Research

Publications

Shifting parental beliefs about child development to foster parental investments and improve school readiness outcomes, with J. List and D. Suskind. Nature Communications, 2021. NBER Working Paper 29394.

Featured among the 50 best recent papers in Social sciences by the editors

Socioeconomic gaps in child development open up early, with associated disparities in parental investments in children. Understanding the drivers of these disparities is key to designing effective policies. We first show that parental beliefs about the impact of early parental investments differ across socioeconomic status (SES), with parents of higher SES being more likely to believe that parental investments impact child development. We then use two randomized controlled trials to explore the mutability of such beliefs and their link to parental investments and child development, our three primary outcomes. In the first trial (NCT02812017 on clinicaltrials.gov), parents in the treatment group were asked to watch a short educational video during four well-child visits with their pediatrician while in the second trial (NCT03076268), parents in the treatment group received twelve home visits with feedback based on their daily interactions with their child. In both cases, we find that parental beliefs about child development are malleable. The first program changes parental beliefs but fails to lastingly increase parental investments and child outcomes. By contrast, in the more intensive program, all pre-specified endpoints are improved: the augmented beliefs are associated with enriched parent-child interactions and higher vocabulary, math, and social-emotional skills for the children.

Evaluating the Short and Medium-Run Impacts of Student Aid: Evidence from an Artefactual Field Experiment, with C. Belzil. Journal of the Economic Science Association (forthcoming).

This paper estimates the short and longer-run impacts of an experiment that provided Canadian high-school students with randomized amounts and types of financial aid. We collected data ten years after the experiment to assess the effects on educational and financial trajectories up to age 30. While the average impacts on enrollment, graduation, debt, and future earnings appear to be limited in the overall sample, we find that loan offers have heterogeneous effects in different provinces, shifting students from two to four-year programs in Quebec while producing the opposite effect in provinces where the cost of education is higher. Our results also reveal that parents respond to grant offers by increasing their transfers, which is particularly true outside of Quebec and among students coming from lower-educated backgrounds. Among the latter population, grant offers reduce the probability of completing the first program of study by 14pp, an effect that is not observed in the rest of the sample.

Working papers

A Nonparametric Finite-Mixture Approach to Instrumented Difference-in-Differences, with an Application to Job Training, with O. Cassagneau-Francis, R. Gary-Bobo, and J.-M. Robin

Under review

We develop a finite-mixture framework for nonparametric difference-in-differences analysis with unobserved heterogeneity correlating treatment and outcome. Our framework includes an instrumental variable for the treatment, and we prove non-parametric identification. We can thus relax the single index and stationarity assumptions of Athey and Imbens (2006) at the cost of adding slightly more structure on unobserved heterogeneity. We apply our framework to evaluate the effect of on-the-job training on wages, using novel French linked employee-employer data. Estimating a parametric version of our model with the help of an EM-algorithm, we find small ATEs and ATTs on hourly wages, around 4% in the year of training, falling to under 2% in the following year.

Separating the Structural and Composition Impacts of Financial Aid on the Choice of Major, with C. Belzil and J. Hansen

Under review

Using the unique design of a field experiment among Canadian high school students combined with early life-cycle data collected 10 years later, we estimate the impacts of financial aid distributed as grants on the distribution of university majors. We find that financial aid raises net university enrollment and graduation rates but attracts marginal entrants with lower STEM enrollment probabilities than the population enrolling under the status quo (the composition effect). Among the latter population, financial aid also reduces STEM enrollment and graduation probabilities (the structural effect). Our results thereby reveal potential unintended consequences of financial aid on students’ educational outcomes.

Leveraging Artificial Intelligence and Field Experiments to Uncover Novel Features of Parental Speech and Foster Child Development, with M. Ahmadi, I. Huda, J. List, A. Muller-Molina, Ajay Sailopal, and D. Suskind

R&R (3rd round) PNAS

Parents play a critical role in shaping children’s skills during the first years of life. Yet, identifying the key contributors to richer learning environments remains difficult due to various unobservable factors. In this paper, we combine field experiments with recent advances in AI to uncover new potential determinants within the home environment. To do so, we develop an acoustic processing model that uses more than 600 hours of recorded parent-child interactions combined with assessment data from two home-visiting field experiments conducted in the Chicagoland area to identify features of parental speech that map into children’s skills. Our two experiments consist of the same intervention focused on helping parents provide nurturing interactions to their child but vary in the population they target. The first experiment targets low-income English-speaking parents of children aged 13-16 months at enrollment (N=206); the second experiment targets low-income Spanish-speaking parents of children aged 24-30 months at enrollment (N=91). We exploit the experimental and natural variation in our data to reveal two causal channels and explore one potential moderator. First, our programs significantly improve parental speech inputs, as measured by acoustic features of speech that are predictive of higher socioemotional skills and adult-child conversational turns. Further, we find that our intervention increases different measures of children’s language skills across the two studies, as well as socioemotional skills in the second experiment. Interestingly, our heterogeneity analyses reveal that some of the interventions’ impacts vary by socioeconomic groups, with patterns across the two experiments suggesting that the mechanisms are context dependent.

