Xuhang Fan (Shawn)

Appearance

Bribing the Self

🕒 Published at:

Gneezy, Uri, Silvia Saccardo, Marta Serra-Garcia和Roel van Veldhuizen. 《Bribing the Self》. Games and Economic Behavior 120 (2020年3月1日): 311–24. https://doi.org/10.1016/j.geb.2019.12.010.

摘要 ​

研究问题：自我欺骗👉有偏建议；自我欺骗能力的相关因素；自我欺骗抑制 ​

自我欺骗如何在帮助专家保持自我形象时导致偏见

自我欺骗能力相关：自我欺骗的存在；自我欺骗能力的局限（自我欺骗需要借口）

Parameter Panel：激励机会的提供时间点；自我欺骗的借口数量

实验综述 ​

专家向客户推荐投资中，根据推荐获得佣金

Treatment：改变专家得知存在激励的时间点 → 当动机先于提供建议时，他们更可能产生偏见

论文意思是结果和自我欺骗理论一致（在评估投资建议前就了解激励会影响顾问对投资的信念和偏好）

自我欺骗只要有哪怕很小的借口都存在，但所有借口消失时会消失

如何约束自我欺骗来促进优质建议

1 Introduction ​

专家建议很重要 ​

在医疗、法律和财务决策等各种领域决策都很重要

【文献】在存在模糊性或主观性的情况下，个体可能会从事自私自利的行为，这使他们能够保持积极的社会或自我形象

(e.g., Kunda, 1990; Konow, 2000; Dana et al., 2007; Haisley and Weber, 2010; Di Tella et al., 2014; Exley, 2015; Grossman and van der Weele, 2017; Gneezy, Saccardo and van Veldhuizen, 2018; Zou, 2018; Bicchieri, Dimant and Sonderegger, 2019)

  • 研究道德顾虑究竟出自对自我形象的担忧(self-image concerns)还是社会形象的担忧(social-image concerns)

  • 探究自我欺骗能力的限制 （所谓借口的多少 ways to convince themselves）

    RiskReturn experiment ：risk-return tradeoff √ trade-off between their commission and the client’s expected loss √

    Dominance experiment ：risk-return tradeoff trade-off between their commission and the client’s expected loss √

    ObviousDominance experiment：None

实验结论 ​

专家们会主动扭曲自己的信念，使自己能够推荐存在激励的选择

这里面的信念扭曲呢？

为减少错误建议比率，任何自我欺骗的借口余地都必须被消除

研究结果可以帮助法规设计、探讨削弱委员会对建议的影响

通过减少判断中的主观性，或通过强化由有偏见的建议引起的自我形象成本，来减少自我欺骗的范围的干预，需要精心设计，以减少不诚实建议的程度

实验设计 ​

The Advice Game（Sender & Receiver Game） ​

选择发送一条投资建议A或B

Bribing%20The%20Self%200457cae3588b4ec5b730ad61c0893ff4/Untitled.png

Excuse Experiment ​

  • Risk-Return

    A —— a 50-50 lottery between $2 and $4

    B —— a 50-50 lottery between $1 and $7

  • Dominance Experiment

    A —— a 50-50 lottery between $2 and $4

    B —— a 50-50 lottery between $2 and $6

  • Obvious-Dominance Experiment

    A —— a 50-50 lottery between $2 and $4

    B —— a 50-50 lottery between $5 and $7

Treatment：根据自欺动机（NO/Before/After） ​

Control T.；Before T.；After T.

结果预测 ​

Hypothesis 1: In the Risk-Return, recommend A(Before) > **A(After)

S-D=Risk Preference+Fairnes

Hypothesis 2A: In the Dominance, recommend A(Before) = A(After)

S-D==Risk Preference

Hypothesis 2B: In the Dominance, recommend A(Before) > A(After)

S-D==Fairness

Hypothesis 3: In the Obvious-Dominance, recommend A(Before) = A(After)

实验结果 ​

The Risk-Return Experiment ​

  • Hypothesis 1. 成立

  • 选择A（激励性选项）比例为：

    Control=30.6%

    Before=61.2%

    After=32.7%

![Bribing%20The%20Self%200457cae3588b4ec5b730ad61c0893ff4/Untitled 1.png](/bribing-the-self/Untitled 1.png)

The Dominance Experiment ​

  • 选择A（激励性选项）比例为：

    Control=13.3%

    Before=53%

    After=24.7%

  • DID 图片

    图片

    Difference-in-difference：compare the effect of the Before and Control treatments on recommendations in the Dominance experiment to those in the Risk-Return experiment

    the coefficient Before treatment X Dominance 接近0且不显著

  • 说明自我欺骗哪怕只有一点点机会都可以实现

![Bribing%20The%20Self%200457cae3588b4ec5b730ad61c0893ff4/Untitled 2.png](/bribing-the-self/Untitled 2.png)

