Prediction, Performativity, and Potential Outcomes: Communicative Rationality in Prediction-Allocation Problems

Sebastian Zezulka, Konstantin Genin

Abstract:

In prediction-allocation problems, predictions are used to allocate social goods. Standard
methodology recommends factoring these problems into two stages: first, individual out-
comes are predicted as accurately as possible; second, allocations are made based on these
predictions. But when predictions inform allocation, they causally influence outcomes.
The “performative” nature of these predictions entangles epistemic and pragmatic con-
siderations, making notions of accuracy ambiguous. Two families of responses propose to
correct the standard procedure. Emphasizing the epistemic, endogenization recommends
making predictions that tend to make themselves true. Pragmatic encroachment recom-
mends making the predictions with the best distributive consequences. We argue that
both responses are misguided. The former undermines the normative goals of allocation.
The latter, despite being instrumentally rational, has a distorting effect on communication.
When predictions are expressions of substantive normative commitments, they cannot help
resolve normative disputes among policymakers and street-level bureaucrats. That is, they
are not communicatively rational. We argue that one does not have to trade off instrumen-
tal and communicative rationality and introduce two criteria of adequacy for predictions:
they should be decision- and discourse-supportive. These criteria can be met by predicting
counterfactual outcomes. We argue that predictions satisfying these criteria can serve as
public reasons.