Unintentional Bias Research Articles

Quantifying Biases in Online Information Exposure

Our consumption of online information is mediated by filtering, ranking, and recommendation algorithms that introduce unintentional biases as they attempt to deliver relevant and engaging content. It has been suggested that our reliance on online technologies such as search engines and social media may limit exposure to diverse points of view and make us vulnerable to manipulation by disinformation. In this article, we mine a massive data set of web traffic to quantify two kinds of bias: (i) homogeneity bias, which is the tendency to consume content from a narrow set of information sources, and (ii) popularity bias, which is the selective exposure to content from top sites. Our analysis reveals different bias levels across several widely used web platforms. Search exposes users to a diverse set of sources, while social media traffic tends to exhibit high popularity and homogeneity bias. When we focus our analysis on traffic to news sites, we find higher levels of popularity bias, with smaller differences across applications. Overall, our results quantify the extent to which our choices of online systems confine us inside “social bubbles.”

Checked Your Bias Lately? Reasons and Strategies for Rural Teachers to Self-Assess for Grading Bias

Multiple factors influence teachers’ grading and scoring of students’ class work, homework, projects and tests. Put simply, bias in grading is giving different grades on student work of essentially equal quality, based on factors irrelevant to the scope and criteria for that work. Grading is to a degree subjective, but it need not be biased. Intentional bias is a common criticism of teachers and has been the topic of numerous studies and reports. However, less attention has been given to unintentional sources of grading bias. For teachers in any school, bias can creep into grading despite a teacher’s best efforts to be fair and impartial. Teachers in rural schools may face more challenges to preventing and reducing grading bias. This paper identifies some of the issues relevant to teachers’ unintentional grading biases and discusses both conventional and innovative ways to reduce it.

Using left ventricular assist devices in advanced heart failure patients

ABSTRACTAdvanced Heart Failure (AHF) is a complex syndrome that affects the physiology of the heart to maintain efficient blood circulation resulting in multiorgan failure and, eventually, death. Left Ventricular Assist Devices (LVADs) have become the cornerstone therapy for AHF patients, both as a bridge to transplantation and as a decisive therapy. Recently the results of the MOMENTUM 3 Trial were published. The trial compared HeartMate 3 LVAD with HeartMate II LVAD in a randomized trial in The Multicenter Study of MagLev Technology in Patients Undergoing Mechanical Circulatory Support Therapy with HeartMate 3 (MOMENTUM 3). Of 366 patients, 190 were assigned to the centrifugal-flow pump group (HeartMate 3) and 176 to the axial-flow (HeartMate II) pump group. In the intention-to-treat population, the primary end point occurred in 151 patients (79.5%) in the centrifugal-flow pump group, as compared with 106 (60.2%) in the axial-flow pump group (P < 0.001 for noninferiority). Reoperation for pump malfunction was less frequent in the centrifugal-flow pump group than in the axial-flow pump group (P < 0.001).The results of the MOMENTUM 3 Trial are a big achievement in the cardiovascular world. Any improvement in LVADs that reduces the risk of stroke, perhaps the most feared complication of these devices, would be meaningful. Besides, given the observed lower rate of pump thrombosis and reoperation for pump malfunction, it already seems likely that the HeartMate 3 will supplant the HeartMate II in clinical practice. In addition, the risks that are associated with reoperation undoubtedly counterbalanced any unintentional bias in performing that intervention.

Undiscovered pathology of transient scaffolding t1remains a driver of failures in clinical trials.

