Wiley

Health Economics

Published by Wiley
Online ISSN: 1099-1050
Discipline: Economics
Learn more about this page
Aims and scope

This Journal publishes articles on all aspects of health economics: theoretical contributions, empirical studies and analyses of health policy from the economic perspective. Its scope includes the determinants of health and its definition and valuation, as well as: the demand for and supply of health care; planning and market mechanisms; micro-economic evaluation of individual procedures and treatments; and evaluation of the performance of health care systems.

Contributions should be original and innovative. As a rule, the Journal does not include routine applications of cost-effectiveness analysis, discrete choice experiments and costing analyses. Editorials, which are regular features, should be concise and topical. Occasionally, commissioned reviews are published, and special issues bring together contributions on a single topic.

 

Editors

Recent publications
Bounds on the Effects of Childhood Sexual Abuse Under Varying Assumptions About Confounding from Unobservables. The figure above depicts the average marginal effect of childhood sexual abuse under varying assumptions about the importance of unobservables. Models controlled for age when outcome was measured, sex, race, highest parental educational attainment, childhood household income, school fixed effects, childhood disability, and other adverse childhood experiences (ACEs) (childhood physical abuse, emotional abuse, knife or gun violence or threat, parental divorce or separation, parental incarceration, and illegal drugs in home). The points marked represent, from left to right: Rmax = 1.3R∼ $\tilde{R}$ and Rmax = 2R∼ $2\tilde{R}$. When δ = 0, Rmax must equal R∼ $\tilde{R}$, but the average marginal effect for δ = 0 is plotted over the same values of Rmax for ease of comparison with the estimates under assumptions of δ > 0. a Panel (d) models log earnings, among those with any earnings. OLS models of the binary outcome “having any earnings” showed no relationship between childhood sexual abuse and having positive earnings. Descriptive statistics for earnings noted here are provided for the sample with positive earnings. Source: Author's calculation from Add Health data
Article
Childhood sexual abuse is a prevalent problem, yet understanding of later‐in‐life outcomes is limited due to unobservable determinants. I examine impacts on human capital and economic well‐being by estimating likely ranges around causal effects, using a nationally representative U.S. sample. Findings suggest that childhood sexual abuse leads to lower educational attainment and worse labor market outcomes. Results are robust to partial identification methods applying varying assumptions about unobservable confounding, using information on confounding from observables including other types of child abuse. I show that associations between childhood sexual abuse and education outcomes and earnings are at least as large for males as for females. Childhood sexual abuse by someone other than a caregiver is as influential or more so than caregiver sexual abuse in predicting worse outcomes. Considering the societal burden of childhood sexual abuse, findings could inform policy and resource allocation decisions for development and implementation of best practices for prevention and support.
 
Article
The year 2022 is the 50th anniversary of the publication of my demand for health model in “On the Concept of Health Capital and the Demand for Health,” Journal of Political Economy 80(2): 223–255, and in The Demand for Health: A Theoretical and Empirical Investigation, NBER Occasional Paper 119 New York: Columbia University Press for the NBER. To mark that occasion, this editorial focuses on the history of the model and its impacts on the field of health economics.
 
Variation in distancing behavior by county Democrat share. This figure shows how the change in daily distance traveled varies with the Democrat share in a county, separately for states with a Democratic or Republican governors. Both the daily distance variable and the Democratic share are residuals from a regression of these variables on state fixed effects, county population, the share of the population that is female, white, black, Hispanic, has less than a high school education, has less than a college education, is under age 29, or is over age 65. The size of the circles in the figure represents the size of the county population
Variation in visitation behavior by county Democrat share. This figure shows how the change in daily visitation varies with the Democrat share in a county, separately for states with a Democratic or Republican governors. Both the daily distance variable and the Democratic share are residuals from a regression of these variables on state fixed effects, county population, the share of the population that is female, white, black, Hispanic, has less than a high school education, has less than a college education, is under age 29, or is over age 65. The size of the circles in the figure represents the size of the county population
Article
In this paper, we examine the relationship between political polarization and individuals' willingness to contribute to the public good by engaging in preventative behaviors against COVID‐19. Using a sample of individuals from close‐election states, we first show that individuals engage in fewer preventative behaviors when the governor of their state is from the opposite party. We also show that this effect is concentrated among moderate individuals who live in polarized states, and that it is strongest when the state has been relatively forceful in combating COVID‐19. We estimate that the opposite‐party effect increased COVID‐19 cases by around 1%.
 
Economic consequences of different levels of government borrowing
Numerical comparative statics of the optimal borrowing
Article
Governments worldwide have issued massive amounts of debt to inject fiscal stimulus during the COVID‐19 pandemic. This paper analyzes fiscal responses to an epidemic, in which interactions at work increase the risk of disease and mortality. Fiscal policies, which are designed to borrow against the future and provide transfers to individuals suffering economic hardship, can facilitate consumption smoothing while reduce hours worked and hence mitigate infections. We examine the optimal fiscal policy and characterize the condition under which fiscal policy improves social welfare. We then extend the model analyzing the static and dynamic pecuniary externalities under scale economies—the decrease in labor supply during the epidemic lowers the contemporaneous average wage rate while enhances the post‐epidemic workforce health and productivity. We suggest that fiscal policy may not work effectively unless the government coordinates working time, and the optimal size of public debt is affected by production technology and disease severity and transmissibility.
 
Article
Emergencies, such as natural and manmade disasters, can present an opportunity or be a detriment to preventive healthcare. While stay‐at‐home orders which some states implemented to mitigate the impact of COVID‐19 are known to reduce acute and routine care, little is known about missed preventive care. Dental care, unlike other forms of preventive care ‐ such as pediatric vaccines and well‐visits, is simpler to analyze as it is not practicable with telehealth. Using weekly foot traffic data by SafeGraph from January 2018 to June 2020, we examine the effect of stay‐at‐home orders on visits to dental offices, finding a 15.4% decline after March 2020 for states with stay‐at‐home orders. Surprisingly, we find that states which allowed dental care during the stay‐at‐home period experienced a further 7.4% decline in visits. Using Michigan Medicaid dental claims for children we find that the decline of 0.25 claims per month is driven primarily by fewer diagnostic and preventive care visits. Though some preventive visits were rescheduled, we estimate only 58% of visits missed in March and April 2020 were made up by the end of the year. These estimates quantify the short‐term declines in preventive dental care, suggesting similar declines in other preventive care.
 
Major flood occurrences in Indonesia and the World. Graphs created by authors based on flood occurrence data by country available at the Emergency Events Database (EM‐DAT) by the Université Catholique de Louvain (UCL), CRED, D. Guha‐Sapir. Disasters qualify into the database if they fulfill at least one of the following criteria: 10 or more people reported killed, 100 or more people reported affected, a declaration of a state of emergency or a call for international assistance (https://www.emdat.be/explanatory‐notes). (a) Includes the top 13 countries in the world in terms of flood occurrence in the past 50 years. (b) Shows the number of large‐scale floods reported every year since 1970 by the Indonesian authorities.
Article
Billions of people live in urban poverty, with many forced to reside in disaster‐prone areas. Research suggests that such disasters harm child nutrition and increase adult morbidity. However, little is known about impacts on mental health, particularly of people living in slums. In this paper we estimate the effects of flood disasters on the mental and physical health of poor adults and children in urban Indonesia. Our data come from the Indonesia Family Life Survey and new surveys of informal settlement residents. We find that urban poor populations experience increases in acute morbidities and depressive symptoms following floods, that the negative mental health effects last longer, and that the urban wealthy show no health effects from flood exposure. Further analysis suggests that worse economic outcomes may be partly responsible. Overall, the results provide a more nuanced understanding of the morbidities experienced by populations most vulnerable to increased disaster occurrence.
 
Parent and Teacher ADHD Assessments by School Starting Age. Based on a sample of all kindergarten‐aged students in the National Longitudinal Survey of Child and Youth (NLSCY) from years 1994–2002. Figure plots binned mean teacher (left) and parent (right) assessments of student ADHD severity by gender using an ADHD behavioral index on a scale of 0–16. The running variable Daysi is the number of days before or after an eligibility cut‐off date that a child's birthday falls, on a range of −180 to 180. Each point represents an averaged assessed ADHD score by binned on Daysi. Shaded regions represent 95 percent confidence intervals, with standard errors clustered on Daysi
Teacher ADHD Assessments. This figure shows reduced form regression discontinuity estimates of teacher assessments and residual teacher assessments overall and by female and male subgroups. Each point represents a teacher (residual) assessment binned by daysi, using an evenly spaced integrated mean squared error method, scaled by a factor of two to avoid over‐smoothing (Calonico et al., 2018). Fitted lines are quadratic polynomial estimates. Error bars correspond to 95 percent confidence intervals, with standard errors clustered on daysi
Special Education Training. Figure plots binned teacher assessments by school starting age and whether a teacher has special education training (blue) or not (black). Each point shows mean assessment errors binned by Daysi before or after an eligibility cut‐off date that a child's birthday falls, on a range of −180 to 180. Plots use a integrated mean squared error (IMSE) evenly‐spaced optimal bin selection method to calculate bin size and flexible global polynomial fits. Error bars represent 95 percent confidence intervals, with standard errors clustered on Daysi
Article
ADHD diagnoses increase discontinuously by a child's school starting age, with young‐for‐grade students having much higher ADHD diagnostic rates. Whether these higher rates reflect over‐diagnosis or under‐diagnosis remains unknown. To decompose this diagnostic discrepancy, we exploit differences in parent and teacher pre‐diagnostic assessments within a regression discontinuity strategy based on school starting age. We show that being young‐for‐grade or male generates over‐assessment of symptoms specifically from teacher assessment. However, under‐assessments of the oldest students in a grade, especially the oldest females, account for a large part of the observed school starting age assessment gap. We argue that this difference by sex and higher school starting age effects in lower‐income schools may exacerbate known gaps in educational attainment by gender and socioeconomic status. Importantly, we fail to find evidence that teachers who receive special education training make such errors.
 
