No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
LinkedOut? A Field Experiment on Discrimination in Job Network Formation
Abstract
We assess the impact of discrimination on Black individuals’ job networks across the U.S. using a two-stage field experiment with 400+ fictitious LinkedIn profiles. In the first stage, we vary race via AI-generated images only and find that Black profiles’ connection requests are 13 percent less likely to be accepted. Based on users’ CVs, we find widespread discrimination across social groups. In the second stage, we exogenously endow Black and White profiles with the same networks and ask connected users for career advice. We find no evidence of direct discrimination in information provision. However, when taking into account differences in the composition and size of networks, Black profiles receive substantially fewer replies. Our findings suggest that gatekeeping is a key driver of Black-White disparities.
ResearchGate has not been able to resolve any citations for this publication.
Low levels of social interaction across class lines have generated widespread concern 1–4 and are associated with worse outcomes, such as lower rates of upward income mobility 4–7 . Here we analyse the determinants of cross-class interaction using data from Facebook, building on the analysis in our companion paper ⁷ . We show that about half of the social disconnection across socioeconomic lines—measured as the difference in the share of high-socioeconomic status (SES) friends between people with low and high SES—is explained by differences in exposure to people with high SES in groups such as schools and religious organizations. The other half is explained by friending bias—the tendency for people with low SES to befriend people with high SES at lower rates even conditional on exposure. Friending bias is shaped by the structure of the groups in which people interact. For example, friending bias is higher in larger and more diverse groups and lower in religious organizations than in schools and workplaces. Distinguishing exposure from friending bias is helpful for identifying interventions to increase cross-SES friendships (economic connectedness). Using fluctuations in the share of students with high SES across high school cohorts, we show that increases in high-SES exposure lead low-SES people to form more friendships with high-SES people in schools that exhibit low levels of friending bias. Thus, socioeconomic integration can increase economic connectedness in communities in which friending bias is low. By contrast, when friending bias is high, increasing cross-SES interactions among existing members may be necessary to increase economic connectedness. To support such efforts, we release privacy-protected statistics on economic connectedness, exposure and friending bias for each ZIP (postal) code, high school and college in the United States at https://www.socialcapital.org .
Social capital—the strength of an individual’s social network and community—has been identified as a potential determinant of outcomes ranging from education to health1–8. However, efforts to understand what types of social capital matter for these outcomes have been hindered by a lack of social network data. Here, in the first of a pair of papers9, we use data on 21 billion friendships from Facebook to study social capital. We measure and analyse three types of social capital by ZIP (postal) code in the United States: (1) connectedness between different types of people, such as those with low versus high socioeconomic status (SES); (2) social cohesion, such as the extent of cliques in friendship networks; and (3) civic engagement, such as rates of volunteering. These measures vary substantially across areas, but are not highly correlated with each other. We demonstrate the importance of distinguishing these forms of social capital by analysing their associations with economic mobility across areas. The share of high-SES friends among individuals with low SES—which we term economic connectedness—is among the strongest predictors of upward income mobility identified to date10,11. Other social capital measures are not strongly associated with economic mobility. If children with low-SES parents were to grow up in counties with economic connectedness comparable to that of the average child with high-SES parents, their incomes in adulthood would increase by 20% on average. Differences in economic connectedness can explain well-known relationships between upward income mobility and racial segregation, poverty rates, and inequality12–14. To support further research and policy interventions, we publicly release privacy-protected statistics on social capital by ZIP code at https://www.socialcapital.org. Analyses of data on 21 billion friendships from Facebook in the United States reveal associations between social capital and economic mobility.
This paper surveys the literature on theories of discrimination, focusing mainly on new contributions. Recent theories expand on the traditional taste-based and statistical discrimination frameworks by considering specific features of learning and signaling environments, often using novel information- and mechanism-design language; analyzing learning and decision making by algorithms; and introducing agents with behavioral biases and misspecified beliefs. This survey also attempts to narrow the gap between the economic perspective on "theories of discrimination" and the broader study of discrimination in the social science literature. In that respect, I first contribute by identifying a class of models of discriminatory institutions, made up of theories of discriminatory social norms and discriminatory institutional design. Second, I discuss the classification of discrimination as direct or systemic, and compare it to previous notions of discrimination in the economic literature.
