Publications

Abstract Background: Declining physician density threatens access to primary care. Although physicians may adjust their labour supply in response to local shortages, little is known about the extent of these adjustments and the role of family circumstances in shaping them. Objective: To examine how French self-employed GPs adjust their labour supply to declining physician density and whether family circumstances explain observed gender differences. Methods: We analyse a representative 4.4% panel of 2898 self-employed French GPs observed from 2010 to 2017 (INSEE Permanent Demographic Sample matched with tax and territorial data). Labour supply is proxied by private practice income. Physician fixed-effects models relate income to local physician density, controlling for individual, household and territorial characteristics. Heterogeneity analyses consider gender and family composition.Results: A 10% decrease in GP density is associated with a 1.7-1.8% increase in private practice income (p < 0.01), indicating partial labour supply adjustment. Male GPs increase income by about 2.2% (p < 0.01), whereas no significant response is observed among female GPs overall. Among physicians without dependent children, however, responses are similar for men (+2.4%) and women (+1.9%) (both p < 0.01), suggesting that gender differences primarily reflect unequal family responsibilities during the child-rearing years rather than gender itself. Conclusions: Self-employed GPs partially adjust their labour supply to declining physician density, but these responses vary with family circumstances. Policies improving access to primary care should therefore complement workforce planning with measures that facilitate the reconciliation of professional and family responsibilities.
Keywords Self-employed GPs Physician shortage Labour supply Capacity constraints Panel
Publication HAL Google Scholar
Abstract We analyze earnings responses to six large payroll tax and income tax reforms in France. Our findings indicate full pass-through to workers when there is a strong and transparent link between contributions and expected benefits. In contrast, employer payroll taxes with no tax-benefit linkage exhibit limited pass-through to workers, while income tax nominally borne by employees show nearly full passthrough. Together with a meta-analysis of the literature, we interpret these results as empirical support for the long-standing hypothesis that tax-benefit linkage matters for the incidence of payroll taxes. In the absence of such linkage, our findings suggest that the individual-level incidence of payroll taxes aligns with their statutory incidence.
Keywords Tax-Benefit Linkage, Payroll Tax, Tax Incidence, Statutory incidence
Publication HAL Google Scholar
Abstract Objective. Eyewitness testimony strongly influences conviction and sentencing; yet false testimony is a leading cause of wrongful convictions. We investigate whether the truth-telling oath — common in legal systems — reduces eyewitness deception. Hypothesis. While prior work emphasizes eyewitness reliability, little is known about whether courtroom procedures like the oath increase truthfulness. Drawing on psychological and economics literature, we hypothesize that oath-taking acts as a commitment device that reduces deception. Method. In a 22 between-subjects experiment combining legal psychology and experimental economics, participants watched a video and answered questions under varying incentive and oath conditions. Performance-based incentives conflicted with honesty, allowing us to isolate the effect of the oath. Results. The oath reduced deception by 39.2% (95% CI [20.1, 59.5]) even under incentives to lie, without impairing truth-tellers’ performance. Conclusions. The courtroom oath is not merely symbolic — it plays a measurable and substantial role in enhancing the integrity of eyewitness testimony. Eyewitness testimony is the most powerful form of evidence in a court of law. Eyewitnesses affect both the odds of conviction and the severity of sentences of the guilty. But eyewitnesses also lie, and false testimony is the primarily cause of wrongful convictions. Most of the extant literature focuses on eyewitness reliability and credibility assessment, but very little is known about the efficiency of the main mechanism used in-field to foster eyewitness honesty: a solemn truth-telling oath—the most ancient and worldwide institution used in the solemn legal ceremony underpinning criminal cases. Herein we examine how the truth-telling oath actually affects the level of eyewitness deception. Using a controlled experimental test designed to address this question, we show that an eyewitness who is exogenously incentivized to lie and takes a solemn oath is significantly less likely to use deception. In contrast with the related literature focusing on the detection of lies, we show that an oath actually works to improve truth-telling. The oath is not just ceremonial, it plays a key role in improving efficiency within the court.
