How an effect varies from one unit to another, carried through the difficulties that real data impose: endogeneity, continuous dose, no experiment, spillovers, and staggered timing. Work under way takes it to dynamic treatments, where a second treatment depends on the first.
The decisions a digital platform makes, one paper each: pricing, comment display, creator selection, targeting, and engineered ties. Work under way asks how these decisions change when AI agents make them.
Deep learning and language models in empirical research, by the role the model plays: the estimator, a graph model of spillovers, and the measurement of language. Work under way uses the model's labels as data and the model as the decision-maker.
Emre Demirkaya, Yingying Fan, Lan Gao, Jinchi Lv, Patrick Vossler, and Jingbo Wang (2024). “Optimal Nonparametric Inference with Two-Scale Distributional Nearest Neighbors.” Journal of the American Statistical Association (Theory and Methods), 119(545), 297–307.
Jingbo Wang and Yufeng Huang (2026). “Scalable Just-in-time Price Elasticity Estimation.” Management Science, forthcoming.
Yan Cheng*‡, Jingbo Wang*, Xinyu Cao*, Zuo-Jun Max Shen, and Michael Xiaoquan Zhang (2025). “Do Bullet Chats Keep Viewers Watching? Estimating Heterogeneous Treatment Effects with a Control Function Approach.”
Yan Cheng*‡, Jingbo Wang*, Xinyu Cao*, Zuo-Jun Max Shen, and Yuhui Zhang (2026). “A Deep-DiD Method to Estimate Heterogeneous Treatment Effects: Application to Content Creator Selection.” Marketing Science, 45(2), 258–279.
Jingbo Wang and Sha Yang (2025). “Estimation of Heterogeneous Treatment Effects in Network-based Quasi-Experiments.”
Chenxi Liao, Jingbo Wang, Ying Xie, and Tianqi Xue‡ (2026). “Do Platform-Engineered Followers Motivate Creators?”
Dynamic treatments planned · Heterogeneous effects of two treatments in sequence, where the second depends on the first.
Andrew Meyer, Jingbo Wang, and Xianchi Dai (2025). “Too Clever by Half: Surprising Language Reduces Consumer Response.”
LLM-labeled regressors planned · Causal regressions in which the regressor is a label produced by a large language model, with measurement error that is not classical.
AI agents planned · The model as the decision-maker.
Cigarette tax pass-through planned · How much of a cigarette excise tax reaches retail prices.