Artificial intelligence (AI)–synthesized text, audio, image, and video are being weaponized for the purposes of nonconsensual intimate imagery, financial fraud, and disinformation campaigns. Our evaluation of the photorealism of AI-synthesized faces indicates that synthesis engines have passed through the uncanny valley and are capable of creating faces that are indistinguishable—and more trustworthy—than real faces.
Comparing levels of discrimination across countries can provide a window into large-scalesocial and political factors often described as the root of discrimination. Because of difficulties inmeasurement, however, little is established about variation in hiring discrimination across countries.We address this gap through a formal meta-analysis of 97 field experiments of discriminationincorporating more than 200,000 job applications in nine countries in Europe and North America. Wefind significant discrimination against nonwhite natives in all countries in our analysis; discriminationagainst white immigrants is present but low. However, discrimination rates vary strongly by country:In high-discrimination countries, white natives receive nearly twice the callbacks of nonwhites; inlow-discrimination countries, white natives receive about 25 percent more. France has the highestdiscrimination rates, followed by Sweden. We find smaller differences among Great Britain, Canada,Belgium, the Netherlands, Norway, the United States, and Germany. These findings challenge severalconventional macro-level theories of discrimination.
This paper studies how exclusive social groups shape upward mobility and whether interactions between low- and high-status peers can integrate the top rungs of the economic and social ladders. Our setting is Harvard in the 1920s and 1930s, where new groups of students arriving on campus encountered a social system centered on exclusive old boys’ clubs. Combining archival and Census records, we first show that students from prestigious private feeder schools are overrepresented in old boys’ clubs, while academic high achievers and ethnic minorities are almost completely absent. Club members earn 30% more than other students and are more likely to work in finance and join country clubs, both characteristic of the era’s elite. We then use random variation in room assignment to show that exposure to high-status peers expands gaps in college club membership, adult social club membership, and finance careers by high school type, with large positive effects for private school students and zero or negative effects for others. To conclude, we turn to more recent cohorts. We show that the link between exclusive college clubs and finance careers persists across the 20th century even as Harvard diversifies, and that elite university students from the highest-income families continue to outearn their peers.
We investigate whether personal information posted by job candidates on social media sites is sought and used by prospective U.S. employers. We create profiles for job candidates on popular social networks, manipulating information protected under U.S. laws, and submit job applications on their behalf to more than 4,000 employers. We estimate employer search activity and bias in interview callbacks. We find evidence of employers searching online for the candidates. At the national level, we find no significant difference in the callback rates for a Muslim versus a Christian candidate, or for a gay versus a straight candidate. However, employers in Republican areas exhibit significant bias against the Muslim candidate relative to the Christian candidate. This bias is significantly larger than the bias in Democratic areas. The results on callback bias are robust to using state- and county-level data, to controlling for firm, job, and geographical characteristics, to including additional interaction effects in the empirical specification, and to several estimation strategies. The results suggest that the online disclosure of certain personal traits can influence the hiring decisions of some U.S. employers, but the likelihood of hiring discrimination via online searches varies across employers.
This paper was accepted by Chris Forman, information systems.
Online correspondence audit studies have emerged as the primary method to examine racial discrimination. Although audits use distinctive names to signal race, few studies scientifically examine data regarding the perception of race from names. Different names treated as black or white may be perceived in heterogeneous ways. I conduct a survey experiment that asks respondents to identify the race they associate with a series of names. I alter the first names given to each respondent and inclusion of last names. Names more commonly given by highly educated black mothers (e.g., Jalen and Nia) are less likely to be perceived as black than names given by less educated black mothers (e.g., DaShawn and Tanisha). The results suggest that a large body of social science evidence on racial discrimination operates under a misguided assumption that all black names are alike, and the findings from correspondence audits are likely sensitive to name selection.
