Outcome bias and hindsight bias are important in philosophical debates and have wide-ranging implications outside of philosophy. Recently, Hedden has articulated a novel line of argumnt that the empirical evidence for what he labels hindsight bias is largely misguided and that empirical researchers who postulate such a bias are engaged in a fallacy fallacy. In this paper, I articulate Hedden’s core insights in terms of two principles and argue that in the relevant empirical research, these (...) principles are often (i) recognized, (ii) invoked in theoretical explanation, and (iii) guiding study designs and interpretation of data. On this basis, I argue that there is ample empirical evidence for outcome and hindsight bias and that cognitive psychologists do not generally exhibit a fallacy fallacy of postulating such biases in cases where there are none. More constructively, the present paper articulates some general principles that help us to better understand when an empirical study provides evidence of these types of bias. Thus, the debate casts light on both the specific evidence for outcome and hindsight bias but also on the general division of labor between philosophy and psychology. (shrink)

We document outcome bias in situations where an agent makes risky financial decisions for a principal. In three experiments, we show that the principal’s evaluations and financial rewards for the agent are strongly affected by the random outcome of the risky investment. This happens despite her exact knowledge of the investment strategy, which can, therefore, be assessed independently of the outcome. The principal thus judges the same decision by the agent differently, depending on factors that the (...) agent has no influence on. The effect of outcomes persists in a setting where principals communicate a preferred investment level. Principals are more satisfied with the agent after a random success when the agent did not follow the requested investment level, than after a failed investment that followed their explicit request. (shrink)

Three experiments demonstrated that decisions resulting in considerable amounts of profit, but missed alternative outcomes of greater profits, were rated lower in quality and produced more regret than did decisions that returned lesser amounts of profit but either did not miss or missed only slightly better alternatives. These effects were mediated by upward counterfactuals and moderated by participants’ orientation to the decision context. That decision evaluations were affected by the availability and magnitude of alternative outcomes rather than the positivity of (...) actual outcomes is counter to the outcome bias effect—a bias in which decisions are rated more positively when they led to more positive outcomes. Experiment 3 demonstrated that these effects represent a bias that occurs even when it is clear that the process by which decisions were made followed rational decision processes. This research suggests that when alter.. (shrink)

In a series of ten preregistered experiments (N=2043), we investigate the effect of outcome valence on judgments of probability, negligence, and culpability – a phenomenon sometimes labelled moral (and legal) luck. We found that harmful outcomes, when contrasted with neutral outcomes, lead to increased perceived probability of harm ex post, and consequently to increased attribution of negligence and culpability. Rather than simply postulating a hindsight bias (as is common), we employ a variety of empirical means to demonstrate that (...) the outcome-driven asymmetry across perceived probabilities constitutes a systematic cognitive distortion. We then explore three distinct strategies to alleviate the hindsight bias and its downstream effects on mens rea and culpability ascriptions. Not all are successful, but at least some prove promising. They should, we argue, be taken into consideration in criminal jurisprudence, where distortions due to the hindsight bias are likely considerable and deeply disconcerting. (shrink)

In a series of ten preregistered experiments (N=2043), we investigate the effect of outcome valence on judgments of probability, negligence, and culpability – a phenomenon sometimes labelled moral (and legal) luck. We found that harmful outcomes, when contrasted with neutral outcomes, lead to increased perceived probability of harm ex post, and consequently to increased attribution of negligence and culpability. Rather than simply postulating a hindsight bias (as is common), we employ a variety of empirical means to demonstrate that (...) the outcome-driven asymmetry across perceived probabilities constitutes a systematic cognitive distortion. We then explore three distinct strategies to alleviate the hindsight bias and its downstream effects on mens rea and culpability ascriptions. Not all are successful, but at least some prove promising. They should, we argue, be taken into consideration in criminal jurisprudence, where distortions due to the hindsight bias are likely considerable and deeply disconcerting. (shrink)

It has consistently been shown that agents judge the intervals between their actions and outcomes as compressed in time, an effect named intentional binding. In the present work, we investigated whether this effect is result of prior bias volunteers have about the timing of the consequences of their actions, or if it is due to learning that occurs during the experimental session. Volunteers made temporal estimates of the interval between their action and target onset , or between two events (...) . Our results show that temporal estimates become shorter throughout each experimental block in both conditions. Moreover, we found that observers judged intervals between action and outcomes as shorter even in very early trials of each block. To quantify the decrease of temporal judgments in experimental blocks, exponential functions were fitted to participants’ temporal judgments. The fitted parameters suggest that observers had different prior biases as to intervals between events in which action was involved. These findings suggest that prior bias might play a more important role in this effect than calibration-type learning processes. (shrink)

Sequences of numbers representing prior resource size were presented to participants in a common-pool resource dilemma. The numbers were sampled from uniform probability distributions with either a low variance (low resource uncertainty) or a high variance (high resource uncertainty). Presentations were both sequential and simultaneous. Three groups of 16 undergraduates either estimated the size of the resource when it did not represent value to them; requested an amount from the resource, identified with a sum of money, when the outcome (...) of the requests only depended on resource size; or requested from the resource (sum of money) when the outcome of the requests depended on both resource size and how much others in a group requested. In support of an individual outcome-desirability bias due to selective recall of the number sequences, after sequential presentation larger requests were observed when resource uncertainty was high than when it was low. No effects of resource uncertainty or presentation were found on the estimates of resource size. Whether or not the outcome of the requests depended on others' requests made little difference. (shrink)

The present comment concerns a recent AI & Ethics article which purports to report evidence of speciesist bias in various popular computer vision (CV) and natural language processing (NLP) machine learning models described in the literature. I examine the authors' analysis and show it, ironically, to be prejudicial, often being founded on poorly conceived assumptions and suffering from fallacious and insufficiently rigorous reasoning, its superficial appeal in large part relying on the sequacity of the article's target readership.

