Pre-existing Biases Research Articles

Gendered language and employment Web sites: How search algorithms can cause allocative harm

Research on algorithms and artificial intelligence in the hiring process tends to focus on applicant screening and is often centered on the employer perspective. The role played by intermediaries, such as employment Web sites in the distribution of information about employment opportunities, tends to be overlooked. This paper examines the role of search algorithms on employment Web sites and their retrieval of employment opportunities for job seekers based on gendered search terms. Through a basic algorithm audit of the search engines used by three major employment Web sites active in the Dutch job market, we explore whether their search algorithms neutralize or reinforce gendered language, in case of the latter thereby naturalizing stigmas and pre-existing bias. According to our findings, employment Web sites can cause allocative harm if they repeatedly fail to show all opportunities relevant to job seekers.

Missing Data in OLAP Cubes

Online analytical processing (OLAP) engines display aggregated data to help business analysts compare data, observe trends, and make decisions. Issues of data quality and, in particular, issues with missing data impact the quality of the information. Key decision-makers who rely on these data typically make decisions based on what they assume to be all the available data. The authors investigate three approaches to dealing with missing data: 1) ignore missing data, 2) show missing data explicitly (e.g., as unknown data values), and 3) design mitigation algorithms for missing data (e.g., allocate missing data into known value categories). The authors evaluate the approach with focus groups and controlled experiments. When one tries to inform decision-makers using the approaches in the research, the authors find that they often alter their decisions and adjust their decision confidence: individual differences of tolerance for ambiguity and pre-existing omission bias in the decision context influence their decisions.

Biased representations of controversial information: Certainty and justification beliefs as moderators

Pre-existing beliefs bias mental representations of socio-scientific controversies in favour of the perspective(s) that endorse(s) them. Several conditions have been proposed to moderate such belief-biased mental representations of controversial information. The present study examined the effects of readers’ prior beliefs on their mental models of textual information on the level of situation-model and text-base representations. The study further investigated the extent that author status, readers’ certainty and justification beliefs moderate the effect of prior beliefs on the readers’ representations. Sixty-two undergraduate students of English as a foreign language read two texts that provided arguments for and against an established controversy in language education. A recognition task was used to assess their situation-model and text-base strengths. The results revealed that readers’ representations were biased towards the information that supported their beliefs at the level of the situation model but not at the level of the text-base. The results further revealed no main or moderating effects for author status on the interaction of beliefs and readers’ mental models of the information. However, readers’ certainty beliefs and beliefs about justification for knowing were shown to moderate the belief-biased representation of the controversial information.

Distributive Justice as a Means of Combating Systemic Racism in Healthcare

Distributive Justice as a Means of Combating Systemic Racism in Healthcare

Assessing attitudes to ED-based HIV testing: Development of a short-structured survey instrument.

IntroductionEmergency Department (ED)-based HIV counseling and testing (HCT) has had a significant impact on improving rates of HIV diagnosis and linkage to care. Unfortunately, expansion of this strategy to low- and middle-income countries has been limited. Successful implementation of ED-based HCT is dependent on patient and provider acceptance of the intervention, and their attitudes and pre-existing biases towards the disease. This study sought to develop validated survey instruments to assess attitudes towards ED-based HCT.MethodsThis cross-sectional study surveyed patients and providers in three EDs in the Eastern Cape province, South Africa. A convenience sample of patients and providers in the ED were surveyed. Exploratory factor analysis was conducted using questions on attitudes to HIV testing to develop validated survey instruments. An ANOVA test assessed variance in attitudes towards HCT based on demographic variables collected.ResultsA total of 104 patient and 132 provider surveys were completed. Exploratory factor analysis resulted in a 17- and 7-question attitudes survey for patients and providers, respectively. Overall, 92.3% of patients and 70.7% of providers supported ED-based HCT, however, both groups displayed only mildly positive attitudes. Questions representing ‘confidentiality’ and ‘stigma around HIV testing’ had the least positive influence on patients’ overall attitudes. Questions representing ‘comfort with HIV testing’ had the least positive influence on providers’ overall attitudes.ConclusionOur study demonstrated ED patients and providers are generally supportive of ED-based HCT. A validated survey instrument was able to provide a standardized approach to identify barriers to HCT implementation in an ED setting, across contexts. For successful implementation, behavioral interventions must focus on strengthening patient beliefs around confidentiality and the consent process, and providers’ comfort levels with providing HIV testing services in the ED.

