Unobservable Dimensions Research Articles

Addressing sample selection bias for machine learning methods

SummaryWe study approaches for adjusting machine learning methods when the training sample differs from the prediction sample on unobserved dimensions. The machine learning literature predominately assumes selection only on observed dimensions. Common approaches are to weight or include variables that influence selection as solutions to selection on observables. Simulation results show that selection on unobservables increases mean squared prediction error using popular machine‐learning algorithms. Common machine learning practices such as weighting or including variables that influence selection into the training or prediction sample often worsen sample selection bias. We propose two control function approaches that remove the effects of selection bias before training and find that they reduce mean‐squared prediction error in simulations. We apply these approaches to predicting the vote share of the incumbent in gubernatorial elections using previously observed re‐election bids. We find that ignoring selection on unobservables leads to substantially higher predicted vote shares for the incumbent than when the control function approach is used.

Information and the trade-off between food safety and food security in rural markets: Experimental evidence from Malawi

We implemented a clustered randomized control trial with experimental auctions among 1,098 rural households in Malawi to evaluate whether providing training about food safety increased demand for safe groundnut. We measured the impact of training on demand for observable and unobservable dimensions of quality, at harvest and in the lean season. We found that the control group who was not trained valued sorted grains (an indicator of observable quality) only. At harvest, both trained and untrained consumers placed statistically equal premiums on food safety labeling (an indicator of otherwise-unobservable quality). However, in the lean season, untrained consumers’ premiums for food safety labeling disappeared, while trained consumers’ premiums for food safety labeling increased.

Factors Associated with Discrepancy of Child-Adolescent/Parent Reported Quality of Life in the Era of COVID-19.

Billions of children/adolescents experienced unprecedented changes in their daily lives that impacted their health-related quality of life (HRQoL) during the first wave of the coronavirus disease 2019. The purpose of this study was to describe child-parent discrepancies in reporting on HRQoL and explore factors associated with such discrepancies at the end of the first lockdown in France. A cross-sectional study was conducted among French school-aged children from 8 to 18 years and their parents living in the Grand Est region in France during the first wave of the epidemic. The impact of individual, self-reported health status and environmental data on discrepant parent-child reports of HRQoL was assessed by multinomial multivariable logistic regression models. A total of 471 parent-child pairs were included. Among 50% of the discordant pairs, parents underestimated HRQoL more frequently than they overestimated it. Home location, social support score, children's education level, parents' education level, tensions and conflicts with neighbors reported by children, whether they had access to a garden, and parents' professional activity were significantly associated with parental overestimation (adjustedOR from 2.08 to 11.61; p < 0.05). Factors associated with parental underestimation were children's education level, SF12 score, home location, the child's gender, parent's level of education, the presence of noise in the residence reported by children, whether a household member was infected with COVID-19, whether they had access to a garden, and family structure (adjustedOR from 1.60 to 4.0; p < 0.05). This study revealed differences between child-reported and parent-reported HRQoL. The COVID-19 pandemic accentuated the discrepancies in observable dimensions and attenuated them in unobservable dimensions of HRQoL but did not impact the directional discrepancy; parents underestimated their child's HRQoL more. These discrepancies appear to be explained by parent and child sociodemographic factors.

Grounded Model of Personal Rivalry and Covert Politics in Field-Identity Formation

Existing literatures by and large portray field-identity formation as an organic, distributed, and democratic meaning-making process—with members across the field negotiating to arrive at a consensus on shared meanings that constitute collective identity. Our inductive study of the formative years (1890-1920) of a century-old academic-professional field revealed however an unobserved dimension of field-identity formation—one marked by personal conflict and collusion to manipulate the field’s identity. Our emergent grounded model, pivoted around an episode of identity contestation in the emerging field, maps how a soured personal relationship and growing rivalry spurred the nascent field’s founder to employ covert politics to mount an overt contestation against his successor in defense of the central feature of the field’s founding identity—ultimately, forcing his successor out of the field and manipulating a consensus over the founding identity. Effectively, we theorize field identity construction as potentially Janus-faced in that it involves actions unfolding, in a complementary and recursive manner, across two levels: an overt level marked by open discourse and debates, to create an illusion of emergence of consensus over the field’s identity; and a covert level marked by personal rivalry and collusion to secure control over the field’s identity. Overall, beyond advancing theory on the hitherto underrecognized psychological and political dimensions of field-identity, our model uncovers the underexplored, dark side of institutional entrepreneurship, and reveals how micro, individual-level interactions inform macro-, field-level processes.

