Poster presented at the Population Association of America Meetings, Online, May 2021

Differential Privacy is an emerging privacy model with increasing popularity in many domains. It functions by adding carefully calibrated noise to data that blurs information about individuals while preserving overall statistics about the population. Theoretically, it is possible to produce robust privacy-preserving visualizations by plotting differentially private data. However, noise-induced data perturbations can alter visual patterns and impact the utility of a private visualization. We still know little about the challenges and opportunities for visual data exploration and analysis using private visualizations. As a first step towards filling this gap, we conducted a crowdsourced experiment, measuring participants' performance under three levels of privacy (high, low, non-private) for combinations of eight analysis tasks and four visualization types (bar chart, pie chart, line chart, scatter plot). Our findings show that for participants' accuracy for summary tasks (e.g., find clusters in data) was higher that value tasks (e.g., retrieve a certain value). We also found that under DP, pie chart and line chart offer similar or better accuracy than bar chart. In this work, we contribute the results of our empirical study, investigating the task-based effectiveness of basic private visualizations, a dichotomous model for defining and measuring user success in performing visual analysis tasks under DP, and a set of distribution metrics for tuning the injection to improve the utility of private visualizations.

Low-skilled prime-age men are less likely to be employed than high-skilled prime-age men, and the differential has increased since the 1970s. I build a search model encompassing three explanations: (1) automation and trade reduced the demand for low-skilled workers; (2) health, welfare, and recreational gaming/computer technology reduced the supply of low-skilled workers; and (3) factors affecting job search, such as online job boards, reduced frictions for high-skilled workers. I find a shift in demand away from low-skilled workers was the leading cause, a shift in supply had little effect, and search frictions actually reduced employment inequality.

Background: States had flexibility in their implementation of the Patient Protection and Affordable Care Act (ACA) Medicaid expansions, which may have led to variation in coverage and changes in access to care for workers with disabilities. Objective/hypothesis: To examine differential trends in health insurance coverage and access to care among workers with disabilities by states’ decisions about expanding Medicaid under the ACA. Methods: We aggregated data from the National Health Interview Survey into groups by time period relative to ACA implementation: pre-ACA (2006–2009), early ACA (2010–2013), and later ACA (2014–2017). We produced health insurance and access statistics for each time period, by state-level Medicaid expansion status. Results: Uninsurance rates decreased after 2014 in all states, regardless of the state's decision whether to expand Medicaid. There was a substantial increase after 2014 in the share of workers with disabilities covered by Medicaid in states that expanded in that year; in other states, workers with disabilities experienced larger increases in privately purchased coverage. At the same time, the share of workers with disabilities reporting cost-related barriers to care declined markedly in 2014 Medicaid expansion states, but it increased slightly in the non-expansion states. Structural barriers to accessing care increased in all states, with the smallest increase in 2014 expansion states. Conclusions: Medicaid coverage and cost-related access to care improved significantly among workers with disabilities in 2014 Medicaid expansion states, both overall and relative to workers with disabilities in non-expansion states.

Every student should have access to a high-quality education. Yet new analysis finds that Virginia’ English learner (EL) student population faces significant barriers to education and the state is falling short in providing adequate resources to assist students in overcoming these barriers. With around 1 in 10 Virginia K-12 school students classified as EL, Virginia is now on par with the national average for representation. EL students are diverse and are more likely to speak a language other than Spanish, compared to EL students nationally.

Universal Basic Income (UBI) is a program in which individuals receive a regular sum of money, usually from the government. The transfer amount is thought to be unconditional of income and enough to cover all subsistence needs. Such a system is easy and cheap to administer because the government does not need to check the eligibility of each applicant. UBI programs are growing as more cities, states and countries (Stockton, California, Newark, New Jersey, Ontario, Canada, Kenya, Finland, Germany, Spain, China, etc) implement experiments of such programs. The idea of a UBI is gaining ground in the U.S.. One of the main responses of the U.S. to high unemployment caused by the COVID-19 pandemic and quarantine was a modified version of a temporary country-wide UBI program in 2020 (CARES Act). 30 mayors across the U.S. created a coalition - Mayors for a Guaranteed Income - to explore cash payment programs and address the racial wealth inequality. UBI is actively discussed to be a potential policy that can mitigate adverse impact of accelerated automation on wages and employment. Thus, it is important to understand what we have learned from UBI experiments, what macroeconomic models predict in the UBI environment, and what is the best approach to implement such programs.