BackgroundPublication and outcome reporting bias is often caused by researchers selectively choosing which scientific results and outcomes to publish. This behaviour is ethically significant as it distorts the literature used for future scientific or clinical decision-making. This study investigates the practicalities of using ethics applications submitted to a UK National Health Service research ethics committee to monitor both types of reporting bias.MethodsAs part of an internal audit we accessed research ethics database records for studies submitting an end (...) of study declaration to the Hampshire A research ethics committee between 1st January 2010 and 31st December 2011. A literature search was used to establish the publication status of studies. Primary and secondary outcomes stated in application forms were compared with outcomes reported in publications.ResultsOut of 116 studies the literature search identified 57 publications for 37 studies giving a publication rate of 32 %. Original Research Ethics Committee applications could be obtained for 28 of the published studies. Outcome inconsistencies were found in 16 of the published studies.ConclusionsThis study showed that the problem of publication and outcome reporting bias is still significant in the UK. The method described here demonstrates that UK NHS research ethics committees are in a good position to detect such bias due to their unique access to original research protocols. Data gathered in this way could be used by the Health Research Authority to encourage higher levels of transparency in UK research. (shrink)

As algorithms have become ubiquitous in consequential domains, societal concerns about the potential for discriminatory outcomes have prompted urgent calls to address algorithmic bias. In response, a rich literature across computer science, law, and ethics is rapidly proliferating to advance approaches to designing fair algorithms. Yet computer scientists, legal scholars, and ethicists are often not speaking the same language when using the term ‘bias.’ Debates concerning whether society can or should tackle the problem of algorithmic bias are (...) hampered by conflations of various understandings of bias, ranging from neutral deviations from a standard to morally problematic instances of injustice due to prejudice, discrimination, and disparate treatment. This terminological confusion impedes efforts to address clear cases of discrimination. -/- In this paper, we examine the promises and challenges of different approaches to disambiguating bias and designing for justice. While both approaches aid in understanding and addressing clear algorithmic harms, we argue that they also risk being leveraged in ways that ultimately deflect accountability from those building and deploying these systems. Applying this analysis to recent examples of generative AI, our argument highlights unseen dangers in current methods of evaluating algorithmic bias and points to ways to redirect approaches to addressing bias in generative AI at its early stages in ways that can more robustly meet the demands of justice. (shrink)

When interests and preferences of researchers or their sponsors cause bias in experimental design, data interpretation or dissemination of research results, we normally think of it as an epistemic shortcoming. But as a result of the debate on science and values, the idea that all extra-scientific influences on research could be singled out and separated from pure science is now widely believed to be an illusion. I argue that nonetheless, there are cases in which research is rightfully regarded as (...) epistemologically deficient due to the influence of preferences on its outcomes. I present examples from biomedical research and offer an analysis in terms of social epistemology. (shrink)

Humans typically display hindsight bias. They are more confident that the evidence available beforehand made some outcome probable when they know the outcome occurred than when they don't. There is broad consensus that hindsight bias is irrational, but this consensus is wrong. Hindsight bias is generally rationally permissible and sometimes rationally required. The fact that a given outcome occurred provides both evidence about what the total evidence available ex ante was, and also evidence about (...) what that evidence supports. Even if you in fact evaluate the ex ante evidence correctly, you should not be certain of this. Then, learning the outcome provides evidence that if you erred, you are more likely to have erred low rather than high in estimating the degree to which the ex ante evidence supported the hypothesis that that outcome would occur. (shrink)

When ethical decisions have to be taken in critical, complex medical situations, they often involve decisions that set the course for or against life-sustaining treatments. Therefore the decisions have far-reaching consequences for the patients, their relatives, and often for the clinical staff. Although the rich psychology literature provides evidence that reasoning may be affected by undesired influences that may undermine the quality of the decision outcome, not much attention has been given to this phenomenon in health care or ethics (...) consultation. In this paper, we aim to contribute to the sensitization of the problem of systematic reasoning biases by showing how exemplary individual and group biases can affect the quality of decision-making on an individual and group level. We are addressing clinical ethicists as well as clinicians who guide complex decision-making processes of ethical significance. Knowledge regarding exemplary group psychological biases (e.g. conformity bias), and individual biases (e.g. stereotypes), will be taken from the disciplines of social psychology and cognitive decision science and considered in the field of ethical decision-making. Finally we discuss the influence of intuitive versus analytical (systematical) reasoning on the validity of ethical decision-making. (shrink)