Measures adopted by the Italian government to fight the COVID‐19 epidemic in early 2020. 13 days after the first community case was identified, the government ordered all educational institutions to close down and switch to online learning (5th March). Less than a week later, it introduced a nationwide lockdown and ordered the closure of all non‐essential commercial activities, such as shops, bars and restaurants (11th March). Eleven days later, it compounded those measures by shutting down all non‐essential production activities (22nd March). Restaurants could offer delivery, while non‐essential production activities could operate if they were supplying goods to firms in essential sectors. As the situation gradually improved, the government lifted the lockdown and allowed most businesses to reopen after almost 2 months of restrictions (fourth May)
Article
Governments around the world have adopted unprecedented policies to deal with COVID‐19. This paper zooms in on business shutdowns and investigates their effectiveness in reducing mortality. We leverage highly granular death registry data for almost 5000 Italian municipalities in a diff‐in‐diff approach that allows us to mitigate endogeneity concerns credibly. Our results, which are robust to controlling for a host of co‐factors, offer strong evidence that business shutdowns effectively curb mortality. We calculate that they may have reduced the death toll from the first wave of COVID‐19 in Italy by about 40%. Our findings also highlight that timeliness is key – by acting 1 week earlier, their effectiveness could have been increased by an additional 5%. Finally, shutdowns should be targeted. Closing service activities with a high degree of interpersonal contact saves the most lives. Shutting down production activities, while substantially reducing mobility, only has mild effects on mortality.
 
Distribution of Hot Days above 80°F Across States and Years. Each bar shows the number of days in a given state and year with at least one county in the state experiencing daily average temperature above 80°F. Sources: GHCN daily (GHCND) weather data 2009–2018.
Distribution of Daily Average Temperature during Pregnancy. Each bar shows the average number of days during pregnancy with mean temperature in the corresponding temperature range defined in absolute temperature bins measured in Fahrenheit (left), and in relative temperature bins measured in terms of deviation from the county‐month historic mean temperature (right). Sources: GHCN daily (GHCND) weather data merged with U.S. National Vital Statistics Birth Data 2009–2018.
Distribution of Residuals in Temperature after Controlling for All Fixed Effects. (a) Days with above 3SD heat and (b) Days with above 90F heat. The residuals are derived from a regression of the number of days in top temperature bins on race‐by‐birth county‐by‐birth month fixed effects and birth state‐by‐birth year fixed effects. Sources: GHCN daily (GHCND) weather data merged with U.S. National Vital Statistics Birth Data 2009–2018.
Article
We provide the first estimates of the impacts of prenatal exposure to extreme temperatures on infant health at birth using the latest national birth data from 2009 to 2018 from all U.S. states. We consistently find that an additional day with mean temperature greater than 80°F or less than 10°F increases preterm births and low birthweight. Strikingly, the adverse effects are borne disproportionately by Black and Hispanic mothers, suggesting that the projected increase in extreme temperatures may further exacerbate the existing birth health disparities across different race/ethnicity groups. We also contribute by investigating the impact of deviations from the normal weather pattern, to identify the extreme weather events after accounting for the adaptation response. We find that prenatal exposure to extreme heat two standard deviations above county's historic average induces preterm births and NICU admissions, particularly for mothers whose pregnancies overlap with summer months. These results are timely and policy relevant, considering the recent weather trends with rising temperatures and frequent extreme weather events.
 
Predicted change in psychological distress by employment status. (a) 2013–2016 (b) 2018–2020. Based on predictions from the regressions in Table 4
Predicted outcomes by financial concern and job loss (a) Probability of psychological distress (b) Change in probability of psychological distress. Predictions based on regression results from table 5. Whiskers denote 95% confidence interval. Difference refers to the difference between those who lost employment or reduced hours during the pandemic versus those who retained employment during the pandemic and had no reduction in work hours. The parameter for No financial concerns difference is given by the coefficient on “Lost employment or reduced work hours during pandemic” in Table 5, while the parameter for Financial concerns Difference is given by the sum of the coefficients on “Lost employment or reduced work hours during pandemic” and “Lost employment or reduced work hours during pandemic X financial concerns” in Table 5
Article
While psychological distress is a common sequelae of job loss, how that relationship continued during the COVID‐19 pandemic is unclear, for example, given higher health risk to working due to disease exposure. This paper examines changes in psychological distress depending on job loss among a cohort of randomly selected residents living in nine predominantly African American low‐income neighborhoods in Pittsburgh PA across four waves between 2013 and 2020. Between 2013 and 2016, we found an increase in psychological distress after job loss in line with the literature. In contrast, between 2018 and 2020 we found change in psychological distress did not differ by employment loss. However, residents who had financial concerns and lost their jobs had the largest increases in psychological distress, while residents who did not have serious financial concerns—potentially due to public assistance—but experienced job loss had no increase in distress, a better outcome even than those that retained their jobs. Using partial identification, we find job loss during the pandemic decreased psychological distress for those without serious financial concerns. This has important policy implications for how high‐risk persons within low‐income communities are identified and supported, as well as what type of public assistance may help.
 
Percentage deviation and geographical pattern – regions. Panel (a) shows the percentage deviation from the national mean, where zero on the x‐axis represents the mean national spending of prescribed drugs per capita per year and the horizontal bars show the percentage deviation in average regional drug spending. The national (weighted) mean was €319 (€1 = 10.5 SEK 2019). Panel (b) shows the geographical pattern of mean regional spending, in euros per capita. The averages in (a) and (b) are pooled over the years 2007–2016, using the full sample of 929,711 individuals.
Percentage deviation from national mean – municipalities. Zero on the x‐axis represents the mean national spending of prescribed drugs per capita per year and the horizontal bars show the percentage deviation in average municipal drug spending. The averages are pooled over the years 2007–2016, using the full sample of 929,711 individuals. The full sample's national (weighted) mean was €319 (€1 = 10.5 SEK 2019)
Event study of pre‐ and post‐move trends – variation across regions. Estimated based on Model 2 with ln(spending+1) as the dependent variable and including independent variables of individual characteristics. The year‐specific thetas are restricted to 5 years before and 5 years after the move
Event study of pre‐ and post‐move trends – variation across municipalities. Estimated based on Model 2 with ln(spending+1) as the dependent variable and including independent variables of individual characteristics
Article
There is substantial variation in drug spending across regions in Sweden, which can be justified if caused by differences in health need, but an indication of inefficiencies if primarily caused by differences in place‐specific supply‐side factors. This paper aims to estimate the relative effect of individual demand‐side factors and place‐specific supply‐side factors as drivers of geographical variation in drug spending in Sweden. We use individual‐level register data on purchases of prescription drugs matched with demographic and socioeconomic data of a random sample of about 900,000 individuals over 2007–2016. The primary empirical approach is a two‐way fixed effect model and an event study where we identify demand‐ and supply‐side effects based on how regional and local migrants change drug spending when moving across regional and municipal borders. As an alternative approach in robustness checks, we also use a decomposition analysis. The results show that the place‐specific supply‐side effect accounts for only about 5%–10% of variation in drug spending and remaining variation is due to individual demand‐side effects. These results imply that health policies to reduce regional variation in drug spending would have limited impact if targeted at place‐specific characteristics.
 
Map of medical cannabis legalized states. Only “Never Treated” states are used as our control group
Map of zip codes close to dispensaries
Map of bordering zip‐code pairs with accessible dispensaries
Article
While many states have legalized medical cannabis, many unintended consequences remain under‐studied. We focus on one potential detriment–the effect of cannabis legalization on automobile safety. We examine this relationship through auto insurance premiums. Employing a modern difference‐in‐differences framework and zip code‐level premium data from 2014 to 2019, we find that premiums declined, on average, by $22 per year following medical cannabis legalization. The effect is more substantial in areas near a dispensary and in areas with a higher prevalence of drunk driving before legalization. We estimate that existing legalization has reduced health expenditures related to auto accidents by almost $820 million per year with the potential for a further $350 million reduction if legalized nationally.
 
