Fairness Bias Research Articles

Integrating Structural Causal Model Ontologies with LIME for Fair Machine Learning Explanations in Educational Admissions

This study employed knowledge discovery in databases (KDD) to extract and discover knowledge from the Benue State Polytechnic (Benpoly) admission database and used a structural causal model (SCM) ontological framework to represent the admission process in the Nigerian polytechnic education system. The SCM ontology identified important causal relations in features needed to model the admission process and was validated using the conditional independence test (CIT) criteria. The SCM ontology was further employed to identify and constrain input features causing bias in the local interpretable model-agnostic explanations (LIME) framework applied to machine learning (ML) black-box predictions. The ablation process produced more stable LIME explanations devoid of fairness bias compared to LIME without ablation, with higher prediction accuracy (91% vs. 89%) and F1 scores (95% vs. 94%). The study also compared the performance of different ML models, including Gaussian Naïve Bayes, Decision Trees, and Logistic Regression, before and after ablation. The limitation is that the SCM ontology is qualitative and context-specific, so the fair-LIME framework can only be extrapolated to similar contexts. Future work could compare other explanation frameworks like Shapley on the same dataset. Overall, this study demonstrates a novel approach to enforcing fairness in ML explanations by integrating qualitative SCM ontologies with quantitative ML/LIME methods.

DEFINING AND DETECTING FAIRNESS BIAS FOR BINARY CLASSIFICATION PROBLEM IN FINANCIAL ANALYSIS

This article aims to present the fairness bias in the models of artificial intelligence. First, it introduces use cases and legislative constrains of automated decision making towards sensitive features. Then using academic datasets, the historical human bias, measures of dataset fairness, and the effective way of choosing the respective metric are presented. And last, different AI models are estimated to show the replication of decision bias from data to models. The design of the research is observational; academical datasets have been used. For the quantitative analysis both descriptive and inferential statistics are applied. The analysis was done for the problem of binary classification mainly focusing on the decision making in finance. The phenomenon of unequal decisions aimed at unprivileged demographic groups was shown and quantified, stating the example given with averaging 8-20% bias between groups, which was also present in even most accurate models – 85% and 90% AUC score.

Fairness-Aware Dynamic Rate Control and Flow Scheduling for Network Utility Maximization in Network Service Chain

Network function virtualization (NFV) decouples the traditional network functions from specific or proprietary hardware, such that virtualized network functions (VNFs) can run in software form. By exploring NFV, a consecutive set of VNFs can constitute a service function chain (SFC) to provide the network service. From the perspective of network service providers, how to maximize the network utility is always one of the major concerns. To this end, there are two main issues need to be considered at runtime: 1) how to handle the unpredictable network traffic burst? and 2) how to fairly allocate resources among various flows to satisfy different traffic demands? In this paper, we investigate a fairness-aware flow scheduling problem for network utility maximization, with joint consideration of resource allocation and rate control. Based on a discrete-time queuing model, we propose a low-complexity online-distributed algorithm using the Lyapunov optimization framework, which can achieve arbitrary optimal utility with different fairness levels by tuning the fairness bias parameter. We theoretically analyze the optimality of the algorithm and evaluate its efficiency by both simulation and testbed-based experiments.

Value-based pricing and cognitive biases: An overview for business markets

To investigate how cognitive biases inhibit value-based pricing practices among managers in business markets, this article considers five different cognitive biases—perceived lack of control, herding, fixed-pie bias, ambiguity aversion, and egocentric fairness bias—and their effects on value-based pricing. Despite recent calls for more research on the psychological aspects of pricing, few studies have focused on business markets. Drawing on research in psychology and marketing for its theoretical foundation, this overview extends the limited body of existing research. The article's key contribution is to explain how psychological challenges affect value-based pricing practices, with implications and suggestions for further research.

Hello handsome! Male's facial attractiveness gives rise to female's fairness bias in Ultimatum Game scenarios-An ERP study.

