Congestion as a Filter: Supply Expansion, Consumer Screening, and Match Quality
How to cite
Wang, Bencehng. 2026. "Congestion as a Filter: Supply Expansion, Consumer Screening, and Match Quality." AEA RCT Registry. AEARCTR-0019480. https://doi.org/10.1257/rct.19480-1.0
Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.
* Primary investigator
Abstract
This completed field experiment studies how changes in advertising supply affect consumer screening and match formation on a large short-video platform in China. The platform sampled 300,000 users from its eligible active-user population, defined as [insert the platform’s exact eligibility criteria], and assigned them by complete randomization at the user level to slot expansion, the default policy, or slot contraction, with 100,000 users in each arm. From June 1 to November 30, 2025, expansion added one potential sponsored position per session, contraction removed one, and the control policy remained unchanged. The recommendation algorithm, auction mechanism, and consumer interface were common across arms. Main outcomes include advertising exposure, clicking, conversion, aggregate matches, attention allocation, and post-transaction feedback.
Intervention
The intervention varied the number of potential sponsored positions available in each user session. Relative to the platform’s default policy, the expansion arm added one potential sponsored position per session, while the contraction arm removed one potential sponsored position. The control arm retained the default policy. The same recommendation algorithm, real-time auction mechanism, campaign constraints, and user interface operated across all three arms. The intervention therefore changed potential advertising capacity before the platform’s existing allocation system selected the advertisements displayed to each user.
Experimental design
Before implementation, 300,000 eligible active users were assigned by computerized complete randomization at the individual-user level to three experimental arms of equal size. Assignment remained fixed throughout the six-month intervention. For every session initiated by an assigned user, the platform applied the corresponding slot-capacity rule before its existing recommendation and auction system selected advertisements. The expansion arm received one additional potential sponsored position per session, the control arm followed the default policy, and the contraction arm received one fewer potential sponsored position. Outcomes were observed before and during the intervention using anonymized platform records.
Design details
Cited by
No citation data available for this registration yet.