Emily Oster

Unobservable Selection and Coefficient Stability: Theory and Evidence

A common approach to evaluating robustness to omitted variable bias is to observe coefficient movements after inclusion of controls. This is informative only if selection on observables is informative about selection on unobservables. Although this link is known in theory in existing literature, very few empirical articles approach this formally. I develop an extension of the theory that connects bias explicitly to coefficient stability. I show that it is necessary to take into account coefficient and R-squared movements. I develop a formal bounding argument. I show two validation exercises and discuss application to the economics literature. Supplementary materials for this article are available online.

Hepatitis B and the Case of the Missing Women

In many Asian countries the ratio of male to female population is higher than in the West: as high as 1.07 in China and India, and even higher in Pakistan. A number of authors (most notably Amartya Sen) have suggested that this imbalance reflects excess female mortality and have argued that as many as 100 million women are “missing.” This paper proposes an explanation for some of the observed overrepresentation of men: the hepatitis B virus. I present new evidence, consistent with an existing scientific literature, that carriers of the hepatitis B virus have offspring sex ratios around 1.50 boys for each girl. This evidence includes both cross‐country analyses and a natural experiment based on recent vaccination campaigns. Hepatitis B is common in many Asian countries, especially China, where some 10–15 percent of the population is infected. Using data on prevalence of the virus by country and estimates of the effect of hepatitis on the sex ratio, I argue that hepatitis B can account for about 45 percent of the “missing women”: around 75 percent in China, between 20 and 50 percent in Egypt and western Asia, and under 20 percent in India, Bangladesh, Pakistan, and Nepal.

Witchcraft, Weather and Economic Growth in Renaissance Europe

In the period between 1300 and 1800 as many as one million people were executed in Europe for the crime of witchcraft. Although a variety of theories about the witch-hunts have been put forward over time, little has been said about the possibility that the witch-hunts were motivated by a desire for vengeance in a time of misfortune. This paper connects the witch-hunts in Europe with deteriorating weather and slow economic growth during this period. The most intense period of witch-hunting coincided with a period of below-average worldwide temperatures known as the little ice age. I find that there is a statistically significant relationship between climate and number of trials: in times of worse climate there were more trials. This finding is robust to different measures of climate. In addition, I find that in periods of slower economic growth there were more witchcraft trials, even controlling for weather patterns. I argue that the two results together suggest that vengeance may have played an important role in witchcraft persecutions.

Fear of health insurance loss among individuals at risk for huntington disease

Genetic testing in Huntington disease, an inherited ultimately fatal neurodegenerative disorder, is infrequent despite wide availability. Factors influencing the decision to pursue testing are largely unknown. We conducted a prospective longitudinal observational study of 1,001 individuals in North America who were at risk for Huntington disease who had not pursued genetic testing prior to enrollment. We evaluated the rationale for remaining untested at baseline, determined the concerns of those who eventually pursued testing, and assessed the populations psychological attributes. We contrasted responses between those who did and did not pursue testing, and between United States and Canadian residents. The principal reasons for remaining untested were comfort with risk and uncertainty and the inability to “undo” knowledge gained. After enrollment, 83 individuals [8.3%] pursued genetic testing. Their greatest concern was losing health insurance, and 41.6% of them [vs. 6.7% of those who did not pursue testing; P < 0.001] reported paying out of pocket for testing or other medical services to conceal their genetic risk from their insurer/employer. Among individuals who were tested, more United States residents [46.1%] than Canadian residents [0.0%; P = 0.02] paid out of pocket for health services or genetic testing. Psychological attributes were similar among individuals who did and did not pursue testing. Individuals at risk for Huntington disease who pursued genetic testing feared losing medical insurance, and many paid out of pocket for medical services. Alleviating the fear of health insurance loss may help those who want to pursue genetic testing for many other conditions. [ClinicalTrials.gov number, NCT0052143].

Approaches and Costs for Sharing Clinical Research Data

The generation, dissemination, and sharing of research data are key ingredients in contributing to scientific progress and the public good. Data sharing has been encouraged to facilitate open science within the clinical research enterprise, improve the development of drugs and devices, and benefit public health.1 But sharing data is complex. Investigators generally tend to guard research data to retain ownership and property rights, avoid competition, protect confidentiality and privacy, or avoid misuse by unqualified persons. Sharing research data also comes at a cost to the sharer. This Viewpoint examines some approaches and cost considerations involved in sharing participant-level clinical research data.

Estimating HIV Prevalence and Incidence in Africa From Mortality Data

Abstract An estimated 33 million people are infected with the HIV virus, with 67% of them in Sub-Saharan Africa. Despite this, knowledge about HIV prevalence in Africa is limited and imperfect. Although population-based testing in recent years has provided reliable information about current prevalence in the general population, we have little reliable data on prevalence in early years of the epidemic. This paper suggests a new methodology for estimating HIV prevalence and incidence using inference from mortality data. This methodology can be used to generate prevalence estimates from early in the epidemic. This information is valuable for understanding how the epidemic has evolved over time and is also likely to be helpful in analyses that explore how policy affects the epidemic or how HIV affects other country-level outcomes.

Knowledge of the Genetic Information Nondiscrimination act among individuals affected by Huntington disease.

The Genetic Information Nondiscrimination Act (GINA) of 2008 was the first US legislation to address genetic discrimination. We sought to assess understanding of GINA among individuals affected by the autosomal dominant condition, Huntington disease (HD). We conducted a cross‐sectional survey of individuals with varying risk of HD to assess their familiarity with GINA. As a control, individuals were surveyed about their familiarity with the Health Insurance Portability and Accountability Act (HIPAA). Those who reported familiarity with GINA were asked about their knowledge of specific provisions of the legislation. The survey was offered to 776 participants and completed by 410 (response rate 53%). Respondents across all groups were less familiar with GINA (41% slightly, somewhat, or very familiar) than with HIPAA (65%; p < 0.0001). Of individuals with or at risk for HD who reported some familiarity with GINA, less than half correctly identified GINAs protections, and less than 15% correctly identified its limitations. Thus, among individuals affected by HD, familiarity with and knowledge of GINA are low. The effectiveness of the legislation may be limited by this lack of knowledge.

Sexually Transmitted Infections, Behavior Change and the Hiv/Aids Epidemic

Forty million people are infected with HIV worldwide; twenty-five million of them are in Sub-Saharan Africa. This paper addresses the question of why Africa has been so heavily affected by HIV, and what explains the variation within Africa. I present a model that decomposes epidemic level into differences in sexual behavior and differences in viral transmission rates. I argue, using evidence drawn from the existing medical literature, that Africa has very high HIV transmission rates, likely due to high rates of other untreated sexually transmitted infections. The difference in transmission is large enough to explain the observed difference in prevalence between the United States and Sub-Saharan Africa. The model also provides a good fit to cross-country data within Africa and suggests that, in contrast to the intra-continental results, differences within that continent can be attributed to difference in sexual behavior and epidemic timing. The results indicate that optimal policy interventions would focus on decreasing transmission rates within Africa, possibly by treating other untreated sexually transmitted infections.

Does disease cause vaccination? Disease outbreaks and vaccination response

Parental fear of vaccines has limited vaccination rates in the United States. I test whether disease outbreaks increase vaccination using a new dataset of county-level disease and vaccination data. I find that pertussis (whooping cough) outbreaks in a county decrease the share of unvaccinated children entering kindergarten. These responses do not reflect changes in the future disease risk. I argue that these facts are best fit by a model in which individuals are both myopic and irrational. This suggests that better promotion of information about outbreaks could enhance the response.