Improving Fairness and Privacy in Selection Problems

Supervised learning models have been increasingly used for making decisions about individuals in applications such as hiring, lending, and college admission. These models may inherit pre-existing biases from training datasets and discriminate against protected attributes (e.g., race or gender). In addition to unfairness, privacy concerns also arise when the use of models reveals sensitive personal information. Among various privacy notions, differential privacy has become popular in recent years. In this work, we study the possibility of using a differentially private exponential mechanism as a post-processing step to improve both fairness and privacy of supervised learning models. Unlike many existing works, we consider a scenario where a supervised model is used to select a limited number of applicants as the number of available positions is limited. This assumption is well-suited for various scenarios, such as job application and college admission. We use ``equal opportunity'' as the fairness notion and show that the exponential mechanisms can make the decision-making process perfectly fair. Moreover, the experiments on real-world datasets show that the exponential mechanism can improve both privacy and fairness, with a slight decrease in accuracy compared to the model without post-processing.

Nativeness, social distance and structural convergence in dialogue

ABSTRACT This study extends the logic of prior studies showing phonetic convergence between interlocutors to the structural domain. We ask whether listeners’ adaptation of the syntactic forms they produce depends on their perceptions about their interlocutor's social proximity and linguistic competence, using structural priming as a measure of convergence. Two experiments compared structural priming in dialogues between native British English speakers and (i) other native British English speakers, (ii) native speakers of North American English, and (iii) non-native speakers of English, to assess to what extent interlocutor characteristics influence structural convergence in dialogue. Our findings suggest that rates of structural convergence depend both on a speaker's pre-existing structural biases for particular verbs, and their perception of (linguistic or social) similarity to their interlocutor. This suggests that low-level mechanisms underlying structural convergence may be mediated by beliefs about how interlocutors are socially situated with respect to each other.

Challenges to Science Communication in a Post-Truth World

The communication of scientific hypotheses, and more broadly of the provisional and non-absolutist nature of science in general, to a lay audience is being increasingly compromised by the public's distrust of experts. This distrust frequently takes the form of hyperskepticism, expressed in unfounded fears and moral panics, and is amplified by a clickbait driven media. Clickbait is a text or a thumbnail link that is designed to entice users to follow the link, where the linked piece of content is often simplistic, salacious or misleading. The sensational headlines and summaries are then widely shared on social media by users with pre-existing political and psychological biases. This “attention economy” can result in the “wisdom of the crowd” – as expressed on social media and clickbait driven websites – drowning out the views of subject experts, and allowing for fearmongering and skepticism or distrust of science to take precedence. This article discusses ways in which scientists and communicators of scientific and other complex topics can help laypersons to understand the ways in which they might be prone to being misled. It also introduces concepts that will contribute to their better understanding and communication of scientific claims in a polarised and politicised digital age.

Shining a light on hippocampal remapping

In this issue of Neuron, McKenzie etal. (2021) test the degree to which pre-existing biases in hippocampal circuits constrict the encoding of new information via artificial induction of place cell remapping. Their results suggest that the hippocampal spatial map encodes new information via pre-existing latent place fields.

Ineffective nuptial gifts suggest female emancipation from sensory exploitation

Male sensory exploitation of female gustatory pre-existing bias has been proposed for the origin of nuptial gifts in insects and spiders. This sexual trait may have been beneficial to both sexes, giving mating and survival advantages to males and providing nutritional resources for females. However, the evolution of deceptive worthless gifts is against females’ interests and may trigger a co-evolutionary change in females’ preferences. We evaluated females’ preferences for nuptial gifts and the adaptive function of the gift in the spider Trechaleoides keyserlingi. The genus belongs to the understudied Neotropical family Trechaleidae in which nuptial gifts are widespread. The family is composed of only two species, and the gift seems to be absent in the sister species, creating a relevant scenario for understanding co-evolutionary processes. In the laboratory, we found that although males invested more in nuptial gifts when encountering mated females compared to unmated, they had similar mating access and duration than males lacking a gift. We also found an absence of female choice between males offering nutritive and worthless gifts. Few females were aggressive and cannibalized males, and we did not find evidence that the gift protected males from cannibalism. In the field, 50% of the gifts were worthless items. This is congruent with the laboratory findings where males offering worthless gifts seem to better attract females, which we discuss in the context of exploitation of female gustatory bias. We therefore propose that females may have evolved indifference for the gift and that gift-giving in this species represents a currently non-functional remnant of a behaviour. Nutritive nuptial gifts can exploit female gustatory preferences, with mutual benefits for both sexes: males can increase mating success and survival, while females increase their fecundity. But males can offer worthless gifts leading females to suboptimal matings, and in turn females can evolve indifference for the trait. The spider genus Trechaleoides is ideal to examine this process because gift-giving behaviour is present in one species and absent in the other. We examined females’ preferences for nuptial gifts and its function for males in the gift-giving species T. keyserlingi. We found that males invest in a gift but gain no reproductive advantage, as females were equally likely to mate with them regardless of whether they offered a gift or whether the gift was nutritive or worthless. We propose that females may have changed their preferences and that the gift is a remnant non-functional trait.