Article
Governments often encourage health service providers to improve quality of care and reduce prices through competition. The efficacy of competition hinges on the assumption that consumers demand high quality care at low prices for any given health condition. In this paper, we examine this assumption by investigating the role of perceived price and quality on consumer choice for four different health conditions across public and private providers. We use a nationally representative survey in Malaysia to elicit respondents' perception on prices and quality, and their preferred choice of provider. We estimate a mixed logit model and show that consumers value different dimensions of quality depending on the health condition. Furthermore, increasing perceived prices for private providers reduces demand for minor, more frequent health conditions such as flu fever or cough, but increases demand for more complex, severe conditions such as coronary artery bypass graft. These findings provide empirical support for price regulation which differentiates the severity of underlying health conditions.
 
Overweight and obesity rate of Chinese juveniles in 1985–2016. We calculated the rates using the data from NCD Risk Factor Collaboration (NCD‐RisC) (2017). The data were accessed at https://www.ncdrisc.org/data‐downloads‐adiposity‐ado.html [Feburary 8, 2022]
Article
In this paper, we examine the causal impact of weight on adolescent mental health. Using the China Family Panel Studies, we find significant negative effects of adolescent weight, instrumented by cohort‐level parental body mass index (BMI), on mental illness. In particular, a one standard deviation increase in adolescent BMI z‐score decreases the K6 score by 0.766 (or 0.232 standard deviations). This finding is contrary to recent evidence from adults. We find this contrast can partly be explained by the different impacts of adolescent weight on self‐image and social relationships. Unlike adults, heavy adolescents feel that they are popular among peers in China.
 
Article
This article examines the effect of recreational cannabis dispensary sales on traffic crashes by employing difference‐in‐differences model that exploits the variation in the timing of recreational marijuana dispensary entry across counties within Colorado. Using marijuana‐related hospital discharge as a measure of marijuana abuse/misuse, the results indicate a sizable rise in marijuana‐related hospital discharges after the entry of retail cannabis stores. However, there is a lack of evidence that traffic crash incidents are affected by the entry. The preferred estimate suggests that, at 90% confidence level, a large increase in traffic crashes by more than 5% can be ruled out.
 
Annual percentage of U.S. hospitals integrated with physicians. Source: American Hospital Association Survey, 2000–2013. Percentages are based on 39,235 total observations of 3926 hospitals
Matched hospitals average ln (operating expenses) by years from integration of integrating hospitals. The figure displays the average of the natural log of hospital expenses for the treatment and control hospitals separately from 5 years prior to integration through 5 years after integration of the treatment hospitals. Year zero is the first of integration for each treatment hospital
Article
This study evaluates whether hospital costs are lower when hospitals integrate with physician practices. It addresses a common element in policy attempts to contain healthcare costs, which is to encourage greater coordination in healthcare delivery. Despite a clear trend toward greater hospital‐physician integration, there is little direct evidence about whether integration lowers hospital costs. The results in this paper show that hospital costs increase by one to three percent after hospital‐physician integration. We also do not find consistent evidence that hospital‐physician integration is associated with higher quality but potentially more costly hospital care. The modest increase in hospital costs appears to derive from an increase in outpatient visits, rather than from higher costs of inpatient care. These findings do not support the hypothesis that increased coordination between hospitals and physicians has led to lower hospital costs.
 
Average service price time trends for broad healthcare types. These figures show the trend in weighted average service prices for each type of healthcare category. Weights are computed using 2011 utilization. In particular, for healthcare type x ∈ {All, Outpatient, Prescriptions, Inpatient} and period t ∈ {2011, 2013}, the average service price of healthcare type x in month m of year t is AverageServicePricextm=∑s∈xClaimsst=2011Claimsxt=2011×CoststmClaimsstm $AverageServicePric{e}_{\mathit{xtm}}={\sum }_{s\in x}\frac{Claim{s}_{st=2011}}{Claim{s}_{xt=2011}}\times \frac{Cos{t}_{\mathit{stm}}}{Claim{s}_{\mathit{stm}}}$
Predicted change in share of type I diabetics in adherence with recommended testing frequency. Plotted are the 95% confidence intervals of the marginal effects of the formulary expansion. Marginal effects are computed using estimates from specifications of Equation (1) using binary indicators for being up to date with lab testing as outcome variables. Patients are up to date with cholesterol and kidney disease testing if they receive at least one lab test annually, blood glucose testing if they receive at least four tests annually, and A1C testing if they receive at least two tests annually
Article
This paper estimates the causal effect of the expansion of Colombia's national prescription drug formulary to include five new types of insulin on the healthcare utilization and costs of type I diabetics and explores the mechanisms through which outpatient cost reductions are realized. We find that expanded coverage generates an increase in the cost of insulin for type I diabetics equal to 17% of their baseline healthcare costs. At the same time, their annual outpatient care utilization falls by 1.9 claims. We devise tests to explore the relative importance of two mechanisms by which the expansion may have lowered type I diabetics' non‐drug healthcare utilization: spillovers from drug to non‐drug spending and rationing of care. We find no evidence that the formulary expansion reduces the rate of complications from diabetes and find substantial declines in non‐drug costs even among the subset of diabetics with no scope for spillovers. We find large reductions in the utilization of discretionary care including diagnostic tests, but no such declines for the use of essential drugs, suggesting that rationing of care is the primary driver of observed cost savings.
 
Timeline of intervention implementation activities
Participation in the 2012 and 2014 surveys
Article
Health insurance enrollment in many Sub‐Saharan African countries is low, even with highly subsidized premiums and exemptions for vulnerable populations. One possible explanation is low service quality, which results in a low valuation of health insurance. Using a randomized control trial in 64 primary health care facilities in Ghana, this study assesses the impact of a community engagement intervention designed to improve the quality of healthcare and health insurance services on households living nearby the facilities. Although the intervention improved the medical‐technical quality of health services, our results show that households' subjective perceptions of the quality of healthcare and insurance services did not increase. Nevertheless, the likelihood of illness and concomitant healthcare utilization reduced, and especially households who were not insured at baseline were more likely to enroll in health insurance. The results show that solely increasing the technical quality of care is not sufficient to increase households' subjective assessments of healthcare quality. Still, improving technical quality can directly contribute to health outcomes and further increase health insurance coverage, especially among the previously uninsured.
 
Concentration curves of total National Health Insurance Fund (NHIF) claim amounts by policyholder. The 45° line would indicate that each policyholder accounts for the same amount of costs.
Concentration curves of total National Health Insurance Fund (NHIF) claim amounts by health facilities. The 45° line would indicate that each health facility accounts for the same amount of costs
Article
Many low‐income countries are in the process of scaling up health insurance with the goal of achieving universal coverage. However, little is known about the usage and financial sustainability of mandatory health insurance. This study analyzes 26 million claims submitted to the Tanzanian National Health Insurance Fund (NHIF), which covers two million public servants for whom public insurance is mandatory, to understand insurance usage patterns, cost drivers, and financial sustainability. We find that in 2016, half of policyholders used a health service within a single year, with an average annual cost of 33 US$ per policyholder. About 10% of the population was responsible for 80% of the health costs, and women, middle‐age and middle‐income groups had the highest costs. Out of 7390 health centers, only five health centers are responsible for 30% of total costs. Estimating the expected health expenditures for the entire population based on the NHIF cost structure, we find that for a sustainable national scale‐up, policy makers will have to decide between reducing the health benefit package or increasing revenues. We also show that the cost structure of a mandatory insurance scheme in a low‐income country differs substantially from high‐income settings. Replication studies for other countries are warranted.
 
Outcomes measured using the GP Patient Survey and Hospital Episode Statistics data
Article
Better integration is a priority for most international health systems. However, multiple interventions are often implemented simultaneously, making evaluation difficult and providing limited evidence for policy makers about specific interventions. We evaluate a common integrated care intervention, multi‐disciplinary group (MDG) meetings for discussion of high‐risk patients, introduced in one socio‐economically deprived area in the UK in spring 2015. Using data from multiple waves of the national GP Patient Survey and Hospital Episode Statistics, we estimate its effects on primary and secondary care utilization and costs, health status and patient experience. We use triple differences, exploiting the targeting at people aged 65 years and over, parsing effects from other population‐level interventions implemented simultaneously. The intervention reduced the probability of visiting a primary care nurse by three percentage points and decreased length of stay by 1 day following emergency care admission. However, since planned care use increased, overall costs were unaffected. MDG meetings are presumably fulfilling public health objectives by decreasing length of stay and detecting previously unmet needs. However, the effect of MDGs on health system cost is uncertain and health remains unchanged. Evaluations of specific integrated care interventions may be more useful to public decision makers facing budget constraints.
 
