The science of experimental economics ${}^{\text{T}}$${}^{\text{T}}$^("T"){ }^{\text {T}}
实验经济学 ${}^{\text{T}}$${}^{\text{T}}$^("T"){ }^{\text {T}}

Received 20 January 2009  收稿日期 2009 年 1 月 20 日
Received in revised form 6 September 2009
2009 年 9 月 6 日以修订形式收到
Accepted 10 September 2009
2009 年 9 月 10 日接受
Available online 3 October 2009
2009 年 10 月 3 日在线提供

B4  B4 厅
C9  C9 系列
Keywords:  关键字：
Methodology  方法论
Experiments  实验
Philosophy of science  科学哲学

© 2009 Elsevier B.V. All rights reserved.
© 2009 爱思唯尔 B.V.保留所有权利。

Some years ago, it was still common to argue that economics could never be an experimental science but was confined to be purely observational or theoretical. Today the integration of experimental economics into mainstream economics is an established fact. Economics is an experimental science, as well as a theoretical and observational one.
几年前，人们仍然普遍认为经济学永远不可能是一门实验科学，而仅限于纯粹的观察或理论。今天，实验经济学融入主流经济学已是一个既定的事实。经济学是一门实验科学，也是一门理论和观察科学。

In this essay we discuss our views on what experiments can contribute to economics. Our focus is mainly on the relationship between theory and experiments. Experiments are historically closely linked to economic theory and recently empirical economists who typically use non-experimental observational data have begun to use experimental methods. Our eclectic views in this essay are those of “practitioners” of experimental economics, not those of professional methodologists. Like most economists we are normally busy doing research, not philosophy of science.
在本文中，我们讨论了我们对实验对经济学有什么贡献的看法。我们的重点主要放在理论与实验之间的关系上。实验在历史上与经济理论密切相关，最近通常使用非实验观察数据的实证经济学家开始使用实验方法。我们在这篇文章中不拘一格的观点是实验经济学的“实践者”的观点，而不是专业方法论者的观点。像大多数经济学家一样，我们通常忙于研究，而不是科学哲学。

0167-2681/$ - see front matter © 2009 Elsevier B.V. All rights reserved.
0167-2681/$ - 见 2009 © Elsevier B.V.保留所有权利。
doi:10.1016/j.jebo.2009.09.008
doi：10.1016/j.jeb.2009.09.008

A large body of experiments has established “anomalies” in individual decision-making (summarized in a wonderful popular book by Thaler, 1991) and the body of evidence from experiments on bargaining, public goods, coordination, markets, auctions and individual decision-making has grown (Kagel and Roth, 1995, 2009; Plott and Smith, 2008). Standards of how to conduct economic experiments have emerged, textbooks teach methods of experimental economics (Davis and Holt, 1993; Friedman and Sunder, 1994; Friedman and Cassar, 2004), and a dedicated field journal, Experimental Economics, has an increasing impact in economics. Experimental economics has come of age.
大量的实验已经确定了个人决策中的“异常”（在 Thaler 的一本精彩的流行书中总结，1991 年），来自讨价还价、公共产品、协调、市场、拍卖和个人决策的实验的证据体已经增加（Kagel 和 Roth，1995 年，2009 年;Plott 和 Smith，2008 年）。如何进行经济实验的标准已经出现，教科书教授实验经济学的方法（Davis 和 Holt，1993 年;Friedman 和 Sunder，1994 年;Friedman 和 Cassar，2004 年），以及专门的领域期刊《实验经济学》在经济学中的影响越来越大。实验经济学已经成熟。

This essay reflects our views on how we see experimental economics now. We want to acknowledge the pioneers who have shaped much of the methodological convictions of most experimental economists (e.g., Smith, 1982; Plott, 1982; Roth, 1995). A second wave of methodological thinking is taking shape in the form of specialized books tackling methodological issues (Guala, 2005; Caplin and Schotter, 2008; Bardsley et al., 2009), special issues (Methodology, e.g., Sugden, 2005a), and methodological articles in major journals (e.g., Samuelson, 2005; Levitt and List, 2007; Schmidt, 2008). Our goal is to contribute to this thinking about the field.
这篇文章反映了我们对现在如何看待实验经济学的看法。我们要感谢那些塑造了大多数实验经济学家的大部分方法论信念的先驱者（例如，Smith，1982;Plott， 1982;Roth， 1995）。方法论思维的第二波浪潮正在以解决方法论问题的专业书籍的形式形成（Guala，2005 年;Caplin 和 Schotter，2008 年;Bardsley et al.， 2009）、特刊（Methodology， e.g.， Sugden， 2005a）和主要期刊上的方法论文章（例如，Samuelson， 2005;Levitt 和 List，2007 年;Schmidt，2008 年）。我们的目标是为这种对该领域的思考做出贡献。