Is AI our Ally in Child Development? Depends on Who you Ask, with S. Camparo, B. Maule, and D. Suskind 

R&R (1st round) Early Childhood Education Journal

As artificial intelligence (AI) continues to integrate into early childhood education, the perspectives of key stakeholders—parents and teachers—play a crucial role in determining its adoption and impact. This study explores public perceptions of AI’s role in child development, with a focus on privacy concerns, willingness to share personal data, and support for AI in early education. Using survey data from 208 participants across three identity groups (parents, teachers, and the general public), we examine how individual identities influence data-sharing intentions and AI acceptance. A core contribution of this research is the examination of gatekeeper saliency, defined as the extent to which an individual’s role as a custodian of a child’s personal data (as a parent or teacher) influences their willingness to share information. Findings reveal that teachers are generally more open to AI implementation in classrooms than parents, who express greater concern over data privacy and the potential risks of AI personalization. Moreover, parental reluctance to share their child’s data mediates their lower support for AI in early childhood education compared to teachers. These results highlight the tension between the perceived benefits of AI-driven learning and the ethical considerations of digital privacy in early childhood settings. Our findings provide critical insights for policymakers, educators, and AI developers seeking to balance innovation with responsible data stewardship in child-centered AI applications.

Correcting Beliefs to Increase Health Investments: A Field Experiment among Unemployed Youth in France, with B. Crépon

This paper tests two programs designed to raise health investments among low-income unemployed youth. In the first program, people received personalized guidance on health insurance. In the second program, they were also examined by a physician. While the first intervention has virtually no impact, the second almost doubles the probability of consulting a psychologist, in line with the prevalence of mental health issues in this population. To further examine how the information provided by the physician affects investment decisions, we specify a Bayesian model where treatment effects depend on people’s priors about their health. Leveraging machine learning to assess its predictions, we find that for people whose beliefs are the most misaligned with the physician’s diagnosis, impacts on mental health investments can be more than twice larger than the average treatment effect. We show that those people have specific characteristics that job centers can use to better target the intervention.

Gender Gaps in Parental Investments in Young Children: Uncovering the Role of Parental Beliefs, with D. Suskind

Recent research shows that differences in childcare responsibilities led to larger employment losses for women during the COVID-19 recession, contributing to a widening of gender inequalities. This paper explores a new potential mechanism for gender gaps in parental investments: differences in beliefs between fathers and mothers of young children. By surveying fathers and mothers about their own inputs as well as their partner’s inputs, we first demonstrate the presence of large discrepancies between the two data sources. Investment gaps vary from zero – when fathers report the inputs of both parents – to 2.2σ – when mothers report both inputs. We then show that fathers are significantly less likely to believe in the importance of parental investments for child development compared to mothers. Differences in beliefs explain up to 13% of the gender gap in the time spent doing educational activities with the child. In comparison, controlling for differences in employment reduces this gap by up to 46%. This novel finding points to the potential for informational parenting interventions to mitigate gender inequality by increasing the enrollment of fathers in their programs.

Estimating Coherency between Survey Data and Incentivized Experimental Data, with C. Belzil and F. Poinas. IZA Discussion Paper 14594, 2021

We estimate the welfare gains of postsecondary financial aid using two datasets obtained from the same students: survey data on postsecondary intentions and incentivized choice data from a field experiment. By decomposing the welfare gain distributions into latent factors, we show that students weight the factors differently in each setting and that about half of the factors affect the welfare gains in the survey and in the experiment in opposite directions. We then examine the drivers of the incoherency in financial aid valuation ranks between the two data sources and discuss implications for ex-ante policy evaluation of financial aid expansions.

Can Information Policies Increase the Take-Up of Grants among Disadvantaged Students? (PDF)

Growing concerns about the capacity of grant policies to reduce socioeconomic gaps in college enrollment call to investigate the conditions required for those policies to reach their target. This paper aims to better understand the role of information barriers in the low take-up of higher education grants among disadvantaged students. Based on a Canadian lab-in-the-field experiment, I model the demand for grants among high-school students as a function of their perceived utility of university, which depends on their level of information on higher education and on the labor market. I use the model to simulate the effects of several information policies that are commonly implemented in high schools to address the difficulties students may face in the transition to higher education (school counselors, ability tests, information on the financial aid system, job presentations by working people), but are rarely studied. Results show that those policies significantly increase students’ willingness to pay for grants, particularly among students living in rural areas (+140%), students coming from low-income families (+65%), and first-generation college students (+34%). Using the structural model to monetize those policies, I find that the value of information is much higher for disadvantaged students. Finally, the results indicate that such policies have the capacity to close socioeconomic gaps in the take-up of grants.

Going Beyond School: Evidence from a Comprehensive Program Targeting Struggling Children in Deprived Areas, with M. Gurgand and N. Guyon

Policy report: IPP Report n°13, March 2016

In deprived areas, some children tend to cumulate academic difficulties with health, social, and sometimes family difficulties. Many policies have tried to help them. Some expensive policies led to great results, but one might wonder whether existing resources could lead to the same results if these children were actually using them. In this paper, we evaluate a French policy consisting of individualized and comprehensive programs that involve both the child, the parents, and the teacher. Interventions range from sports activities or health diagnosis for the child to administrative assistance for the parents and they leverage existing local resources. Our identification strategy relies on propensity score matching combined with difference-in-differences estimation. To implement it, we collected longitudinal data between Fall 2012 and Spring 2014 on a large set of cognitive and non-cognitive skills among children aged 7 to 10, and we surveyed their parents and teacher, both in areas covered by the policy and in areas that were similar but not covered. We find no impact of the policy on children's behavior and cognitive skills, or on their parents’ relation to school, and negative impacts on the relation to other children and on school motivation. By contrast, school attendance increases among treated children. Comparison with other comprehensive programs suggests that more intensive or earlier interventions might be required to significantly improve the situation of deprived children in the short run.