Future-bias is the preference, all else being equal, for negatively valenced events be located in the past rather than the future, and positively valenced ones to be located in the future rather than the past. Strong risk aversion is the preference to pay some cost to mitigate the badness of the worst outcome. People who are both strongly risk averse and future-biased can face a series of choices that will guarantee them more pain, for no compensating benefit: they (...) will be pain pumped. Thus, combining these preferences is rationally impermissible. Dougherty (2011) argues that this gives us reason to think that future-bias is rationally impermissible. This argument, and a similar one presented by Greene and Sullivan (2015), relies on the idea that if several preferences are not rationally combinable, then in the absence of some explanation of why they are not combinable, we should conclude that (at least) one of them is rationally impermissible. We take up this question by, inter alia, reflecting on the empirical results of a study we ran that probes people’s risk averse and future-biased preferences. We argue that the data, in conjunction with other considerations, suggests that we should not infer from the irrationality of the combination of these preferences to the irrationality of one of the preferences singly. We also argue that given the descriptive data about these preferences, there is no reason to think that, if one of them is rationally impermissible, then it is future-bias. (shrink)

In this article, I describe how gender bias can affect the design, testing, clinical trials, regulatory approval, and clinical use of implantable devices. I argue that bad outcomes experienced by women patients are a cumulative consequence of small biases and inattention at various points of the design, testing, and regulatory process. However, specific instances of inattention and bias can be difficult to identify, and risks are difficult to predict. This means that even if systematic gender bias in (...) implant design is an ethical issue, it is one with no clearly blameworthy player. From a practical perspective, there is no single obvious point at which to intervene. Philosophers working in other areas have explored structurally similar moral problems—sometimes referred to as “moral aggregation problems”—such as the type of environmental harm caused by small actions of many players. I describe key features of these type of problems and strategies to address them. I then draw on these to suggest an approach to gender bias in medical implant design and use. (shrink)

Some recent work in philosophy of religion addresses what can be called the “axiological question,” i.e., regardless of whether God exists, would it be good or bad if God exists? Would the existence of God make the world a better or a worse place? Call the view that the existence of God would make the world a better place “Pro-Theism.” We argue that Pro-Theism is not implausible, and moreover, many Theists, at least, (often implicitly) think that it is true. That (...) is, many Theists think that various good outcomes would arise if Theism is true. We then discuss work in cognitive science concerning human cognitive bias, before discussing two noteworthy attempts to show that at least some religious beliefs arise because of cognitive bias: Hume’s, and Draper’s and Nichols’s. We then argue that, as a result of certain cognitive biases that result when good outcomes might be at stake, Pro-Theism causes many Theists to inflate the epistemic probability that God exists, and as a result, Theists should lower the probability they assign to God’s existence. Finally, based our arguments, we develop a novel objection to Pascal’s wager. (shrink)

When Artificial Intelligence (AI) is applied in decision-making that affects people’s lives, it is now well established that the outcomes can be biased or discriminatory. The question of whether algorithms themselves can be among the sources of bias has been the subject of recent debate among Artificial Intelligence researchers, and scholars who study the social impact of technology. There has been a tendency to focus on examples, where the data set used to train the AI is biased, and denial (...) on the part of some researchers that algorithms can also be biased. Here we illustrate the point that algorithms themselves can be the source of bias with the example of collaborative filtering algorithms for recommendation and search. These algorithms are known to suffer from cold-start, popularity, and homogenizing biases, among others. While these are typically described as statistical biases rather than biases of moral import; in this paper we show that these statistical biases can lead directly to discriminatory outcomes. The intuitive idea is that data points on the margins of distributions of human data tend to correspond to marginalized people. The statistical biases described here have the effect of further marginalizing the already marginal. Biased algorithms for applications such as media recommendations can have significant impact on individuals’ and communities’ access to information and culturally-relevant resources. This source of bias warrants serious attention given the ubiquity of algorithmic decision-making. (shrink)

Most of us are hedonically future-biased: other things being equal, we prefer pains to be in the past and pleasures to be in the future. Recently, various authors have argued that future bias is irrational, and that we should be temporally neutral instead. I argue that instead of temporal neutrality, the putative counterexamples and the rationales offered for them only motivate a more narrow principle I call Only Action Fixes Utility: it is only when you act on the basis (...) of assigning a utility to an outcome that rationality requires you to give it the same value retrospectively and prospectively, other things being equal. When hedonic experiences are untethered from action, hedonic future bias is rationally permissible. I support this principle by appeal to additional scenarios and more general asymmetries between agential and experiential goods. (shrink)

Political fact-checkers evaluate the truthfulness of politicians’ claims. This paper contributes to an emerging scholarly debate on whether fact-checkers treat political parties differently in a systematic manner depending on their ideology (bias). We first examine the available approaches to analyze bias and then present a new approach in two steps. First, we propose a logistic regression model to analyze the outcomes of fact-checks and calculate how likely each political party will obtain a truth score. We test our model (...) with a sample of fact-checks from Newtral, a major Spanish fact-checker. Our model would signal bias under two assumptions: a) all political parties are on average equally accurate in their statements; b) the verification method gives precise instructions and is implemented systematically. We investigate this second assumption with a series of interviews with Newtral fact-checkers. We show that standard verification protocols are so loosely implemented that fact-checks reflect a set of journalistic decisions, rather than a bias in the statistical sense. We call for a more rigorous definition of verification methods as a pre-requisite for an unbiased assessment of politician’s claims. (shrink)