People find jobs through their social networks using ties of different strengths. Intuitively weak ties might be less useful because people communicate less often with them, or more useful because they provide novel information. Granovetter's early work showed that more job-seekers get help via acquaintances than friends (Granovetter, 1973). However, recent work on job-finding (Gee et al., 2017) shows an apparent paradox of weak ties in the United States: most people are helped through one of their numerous weak ties, but a single stronger tie is significantly more valuable at the margin. Although some studies have addressed the importance of weak ties in job finding within specific countries, this is the first paper to use a single dataset and methodology to compare the importance of weak ties across countries. Here, we use de-identified data from almost 17 million social ties in 55 countries to document the widespread existence of this paradox of weak ties across many societies. More people get jobs where their weak ties work. However, this is not because weak ties are more helpful than strong ties – it is because they are more numerous. In every country, the likelihood of going to work where an individual friend works is increasing – not decreasing – with tie strength. Yet, there is substantial variation in the added value of a strong tie at the margin across these countries. We show that the level of income inequality in a country is positively correlated with the added value of a strong tie, so that individual strong ties matter more when there is greater income inequality.
Social networks are important for finding jobs, but which ties are most useful? Granovetter (1973) suggested that ``weak ties'' are more valuable than ``strong ties,'' since strong ties have redundant information, while weak ties have new information. Using six million Facebook users' data we find evidence for the opposite. We proxy for job help by identifying people who eventually work with a pre-existing friend. Using objective tie strength measures and our job help proxy, we find that most people are helped through one of their numerous weak ties, but a single stronger tie is significantly more valuable at the margin.
Cooperation is central to human societies. Yet relatively little is known about the cognitive underpinnings of cooperative decision making. Does cooperation require deliberate self-restraint? Or is spontaneous prosociality reined in by calculating self-interest? Here we present a theory of why (and for whom) intuition favors cooperation: cooperation is typically advantageous in everyday life, leading to the formation of generalized cooperative intuitions. Deliberation, by contrast, adjusts behaviour towards the optimum for a given situation. Thus, in one-shot anonymous interactions where selfishness is optimal, intuitive responses tend to be more cooperative than deliberative responses. We test this 'social heuristics hypothesis' by aggregating across every cooperation experiment using time pressure that we conducted over a 2-year period (15 studies and 6,910 decisions), as well as performing a novel time pressure experiment. Doing so demonstrates a positive average effect of time pressure on cooperation. We also find substantial variation in this effect, and show that this variation is partly explained by previous experience with one-shot lab experiments.
We examine the trade-offs associated with using Amazon.com's Mechanical Turk (MTurk) interface for subject recruitment. We first describe MTurk and its promise as a vehicle for performing low-cost and easy-to-field experiments. We then assess the internal and external validity of experiments performed using MTurk, employing a framework that can be used to evaluate other subject pools. We first investigate the characteristics of samples drawn from the MTurk population. We show that respondents recruited in this manner are often more representative of the U.S. population than in-person convenience samples—the modal sample in published experimental political science—but less representative than subjects in Internet-based panels or national probability samples. Finally, we replicate important published experimental work using MTurk samples.
We develop empirical tools for studying discrimination in multiphase systems and apply them to the setting of foster care placement by child protective services. Leveraging the quasi-random assignment of two sets of decision-makers—initial hotline call screeners and subsequent investigators—we study how unwarranted racial disparities arise and propagate through this system. Using a sample of over 200,000 maltreatment allegations, we find that calls involving Black children are 55% more likely to result in foster care placement than calls involving white children with the same potential for future maltreatment in the home. Call screeners account for up to 19% of this unwarranted disparity, with the remainder due to investigators. Unwarranted disparity is concentrated in cases with potential for future maltreatment, suggesting that white children may be harmed by “underplacement” in high-risk situations.
