In this thesis I investigate three situations in which a principal must make a public decision. The optimal decision from the principal's point of view depends on informationheld only by agents, who have different preferences from the principal about how the information is used.
In the first two situations (Chapters 2 and 3) the principals and agents - legislatures and bureaus, respectively - are each part of the government and interact to create public policy. In Chapter 2 the bureau has private information about the cost of a public project, performed for multiple legislative principals who can each seek out cost information through oversight. The multiplicity of principals can cause the level of oversight to be inefficiently low due to a collective action problem. Further, the inefficiency becomes more likely as oversight becomes a more important part of the principals' utility functions, and as the oversight technology becomes more effective. For some parameters an increase in the effectiveness of the auditing technology reduces the welfare of the principals collectively.
In Chapter 3 the bureau has substantive expertise about the effects of various policy choices. The principal can delegate policy making authority to the bureau to tap its expertise, but bureaus are imperfectly controlled by statutory restrictions. On the other hand, the scope for delegation can be reduced endogenously if the legislaturechooses to acquire its own substantive expertise. I examine how strategic accounting for both bureaucratic subversion and costly development of legislative expertise affectthe legislature's delegation decision. I also show that legislatures may in fact want subversion to be "cheap," while bureaucrats may want their own authority constrainedand subversion to be costly.
In the third situation (Chapter 4) the information desired by the principal is the valuation of an excludable public good for each member of society. I experimentallycompare three collective choice procedures for determining public good consumption and cost shares. The first, Serial Cost Sharing, has attractive incentive properties butis not efficient; the other two are "hybrid" bidding procedures that never exclude any agents but are manipulable. I characterize Bayesian Nash equilibria in the hybrid mechanisms, and prove some more general properties as well. Serial Cost Sharing tends to elicit values successfully, but is outperformed on several efficiency criteria by a hybrid mechanism - despite its incentive problems and coordination problems due to multiple equilibria.