We begin with our perspectives on economic theory and experiments (Sections 2 and 3). We then move to a discussion of what theories can do for experiments, and what experiments can do for theories (Sections 4 and 5). Section 6 discusses how we should evaluate theories in light of experimental data. We summarize with our “Top Ten” list of things to do (and avoid) for experimentalists and theorists alike in Section 7. Section 8 concludes.
我们从我们对经济理论和实验的看法开始（第 2 节和第 3 节）。然后，我们开始讨论理论可以为实验做些什么，以及实验可以为理论做些什么（第 4 节和第 5 节）。第 6 节讨论了我们应该如何根据实验数据评估理论。我们在第 7 节中总结了实验者和理论家应该做（和避免）的“十大”清单。第 8 节总结道。

Economic theory provides a framework and tools for describing and analyzing economic situations. It makes behavioral assumptions to derive predictions and to provide explanations of economic and social phenomena. More specifically, an economic theory (model) is a description of a social situation, which involves specifying the actors, the choices they face, their information, and how they evaluate each possible outcome.
经济理论为描述和分析经济形势提供了框架和工具。它做出行为假设以得出预测并提供对经济和社会现象的解释。更具体地说，经济理论（模型）是对社会状况的描述，涉及指定参与者、他们面临的选择、他们的信息以及他们如何评估每种可能的结果。

Behavioral assumptions enter in three ways: a first set of assumptions specifies how the individuals evaluate each possible outcome. The evaluations of outcomes are described by the preferences of the individual, including attitudes towards risk and uncertainty (“risk preferences”), the future (“time preferences”), and the extent to which outcomes and the behavior of others are relevant (“social preferences”). A second set of assumptions explicates the individuals’ cognitive abilities (their degree of logical sophistication) and how they form beliefs about uncertain states of the world. A third set of assumptions specifies how the individuals will behave. These behavioral predictions are derived by applying “solution concepts”, which describe how assumptions on preferences and beliefs translate into outcomes.
行为假设以三种方式进入：第一组假设指定了个体如何评估每个可能的结果。对结果的评估由个人的偏好来描述，包括对风险和不确定性的态度（“风险偏好”）、未来（“时间偏好”）以及结果和他人行为的相关程度（“社会偏好”）。第二组假设解释了个人的认知能力（他们的逻辑复杂程度）以及他们如何形成对世界不确定状态的信念。第三组假设指定了个体的行为方式。这些行为预测是通过应用“解决方案概念”得出的，“解决方案概念”描述了对偏好和信念的假设如何转化为结果。

This description of economic theory is a very abstract framework, which is instantiated with more specific assumptions to derive predictions. We describe “standard economic theory” as theory which assumes:
这种对经济理论的描述是一个非常抽象的框架，它通过更具体的假设来实例化以得出预测。我们将“标准经济理论”描述为假设：
(i) Economic decision makers are cognitively sophisticated-they are assumed to be rational; their preferences are complete, transitive and obey the other axioms of expected utility theory. Individuals also have rational expectations about relevant states of the world and the behavior of others. In strategic situations rationality is common knowledge, that is, everybody is rational, knows that everybody is rational, knows that others know that everybody is rational and so on.
（i） 经济决策者在认知上是复杂的——他们被认为是理性的;他们的偏好是完全的、可传递的，并服从预期效用理论的其他公理。个人对世界的相关状态和他人的行为也有理性的期望。在战略形势下，理性是常识，也就是说，每个人都是理性的，知道每个人都是理性的，知道别人知道每个人都是理性的，等等。
(ii) Risk preferences obey the expected utility principle. Time preferences exhibit exponential discounting. Social preferences are assumed to be zero (or already captured in the utility function). In applications, functional forms for the utility function are specified, e.g., it is assumed that people have Cobb-Douglas utility functions or that they have constant relative or absolute risk aversion.
（ii） 风险偏好遵循预期效用原则。时间偏好表现出指数折扣。社交首选项假定为零（或已在实用程序函数中捕获）。在应用程序中，指定了效用函数的函数形式，例如，假设人们具有 Cobb-Douglas 效用函数或他们具有恒定的相对或绝对风险厌恶。
(iii) Solution concepts are typically equilibrium concepts, like competitive equilibrium, Nash equilibrium, or subgame perfect Nash equilibrium.
（iii） 解概念通常是均衡概念，如竞争均衡、纳什均衡或子博弈完美纳什均衡。