Keywords Eye-witness testimony, Truth-telling oath, Controlled experiment
Publication HAL Google Scholar
Abstract Because they play an increasingly important role in determining access to credit, credit scoring models are under growing scrutiny from banking supervisors and internal model validators. These authorities need to monitor the model performance and identify its key drivers. To facilitate this, we introduce the explainable performance (XPER) methodology to decompose a performance metric (e.g., area under the curve (AUC), [Formula: see text]) into specific contributions associated with the various features of a forecasting model. XPER is theoretically grounded on Shapley values and is both model-agnostic and performance metric-agnostic. Furthermore, it can be implemented either at the model level or at the individual level. Using a novel data set of car loans, we decompose the AUC of a machine-learning model trained to forecast the default probability of loan applicants. We show that a small number of features can explain a surprisingly large part of the model performance. Notably, the features that contribute the most to the predictive performance of the model may not be the ones that contribute the most to individual forecasts (Shapley additive explanation). Finally, we show how XPER can be used to deal with heterogeneity issues and improve performance. This paper was accepted by Kay Giesecke, finance. Funding: The authors thank the Institut Universitaire de France, the Autorité de contrôle prudentiel et de résolution Chair in Regulation and Systemic Risk, the HEC-Deloitte Chair on Artificial Intelligence for Business Innovation, the Excellence Initiative of Aix-Marseille University [A*MIDEX], and the French National Research Agency [Grants AMSE ANR-17-EURE-0020, Ecodec ANR-11-LABX-0047, and MLEforRisk ANR-21-CE26-0007] for supporting our research. Supplemental Material: The online appendices and data files are available at https://doi.org/10.1287/mnsc.2023.02025 .
Keywords Explainability, Credit scoring, Performance metrics, Shapley values
Publication HAL Google Scholar
Abstract The modernisation theory of regime change is often perceived to be a murky paradigm, lacking theoretical or empirical foundations. In response, we clarify the links between education and regime change.More specifically, we propose that education contributes indirectly to the collapse of autocratic regimes because educated people engage in non-violent (civil) resistance that reduces the effectiveness of the security apparatus. We empirically test the validity of this 'defanging effect' of education. We indeed find that the combination of high autocracy and high education levels tends to trigger non-violent campaigns, which in turn increases the likelihood of a regime change, likely to be associated with political liberalisation.
Keywords Regime change, Modernisation, Education, Democratisation, Civil resistance, Autocracy
Publication HAL Google Scholar
Abstract Preventive behaviors are crucial for controlling the spread of infectious diseases. Until now, most of the literature on the understanding of the willingness to comply with preventive behaviors at the individual level has focused on either one of those behaviors or studied several behaviors but independently. However, preventive behaviors might not be independent of each other, and the question of the relationship between these various behaviors deserves to be further investigated. The COVID-19 pandemic represents an interesting setting to study compliance with preventive behaviors when several prophylactic measures aiming to reduce the same infection risk are available. The aim of this study is to investigate how economic and social preferences may shape the relationship between three types of COVID-19 preventive behaviors among a representative sample of the French population: 1) compliance with restrictions on movement, 2) adherence to barrier gestures and 3) COVID-19 testing. Using a Latent Class Analysis, we identify four groups of individuals with diverging patterns of compliance with preventive behaviors, differing both in terms of intensity and types of prophylactic measures followed: individuals who apply all preventive behaviors, those who reject them all, individuals who do not respect restrictions on movement but still protect themselves and others by applying barrier gestures, and those who do not use barrier gestures but comply with restrictions on movement. Our results support the existence of a risk compensation process leading some individuals to tailor their menu of preventive behaviors until they reach the risk threshold they are willing to handle. The composition of the menu of preventive behaviors appears to be linked with individuals' economic and social preferences including risk and time preferences, prosociality, and interpersonal trust. Exploring heterogeneity in preventive behaviors may inform the design of targeted prevention and communication campaigns that are better tailored to achieve public health goals.
Keywords Infectious diseases, Economic and social preferences, Latent class analysis, Risk compensation, Preventive behaviors
Publication HAL Google Scholar
Abstract SkyRank is an open source Python framework designed to implement, compare, and visualize various Skyline ranking algorithm implementations, including dp-idp improved with dominance hierarchy, RankSky, CoSky, and DeepSky. While these methods have been independently introduced in the literature, SkyRank provides a unified, reproducible, and extensible platform to evaluate and visualize them interactively. SkyRank is designed for researchers, students, and practitioners working on multi-criteria decision making, Pareto dominance, and database preference queries. It includes a modular benchmarking backend, a LaTeX-compatible graph generator, and two graphical user interfaces (based on PyQt5 and Tkinter) for interactive data exploration.