In this article, we argue that accounting for sibling complexity is a necessary step toward more accurate assessments of family structure. First, we argue that current conceptualizations of family structure are rooted in (and reinforce) Eurocentric definitions of family, and we highlight contradictions between family theory and measurements of family structure. Second, we discuss the prevalence of diverse sibling compositions in families and show the informatice value of accounting for sibling complexity. Third, we explore the barriers to accounting for sibling structure by evaluating the extent to which complex sibling compositions are captured in publicly available secondary datasets recently used to study families. Finally, we consider both theoretical and methodological implications of failing to account for sibling complexity in family research and offer recommendations for future data collection efforts.

Despite research going back over a century showing the U.S. census counts some groups more accurately than others at the national level, little is known about how undercount varies within the country. I focus on a population easily measured with administrative data yet known to suffer high levels of undercount—native-born young children—to document state-level variation in undercount by race and Hispanic status. Although the race-specific analysis is only possible for the 2000 census, the patterns I show for all children are similar to those in 2010, implying the results from 2000 are likely relevant to today. Undercount levels vary widely across states, with non-Black children having the highest rates in the south and southwest, and Black children in the northeast. Results by Hispanic status show non-Black Hispanic young children are highly undercounted in several states with high Hispanic populations, but not all, and are also highly undercounted in the northeast and New England. In several states with high non-Black Hispanic undercounts, non-Black non-Hispanic children are also undercounted at a high rate. I find a very strong correlation between the fraction of births to foreign-born mothers in the state and the undercount of Black and non-Black children—in fact, it is the strongest correlate with the undercount of native-born Black children of those I investigate. The fraction of foreign-born mothers does not correlate with the undercount of non-Black Hispanic and non-Hispanic young children, although Hispanic status of the parents do. My results suggest a group-specific, local focus for future work is needed to determine the causes of census undercount.

—Flow map is an effective way to visualize object movements across space over time. It aims to model the paths from destinations to origins with quality constraints satisfied, which is similar to river system extraction in a digital elevation model (DEM). In this paper, we present a novel and automated approach called RFDA-FM for spatial flows from one origin to multiple destinations using a river flow directions assignment algorithm in DEM. The RFDA-FM first models the mapping space as a flat surface by DEM. An improved maze solving algorithm (MSA) is then introduced to assign the flow directions by constraining its searching directions, direction weights and searching range. The paths from the destinations to the origin are obtained iteratively based on the improved MSA according the path importance. Finally, these paths are rendered with varied widths and smoothed according to their volume using the Bézier curves. The evaluation results indicate that the flow maps generated by RFDA-FM can have a higher quality on uniform distribution of edge lengths and avoidance of self-intersections and acute angles by comparing to the existing approaches. The experiments demonstrate that RFDA-FM is also applicable for heterogeneous mapping space or mapping space with obstacle areas.

The United States has been experiencing a drug overdose mortality epidemic marked by the introduction and spread of opioids across rural and urban communities over the past 20 years. The current Coronavirus (COVID-19) pandemic has overshadowed the opioid epidemic but aggravated the opioid problem by hindering access to health services and increasing the number of people out of work. Research on the geography of the opioid epidemic has focused on the association between declining local economic opportunities and increases in drug overdose mortality since 2000, but the link has not always been strong. This study identifies two phases comprising the epidemic and examines their differing demographic and geographic natures. Results show that in the first phase, beginning around 2000 and ending in the early 2010s, drug overdose mortality rates soared among the middle aged as prescription opioid painkillers drove the epidemic. Physical disability is associated with chronic pain, and during this period, drug overdose deaths rose most in areas with high physical disability rates. We found little evidence that the aggravation of local economic problems was associated with increases in drug overdose mortality in the 2000s. Since the early 2010s, opioid drug reformulation and declining prescription rates have resulted in ebbing mortality from prescription opioids. At the same time, illicit opioids such as heroin and, increasingly, fentanyl and related synthetic opioids rapidly entered the scene—causing a growing share of drug overdose deaths, marking the beginning of what we call the illicit opioid phase of the epidemic. In this phase, drug overdose mortality has risen, particularly among young adult males, and its geography has shifted markedly. Physical disability seems less a factor in rising drug overdose mortality than local (county) economic hardship and outmigration in the northeastern quadrant of the United States, where the marketing of these drugs currently appears most developed.