It is important to note that these assumptions are psychological (behavioral) because they specify how individuals evaluate outcomes, reason, and reach decisions. This standard model is an extremely powerful framework which provides intellectual clarity and tractable formalizations. This framework has generated considerable theoretical advances in all areas of economics (see, e.g., the textbooks by Mas-Colell et al., 1995 and Romer, 2005, and the essays by Becker, 1993 and Lazear, 2000). This framework has made economics the most influential of the social sciences.
重要的是要注意，这些假设是心理（行为）的，因为它们指定了个人如何评估结果、推理和达成决策。这个标准模型是一个非常强大的框架，它提供了智力上的清晰度和易于处理的形式化。这个框架在经济学的所有领域都产生了相当大的理论进步（例如，参见 Mas-Colell 等人（1995 年和 Romer，2005 年）的教科书，以及 Becker（1993 年）和 Lazear（2000 年）的论文）。这个框架使经济学成为最具影响力的社会科学。

More generally, a theory (model) is an abstraction from the world. Models are, by definition, wrong (descriptively inaccurate). However, this does not mean they are useless. In particular, theories make the logical implications of behavioral assumptions explicit and concrete. The standard model can be seen as a boundary case, but understanding boundary cases is critical for scientific development. This does not relieve theorists from developing models based on more descriptively accurate assumptions, but it provides an important benchmark against which these theorists can compare the implications of their new models.
更一般地说，理论（模型）是从世界中提取的抽象。根据定义，模型是错误的（描述性不准确）。然而，这并不意味着它们毫无用处。特别是，理论使行为假设的逻辑含义明确而具体。标准模型可以看作是一个边界案例，但理解边界案例对于科学发展至关重要。这并不能免除理论家基于描述性更准确的假设开发模型的义务，但它提供了一个重要的基准，这些理论家可以据此比较其新模型的含义。

In summary, economic theory provides a framework for analysis. The conceptual distinction between preferences, beliefs, and constraints is an extremely useful framework that has served economics well. The behavioral assumptions of the standard model are boundary cases, but the standard model has been invaluable in providing testable predictions and developing theoretical alternatives.
总之，经济理论提供了一个分析框架。偏好、信念和约束之间的概念区别是一个非常有用的框架，对经济学有很好的帮助。标准模型的行为假设是边界情况，但标准模型在提供可测试的预测和开发理论替代方案方面具有不可估量的价值。

Experiments are a controlled data generating process. “Control” means that most factors which influence behavior are held constant and only one factor of interest (the “treatment”) is varied at a time. This controlled variation of factors is crucial for being able to draw causal inferences. In rare cases this controlled data generating process occurs naturally (called a “natural experiment”). In the majority of cases, however, it is the researcher who controls the data generating process. Notice that our definition of an experiment is very general-it applies to all experimental sciences, and hence also to economics. Also, neither in our definition, nor in our discussion of experiments do we make a distinction between lab or field experiment or between a classroom experiment, one conducted in an fMRI scanner, or an experiment conducted with students, professionals, or Joe The Plumber. Finally, what “control” means depends inextricably on the exact research question. The question defines which arguments need to be held constant (and which varied systematically).
实验是一个受控的数据生成过程。“控制”意味着影响行为的大多数因素保持不变，并且一次只有一个感兴趣的因素（“治疗”）发生变化。这种受控的因子变化对于能够得出因果推理至关重要。在极少数情况下，这种受控的数据生成过程是自然发生的（称为“自然实验”）。然而，在大多数情况下，研究人员控制着数据生成过程。请注意，我们对实验的定义非常笼统——它适用于所有实验科学，因此也适用于经济学。此外，无论是在我们的定义中，还是在我们对实验的讨论中，我们都没有区分实验室或现场实验，也没有区分课堂实验，即在 fMRI 扫描仪中进行的实验，或与学生、专业人士或水管工 Joe 一起进行的实验。最后，“控制”的含义与确切的研究问题密不可分。这个问题定义了哪些参数需要保持不变（哪些参数系统地变化）。

