«Forthcoming, American Economic Review Abstract We conduct an experiment assessing the extent to which people trade off the economic costs of ...»
Preferences for Truthfulness: Heterogeneity Among and Within Individuals
Rajna Gibson*, Carmen Tanner**, and Alexander F. Wagner***
May 2, 2012
Forthcoming, American Economic Review
We conduct an experiment assessing the extent to which people trade off the economic
costs of truthfulness against the intrinsic costs of lying. The results allow us to reject a
type-based model. People's preferences for truthfulness do not identify them as only either
"economic types" (who care only about consequences) or "ethical types" (who care only about process). Instead, we find that preferences for truthfulness are heterogeneous among individuals. Moreover, when examining possible sources of intrinsic costs of lying and their interplay with economic costs of truthfulness, we find that preferences for truthfulness are also heterogeneous within individuals.
Keywords: Truthfulness, heterogeneous preferences, type-based models, costs of lying, honesty, incentives, earnings management JEL-Codes: A13, C91, G30, M14 We thank the Research Priority Program Finance and Financial Markets of the University of Zurich, the NCCR FINRISK, the Swiss Finance Institute, and the Swiss National Science Foundation (Grant #PP001for financial support. Nicolas Berkowitsch provided excellent research assistance. We thank Marcel Morf, who programmed the first version of the code for the experiments. We thank the Editor (Joel Sobel) and three anonymous referees for excellent remarks that have greatly improved the paper. Rainer Winkelmann, who went far beyond the call of duty in helping us clarify the empirical framework and its relation to our preferred preference specification, deserves our special thanks. We also thank Peter Bossaerts, Chiddi Chidambaran, Alain Cohn, François Degeorge, Ernst Fehr, Gerlinde Fellner, Laurent Frésard, Michelle Sovinsky, Susanne Neckermann, Pawel Polak, Bernard Raffournier, Jean-Charles Rochet, Meir Statman, and Mei Wang for fruitful discussions and comments. Useful comments were provided by seminar participants at the European Finance Association meetings, the International Conference on Psychology, the Conference of the Society of Judgment and Decision Making, the European Association of Experimental Social Psychology, la Universidad Carlos III, Fudan University, the University of Fribourg, Northwestern University, Rutgers University, the University of St. Gallen, and the University of Zurich. Part of this work was conducted while Gibson was visiting UCLA.
* Swiss Finance Institute -- University of Geneva. Mail: University of Geneva, 40 Bd Pont d’Arve, CH-1211 Genève 4, Switzerland, +41 (022) 379 8983, Rajna.Gibson@unige.ch ** University of Zurich. Mail: University of Zurich, Department of Banking and Finance, Plattenstrasse 32, CH-8032 Zurich, Switzerland, +41 (044) 634 4016, firstname.lastname@example.org *** Swiss Finance Institute - University of Zurich, Harvard University, and CEPR. Mail: University of Zurich, Department of Banking and Finance, Plattenstrasse 14, CH-8032, Zurich, Switzerland, +41 (044) 634 3963, email@example.com Standard economic models of self-interested utility maximization, which emphasize the role of consequences in determining agents’ actions, predict a grim inexorability to all economic systems. These models are based on an assessment of humans as self-interested agents who behave dishonestly for cogent reasons. These hypothetical persons prioritize the outcomes of their actions and only forgo materially beneficial lying if strategic or reputational considerations arise. Some researchers, such as Bhide and Stevenson (1990), assert that these reputational forces are often weak, implying that honesty simply does not seem to pay.
Examples of disastrous dishonesty based on such self-interest abound in the corporate world. Deliberate deception has augmented the economic effects of regulatory failure, of a deteriorating macro-economy, and of inadequate models, in, for example, the subprime crisis.
Yet, truthfulness also appears to prosper in society. Whistleblowers often jeopardize their careers and friendships when they truthfully reveal the wrongdoing of their companies. Some CEOs are regarded as particularly virtuous (Treviño and Brown 2004).
