AIDS policy and psychology: a mechanism-design approach.

Economic theorists have given little attention to health-related externalities, such as those involved in the spread of AIDS. One reason for this is the critical role played by psychological factors, such as fear of testing, in the continued spread of the disease. We develop a model of AIDS transmission that acknowledges this form of fear. In this context we design a mechanism that not only encourages testing but also slows the spread of the disease through voluntary transmission. Our larger agenda is to demonstrate the power of psychological incentives in the public health arena.

1. Introduction

* Economic theorists have devoted great effort to designing mechanisms to reduce the damage caused by externalities. How best to slow the spread of AIDS would seem to be an important case in point. Yet despite the pioneering efforts of Philipson and Posner (1995) and Kremer (1996), economic theorists have largely ignored this question. Indeed they have given little attention to any health-related externalities, despite their profound social importance.

One factor that differentiates health-related from standard externalities is the central role played by psychological factors. This critical role of psychology is hinted at by Philipson and Posner when they discuss the role of fear in limiting the efficacy of certain AIDS policies. In particular, they discuss the potential impact of verifiable "AIDS-cards" that offer proof to all that one does not have the virus. In an idealized version of this scheme, they speculate that there would be assortative matching, with those who were verified to be clear of the disease matching only with others of their type. Yet, following the empirical findings of Lyter et al. (1987), they argue that not many would be willing to take such a test for psychological reasons: (1)

many people are fearful of tests which may show they are doomed even if the probability of that result is very low (Philipson and Posner, 1995, p. 472).

In our view, psychological realities of this type need no longer be seen as barriers to progress in economic theory. Rather, they are profoundly enriching. The time has come not only to acknowledge their importance, but also to incorporate them into policy analysis. In this spirit we present a theoretical approach to AIDS policy that explicitly incorporates fear of the form suggested by Philipson and Posner. We use our approach to reassess the potential for certification policies to reduce the spread of the disease. We outline circumstances in which a variant of the AIDS-card scheme may be a very effective policy tool, even when fear is profound.

We begin our analysis in Section 2 by developing a strategic model of the spread of AIDS, and the potential role of certification in limiting its spread. We confirm the conjecture of Philipson and Posner that in the absence of fear, there is an equilibrium in which all agents test, and matching is assortative. In Section 3 we incorporate into the model a fear-induced preference for late resolution of uncertainty and confirm that it may indeed render the certification policy ineffective. If fear is sufficiently important, then not everyone tests, and the disease continues to be spread by those who are HIV-positive.

What can a policy maker do to reduce the spread of AIDS? Section 4 lays out our vision of the feasible set of policies. We assume that the policy maker is able both to assess the health status of private agents and to pass certifiable messages back to them based on the results of these assessments. We analyze two different classes of policy: conditional mechanisms that condition the procedure of sending messages on the test results of all individuals, and unconditional mechanisms that have less flexibility in their choice of messages. Throughout, we assume that the policy maker has no direct control of any subsequent social interactions among the private agents. To affect change, the policy maker must not only induce voluntary testing and certification, but also structure the certification process so that it changes subsequent patterns of sexual behavior. (2)

Section 5 explores the details of policy design in an important special case. We consider a policy maker whose fundamental goal is to stop the spread of AIDS. We provide conditions under which certification policies can be used to establish an equilibrium with absolutely no new infections. This provides a positive answer to the key question of feasibility. Under the assumed conditions, which may include the presence of a high level of fear, policies exist that can stop cold the spread of infection.

In addition to establishing the value of psychological incentives, Section 5 also clarifies the distinction between the conditional and unconditional mechanisms. Unconditional mechanisms are blunt tools. The only way that they can be used to stop the spread of AIDS involves simultaneously blocking many ex post advantageous trades by inducing doubt concerning the HIV status of potential partners. Conditional mechanisms are considerably more delicate, and they enable the policy maker to not only stop the spread of AIDS but also achieve the secondary goal of allowing as many risk-free matches as possible to take place, while keeping fear to a minimum.

While conditional mechanisms have clear advantages over unconditional mechanisms, they also have practical disadvantages. One disadvantage is that they are far more complex, since the test result of any one individual depends on the results of others. This would make them far more difficult to mechanize than the unconditional mechanisms. If the tests cannot be mechanized, it is likely that the policy would have to be left in the hands of individual physicians. This raises the question of credibility. We show in Section 6 that a caring physician with a healthy patient would not agree to pass on the ambiguous message called for by our mechanisms. (3) In practical terms, our unconditional mechanism may be more readily implemented than our conditional mechanism.

Although the details are somewhat intricate, the fundamental point of our analysis is simple. There are strong health-based incentives to test for AIDS, but fear may override these incentives. Our resolution of the problem is to decrease the informativeness of a bad test result, mitigating the fear of bad news and thereby allowing the health-based incentives to reassert their primacy. A similar approach to policy may be of value in the many other medical settings in which fear-based avoidance behavior is believed to play a role. For example, according to Dr. Timothy Johnson, ABCNEWS's medical editor (see Cohen, 2002), "Study after study has shown that men are more reluctant to face up to worrisome symptoms or go to the doctor for checkups. And that is probably one big reason why men's life expectancy, which in the early 1900s was virtually the same for both sexes, now lags behind by approximately six years" (4) In addition, there may be other policy options worthy of exploration, such as using "fear appeals" to induce a private desire for knowledge (Witte, 1998; Witte and Allen, 2000; and Caplin, 2003). As research on these subjects develops, we believe that "behavioral epidemiology" will take its place alongside economic epidemiology (pioneered by Philipson (2000) and others) as a guide to policy makers trying to influence health outcomes.

The theoretical analysis that follows has a certain amount of novelty. To the best of our knowledge, ours is the first example in which the Kreps and Porteus (1978) model of preferences over the timing of the resolution of uncertainty has been placed into a mechanism-design framework. This combination calls for use of the psychological game apparatus of Geanakoplos, Pearce, and Stacchetti (1989). Yet our article represents only the smallest of first steps in the larger program of incorporating psychological phenomena into policy analysis. Economic theory itself is in need of substantial expansion if we are to incorporate a sophisticated understanding of psychological phenomena into our analysis.

2. The basic model

* We model the social context underlying the spread of AIDS as an extensive game with imperfect information but perfect recall (see Osborne and Rubinstein, 1994). We denote this game by [GAMMA].

* The extensive-game form. Society consists of a fixed finite set of individuals. (5) We focus in particular on the three-player case, since this is the smallest number in which competitive forces can come into play to induce testing. In Section 6 we comment on how the model can be extended to allow for a large population of players. The game itself is played in four stages as follows:

(i) Determination of players' types. Each player has probability p of being infected and the probabilities are independently distributed across the three individuals. We let t = ([t.sub.1], [t.sub.2], [t.sub.3]) denote the vector of types, with [t.sub.1] [member of] {(+), (-)}, where (+) stands for infected" (HIV positive) and (-) stands for" healthy" (HIV negative). Players are ignorant of their types, but the probability distribution according to which their types is determined is common knowledge.

(ii) Private testing decisions. At the second stage of the game, all three players simultaneously decide whether or not to test for AIDS. We let [a.sup.0] denote the vector of testing decisions, with [a.sup.0.sub.i] [member of] {T, N T}, where T represents a decision to test and NT represents a decision not to test. Each player observes only his own action and is left to infer the decisions of others.