Erik R. Seeman, Speaking with the Dead in Early America

American Religion 2, no. 2 (Spring 2021), pp. 169–174 Copyright © 2021, The Trustees of Indiana University • doi: 10.2979/amerreli.2.2.19 Book Review Erik R. Seeman, Speaking with the Dead in Early America (Philadelphia: University of Pennsylvania Press, 2019) Ava Chamberlain Wright State University, Dayton, USA While reading this fascinating new book I sensed the presence of the dead all around me. Deceased loved ones appeared repeatedly in my dreams. In the family photographs and heirlooms displayed in my home, I recognized my own desire to communicate with the dead. To feel the loss wrought by the coronavirus pandemic, I began keeping a tally of the relentlessly increasing number of COVID-19 deaths. As Erik Seeman perceptively observes, “the living almost always miss the dead, and some mourners will pursue continuing relationships with the deceased” (272). Taking this ordinary human experience as his object of analysis, he surveys early American history from a fresh vantage point that reveals important new dimensions of Protestant lived religion. In Speaking with the Dead in Early America Seeman aims to write “a history of Protestant communication with the dead before the advent of Spiritualism” (5). Having explored African and Native American death ways in two earlier works, he limits the scope of this history to white residents of America’s northern colonies and states. This seemingly conventional focus belies the book’s truly innovative character. By employing new source material and reading well-known sources in new ways, Seeman reveals an unobserved dimension of the lived religion of early American Protestants. American Religion 2:2 170 Seeman finds evidence of communication with the dead in a variety of interesting locations. He approaches the burial grounds scattered throughout New England as a “physical archive” in which the voices of the dead are preserved on “talking gravestones” (73). These voices speak through the epitaphs engraved on the gravestones. Often written in the form of direct address to the viewer, they suggest that the bereaved regularly visited graveyards to hear their loved ones speak and to feel their presence. Similarly, needlework stitched by young women during periods of mourning facilitated communication between the living and the dead. Although commonly written in the third person, these embroidered epitaphs spoke to the viewer by giving “lasting material expression to feelings of grief, loneliness, and resignation to God’s will” (194). Displayed within the home, they provided bereaved family members a tangible connection with their dear departed. Seeman locates evidence of communication with the dead in a wide variety of texts. If you enjoy a good ghost story this book is for you, for it is filled with accounts of “shades,” to use a popular early American term for ghost, appearing at bedsides and in graveyards to comfort the living, to reveal visions of heaven, and to avenge secret crimes, especially murders. Read widely on a popular level, these stories likewise fascinated educated male elites determined to demonstrate the reality of the supernatural using an empirical methodology. Ghosts also made regular appearances in elegiac verse written to commemorate the dead. Seeman observes that the dead frequently speak directly to the living in this popular literary genre, which mourners read at graveside funerals effectively giving voice to the soon-to-be interred corpse. Elegies also permit dead women to speak with an authority often denied the living. Although less commonly the subject of mourning elegies than men, “when [women] did get such attention, nearly half of them spoke from beyond the grave” (67). Seeman emphasizes women’s persistent interest in communication with the dead. As elites increasingly questioned the veracity of supernatural tales appearing in popular print, publishers began catering to a female audience. Stories about guardian angels illustrate this trend. Seeman traces the emergence of the belief that humans become angels in heaven, who in turn return to earth as guardian angels. Women eagerly consumed these stories, especially when the angels were infants returning to comfort their grieving mothers. Women’s letters and journals testify to the sincerity of this belief, thereby revealing that a “dialectical relationship existed between imaginative literature and lived religion” (242). More important, Seeman argues, women’s engagement with the dead “drove important...