This paper presents a series of studies (total N=579) which demonstrate that folk judgments concerning the reasonableness of decisions and actions depend strongly on whether they engender positive or negative consequences. A particular decision is deemed more reasonable in retrospect when it produces beneficial consequences than when it produces harmful consequences, even if the situation in which the decision was taken and the epistemic circumstances of the agent are held fixed across conditions. This finding is worrisome for the law, where (...) the reasonable person standard plays a prominent role. The legal concept of reasonableness is outcome-insensitive: whether the defendant acted in a reasonable fashion or not depends exclusively on her context of action, no matter how things play out. Folk judgments of reasonableness are thus inconsistent with the legal concept of reasonableness. Problematically, in common law jurisdictions, the decision whether a defendant’s behavior was reasonable or not is frequently (though not necessarily) delegated to a lay jury. (shrink)

In recent decades, economists have developed methods for measuring the country-wide level of inequality of opportunity. The most popular method, called the ex-ante method, uses data on the distribution of outcomes stratified by groups of individuals with the same circumstances, in order to estimate the part of outcome inequality that is due to these circumstances. I argue that these methods are potentially biased, both upwards and downwards, and that the unknown size of this bias could be large. To (...) argue that the methods are biased, I show that they ought to measure causal or counterfactual quantities, while the methods are only capable of identifying correlational information. To argue that the bias is potentially large, I illustrate how the causal complexity of the real world leads to numerous non-causal correlations between circumstances and outcomes and respond to objections claiming that such correlations are nonetheless indicators of unfair disadvantage, that is, inequality of opportunity. (shrink)

In this paper, we consider three competing explanations of the empirical finding that people’s causal attributions are responsive to normative details, such as whether an agent’s action violated an injunctive norm—the intervention view, the bias view, and the responsibility view. We then present new experimental evidence concerning a type of case not previously investigated in the literature. In the switch version of the trolley problem, people judge that the bystander ought to flip the switch, but they also judge that (...) she is more responsible for the resulting outcome when she does so than when she refrains. And, as predicted by the responsibility view, but not the intervention or bias views, people are more likely to say that the bystander caused the outcome when she flips the switch. (shrink)

Across the world, artificial intelligence (AI) technologies are being more widely employed in public sector decision-making and processes as a supposedly neutral and an efficient method for optimizing delivery of services. However, the deployment of these technologies has also prompted investigation into the potentially unanticipated consequences of their introduction, to both positive and negative ends. This paper chooses to focus specifically on the relationship between gender bias and AI, exploring claims of the neutrality of such technologies and how its (...) understanding of bias could influence policy and outcomes. Building on a rich seam of literature from both technological and sociological fields, this article constructs an original framework through which to analyse both the perpetuation and mitigation of gender biases, choosing to categorize AI technologies based on whether their input is text or images. Through the close analysis and pairing of four case studies, the paper thus unites two often disparate approaches to the investigation of bias in technology, revealing the large and varied potential for AI to echo and even amplify existing human bias, while acknowledging the important role AI itself can play in reducing or reversing these effects. The conclusion calls for further collaboration between scholars from the worlds of technology, gender studies and public policy in fully exploring algorithmic accountability as well as in accurately and transparently exploring the potential consequences of the introduction of AI technologies. (shrink)

Extant research has shown that ordinary causal judgments are sensitive to normative factors. For instance, agents who violate a norm are standardly deemed more causal than norm-conforming agents in identical situations. In this paper, we explore two competing explanations for the Norm Effect: the Responsibility View and the Bias View. According to the former, the Norm Effect arises because ordinary causal judgment is intimately intertwined with moral responsibility. According to the alternative view, the Norm Effect is the result of (...) a blame-driven bias. In a series of five preregistered experiments (N = 2688), we present evidence that predominantly favours the Bias View. In particular, and against predictions made by the Responsibility View, we show that participants deem agents who violate nonpertinent or silly norms – norms that do not relate to the outcome at hand – as more causal, and that this effect cannot be explained in terms of plausible mediators such as the agent’s foreknowledge and desire, or the foreseeability of harm. We close with a discussion regarding the implications of these findings, in as regards the just assessment of proximate cause in the law. (shrink)

Components of artificial intelligence (AI) for analysing social big data, such as natural language processing (NLP) algorithms, have improved the timeliness and robustness of health data. NLP techniques have been implemented to analyse large volumes of text from social media platforms to gain insights on disease symptoms, understand barriers to care and predict disease outbreaks. However, AI-based decisions may contain biases that could misrepresent populations, skew results or lead to errors. Bias, within the scope of this paper, is described (...) as the difference between the predictive values and true values within the modelling of an algorithm. Bias within algorithms may lead to inaccurate healthcare outcomes and exacerbate health disparities when results derived from these biased algorithms are applied to health interventions. Researchers who implement these algorithms must consider when and how bias may arise. This paper explores algorithmic biases as a result of data collection, labelling and modelling of NLP algorithms. Researchers have a role in ensuring that efforts towards combating bias are enforced, especially when drawing health conclusions derived from social media posts that are linguistically diverse. Through the implementation of open collaboration, auditing processes and the development of guidelines, researchers may be able to reduce bias and improve NLP algorithms that improve health surveillance. (shrink)