AIMTo statistically examine the released clinical trials and meta-analyses of polymeric bioresorbable scaffolds resuming the main accomplishments in the field with a translation to the routine clinical practice.METHODSThe statistical power in clinical trials such as ABSORB Japan, ABSORB China, EVERBIO II, AIDA, and few meta-analyses by the post hoc odds ratio-based sample size calculation, and the patterns of artery remodeling published in papers from ABSORB A and B trials were evaluated.RESULTSThe phenomenal admiration from the first ABSORB studies in 2006-2013 was replaced by the tremendous disappointment in 2014-2017 due to reported relatively higher rates of target lesion failure (a mean prevalence of 9.16%) and device thrombosis (2.38%) in randomized controlled trials. Otherwise, bioresorbable vascular scaffold (BVS) performs as well as the metallic drug-eluting stent (DES) with a trend toward some benefits for cardiac mortality [risk ratio (RR), 0.58-0.94, P > 0.05]. The underpowered design was confirmed for some studies such as ABSORB Japan, ABSORB China, EVERBIO II, AIDA trials, and meta-analyses of Polimeni, Collet, and Mahmoud with some unintentional bias (judged by the asymmetrical Funnel plot). Scaffold thrombosis rates with Absorb BRS were comparable with DES performed with a so-called strategy of the BVS implantation with optimized pre-dilation (P), sizing (S) and post-dilation (P) (PSP) implantation (RR, PSP vs no PSP 0.37) achieving 0.35 per 100 patient-years, which is comparable to the RR 0.49 with bare-metal stents and the RR 1.06 with everolimus DES. Both ABSORB II and ABSORB III trials were powered enough for a five-year follow-up, but the results were not entirely conclusive due to the mostly non-significant fashion of data. The powered meta-analyses were built mostly on statistically poor findings.CONCLUSIONThe misunderstanding of the pathology of transient scaffolding drives the failures of the clinical trials. More bench studies of the vascular response are required. Several next-generation BVS including multifunctional electronic scaffold grant cardiology with a huge promise to make BVS technology great again.

Prevalence of nonsuicidal self-injury in chinese middle school and high school students: A meta-analysis.

Recent studies have revealed a high rate of nonsuicidal self-injury (NSSI) behavior in Chinese students, but quantitative syntheses of pooled prevalence are sparse. There have been several NSSI prevalence meta-analyses in other populations. However, given the existence of cultural diversity, racial difference, educational system difference and so on, these results may not be ideal for Chinese populations. Furthermore, the above-mentioned meta-analyses did not include Chinese database which may have led to unintentional bias. Thus, we conducted this meta-analysis to estimate the prevalence of NSSI in Chinese middle-school students.The databases searched included PubMed, Embase, Web of Science, CBM (Chinese database), Chinese National Knowledge Infrastructure (CNKI), Wanfang Data (Chinese database) and the Weipu database (Chinese database). The search terms included: self-injury/self-harm/self-abuse/nonsuicidal self-injury/deliberate self-harm, adolescen∗/youth/teen/students, and China/Chinese. All relevant articles published between January 2000 to November 2017, in either Chinese or English, were included. Two investigators were engaged in this process, and any disagreements were settled by a third investigator. A random effects model was then used to calculate the pooled prevalence.A total of 420 studies with 160,348 participants were retrieved. The pooled prevalence was 22.37% (95% CI: 18.84%–25.70%). Substantial heterogeneity in prevalence estimates was revealed. Subgroup analyses showed that the pooled estimate of prevalence of life time NSSI was 14.5% (95%CI: 0.06%–22.7%), and 6–24 months NSSI was23.3% (95%CI: 20.5%–26.1%). The prevalence for males was 20.6% (95% CI: 16.1%–25.0%), and for females was 21.9% (95% CI: 17.6%–26.2%).The prevalence of NSSI in Chinese middle-school students is relatively high. More attention should be paid to the current situation.

Coordinating judgmental forecasting: Coping with intentional biases

Human judgment, an almost inextricable ingredient in demand forecasting, introduces many unintentional and intentional biases to the forecasting and operations planning process. In the present research, we isolate intentional biases from this process and relate them to heterogeneous departmental roles and incentives. Through a laboratory experiment, which simulates forecasting operations planning in an interdepartmental decision-making context, we examine the effects of departmental roles, incentives and various weighting schemes on forecasting behavior and performance. We find that departmental roles, even without role-specific incentives, entail intentional biases of 8% of the forecast, and that role-specific incentives increase these biases to 14%. We further test the claim that accuracy-weighted schemes can remove biases in forecasting, and conclude that they halve, but don’t fully remove them. Finally, a weighting scheme that explicitly corrects biased inputs shows great promise in reducing intentional and unintentional biases. In our experiment, this scheme reduces biases by 35%. This shows the importance of disentangling intentional and unintentional biases for more effective forecasting adjustments. Our insights have substantial ramifications for the design of the forecasting operations planning process in dynamic business environments determined by high levels of role- and incentive-based heterogeneity.

