Publications
Despite their high predictive performance, random forest and gradient boosting are often considered as black boxes which has raised concerns from practitioners and regulators. As an alternative, we suggest using partial linear models that are inherently interpretable. Specifically, we propose to combine parametric and non-parametric functions to accurately capture linearities and non-linearities prevailing between dependent and explanatory variables, and a variable selection procedure to control for overfitting issues. Estimation relies on a two-step procedure building upon the double residual method. We illustrate the predictive performance and interpretability of our approach on a regression problem.
Objectives This study aims to establish an inventory of the teaching of Therapeutic Patient Education (TPE) in the initial training of French midwives. Method: A descriptive quantitative study was conducted in France. An online questionnaire comprising 27 questions was distributed to 35 French midwifery schools. Results: Out of 19 schools that responded to the survey, 11 taught TPE, 8 did not address it in training. This teaching is mainly transversal. The obstacles to the teaching of TPE are the current density of the program, the absence of a text regulating this teaching and the difficulties in circumscribing the field of TPE in relation to that of prevention, promotion and health education. The simulation is used in only one school. Discussion: This survey shows a willingness of educational teams to invest in the teaching of health education, including TPE. For this, it is a question of strengthening the training of teachers in order to clarify the areas of intervention of the midwife calling for health promotion, prevention and health education; to offer specific internships to students and to use simulation. Extending the duration of initial training is an opportunity to plan specific teaching and to discuss the place of the health service.
The Balassa-Samuelson effect is still an important phenomenon in the theory of economic development, as Balassa states, "As economic development is accompanied by greater inter-country differences in the productivity of tradable goods, differences in wages and service prices increase, and correspondingly so do differences in purchasing power parity and exchange rates." To the best of our knowledge, the Balassa-Samuelson effect has not been formally examined in the framework of optimal growth theory. By embedding the Balassa-Samuelson's original model in an optimal growth model setting, we investigate the validity of the Balassa-Samuelson effect in such a case and show that the Balassa-Samuelson effect follows from one of the properties of the optimal steady state.
We use data from Pakistan to establish a reciprocal exchange relationship between the judiciary and the government. We document large transfers in the form of expensive real estate from the government to the judiciary, and reciprocation in the form of pro-government rulings from the judiciary to the government. Our estimates indicate that the allocation of houses to judges increases pro-government rulings and reduces decisions on case merits. The allocation also incurs a cumulative cost of 0.03% of GDP to the government. However, it allows the government to expropriate additional land worth 0.2% of GDP in one year.
We use a distinctive methodology that leverages a fixed population of Twitter users located in France to gauge the mental health effects of repeated lockdown orders. To do so, we derive from our population a mental health indicator that measures the frequency of words expressing anger, anxiety and sadness. Our indicator did not reveal a statistically significant mental health response during the first lockdown, while the second lockdown triggered a sharp and persistent deterioration in all three emotions. Our estimates also show a more severe deterioration in mental health among women and younger users during the second lockdown. These results suggest that successive stay-at-home orders significantly worsen mental health across a large segment of the population. We also show that individuals who are closer to their social network were partially protected by this network during the first lockdown, but were no longer protected during the second, demonstrating the gravity of successive lockdowns for mental health.
This study investigates the effects of the investment-based presence of multinational enterprises (MNEs) on poverty in developing countries. The relationship is decomposed into different pathways corresponding to various facets of firms’ presence and activities, and monetary and multidimensional poverty. We hypothesize that depending on the pathways, the effects can be positive or negative in terms of poverty alleviation, and an overall conclusion has to be nuanced. The hypotheses are tested across 431 Indonesian administrative districts, observed in 2008, 2014 and 2018. Pooled instrumental variable regressions show that a higher presence of foreign MNEs does not reduce the number of people below the poverty line. It raises the depth and severity of poverty, and the population is also more exposed to pollutions. These results inform the ongoing debate, and offer important implications for policy makers eager to attract foreign direct investments, as well as for MNEs’ managers concerned with social responsibility and achieving sustainable development goals in host developing countries.
This article studies the behavioral and socio-demographic determinants of reported compliance with prophylactic measures against COVID-19: barrier gestures, lockdown restrictions and mask wearing. The study contrasts two types of measures for behavioral determinants: experimentally elicited preferences (risk tolerance, time preferences, social value orientation and cooperativeness) and stated preferences (risk tolerance, time preferences, and the GSS trust question). Data were collected from a representative sample of the inland French adult population (N=1154) surveyed during the first lockdown in May 2020, and the experimental tasks were carried out on-line. The in-sample and out-of-sample predictive power of several regression models - which vary in the set of variables that they include - are studied and compared. Overall, we find that stated preferences are better predictors of compliance with these prophylactic measures than preferences elicited through incentivized experiments: self-reported level of risk, patience and trust are predicting compliance, while elicited measures of risk-aversion, patience, cooperation and prosociality did not.
In current economic conditions, financial stability is paramount to the proper functioning of open markets. Financial stability must be balanced with financial flexibility. This relationship is deeply affected by financial fragmentation. This is why Central Banks have focused on these issues in the last decade in particular. Both financial stability and financial fragmentation have unintended consequences on optimal currency areas. In this paper, we survey the original optimal currency areas literature and relate it with the new literature on financial stability and financial fragmentation. We highlight the importance of new macroprudential policies both at the national and regional levels.
Objective This study aims to develop high-performing Machine Learning and Deep Learning models in predicting hospital length of stay (LOS) while enhancing interpretability. We compare performance and interpretability of models trained only on structured tabular data with models trained only on unstructured clinical text data, and on mixed data. Methods The structured data was used to train fourteen classical Machine Learning models including advanced ensemble trees, neural networks and k-nearest neighbors. The unstructured data was used to fine-tune a pre-trained Bio Clinical BERT Transformer Deep Learning model. The structured and unstructured data were then merged into a tabular dataset after vectorization of the clinical text and a dimensional reduction through Latent Dirichlet Allocation. The study used the free and publicly available Medical Information Mart for Intensive Care (MIMIC) III database, on the open AutoML Library AutoGluon. Performance is evaluated with respect to two types of random classifiers, used as baselines. Results The best model from structured data demonstrates high performance (ROC AUC = 0.944, PRC AUC = 0.655) with limited interpretability, where the most important predictors of prolonged LOS are the level of blood urea nitrogen and of platelets. The Transformer model displays a good but lower performance (ROC AUC = 0.842, PRC AUC = 0.375) with a richer array of interpretability by providing more specific in-hospital factors including procedures, conditions, and medical history. The best model trained on mixed data satisfies both a high level of performance (ROC AUC = 0.963, PRC AUC = 0.746) and a much larger scope in interpretability including pathologies of the intestine, the colon, and the blood; infectious diseases, respiratory problems, procedures involving sedation and intubation, and vascular surgery. Conclusions Our results outperform most of the state-of-the-art models in LOS prediction both in terms of performance and of interpretability. Data fusion between structured and unstructured text data may significantly improve performance and interpretability.
This paper combines a database on non-tariff measures (NTMs) with Morocco’s firm-level census to explore the effect of regulatory harmonization with the E.U. on firms’ outcomes. Exploiting cross-sectoral variation in the timing and extent of regulatory harmonization, we find that harmonization waves correlate with rises in productivity, with higher markups and with greater numbers of exporting firms. These effects were reinforced by an induced market-structure change: harmonization temporarily protected the Moroccan market from competition from low-end producers in other developing countries, who took time to adapt. We identify these effects through changes in both trade patterns and firm-level outcomes.