Numerous journalists risk their lives to report the truth about political repression, economic crimes, and human rights violations.
To explain otherwise puzzling behavior both in the field and in experiments, several authors have proposed the idea that some people experience intrinsic costs when they lie.
For example, in a cheap-talk sender-receiver game, Gneezy (2005) found that many subjects told the truth.1 Of various possible explanations for this result, he inferred that the most plausible was that “people have non-consequential preferences in which they treat the same monetary outcome differently, depending on the process that leads up to it” (p. 392).
Moreover, in Gneezy’s interpretation, “different people weigh these preferences differently” (ibid.). That is, a model in which agents exhibit (continuously) heterogeneous preferences for truthfulness could explain his data.
Because Gneezy’s experiment was set in a strategic context, social preferences may also have been active. Therefore, Gneezy also emphasized the joint relevance of processSimilar results on truth-telling have been obtained in other studies (Evans, Hannan, Krishnan, and Moser 2001; Sánchez-Pagés and Vorsatz 2007). Only a few researchers, such as Baiman and Lewis (1989), have found that people will lie even for just a tiny monetary payoff. See the edited volume by Zak (2008) for numerous additional examples.
dependent preferences and of consequences to oneself and to others.2 But it is precisely because both lying aversion and social preferences operated in his experiment that the two channels were difficult to isolate. Indeed, Hurkens and Kartik (2009, p. 180) showed that Gneezy’s (2005) empirical observations were consistent with the “hypothesis that people are one of two kinds: either a person will never lie, or a person will lie whenever she prefers the outcome obtained by lying over the outcome obtained by telling the truth.” Based on this existing evidence, it is, therefore, possible that the world is populated, in the spirit of type-based models such as that of Koford and Penno (1992), by exactly two fixed types: “economic types” and “ethical types” (in Gneezy’s terminology). Alternatively, these two types can, respectively, be characterized as consequentialists (who care about consequences to themselves and to others, but not about the process by which these consequences are achieved) and as non-consequentialists (who care only about the process, but not about consequences).
The two-type-based model and the model with heterogeneous preferences for truthfulness lead to very different implications, particularly for agent selection and incentive design. Therefore, it is important to determine which of these two models offers a more accurate description of reality.
To address this question, we conducted a decision-theoretic laboratory experiment in which each participant was placed in the situation of a CEO who had to announce his/her firm’s earnings to a passive market. The participants were informed of the true level of earnings. They were also told that falsely reporting higher earnings was legal and would lead to higher actual payoffs than honestly announcing the lower earnings. We considered that economic types would always lie in our experiment because truthfulness was designed to be economically costly; we considered that ethical types would always tell the truth. If, by contrast, individuals varied continuously in the extent to which they were driven by preferences for truthfulness, they would trade off the economic costs of truthfulness with the costs of lying; those with intermediately strong preferences for truthfulness would exhibit the most changes in behavior as economic costs changed.
A long-standing literature considers the role of preferences that depend on process and/or on consequences for others; it recognizes that people do not necessarily maximize utility according to the material consequences of their actions. For example, Rabin (1995) demonstrated how fairness considerations can explain why people are willing to reward or punish others even when this requires a sacrifice of their own well-being.
The simplicity of our experimental setup--involving a decision-making situation with no counterparties--allowed us to isolate motivations for truthtelling that are non-strategic and not driven by social preferences; it permitted us to sidestep issues that occur in strategic contexts. Moreover, in our experiment, we observed individuals’ behaviors. This setup enabled us to provide evidence, stronger than that developed in existing works, regarding heterogeneity in preferences for truthfulness.
We observed that, in a situation where the standard economic model predicts that everybody will lie, 32% of the participants chose not to do so, thus forgoing a larger variable compensation. Importantly, the aggregate percentage of truthtellers decreased as the costs of truthfulness increased. Our individual-level regressions imply that the marginal effect of a cost increase on the probability of an individual’s telling the truth is significantly negative, even after controlling for various demographic and psychological factors. These results are at odds with the type-based model but are consistent with a model that posits heterogeneous preferences for truthfulness.