The Validity of Social Media–Based Career Information

The use of social media expands the availability and sources of career information. However, the authorship of this information has changed from traditional print media and multimedia sources created by experts to social media–based career information created by the users themselves. Although variability in career information validity has been an issue for some time, rapid growth in the use of social media creates some unique challenges. The ease with which social media–based career information can spread creates the potential for rapid widespread dissemination of disinformation and biased perceptions. Potential sources of invalidity include intentional bias (with or without profit motive), unintentional bias, restricted range of experience, out‐of‐date information, popularity bias, similarity bias, and context deficiency. The authors examine potential sources of social media–based career information invalidity and suggest implications for practice to help individuals make the best use of such information.

Cultural cognition and climate change education in the U.S.: why consensus is not enough

Integrating climate change into environmental education programs and formal science classrooms can be difficult, as the issue remains controversial and highly politicized among the American public. This study proposes that the same cultural values that shape worldview differences and divide public opinion on anthropogenic climate change will influence if and how science educators support education on the topic. An online survey with quantitative and qualitative measures was distributed among science educators in the southeastern United States to test what, if any, impact cultural cognition has on their opinions about climate change education. The results suggest that respondents’ cultural values have a significant influence on their intentions to support climate change education and preferences for curricula content. The findings also suggest potential avenues for the field of environmental education to develop climate change materials that may help decrease unintentional biases among science educators and more effectively engage their support in teaching its causes and potential solutions regardless of worldview differences.

Testing an Intervention for Recognizing and Reporting Subtle Gender Bias in Promotion and Tenure Decisions

ABSTRACTWomen make up the majority of doctoral degree earners yet remain underrepresented in tenure-track positions within the academy. Gender disparities result in part from the accumulation of subtle, typically unintentional biases that pervade workplace structures, practices, and patterns of interactions that inadvertently favor men. However, the subtle nature of gender bias makes it difficult to detect and thus diminishes the likelihood of action to address it. We experimentally evaluated the effectiveness of a brief intervention, the Workshop Activity for Gender Equity Simulation in the Academy (WAGES-Academic), which was designed to increase recognition of subtle gender bias in the academic workplace. Participants (N = 177) completed either the WAGES intervention or one of two control conditions and later evaluated promotion and tenure materials of a woman faculty member who received either a blatant sexist, subtle sexist, or nonsexist review. Consistent with hypotheses, WAGES participants (vs. controls) detected more subtle gender bias (ps < .02) and were subsequently more likely to report concerns about bias (ps < .04). Results suggest that low-cost interventions that educate individuals about subtle bias in a nonthreatening way may increase detection and reporting of gender bias in higher education institutions.

The impact of psychopathy and warnings on faking behavior: A multisaturation perspective

Evidence suggests that individuals can and do present themselves positively on personality assessments when motivated to do so. This faking can reduce the validity of personality assessments and is of special concern in high stakes situations where critical decisions are being made at least partially on personality scores (e.g., personnel selection). In the current study, we take a multisaturation perspective of faking, and examine how psychopathy might be related to faking on normal range personality traits measured using single-stimulus or forced-choice personality assessments in a simulated selection context. To examine whether warnings interacted with psychopathy in predicting faking behavior, we included a warning condition. Findings suggest that faking on the single-stimulus personality assessment was more affected by elevated psychopathy such that those higher in psychopathy were more likely to fake than those lower in psychopathy, however psychopathy was also associated with faking behavior on the forced choice measure. This may result in an unintentional bias toward selecting employees with higher psychopathy when evaluating candidates with normal range personality assessments. Warning condition did not play a significant role either as a main effect or in interaction with psychopathy.