Offices are social places. Employees and managers take breaks together and talk about family and hobbies. In this study, we show that employees’ social interactions with their managers can be advantageous for their careers, and that this phenomenon contributes to the gender pay gap. We use administrative and survey data from a large financial institution and exploit quasi-random variation induced by the rotation of managers. We provide evidence that when employees have more face-to-face interactions with their managers, they are promoted at a higher rate. This mechanism could explain a third of the gender gap in promotions at this firm. (JEL G21, J16, J31, J71, M12, M51, Z13)
We review the recent literature on rational inattention, identify the main theoretical mechanisms, and explain how it helps us understand a variety of phenomena across fields of economics. The theory of rational inattention assumes that agents cannot process all available information, but they can choose which exact pieces of information to attend to. Several important results in economics have been built around imperfect information. Nowadays, many more forms of information than ever before are available due to new technologies, and yet we are able to digest little of it. Which form of imperfect information we possess and act upon is thus largely determined by which information we choose to pay attention to. These choices are driven by current economic conditions and imply behavior that features numerous empirically supported departures from standard models. Combining these insights about human limitations with the optimizing approach of neoclassical economics yields a new, generally applicable model. (JEL D83, D91, E71)
The authors analyzed data from multiple large-scale randomized experiments on LinkedIn’s People You May Know algorithm, which recommends new connections to LinkedIn members, to test the extent to which weak ties increased job mobility in the world’s largest professional social network. The experiments randomly varied the prevalence of weak ties in the networks of over 20 million people over a 5-year period, during which 2 billion new ties and 600,000 new jobs were created. The results provided experimental causal evidence supporting the strength of weak ties and suggested three revisions to the theory. First, the strength of weak ties was nonlinear. Statistical analysis found an inverted U-shaped relationship between tie strength and job transmission such that weaker ties increased job transmission but only to a point, after which there were diminishing marginal returns to tie weakness. Second, weak ties measured by interaction intensity and the number of mutual connections displayed varying effects. Moderately weak ties (measured by mutual connections) and the weakest ties (measured by interaction intensity) created the most job mobility. Third, the strength of weak ties varied by industry. Whereas weak ties increased job mobility in more digital industries, strong ties increased job mobility in less digital industries.
We develop new quasi-experimental tools to measure disparate impact, regardless of its source, in the context of bail decisions. We show that omitted variables bias in pretrial release rate comparisons can be purged by using the quasi-random assignment of judges to estimate average pretrial misconduct risk by race. We find that two-thirds of the release rate disparity between White and Black defendants in New York City is due to the disparate impact of release decisions. We then develop a hierarchical marginal treatment effect model to study the drivers of disparate impact, finding evidence of both racial bias and statistical discrimination. (JEL J15, K42)
We study the results of a massive nationwide correspondence experiment sending more than 83,000 fictitious applications with randomized characteristics to geographically dispersed jobs posted by 108 of the largest U.S. employers. Distinctively Black names reduce the probability of employer contact by 2.1 percentage points relative to distinctively white names. The magnitude of this racial gap in contact rates differs substantially across firms, exhibiting a between-company standard deviation of 1.9 percentage points. Despite an insignificant average gap in contact rates between male and female applicants, we find a between-company standard deviation in gender contact gaps of 2.7 percentage points, revealing that some firms favor male applicants and others favor women. Company-specific racial contact gaps are temporally and spatially persistent, and negatively correlated with firm profitability, federal contractor status, and a measure of recruiting centralization. Discrimination exhibits little geographical dispersion, but two-digit industry explains roughly half of the cross-firm variation in both racial and gender contact gaps. Contact gaps are highly concentrated in particular companies, with firms in the top quintile of racial discrimination responsible for nearly half of lost contacts to Black applicants in the experiment. Controlling false discovery rates to the 5% level, 23 companies are found to discriminate against Black applicants. Our findings establish that discrimination against distinctively Black names is concentrated among a select set of large employers, many of which can be identified with high confidence using large-scale inference methods.