Old Wine in a New Bottle: How to Teach the Comparative Politics of the Middle East with Fiction

U.S. undergraduates often hold ill-informed and stereotypical perceptions about the Middle East. While theories of comparative politics can help undergraduates gain a more nuanced view of the region, these theories often strike students as extremely abstract and unrelated to actual experience. The use of novels from the Middle East can help instructors overcome both problems. This paper demonstrates how one such work, In the Country of Men by Hisham Matar, can illustrate different comparative politics theories while challenging preexisting biases about the region that students may hold.

Are we paying-to-play? A quantitative assessment of Canadian open access research in ecology and evolution

Open access (OA) allows for peer-reviewed research to be freely accessed and there has been a collective shift from both researchers and publishers towards more OA publishing. OA typically occurs either through article-processing charges (the gold road) or via self-archiving (the green road); the former can be expensive, while the latter has seen minimal uptake. The gold road of OA has led to predatory publishers and, to some, questionable publications. Here, I used publicly available grant information in Canada and combined this with individual publishing statistics to test a variety of factors and their influence on OA publishing. I showed that an individual’s award amount, H-index, and gender did not influence the proportion of OA articles they published, but an individual’s H-index scaled with the number of OA publications. Institute size influenced OA publishing patterns, with researchers at large universities (i.e., >20 000 full-time students) publishing proportionately more OA articles than medium and small institutes. I discuss the potential for this pattern to build on pre-existing systemic biases when it comes to funding and publishing.

Bias Preservation in Machine Learning: The Legality of Fairness Metrics Under EU Non-Discrimination Law

Western societies are marked by diverse and extensive biases and inequality that are unavoidably embedded in the data used to train machine learning. Algorithms trained on biased data will, without intervention, produce biased outcomes and increase the inequality experienced by historically disadvantaged groups. Recognising this problem, much work has emerged in recent years to test for bias in machine learning and AI systems using various fairness and bias metrics. Often these metrics address technical bias but ignore the underlying causes of inequality. In this paper we make three contributions. First, we assess the compatibility of fairness metrics used in machine learning against the aims and purpose of EU non-discrimination law. We show that the fundamental aim of the law is not only to prevent ongoing discrimination, but also to change society, policies, and practices to ‘level the playing field’ and achieve substantive rather than merely formal equality. Based on this, we then propose a novel classification scheme for fairness metrics in machine learning based on how they handle pre-existing bias and thus align with the aims of non-discrimination law. Specifically, we distinguish between ‘bias preserving’ and ‘bias transforming’ fairness metrics. Our classification system is intended to bridge the gap between non-discrimination law and decisions around how to measure fairness in machine learning and AI in practice. Finally, we show that the legal need for justification in cases of indirect discrimination can impose additional obligations on developers, deployers, and users that choose to use bias preserving fairness metrics when making decisions about individuals because they can give rise to prima facie discrimination. To achieve substantive equality in practice, and thus meet the aims of the law, we instead recommend using bias transforming metrics. To conclude, we provide concrete recommendations including a user-friendly checklist for choosing the most appropriate fairness metric for uses of machine learning and AI under EU non-discrimination law.

