More than eighty years after Justice Brandeis coined the phrase “laboratories of democracy,” the concept of policy experimentation retains its currency as a leading justification for decentralized governance. This Article examines the downsides of experimentation, and in particular the potential for decentralization to lead to the production of information that exacerbates public choice failures. Standard accounts of experimentation and policy learning focus on information concerning the social welfare effects of alternative polices. But learning can also occur along a political dimension as information about ideological preferences, campaign techniques, and electoral incentives is revealed. Both types of information can be put to use in the policy arena by a host of individual and institutional actors that have a wide range of motives, from public-spirited concern for the general welfare to a desire to maximize personal financial returns. In this complex environment, there is no guarantee that the information that is generated by experimentation will lead to social benefits. This Article applies this insight to prior models of federalism developed in the legal and political science literatures to show that decentralization can lead to the over-production of socially harmful information. As a consequence, policy makers undertaking a decentralization calculation should seek a level of decentralization that best balances the costs and benefits of information production. To illustrate the legal and policy implications of the arguments developed here, this Article examines two contemporary environmental rulemakings of substantial political, legal, and economic significance: a rule to define the jurisdictional reach of the Clean Water Act; and a rule to limit greenhouse gas emissions from the electricity generating sector.