Employment interruption is a common experience in today’s labor market, most frequently due to unemployment from job loss and temporary lapses to care for family or children. Although existing research shows that employment lapses cause disadvantages at the hiring interface compared to individuals with no employment disruptions, competing theories predict different mechanisms explaining these hiring penalties. In this study, the author uses an original conjoint survey experiment to causally assess perceptions of fictitious job applicants, focusing on a comparison of unemployed applicants and nonemployed caregiver applicants, who left work to care for family, to currently employed applicants. The author examines whether disadvantages for job applicants with employment gaps are receptive to positive information (and therefore represent a form of “informational bias”) or are resistant to information (reflecting “cognitive bias”) and further assesses which types of information affect or do not affect levels of bias in fictitious hiring decisions. Results show that positive information on past job performance and social skills essentially eliminates disadvantages faced by unemployed job applicants, but nonemployed caregiver applicants remain disadvantaged even with multiple types of positive information. These findings suggest that unemployed applicants face informational biases but that nonemployed caregiver applicants face cognitive biases that are rigid even with rich forms of positive or counter-stereotypical information. This study has implications for understanding the career consequences of employment disruption, which is especially relevant to consider in light of labor market disruptions during the recent pandemic.

The Affordable Care Act (ACA) not only changed the landscape of health insurance coverage in the United States but also affected the relationship between working decisions and health insurance. In this paper, we estimate the impact of the ACA on the near elderly (ages 60–64) in the 5 years after the implementation of its key provisions in early 2014. We exploit variation across geographic areas in the preexisting level of uninsurance and use 65–69-year-olds, whose insurance coverage was unaffected by the ACA, as a within-region control group. Our findings indicate that the ACA increased health insurance coverage among the near elderly by 4.5 percentage points and reduced their labor force participation rate by 0.6 percentage points.

Founded in family systems theory, the spillover hypothesis suggests that marital relationships are linked to parenting. The current study used meta-analysis to extend this literature and examine links between marital positivity and marital negativity with absolute levels of parental differential treatment (PDT) of siblings. Multilevel data included 2575 effect sizes nested within 45 sources (articles/raw datasets), nested within 12 unique samples. Lower marital positivity and greater marital negativity were linked to greater differences in treatment. Those main effects, however, were moderated by several source and effect size characteristics. For example, links were stronger when effect sizes were based on differential positive interactions. Caution is warranted in most cases; effect sizes were generally small. Findings, however, suggest that parents should be aware of how they may treat their children differently in connection with poorer marital relationships. Future research on PDT should include a greater focus on the domains of differential treatment.

Immigration policy debates currently focus on restricting immigration in favor of the highly skilled with the assumption that highly skilled immigrants will be better able to join the labor market and contribute to the economy. However, few studies empirically test the impact of immigrant selection policy by comparing labor market outcomes of immigrants from a single origin in multiple destinations. Fewer still address how race (specifically blackness) may impact the utility of these selection policies. This paper fills this gap by determining Ethiopian immigrants’ labor force participation, occupational status, and self-employment in the United States and Israel—countries with and without immigrant selection policies respectively. We find that Ethiopians experience similar labor market disadvantages relative to the native-born in both countries. These results indicate that rather than selection policy being the driver of labor market success, racial discrimination likely plays the largest role in determining Ethiopian (black) African immigrants’ labor market incorporation in both places.

We describe the artifact, publicly available at [1], that implements the proposal in [2], and the reproduction of the experimental results. It is an extended and distributed version of the Mondrian anonymization algorithm. Our solution anonymizes large datasets by partitioning data among workers in a distributed setting. It provides parallel execution on a dynamically chosen number of workers, limiting their interaction and data exchange.