A coherent practice of mens rea (‘guilty mind’) ascription in criminal law presupposes a concept of mens rea which is insensitive to the moral valence of an action’s outcome. For instance, an assessment of whether an agent harmed another person intentionally should be unaffected by the severity of harm done. Ascriptions of intentionality made by laypeople, however, are subject to a strong outcome bias. As demonstrated by the Knobe effect, a knowingly incurred negative side effect is standardly (...) judged intentional, whereas a positive side effect is not. We report the first empirical investigation into intentionality ascriptions made by professional judges, which finds (i) that professionals are sensitive to the moral valence of outcome type, and (ii) that the worse the outcome, the higher the propensity to ascribe intentionality. The data shows the intentionality ascriptions of professional judges to be inconsistent with the concept of mens rea supposedly at the foundation of criminal law. (shrink)

The delineation of constituency boundaries and variations in vote distribution across districts often favor certain parties at the expense of others. Applying a hitherto under-utilized formula (Brookes, 1959; Johnston et al., 1999), this study investigates whether the mechanism translating votes into seats in Japan's single-member districts results in systematic partisan advantage that may influence election outcomes. Simulations are conducted for the 2003 and 2005 general elections under two scenarios: where the governing coalition and the main opposition party receive equal vote (...) shares, and where their vote shares are reversed from the actual results. Components of electoral bias are then disaggregated into size and distribution effects, and the impact of malapportionment, electorate size, turnout, and the role of third party/independent candidates on overall electoral bias is examined. Results show that while partisan bias exists, disadvantages toward one party in some components are likely to cancel out benefits derived from others, producing a relatively small net effect. Furthermore, electoral bias in Japan is found to award sectoral rather than partisan seat bonuses. (shrink)

Data from medical research are typically summarized with various types of outcome measures. We present three arguments in favor of absolute over relative outcome measures. The first argument is from cognitive bias: relative measures promote the reference class fallacy and the overestimation of treatment effectiveness. The second argument is decision-theoretic: absolute measures are superior to relative measures for making a decision between interventions. The third argument is causal: interpreted as measures of causal strength, absolute measures satisfy a (...) set of desirable properties, but relative measures do not. Absolute outcome measures outperform relative measures on all counts. (shrink)

More than 40 primary studies, and three recent systematic reviews and meta-analyses, have shown a clear association between pharmaceutical industry funding of clinical trials and pro-industry results. Industry sponsorship biases published scientific research in favour of the sponsors, a result of the strong interest commercial sponsors have in obtaining favourable results.Three proposed remedies to this problem are widely agreed upon among those concerned with the level of sponsorship bias: financial disclosure, reporting standards and trial registries. This paper argues that (...) all of these remedies either fail to address the mechanisms by which pharmaceutical companies’ sponsorship leads to biased results—design bias, multiple trials with predictable outcomes, fraud, rhetorical effects and publication bias—or else only inadequately address those mechanisms. As a result, the policies normally proposed for dealing with sponsorship bias are unable to eliminate it. Only completely separating public clinical research from pharmaceutical industry funding can eliminate sponsorship bias. (shrink)

Bermúdez argues that a framing effect is rational, which will be true if one accepts that the biased editing phase is rational. This type of rationality was called procedural by Simon. Despite being procedurally rational in the evaluation phase framing effect stems from biased way we set a reference point against which outcomes are compared.

In this paper I argue that some examples of what we label ‘algorithmic bias’ would be better understood as cases of institutional bias. Even when individual algorithms appear unobjectionable, they may produce biased outcomes given the way that they are embedded in the background structure of our social world. Therefore, the problematic outcomes associated with the use of algorithmic systems cannot be understood or accounted for without a kind of structural account. Understanding algorithmic bias as institutional (...) class='Hi'>bias in particular (as opposed to other structural accounts) has at least two important upshots. First, I argue that the existence of bias that is intrinsic to certain institutions (whether algorithmic or not) suggests that at least in some cases, the algorithms now substituting as pieces of institutional norms or rules are not “fixable” in the relevant sense, because the institutions they help make up are not fixable. Second, I argue that in other cases, changing the algorithms being used within our institutions (rather than getting rid of them entirely) is essential to changing the background structural conditions of our society. (shrink)

