Accounting numbers are frequently used in evaluating management performance, and performance evaluation is an important ingredient in motivating managers. Three significant factors generally create difficulties in developing performance measures for a given manager. First, the actions and strategies implemented by the manager are not observable directly, so the manager cannot be compensated directly for his input into the firm. Second, the full consequences of the manager's actions are not observable, in large part because the impact of those actions extend beyond his subunit of the firm and beyond his time as manager of that subunit. Third, uncontrollable events influence the consequences that are observed. The agency theory literature has explored extensively the implications of the nonobservability of the manager's actions and the fact that performance measures are influenced by unobservable, uncontrollable events. However, this literature has given only limited attention to the fact that performance measures frequently are incomplete or imperfect representations of the economic consequences of the manager's actions. On the other hand, discussions of performance evaluation in management accounting texts often raise issues regarding the incompleteness and imperfectness of the accounting numbers that are used as performance measures. For example, divisional accounting profit is described as a short-term financial measure that may induce managers to ignore the future economic consequences of their current actions. More generally, management accounting texts discuss various problems that arise in inducing managers to have goals that are congruent with those of the firm's owners. These discussions typically follow one of two tacks. First, most texts discuss alternative methods for measuring various accounting numbers. For ex-ample, discussions of divisional profit measures often consider direct costing versus absorption costing, the elimination of allocated fixed costs, market versus cost based transfer prices, and the inclusion of interest charges for assets used. The objective here is to create a single measure that is as congruent with the firm's objectives as possible. Second, some texts discuss the use of additional, often nonfinancial, performance measures. Kaplan and Atkinson (1989,536) refer to General Electric and McDonald's as leaders in the use of such measures, and Anthony et al. (1992, 651) provide the following summary of their measures. McDonald's evaluated its store managers on product quality, service, cleanliness, sales volume, personnel training, and cost control. When General Electric decentralized in the 1950s, it identified multiple measures of divisional performance: profitability, market position, productivity, product leadership, personnel development, employee attitudes, and public responsibility. This paper uses an agency theory model to explore the economic impact of variations in performance measure congruence and the use of multiple performance measures to deal with both problems of goal congruence and the impact of uncontrollable events on performance measures. To address the congruency issues, we use a multidimensional representation of the manager's actions. Most of the agency theory literature has examined models in which the manager's action space is either single dimensional or finite. Our approach is similar to the multi-task model examined by Holmstrom and Milgrom [HM] (1991). Our analysis differs from theirs in that we focus on performance measure issues and consider measures that may be influenced by more than one element of the manager's action. The key characteristics of a single performance measure are its congruence with the principal's expected gross payoff and its noisiness (due to uncontrollable events). The first-best result is achieved if, and only if, the performance measure is perfectly congruent and noiseless. A contract based on a noncongruent measure induces suboptimal effort allocation across tasks, whereas performance measure noise results in suboptimal effort intensity. We characterize the value of providing additional performance measures, and illustrate the use of additional measures to reduce risk and noncongruity (due to myopia and window dressing). The value of an additional measure is zero if, and only if, the existing measures constitute a sufficient statistic for the additional measure with respect to the manager's action.
