Testing theories of reciprocity: Do motivations matter?

Abstract

One of the key issues for understanding reciprocity is how agents evaluate the kindness of an action. In this paper we investigate experimentally the hypothesis that the motivation driving an action is relevant for its perceived kindness and, as a consequence, for reciprocal behavior. In particular, we examine the hypothesis that, for a given distributional outcome, positive reciprocity is less strong in response to strategically motivated actions than to non-strategically motivated actions. Our results indicate that, both at the aggregate and the individual level, reciprocity is significantly stronger when strategic motivations can be ruled out. These findings suggest that intentions matter and, in particular, that models of intention-based reciprocity should take into account the nature of the motivations behind choices.

Introduction

A large number of experimental and field studies indicate that economic decisions are in many cases motivated not only by material self interest, but also by concerns for fairness. This evidence has led to the development of several recent theoretical models that incorporate fairness as a determinant of economic behavior (see e.g. Fehr and Gächter, 2000, Sobel, 2005, Fehr and Schmidt, 2006, for recent surveys). Alternative theoretical approaches differ with respect to how fairness is defined. In particular, two main classes of models can be distinguished: models that focus on distributional concerns, and models that focus on intention-based reciprocity.¹

In the distributional approach, fairness refers to the distribution of material payoffs. Economic agents are motivated not only by their own material gain, but also by how their payoff compares with that of other agents. Fehr and Schmidt (1999) assume that the utility of a subject depends on the difference between his own payoff and that of other subjects, so that agents have egalitarian preferences. Bolton and Ockenfels (2000) assume that the utility function of a subject depends on his own payoff relative to the average overall payoff, so that agents care about their own relative status. In these models, fairness-related preferences depend only on the final distribution of payoffs, so that agents are not concerned about how a given distribution has been obtained.

In the reciprocity approach, fairness refers to the intentions of other agents. Agents derive utility from rewarding kind actions and punishing unkind actions, even if this is costly in terms of material payoffs (e.g. Rabin, 1993, Dufwenberg and Kirchsteiger, 2004). Preferences depend on the perceived kindness of an action and, therefore, on the beliefs about other agents’ intentions (why an agent has chosen a given action).² In these models, actions with identical outcomes may elicit different reciprocating responses depending on how they are interpreted. A key question for intention-based reciprocity models is therefore how agents evaluate the kindness of a particular action.

One way of assessing the kindness of an action is to compare the action intentionally chosen with the alternative actions that could have been chosen, thus focusing on the strategy space of the first mover. Both intentionality, intended as free-will, and the set of alternative possibilities therefore may contribute to define the perceived kindness of an action. This implies two testable predictions. First, there should be no intention-based reciprocal behavior when the action of the first mover is not chosen intentionally, for example because it is the only available option or it is determined exogenously, by a disinterested third party or by chance. Second, the perceived kindness of an intentionally chosen action depends on the characteristics of the alternative actions that were available to the agent but were not chosen.

At the empirical level, a first group of experimental studies has investigated the role of intention-based fairness by focusing on the first prediction, testing the relevance of first mover's intentionality (the so-called attribution hypothesis). A control treatment where the sender can intentionally choose what action to take among a set of alternatives (thus signalling her intentions) is compared with a treatment where the sender cannot choose, either because she does not have alternative options, as in McCabe et al. (2003), or because her choice is determined randomly, as in Blount (1995) and Falk et al. (2008). The evidence, however, is mixed, and different results are obtained for positive and negative reciprocity.³ It is important to observe that the notion of intentions investigated in this literature refers to the attribution of first mover's intentionality (free will).

