diffusion, such adoption patterns are not incontrovertible proof of learning-based diffusion.
It is entirely possible that similar states or countries would adopt the same policies over time
regardless of the actions of others. If so, then the “spread” of a policy within a geographical
region, from one neighboring state to another, or from one group of “peer” states to another
set of peers may have little to do with learning from others and much more to do with
whether the political opportunity for such an adoption has yet arisen within each individual
government. This distinction is important not only for methodological reasons, but also for
its normative implications. If policymakers do not (or need not) rely on lessons learned from
others, then decentralization in federal systems becomes less attractive, and studies of policy
adoptions would benefit more from an internal focus than from cross-country comparisons.
We therefore believe that it is necessary to develop a baseline conception of internal learn-
ing and policy adoption, for comparison to actual learning from others. We do so here by
developing a game-theoretic model of learning and policy diffusion, which we contrast with
a decision-theoretic model that features no cross-state learning.
2
We then note directions for
research that would distinguish learning-based diffusion of innovations from isolated adop-
tions. Only after properly characterizing instances of learning-based diffusion can scholars
adequately address the questions of when, how, and why such diffusion takes place.
Beyond the benefits of distinguishing learning-based diffusion from internal policy adop-
tions, our model generates expectations about which governments are most likely to adopt
which policies. For example, we highlight conditions under which policymakers engage in
costly policy experimentation rather than staying with the status quo. We explain which
policymakers adopt policies that give them a lower expected payoff in the present in hopes of
gaining knowledge about the best policy for the future, and contrast them with policymakers
who free-ride on the experiments and information provision of others.
Specifically, in this paper we develop a model of decentralized policymaking with learn-
ing externalities. The model describes policymaking across multiple jurisdictions over two
periods. To focus on the information aspects of the interaction between policymakers, the
policy choices do not impose externalities on other players in a direct fashion. However, the
players’ strategies create informational externalities by inducing learning over policy options.
Policymakers in the game choose from among three policies. Each policy has two com-
2
For simplicity, we often refer to the governments in our model as “states.” They could alternatively be
thought of as localities or countries. As the model is generic, they could likewise be thought of as firms,
organizations, or individuals adopting any of a variety of innovations. Our model has analytic value relative
to standard R&D models when learning agents’ utilities are influenced not only by the “common” valence
component of a policy but also by a spatial component.
3
Convention