Keywords Ranking, Open source, Algorithm, Dp-idp, RankSky, CoSky, DeepSky, Skyline
Publication HAL Google Scholar
Abstract SkyRank is an open source Python framework designed to implement, compare, and visualize various Skyline ranking algorithm implementations, including dp-idp improved with dominance hierarchy, RankSky, CoSky, and DeepSky. While these methods have been independently introduced in the literature, SkyRank provides a unified, reproducible, and extensible platform to evaluate and visualize them interactively. SkyRank is designed for researchers, students, and practitioners working on multi-criteria decision making, Pareto dominance, and database preference queries. It includes a modular benchmarking backend, a LaTeX-compatible graph generator, and two graphical user interfaces (based on PyQt5 and Tkinter) for interactive data exploration.
Keywords Ranking, Open source, Algorithm, Dp-idp, RankSky, CoSky, DeepSky, Skyline
Publication HAL Google Scholar
Abstract Honest and dishonest behaviors may both diffuse among the members of an organization. Knowing which of the two spreads faster is important because it impacts the extent to which managers will need to resort to other, potentially more costly solutions to curb dishonest behavior. Assessing empirically which of honest behavior or dishonest behavior spreads faster is challenging because this requires field measurements of social relationships and dishonest behavior of individual members, which poses both measurement and inference problems. We examine an original fine-grained data set from a large company that allows for identifying agents likely to be dishonest and interactions among employees while offering a natural experiment that circumvents the inference problems associated with identifying peer-to-peer diffusion. We find (1) that dishonest behavior diffuses, whereas honest behavior does not; (2) that diffusion likely operates through spreading information about opportunities for collusion; and (3) that policies that screen on dishonesty at hiring may be efficient to curb dishonest behavior in environments with high turnover. This paper was accepted by Lamar Pierce, organizations. Funding: R. Ferrali acknowledges support from the Mamdouha S. Bobst Center for Peace and Justice [Grant 2016-12-Bobst], the French National Research Agency [Grant ANR-17-EURE-0020], and Aix-Marseille University—A*MIDEX [Excellence Initiative]. Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2023.01981 .
Keywords Dishonesty, Natural experiment, Organization Behavior, Network analysis
Publication HAL Google Scholar
Abstract Background It is commonly believed that Africa largely evaded the worst of the COVID-19 pandemic, with fewer cases than other continents. However, regional comparisons that ignore differences in testing intensity may misrepresent dynamics. Studying the spread and case-fatality relationship during COVID-19 across WHO regions requires explicitly adjusting for time-varying test volumes. Methods We build a weekly panel dataset spanning May 2020 to December 2021 for the WHO regions: Africa, Eastern Mediterranean, South-East Asia, the Americas, Western Pacific, and Europe. Data on tests, confirmed cases, and COVID-19-attributed deaths were sourced from Our World in Data. We apply a novel metric that corrects for fluctuating test volumes to quantify week-to-week acceleration in infections and in mortality. We then compare the frequency, magnitude, and timing of these acceleration episodes across regions.Results Accounting for testing dynamics, we show that Africa exhibits multiple infectionacceleration episodes whose magnitude and frequency match those in other regions. Mortality accelerations in Africa closely follow infection surges, with an average lag of ten weeks. A positive correlation between infection acceleration in Africa and the Americas further indicates synchrony. These findings hold when using a larger secondary dataset of 140 countries. Conclusions Contrary to prevailing assumptions, Africa was not spared from the pandemic's severe dynamics. Infection surges were on par with those elsewhere and were followed by mortality accelerations. These results underscore that accounting for testing variability is essential to accurately assess pandemic progression, and they highlight the urgent need to strengthen surveillance and healthcare capacity across all regions.The relatively low number of reported COVID-19 cases and deaths in Africa has prompted debates about whether the continent was spared the worst of the pandemic, a phenomenon described by some as the African "puzzle" 1,2 or "paradox" 3 . Early media reports and research articles speculated that Africa's younger population, lower population density in rural areas, and prior experience with infectious diseases and their pharmaceutical treatments might have mitigated the severe impacts observed in other regions.However, emerging evidence from seroprevalence studies indicates that far more individuals in Africa were exposed to SARS-CoV-2 than is reflected in official surveillance data, especially during the pandemic's first 2 years. For example, the ratio of seroprevalence to confirmed cases has been estimated to be as high as 100:1 4 . This gap between seroprevalence estimates and reported cases grew as the pandemic continued 5 , suggesting substantial under-reporting in surveillance data. For example, while 6 mention low testing rates as a likely source of under-reporting in their discussion of the
Publication HAL Google Scholar
Abstract This study provides a life-course analysis of the relationship between self-employment, health, and health care use among individuals aged 50 and older in Europe. Using data from the Survey of Health, Ageing, and Retirement in Europe (SHARE), we apply first-difference and dynamic panel data models that go beyond standard approaches in mitigating endogeneity concerns. Our findings show that the self-employed enjoy better health at younger ages, consistent with a selection effect. In addition, they experience a steeper decline in physical health over time. We also document two distinct phases of health care use: during working life, the self-employed are more likely to be hospitalised, suggesting delayed care until acute needs arise; after retirement, the number of medical visits increases, consistent with a lower opportunity cost of care.