Beliefs like the Gambler's Fallacy and the Hot Hand have interested cognitive scientists, economists, and philosophers for centuries. We propose that these judgment patterns arise from the observer's mental models of the sequence-generating mechanism, moderated by the strength of belief in an a priori base rate. In six behavioral experiments, participants observed one of three mechanisms generating sequences of eight binary events: a random mechanical device, an intentional goal-directed actor, and a financial market. We systematically manipulated participants’ beliefs about the (...) base rate probabilities at which different outcomes were generated by each mechanism. Participants judged 18 sequences of outcomes produced by a mechanism with either an unknown base rate, a specified distribution of three equiprobable base rates, or a precise, fixed base rate. Six target sequences ended in streaks of between two and seven identical outcomes. The most common predictions for subsequent events were best described as pragmatic belief updating, expressed as an increasingly strong expectation that a streak of identical signals would repeat as the length of that streak increased. The exception to this pattern was for sequences generated by a random mechanical device with a fixed base rate of.50. Under this specific condition, participants exhibited a bias toward reversal of streaks, and this bias was larger when participants were asked to make a dichotomous choice versus a numerical probability rating. We review alternate accounts for the anomalous judgments of sequences and conclude with our favored interpretation that is based on Rabin's version of Tversky & Kahneman's Law of Small Numbers. (shrink)

For nearly five years, bioethicists and neurologists debated whether Jahi McMath, an African American teenager, was alive or dead. While Jahi's condition provides a compelling study for analyzing brain death, circumscribing her life status to a question of brain death fails to acknowledge and respond to a chronic, if uncomfortable, bioethics problem in American health care—namely, racial bias and unequal treatment, both real and perceived. Bioethicists should examine the underlying, arguably broader social implications of what Jahi's medical treatment and (...) experience represented. On any given day, disparities in the quality of health care and health outcomes for people of color in comparison to whites are evidenced in American hospitals and clinics. These disparities are not entirely explained by differences in patient education, insurance status, employment, income, expressed preference for treatments, and severity of disease. Instead, research indicates that, even for African Americans able to gain access to health care services and navigate institutional nuances, disparities persist across a broad range of services, including diagnostic screening and general medical care, mental health diagnosis and treatment, pain management, HIV‐related care, and treatments for cancer, heart disease, diabetes, and kidney disease. (shrink)

One of the predominant debates in AI Ethics is the worry and necessity to create fair, transparent and accountable algorithms that do not perpetuate current social inequities. I offer a critical analysis of Reuben Binns’s argument in which he suggests using public reason to address the potential bias of the outcomes of machine learning algorithms. In contrast to him, I argue that ultimately what is needed is not public reason per se, but an audit of the implicit moral assumptions (...) of societies within which algorithms are built and applied. Public justification is appealing since it offers us the possibility to align the decision-making outcomes of the algorithm with the core moral values of stakeholders within a constitutional democratic society. My concern is that the common moral principles that form the foundation of public reason are not necessarily neutral, as they still express specific moral ideals and normative standards even if there is moral agreement by society as a whole, or among different stakeholders within society. Appealing to such normative standards may thus still lead to algorithmic outcomes being biased as common moral values may very well still be discriminatory even though they are formed from a consensus, and even if public reason is applied as a kind of filter for potential algorithmic outcomes. Hence, I argue that these implicit moral norms within society that we take as a given in public reasoning, need to be audited from generation to generation in order to effectively mitigate potential algorithmic bias. (shrink)

The ratio bias––according to which individuals prefer to bet on probabilities expressed as a ratio of large numbers to normatively equivalent or superior probabilities expressed as a ratio of small numbers––has recently gained momentum, with researchers especially in health economics emphasizing the policy importance of the phenomenon. Although the bias has been replicated several times, some doubts remain about its economic significance. Our two experiments show that the bias disappears once order effects are excluded, and once salient (...) and dominant incentives are provided. This holds true for both choice and valuation tasks. Also, adding context to the decision problem does not change this outcome. No ratio bias could be found in between-subject tests either, which leads us to the conclusion that the policy relevance of the phenomenon is doubtful at best. (shrink)

Ethical decision theory may not be sufficiently well developed to furnish reliable guidelines to people involved in complex decision making that involves conflict between ethical considerations and business imperatives such as making a profit. In conditions of ethical uncertainty hindsight bias may occur, and this study reports on an exploration of hindsight bias effects among participants in continuing education in business programmes. Perceptions of business ethics were found to differ among groups within the sample depending on what they (...) thought had been the outcome of the ethical decision concerned. (shrink)

Does hindsight knowledge make research seem more ethical and predictable? In line with the notion of hindsight bias, students in 3 experiments knowing the outcome of an animal experiment judged the results as more foreseeable and ethical relative to students who did not know the outcome. Via self to other comparisons, students evaluate themselves more favorably compared to a peer but exhibited hindsight bias in doing so. Uniquely, the findings reveal the possibility that students deem themselves (...) to be more skeptical and objective than their peers. Implications for teaching animal ethics and for Institutional Review Boards are discussed. (shrink)

