Explore the vast range of titles available.

In this paper, we review algorithmic bias in education, discussing the causes of that bias and reviewing the empirical literature on the specific ways that algorithmic bias is known to have manifested in education. While other recent work has reviewed mathematical definitions of fairness and expanded algorithmic approaches to reducing bias, our review focuses instead on solidifying the current understanding of the concrete impacts of algorithmic bias in education—which groups are known to be impacted and which stages and agents in the development and deployment of educational algorithms are implicated. We discuss theoretical and formal perspectives on algorithmic bias, connect those perspectives to the machine learning pipeline, and review metrics for assessing bias. Next, we review the evidence around algorithmic bias in education, beginning with the most heavily-studied categories of race/ethnicity, gender, and nationality, and moving to the available evidence of bias for less-studied categories, such as socioeconomic status, disability, and military-connected status. Acknowledging the gaps in what has been studied, we propose a framework for moving from unknown bias to known bias and from fairness to equity. We discuss obstacles to addressing these challenges and propose four areas of effort for mitigating and resolving the problems of algorithmic bias in AIED systems and other educational technology.

Funnel plots, and tests for funnel plot asymmetry, have been widely used to examine bias in the results of meta-analyses. Funnel plot asymmetry should not be equated with publication bias, because it has a number of other possible causes. This article describes how to interpret funnel plot asymmetry, recommends appropriate tests, and explains the implications for choice of meta-analysis model

Researchers often lack the necessary data to credibly estimate racial discrimination in policing. In particular, police administrative records lack information on civilians police observe but do not investigate. In this article, we show that if police racially discriminate when choosing whom to investigate, analyses using administrative records to estimate racial discrimination in police behavior are statistically biased, and many quantities of interest are unidentified—even among investigated individuals—absent strong and untestable assumptions. Using principal stratification in a causal mediation framework, we derive the exact form of the statistical bias that results from traditional estimation. We develop a bias-correction procedure and nonparametric sharp bounds for race effects, replicate published findings, and show the traditional estimator can severely underestimate levels of racially biased policing or mask discrimination entirely. We conclude by outlining a general and feasible design for future studies that is robust to this inferential snare.

Flaws in the design, conduct, analysis, and reporting of randomised trials can cause the effect of an intervention to be underestimated or overestimated. The Cochrane Collaboration’s tool for assessing risk of bias aims to make the process clearer and more accurate

Publication bias remains a controversial issue in psychological science. The tendency of psychological science to avoid publishing null results produces a situation that limits the replicability assumption of science, as replication cannot be meaningful without the potential acknowledgment of failed replications. We argue that the field often constructs arguments to block the publication and interpretation of null results and that null results may be further extinguished through questionable researcher practices. Given that science is dependent on the process of falsification, we argue that these problems reduce psychological science's capability to have a proper mechanism for theory falsification, thus resulting in the promulgation of numerous \"undead\" theories that are ideologically popular but have little basis in fact.

We study the out-of-sample and post-publication return predictability of 97 variables shown to predict cross-sectional stock returns. Portfolio returns are 26% lower out-of-sample and 58% lower post-publication. The out-of-sample decline is an upper bound estimate of data mining effects. We estimate a 32% (58%-26%) lower return from publication-informed trading. Post-publication declines are greater for predictors with higher in-sample returns, and returns are higher for portfolios concentrated in stocks with high idiosyncratic risk and low liquidity. Predictor portfolios exhibit post-publication increases in correlations with other published-predictor portfolios. Our findings suggest that investors learn about mispricing from academic publications.

David Moher and colleagues introduce PRISMA, an update of the QUOROM guidelines for reporting systematic reviews and meta-analyses

Global climate model (GCM) output typically needs to be bias corrected before it can be used for climate change impact studies. Three existing bias correction methods, and a new one developed here, are applied to daily maximum temperature and precipitation from 21 GCMs to investigate how different methods alter the climate change signal of the GCM. The quantile mapping (QM) and cumulative distribution function transform (CDF-t) bias correction methods can significantly alter the GCM’s mean climate change signal, with differences of up to 2°C and 30% points for monthly mean temperature and precipitation, respectively. Equidistant quantile matching (EDCDFm) bias correction preserves GCM changes in mean daily maximum temperature but not precipitation. An extension to EDCDFm termed PresRat is introduced, which generally preserves the GCM changes in mean precipitation. Another problem is that GCMs can have difficulty simulating variance as a function of frequency. To address this, a frequency-dependent bias correction method is introduced that is twice as effective as standard bias correction in reducing errors in the models’ simulation of variance as a function of frequency, and it does so without making any locations worse, unlike standard bias correction. Last, a preconditioning technique is introduced that improves the simulation of the annual cycle while still allowing the bias correction to take account of an entire season’s values at once.

Influential models of visual working memory treat each item to be stored as an independent unit and assume that there are no interactions between items. However, real-world displays have structure that provides higher-order constraints on the items to be remembered. Even in the case of a display of simple colored circles, observers can compute statistics, such as mean circle size, to obtain an overall summary of the display. We examined the influence of such an ensemble statistic on visual working memory. We report evidence that the remembered size of each individual item in a display is biased toward the mean size of the set of items in the same color and the mean size of all items in the display. This suggests that visual working memory is constructive, encoding displays at multiple levels of abstraction and integrating across these levels, rather than maintaining a veridical representation of each item independently.

Although you might not think it, it's hard to catch HIV. Less than 1% of unprotected sexual exposures result in infection. What then leads to transmission? Carlson et al. determined the amino acid sequence of viruses infecting 137 Zambian heterosexual couples in which one partner infected the other (see the Perspective by Joseph and Swanstrom). The authors then used statistical modeling and found that transmitted viruses are typically the most evolutionarily fit. That is, compared to other viral variants in the infected person, the transmitted virus most closely matches the most common viral sequence found in the Zambian population. Science , this issue 10.1126/science.1254031 ; see also p. 136 An analysis of discordant couples reveals that transmitted HIV-1 viruses are typically the most evolutionarily fit. [Also see Perspective by Joseph and Swanstrom ] Heterosexual transmission of HIV-1 typically results in one genetic variant establishing systemic infection. We compared, for 137 linked transmission pairs, the amino acid sequences encoded by non-envelope genes of viruses in both partners and demonstrate a selection bias for transmission of residues that are predicted to confer increased in vivo fitness on viruses in the newly infected, immunologically naïve recipient. Although tempered by transmission risk factors, such as donor viral load, genital inflammation, and recipient gender, this selection bias provides an overall transmission advantage for viral quasispecies that are dominated by viruses with high in vivo fitness. Thus, preventative or therapeutic approaches that even marginally reduce viral fitness may lower the overall transmission rates and offer long-term benefits even upon successful transmission.