Survival proabilities for patients admitted during strike periods relative to patients admitted in other periods. Length of stay capped at 30 days. Admission during strikes includes physicians, nurses or DTT strikes
Relation between Hospital Quality and Physicians' strike impact (a)/Exposure to strikes (b); and relation between hospital occupancy rates and Physicians' strike impact (c)/Exposure to strikes (d). Each point represents one of the 46 hospitals included in this study. “Hospital quality” defined as the hospital fixed effects estimated based on the previous regression model during periods without strikes. “Occupancy rates for hospital beds” represents average occupancy rates in December 2018. “Strike impact” is the impact of exposure to physicians' strike on the admission day, estimated for each of the 46 hospitals. “Patients exposed to strikes” defined as the proportion of hospital admissions exposed to strikes on admission days
Article
Hospital strikes in the Portuguese National Health Service (NHS) are becoming increasingly frequent. This paper analyses the effect of different health professionals' strikes (physicians, nurses, and diagnostic and therapeutic technicians (DTT) ‐ DTT) on patient outcomes and hospital activity. Patient‐level data, comprising all NHS hospital admissions in mainland Portugal from 2012 to 2018, is used together with a comprehensive strike dataset with almost 130 protests. Data suggests that hospital operations are partially disrupted during strikes, with sharp reductions in surgical admissions (up to 54%) and a decline on both inpatient and outpatient care admissions. The model controls for hospital characteristics, time and regional fixed effects, and case‐mix changes. Results suggest a modest increase in hospital mortality limited for patients admitted during physicians' strikes, and a slight reduction in mortality for patients already at the hospital when a strike takes place. Increases in readmission rates and length of stay are also found. Results suggest that hospitals and legal minimum staffing levels defined during strikes are not flexible enough to accommodate sudden disruptions in staffing, regardless of hospital quality in periods without strikes.
 
Article
In this perspective, the assertion that race‐free risk assessment would harm patients of all races is critiqued from the viewpoint that race is not just another covariate in our arsenal. Although race may be associated with outcome, it is nevertheless a proxy for a myriad of other potential explanatory variables that could be genetic/biological but in many circumstances are more likely to be sociological/socioeconomic. It is argued that the pursuit of health maximization through the use of socially constructed variables like race must be done sensitively, recognizing that racial covariates in the medical arena can be subject to structural, institutional or personal biases. Even when such biases are thought to be minimized, the appearance of such bias may be sufficient to justify the removal of its use, particularly where employing a racial covariate could further increase existing disparities. While racial covariates may have descriptive value in helping to understand such disparities, it is beholden on the scientific community to explore alternatives to racial covariates that may provide the same or perhaps even better prognostic value in our analyses.
 
Article
We investigate the impact of the Great Recession in Italy on the incidence of chronic diseases using new individual longitudinal data from Electronic Health Records. We exploit the exogenous shock in the economic conditions occurred in 2008 to estimate heterogeneous effects of an unprecedented rise in local unemployment rates in an individual fixed‐effects model. Our results document that harsh economic downturns have a negative long‐lasting effect on cardiovascular disease and a temporary effect on depression. This effect is heterogeneous across gender, increases with age and is stronger right before the retirement age. An important policy recommendation emerging from this study is that prolonged economic downturns constitute an additional external risk for individual health and not a temporary benefit.
 
Distribution of the observed versus predicted utilities using the econometric techniques. ALDVMM, adjusted limited dependent variable mixture model; Betamix, mixture beta regression model; CLAD, censored least absolute deviation; GLM, generalized linear model; GLOGIT, generalized logistic regression; MR, median regression
Distribution of the observed versus predicted utilities using direct mapping with machine learning techniques. LASSO 1: LASSO technique is used for prediction. Explanatory variables (without interactions) are only considered. LASSO 2: LASSO technique is used for prediction. Explanatory variables and their two‐way interactions are considered. LASSO, least absolute shrinkage and selection operator; NN, neural networks; QRNN, quantile (median) regression neural networks
Distribution of the observed versus predicted utilities using indirect mapping with machine learning techniques. LASSO 1: LASSO technique is used for prediction. Explanatory variables (without interactions) are only considered. LASSO, least absolute shrinkage and selection operator; NN, neural networks; QRNN, quantile (median) regression neural networks
Article
Non‐preference‐based patient‐reported outcome measures (PROMs) are popular in health outcomes research. These measures, however, cannot be used to estimate health state utilities, limiting their usefulness for economic evaluations. Mapping PROMs to a multi‐attribute utility instrument is one solution. While mapping is commonly conducted using econometric techniques, failing to specify the complex interactions between variables may lead to inaccurate prediction of utilities, resulting in inaccurate estimates of cost‐effectiveness and suboptimal funding decisions. These issues can be addressed using machine learning. This paper evaluates the use of machine learning as a mapping tool. We adopt a comprehensive approach to compare six machine learning techniques with eight econometric techniques to map the Patient‐Reported Outcomes Measurement Information System Global Health 10 (PROMIS‐GH10) to the EuroQol five dimensions (EQ‐5D‐5L). Using data collected from 2015 Australians, we find the least absolute shrinkage and selection operator (LASSO) model out‐performed all machine learning techniques and the adjusted limited dependent variable mixture model (ALDVMM) out‐performed all econometric techniques, with the LASSO performing better than ALDVMM. The variable selection feature of LASSO was then used to enhance the performance of the ALDVMM in a hybrid model. Our analysis identifies the potential benefits and challenges of using machine learning techniques for mapping and offers important insights for future research.
 
Article
Driving under the influence of alcohol is a major cause of fatalities worldwide. There have been a range of legislative and policy interventions aiming to address this. Bar closing hours is one policy with clear implications for drink driving. Existing evidence, largely drawn from one‐off policy changes in urban settings, reports mixed evidence that is difficult to generalize. We return to this issue using a setting, Norway, that is advantageous due to large temporal and regional variation in closing times, frequent changes, and a lack of confounding policy changes. We demonstrate an average zero effect of closing hours on traffic accidents that masks large variations in effects: in terms of population density; accident severity; and direction of change in closing hours. Extensions in closing hours in populous municipalities decrease accidents, whereas the opposite is true for rural municipalities. Our findings suggest that estimates from single policy changes may be difficult to generalize, while demonstrating that closing hours can generate large effects on traffic accidents.
 
Article
This is the first study to comprehensively examine the impact of job losses during the Great Recession on mental health, physical health, health behavior, and risky health behavior of young adults (ages 18–27). We employ U.S. longitudinal data with individual fixed effects to control for time‐invariant factors that may bias the results. We find that job losses during the recession of young adults living by themselves led to increased onset of doctor‐diagnosed mental health problems and worries related to jobs. Poorer individuals suffered more from increased worries, obesity, and binge drinking. In contrast, for those living with their parents, job loss of young adults did not negatively affect their own health. Instead, fathers' job losses led to worse mental health, physical health, and health behavior for young adults. Overall, the results suggest that when living on their own, young adults were responsible for their households' livelihood, and consequently, own job losses led to stress and negative health outcomes. However, when living with parents, they were financially reliant on their parents. Therefore, own job losses did not affect health, but job losses of fathers, the primary income earners for most households, worsened the health of young adults.
 
Example of the choice context and attributes
Article
Non‐pecuniary sources of motivation are a strong feature of the health care sector and the impact of competitive incentives on behavior may be lower where pecuniary motivation is low. This paper measures the marginal utility of income (MUY) of physicians from a stated‐choice experiment, and examines whether this measure influences the association between competition faced by physicians and the prices they charge. We find that physicians are more likely to exploit a lack of competition with higher prices if they have a high MUY.
 
Sample selection and treatment noncompliance
APPROACH project counties in Guizhou province
Article
Health care in China suffers from substantial allocative inefficiency in the delivery system and technical inefficiency within hospitals. To ameliorate this problem in rural areas, the Analysis of Provider Payment Reforms on Advancing China's Health (APPROACH) project shifted the payment method of China's rural health insurance scheme for county hospitals from fee‐for‐service to a novel global budget. In particular, APPRAOCH global budget incentivized system‐level allocative efficiency by reimbursing county hospitals at higher tariffs for gatekeeping and averting out‐of‐county (OOC) admissions among local patients they could treat. APPROACH conducted a large‐scale randomized controlled trial of the global budget in 56 counties (22 million enrollees) of Guizhou province during 2016–2017. Applying randomization inference to claims data, we find a significant shift of inpatient utilization and expenditure from OOC hospitals to county hospitals. At county hospitals, average expenditure per admission and length of stay decreased, though not significantly. Effects on readmissions show no clear sign of compromised quality. We further find limited effect heterogeneity with respect to treatment and hospital characteristics. Overall, APPROACH global budget may offer a framework for improving health care efficiency without sacrificing quality.
 
Article
This paper documents how substance abuse treatment (SAT) providers and services respond to increases in population‐level opioid addiction. I do this by exploiting the implementation of Medicare Part D as an exogenous increase in the availability of prescription opioids. Starting in 2006, states with higher shares of the population eligible for Medicare Part D experienced increases in residential and hospital inpatient SAT facilities, beds dedicated to SAT, and SAT facilities offering medication‐assisted treatment, relative to states with lower shares. These results suggest that the supply of SAT in the United States is capable of responding significantly to changes in demand.
 
Google trends of the term “Parto Cesárea” (Caesarean section) in São Paulo. Source: Google trends
Overall (a), Primary (b), and Repeated (c) C‐section rates in São Paulo (bold line) compared to synthetic São Paulo (dashed line). The vertical dashed line denotes when Law 17,137 was implemented
Article
Caesarean section (C‐section) rates continue to rise globally. Yet, there is little consensus about the key determinants of rising C‐section rates and the sources of variation in C‐section rates across the world. While C‐sections can save lives when medically justified, unnecessary surgical procedures can be harmful for women and babies. We show that a state‐wide law passed in São Paulo (Brazil), which increased women's autonomy to choose to deliver via C‐section even when not medically necessary, is associated with a 3% increase in overall C‐section rates. This association was driven by a 5% increase in primary C‐sections, rather than repeated C‐sections. Since the law emphasizes women's autonomy, these results are consistent with mothers' demand being an important contributor to high C‐section rates in this context.
 
