Travail et économie publique (interne)

We study the incidence of subsidizing housing retrofits. We leverage cross-sectional and temporal variation in the main policy instrument aimed at fostering housing energy efficiency investment by French households. Under the new, means-tested scheme, households who re- ceive higher subsidies that are discontinuously face an even lower total price for a given ren- ovation step. We show that this over-100% pass-through rate is consistent with rationed local markets where renovation firms can exert market power and discriminate across households’ eligibility status, and where demand for renovation is steep. We corroborate this intuition empirically. Next, we exploit the differential exposure of firms to the reform. Using a triple difference strategy across eligible and non-eligible firms in the construction sector and shock intensity categories, we show that price increases anticipated by firms are a decreasing func- tion of the magnitude of the shock, with more negative shocks associated with larger price increases, which is consistent with the cross-sectional results.

We show that a gender quota system in competitive higher education institutions could be an efficient solution to address the issue of women under-representation in math-intensive fields and elite environments. We use original hand-collected historical data from the entrance exam for one of the most competitive graduate schools in France to evaluate the effect of a change in admission policy that removed a hard gender-based quota system. We document that the end of the quota led to a sharp decline in the percentage of admitted female candidates, but only in math-intensive fields. We then focus on the mathematics entrance exam to delve into the mechanisms. We show that roughly half of this fall can be mechanically explained by a gender performance gap. However, we also uncover an endogenous response by female candidates: there are fewer female candidates at the entrance exam once it became mixed-gender, and this turning-away mechanism is mostly driven by potentially high-achieving female candidates. This detrimental endogenous response of women in a real-life context is important. As this elite institution leads to high-level academic careers in France, we show that the removal of the gender quota increased the gender gap in academic careers for affected students.

We investigate the value of retraining programs in facilitating the mobility of workers into occupations in high demand. By sending 5,000 fictitious job applications to firms posting ads in six tight labor market occupations, we randomly vary the candidates’ training and experience to compare labor demand for four profiles, all aged 21: an textit{incumbent} with both initial training and experience in the posted occupation, and three movers who initially trained and worked in a neighboring, tighter occupation. The movers differ by the extent of retraining they have undergone for the target occupation. Callback rates vary significantly, with the incumbent receiving the highest callbacks, closely followed by the long-retraining mover who underwent several weeks or months of retraining (59% and 51% callback rates, respectively). Untrained movers and short-retraining movers have significantly lower callback rates (30%). We observe notable heterogeneity across occupations, with long retraining yielding much more limited returns in some. We develop and test a matching model, predicting that the effect of retraining on callbacks should increase and then decrease with labor market tightness. Using variation in tightness within occupations, we find that even in the tight labor markets studied, the effect of retraining on callbacks continues to increase with rising tightness.

In this paper, I build an approach to measure proportional treatment effects with non-linear difference-in-differences models. TWFE (two-way fixed effect) linear estimators do not recover difference-in-differences estimates in presence of staggered treatment. I show that this issue extends to Poisson-Pseudo Maximum Likelihood estimators. In the linear case, robust estimators exists to recover correct DiD estimates, but these approaches do not extend to Poisson, as aggregation is challenging in the non-linear case. This paper develops an estimator robust to TWFE staggered bias for PPML, which recovers a quantity with similar interpretation as in the canonical 2-by-2 model.

Teacher salaries have declined relative to private-sector alternatives, coinciding with rising teacher exit rates. This has spurred the question of whether the quality of education has fallen. This paper contributes to this debate by examining how wages influence the turnover of high-productivity teachers. I leverage a reform in France that raised wages in highly disadvantaged schools and effectively acts as both an increase in the wage gap between disadvantaged and advantaged schools, and a decrease in the wage gap between disadvantaged schools and private sector outside options. Using administrative data on middle school teachers and students in France, I find that increasing the salary at disadvantaged schools improved the average teacher productivity in these schools relative to advantaged schools by as much as 20 percent. The effect is driven by an increase in the inflow and retention of high-productivity teachers and a break in the priorly increasing relative trend in exits from the teaching profession of high-productivity teachers in disadvantaged schools. To validate that wage sensitivity within the educational sector translates to outside options, I use employer-employee matched data to build occupational transition matrices between teaching and alternative occupations. Using a shift-share instrument of local outside options salary, I confirm that teachers are more likely to exit when outside options salaries are higher.

Some families of causal parameters are a function of the joint distribution of potential outcomes and thus can only be partially identified. Framing identification as an optimal transport problem, I construct sharp bounds that are valid even under arbitrary model misspecification. Both continuous and discrete covariates can be selected through the multiplier bootstrap and then used to tighten the identified set, which does not inherit any bias from incorrect model selection. This approach requires a so-called distributional difference-in-differences assumption, which holds by construction in RCTs and can be defended in many nonexperimental settings with panel data. I finally study applications with real and simulated data.

Development is associated with job status change. As countries grow richer, men tend to leave self-employment for wage jobs, and women tend to move away from unpaid productive work into inactivity and later, wage work (Bandiera et al. 2022). In this paper, I examine whether these two transitions are correlated at the household level through firm-specific value of spousal inactivity. Using harmonized occupational data linked across spouses, I show that unpaid women are consistently married to self-employed men, across all levels of development. This can be the consequence of assortative matching as much as the fruit of an added value of spouses being available for family firms, and I will present a strategy to identify the two separately. I will present preliminary results that women exhibit an added worker response to hardship at husbands firms only in regions where they have such opportunities. If the do not, their usbands have to self-insure through increased hours at declining firms.

Violence against women is pervasive worldwide, yet it remains largely underreported and rarely prosecuted. This paper examines how news coverage of violence against women influences its reporting and judicial treatment by combining novel administrative data on nearly all cases handled by the French criminal justice system (2012–2019) with high-frequency data on TV news broadcast content. I start by documenting striking judicial patterns: despite a growing number of reported cases of violence against women, perpetrators are rarely prosecuted, with 77% of cases dismissed by prosecutors. Leveraging as-good-as-random variation in TV news content, I find that news coverage of crimes against women has a limited effect on reporting but significantly reduces the propensity to dismiss cases in the short run—a key stage of the criminal justice process. This effect is driven by intimate partner violence and mostly arises after #MeToo, following news highlighting police or judicial accountability and in cases involving greater prosecutorial discretion. Although judges’ sentencing decisions remain directly unaffected, cases prosecuted following news coverage are as likely to result in conviction as other cases. Overall, these findings suggest that increasing the visibility of violence against women could be a powerful tool to strengthen the judicial response and combat this violence.