Our primary contribution, therefore, is to provide evidence for the notion that people occupy a spectrum of preferences for truthfulness rather than only two opposite positions.3 As a secondary contribution, we examine potential sources of the heterogeneity in preferences for truthfulness. Tendencies towards impression management and selfdeception offer no explanatory power; however, one measure of one source of intrinsic costs of lying, an index of “protected values of truthfulness,” seems to organize the data well. We also find substantial evidence of non-separability between this measure of intrinsic costs of lying and economic costs of truthfulness in the utility function. In other words, total preferences for truthfulness not only display heterogeneity among, but also within, individuals. We do not have adequate measures of other possible sources of intrinsic costs of lying, including, in particular, measures of expressive preferences.
Therefore, we acknowledge that other preference formulations could potentially explain our empirical evidence.
Section I presents the basic tradeoff and the hypotheses. Section II describes the experiment. Section III discusses the main results. Section IV explores possible sources of Gneezy, Imas and Madarász (2012) found that participants who lied in a sender-receiver game were more likely to later donate to charity than those who chose to tell the truth. This result also suggests that individuals cannot only be classified as “ethical types” and “economic types.” intrinsic costs of lying and their interaction with economic costs of truthfulness. Section V concludes.
I. The tradeoff
Consider an agent who decides whether to tell the truth, T=1, or to lie, T=0. Lying, the agent receives a certain income m. There are Economic Costs Of Stating the Truth, for which we use the term ECOST. The agent receives funds m-ECOST when he tells the truth.
We model preferences for truthfulness by positing that the agent also experiences total costs of lying, C i. (For the moment, C i is given. We discuss in Section IV how these total costs of lying may arise from the interplay between the intrinsic costs of lying and the extrinsic economic costs of truthfulness.) If types are continuous, C i can take on any value, positive or negative. By contrast, in the two-type model, there are only “ethical types” who have C i and “economic types” who have C i 0. Let the global utility
where T is the choice variable.4 For simplicity, and because wealth effects are unlikely in our experiment, we assume the agent has a constant marginal utility of money b0. We also assume that all participants have the same b.5 The difference between the utilities of truthtelling and of lying is given by Yi * C i bECOST.
(2) An individual exhibits truthfulness when Yi* 0. This implies that truthfulness can, in this framework, only arise as optimal behavior if there is a positive total cost of lying. While social preferences are known to contribute to behavior (e.g., Fehr and Fischbacher 2002, 2003), our experiment is designed to eliminate any role for altruism, reciprocity, guilt Truthfulness here is a matter of preference. Alternatively, we could posit a constraint involving a need to maintain a minimum level of truth-telling. Within this simple context, the two formulations are identical.
Rabin (1995) showed that moral preferences and moral constraints can result in different incentives for information collection.
It is standard to assume that, abstracting from the preference feature of interest (for example, inequity aversion), all participants have equal marginal utility of money. See, for example, Fehr and Schmidt (1999).
aversion (Charness and Dufwenberg 2006), and related factors, as well as any role for strategic concerns that might arise with repeated interaction.
Consider now a population of individual decision-makers (whose distribution of Ci is not known), each of them facing various economic costs of truthfulness. A type-based model, such as that of Koford and Penno (1992), implies that ethical types, with their overwhelming preferences for truthfulness, would always choose T=1, and this choice would be invariant to ECOST. Conversely, economic types would always lie when profitable. (At ECOST = 0, they would perceive no advantage or disadvantage to either telling the truth or lying; but at all other levels of economic costs of truthtelling, the utility difference Yi* would be negative.) Aggregating across the population of individuals, this
implies the following hypothesis:
Hypothesis TYP (Type-based model): The fraction of the population telling the truth remains constant across varying economic costs of truthfulness.