The degree to which predictions can influence the course of events depends on the power of those who make the prediction. The traffic predictions of a service used by millions of drivers has the ability to significantly shape traffic. Were a competing service to enter the market, its predictions would affect a smaller user base and hence show a lesser effect on traffic patterns.

If power is the cause of performativity, we can ask if it is possible to measure power through the strength of performativity. This idea is the starting point for the definition of performative power by Hardt, Jagadeesan, and Mendler-Dünner []. Performative power avoids difficulties with traditional economic tools in the context of digital platforms and digital economies.

The first work on performative power was conceptual and theoretical in nature. What remained open was how to effectively estimate performative power in practice.

Mendler-Dünner, Carovano, and Hardt took on this problem, showing how to estimate performative power using a browser extension that executes subtle randomized controlled trials on a participant’s search results []. Inspired by the Google Shopping case of the European Commission, the study serves as a first blueprint for future investigations of performative power.

On the topic of competition, group leader Ana-Andreea Stoica led a project about how to estimate causal effects of treatments when experimenters compete for a pool of subjects.

The situation arises, for example, in targeted advertising where advertisers bid on subjects. The causal effect of an ad on clicks depends on the position in which it is displayed. The work characterizes the optimal estimator at equilibrium in a novel and intriguing mix of game theory and statistics [].

Stoica also studied artificial intelligence in the context of the European Digital Markets Act in an interdisciplinary collaboration with legal experts []. This work explores the role that LLM-based platforms play as gatekeepers.