The current study delineated how male proposers’ facial attractiveness affect female responders’ fairness considerations and their subsequent decision outcome during the Ultimatum Game (UG). Event Related Potentials (ERPs) were recorded from 17 female subjects, who played the role as responders and had to decide whether to accept offers from either attractive or unattractive male proposers. Behavioral data (Acceptance Ratio and Response time) revealed that, more offers were accepted from attractive-face conditions; subjects typically responded quicker to unfair offers from unattractive proposers as compared with slower to unfair offers from attractive proposers. The ERP data demonstrated similar N2 amplitudes elicited by both attractive and unattractive faces, and a larger early frontal LPP elicited by the attractive faces compared with unattractive ones, but no significant differences of both late posterior LPP and typical parietal LPP amplitudes were observed between these two face conditions, which was different from our previous study with similar paradigm but male participants. The results suggest that, in comparison to males, females might not experience the potential attention bias towards unattractive opposite-sex faces and are less likely to possess an enhanced processing and evaluation of those faces. This phenomenon might be explained by endogenous gender differences in mate preference. The feedback-related negativity (FRN) and P300 responses during an offer presentation were further measured in both attractive-face and unattractive-face conditions and the results demonstrated that the amplitudes elicited by fair and unfair offers were not statistically different in the former condition, but were different in the latter condition. More specifically, unfair offers generated larger FRN and smaller P300 than fair ones in the unattractive-face condition. Findings suggest that, although females tend to possess less salient evaluation of male’s facial attractiveness, the attractiveness of male proposers would still attenuate female responders’ fairness consideration during the UG.

Judicial error and cooperation

Cooperation can be induced by an authority with the power to mete out sanctions for free riders, but law enforcement is prone to error. This paper experimentally analyzes preferences for and consequences of errors in formal sanctions against free riders in a public goods game. With type I errors, even full contributors to the public good may be punished. With type II errors, free riders may go unpunished. We find that judicial error undermines cooperation and that the effects of type I and II errors are symmetric. To investigate their relative (dis-)like for error, we let subjects choose what type of error to prevent. We find that subjects prefer type II over type I errors. However, the strength of preferences for preventing type I errors is fully in line with a motive to maximize income and does not indicate any additional psychological or fairness bias against type I errors.

Judicial Error and Cooperation

Cooperation can be induced by an authority with the power to mete out sanctions for free riders, but law enforcement is prone to error. This paper experimentally analyzes preferences for and consequences of errors in formal sanctions against free riders in a public goods game. With type I errors, even full contributors to the public good may be punished. With type II errors, free riders may go unpunished. We find that judicial error undermines cooperation and that the effects of type I and II errors are symmetric. To investigate their relative (dis-)like for error, we let subjects choose what type of error to prevent. By use of an incentive-compatible mechanism, we find that subjects prefer type II over type I errors. We find that the strength of this preference is fully in line with a motive to maximize income and does not indicate any additional psychological or fairness bias against type I errors.

Quantification of fairness bias in relation to decisions using a relativistic fairness-equity model

This study quantifies the fairness bias in relation to decision by a stimulus-response function using a relativistic fairness-equity model. The interrelationship between fairness and decision is quantified by using an Ultimatum Game (UG) experimental paradigm, in which an amount of money is shared between two parties, while the human subjects are asked to accept or reject the share. The results showed that the fairness perception is shifted upward (toward a higher positive fairness baseline in the y -intercept of the stimulus-response function) for acceptance trials, without changing the slope (which corresponds to the fairness sensitivity). On the other hand, the fairness perception is shifted downward (toward a negative fairness baseline in the y -intercept) for the rejection trials. The analysis also showed that the fairness crossover point is shifted to the left for the acceptance trials, while the fairness crossover point is shifted to the right for the rejection trials. This suggests that human subjects were more lenient to fairness when they considered slightly inequitable offers as fair in their decision to accept the offers (quantified by the fairness crossover point being shifted to the left for the acceptance trials). On the other hand, it suggests that they were greedy when they considered hyper-equitable offers as unfair (quantified by the fairness crossover point being shifted to the right for the rejection trials). The analysis also showed that there is a singularity point, in which the most equitable offer (even-split) is always considered as the fairest, even when they rejected the offers. This absolute equity is rated as the fairest (even fairer than any of the hyper-equitable offers) independent of whether the subjects decided to accept or reject the offers. These results suggested that when human subjects decided to accept or reject the offer, they included both self-regarding and other-regarding concerns, by using both self-centered and other-centered frames of reference in assessing fairness. The inclusiveness of both parties in the fairness consideration provides an optimal solution to maximize the gains for both parties at the most equitable offer (even-split) without creating conflict-of-interest. The changes in fairness perception are quantified by the shifting of the stimulus-response curve up/down (changing the fairness baseline) or left/right (changing the fairness leniency), without changing the slope (the fairness sensitivity), when the decision is made to accept or reject the offers.

