# Performative Prediction **Entity class:** Machine-learning framework **Domains:** Prediction / feedback / causal inference / distribution shift ## Definition **Performative prediction** describes a prediction that changes the future population or outcome it is meant to predict because institutions act on the prediction. Perdomo, Zrnic, Mendler-Dünner, and Hardt formalized the term in 2020: model deployment induces a new data distribution, so repeated retraining can converge on the world produced by the model rather than a fixed external reality. In preventive security, a score can redirect observation, questioning, investigation, restriction, or assistance. Those interventions alter conduct and determine which outcomes become visible. The returned data are therefore partly generated by the deployed system. This is the decision-stage mechanism connecting [[wiki/Observation-Induced Ground Truth|Observation-Induced Ground Truth]] to a [[wiki/Runaway Predictive Feedback Loop|Runaway Predictive Feedback Loop]]. The harm question is whether the system can distinguish a changed threat environment from a changed measurement environment. [[wiki/The Second Error Function|The Second Error Function]] must count state changes, opportunity losses, and new observations produced by intervention so that apparent confirmation is not mistaken for independent validation. ## Relationships - **feedback mechanisms:** [[wiki/Observation-Induced Ground Truth|Observation-Induced Ground Truth]], [[wiki/Runaway Predictive Feedback Loop|Runaway Predictive Feedback Loop]], and [[wiki/Selective Labels|Selective Labels]]. - **measurement:** [[wiki/The Second Error Function|The Second Error Function]] and [[wiki/Security Harms Topology|Security Harms Topology]]. - **analysis:** [[research/Harms Incurred While Bringing Preventive Systems Online|Harms Incurred While Bringing Preventive Systems Online]]. - **collection:** [[collections/Terrorism, Counterterrorism, and the Intelligence Environment|Terrorism, Counterterrorism, and the Intelligence Environment]]. ## Sources / Provenance - [Perdomo, Zrnic, Mendler-Dünner, and Hardt — “Performative Prediction,” Proceedings of Machine Learning Research 119](https://proceedings.mlr.press/v119/perdomo20a.html) (2020-07-13–18). **As of:** 2026-09-23