Significance
Although previous attempts have been made to measure everyday discrimination against African Americans, these approaches have been constrained by distinct methodological challenges. We present the results from an audit or correspondence study of a large-scale, nationally representative pool of the American public. We provide evidence that in simple day-to-day interactions, such as sending and responding to emails, the public discriminates against Black people. This discrimination is present among all racial/ethnic groups (aside from among Black people) and all areas of the country. Our results provide a window into the discrimination that Black people in the United States face in day-to-day interactions with their fellow citizens.
Gender differences in professional networks have been shown to contribute to men and women's disparate labor market outcomes. This gap could be due to differences in network access, differences in network usage, or both. Using novel administrative data from a student-alumni professional networking website, we study gender differences in student network usage, holding network access fixed. Focusing on messages sent by students to alumni, we document that male and female students network similarly, in terms of both the number of messages sent and the specific questions asked. Furthermore, there are only small gender differences in question tone.
Although there is ample evidence of discrimination against women in the workplace, it can be difficult to understand what factors contribute to discriminatory behavior. We use an experiment to both document discrimination and unpack its sources. First, we show that, on average, employers prefer to hire male over female workers for male-typed tasks, even when the two workers have identical résumés. Second, and most critically, we use a control condition to identify that this discrimination is not specific to gender. Employers are simply less willing to hire a worker from a group that performs worse on average, even when this group is, instead, defined by a nonstereotypical characteristic. In this way, beliefs about average group differences are the key driver of discrimination against women in our setting. We also document some evidence for in-group preferences that contribute to the gender discrimination observed. Finally, our design allows us to understand and quantify the extent to which image concerns mitigate discriminatory behavior.
This paper was accepted by Yan Chen, decision analysis.
We model the dynamics of discrimination and show how its evolution can identify the underlying source. We test these theoretical predictions in a field experiment on a large online platform where users post content that is evaluated by other users on the platform. We assign posts to accounts that exogenously vary by gender and evaluation histories. With no prior evaluations, women face significant discrimination. However, following a sequence of positive evaluations, the direction of discrimination reverses: women’s posts are favored over men’s. Interpreting these results through the lens of our model, this dynamic reversal implies discrimination driven by biased beliefs. (JEL C93, D83, J16, J71)
We present data from a nationally representative 2017 survey of American adults. For heterosexual couples in the United States, meeting online has become the most popular way couples meet, eclipsing meeting through friends for the first time around 2013. Moreover, among the couples who meet online, the proportion who have met through the mediation of third persons has declined over time. We find that Internet meeting is displacing the roles that family and friends once played in bringing couples together.
We conduct laboratory experiments that explore how gender stereotypes shape beliefs about ability of oneself and others in different categories of knowledge. The data reveal two patterns. First, men’s and women’s beliefs about both oneself and others exceed observed ability on average, particularly in difficult tasks. Second, overestimation of ability by both men and women varies across categories. To understand these patterns, we develop a model that separates gender stereotypes from misestimation of ability related to the difficulty of the task. We find that stereotypes contribute to gender gaps in self-confidence, assessments of others, and behavior in a cooperative game. (JEL C92, D83, D91, J16).
We propose generalized random forests, a method for nonparametric statistical estimation based on random forests (Breiman [Mach. Learn. 45 (2001) 5–32]) that can be used to fit any quantity of interest identified as the solution to a set of local moment equations. Following the literature on local maximum likelihood estimation, our method considers a weighted set of nearby training examples; however, instead of using classical kernel weighting functions that are prone to a strong curse of dimensionality, we use an adaptive weighting function derived from a forest designed to express heterogeneity in the specified quantity of interest. We propose a flexible, computationally efficient algorithm for growing generalized random forests, develop a large sample theory for our method showing that our estimates are consistent and asymptotically Gaussian and provide an estimator for their asymptotic variance that enables valid confidence intervals. We use our approach to develop new methods for three statistical tasks: nonparametric quantile regression, conditional average partial effect estimation and heterogeneous treatment effect estimation via instrumental variables. A software implementation, grf for R and C++, is available from CRAN.