Article
We study whether primary care physicians (PCPs) exercise left digit bias with respect to patients' age. Relying on a comprehensive administrative visit level data from a large Israeli HMO, we measure the intensity of patients' medical examination in visits that take place around a decadal birthday—a birthday that ends with zero—within a regression discontinuity framework. We find that in standard settings with clear patient information there is no evidence that PCPs exhibit left digit bias. However, when PCPs meet unfamiliar patients seeking immediate care, they are more likely to use basic diagnostic tests just above the decadal birthday threshold, indicating that under these circumstances, PCPs do use left digit bias.
 
Research Design. This figure depicts our difference‐i n‐differences (DiD) research design. We select (11) treatment and control state matched pairs with each pair exhibiting similar pre‐treatment time trends and covariate values. We fit individual state time trends for each state in the sample. Following the treatment date, we fit (2) additive post ‐treatment trends, one trend for the set of treatment states and one trend for the set of control states. The overall treatment effect of reform is computed as the difference in the two post‐treatment trends. Controls for nursing home covariates and other controls (e.g., overall year dummies) are implied and suppressed in the graph. The graph depicts the research design for a single pair of states
Nursing Care in Treatment and Control States. We depict trends in total direct care and nursing assistants in both treatment (reform) and control (non‐reform) states. Separate comparison trends are depicted for NHs affiliated with a large chain and NHs that are independent or affiliated with a small chain. Within the sample of large chain affiliated nursing homes it appears that inputs declined and converged to the levels in control states in the time period around reform (2003–2006). This is true for both direct care and CNA care. Within the sample of SC/I homes there is no such convergence. It is also worth noting that within the sample of LC‐affiliated homes that inputs are strictly greater in treatment states prior to reform. Within the sample of SC/I homes, inputs are strictly lower in treatment states throughout the sample period
RN and LPN Nursing Care in Treatment and Control States. We depict trends in RN care and LPN care in both treatment (reform) and control (non‐reform) states. Separate comparison trends are depicted for NHs affiliated with a large chain and NHs that are independent or affiliated with a small chain. Within both samples it appears that NHs in treatment states substituted LPN for RN care across all years as evidenced by fewer RN inputs and more LPN inputs relative to NHs in control states. It is unclear (from the charts), whether the extent of this substitution grew or contracted over time, after tort reform
Citations and Infections in Treatment and Control State. We depict trends in total citations and the rate of UTI infections in both treatment (reform) and control (non‐reform) states. Separate comparison trends are depicted for NHs affiliated with a large chain (LC) and NHs that are independent or affiliated with a small chain (SC/I). Within the sample of large chain‐affiliated nursing homes it appears that the trends in citations declined below that of control states sometime after the reform period. For infections, there is less of an apparent trend in the infection rates between control and treatment states. Within the sample of SC/I‐affiliated homes, there is no apparent divergence in the trend in total citations. Conversely, there is an increase in the relative rate of infections in treatment states
Restraints and Falls in Treatment and Control States. We depict trends in the use of restraints and falls in both treatment (reform) and control (non‐reform) states. Separate comparison trends are depicted for NHs affiliated with a large chain (LC) and NHs that are independent or affiliated with a small chain (SC/I). Within both samples, it is apparent that the use of restraints declined across all categories of nursing homes and that relatively speaking they declined at a greater rate in treatment states. This pattern holds among both LC and SC/I samples. Likewise the rate of falls increased over time, especially within the sample of LCs. Here the relative rate of increase appears to be greater in reform relative to the control states. Within the SC/I sample, the rate of falls increases in the second half of the sample period for both the reform and control groups
Article
We provide time series evidence of tort reform's impact on inputs and quality in the nursing home industry. Between 2000 and 2010, 11 state reforms capped noneconomic damages for health care services. Small chain and unaffiliated nursing homes enjoyed “judgment proof standing” and were less apt to be sued, prior to reform. We find that the managers of such homes were relatively unresponsive to the implementation of state caps on noneconomic damages. Large “deep‐pocketed” chain‐affiliated homes lacked judgment proof standing and implemented greater reductions in their nursing inputs in the aftermath of tort relief. However, we find little evidence of service quality erosion across four measured dimensions of care outcomes. Our findings are consistent with a “defensive care” model in which large chain homes employ unproductive inputs in an effort to meet a negligence standard of care.
 
Cantonal averages of the yearly variation in price components and registered staffing ratios across nursing homes. The number of registered nurses and care prices are weighted by case‐mix. Prices are deflated using consumer price indices (Federal Statistical Office (2021), base year 2012). The size of the circles reflects the underlying number of nursing homes in the respective canton. Cantons: AG=Aargau, AR: Appenzell Ausserrhoden, BE: Bern, BL: Basel‐Landschaft, BS: Basel‐Stadt, GR: Grisons, JU: Jura, SG: St. Gallen, SO: Solothurn, TG: Thurgau, VD: Vaud, VS: Valais. Source: FOPH (2021), years 2012–2017, own calculations
Article
Many countries limit public and private reimbursement for nursing care costs for social or financial reasons. Still, quality varies across nursing homes. We explore the causal link between case‐mix adjusted nurse staffing ratios as an indicator of care quality and different price components in Swiss nursing homes. The Swiss reimbursement system limits and subsidizes the care price at the cantonal level, which implicitly limits staffing ratios, while the residents cover the nursing home‐specific lodging price privately. To estimate causal effects, we exploit (i) the exogeneity of the Swiss care price regulation, (ii) nursing‐home fixed effects estimations and (iii) instrumental variables for the lodging price. Our estimates show a positive impact of prices on certified staffing ratios. We find that a 10% increase in care prices increases certified staffing ratios by 3–4%. A comparable 10% increase in lodging prices raises certified staffing ratios by 1.5–10% (depending on the model). Our findings highlight that price limits for nursing care impose a limit on staffing ratios. Furthermore, our results indicate that providers circumvent price limits by increasing lodging prices that are privately covered. Thus, this cost shifting implicitly shifts the financial burden to the residents.
 
Article
In this study, we find that children in Vietnam who were born in December of a given year have better health outcomes than those born in January of the following year. Children born in December are taller, heavier, and less likely to be underweight and suffer from stunted growth than those born in January of the following year, though these two groups of children differ in age by only 1 month. We argue that the effect of being born in December compared to January on children's health is translated through early preschool attendance. In Vietnam, children born in December are more likely to start preschool as well as primary school 1 year earlier than those born in January of the following year. Thus, the health benefit for a child born in December would come from earlier and longer exposure to preschool. Importantly, we find that the positive effect of preschool persists over time as children grow. A possible major reason why preschool attendance improves health is the nutrition provided for children in preschools.
 
Mean utilities for all six health states elicited in this experiment
Article
Time trade‐off utilities have been suggested to be biased upwards. This bias is a result of the method being applied assuming linear utility of life duration, which is violated when individuals discount future life years or are loss averse for health. Applying a “corrective approach”, that is, measuring individuals' discount function and loss aversion and correcting time trade‐off utilities for these individual characteristics, may reduce this bias in utilities. Earlier work has developed this approach for time trade‐off in a student sample. In this study, the corrective approach was extended to composite time trade‐off (cTTO) methodology, which enabled correcting utilities for health states worse than dead. In digital interviews a sample of 150 members of the general public completed cTTO tasks for six health states, and afterward they completed measurements of loss aversion and discounting. cTTO utilities were corrected using these measurements under multiple specifications. Respondents were also asked to reflect on and adjust their cTTO utilities directly. Our results show considerable loss aversion and both positive and negative discounting were prevalent. As predicted, correction generally resulted in lower utilities. This was in accordance with the direction of adjustments made by respondents themselves.
 
Dynamic treatment effects by age. This figure presents the dynamic treatment effects by age using the interaction‐weighted (IW) estimations proposed by Sun and Abraham (2021) along with the 95% CI. The reference line denotes age equal to 62. All models include marital status, education years, race, gender, total assets (ln), veteran status, and cohort fixed effects. Standard errors are clustered by individual
Dynamic treatment effects by cohort. This figure presents the dynamic treatment effects by cohort using the interaction‐weighted (IW) estimations proposed by Sun and Abraham (2021) along with the 95% CI. The reference line denotes birth year equal to 1938. All models include marital status, education years, race, gender, total assets (ln), and veteran status. Standard errors are clustered by individual
Article
Governments are under pressure to raise the retirement age in response to an aging population and low fertility rates. However, the literature has not reached a consistent conclusion on the health effects of extending working lives. Furthermore, while most studies have concentrated on post‐retirement health consequences, the health outcomes during the transition from work to retirement have been overlooked. Therefore, this article focuses on the transition period — the time between the early benefit age and full retirement age. Exploiting the increase in retirement age introduced by the U.S. Social Security Amendments of 1983, the difference‐in‐difference estimation finds that the reform successfully encouraged more people to work longer and claim benefits later, whilst having no adverse influence on health during the transition period. This paper infers that the desirable impacts of the 1983 amendments could be partly attributed to the adequate preparation time the reform left to the public.
 