Objectives: To investigate whether different methods of obtaining informed consent affected recruitment to a study of delirium in older, medically ill hospital inpatients.Design: Open randomised study.Setting: Acute medical service for older people in an inner city teaching hospital.Participants: Patients 70 years or older admitted to the unit within three days of hospital admission randomised into two groups.Intervention: Attempted recruitment of subjects to a study of the natural history of delirium. This was done by either a formal test of capacity, followed (...) by either a request for consent or an attempt at obtaining assent from a proxy, or a combined informal capacity/consent process.Main outcome measures: Prevalence and severity of delirium, and, as case mix measures, length of hospital stay and destination on discharge.Results: Recruitment of subjects through establishing formal capacity and then informed consent was less successful and, compared with those recruited through the usual combined capacity/consent approach, yielded a sample with less cognitive impairment, lower severity of delirium, lower probability of case note diagnosis of delirium and lower rate of entering a care home.Conclusions: Methods of obtaining informed consent may significantly influence the case mix of subjects recruited to a study of delirium. Stringent testing of capacity may exclude patients with delirium from studies, thus rendering findings less generalisable. A different method is necessary to achieve an ethical balance between respecting autonomy through obtaining adequate informed consent and avoiding sample bias. (shrink)

We are far too willing to reject the belief that much of what we see in life is random.Uncertainty is an everyday experience in medical research and practice, but theory and methods for reasoning clearly about uncertainty were developed only recently. Confirmation bias, selective memory, and many misleading heuristics are known enemies of the insightful clinician, researcher, or citizen, but other snares worth exposing may lurk in how we reason about uncertainty in our everyday lives. Here we draw attention (...) to a cognitive bias that is little known—or possibly unknown—to those outside the field of probability pedagogy. Described by Konold as the... (shrink)

In this paper we investigate risk prediction of criminal re-offense among juvenile defendants using general-purpose machine learning algorithms. We show that in our dataset, containing hundreds of cases, ML models achieve better predictive power than a structured professional risk assessment tool, the Structured Assessment of Violence Risk in Youth, at the expense of not satisfying relevant group fairness metrics that SAVRY does satisfy. We explore in more detail two possible causes of this algorithmic bias that are related to biases (...) in the data with respect to two protected groups, foreigners and women. In particular, we look at the differences in the prevalence of re-offense between protected groups and the influence of protected group or correlated features in the prediction. Our experiments show that both can lead to disparity between groups on the considered group fairness metrics. We observe that methods to mitigate the influence of either cause do not guarantee fair outcomes. An analysis of feature importance using LIME, a machine learning interpretability method, shows that some mitigation methods can shift the set of features that ML techniques rely on away from demographics and criminal history which are highly correlated with sensitive features. (shrink)

An unprocessed risk is a collection of simple lotteries with a reduction-rule that describes the actual-payoff to the decision-maker as a function of realized lottery outcomes. Experiments reveal that the willingness to pay for unprocessed risks is consistently biased toward the payoff-level in the unprocessed representation. The anchoring-to-frame bias in cases of positive framing is significantly weaker than in cases of negative framing suggesting that rational negativity bias may reflect in asymmetric violations of rationality.

The link between gaming and negative outcomes has been explored by previous research and has led to the widespread adverse attitude toward gaming (ATG) and gamers, especially from those who are unfamiliar with this activity. By implementing an audit study with gamers and non-gamers as participants (N = 1,280), we found that non-gamer participants rated gamers less as similar to their ideal marriage partners compared to non-gamers, while gamer participants did not differentiate between gamers and non-gamers in the ideal marriage (...) partners similarity rating (IMPSR). The findings also revealed that the difference in IMPSR between gamer and non-gamer participants toward gamers was completely mediated by their ATG. These results imply that non-gamers consider gaming as an undesired characteristic, and this is due to the relatively negative attitude of non-gamers toward gaming. Current study provides a new perspective on exploring the effect of gaming by investigating the social interaction between gamers and non-gamers in real-world and suggests that the unfamiliarity of gaming can lead to the negative ATG, which may, ultimately, place gamers at a disadvantage in the context of mate selection. (shrink)

This book explores the ethical problems of algorithmic bias and its potential impact on populations that experience health disparities by examining the historical underpinnings of explicit and implicit bias, the influence of the social determinants of health, and the inclusion of racial and ethnic minorities in data. Over the last twenty-five years, the diagnosis and treatment of disease have advanced at breakneck speeds. Currently, we have technologies that have revolutionized the practice of medicine, such as telemedicine, precision medicine, (...) big data, and AI. These technologies, especially AI, promise to improve the quality of patient care, lower health care costs, improve patient treatment outcomes, and decrease patient mortality. AI may also be a tool that reduces health disparities; however, algorithmic bias may impede its success. This book explores the risks of using AI in the context of health disparities. It is of interest to health services researchers, ethicists, policy analysts, social scientists, health disparities researchers, and AI policy makers. (shrink)

