Overview and details of the sessions of this conference. Please select a date or location to show only sessions at that day or location. Please select a single session for detailed view (with abstracts and downloads if available).
Design of panel experiments with spatial and temporal interferences
Tu Ni1, Iavor Bojinov2, Jinglong Zhao3
1National University of Singapore; 2Harvard University; 3Boston University
Interference poses challenges in panel experiments. Aggregating units into clusters is common, but optimal aggregation level is unclear. We propose a randomized design for grid-based units. Our design features randomized spatial clustering and balanced temporal randomization. Theoretical performance, inferential techniques, and simulations validate its superiority.
1The University of Hong Kong; 2Purdue University; 3Boston University; 4Tencent Inc.
Challenge: Estimating long-term treatment effects in early-stage experiments is costly. Methodology: We propose a surrogate model using short-term data and historical observations. Results: Verified on WeChat, our method effectively estimates long-term treatment effects. Implications: Our approach reduces experiment duration and provides efficient empirical estimation of long-term effects.
Content promotion for online content platforms with the diffusion effect
Yunduan Lin1, Mengxin Wang1, Max Shen1, Heng Zhang2, Renyu Zhang3
1UC Berkeley; 2Arizona State University; 3Chinese University of Hong Kong
Problem: Content platforms lack effective promotion policies utilizing the diffusion effect. Methodology: We propose a diffusion model, formulate the optimization problem, and introduce D-OLS estimators. Results: We prove submodularity and achieve a 1-1/e-approximation solution. D-OLS estimators are consistent and efficient. Our model improves adoption by 22.48% compared to existing policies. Implications: Our diffusion model enhances content promotion for online platforms.