Equilibria Under Negligence Liability

AbstractIn many accident contexts, the expected accident harm depends on observable as well as unobservable dimensions of the precaution exercised by the parties involved. The observable dimensions are commonly referred to as the ‘care’ levels and the unobservable aspects as the ‘activity’ levels. In a seminal contribution, Shavell, S (1980). Strict liability versus negligence. J. Leg. Stud. 9: 1–25 extended the scope of the economic analysis of liability rules by providing a model that allows for the care as well as activity level choices. Subsequent works have used and extended Shavell’s model to predict outcomes under various liability rules, and also to compare their efficiency properties. These works make several claims about the existence and efficiency of equilibria under different liability rules, without providing any formal proof. In this paper, we re-examine the prevalent claims in the literature using the standard model itself. Contrary to these prevalent claims, we show that the standard negligence liability rules do not induce equilibrium for all of the accident contexts admissible under the model. Under the standard model, even the ‘no-fault’ rules can fail to induce a Nash equilibrium. In the absence of an equilibrium, it is not plausible to make a claim about the efficiency of a ruleper-seorvis-a-visother rules. We show that even with commonly used utility functions that meet all of the requirements of the standard model, the social welfare function may not have a maximum. In many other situations fully compatible with the standard model, a maximum of the social welfare function is not discoverable by the first order conditions. Under the standard model, even individually optimum choices might not exist. We analyze the underlying problems with the standard model and offer some insights for future research on this subject.

Equilibria under Liability Rules: How the Standard Claims Fall Apart

In many accident contexts, the accident harm depends on observable as well as unobservable dimensions of the precaution exercised by the parties involved. The observable dimensions are commonly referred to as the `care' levels and the unobservable aspects as the `activity' levels. In a seminal contribution, Shavell (1980) extended the scope of economic analysis of liability rules by providing a model that allows for the care as well as activity level choices. Subsequent works have used and extended Shavell's model to predict outcomes under various liability rules and also to compare their efficiency properties. These works make several claims about the existence and efficiency of equilibria under different liability rules, without providing any formal proof. In this paper, we re-examine the prevalent claims in the literature using the standard model itself. Contrary to prevalent claims, we show that the standard negligence liability rules do not induce equilibrium for all the accident contexts admissible under the model. Under the standard model, even the `no-fault' rules can fail to induce a Nash equilibrium. In the absence of an equilibrium, it is not plausible to make a claim about efficiency of a rule per-se or vis-a-vis other rules. We show that even with commonly used utility functions that meet all the requirements of the standard model, the social welfare function may not have a maximum. In many other situations fully compatible with the standard models, a maximum of the social welfare function is not discoverable by the first order conditions. Under the standard models, even individually optimum choices might not exist. We analyze the underlying problems with the standard models and offer some insights for future research on this subject.

Do Preferences for Urban Amenities Really Differ by Skill?

City-level policies often aim at attracting skilled workers by improving urban amenities. However,due to endogeneity problems, studies relying on revealed preferences have difficulties in providingevidence for the basic premise that skilled workers place a higher value on urban amenities thanless skilled individuals. Therefore, we use a stated-preference experiment to directly examinepreferences for urban amenities. In a custom survey, we elicit hypothetical job choices betweentwo cities that differ in wages and a set of urban amenities. We find that amenities are importantdeterminants of city choice, with respondents willing to forgo a significant fraction of their wage tolive in a city with better amenities. Most strikingly, we do not find any preference heterogeneitybetween workers differing by education or creative class membership. Instead, we uncover largeheterogeneities mainly along family-related mobility constraints and unobserved dimensions. Ourresults imply that there is not much scope for amenity-oriented policies to improve the local skillmix. Rather, the urban skill bias reflects the incapability of less skilled individuals to afford livingin and moving to their preferred places, resulting in significant welfare losses.

Labor Market Segmentation and Immigrant Competition: A Quantal Response Statistical Equilibrium Analysis.

