Principal-Agent Models: An Expansion?

Principal-agent models have been the basis for an extensive set of studies relating bureaucracy to elected officials. Yet despite the outpouring of research, there has been little attempt to test the basic assumptions of the principal-agent model. The model makes two assumptions: that goal conflict exists between principals and agents and that agents have more information than their principals, which results in an information asymmetry between them. But how valid are these assumptions? Can instances be found in which these assumptions do not hold? What happens when we vary these assumptions? In this article, we present both a critique of the traditional principal-agent model and a presentation of a broader theoretical framework for conceptualizing bureaucratic politics. Principal-agent models, derived from such disparate disciplines as law, finance, accounting, and economics, have become the basis for an extensive set of studies relating bureaucracy to elected officials (see Mitnick 1973, 1975, and 1980; Moe 1982, 1984, and 1985; Wood 1988; Wood and Waterman 1991, 1993, and 1994; Scholz and Wei 1986). Such models also have been extended to presidents' decisions to use force (Downs and Rocke 1994) and to the Supreme Court and its relationship to lower courts (Songer, Segal, and Cameron 1994). Despite this outpouring of research, few studies have directly examined the basic terminology and assumptions of the principalagent model. In this article our objectives are to examine principal-agent theory critically, to relax some of its restrictive assumptions, and in the process to present a more general theory of relationships between the bureaucracy and its political environment. Although our critique may be relevant to other uses of the principal-agent model, we limit our assessment to its use as a J-PART 8(1998):2:173-202 theory of bureaucracy. 173/ Journal of Public Administration Research and Theory Principal-Agent Models: An Expansion? THE PRINCIPAL-AGENT MODEL Despite the widespread referencing of the principal-agent model, only in rare instances does a researcher actually discuss the model and how its assumptions fit the problem to be studied. For this reason it is useful to review the model in its various incarnations and to examine its basic assumptions. The principalagent model, as applied in such disciplines as sociology, political science, and public administration, is in essence a theory about contractual relationships between buyers and sellers (see Ross 1973; Pratt and Zeckhauser 1985). As described by Charles Perrow: In its simplest form, agency theory assumes that social life is a series of contracts. Conventionally, one member, the 'buyer' of goods or services is designated the 'principal,' and the other, who provides the goods or service is the 'agent'—hence the term 'agency theory.' The principal-agent relationship is governed by a contract specifying what the agent should do and what the principal must do in return. (1986, 224) A common application in economics is the market for professional services, say between a patient—the principal—and a physician—the agent (Evans 1980). Assuming that both are rational utility maximizers, a patient and a physician are likely to have different goals. The patient would like to be made healthy but pay as little as possible. The physician would be interested in income so, therefore, faces the temptation to provide more medical services than are necessary or to charge a higher price than is warranted. In this exchange, patients are at a disadvantage because they cannot directly evaluate the services provided by the physician. In short, an information asymmetry exists, with an advantage to the physician. Principals seek to manipulate and mold the behavior of agents so that they will act in a manner consistent with the principals' preferences. The contractual arrangement is one tool for accomplishing this goal. The contractual arrangement played an important role in Mitnick's (1973 and 1975) formulation of an institutional or regulatory principal-agent model. Rather than focus on buyers and sellers in an exchange, Mitnick examined the relationship between agents in the regulatory bureaucracy and their political principals (e.g., legislators and interest groups). Mitnick's (1980, 146) framework included "a typology of agency relationships generated from such dimensions as the level of consent between agent and principal regarding the agent's actions (e.g., whether or not a contract exists, a contract which may be formal or informal), the source of specification of the agent's acts . . . . and the level of discretion possessed by the agent." Mitnick noted 174/J-PART, April 1998 Principal-Agent Models: An Expansion? that agents could be motivated by the public interest or by their own narrow self-interest. Since agents enjoy information advantages over their political principals, he added, "The regulators are thus seen as agents to be policed to adherence to the 'public interest* goals of some principal 'public.'" But "such policing is not costless. . . . Principals must pay specification costs to identify acts of the agent that would satisfy the principal's preferences, and policing costs in monitoring and enforcing compliance" (p. 150). Given the existence of these costs, Mitnick concluded (p. 151), "The rational principal attempts policing only if he expects a net return . . . We suggest that because of bounded rationality and information costs, not to speak of fundamental problems in valuation, the return to the given public interest criteria may be more difficult to measure and predict than certain obvious indicators wrongly assumed to be correlated with return to that interest." In other words, in the principal-agent relationship there is an "inevitability of control loss" phenomenon (p. 17). The idea that costs are involved in monitoring an agent, and that principals rationally can decide not to monitor their agents' behavior, explains why the public interest is so often made subservient to private interests in the regulatory arena (one of the main concerns of the regulatory literature). It also raises an important distinction between the economic and the institutional or regulatory principal-agent models. In the institutional model, if a political principal such as the legislature decides that it is not in its rational self-interest to police or monitor its bureaucratic agents, that principal is unlikely to directly bear any cost incurred by the agent's continued shirking. Instead, the bulk of that cost is passed along to the general public, which in a democracy is the legislative branch's political principal. Legislators would pay only a direct cost for their lax oversight of the bureaucracy if the public became aware of it (e.g., if there were a scandal) and sought retribution against legislators at the polls. On the other hand, in the economic model, if the buyer did not adequately monitor the agent's behavior, then the buyer (i.e., the principal) would directly bear the cost, for example, by paying higher fees for services rendered. Likewise, as Mitnick notes, the nature of the contractual arrangement in the regulatory arena can be either formal or informal, whereas formal contracts would be more likely to be adopted in the market-place setting. As these examples demonstrate, important differences emerge once we begin to apply the principal-agent model from market place transactions to the regulatory arena. When political scientists first applied principal-agent theory to bureaucratic settings, further differences emerged from the 175/ J-PART, April 1998 Principal-Agent Models: An Expansion? economic model. In its new formulation, as with buyers and sellers, it is assumed that politicians (principals) and bureaucrats (agents) do not necessarily share similar goals, or to use Mitnick's (1986, 4) terminology, "have potentially differing preferences. . . ." If we assume that they are rational utility maximizers (politicians maximizing reelection chances and bureaucrats maximizing budgets), politicians have an interest in policies that benefit their constituents but have no interest in paying excessively for them. Because politicians and political coalitions change over time and bureaucracies develop separate interests through institutionalization and changing external relationships, a potential conflict occurs when the goals and objectives of principals and agents are at odds. Over time politicians may seek to alter established policy toward their preferred objectives, which may or may not be the same as those of the original legislation or political coalition. Bureaucratic interests also diverge from the original policy through time as politicians develop expertise about how the policy should be implemented or gain support from constituencies that favor different approaches. Even if no policy disagreement exists, principal-agent theory suggests that bureaucrats are likely to shirk, to produce outputs at a higher than needed cost, or to produce a level of outputs that is lower than desired. Agency theory posits a dynamic process of interaction between principals and agents, which develops through time. In this process, bureaucrats are assumed to have distinct informational and expertise advantages over politicians. They better understand the policy and the organizational procedures that are required to implement it. They have both the opportunity and the incentive to manipulate politicians and processes for political gain (see Niskanen 1971; Miller and Moe 1983). For some policies, especially those of a technical nature, bureaucracies are more knowledgeable about organizational needs than politicians are, so politicians are reluctant to intervene. Therefore, the key question for agency theory is, How can politicians vested with contemporaneous legitimacy overcome these uncertainties and the bureaucracy's inherent tendency to shirk? (See Wood and Waterman 19