Many scientific and engineering challenges---ranging from personalized
medicine to customized marketing recommendations---require an understanding of
treatment effect heterogeneity. In this paper, we develop a non-parametric
causal forest for estimating heterogeneous treatment effects that extends
Breiman's widely used random forest algorithm. Given a potential outcomes
framework with unconfoundedness, we show that causal forests are pointwise
consistent for the true treatment effect, and have an asymptotically Gaussian
and centered sampling distribution. We also discuss a practical method for
constructing asymptotic confidence intervals for the true treatment effect that
are centered at the causal forest estimates. Our theoretical results rely on a
generic Gaussian theory for a large family of random forest algorithms that, to
our knowledge, is the first set of results that allows any type of random
forest, including classification and regression forests, to be used for
provably valid statistical inference. In experiments, we find causal forests to
be substantially more powerful than classical methods based on nearest-neighbor
matching, especially as the number of covariates increases.
We present new evidence on the evolution of black-white earnings differences among all men, including both workers and nonworkers. We study two measures: (i) the level earnings gap-the racial earnings difference at a given quantile; and (ii) the earnings rank gap-the difference between a black man's percentile in the black earnings distribution and the position he would hold in the white earnings distribution. After narrowing from 1940 to themid-1970s, the median black-white level earnings gap has since grown as large as it was in 1950. At the same time, the median black man's relative position in the earnings distribution has remained essentially constant since 1940, so that the improvement then worsening of median relative earnings have come mainly from the stretching and narrowing of the overall earnings distribution. Black men at higher percentiles have experienced significant advances in relative earnings since 1940, due mainly to strong positional gains among those with college educations. Large relative schooling gains by blacks at the median and below have been more than counteracted by rising return to skill in the labor market, which has increasingly penalized remaining racial differences in schooling at the bottom of the distribution. © The Author(s) 2018. Published by Oxford University Press on behalf of President and Fellows of Harvard College. All rights reserved.
"Ban the Box" (BTB) policies restrict employers from asking about applicants' criminal histories on job applications and are often presented as a means of reducing unemployment among black men, who disproportionately have criminal records. However, withholding information about criminal records could risk encouraging racial discrimination: employers may make assumptions about criminality based on the applicant's race. To investigate BTB's effects, we sent approximately 15,000 online job applications on behalf of fictitious young, male applicants to employers in New Jersey and New York City before and after the adoption of BTB policies. These applications varied whether the applicant had a distinctly black or distinctly white name and the felony conviction status of the applicant. We confirm that criminal records are a major barrier to employment: employers that asked about criminal records were 63% more likely to call applicants with no record. However, our results support the concern that BTB policies encourage racial discrimination: the black-white gap in callbacks grew dramatically at companies that removed the box after the policy went into effect. Before BTB, white applicants to employers with the box received 7% more callbacks than similar black applicants, but BTB increased this gap to 43%. We believe that the best interpretation of these results is that employers are relying on exaggerated impressions of real-world racial differences in felony conviction rates.
We propose strategies to estimate and make inference on key features of heterogeneous effects in randomized experiments. These key features include best linear predictors of the effects using machine learning proxies, average effects sorted by impact groups, and average characteristics of most and least impacted units. The approach is valid in high dimensional settings, where the effects are proxied by machine learning methods. We post-process these proxies into the estimates of the key features. Our approach is agnostic about the properties of the machine learning estimators used to produce proxies, and it completely avoids making any strong assumption. Estimation and inference relies on repeated data splitting to avoid overfitting and achieve validity. Our variational inference method is shown to be uniformly valid and quantifies the uncertainty coming from both parameter estimation and data splitting. In essence, we take medians of p-values and medians of confidence intervals, resulting from many different data splits, and then adjust their nominal level to guarantee uniform validity. The inference method could be of substantial independent interest in many machine learning applications. Empirical applications illustrate the use of the approach.