ВІЗУАЛІЗАЦІЇ «МАЗЕПИНСЬКОЇ ЛЕГЕНДИ»: ЗМІСТ, ДЖЕРЕЛА, ЕВОЛЮЦІЯ, ТРАДИЦІЙНІ ТА МОДЕРНІЗАЦІЙНІ ІНТЕРПРЕТАЦІЇ

The omnipresence of modern technologies and free access to information granted to general public facilitated demythologization of our history, refutation of ingrained archaic stereotypes and propagation of historical knowledge and better understanding of intellectual and cultural heritage. However, lack of systemic approach to processing of the information found, partial coverage of disparate topics and blatant ignorance of how the source interpretations evolved, repeated public indoctrination with second-hand interpretations (i.e. referring to previous interpretations) without explanation of the original semantics of text or image, as well as pathological graphorrhea of certain authors promoted by facilities and resources of numerous social networks, all result in profanation of the epistemological process. In certain cases, this situation is aggravated by the controversy embedded in the written or visual source itself, as well as by the intentional or unintentional bias against a certain historical figure described or depicted by the source, that is smothered in the current confrontation between different myth-making systems or information war. This article exposes the core and sources of the so-called "Hippolytus legend" and its visualization in the XVII-XIX paintings. It puts under scrutiny when and under what circumstances this work of ancient literature transformed into "the Legend of Mazepa". It also exposes the evolution in ХІХ - ХХІ centures of the new edition of the romantic legend presented by the literature, works of fine arts and decorative and applied arts. Various versions of traditional visual interpretations have been analysed along with the underlying reasoning and causes for the emergence of contemporised visual presentation of the well-known image of Mazepa tied to the horse’s croup. The semantic modifications of so-called distorted images of "modern Mazepas" have also been exposed. The historical figures had been most often branded as "modern Mazepas" have also been put under scrutiny. Such a detailed analysis of both Mazepa's literary and visual image and the peculiarities of the emergence and subsequent interpretations of the "Mazepa's lore" are crucial for understanding of the ancient texts and images that are not always correctly interpreted nowadays, thus spawning numerous new fabrications and myths.

Blinded Visual Scoring of Images Using the Freely-available Software Blinder.

In nearly all subfields of biomedical sciences, there are phenotypes that are currently classified by expert visual scoring. In research applications, these classifications require the experimenter to be blinded to the treatment group in order to avoid unintentional bias in scoring. Currently, many labs either use laborious and tedious methods to manually blind the images, require multiple experimenters to gather and score the data blindly or fail to properly blind the data altogether. In this protocol, we present a simple, freely available software that we created that allows the experimenter to blindly score images. In our protocol, the user loads unblinded images and defines a scoring system. The software then shows the user the images in a random order, allowing the user to select a score from their defined scoring system for each image. Furthermore, the software has an optional "quality control" mechanism where the user will be shown some images multiple times to test the robustness of the visual scoring. Finally, the software summarizes the results in an exportable file that includes unblinded summary data for each group and a full list of images with their scores. In this protocol, we briefly present directions for using the software, potential applications, and caveats/limitations to this approach.

The cohort multiple randomized controlled trial design was found to be highly susceptible to low statistical power and internal validity biases

ObjectivesThe “cohort multiple randomized controlled trial” (cmRCT) is a recent innovation by which novel interventions are trialed within large longitudinal cohorts of patients to gain efficiencies and align trials more closely to standard clinical practice. The use of cmRCTs is outpacing its methodological understanding, and more appropriate methods for designing and analyzing such trials are urgently needed. Study Design and SettingWe established the UK Comprehensive Longitudinal Assessment of Salford Integrated Care cohort of 4,377 patients with long-term conditions within which we are conducting a cmRCT (“Proactive Telephone Coaching and Tailored Support”) of telephone-based health coaching. ResultsWe identify some key methodological challenges to the use of the cmRCT in actual practice. Principal are issues around statistical power, sample size, and treatment effect estimation, for which we provide appropriate methods. Sampling procedures commonly applied in conventional RCTs can result in unintentional selection bias. The fixed data collection points that feature in cmRCTs can also threaten validity. ConclusionThe cmRCT may offer advantages over conventional trial designs. However, a cmRCT requires appropriate power calculation, sampling, and analysis procedures; else, studies may be underpowered or subject to validity biases. We offer solutions to some of the key issues, but further methodological investigations are needed. Cohort multiple RCT–specific Consolidated Standards of Reporting Trials guidance may be indicated.