Economic experiments are experiments motivated by (and designed to answer) economic questions. Economic experiments usually involve controlling the choice sets (what decision makers can do), the information conditions (what decision makers know), and the monetary incentive structure (how decisions translate into payoffs) (Smith, 1976, 1982). Thus, an experiment implements the most important aspects of an economic model as introduced above.
经济实验是由经济问题激发（并旨在回答）经济问题的实验。经济实验通常涉及控制选择集（决策者可以做什么）、信息条件（决策者知道什么）和货币激励结构（决策如何转化为回报）（Smith， 1976， 1982）。因此，实验实现了上面介绍的经济模型的最重要方面。

Like theoretical models, experiments are simplifications of the world. They are equally (or sometimes more) descriptively inaccurate than economic models. They often (but not always) involve very little context, artificial settings, and abstract instructions. That said, just as descriptively inaccurate theories are useful, descriptively inaccurate experiments are useful as well. They provide tests of theories, examples or illustrations of phenomena, and techniques to elicit and measure preferences in various populations. They can provide “existence proofs” or “nonexistence proofs” of theoretical constructs.
与理论模型一样，实验是世界的简化。它们在描述上与经济模型一样（有时甚至更多）不准确。它们通常（但并非总是）涉及很少的上下文、人为设置和抽象指令。也就是说，正如描述不准确的理论是有用的，描述不准确的实验也是有用的。它们提供理论测试、现象示例或插图，以及引发和测量不同人群偏好的技术。他们可以提供理论结构的“存在证明”或“不存在证明”。

In summary, experiments are controlled data collection processes. Like theories, experiments can be descriptively inaccurate, and yet still critical for scientific progress.
总之，实验是受控的数据收集过程。就像理论一样，实验可能在描述上不准确，但对科学进步仍然至关重要。

The role of economic theory for experiments is most straightforward when experiments are used to test a theory. Theory tells the researcher the relevant variables which need to be controlled or manipulated, and makes predictions of what will happen in the experiment. A famous example is tests of competitive market theory, initiated in their modern form by Smith (1962). Market theories are designed to apply to any market and any supply and demand function. Smith “induced” particular supply and demand functions and used competitive price theory to derive the equilibrium price and quantity. Together with assumptions on trading rules (which typically are not part of the theory) and some further auxiliary assumptions (like privacy of payoffs) the experiments test whether actual prices and quantities correspond to those predicted by theory. Economic theory plays a particularly important role when the task is to discriminate between competing theories. The theories allow the researcher to identify the parameters (or experimental treatments) in which competing theories make different predictions.
当使用实验来检验理论时，经济理论对实验的作用最为直接。理论告诉研究人员需要控制或纵的相关变量，并预测实验中将发生的情况。一个著名的例子是竞争性市场理论的测试，由 Smith （1962） 以现代形式发起。市场理论旨在适用于任何市场和任何供求函数。史密斯“诱导”了特定的供求函数，并使用竞争价格理论来推导出均衡价格和数量。结合对交易规则的假设（通常不是理论的一部分）和一些进一步的辅助假设（如收益的隐私），实验测试了实际价格和数量是否与理论预测的价格和数量相符。当任务是区分相互竞争的理论时，经济理论起着特别重要的作用。这些理论允许研究人员确定竞争理论做出不同预测的参数（或实验处理）。

Even when experiments are not explicitly designed to test theory, theory typically provides a framework to derive benchmark predictions (for instance, what does expected utility or prospect theory predict? What will the outcome be if everyone is rational and self-interested, or boundedly rational and inequity averse? How would a time-(in-)consistent person behave?). In this role, economic theory is useful even if we know from other data that the theory is often descriptively wrong, as it provides predictions which can be tested.
即使实验没有明确地设计为测试理论，理论通常也提供了一个框架来推导出基准预测（例如，预期效用理论或前景理论预测什么？如果每个人都是理性和自利的，或者是有限理性和厌恶不平等的，结果会是什么？一个时间（不一致）的人会如何表现？在这个角色中，即使我们从其他数据中知道该理论在描述上经常是错误的，经济理论也是有用的，因为它提供了可以检验的预测。