We examine a vast, interdisciplinary, and increasingly global literature concerning skin color and colorism, which are related to status throughout the world. The vast majority of research has investigated Western societies, where color and colorism have been closely related to race and racism. In Latin America, the two sets of concepts have particularly overlapped. In the rest of the world, particularly in Asia, color and colorism have also been important but have evolved separately from the relatively new concepts of race and racism. In recent years, however, color consciousness and white supremacy appear to have been increasingly united, globalized, and commodified, as exemplified by the global multibillion-dollar skin lightening industry. Finally, we document the growing methodological attention to measurements of skin color and social science data that incorporate skin color measures. Expected final online publication date for the Annual Review of Sociology Volume 43 is July 30, 2017. Please see http://www.annualreviews.org/page/journal/pubdates for revised estimates.
In an experiment on Airbnb, we find that applications from guests with distinctively African American names are 16 percent less likely to be accepted relative to identical guests with distinctively white names. Discrimination occurs among landlords of all sizes, including small landlords sharing the property and larger landlords with multiple properties. It is most pronounced among hosts who have never had an African American guest, suggesting only a subset of hosts discriminate. While rental markets have achieved significant reductions in discrimination in recent decades, our results suggest that Airbnb's current design choices facilitate discrimination and raise the possibility of erasing some of these civil rights gains.
Referred workers are more likely than nonreferred workers to be hired, all else equal. In three field experiments in an online labor market, we examine why. We find that referrals contain positive information about worker performance and persistence that is not contained in workers’ observable characteristics. We also find that referrals perform particularly wellwhen working directlywith their referrers.However, we do not find evidence that referrals exert more effort because they believe their performance will affect their relationship with their referrer or their referrer’s position at the firm.
We integrate tools to monitor information acquisition in field experiments on discrimination and examine whether gaps arise already when decision makers choose the effort level for reading an application. In both countries we study, negatively stereotyped minority names reduce employers' effort to inspect resumes. In contrast, minority names increase information acquisition in the rental housing market. Both results are consistent with a model of endogenous allocation of costly attention, which magnifies the role of prior beliefs and preferences beyond the one considered in standard models of discrimination. The findings have implications for magnitude of discrimination, returns to human capital and policy.
Systems as diverse as genetic networks or the World Wide Web are best described as networks with complex topology. A common property of many large networks is that the vertex connectivities follow a scale-free power-law distribution. This feature was found to be a consequence of two generic mech-anisms: (i) networks expand continuously by the addition of new vertices, and (ii) new vertices attach preferentially to sites that are already well connected. A model based on these two ingredients reproduces the observed stationary scale-free distributions, which indicates that the development of large networks is governed by robust self-organizing phenomena that go beyond the particulars of the individual systems.
The publication of Daniel Kahneman's book, Thinking, Fast and Slow, is a major intellectual event. The book summarizes, but also integrates, the research that Kahneman has done over the past forty years, beginning with his path-breaking work with the late Amos Tversky. The broad theme of this research is that human beings are intuitive thinkers and that human intuition is imperfect, with the result that judgments and choices often deviate substantially from the predictions of normative statistical and economic models. In this review, I discuss some broad ideas and themes of the book, describe some economic applications, and suggest future directions for research that the book points to, especially in decision theory.
In modern face recognition, the conventional pipeline consists of four stages: detect => align => represent => classify. We revisit both the alignment step and the representation step by employing explicit 3D face modeling in order to apply a piecewise affine transformation, and derive a face representation from a nine-layer deep neural network. This deep network involves more than 120 million parameters using several locally connected layers without weight sharing, rather than the standard convolutional layers. Thus we trained it on the largest facial dataset to-date, an identity labeled dataset of four million facial images belonging to more than 4,000 identities, where each identity has an average of over a thousand samples. The learned representations coupling the accurate model-based alignment with the large facial database generalize remarkably well to faces in unconstrained environments, even with a simple classifier. Our method reaches an accuracy of 97.25% on the Labeled Faces in the Wild (LFW) dataset, reducing the error of the current state of the art by more than 25%, closely approaching human-level performance.