Implicit and Explicit Attitudes Toward Prosecutors and Defense Attorneys

Research consistently finds that jurors bring preexisting attitudes and opinions, which can influence trial outcomes, into the courtroom. This research seeks to understand if juror decision-making is influenced by implicit and explicit perceptions of legal actor trustworthiness. Participants were recruited from Amazon’s Mechanical Turk online platform and were randomly assigned to read a summarized trial scenario and render a guilty or not guilty verdict. Two of the three trials differed according to presentation of “compromising” evidence by either the prosecution or the defense, with the third serving as a control condition. To establish if participants had trustworthy or untrustworthy implicit attitudes toward prosecutors or defense attorneys, participants completed an Implicit Association Test (IAT). Participants then reported their explicit attitudes toward prosecutors and defense attorneys, in addition to providing their demographic information and attitudes toward the criminal legal system. Results indicate that implicit and explicit biases, as well as certain socio-demographics, are associated with verdicts in the trial scenarios. However, in particularly ambiguous cases (control condition), preexisting implicit biases of legal actor trustworthiness appear to inform verdicts more than explicit attitudes. These results suggest that attorney reliance on explicit attitudes during voir dire may be more informative, except in cases in which the evidence for both the prosecution and defense is particularly ambiguous.

Student-Reported Attitudes during an Interprofessional Palliative Care Learning Experience: Implications for Dual-Professional Identity, Interdisciplinary Bias, and Patient Outcomes.

Background: The geriatric population in the United States is in need of palliative care (PC), yet it is not consistently established in the curriculum across health care training programs. There is a clarion call to reform the education of health care students using interprofessional education (IPE). The Joint Commission reported that communication errors represent two-thirds of the causes behind provider sentinel events in health care.Objective: The purpose of this study was to design, implement, and assess an IPE curriculum on PC to understand interprofessional student attitudes.Design/Setting: Three professors conducted a mixed-methods study at a California university involving an IPE PC event for 40 nursing and speech-language pathology students, and administered the Interprofessional Attitudes Survey (IPAS) and reflective questions.Results: Qualitative findings indicated that students increased their knowledge about PC and the purpose/value of IPE. Four out of the five IPAS subscales had positive outcomes: teamwork and roles/responsibilities, patient-centeredness, diversity/ethics, and community-centeredness. Interprofessional-biases subscale revealed that 33% of the participants reported biases toward students from other health care disciplines, and 35% reported that students from other health care disciplines held similar biases toward them. However, only 25% did not believe that the interdisciplinary biases interfered with patient outcomes.Conclusion: The study identified the existence of interprofessional biases and prejudices that may impede collaboration among health care professionals resulting in reduced health care outcomes. Faculty and health educators are encouraged to embed IPE into a multidisciplinary curriculum that dismantles preexisting interdisciplinary biases and stereotypes, and constructs dual-professional identity. IRB ID #904203-1

Traits for depression related to agentic and external control

Depression has been linked to weakened perceptions of control. The experimental evidence derives from tasks with exposure to stable action-outcome contingencies. One assumption has been that performance represents a general cognitive bias that might manifest itself by a global performance difference. Another view is that people have specific situational perceptions of control reflecting their recent actions and the contingencies to which they are currently experiencing. In an experiment with N = 179, participants acquired one of four action-outcome sequences (Constant or Variable). We measured how learning was reflected in ratings of control and probability of responding in relation to mood. In three experimental treatments, the overall contingency across training involved an average moderate degree of control (ΔP = 0.25), but differed in how control varied (Constant or one of two Variable treatments). A fourth, control treatment involved a Constant zero degree of control (ΔP = 0.00). Participants rated their control before, during and after each sequence, providing measures of pre-existing bias, ratings of control in specific situations and generalised control perceptions. Specific control ratings were only influenced by the contingency experience and not pre-existing bias. Higher scores on the Beck’s depression inventory were associated with weakened association between action and context ratings. Overall, these data suggest that human agency is related to rates of responding and that mood is related to a difference in sensitivity to the ratings of and responding to the context.

