Technical Reports and Working Papers

Measurement and modeling of social networks and peer influence

Much of our work studies social networks when complete network data are unavailable. We develop methods to measure network structure and examine how connections shape outcomes, including the spread of information and behavior.

• The Uneven Rise of AI-Generated Content on the Web
• Non-robustness of diffusion estimates on networks with measurement error (Revise & Resubmit, Econometrica.)
• Randomized recruitment driven sampling
• Scalable Spatial Stream Network (S3N) Models
• Model-based inference and experimental design for interference using partial network data (R package, code)
• Asymptotically normal estimation of local latent network curvature (R package, code)
• General covariance-based conditions for Central Limit Theorems with dependent triangular arrays
• Estimating spillovers using imprecisely measured networks (code)
• Examining Racial Segregation in Associative Networks on Twitter

Rashomon sets, interpretability, and robust inference

Different models can fit the same data well while giving different answers to a scientific question. Our work uses Rashomon sets to examine these differences and develop estimates that account for them.

• Rashomon-Seeded Annealing for Robust Bayesian Inference in Factorial Designs
• When you come to a fork in the road, take it: the Rashomon effect for social scientists

Decision-making under uncertainty

These papers study inference and decisions when data are incomplete, costly to collect, or produced by predictive models. Applications include survey design, active learning, and deciding when prediction models need to be updated.

• From Census to Sensors: The Bureaucratic Regime and the Data Revolution
• When Surveys Become Conversations: Adaptive Matrix Validation for AI-Assisted Interviews
• Spatially robust inference with predicted and missing at random labels
• Adaptive active learning for regression via reinforcement learning
• How to measure obesity in public health research? Problems with using BMI for population inference
• Some models are useful, but for how long?: A decision theoretic approach to choosing when to refit large-scale prediction models
• Do We Really Even Need Data? A Modern Look at Drawing Inference with Predicted Data
• A unified framework for inference with general missingness patterns and machine learning imputation
• A moment-based generalization to Post-Prediction Inference

Global health methodology and data collection strategies

Our global health work develops methods for collecting and analyzing data where routine health information is limited. Recent projects include estimating causes of death from verbal autopsies and designing responses to infectious disease outbreaks.

• Multimodal Artificial Intelligence for Cause-of-Death Assignment From Verbal Autopsy Records: A National Benchmark Study in South Africa
• LAVA: Language Model Assisted Verbal Autopsy for Cause-of-Death Determination
• Feasible contact tracing
• Smart containment with active learning: A proposal for a data-responsive and graded approach to COVID-19 (more details)
• Quantifying the contributions of training data and algorithm logic to the performance of automated cause-assignment algorithms for verbal autopsy
• Bayesian age category reconciliation for age- and cause-specific under-five mortality estimates
• Verbal autopsy in civil registration and vital statistics: The Symptom-Cause Information Archive

Please see arXiv for the most up to date list of working papers.