Research
Working paper :
PAVED with good intentions ? An evaluation of a French police predictive policing system
Abstract
From late 2017 to early 2019, one of the two french law enforcement agencies tested in 11 out of 101 departments a predictive policing system named PAVED. The system design by the Gendarmerie predicts burglaries and vehicle thefts with the stated objective of better allocating patrols and thus increasing deterrence. We use month-law enforcement jurisdiction area panel data to evaluate whether the system produces the expected reduction in these thefts and whether this effect is due to a deterrent effect or a displacement effect. Both the standard Two Way Fixed Effect and Synthetic Difference-in-Difference estimations consistently indicate a significant reduction of vehicle thefts in the treated \textit{Gendarmerie} areas but no detectable effect on burglaries. Most of our results indicate that vehicle theft has not increased in areas near the treated areas, suggesting that the reduction in vehicle theft is due to a deterrent effect rather than a displacement effect.
Link
The heterogeneous effect of dropping out for higher education students : the French case
Abstract
For higher education system with general and vocational degrees, the question of how to allow resources for dropout policy is fundamental. In France, the university concentrates most of the focus and resources compared to the other vocational track. I use this setting to estimate the heterogeneous effect of dropping out on labor market outcomes (rate of employment and wages), conditional on the followed degree. I use a causal Random Forest methodology in order to account for the heterogeneous students composition of these degrees, with the distance to the closest higher education institution at 6th grade as an instrument for dropping out. I find that using 2SLS lead to underestimate the overall effect of dropping by 9 percentage points for the rate of employment, and by 4 percentage points for the average wage. Vocational degrees dropouts are actually more penalized than university dropouts on their average wage, but not on their time in employment. Finally, using a multidimensional categorization of students can be beneficial for creating targeted dropout policy.
Link
Work in progress :
The reproduction of place-based resource misallocation : an application to predictive policing
Abstract
In many applications of machine learning algorithms for public policy, the representativity of the database are highly dependant of public actions and resource allocation. Having a non representative database lead to misprediction, which undermine the efficiency of the predictive system. In this paper, I propose a measure of resource misallocation based on a public policymaker objective, as well as a methodology to estimate this unbalance reproduction from the data to the predictions. Finally, I propose an application of this methodology to predictive policing, using the NYPD data.
What we do in the shadows : the effect of street lighting on criminality (with Alejandro Gimenez Santana, Adriana Santos, Khalil Zlaoui)
Abstract :
While heavily discussed in criminology literature, the efficiency of public lighting for the reduction of street criminality lacks causal evidence. In this paper, we used the staggered lighting replacement of street poles in the city of Newark, NJ, to study the effect of better and brighter lighting on violent crime. This program implemented new LED lights, which are more cost effective, environmentally friendly and reduce light pollution, while still providing natural lighting at night. The replacement of the poles happened between 2019 and 2021 and concern the entirety of the city’s geographical area. We implement a difference-in-difference model with fixed effects based on the replacement schedule and mapping to estimate the effect of the new LED poles on aggravated assault, robbery and homicides. We find a negative and significant reduction of 2.34 crime for every 100 poles changed, with most of the effect concentrated on 2019 and 2021. Most of this effect is due to a decrease in the aggravated assaults.
Social segregation in secondary education based on income brackets ( with Pr. Pierre Courtioux and Pr. Tristan-Pierre Maury)
Abstract
In this paper, we affect income brackets on students from French secondary education students using an external database on income, living conditions and geographical situation of households. The income brackets are affected using a multinomial logit estimation of the propensity to belong in a given bracket, followed by a boostrap step. Using this framework, we compute different segregation indexes and study their evolution through time.