Using Machine Learning to Advance Early Warning Systems: Promise and Pitfalls

Background/Context Early warning indicators (EWI) are often used by states and districts to identify students who are not on track to finish high school, and provide supports/interventions to increase the odds the student will graduate. While EWI are diverse in terms of the academic behaviors they capture, research suggests that indicators like course failures, chronic absenteeism, and suspensions can help identify students in need of additional supports. In parallel with the expansion of administrative data that have made early versions of EWI possible, new machine learning methods have been developed. These methods are data-driven and often designed to sift through thousands of variables with the purpose of identifying the best predictors of a given outcome. While applications of machine learning techniques to identify students at-risk of high school dropout have obvious appeal, few studies consider the benefits and limitations of applying those models in an EWI context, especially as they relate to questions of fairness and equity. Focus of Study In this study, we will provide applied examples of how machine learning can be used to support EWI selection. The purpose is to articulate the broad risks and benefits of using machine learning methods to identify students who may be at risk of dropping out. We focus on dropping out given its salience in the EWI literature, but also anticipate generating insights that will be germane to EWI used for a variety of outcomes. Research Design We explore these issues by using several hypothetical examples of how ML techniques might be used to identify EWI. For example, we show results from decision tree algorithms used to identify predictors of dropout that use simulated data. Conclusions/Recommendations Generally, we argue that machine learning techniques have several potential benefits in the EWI context. For example, some related methods can help create clear decision rules for which students are a dropout risk, and their predictive accuracy can be higher than for more traditional, regression-based models. At the same time, these methods often require additional statistical and data management expertise to be used appropriately. Further, the black-box nature of machine learning algorithms could invite their users to interpret results through the lens of preexisting biases about students and educational settings.

Sign tracking predicts cue-induced but not drug-primed reinstatement to methamphetamine seeking in rats: Effects of oxytocin treatment.

The incentive sensitisation theory of addiction posits that drug-associated stimuli become imbued with incentive motivational properties, driving pathological drug seeking. However, pre-existing variability in the incentive salience to non-drug reward cues ('sign trackers' (STs); 'goal trackers' (GTs)) is also predictive of the desire for and relapse to cocaine and opioids. Here, we asked whether variation in propensity to attribute incentive salience to a food cue is predictive of reinstatement to the highly addictive psychostimulant methamphetamine (METH), and whether treatment with the promising anti-addiction therapy oxytocin differentially reduces METH behaviour between STs and GTs. Rats were trained to associate a Pavlovian cue with delivery of a sucrose pellet over 8 days. They then received jugular vein catheters for intravenous METH self-administration, followed by behavioural extinction, and cue-induced and METH-primed reinstatement to METH-seeking behaviours. Oxytocin was administered prior to self-administration and reinstatement tests. Despite the self-administration of similar amounts of METH, STs reinstated more to METH cues than did GTs, yet METH-priming reinstated STs and GTs similarly. Furthermore, oxytocin attenuated cue-induced reinstatement more so in STs than in GTs, and reduced METH-primed reinstatement to a greater extent in the top quartile of reinstaters, indicating that oxytocin treatment may be most effective for those at highest risk of addiction. This pre-existing bias towards reward cues presents a possible tool to screen for METH addiction susceptibility and may be useful for understanding the neurobiology of addiction and for pharmacotherapeutic discovery.

Convergent probabilistic cues do not trigger syntactic adaptation: Evidence from self-paced reading.

Previous work has ostensibly shown that readers rapidly adapt to less predictable ambiguity resolutions after repeated exposure to unbalanced statistical input (e.g., a high number of reduced relative-clause garden-path sentences), and that these readers grow to disfavor the a priori more frequent (e.g. main verb) resolution after exposure (Fine, Jaeger, Farmer, & Qian, 2013). However, recent work has failed to replicate effects indicating a penalty for the a priori preferred, more frequent continuation, despite finding a speedup in syntactic repair times after initial exposure to the dispreferred, infrequent structure (Harrington Stack, James, & Watson, 2018). The current study reports three self-paced reading experiments that test whether co-occurring cues (explicit comprehension questions, preceding semantic cues, and font color) help facilitate adaptation to reduced relative/main verb garden-path sentences. Results suggest that readers do not overcome preexisting expectation biases by rapidly adapting to statistically novel linguistic contexts even with convergent probabilistic cues. An emphasis is placed on the difference between syntactic satiation effects and expectation adaptation, the latter of which we argue can only be determined through a penalty for an a priori preferred resolution after repeated exposure to its a priori less-preferred counterpart. (PsycInfo Database Record (c) 2020 APA, all rights reserved).

An ode to conference proceedings from the U.S. Southwest

In this era of online indexing and scholarly social media, it is easier than ever for researchers to find papers and authors that they are interested in. But as access to information expands, choices over what information to consume are increasingly driven by pre-existing prejudices and confirmation bias. In this context, it takes ever more discipline for scholars to not become locked in their own echo chambers. All of us can see the damage that this same tendency is doing to our public life, so we should try to guard against the same thing happening in scholarship.