Article
Mental health disorders are among the leading causes of disease burden worldwide. Recently, attention has been drawn to the Internet and social media as determinants of the increase in mental health conditions in recent years. In this paper, I analyze the causal effect of broadband Internet access on the mental health of adults. I leverage confidential information on the coordinates of respondents to the German Socio‐Economic Panel (GSOEP) and exploit technological features of the German telecommunication network to instrument for broadband Internet access. The results are suggestive that broadband Internet leads to worse mental health for women (primarily those aged 17–30) but not for men, thus widening the gender gap in mental disorders. Looking at sub‐facets of mental health, broadband access leads to a worsening of socializing behavior and ability to cope with emotional problems. The fact that the results are concentrated among the younger cohorts of women is suggestive that high Internet usage intensity amplifies the negative effect of broadband internet access on mental health.
 
Parents' psychological distress score by COVID‐related job loss. This figure shows the mean K10 score of the parents at baseline and the COVID‐19 follow up by job loss. K10 scores are the Kessler 10 scores
Children's’ life satisfaction score by COVID‐related job loss. This figure shows the mean Huebner's student life satisfaction score of the children in our sample at baseline and the COVID‐19 follow up by job loss
Children's Hope score by COVID‐related job loss. This figure shows the mean “Hope” score of the children in our sample at baseline and the COVID‐19 follow up by job loss. “Hope” is an index, calculated by adding the respondents score for following four statements: (i) they feel positive about their future, (ii) if they try hard, they can improve their situation in life, (iii) they like to make plans for their future studies and work, (iv) they have opportunities to develop job skills. The statements are scored from 1–6, with a score of 1 if they strongly disagree and a score of 6 if they strongly agree. A higher score indicates greater ‘hope’
Children's mean depression score by COVID‐related job loss. This figure shows the mean depression score of the children in our sample at baseline and the COVID‐19 follow up by job loss. Depression is the score of the respondents on how often they have noticed their child getting depressed during the lockdown scored from 1 to 5. A higher score indicates higher frequency with which the parent has noticed their child getting depressed
Article
We combine data collected just prior to the unfolding of COVID‐19 with follow‐up data from July 2020 to document the adverse economic effects of the pandemic and resulting impact on parental and child mental well‐being in peri‐urban Pakistan. 22% of the households in our sample are affected by job loss, with monthly income down 38% on average. Our difference‐in‐difference results show that job loss is associated with a 0.88 standard deviation (SD) increase in adult mental distress scores (K10), a 0.43 SD reduction in a Hope index of children's aspirations, agency and future pathways, and a 0.40 SD increase in children's depression symptoms. In addition, we observe higher levels of parental stress and anger reported by children, as well as an increase in reported prevalence of domestic violence. Overall, we document that the pandemic has disproportionately and negatively affected the economic and mental well‐being of the most vulnerable households in our sample.
 
Trends in secondary school enrollment and transition for Zimbabwe. The estimates here are based on data from Riddell & Nyagura, 1991, which in turn are curated from the UN statistical yearbooks from 1970 to 1988. The transition rate is the percentage of students who graduated from grade seven (highest grade in primary school) who end up enrolling in grade eight (secondary school). Secondary school enrollment is measured in thousands of students.
Effects of the Reform on Education (First Stage effects). Authors' estimate based on Word Health Survey data. The y‐axis represents the highest grade attained by individuals in our samples and the x‐axis represents the age in 1980 when the reform was implemented. All estimations include gender, living in rural area, fixed effects for region and survey rounds. Standard errors are clustered by the age of respondent in 1980.
Values of mental health indices by treatment status. Based on Zimbabwe World Health Survey data. The figures plot the distribution of our measures of mental health: depression and anxiety indices, which measure the severity of symptoms related to depression and anxiety, in the top panels, and probability of having depression‐ or anxiety‐related symptoms in the bottom panels, by treatment status. Treated includes individuals aged 15 years and younger in 1980 and the untreated group consists of individuals aged 16 years and older in 1980.
The impact of education on mental health. Based on author calculations using World Health Survey Zimbabwe. The figures plot different measures of mental health against the running variable (age in 1980). The upper panels represent depression index (left) which measures the severity of the symptoms and an indicator variable for having any depression‐related symptoms (right). The lower panels presents graphs for similar outcomes pertaining to anxiety‐related symptoms.
Heterogeneous impacts of education on mental health. The definitions of the mental health measures are the same as those used in Table 3. The sample includes women who were between the ages of 0 and 30 in 1980. All specifications exclude those who were 14 and 15 years old in 1980, and control for categorical variables for living in a rural area, fixed effects for survey round, region, linear age trends, and rainfall/temperature shock in the year of birth. Results are based on data from World Health Survey from Zimbabwe. Standard errors are clustered by the age of the respondent in 1980.
Article
We analyze the role of education as a determinant of mental health. To do this, we leverage the age‐specific exposure to an educational reform as an instrument for years of education and find that the treated cohorts gained more education. This increase in education had an effect on mental health more than 2 decades later. An extra year of education led to a lower likelihood of reporting any symptoms related to depression (11.3%) and anxiety (9.8%). More educated people also suffered less severe symptoms – depression (6.1%) and anxiety (5.6%). These protective effects are higher among women and rural residents. The effects of education on mental well‐being that we document are potentially mediated through better physical health, improved health behavior and knowledge, and an increase in women's empowerment.
 
Article
Until recently, there has been a consensus that clinicians seeking to assess patient risks of illness should condition risk assessments on all observed patient covariates with predictive power. The broad idea is that knowing more about patients enables more accurate predictions of their health risks and, hence, better clinical decisions. This consensus has recently unraveled with respect to a specific covariate, namely race. There have been increasing calls for race‐free risk assessment, arguing that using race to predict health risks contributes to racial disparities and inequities in health care. In some medical fields, leading institutions have recommended race‐free risk assessment. An important open question is how race‐free risk assessment would affect the quality of clinical decisions. Considering the matter from the patient‐centered perspective of medical economics yields a disturbing conclusion: Race‐free risk assessment would harm patients of all races.
 
Proportional Hazards (PH) assumption test for having a child with cerebral palsy (CP) in divorce risk analysis. CP this figure shows the log‐log survival curves for the treatment and comparison groups (i.e., having a child without CP or with CP on the left, and no CP, mild CP or severe CP on the right) regarding the risk of the divorce/separation. The hazard curves are adjusted for all covariates from the full specification (see columns 3 and 4 in Table 2). A parallel development of the plots indicates that the PH assumption is fulfilled. Mild CP encompasses individuals at gross motor function classification system (GMFCS) levels I‐III or individuals with subtypes of CP that are associated with lower GMFCS‐levels whereas severe CP encompasses individuals whose gross motor function is more severely affected at GMFCS levels IV‐V or individuals with subtypes of CP that are associated with higher GMFCS‐levels.
Smoothed hazard function and cumulated failure estimates of divorce. Cerebral palsy (CP). This figure shows smoothed hazard rates and cumulated failure estimates of divorce/separation for parents of a child with no CP, mild CP, or severe CP, based on the analysis in column 4, Table 2. The smoothed HR on the left show the likelihood of divorce/separation across the severity groups in each period, given that a divorce did not happen in any previous period. The cumulated failure rates display the proportion of divorced parents across time. Mild CP encompasses individuals at gross motor function classification system (GMFCS) levels I‐III or individuals with subtypes of CP that are associated with lower GMFCS‐levels whereas severe CP encompasses individuals whose gross motor function is more severely affected at GMFCS levels IV‐V or individuals with subtypes of CP that are associated with higher GMFCS‐levels.
Proportional Hazards (PH) assumption test for having a child with cerebral palsy (CP) in likelihood of additional children analysis. CP this figure shows the log‐log survival curves for the treatment and comparison groups (i.e., having a child without CP or with CP on the left, and no CP, mild CP or severe CP on the right) regarding the likelihood of having additional children. The hazard curves are adjusted for all covariates from the full specification (see columns 3 and 4 in Table 2). A parallel development of the plots indicates that the PH assumption is fulfilled. The decreasing hazards show that the likelihood of having additional children decreases over time. Mild CP encompasses individuals at gross motor function classification system (GMFCS) levels I‐III or individuals with subtypes of CP that are associated with lower GMFCS‐levels whereas severe CP encompasses individuals whose gross motor function is more severely affected at GMFCS levels IV‐V or individuals with subtypes of CP that are associated with higher GMFCS‐levels.
Smoothed hazard function and cumulated failure estimates of having additional children cerebral palsy (CP) this figure shows smoothed hazard ratios|hazard rates (HR) and cumulated failure estimates of divorce/separation for parents of a child with no CP, mild CP, or severe CP, based on the analysis in column 4, Table 5. The smoothed HR on the left show the likelihood of having an additional child after the birth of the child in our sample with no CP, mild CP, or severe CP in each period, given that the parents did not had an additional child in any previous period. The cumulated failure rates display the proportion of divorced parents across time. Mild CP encompasses individuals at gross motor function classification system (GMFCS) levels I‐III or individuals with subtypes of CP that are more likely to be associated with lower GMFCS‐levels whereas severe CP encompasses individuals whose gross motor function is more severely affected (GMFCS IV‐V) and who require wheelchairs or individuals with subtypes of CP that are more likely to be associated with higher GMFCS‐levels.
Article
This study analyzes the relationship of having a child with the early‐onset disability cerebral palsy (CP) and the parental decision to divorce and to have additional children. We use longitudinal matched case‐control data from multiple linked Swedish National Population Registers between 2001 and 2015 and perform Cox proportional hazards regressions with interval‐censoring. Although we do not find a general excess parental divorce risk on CP relative to the comparison group without CP, we find that having a child with CP increases the risk of divorce for parents with low education. We also find that having a child with CP reduces the likelihood of having additional children, especially for mothers in the older age range (maternal age at delivery >33 years) and parents with low education. The severity level of the disability, as indicated by gross motor function, is not related to the results. These findings should be understood in the Swedish context, which provides extensive welfare support (e.g., personal assistance). If future studies would find adverse results in countries with less social care and benefits, our results may indicate that it is possible to mitigate negative consequences for the family of a child with disability.
 