Reducing Stereotyping When Developing Leaders.

Stereotyping is a human condition. When it leads to identifying contingencies-factors that inhibit or advance leadership capacity-it threatens leaders' abilities and performance. Professional development educators are called to awareness of unintentional bias that can affect the predicaments aligned with stereotyping and foster a learning culture that advances trust and hope. The impact of marginalization, segregation, and underrepresentation of individuals in health care settings is presented. J Contin Educ Nurs. 2017;48(11):492-493.

Nests of red wood ants (Formica rufa-group) are positively associated with tectonic faults: a double-blind test.

Ecological studies often are subjected to unintentional biases, suggesting that improved research designs for hypothesis testing should be used. Double-blind ecological studies are rare but necessary to minimize sampling biases and omission errors, and improve the reliability of research. We used a double-blind design to evaluate associations between nests of red wood ants (Formica rufa, RWA) and the distribution of tectonic faults. We randomly sampled two regions in western Denmark to map the spatial distribution of RWA nests. We then calculated nest proximity to the nearest active tectonic faults. Red wood ant nests were eight times more likely to be found within 60 m of known tectonic faults than were random points in the same region but without nests. This pattern paralleled the directionality of the fault system, with NNE–SSW faults having the strongest associations with RWA nests. The nest locations were collected without knowledge of the spatial distribution of active faults thus we are confident that the results are neither biased nor artefactual. This example highlights the benefits of double-blind designs in reducing sampling biases, testing controversial hypotheses, and increasing the reliability of the conclusions of research.

Linking fields with GMA: Sustainability, companies, people and Operational Research

This article is about making decisions concerning the management of sustainability, decisions that may influence the use or protection of natural resources or address difficult societal choices. Managers have more and more to tackle a diversity of problems in a rigorous and transparent way. One of the distinctive features of these decisions is that managers must give attention to both the values of people affected and factual information concerning the potential consequences of actions. This imposes the adoption of new methods for structuring spaces, strategy alternatives, and organizational planning. The support from operational research analysts becomes increasingly important, as we are dealing with people mostly without strong quantitative or model-building backgrounds. With the presence of different perspectives and mental models, behavior elements are at the core of the problem and unintentional biases in model use may occur. Our intention is to help promote the transference of knowledge to and within companies so that they may assure resilience. We found in general morphological analysis a great help for that. We want to make available a meta-model based on Operational Research for fields involving public resources and multiple interests to aid current and future managers of companies. We conclude the article with two case studies to illustrate our approach.

Abstract IA21: Cancer and the LGBT community

Abstract In response to the clear and persistent trends the elimination of cancer-related health disparities among racial/ethnic minorities has been a national public health priority. For the first time Healthy People 2020, the nation's roadmap for improving the health of Americans, also includes sexual minorities as population of priority. Although precise data on cancer rates among lesbians are lacking, evidence suggests that sexual minority status may contribute to excess risk for the development of certain types of cancers, including breast cancer, anal cancer, lung cancer and cancers associated with HIV/AIDS. The causes of these disparities are complex and likely influenced by the same factors that drive cancer disparity rates among African American and other underserved populations. These factors include poor continuity of care; socioeconomic factors, such as lack of adequate insurance coverage; medical mistrust; cultural and emotional factor; and providers' unintentional bias and inadequate cultural competency. The purpose of this presentation is to discuss risk factors for cancer disparities among LGBT individuals and to describe existing cancer risk reduction interventions targeted toward this community. Citation Format: Alicia K. Matthews. Cancer and the LGBT community. [abstract]. In: Proceedings of the Ninth AACR Conference on the Science of Cancer Health Disparities in Racial/Ethnic Minorities and the Medically Underserved; 2016 Sep 25-28; Fort Lauderdale, FL. Philadelphia (PA): AACR; Cancer Epidemiol Biomarkers Prev 2017;26(2 Suppl):Abstract nr IA21.