The Obvious-Dominance Experiment ​

  • 选择A（激励性选项）比例为：

    Control=16%

    Before=30.5%

    After=29.4%

![Bribing%20The%20Self%200457cae3588b4ec5b730ad61c0893ff4/Untitled 3.png](/bribing-the-self/Untitled 3.png)

Mechanisms of Self-Deception: Belief Distortion and Choice for Self ​

自我欺骗的直接证据

实验设计 ​

Three Experiments (RiskReturn, Dominance and ObviousDominance) × Before & After

  • 被试决策前提取信念：你认为被试更喜欢哪个选择？

    paradigm Babcock et al. (1995)

    没有激励信念？？那被试完全可以欺骗实验者啊

    作者做了一个辩解：存在文献说明激励和不激励信念结果都一样(e.g., Friedman and Massaro, 1998; Sonnemans and Offerman, 2001; Trautmann and van de Kuilen, 2014; Hollard, Massoni and Vargnaud, 2016)

    但是这里是有欺骗实验者的动机的！不应该这样

  • 最后要求被试自己做一个选择获得收益（随机抽1实现）

Risk-Return experiment ​

investment A was a 50-50 lottery between $0.50 and $1

Investment B was a 50-50 lottery between $0.25 and $1.75

Dominance experiment ​

investment A was a 50-50 lottery between $0.50 and $1

Investment B was a 50-50 lottery between $0.50 and $1.50

Obvious-Dominance Experiment ​

investment A was a 50-50 lottery between $0.50 and $1

Investment B was a 50-50 lottery between $1.25 and $1.75

实验结果 ​

Beliefs ​

自欺动机强的普遍推荐A率高，而且Before和After差距大

  • RiskReturn

    Preferrable Choice(Before)==75.2%; Preferrable Choice(After)==31.4%;

  • Dominance

    Preferrable Choice(Before)==46.6%; Preferrable Choice(After)==8.1%;

  • ObviousDominance

    Preferrable Choice(Before)==29.9%; Preferrable Choice(After)==10.6%;

![Bribing%20The%20Self%200457cae3588b4ec5b730ad61c0893ff4/Untitled 4.png](/bribing-the-self/Untitled 4.png)

![Bribing%20The%20Self%200457cae3588b4ec5b730ad61c0893ff4/Untitled 5.png](/bribing-the-self/Untitled 5.png)

![Bribing%20The%20Self%200457cae3588b4ec5b730ad61c0893ff4/Untitled 6.png](/bribing-the-self/Untitled 6.png)

Recommendations ​

网上实验结果复现了线下的结果；但是Before和After差距变小

Choice for Self. ​

  • RiskReturn

    Choose A=58.1%(Before); 30.7%(After)

  • Dominance

    Choose A=19.6%(Before); 7.4%(After)

  • ObviousDominance

    no gap in A choices

    Choose A=11.7%(Before); 9.9%(After)

![Bribing%20The%20Self%200457cae3588b4ec5b730ad61c0893ff4/Untitled 7.png](/bribing-the-self/Untitled 7.png)

Discussion ​

Stylized Framework ——Benabou (2015) ​

  • 专家会从选择、推荐选项一致中获得自我形象效用
  • 自我欺骗采取观察者有偏信念更新（Biased Updating）的形式

Cost Of Self-Deception ​

文章的理论模型假设自我欺骗成本是常数（默认是0）

假设在后期研究中可以放宽来缩小 the scope for self-deception

Conclusion ​

相关文献：专家的错误建议；激励扭曲判断 ​

(Steinman et al., 2001; Moore et al., 2010; Cain, Loewenstein and Moore, 2011; see also Malmendier and Schmidt, 2017)

Belief-Based Utility ​

(e.g., Loewenstein and Molnar, 2018; B ́enabou and Tirole, 2016; Golman et al., 2016; Mobius et al. 2014)

过度治疗的抑制 ​

先让专家有选择，再让专家得知欺骗动机

评价和反思 ​

自我欺骗的信念测度问题 ​

  • 自我欺骗的是不是要测度出信念的矛盾（Belief Distortion）

    本文只有一个信念（比较的是Before和After）

  • 自我欺骗的信念测度不用激励？

    没有激励很可能欺骗研究者（换成精准测度范式的话又和Nudge相似）

没有提供个体层面的精准测度 ​

  • Before和After的存在的确促进了有偏推荐——但是一部分本就风险偏好的人他们的选择被忽视了（只要你本来就想推荐风险偏好的选项，那就不需要做任何trade-off）