BackgroundDouble-blind peer review has been proposed as a possible solution to avoid implicit referee bias in academic publishing. The aims of this study are to analyse the demographics of corresponding authors choosing double-blind peer review and to identify differences in the editorial outcome of manuscripts depending on their review model.MethodsData includes 128,454 manuscripts received between March 2015 and February 2017 by 25 Nature-branded journals. We investigated the uptake of double-blind review in relation to journal tier, as well as (...) gender, country, and institutional prestige of the corresponding author. We then studied the manuscripts’ editorial outcome in relation to review model and author’s characteristics. The gender (male, female, or NA) of the corresponding authors was determined from their first name using a third-party service (Gender API). The prestige of the corresponding author’s institutions was measured from the data of the Global Research Identifier Database (GRID) by dividing institutions in three prestige groups with reference to the 2016 Times Higher Education (THE) ranking. We employed descriptive statistics for data exploration, and we tested our hypotheses using Pearson’s chi-square and binomial tests. We also performed logistic regression modelling with author update, out-to-review, and acceptance as response, and journal tier, author gender, author country, and institution as predictors.ResultsAuthor uptake for double-blind submissions was 12% (12,631 out of 106,373). We found a small but significant association between journal tier and review type (p value < 0.001, Cramer’s V = 0.054, df = 2). We had gender information for 50,533 corresponding authors and found no statistically significant difference in the distribution of peer review model between males and females (p value = 0.6179). We had 58,920 records with normalised institutions and a THE rank, and we found that corresponding authors from the less prestigious institutions are more likely to choose double-blind review (p value < 0.001, df = 2, Cramer’s V = 0.106). In the ten countries with the highest number of submissions, we found a large significant association between country and review type (p value < 0.001, df = 10, Cramer’s V = 0.189). The outcome both at first decision and post review is significantly more negative (i.e. a higher likelihood for rejection) for double-blind than single-blind papers (p value < 0.001, df = 1, Cramer’s V = 0.112 for first decision; p value < 0.001; df = 1, Cramer’s V = 0.082 for post-review decision).ConclusionsThe proportion of authors that choose double-blind review is higher when they submit to more prestigious journals, they are affiliated with less prestigious institutions, or they are from specific countries; the double-blind option is also linked to less successful editorial outcomes. (shrink)

Most phobias are focussed on a small number of fear-inducing stimuli (e.g., snakes, spiders). A review of the evidence supporting biological and cognitive explanations of this uneven distribution of phobias suggests that the readiness with which such stimuli become associated with aversive outcomes arises from biases in the processing of information about threatening stimuli rather than from phylogenetically based associative predispositions or “biological preparedness.” This cognitive bias, consisting of a heightened expectation of aversive outcomes following fear-relevant stimuli, generates and (...) maintains robust learned associations between them. Some of the features of such stimuli which determine this expectancy bias are estimates of how dangerous they are, the semiotic similarity between them and their aversive outcomes, and the degree of prior fear they elicit. Ontogenetic and cultural factors influence these features of fear-relevant stimuli and are hence important in determining expectancy bias. The available evidence does not exclude the possibility that both expectancy biases and specific evolved predispositions coexist, but the former can explain a number of important findings that the latter cannot. (shrink)

Background: The pregnancy and psychological status of infertile couples has always been a concern, but there is no clear evidence for the efficacy of psychotherapy for infertile couples. This study aimed to summarize the current evidence of the effects of psychotherapy on psychological and pregnancy outcomes for infertile couples. Method: We searched Ovid MEDLINE, Ovid EMbase, The Cochrane Library, and Web of Science for articles published from 1946 to June 26, 2020. The pregnancy outcomes, psychological outcomes, and acceptability were involved (...) in the study. Results: Overall, 29 studies with a combined total of 3,522 adult participants were included in the meta-analysis. Compared with a placebo, psychotherapy was associated with the pregnancy rate [risk ratio = 1.43, 95% CI [1.07, 1.93]], total psychological scales associated with infertility [standardized mean difference = −0.33 95% CI [−0.63, −0.02]], subsymptoms of psychological scores using the 28-item version of GHQ, and the integrative body–mind–spirit ; total psychological scales associated with infertility using other treatments and more than a month; and anxiety using BMS had significant statistical significance. The funnel plots of all outcomes were approximately symmetrical, and no significant publication bias was found. Conclusions: The study showed that psychotherapy can lead to improvements in the pregnancy rate for infertile patients, especially for patients receiving assisted fertility. In addition, it may help improve total psychological scales associated with infertility and depression. CBT and BMS play an important role in improving rate of pregnancy, and BMS is associated with reducing anxiety. Although psychological interventions had limited effects on the pregnancy outcomes of infertility, our study still recommended that psychotherapies, in particular CBT and BMS, were applied to the therapeutic regimen for infertility, especially for patients receiving assisted fertility. (shrink)

It has long been known that scientists have a tendency to conduct experiments in a way that brings about the expected outcome. Here, we provide the first direct demonstration of this type of experimenter bias in experimental philosophy. Opposed to previously discovered types of experimenter bias mediated by face-to-face interactions between experimenters and participants, here we show that experimenters also have a tendency to create stimuli in a way that brings about expected outcomes. We randomly assigned undergraduate (...) experimenters to receive two different hypotheses about folk intuitions of consciousness, and then asked them to design experiments based on their hypothesis. Specifically, experimenters generated sentences ascribing intentional and phenomenal mental states to groups, which were later rated by online participants for naturalness. We found a significant interaction between experimenter hypothesis and participant ratings indicating a general tendency for experimenters to obtain the result that they expected. These results indicate that experimenter bias is a real problem in experimental philosophy since the methods and design employed here mirror the predominant survey methods of the field as a whole. The bearing of the current results on Knobe and Prinz’s :67–83, 2008) group mind hypothesis is discussed, and new methods for avoiding experimenter bias are proposed. (shrink)