Accountability and the fairness bias: the effects of effort vs. luck

According to the principle of accountability a person’s fair allocation takes into account the input-relevant variables she can influence, like effort, but not the variables she cannot influence, like luck. We study redistribution behavior in a real effort-task experiment, where luck influences production in three versions that differ in their effort-relatedness: luck comes either as wage luck, endowment luck or payment luck. We confirm that in a base treatment, i.e., in absence of luck, subjects base their allocation decisions largely on effort. When luck is present behavior changes. While we find that redistribution behavior of high performers in the real effort-task is affected by wage luck, we find that endowment luck affects mostly redistribution behavior of low performers. Furthermore, when introducing payment luck, i.e., redistribution of certain versus probabilistic payments, we find that allocations are more selfish for probabilistic payments.

Quantification of fairness bias by a Fairness-Equity Model

Article accompanying a poster presentation for the 2011 Computational Neuroscience Meeting. This article discusses the quantification of fairness bias by a fairness-equity model.

Fair Air: Distributive Justice and Environmental Economics

Are fairness concerns of relevance to environmental economics and, if so, are they sufficiently structured to improve analysis in this field? On both of these questions, we answer in the affirmative, arguing that people's fairness views are based on both general rules and the context, where context refers to the set of variables and persons employed to interpretandapplytheprinciples.Thefairnessrulesanalyzedareaccountability(i.e.,rewards that are proportional to contributions individuals control), efficiency, need and equality. We conclude that stakeholders typically exhibit a "fairness bias", i.e., they tend, consciously or not, to interpret and apply fairness principles in a self-serving manner, whereas the views of spectators, or impartial third parties, tend to converge significantly more. Further, we argue that fairness considerations are relevant to both descriptive and prescriptive analysis in environmental economics. These fairness concerns are reflected in the behavior of private and public decision-makers and have potentially important policy implications through the overall social objective function.

Commentaries and Rejoinder to “The Dishonesty of Honest People”

In their clever and insightful article, Mazar, Amir, and Ariely (2008) propose that people balance two competing desires when deciding whether to behave dishonestly: the motivation for personal gain and the desire to maintain a positive self-concept. Their studies focus on the latter fac tor, showing that people behave dishonestly when it pays, but only to the extent that they can do so without violating their perception of themselves as honest. The research is innovative and important. It has already had an influence on policies dealing with conflicts of interest in medicine (Association of American Medical Colleges 2007) and, even before its own publication, has spawned significant follow-up research (Vohs and Schooler 2008). In our opinion, the main limitation of Mazar, Amir, and Ariely's article is not in the perspective it presents but rather in what it leaves out. Although it is important to understand the psychology of rationalization, the other fac tor that Mazar, Amir, and Ariely recognize but then largely ignore?namely, the motivation to behave dishonestly?is arguably the more important side of the dishonesty equation. The motivation side is especially important, in part because the propensity to rationalize is itself a function of the motivation to do so. Given sufficient motivation, people can persuade themselves of almost anything, including why behavior they normally would consider unethical is morally acceptable. Research on the self-serving fairness bias (for a summary, see Babcock and Loewenstein 1997) shows that people tend to conflate what is fair with what is in their per sonal interest, and the same is no doubt true of people's judgments of what is ethical. Given a sufficiently powerful motivation to commit an act of fraud, in general, people are more than capable of rationalizing why it does not conflict with their own ethical precepts. Furthermore, after people have taken the first step toward unethical behavior, a large body of research shows that subsequent steps into the abyss of immorality become progressively easier (e.g., Lifton 1986; Milgram 1963). HYPERMOTIVATION

Blind Spots: The Effects of Information and Stakes on Fairness Bias and Dispersion

Mounting empirical research provides evidence of fairness bias and its economic and social effects, where fairness bias refers here to a deviation of claims from unbiased justice due to a personal stake. A far less appreciated issue is dispersion of fairness views and claims, which is also important for its effects on disagreements, empirical analysis, and philosophical theories. This study undertakes a systematic analysis of the effects on fairness bias and dispersion of two variables: stakes and information. Most philosophical and social science analyses related to justice and bias associate heightened bias with increased information and, conversely, impartiality with the elimination of certain information. Less attention has been paid to the opposing impact of information, which is to supply the facts needed to achieve justice more reliably. An important open question is whether, on balance, increased information helps agents to achieve fairer outcomes or whether biased use of such information contributes to less fair outcomes. This study focuses on a set of previously reported experiments that share certain features and subjects them to a new analysis. The results of this analysis suggest that, although information is often used in a self-serving way, increased information can, under certain conditions, contribute to fairness claims becoming less biased and less dispersed, both for stakeholders as well as impartial spectators.

