Effects of Alternate Information Structures in a Decomposed Organization: A Laboratory Experiment

Abstract

The literature on the deomposition of mathematical programs as models for organizational design and resource allocation in decentralized organizations is extensive. Although models differ in detail, all conceptualize the allocation problem as a multi-level managerial coordination procedure, involving local (divisional) and global (organization-wide) resources, in which informational automony is to be maintained. That is, coordination of resource usage by relatively autonomous divisions is to be effected in these models without any one agent in the organization accumulating complete knowledge of the technical resource transformations, payoff or cost coefficients and detailed plans that divisions utilize in converting resources to useful ends. Unfortunately there has been little empirical investigation into the implementational problems of applying these models in decentralized organizations. This study reports on an experiment with human subjects as decision makers in a simulated decentralized organization. The formulation of the overall resource-allocation problem as a linear program permitted two forms of coordination: price-directive in which transfer pricing is used to allocate resources, and resource-directive in which a rationing or budgeting approach is used. Both schemes can be shown to solve the overall organizational problem but impose different information, communication, and decision-making structures upon the subject managers. The experiment was designed to examine comparative managerial performance by the subjects, as central coordinating agents, under the alternative transfer pricing and budgeting schemes. Within each of these schemes two levels of decision time pressure were also introduced to examine its impact upon subject performance. The purpose of the investigation was to study the influence of organizational design (allocation scheme) and situational factors (time pressure) upon human decision-making under carefully controlled experimental conditions. The experimental setting was that of a decentralized university incorporating three subordinate college divisions and one coordinating agent, the president's office. The colleges were modeled as linear programs and the coordination function was assigned to the experimental subjects. In the price-directive scheme a subject assigned a transfer price to each of two global resources in each planning iteration. In the resource-directive scheme a subject directly allocated amounts of the two global resources to each of the colleges in each planning iteration. Consistent with decomposition theory, feedback to the subject in the form of aggregate resource demands (price-direction) or individual bids for higher resource allocations (resource-direction) was given to initiate a new planning iteration. The budgetary goal utilized by subjects was to maximize net dollar contribution from colleges to the university under prespecified quality-of-education constraints. After several planning iterations each subject finalized the resource allocations, terminating the experiment. The hypotheses were that resource-directive subjects would outperform price-directive subjects, that high decision time pressure would exacerbate decision making and that subjects would outperform decomposition algorithms in early iterations. Data from the experiment did not support the first two hypotheses; the third was confirmed. Aside from concluding that strategic factors (organizational design) and tactical factors (time pressure) strongly influence decision making behavior, several specific implications can be tentatively drawn. In similar settings transfer pricing schemes may be preferable to traditional budgeting schemes for planning resource allocations. Furthermore, the potential exists for profitable use of man-machine procedures for resource allocation involving decision support technology in the form of decomposition models to augment human decision heuristics. Finally, experimental methods offer a vehicle for addressing human factors and implementational considerations missing from current analytic models.