But perhaps the most critical role of economic theory is to formulate new models consistent with robustly observed behavioral regularities. Experiments can only provide facts; explanations have to come from theory. Prominent examples are prospect theory to explain violations of expected utility theory (Kahneman and Tversky, 1979; Tversky and Kahneman, 1992; Schmidt et al., 2008), the theory of inequity aversion to explain other-regarding preferences (Bolton, 1991; Fehr and Schmidt, 1999; Bolton and Ockenfels, 2000) and hyperbolic discounting to explain present-bias and excessive exponential discount rates (e.g., Loewenstein and Prelec, 1992; Laibson, 1997). These theoretical advances are useful for (at least) three reasons. First, as with any theoretical advance, the new theory typically organizes a body of empirical evidence better than the previous theory, often including the previous theory as a special case. Second, these new theories provide new testable predictions, which moves scientists to the next step of the dialectic between theory and data. Finally, as with the standard theory, new theories illuminate particular behavioral phenomena. Thus if and when the new theories are disproved, they provide a language to talk about the results which they helped to organize.
但也许经济理论最关键的作用是制定与稳健观察到的行为规律相一致的新模型。实验只能提供事实;解释必须来自理论。突出的例子是前景理论，以解释违反预期效用理论的行为（Kahneman 和 Tversky，1979 年;Tversky 和 Kahneman，1992 年;Schmidt et al.， 2008）、解释他人偏好的不平等厌恶理论（Bolton， 1991;Fehr 和 Schmidt，1999 年;Bolton 和 Ockenfels，2000 年）和双曲线贴现来解释现在偏差和过度的指数贴现率（例如，Loewenstein 和 Prelec，1992 年;Laibson， 1997）。这些理论进步是有用的，（至少）有三个原因。首先，与任何理论进步一样，新理论通常比以前的理论更好地组织了一系列经验证据，通常将以前的理论作为特例包括在内。其次，这些新理论提供了新的可检验预测，这使科学家们进入了理论和数据之间辩证法的下一步。最后，与标准理论一样，新理论阐明了特定的行为现象。因此，如果新理论被反驳，它们就会提供一种语言来谈论它们帮助组织的结果。

Consider the theory of inequity aversion (Bolton, 1991; Fehr and Schmidt, 1999; Bolton and Ockenfels, 2000). Prior to these theories, experiments and observational data had produced a body of contradictory-looking evidence with ad hoc explanations: people behaved non-selfishly sometimes (e.g., in ultimatum games and public goods games with punishment), but (with experience) tended to conform to the self-interested prediction (e.g., in public goods games without punishment and in market experiments even with unfair outcomes). Theories of inequity aversion rationalized both sets of findings.
考虑一下不平等厌恶理论（Bolton， 1991;Fehr 和 Schmidt，1999 年;Bolton 和 Ockenfels，2000 年）。在这些理论之前，实验和观察数据已经产生了大量看起来相互矛盾的证据，并带有临时的解释：人们有时表现得非自私（例如，在最后通牒游戏和有惩罚的公共产品游戏中），但（根据经验）倾向于符合自利的预测（例如，在没有惩罚的公共产品游戏中，在市场实验中即使有不公平的结果）。不平等厌恶理论使这两组发现合理化。

These new theories generated new predictions, which were themselves tested (e.g., Blount, 1995; Engelmann and Strobel, 2004; Falk et al., 2008) and inconsistencies discovered. Subsequently new theories of reciprocity were developed to account for these new observations (e.g., Charness and Rabin, 2002; Dufwenberg and Kirchsteiger, 2004; Falk and Fischbacher, 2006; Cox et al., 2007). Just as the existence of inequity aversion theories did not violate the usefulness of standard theories in
这些新理论产生了新的预测，这些预测本身也经过了测试（例如，Blount，1995 年;Engelmann 和 Strobel，2004 年;Falk et al.， 2008） 和发现的不一致之处。随后，新的互惠理论被开发出来来解释这些新的观察结果（例如，Charness 和 Rabin，2002 年;Dufwenberg 和 Kirchsteiger，2004 年;Falk 和 Fischbacher，2006 年;Cox et al.， 2007）。正如不平等厌恶理论的存在并不违反标准理论的有用性一样