SABUL: A Transport Protocol for Grid Computing

This paper describes SABUL, an application-level data transfer protocol for data-intensive applications over high bandwidth-delay product networks. SABUL is designed for reliability, high performance, fairness and stability. It uses UDP to transfer data and TCP to return control messages. A rate-based congestion control that tunes the inter-packet transmission time helps achieve both efficiency and fairness. In order to remove the fairness bias between flows with different network delays, SABUL adjusts its sending rate at uniform intervals, instead of at intervals determined by round trip time. This protocol has demonstrated its efficiency and fairness in both experimental and practical applications. SABUL has been implemented as an open source C++ library, which has been successfully used in several Grid computing applications.

Judgments of Fairness by Just World Believers

The author investigated the effects of just world belief on the egocentric fairness bias among Japanese undergraduates. After responding to the Just World (JW) Scale (Rubin & Peplau, 1975) and to a set of fair and unfair behaviors, the participants were divided into 2 groups (low JW and high JW) according to their responses to the scale. The egocentric fairness bias was confirmed; the participants tended to consider their own behaviors as fair rather than unfair and other people's behaviors as unfair rather than fair. High-JW participants regarded their own behaviors as fairer and those of other people as less fair than did low-JW participants. The egocentric fairness bias was stronger in the participants who believed strongly in a just world than in those whose belief in a just world was weaker.

Explaining Bargaining Impasse: The Role of Self-Serving Biases

The authors review studies conducted by themselves and coauthors that document a 'self-serving' bias in judgments of fairness and demonstrate that the bias is an important cause of impasse in negotiations. They discuss experimental evidence showing that (1) the bias causes impasse; (2) it is possible to reduce impasses by debiasing bargainers; and (3) the bias results from selective evaluation of information. The authors also review results from a field study of negotiations between teachers' unions and school boards in Pennsylvania that both document the fairness bias in a naturalistic setting and demonstrates its impact on strikes.

Cognitive biases in negotiation : The determinants of fixed-pie assumption and fairness bias

Cognitive biases in negotiation : The determinants of fixed-pie assumption and fairness bias

Frequentistic adverbs as measures of egocentric biases

Research by Messick et al. (1985) concerning the egocentric fairness bias was replicated using linguistic choices rather than rating scales. Subjects were asked to choose from among six frequency quantifiers to describe the estimated frequency of moral and immoral behaviours as performed by ‘I’ or ‘Other people’. Results replicate the dual slope phenomenon found by previous researchers. Subjects chose quantifiers referring to higher frequencies when estimating frequencies for ‘I’–moral behaviour than for ‘Other people’—moral behaviour, and quantifiers referring to lower frequencies when estimating frequencies for ‘I’—immoral behaviour than for ‘Other people’—immoral behaviour. This ‘I’–‘Other’ differentiation was greater for the immoral behaviours than the moral behaviours.

The Refugee Act of 1980: Problems and Congressional Concerns

The Refugee Act of 1980 was the most comprehensive U.S. law ever enacted concerning refugee admissions and resettlement. But each one of its principles has raised problems and encouraged new debate with respect to its implementation or its continuance. Levels of refugee flow are clearly a concern in Congress, linked in the eyes of many with broader issues of border control, fiscal policy, and the appropriate role of the legislative branch. Further attempts to limit that flow through some sort of numerical cap can be expected, as well as further debate on the reach of the refugee definition. Closely associated with the issue of flow and the validity of particular claims is the asylum issue. The right of asylum, having been incorporated into domestic law, will not be abolished. However, concern over numbers, over the inefficiency of present procedures, and over their hotly debated fairness or political bias is certain to generate renewed effort to reform asylum adjudication procedures. Finally, the method and financing of resettlement assistance to refugees is likely to remain the subject of debate. Divorcing refugee assistance from refugee admissions will continue to be discussed. Congressional battles will also be waged over appropriate levels of funding for refugee programs and quite possibly over the design of those programs, especially as it affects their coordination with state and local governments, the role of the private voluntary agencies, and the mutual assistance agencies.