The dangers of faulty, biased, or malicious algorithms requires independent oversight

The recent crash of a driverless car sends a clear warning about how algorithms can be deadly (1). Similarly, there are clear dangers in vital national services, such as communications, financial trading, healthcare, and transportation. These services depend on sophisticated algorithms, some relying on unpredictable artificial intelligence techniques, such as deep learning, that are increasingly embedded in complex software systems (2⇓–4). As search algorithms, high-speed trading, medical devices, and autonomous aircraft become more widely implemented, stronger checks become necessary to prevent failures (5, 6). Proper independent oversight and investigation of flawed algorithms can help anticipate and improve quality, hence avoiding failures that lead to disaster. Image courtesy of Shutterstock/joloei. What might help are traditional forms of independent oversight that use knowledgeable people who have powerful tools to anticipate, monitor, and retrospectively review operations of vital national services. The three forms of independent oversight that have been used in the past by industry and governments—planning oversight, continuous monitoring by knowledgeable review boards using advanced software, and a retrospective analysis of disasters—provide guidance for responsible technology leaders and concerned policy makers (7). Considering all three forms of oversight could lead to policies that prevent inadequate designs, biased outcomes, or criminal actions. There is a long history of analyses of how poor design, unintentional bias, and malicious interventions can cause algorithms to trigger huge financial losses, promote unfair decisions, violate laws, and even cause deaths (8). Helen Nissenbaum, a scholar who focuses on the impact of technology on culture and society, identified the sources of bugs and biases in software, complaining about the “systematic erosion … [↵][1]1Email: ben{at}cs.umd.edu. [1]: #xref-corresp-1-1

MISSING DATA IMPUTATION IN THE ELECTRONIC HEALTH RECORD USING DEEPLY LEARNED AUTOENCODERS.

Electronic health records (EHRs) have become a vital source of patient outcome data but the widespread prevalence of missing data presents a major challenge. Different causes of missing data in the EHR data may introduce unintentional bias. Here, we compare the effectiveness of popular multiple imputation strategies with a deeply learned autoencoder using the Pooled Resource Open-Access ALS Clinical Trials Database (PRO-ACT). To evaluate performance, we examined imputation accuracy for known values simulated to be either missing completely at random or missing not at random. We also compared ALS disease progression prediction across different imputation models. Autoencoders showed strong performance for imputation accuracy and contributed to the strongest disease progression predictor. Finally, we show that despite clinical heterogeneity, ALS disease progression appears homogenous with time from onset being the most important predictor.

High prevalence of congenital thrombophilia in patients with pregnancy-related or idiopathic venous thromboembolism/pulmonary embolism.

Congenital thrombophilia which is characterized by deficiencies in proteins such as antithrombin (AT), protein C (PC) and protein S (PS), is a major cause of venous thromboembolism (VTE). A total of 130 patients with VTE were evaluated for congenital thrombophilia based on the activity of AT, PC, or PS. Fifteen VTE patients with congenital AT deficiency (11.5%), 16 with congenital PC deficiency (12.3%) and eight with congenital PS deficiency (6.2%) were diagnosed using DNA analysis. The frequency of congenital AT deficiency was significantly higher in subjects with pregnancy-related and idiopathic VTE than in those with VTE due to other causes, and congenital PC and PS deficiency were frequently associated with idiopathic VTE. Among the groups examined, the plasma levels of AT were the lowest in subjects with pregnancy-related VTE. Although our findings may have been influenced by some unintentional bias, congenital thrombophilia is nevertheless a major cause of VTE in pregnant patients as well as in young or middle-aged patients without any underlying diseases.