Competition between and within groups of workers takes place in labor markets that are segmented along various, often unobservable dimensions. This paper proposes a measure of the intensity of competition in labor markets on the basis of limited data. The maximum entropy principle is used to make inferences about the unobserved mobility decisions of workers in US household data. The quantal response statistical equilibrium class of models can be seen to give robust microfoundations to the persistent patterns of wage inequality. An application to labor market competition between native and foreign-born workers in the United States shows that this class of models captures a substantial proportion of the informational content of observed wage distributions.

Quantifying the Immediate Effects of the COVID-19 Pandemic on Scientists

The COVID-19 pandemic has undoubtedly disrupted the scientific enterprise, but we lack empirical evidence on the nature and magnitude of these disruptions. Here we report the results of a survey of approximately 4,500 Principal Investigators (PIs) at U.S.- and Europe-based research institutions. Distributed in mid-April 2020, the survey solicited information about how scientists' work changed from the onset of the pandemic, how their research output might be affected in the near future, and a wide range of individuals' characteristics. Scientists report a sharp decline in time spent on research on average, but there is substantial heterogeneity with a significant share reporting no change or even increases. Some of this heterogeneity is due to field-specific differences, with laboratory-based fields being the most negatively affected, and some is due to gender, with female scientists reporting larger declines. However, among the individuals' characteristics examined, the largest disruptions are connected to a usually unobserved dimension: childcare. Reporting a young dependent is associated with declines similar in magnitude to those reported by the laboratory-based fields and can account for a significant fraction of gender differences. Amidst scarce evidence about the role of parenting in scientists' work, these results highlight the fundamental and heterogeneous ways this pandemic is affecting the scientific workforce, and may have broad relevance for shaping responses to the pandemic's effect on science and beyond.

Hospital competition under pay‐for‐performance: Quality, mortality, and readmissions

AbstractHealth outcomes, such as mortality and readmission rates, are commonly used as indicators of hospital quality and as a basis to design pay‐for‐performance (P4P) incentive schemes. We propose a model of hospital behavior under P4P where patients differ in severity and can choose hospital based on quality. We assume that risk‐adjustment is not fully accounted for and that unobserved dimensions of severity remain. We show that the introduction of P4P which rewards lower mortality and/or readmission rates can weaken or strengthen hospitals' incentive to provide quality. Since patients with higher severity have a different probability of exercising patient choice when quality varies, this introduces a selection bias (patient composition effect) which in turn alters quality incentives. We also show that this composition effect increases with the degree of competition. Critically, readmission rates suffer from one additional source of selection bias through mortality rates since quality affects the distribution of survived patients. This implies that the scope for counterproductive effects of P4P is larger when financial rewards are linked to readmission rates rather than mortality rates.

Multidimensional Mortality Selection: Why Individual Dimensions of Frailty Don't Act Like Frailty.

Theoretical models of mortality selection have great utility in explaining otherwise puzzling phenomena. The most famous example may be the Black-White mortality crossover: at old ages, Blacks outlive Whites, presumably because few frail Blacks survive to old ages while some frail Whites do. Yet theoretical models of unidimensional heterogeneity, or frailty, do not speak to the most common empirical situation for mortality researchers: the case in which some important population heterogeneity is observed and some is not. I show that, when one dimension of heterogeneity is observed and another is unobserved, neither the observed nor the unobserved dimension need behave as classic frailty models predict. For example, in a multidimensional model, mortality selection can increase the proportion of survivors who are disadvantaged, or "frail," and can lead Black survivors to be more frail than Whites, along some dimensions of disadvantage. Transferring theoretical results about unidimensional heterogeneity to settings with both observed and unobserved heterogeneity produces misleading inferences about mortality disparities. The unusually flexible behavior of individual dimensions of multidimensional heterogeneity creates previously unrecognized challenges for empirically testing selection models of disparities, such as models of mortality crossovers.

