« The Industry Standard returns as prediction market | Main | Followup: Second GiveWell founder admits deception »

February 05, 2008

ICD: A 5-step program?

Bob Gibbons presented an intriguing framework for an incentive-centered design program during a talk he gave to our STIET seminar on 31 January. He wasn't thinking about information system design problems, but organizational design. But his fundamental concern was the same: an organization's performance depends on the incentives and how agents respond to them. Like us, he took an explicitly multidisciplinary perspective.

Before I summarize his framework,I'll mention that his multidisciplinary lens is somewhat different than what my group of colleagues and students and I usually do. Our focus so far has been on the interaction between economics (rational choice theory) and computation (information processing). We've talked for a while, and a few of us are starting to integration social psychology perspectives as well (especially to think about non-monetary and intrinsic incentives). We also see a role for personality psychology, and maybe cognitive.

Bob, like me, starts from the foundation of economics. (By the way, Bob was hired by MIT as an assistant professor in Economics while I was there as a grad student, so we got to know each other a bit -- as he reminded me the other day, we played basketball together. He's now a full professor in the business economics group of the Sloan School of Management at MIT.) But then he moves into politics and complex systems perspectives. In particular, he relies on March's approach to control and hierarchies in organizations, and on Winter's work that predicts path dependence.

On to Bob's 5-step program:

  1. The formal is flawed
  2. The relational is required
  3. The formal and relational interact
  4. Institutional design
  5. Building & changing relationships


These suggest a way of explaining why cross-disciplinary approaches are important for ICD, and a framework for moving forward. I can barely do these justice in a short note; Bob gave a detailed 80-minute talk with this outline. I'll try:

Formal is flawed. Formal models that rely on a few narrowly drawn incentive instruments are incapable of doing a very convincing job of describing complex incentive problems. For example, for the standard price model, Bob poses this challenge: "Find an employee with fabulous incentives created solely by a formula." Formulae are not enough: there are too many measurement problems, contingencies, etc.

Relational is required. A quote from Leamer (2007) summarizes the point better than I can: "Most exchanges take place within the context of long-term relationships that create the language needed for buyer and seller to communicate, that establish the trust needed to carry out the exchange, that allow ongoing servicing of implicit or explicit guarantees, that monitor the truthfulness of both parties, and that punish those who mislead.�? A point that Bob made is that most organizational interactions are more like long-term repeated games than they are like one-shot strategic interactions. As a general matter, long-term relationships can lead to a wide variety of outcomes (cf. the Folk Theorem). And, another general implication of repeated games is that the "shadow of the future" is pivotal, so investing in and respecting the relationship is crucial.

Formal and relational interact. The way that relations are structured can have a strong (even dispositive) impact on the effectiveness of the formal incentives. For example, because of the shadow of the future, it can make sense to include a subjective bonus in a compensation plan (that is, the principal says something like "trust me to honestly assess your performance and pay you an ex post bonus based on it" -- trust because the performance is not verifiable by a court and thus not contractible).

(Intermediate summary: Some prices can be chosen, but not the right ones because of gap between performance goals and contractible measures. Relationships help, but not enough. Reliance on relationships affects the desired structure of formal incentives.)

Institutional design. Cyert and March (1963): An organization "is basically a coalition without a generally shared, consistent set of goals. Consequently, we cannot assume that a rational manager can treat the organization as a simple instrument in his dealings with the external world. Just as he needs to predict and attempt to manipulate the ‘external’ environment, he must predict and attempt to manipulate his own firm.�? And here's where politics, authority and control come in: Pfeffer (1981): "it is necessary to understand who participates in decision making, what determines each player’s stand on the issues, what determines each actor’s relative power, and how the decision process arrives at a decision.�? Gibbons' conclusion: Choose the formal to facilitate the relational. He didn't spend much time on this idea, but one of his examples is that the best allocation of control for spot conditions may not be best for relational decisions.

Building and changing relationships. The driving point here is that seemingly similar organizations experience persistent performance differences. Bob explains this as a consequence of path dependence. The paths may differ in (among other things?) the extent to which they rely on formal and on relational incentives. He suggests a stylized, extreme case, in which a concave possibilities frontier between a very controlled firm and a very decentralized firm is traced by relational restructurings, whereas primary reliance on formal incentives carves out a path convex and far inside the frontier. Where the firm ends up, then, depends on the mix of formal and relational incentive structures it employs. (See his slides 49 and 50.)

This is all a bit vague, largely because I don't have a good grasp of the ideas yet. I'll follow Bob's work and see if I can make it more concretely useful for the ICD research programme.

Posted by jmm at February 5, 2008 12:26 AM

Comments

Login to leave a comment. Create a new account.