Mobility Indicators in Argentina for 2020. The figure plots indicators of spatial mobility. Walking and driving data were obtained through the Apple Mobility Trends Report, using baseline volume from January 13, 2020. The public transit indicator comes from Moovit, using baseline volume in the week of January 15, 2020. An indicator with value of 100 means that mobility on that day was the same as the reference date
Event‐study Analysis. The green line shows the simple average of walking, driving, and public transit mobility indicators shown in Figure 1. Blue dots correspond to the point estimates obtained using Equation (1), and blue bars show the associated 95 percent confidence intervals. The vertical dashed line marks week 11, when mobility restrictions were first imposed. Weeks 6–8 are missing because telemedicine calls were not recorded those weeks in 2019. Week 17 is missing because telemedicine calls were not recorded that week in 2020
Treatment Effects: Heterogeneity. Panel (a) shows ordinary‐least‐squares estimates of θ in Equation (2) for the different subgroups specified on the x‐axis. “Baseline” reports the main results of Table 2. (For reference we place a horizontal dotted bar at that level.) “General Medicine”, “Ob/Gyn,” and “Pediatrics” show the effect on log (calls) related to general medicine, obstetric or gynecological care, and pediatric consultations. The categories labeled “< 18,” “18–24,”“25–39,”“40–54,” and “55–64” show the estimated effect on people in those age categories. “Male” and “Female” estimate the effect for patients of either sex. Finally “Preexist. Condit.” shows the estimated increase in calls by patients with a medical condition preexisting at the time of the call. Panel (b) depicts the same estimates when the outcome is log (number of first‐time callers). Blue bars report the 95 percent confidence intervals
Article
Telemedicine can expand access to health care at relatively low cost. Historically, however, demand for telemedicine has remained low. Using administrative records and a difference‐in‐differences methodology, we estimate the change in demand for telemedicine experienced after the onset of the COVID‐19 epidemic and the imposition of mobility restrictions. We find that the number of telemedicine calls made during the pandemic increased by 230 percent compared to the pre‐pandemic period. The effects were mostly driven by older individuals with preexisting conditions who used the service for internal medicine consultations. The demand for telemedicine remained relatively high even after mobility restrictions were relaxed, which is consistent with telemedicine being an “experience good.” These results are a proof of concept for policy makers to use such relatively low‐cost medical consultations, made possible by new technologies, to provide needed expansion of access to health care.
 
Distribution of the GHQ score
Uber UK entry date by area
Correlation between Uber diffusion and the GHQ score. The figure presents the coefficients on Uber diffusion from a set of regressions (for each SOC) of the GHQ score on a dummy for Uber diffusion, controlling for female, white, age, college, household income (/1000), share of workers who are female/have a college degree/are in each age group, the mean household income of workers in the TTWA‐year, population (ln) in the TTWA‐year, and year and TTWA fixed effects. The 2‐digit SOC codes are used, and in addition, we distinguish between self‐employed and salaried drivers in SOC 8214. Standard errors are clustered at the TTWA level. *p < 0.10, **p < 0.05, ***p < 0.01
Article
While the spread of digital technologies and the growth of associated atypical forms of work are attracting increasing attention, little is known about the impact of these new forms of work on psychological well‐being. This paper examines the effect of Uber diffusion on the mental health of drivers, taking advantage of the rollout of Uber across UK regions. We match individual‐level information on health and sociodemographic characteristics from the UK Household Longitudinal Study (Understanding Society) between 2009 and 2019 with data on the diffusion of Uber across the country. We first show that Uber diffusion is positively associated with mental health, as measured by the General Health Questionnaire, in the population group of self‐employed drivers. We argue that this positive correlation captures a selection effect (of comparatively healthier individuals into the category of self‐employed drivers after Uber entry) and the omission of unobserved factors, rather than a causal effect. Indeed, we do not observe any improvement in mental health for workers who were already self‐employed drivers before Uber entry. In parallel with this, among individuals who remained salaried drivers over time, our results suggest there may be a decline in mental health after Uber's introduction, probably because they feel the competition from Uber drivers.
 
Variation in coded severity within patients. Coded severity defined as Hierarchical Condition Category (HCC) score. Graph to illustrate variation in dependent variable. The average patient HCC score was 1.047. The average within‐panel range in a patient's HCC score from 2010 to 2015 was 0.987; the 95th percentile of this within‐panel range was 2.894
Event study of coded severity. This figure displays the event study specification of the effect of integration on the coded severity of an integrated physician's patients. The point estimates are the regression coefficients from Equation (2). The y‐axis displays the dependent variable (natural log of HCC scores). The x‐axis displays the time relative to when the physician became hospital‐integrated (t = 0 represents the first year of integration). These estimates imply that patients' risk scores increased between 2% and 4% after their primary care physician vertically integrated with a hospital
Event study of standard deviation in coded severity. This figure displays the event study specification of the effect of integration on the variability in risk scores within a physician's panel of patients. For each physician in each year, we calculated the standard deviation of his or her patient's Hierarchical Condition Category (HCC) scores. We then estimated a physician‐year level event study model to determine whether integration exerted an influence on the variability of HCC scores. The sample mean of standard deviation in the last pre‐period was 0.664. The point estimates are the regression coefficients from Equation (3). The y‐axis displays the dependent variable (natural log of HCC scores). The x‐axis displays the time relative to when the physician became hospital‐integrated (t = 0 represents the first year of integration). These estimates demonstrate that the variability of HCC scores within a physician's panel of patients (even holding the panel of patients constant) decreased modestly after the physician vertically integrated with a hospital
Article
Hospital‐physician integration has surged in recent years. Integration may allow hospitals to share resources and management practices with their integrated physicians that increase the reported diagnostic severity of their patients. Greater diagnostic severity will increase practices' payment under risk‐based arrangements. We offer the first analysis of whether hospital‐physician integration affects providers' coding of patient severity. Using a two‐way fixed effects model, an event study, and a stacked difference‐in‐differences analysis of 5 million patient‐year observations from 2010 to 2015, we find that the integration of a patient's primary care doctor is associated with a robust 2%–4% increase in coded severity, the risk‐score equivalent of aging a physician's patients by 4–8 months. This effect was not driven by physicians treating different patients nor by physicians seeing patients more often. Our evidence is consistent with the hypothesis that hospitals share organizational resources with acquired physician practices to increase the measured clinical severity of patients. Increases in the intensity of coding will improve vertically‐integrated practices' performance in alternative payment models and pay‐for‐performance programs while raising overall health care spending.
 
Estimates of the net benefit separation (NBS) comparing radiation therapy (RT) to control as a function of the WTP, λ. Cost‐effectiveness determination curves (CED) curves are provided for patient cohorts conditional on cancer stage and Charlson comorbidity index. Results for patients with stage I cancer are provided on the left, while those for patients with stage II cancer are on the right. Estimated NBS within the range of primary interest are denoted in black. Gray indicates estimated NBS outside of this region, provided to observe the behavior of the NBS
Estimates of the net benefit separation (NBS) comparing chemotherapy (CT) to control as a function of the WTP, λ. Cost‐effectiveness determination curves (CED) curves are provided for patient cohorts conditional on cancer stage and Charlson comorbidity index. Results for patients with stage I cancer are provided on the left, while those for patients with stage II cancer are on the right. Estimated NBS within the range of primary interest are denoted in black. Gray indicates estimated NBS outside of this region, provided to observe the behavior of the NBS
Article
To make informed health policy decisions regarding a treatment, we must consider both its cost and its clinical effectiveness. In past work, we introduced the net benefit separation (NBS) as a novel measure of cost‐effectiveness. The NBS is a probabilistic measure that characterizes the extent to which a treated patient will be more likely to experience benefit as compared to an untreated patient. Due to variation in treatment response across patients, uncovering factors that influence cost‐effectiveness can assist policy makers in population‐level decisions regarding resource allocation. In this paper, we introduce a regression framework for NBS in order to estimate covariate‐specific NBS and find determinants of variation in NBS. Our approach is able to accommodate informative cost censoring through inverse probability weighting techniques, and addresses confounding through a semiparametric standardization procedure. Through simulations, we show that NBS regression performs well in a variety of common scenarios. We apply our proposed regression procedure to a realistic simulated data set as an illustration of how our approach could be used to investigate the association between cancer stage, comorbidities and cost‐effectiveness when comparing adjuvant radiation therapy and chemotherapy in post‐hysterectomy endometrial cancer patients.
 