Teacher Discrimination of High School Students, and Its Effect on Labor Market Outcomes

Using blind and non-blind evaluations, we show that the common identification strategy of gender bias in schools cannot identify a weighted average of gender bias for different ability levels due to the discrete nature of grades. We propose a new method, similar in spirit to matching, that enables the identification of heterogeneity in gender bias along both observable and unobservable dimensions. Utilizing our estimator, we quantify the difference-in-difference estimator to have a bias of approximately 30 percent of the true effect in our Norwegian high school data, and we find that gender bias can explain most of the average grade gap between male and female students. Low-ability students are exposed to the highest levels of bias in a classroom setting, and a one-standard-deviation-higher expected bias leads to students, on average, losing $750–1,000 in labor income each of the first five years after high school graduation.

Worker–Plant Matching and Ownership Change*

AbstractIs a change in ownership an opportunity for the new owners to make systematic changes to the workforce of the acquired plant? Using matched employer–employee data, we document changes to the workforce along observable and unobservable dimensions of worker quality around the time of ownership change. We observe above‐average separations of workers around domestic acquisitions. This is associated with a decline in unobserved worker quality in the plant. Foreign acquisitions are not associated with above‐average worker turnover; instead, new foreign owners share rents with the high‐skilled workers who are already in the plant before the acquisition.

The Impact of Risk Retention Regulation on the Underwriting of Securitized Mortgages

The Dodd-Frank Act requires securitization sponsors to retain not less than a 5% share of the aggregate credit risk of the assets they securitize. This paper examines how the implementation of risk-retention requirements affected the market for securitized mortgage loans. Using a difference-in-difference empirical framework, I find that risk retention implementation is associated with mortgages being issued with markedly higher interest rates, yet notably lower loan-to-value ratios and higher income to debt-service ratios. In addition, after controlling for observable loan characteristics, loans subject to risk retention requirements appear to be less likely to become troubled. These findings suggest that the risk retention rules have made securitized loans safer in both observable and unobservable dimensions, yet are more expensive to borrowers. Further evidence suggests that the risk-retention rules are binding, with the amount of risk being retained following implementation roughly three times that of before, while lenders also seemed to accelerate the securitization of originated loans during the months immediately before the rules took effect.

Covariance resampling for particle filter – state and parameter estimation for soil hydrology

Abstract. Particle filters are becoming increasingly popular for state and parameter estimation in hydrology. One of their crucial parts is the resampling after the assimilation step. We introduce a resampling method that uses the full weighted covariance information calculated from the ensemble to generate new particles and effectively avoid filter degeneracy. The ensemble covariance contains information between observed and unobserved dimensions and is used to fill the gaps between them. The covariance resampling approximately conserves the first two statistical moments and partly maintains the structure of the estimated distribution in the retained ensemble. The effectiveness of this method is demonstrated with a synthetic case – an unsaturated soil consisting of two homogeneous layers – by assimilating time-domain reflectometry-like (TDR-like) measurements. Using this approach we can estimate state and parameters for a rough initial guess with 100 particles. The estimated states and parameters are tested with a forecast after the assimilation, which is found to be in good agreement with the synthetic truth.

Marginalized Particle Flow Filter

As an alternative to the Kalman filter and the particle filter, the particle flow filter has recently attracted interest for solving the curse of dimensionality of the particle filter. Compared with the particle filter, the particle flow filter can obtain a better performance in high-dimensional state spaces with fewer samples. However, for some unobservable state dimensions, the flow operation wastes computational resources. In this paper, we propose a marginalized particle flow filter to handle the unobservable sub-state estimation. In contrast to the standard particle flow filter, we only migrate those observable dimensions of each particle according to homotopy theory and estimate the unobservable dimensions using the Kalman filter. The proposed algorithm can enhance the estimation quality of the unobservable state space and reduce the runtime of the particle flow filter. We evaluate the performance of the proposed algorithm through a multi-target tracking simulation.