Keywords Self-employment, SHARE, Health care use, Health
Publication HAL Google Scholar
Abstract This study provides a life-course analysis of the relationship between self-employment, health, and health care use among individuals aged 50 and older in Europe. Using data from the Survey of Health, Ageing, and Retirement in Europe (SHARE), we apply first-difference and dynamic panel data models that go beyond standard approaches in mitigating endogeneity concerns. Our findings show that the self-employed enjoy better health at younger ages, consistent with a selection effect. In addition, they experience a steeper decline in physical health over time. We also document two distinct phases of health care use: during working life, the self-employed are more likely to be hospitalised, suggesting delayed care until acute needs arise; after retirement, the number of medical visits increases, consistent with a lower opportunity cost of care.
Keywords Self-employment, SHARE, Health care use, Health
Publication HAL Google Scholar
Abstract We study candidates' position adjustments in response to information about voters' preferences. Repositioning allows candidates to move closer to the median voter, but it incurs financial and electoral costs. In a subgame-perfect equilibrium, candidates diverge from the center ex ante if the costs of adjustment are sufficiently large. This allows them to increase the chances of a costless victory when the information is strongly in their favor. Our theory highlights a dynamic of moderation during the campaign stage in competitive elections, as well as a prominent role for minor adjustments made preemptively by the favored candidate. JEL Classification: C72, D72, D82 Model. We enrich the Downs-Hotelling framework by introducing an information shock, creating a two-stage game. The shock reveals the location of the median voter. This captures the idea that voters' aggregate preferences fluctuate over time and that their current leanings are disclosed during the electoral cam-This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made.
Keywords Re-positioning, Spatial voting, Imperfect information, Flip-flop
Publication HAL Google Scholar
Abstract Background: Breast cancer and its treatment may contribute to an increased risk of unemployment, influenced by both disease-related factors and socioeconomic determinant. Few longitudinal studies have examined employment outcomes among women diagnosed with cancer. This retrospective study investigated long-term employment among breast cancer survivors (BCS) and assessed disease specific and socioeconomic factors associated with employment. Design and methods: Registry-based data included working age BCS in Norway 2004–2008 alive at 6 years follow-up ( N = 3560). The employment status on each BCS was compared to two matched non-cancer controls ( N = 7081) by means of logistic regression analyses with marginal effects. Separate analyses by employment status at the time of diagnosis were conducted. Results: Among BCS employed at diagnosis, 73.7%, 71.5% and 71.8% of BCS were in employment at 1, 2 and 6 years after diagnosis, respectively. BCS employed at diagnosis had significantly lower probability of being employed at all follow-up time points, compared to controls. BCS outside employment at the time of diagnosis experienced lower probability of employment compared to controls. BCS with secondary or higher education had higher probability of employment compared to BCS with basic education, and BCS living in families with children were more likely to enter employment during follow-up compared to BCS without children. Conclusions: BCS employed at diagnosis had a subsequent risk of unemployment, and BCS not employed at diagnosis had lower probability of entering employment. Additional risk factors are high age, low education, and being single without children. Significance for public health: The risk of unemployment after a breast cancer diagnosis was increased. Job loss is costly economically and socially, both for individuals and for society. Early focus on employment particularly among employees with low education and with little family support may alleviate this problem.
Publication HAL Google Scholar