Multi-task agency: a combinatorial model

An agent allocates unobservable effort over a portfolio of projects; the principal observes only which projects succeed and which fail. An explicit solution to the multi-task agency problem is found using Mobius inversion in the lattice of project portfolios. An application is made to the structure and management of scientific research organisations. Optimal structures are compared with those derived by standard cost-benefit techniques. Systematic biases are identified in the selection of project type, in the level of investment in scientific infrastructure such as libraries, and in the diversification of tasks within the organisation. (C) 2001 Elsevier Science B.V. All rights reserved. JEL classification: D82; H43.