Performance analysis of a novel oil-free rotary compressor

A novel rotating piston-type compressor is presented. The conventional rolling piston architecture was redesigned to allow oil-free operation. The spring activated vane was replaced by a directly driven swinging vane to provide constant contact to the rotating piston. The model of the compressor was implemented in the AMESim simulation platform. Experiments were conducted on prototypes and the data were evaluated based on the simulation results. Directly unobservable geometric dimensions were estimated by adjusting the mathematical model parameters to the measured thermodynamic state variables with the use of a multi-parametric genetic algorithm. The simple genetic algorithm proved fast and adequate solutions. According to the collected results, the oil-free rotating piston architecture is significantly more sensitive to the sealing clearances compared to the conventional oil lubricated rolling piston compressors. Therefore, the theoretically estimated performance can only be achieved with extremely small manufacturing tolerances, which has to be maintained during the operation.

The wage penalty for motherhood: Evidence on discrimination from panel data and a survey experiment for Switzerland

Background: Survey-based research finds a sizeable unexplained wage gap between mothers and nonmothers in affluent countries. The source of this wage gap is unclear: It can stem either from the unobserved effects of motherhood on productivity or from employer discrimination against mothers. Objective: This paper opens the black box of the motherhood wage gap by directly measuring discrimination in Switzerland based on two complementary methods. Methods: We first use two longitudinal population surveys to establish the size of the wage residual for motherhood. We then run a factorial survey experiment among HR managers (N=714) whom we asked to assign a starting wage to the resumes of fictitious job candidates. Results: The population surveys show an unexplained wage penalty per child of 4% to 8%. The factorial survey experiment shows that recruiters assign wages to mothers that are 2% to 3% below those of nonmothers. The wage penalty is larger for younger mothers, 6% for ages 40 and less, but disappears for older mothers or mothers in a blue-collar occupation. Conclusions: The motherhood wage gap found in panel studies cannot be reduced to unobserved dimensions of work productivity. The experimental evidence shows that recruiters discriminate against mothers. Contribution: Our paper’s novelty is to uncover wage discrimination against mothers by combining two different methods. Our national panel surveys mirror the supply side of the labor market and provide us with strong external validity. The factorial survey experiment on recruiters informs on the demand side of the labor market and shows a causal effect.

The effects of foreign aid on economic growth in developing countries

Decades of research regarding the effect of foreign aid on economic growth in less developed countries have produced inconclusive results. Research in this literature has been plagued by a variety of empirical impediments. Among them, measurement and endogeneity issues, sparse sets of control variables that may be correlated with both foreign aid and economic growth such as institutional quality, and disagreements regarding the appropriate econometric methodology. This paper highlights a further issue: the heterogeneous effects of foreign aid on growth across less developed countries. Previous studies have pooled all of the developing countries together, treating them as homogenous, despite that developing countries are vastly different across both observable and unobservable dimensions. Developing countries differ in their stages of development, per capita income, socio-economic, financial, and political characteristics. For this reason, the World Bank (2012) broadly classifies the developing countries into two categories: low income developing countries (LIDCs) and high income developing countries (HIDCs) based on per capita income. Our hypothesis is that the relationship between foreign aid and economic growth should be different among LIDCs and HIDCs. For this reason, we analyze the relationship between foreign aid and economic growth separately for LIDCs and HIDCs, producing estimates using samples which are more likely to be homogenous. Integrating the fullest set of control variables thus far in the literature such as unemployment rate, capital formation, government budget surplus, inflation rate, degree of trade openness, and corruption, and using GMM methodology in a dynamic setting we find that foreign aid has positive effects on growth in high-income developing countries and negative effects on growth in low-income developing countries. We also find that higher unemployment rates, higher inflation, and higher levels of corruption reduce economic growth in both LIDCs and HIDCs. Additionally, higher level of capital formation, a larger budget surplus, and higher degrees of trade openness contribute positively to economic growth in both LIDCs and HIDCs. We do not find any evidence that time trends in the data affect our results. The result remains after accounting for endogeneity concerns and when a measure of institutional quality proxied by corruption. The finding implies that foreign aid has beneficial effects in high-income developing countries which are at latter stages of development. This suggests that countries need to gain some traction before foreign aid can help.