Effect of punitive prenatal substance use policies (PSUPs) on the log of outcomes. The rate of neonatal drug withdrawal syndrome was drawn from the 2008–2018 Healthcare Cost and Utilization Project (HCUP) FastStats. The proportion of births with low gestational age, low birth weight, and any prenatal care were drawn from the 2008–2019 National Vital Statistics System (NVSS) Natality Files and reflect conception years 2008–2018. Coefficient estimates and 95% confidence intervals are based on event study models that include PSUP lags and leads (Equation 2). Models are weighted by the number of births in a state‐year, include State and year fixed‐effects, and account for control variables as well as for the priority treatment PSUP. Always treated punitive states are dropped
Effect of priority substance use disorder (SUD) treatment prenatal substance use policies (PSUPs) on the log of outcomes. The rate of neonatal drug withdrawal syndrome was drawn from the 2008–2018 Healthcare Cost and Utilization Project (HCUP) FastStats. The proportion of births with low gestational age, low birth weight, and any prenatal care were drawn from the 2008–2019 National Vital Statistics System (NVSS) Natality Files and reflect conception years 2008–2018. Coefficient estimates and 95% confidence intervals are based on event study models that include PSUP lags and leads (Equation 2). Models are weighted by the number of births in a state‐year, include state and year fixed‐effects, and account for control variables as well as for the punitive PSUP. Always treated priority treatment states are dropped
Effect of punitive prenatal substance use policies (PSUPs) on the log of neonatal drug withdrawal syndrome, by subpopulation. The rate of neonatal drug withdrawal syndrome was drawn from the 2008–2018 Healthcare Cost and Utilization Project (HCUP) FastStats. Coefficient estimates and 95% confidence intervals are a linear combination estimates from the four intervention years from event study models (Equation 2). Models are weighted by the number of births in a state‐year, include state and year fixed‐effects, and account for control variables as well as for the priority substance use disorder (SUD) treatment PSUP. Always treated punitive states are dropped
Effect of priority substance use disorder (SUD) treatment prenatal substance use policies (PSUPs) on the log of neonatal drug withdrawal syndrome, by subpopulation. The rate of neonatal drug withdrawal syndrome was drawn from the 2008–2018 Healthcare Cost and Utilization Project (HCUP) FastStats. Coefficient estimates and 95% confidence intervals are a linear combination estimates from the four intervention years from event study models (Equation 2). Models are weighted by the number of births in a state‐year, include state and year fixed‐effects, and account for control variables as well as for the punitive PSUP. Always treated priority SUD treatment states are dropped
Article
We study the effect of punitive and priority treatment policies relating to illicit substance use during pregnancy on the rate of neonatal drug withdrawal syndrome, low birth weight, low gestational age, and prenatal care use. Punitive policies criminalize prenatal substance use, or define prenatal substance exposure as child maltreatment in child welfare statutes or as grounds for termination of parental rights. Priority treatment policies are supportive and grant pregnant women priority access to substance use disorder treatment programs. Our empirical strategy relies on administrative data from 2008 to 2018 and a difference‐in‐differences framework that exploits the staggered implementation of these policies. We find that neonatal drug withdrawal syndrome increases by 10%–18% following the implementation of a punitive policy. This growth is accompanied by modest reductions in prenatal care, which may reflect deterrence from healthcare utilization. In contrast, priority treatment policies are associated with small reductions in low gestational age (2%) and low birth weight (2%), along with increases in prenatal care use. Taken together, our findings suggest that punitive approaches may be associated with unintended adverse pregnancy outcomes, and that supportive approaches may be more effective for improving perinatal health.
 
Panel A illustrates the state‐level variation in recreational cannabis law (RCL) implementation among sample states for our study window (2011–2019). RCL implementation data were retrieved from the RAND Corporation's Opioid Policy Tools and Information Center . Displayed are the 11 RCL implementers in our sample and their implementation time across our study window. Michigan is shown in this panel because its RCL allowed for non‐dispensary cannabis possession in the last year of our study window. Our sample includes 1834 state‐quarter observations across 11 years of State Drug Utilization Data (SDUD) data. Panel B plots the total number of Medicaid enrollees, ages 21 and older, who lived in states with an active RCL within our time frame. Medicaid enrollment data were sourced from reports provided by the Centers for Medicare and Medicaid Services (CMS).
Event study estimates of recreational cannabis laws (RCLs)' impact on logged drug utilization – number of prescriptions written – per Medicaid enrollee. Estimates correspond to the baseline model described in Section 2.2. Point estimates are shown on period‐specific coefficient labels and standard errors can be found in Appendix Table A2. Coefficients are transformed to represent percentage changes in rates and standard errors are expanded using the delta method to account for nonlinear transformation of the outcome variable. Drug classes are displayed as the title for each subplot.
Article
The potential substitution of cannabis for prescription medication has attracted a substantial amount of attention within the context of medical cannabis laws (MCLs). However, much less is known about the association between recreational cannabis laws (RCLs) and prescription drug use. With recent evidence supporting substitution of cannabis for prescription drugs following MCLs, it is reasonable to ask what effect RCLs may have on those outcomes. We use quarterly data for all Medicaid prescriptions from 2011 to 2019 to investigate the effect of state‐level RCLs on prescription drug utilization. We estimate this effect with a series of two‐way fixed effects event study models. We find significant reductions in the volume of prescriptions within the drug classes that align with the medical indications for pain, depression, anxiety, sleep, psychosis, and seizures. Our results suggest substitution away from prescription drugs and potential cost savings for state Medicaid programs.
 
The upper and lower bounds of Ec ${E}_{c}$ and Wc ${W}_{c}$ illustrated
Illustrating the behavior of Wc ${W}_{c}$ and Ec ${E}_{c}$ for an arbitrary binary health variable
Article
The concentration index, including its normalization, is prominently used to assess socioeconomic inequalities in health and health care. Wagstaff's and Erreygers' normalizations or corrections of the standard concentration index are the most suggested approaches when analyzing binary health variables encountered in many health economics and health services research. In empirical applications of the corrected or normalized concentration indices, researchers interpret them similarly to the standard concentration index, which may be problematic as this ignores their underlying behaviors. This paper shows that the empirical bounds of the standard concentration index, including the corrected indices, depend not only on the sample size directly but also on the sampling weight. Notably, the paper highlights critical challenges for assessing and interpreting the popular Wagstaff's and Erreygers' corrected concentration indices with binary health variables. Specifically, it shows that it might be misleading, for example, to assess socioeconomic health inequalities using the magnitude of the “symmetric” Erreygers' corrected concentration index in the face of progressive improvements in the binary health variable. Also, Wagstaff's normalized concentration index may give a spurious “concentration” of the binary health variable among the rich or the poor in certain rare instances.
 
Density of days of illness. The figure plots the density of days of illness in a month reported by individuals between the ages of 15–65 years.
Seasonality in illness. The figure plots the average proportion of individuals, between the ages of 15–65 years, reporting positive illness days by month.
Seasonality in days of work. The figure shows the average days spent by males and females, between the ages of 15–65 years, on different activities by month.
Age‐gender division of labor. The figure compares the market and non‐market labor supply of individuals between the ages of 15–65 years across gender and age‐cohorts. The y‐axis plots the average number of days worked by an individual in wage‐labor, own‐farm, domestic and livestock activities in a month, by gender and across different age‐cohorts.
Article
In developing countries where medical infrastructure, service delivery systems, and the markets for health insurance are underdeveloped, one important mechanism to cope with the consequences of health shocks is the intra‐household substitution of labor. This paper studies the impact of short‐term illness shocks on labor supply and wage earnings of prime‐aged individuals using data from agricultural households in India. It also documents the compensating intra‐household labor supply responses of other non‐ill members of the household. We find that an illness shock reduces an individual's monthly wage earnings by 7.1% via the decline in the individual's days of employment in the labor market. Further, an illness shock to the household head causes a compensating increase in the wage labor supply of the wife. An illness shock to the wife, however, induces the household head to devote more time to domestic activities. The compensating labor supply responses are only partially able to insure the loss in total wage income of the household. Our results indicate that the gender‐based specialization of labor weakens in the event of an illness shock.
 
Journal metrics
36 days
Submission to first decision
16%
Acceptance rate
$4,700 / £3,150 / €3,900
APC
2.395 (2021)
Journal Impact Factor™
4.2 (2021)
CiteScore
Top-cited authors
Nancy Devlin
  • University of Melbourne
Mandy Ryan
  • University of Aberdeen
Karen Gerard
  • University of Southampton
Yan Feng
  • Queen Mary, University of London
Esther W. de Bekker-Grob
  • Erasmus University Rotterdam