Core Projects - Data, Methods, and Digital Infrastructure for Resilience

BRIDGE (Building Resilience by Integrating Data on firms, Geographies & Employees) investigates how major economic and digital shocks reshape adaptation, well-being, and performance of firms, workers, and territories. The project is based on secure access to high-granularity French administrative microdata (via CASD), enabling the integration of matched employer–employee information with firm accounts, customs microdata, workplace injury records, and working conditions surveys.

The project is structured around six research questions: (1) whether pre-COVID ICT adoption increased export resilience and through which channels; (2) the impact of robots/automation on employment composition and workplace accidents; (3) the determinants of jobs perceived as unproductive or socially “useless”; (4) the effects of labour-market flexibilization reforms on productivity and innovation; (5) energy and transport vulnerability and inequality in island regions to inform a just transition; and (6) whether local employer market power (monopsony) affects occupational safety and health.

As large-scale surveys become increasingly essential for understanding societal trends and behaviors, the rising costs, respondent fatigue, and declining response rates pose a significant challenge. Traditional methods utilized for reducing the burden of participants, such as matrix sampling, often result in higher variance or limited analytical power. This project proposes an innovative approach using matrix completion techniques to reconstruct incomplete survey data, allowing for the reduction of survey length and subsequent prediction of the missing entries without compromising data quality or analytical validity. This work systematically induces missingness in survey responses to benchmark multiple state-of-the-art imputation methods (including classical techniques, low-rank method, and deep learning models) against ground truth datasets.  

The project will benchmark these methods using metrics such as accuracy, variance, and fitness for purpose, assessing the impact of survey length on results and to what extent it is possible to replicate those obtained from full surveys. Furthermore, a key aspect of the study is evaluating how these imputation methods perform across different survey types, spanning over attitudinal, behavioral, and socio-demographic questions, and determining the boundary conditions under which prediction techniques are most effective. 

This project aims to deliver a cost-effective and generalizable framework for survey length reduction leveraging matrix completion methods. This framework will help address the scalability challenges faced by data-driven research in the social sciences and will be applicable to other fields such as climate change, public health, and economics. Finally, we evaluate fairness and inclusion in survey designs by comparing performance across non-representative and representative samples. Ultimately, this work rethinks social sciences research and strengthens policy making through preserving actionable data while reducing overall cost for surveys. It also offers tools for researchers and practitioners to understand the limits of data imputation and identify the most efficient survey design methods.

This project aims to address the critical challenges of privacy constraints and data scarcity in urban mobility and telecommunications traffic through the advancement of generative Artificial Intelligence. Building upon the successful results of the previous Open Lab project, this research focuses on developing scalable systems capable of generating realistic, privacy-preserving synthetic data.

The methodological innovation moves in three primary directions. First, it seeks to generalize across irregular domains by replacing traditional convolutional architectures with Graph Neural Networks (GNNs) to effectively model road networks and census zones without the limitations of rigid grids. Second, the project introduces semantic conditioning, which integrates heterogeneous urban descriptors such as demographics, land use, and infrastructure. This is achieved by using satellite imagery and attention mechanisms to fuse diverse data sources for more controlled data generation. Finally, the research explores zero-shot and few-shot generalization strategies to transfer learned patterns from data-rich cities to new urban contexts where real traffic data is unavailable.

The expected outcomes include a general-purpose diffusion framework, innovative conditioning modules, and the release of open-source code to support further research in digital resilience and data-driven urban planning.

In recent decades, high-profile thefts and restitutions of archaeological objects have intensified public scrutiny of provenance, legality, and ethical collecting practices. While these events primarily involve museums, cultural authorities, and law-enforcement agencies, their effects on the antiquities market remain only partially understood.

This project investigates whether and how the antiquities market reacts to publicly reported thefts and restitutions, conceptualised as visibility-enhancing events. Focusing on Etrusco-Italic antiquities within a defined chronological framework, the project examines whether auction prices change following such events. It also investigates whether objects with well-documented provenance command price premiums and whether different types of restitutions or scandals generate distinct market responses. The study will also explore, where data permit, a reverse relationship: whether rising market prices increase the likelihood of theft, suggesting a feedback loop between market dynamics and illicit activity.

Methodologically, the project integrates data from international and national art-theft and restitution databases, auction house catalogues and pricing databases, and legal and media records documenting the timing and visibility of key events. Econometric techniques are employed to isolate price effects while controlling for object-specific characteristics. Selected case studies, developed in collaboration with museums and cultural authorities, provide institutional and historical context.

By empirically linking thefts, restitutions, and price dynamics, the project contributes to cultural economics and provenance research. Its findings are relevant to scholars, policymakers, museums, and market participants, offering evidence-based insights into how legal, ethical, and reputational risks are internalised by the antiquities market and how these dynamics may inform strategies for cultural heritage protection and market regulation.

Traditional laboratory experiments have extensively explored the ”manna from heaven” effect, examining how individuals’ behavior changes based on whether an endowment is earned or given. However,this phenomenon has not been studied in online experiments, where participants are paid based on the time spent on platforms, which may influence the perception of ownership and entitlement over the endowment without requiring effort. This study aims to address this gap by using a between-subjects design with three treatment conditions: Control, Time Treatment (participants wait before receiving the endowment), and Real-Effort Task (participants complete a Real-effort task before receiving the endowment). The experiment is conducted on Prolific Platform and it investigates whether time spent on the platform creates a psychological attachment to the endowment, potentially reducing the need for effort-based tasks. By exploring this in an online setting, the study provides insights into whether the well-established use of Real-Effort Task in laboratory experiments should be applied in online experimental contexts as well.