Objective: To re-evaluate the effect of Medicaid on poverty using a poverty measure that accounts for health insurance needs and benefits and an evaluation approach that reflects disparities in access to alternative coverage. Data Sources: The Current Population Survey (CPS) for calendar year 2015. Study Design: We estimate the effect of losing Medicaid on poverty, combining two previous approaches: (1) A propensity impact, which simulates a no-Medicaid counterfactual incorporating changes to health insurance and medical out-of-pocket spending, using the Supplemental Poverty Measure (SPM). This measure does not reflect a need for health care access nor how health benefits meet that need. (2) An accounting impact, which assumes that those losing Medicaid remain uninsured and does not incorporate any behavioral changes, using the health-inclusive poverty measure (HIPM). This measure includes a need for health insurance in the threshold and health insurance benefits in resources. Data Collection/Extraction Methods: Not applicable. Principal Findings: Using the propensity-matched approach, we attributed a 2.5 percentage point reduction in health-inclusive poverty among those younger than age 65 to the Medicaid program, between the 1.0-point SPM propensity-match impact and the 3.9-point HIPM accounting impact. Medicaid's antipoverty impact and HIPM-SPM differences are greater among those who would become uninsured. HIPM propensity-matched estimates reveal much larger impacts of Medicaid on poverty disparities linked to race/ethnicity and single parenthood than SPM-based propensity estimates. Conclusions: Both the poverty measure and the method used to estimate the counterfactual make substantial, policy-relevant differences to estimates of Medicaid's impact on poverty. A poverty measure that fails to incorporate health insurance needs and benefits substantially underestimates Medicaid's effect. Failing to consider adjustments in insurance coverage and out-of-pocket spending substantially overestimates Medicaid's effect and underestimates its reduction of disparities.

View recommendations have emerged as a powerful tool to assist data analysts in exploring and understanding big data. Existing view recommendation approaches proposed a variety of utility functions in selecting useful views. However, the suitability of the utility functions and their tunable parameters for an analysis is usually dependent on the analysis context, such as the user, the data and the analysis task. In order to provide context-aware view recommendation, we formulate a new Interactive View Recommendation (IVR) paradigm, where the system interacts with the user to discover the utility functions that are most suitable in the current analysis context. We further develop an IVR framework, coined ViewSeeker, which leverages user feedback on intelligently selected example views to discover the most suitable utility functions. Finally, we implemented a prototype of ViewSeeker and verified its efficiency and effectiveness using two real-world datasets.

Although scholars have studied education's effects on many different outcomes, little attention has been paid to its effects on adults’ economic views. This article examines those effects. It presents results based on longitudinal data which suggest that secondary education has a little-appreciated consequence: it makes Americans more opposed to redistribution. Placebo tests and other analyses confirm this finding. Further investigation suggests that these conservative effects of education operate partly by changing the way that self-interest shapes people's ideas about redistribution.

This paper studies how the micro behaviors of oil tankers and their interactions with exporters trigger shipping market congestion and pose significant implications for oil trade, prices, and final good production. By constructing a unique shipping dataset that tracks 757 oil tankers during 2017­2020, we document three novel facts: shipping market congestion, prevalence of market switching behavior by oil tankers, and propagation of oil trade shocks through a transportation network. By embodying the central mechanisms underlying the novel facts, we develop in a pioneering way a search and matching model that connects oil exports, transportation, and final good production, with a focus on the formation of shipping market congestion. Counterfactual exercises unveil that while congestion attenuates over one third of the volatility in oil trade and final good production that would otherwise have occurred in a congestion­free environment due to the impacts of oil trade shocks, this comes at the cost of exaggerated oil price fluctuations. Furthermore, compared to exogenous trade shocks, the corresponding endogenous adjustment of the oil transportation network only plays a minimal role in creating fluctuations of those macro aggregates, including oil prices.

This document reports on the implementation efforts associated with the development of two tools supporting novel techniques for data sanitization. The proposed approaches specifically focus on scenarios where different parties aim at collaborating to anonymize a dataset or to compute (in a privacy preserving manner) statistics over private data. The document first illustrates an anonymization approach that operates in a distributed scenario. The proposal specifically extends the Mondrian algorithm to leverage the presence of multiple workers to improve scalability and enable the efficient anonymization of large data collections. The developed tool is able to compute a k-anonymous and `-diverse version of the dataset relying on an arbitrary number of workers, without affecting information loss. The tool has received the Best Artifact Award at the IEEE PerCom 2021 Conference. The document then presents a novel solution for computing the differentially private median over the union of two private datasets. The performance of the developed solution outperforms existing approaches, while maintaining utility comparable to computations performed in a centralized scenario.