A second group of experimental studies investigates the role of fairness intentions by focusing on the second testable prediction, testing the relevance of the alternative actions available to the first mover for the perceived kindness of a chosen action. In these studies, the strategy space of the first mover is manipulated in ways that are strategically irrelevant, but potentially relevant for assessing the fairness of intentions (e.g. Andreoni et al., 2002, Brandts and Solà, 2001, Falk et al., 2003). These studies generally indicate that the perceived fairness of intentions is sensitive to alternative strategy spaces. Bolton and Ockenfels (2005) provide evidence that both distributional factors (relative shares) and strategy spaces (available actions) matter for fairness behavior. Bolton et al. (2005) study experimentally the influence of procedural fairness on the pattern of acceptance and resistance to different outcomes, finding that choice behavior is sensitive to procedural fairness. Overall, however, the evidence on the role of non-distributional factors for models of social preferences is not conclusive. In particular, what determines the perceived kindness of an action remains an open question.

In this paper we propose a new approach for assessing the relevance of intention-based theories of fairness. We formulate and test experimentally the hypothesis that the nature of the motivations driving an action plays an important role for its perceived kindness and, as a consequence, for the reciprocal response to that action. We therefore focus on the behavioral relevance for reciprocity of the type of motivation driving the action an agent is responding to. Following Falk and Fischbacher (2006), we propose a framework to model explicitly the effect of the nature of motivations on reciprocal behavior, and test the hypothesis that, for a given distributional outcome, an action is perceived to be less kind if it is strategically motivated (driven by the expectation of a higher future payoff), than if it is not strategically motivated.

To clarify, consider as an example the sequential game in Fig. 1. Player 1 moves first, choosing between the actions K and G. If player 1 chooses K, the game ends. If player 1 chooses G, player 2 chooses between k and g, and the game ends. Payoffs in monetary units are indicated by the numbers at the end of each path.⁴ Should G be perceived as a kind action? It clearly depends on the strategy space of player 1. If the action K was not available, so that the first mover could only choose G, his action would be perceived differently (presumably, as less kind). Intentionality should matter. Similarly, if the payoffs from choosing K were different, or there were other alternative actions with different payoffs, intentionally choosing G could be perceived differently. It is important to observe, however, that the perceived kindness of action G also depends on what player 1 believes that player 2 will do. Assume that player 1 believes that player 2 can only choose k. Clearly, relative to the original setting, intentionally choosing action G would be perceived differently (presumably, as more kind) by player 2. This hypothesis, the different effect of strategic and non-strategic motivations on the perceived kindness of an action, has received relatively little attention in the empirical literature.

In order to test this hypothesis, we propose a new experimental design, based on a symmetric gift-exchange game, that allows to manipulate the beliefs of the first mover about the strategy space of the second mover. The game is in two phases: in phase 1 a sender gives to a receiver; in phase 2 the same game as in phase 1 is played with reversed roles. We compare two treatments varying the information sets of the players: in one treatment, in phase 1 subjects are not aware of phase 2; in the second treatment, in phase 1 subjects are aware of phase 2. As a consequence, in the first treatment giving in phase 1 cannot be strategically motivated, while in the second treatment giving in phase 1 can be strategically motivated, i.e. driven by the expectation of a higher payoff in phase 2. The two treatments only differ with respect to the nature of the motivations of the sender in phase 1, whereas the distributional outcome of the sender's action and the sender's willingness to give are kept constant across treatments. We expect reciprocity to be stronger in the first treatment, where the sender's action cannot be perceived to be driven by strategic motivations, than in the second treatment, where it can be perceived to be strategically motivated.

The results indicate that the type of motivation behind the choice of an agent has a significant impact on the reciprocal behavior of other agents. When the experimental design rules out the attribution of strategic motivations, second movers’ responses are characterized by significantly stronger positive reciprocity. This result holds both for strategy profiles and for actual decisions. In particular, at the individual level, a large number of second movers display unconditional behavior when first movers’ can be strategically motivated.

The paper is structured as follows. Section 2 presents the theoretical framework of the analysis. Section 3 describes the experimental design. Section 4 discusses the predictions to be tested. Section 5 presents the results. Section 6 concludes with a discussion of the main findings and the implications of the analysis.