.10239] Fair Division of a Graph

23 days ago by Vaguery

We consider fair allocation of indivisible items under an additional constraint: there is an undirected graph describing the relationship between the items, and each agent's share must form a connected subgraph of this graph. This framework captures, e.g., fair allocation of land plots, where the graph describes the accessibility relation among the plots. We focus on agents that have additive utilities for the items, and consider several common fair division solution concepts, such as proportionality, envy-freeness and maximin share guarantee. While finding good allocations according to these solution concepts is computationally hard in general, we design efficient algorithms for special cases where the underlying graph has simple structure, and/or the number of agents -or, less restrictively, the number of agent types- is small. In particular, despite non-existence results in the general case, we prove that for acyclic graphs a maximin share allocation always exists and can be found efficiently.

assignment-problems
fairness
operations-research
planning
collective-behavior
cake-cutting
game-theory
constraint-satisfaction
rather-interesting
variant-problems
mathematical-recreations
to-write-about
to-simulate
23 days ago by Vaguery

Camera Above the Classroom

24 days ago by csantos

"Since discovering that photo on Weibo, Jason’s been observing the cameras installed in the classroom, trying to guess their functions. The smaller one at the front is used for facial recognition, he deduces from the picture on Weibo. The bigger one at the back of the classroom is used for livestreaming, judging from the angle of the video he previously spotted on his teacher’s laptop. He’s not sure about the third camera on the window side of the classroom, though he thinks it’s a backup."

surveillance
classroom
doom
China
ComputerVision
ethics
fairness
MachineLearning
24 days ago by csantos

Twitter

24 days ago by aarontay

@aarontay @karenhytteballe Yes, absolutely. #FAIRness of #researchdata is independent of their #openness. Your…

researchdata
openness
FAIRness
from twitter_favs
24 days ago by aarontay

Equality of opportunity in supervised learning | the morning paper

4 weeks ago by tarakc02

> Equalized odds requires both the fraction of non-defaulters that qualify for loans and the fraction of defaulters that qualify for loans to be constant across groups. This cannot be achieved with a single threshold for each group, but requires randomization. There are many ways to do it; here we pick two thresholds for each group, so above both thresholds people always qualify, and between the thresholds people qualify with some probability.

> Equalized odds enforces that the accuracy is equally high in all demographics, punishing models that perform well only on the majority.

> In some situations… the equalized odds predictor can be thought of as introducing some sort of affirmative action: the optimally predictive score R* is shifted based on A. This shift compensates for the fact that, due to uncertainty, the score is in a sense more biased than the target label (roughly, R* is more correlated with A than Y is correlated with A). Informally speaking, our approach transfers the burden of uncertainty from the protected class to the decision maker.

fairness
affirmative-action
algorithmic-bias
bias
FATML
> Equalized odds enforces that the accuracy is equally high in all demographics, punishing models that perform well only on the majority.

> In some situations… the equalized odds predictor can be thought of as introducing some sort of affirmative action: the optimally predictive score R* is shifted based on A. This shift compensates for the fact that, due to uncertainty, the score is in a sense more biased than the target label (roughly, R* is more correlated with A than Y is correlated with A). Informally speaking, our approach transfers the burden of uncertainty from the protected class to the decision maker.

4 weeks ago by tarakc02

Delayed impact of fair machine learning | the morning paper

4 weeks ago by tarakc02

In the larger universe of papers looking at threshold-based metrics of fairness, this one starts with the idea that the goal of fairness is to "promote the long-term well-being" of protected groups. They then introduce a feedback loop between decisions that get made with the help of algorithms and the probability scores output by those algorithms (in the given example, if someone is approved for a loan and then pays it off, that increases their credit score, while if they default, that will decrease their credit score). Then, defining well-being in terms of group change in mean credit score, the authors go on to show that various fairness metrics may in fact make protected groups worse off under specific conditions, and more generally show how to use the feedback model framework to evaluate different fairness criteria. The tool they introduce to do this is the "Outcome Curve," a concave curve whose x-axis is the "selection rate" given by various decision-making thresholds, and the y-axis is the expected change in mean (credit) score. Important points on this curve include the point that maximizes the firm's profit, the point that maximizes the group's well-being, regions where groups are made worse off than if the firm had just maximized profit, and regions where the group is made worse off in an absolute sense. There is a brief section describing "measurement error," in particular where scores systematically underestimate a group's repayment probabilities, and I am only mentioning this because that section, and the accompanying analysis using the outcome curve framework, so echoes the debates about affirmative action that were going on when I was growing up in the '90s. Finally, there is an empirical section using the framework along with historical credit data. The results are all conditional on the assumed costs and benefits to both parties of a repaid loan and a defaulted loan, which I guess points to a challenging part of using this framework. The paper did not answer all of my questions, but the focus on group-level outcomes over time, and the focus on the feedback loop of decision making systems, were helpful.

fairness
FATML
algorithmic-bias
fairml
affirmative-action
4 weeks ago by tarakc02

**related tags**

Copy this bookmark: