class: left, top, title-slide # Lie for me ## An experiment about delegation, efficiency, and morality ### Matteo Ploner (CEEL-DEM, University of Trento) ### EcFin Research Seminar
University of Portsmouth ### May 18, 2022 --- layout: true <div class="my-header"> </div> <div class="my-footer"><span>Matteo Ploner (UNITN) - Lie for me</span></div> --- class: top, left ## Motivation <div style="float:right; margin:0 0 0px 0px"> <span class="iconify" data-icon="twemoji:thinking-face" data-width="150"></span> </div> - Moral dilemmas - Tension between *self interest* and -- - *Ethics* - How should I behave? -- - *Reputation* - How do I look to others? -- - *Self-image* - How do I look to myself? -- - A possible solution → *shift the blame* to others - Delegation as a moral wiggle-room - <i>waste management, animal welfare, exploitation of labour/resources, ...</i> --- ## Related Contributions .panelset[ .panel[.panel-name[Deception in the Lab] <!-- --> .pull-left[ <iframe src="https://livedataoxford.shinyapps.io/experimental_economics_iframe2/" height="420" width="800"> <span style="font-size:30%">http://www.preferencesfortruthtelling.com/index</span> </iframe> ] .pull-right[ - Individuals in the lab lie, though not to the full extent - A trade-off between *material benefits* and *reputation concerns + moral costs* ] ] .panel[.panel-name[Altruistic lies] - *Moral character = honesty + benevolence* <a name=cite-levine_are_2014></a>([Levine and Schweitzer, 2014](https://www.sciencedirect.com/science/article/pii/S0022103114000328)) - When the two are in conflict, benevolence may prevail - → altruistic lies that benefits only the others may be perceived as more moral than honesty - Mixed evidence - <span class="iconify" data-icon="heroicons-solid:thumb-up"></span><a name=cite-michailidou2019d></a><a name=cite-weisel_collaborative_2015></a>[Michailidou and Rotondi (2019)](#bib-michailidou2019d); [Weisel and Shalvi (2015)](http://www.ncbi.nlm.nih.gov/pubmed/26261341 http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=PMC4553769) - <span class="iconify" data-icon="heroicons-solid:thumb-down"></span><a name=cite-buckle_lying_2021></a>[Buckle, Füllbrunn, and Luhan (2021)](https://www.sciencedirect.com/science/article/pii/S0165176520304377) ] .panel[.panel-name[Delegation] - People delegate others to shift responsibility attribution <a name=cite-bartling_shifting_2011></a>([Bartling and Fischbacher, 2011](#bib-bartling_shifting_2011)) - People may delegate agents to perform unethical actions <a name=cite-erat_avoiding_2013></a><a name=cite-kandul2018care></a>([Erat, 2013](#bib-erat_avoiding_2013); [Kandul and Kirchkamp, 2018](#bib-kandul2018care)) - Especially when consequences for the deceived party are large - "Surrogate liars" lie to a large extent when this benefits both themselves and their principal ] ] --- ## Main Research Questions .bg-black-10.b--dark-green.ba.bw2.br3.shadow-5.ph4.mt5[ - Do people "hire" others to be their *surrogate liars*? <span class="iconify" data-icon="heroicons-solid:thumb-down"></span> ] -- -- .bg-black-10.b--dark-green.ba.bw2.br3.shadow-5.ph4.mt5[ - Do *expectations* of those delegating matter when being a surrogate liar? <span class="iconify" data-icon="heroicons-solid:thumb-up"></span> ] -- .bg-black-10.b--dark-green.ba.bw2.br3.shadow-5.ph4.mt5[ - Do individual *social attitudes* impact on the likelihood of being a surrogate liar? <span class="iconify" data-icon="heroicons-solid:thumb-up"></span> ] --- class: inverse, center, middle # Design --- class: smaller-page ## Interaction setting <div style="float:right; margin:0 0 0px 0px"> <img src="scheme.png" width=450> </div> - Two players: *Principal* and *Agent* -- - **Lottery** - Two outcomes `\(p(H)=p(L)=1/2\)` - `\(H>L\)` - *Private toss* of a fair die and *self-report* (die-under-the-cap) - If Even <span class="iconify" data-icon="bi:dice-2" data-inline="false"></span> <span class="iconify" data-icon="bi:dice-4" data-inline="false"></span> <span class="iconify" data-icon="bi:dice-6" data-inline="false"></span> `\(\Rightarrow\)` **H**(igh) - If Odd <span class="iconify" data-icon="bi:dice-1" data-inline="false"></span> <span class="iconify" data-icon="bi:dice-3" data-inline="false"></span> <span class="iconify" data-icon="bi:dice-5" data-inline="false"></span> `\(\Rightarrow\)` **L**(ow) -- - **Choice task** - Principal can choose between tossing the die herself (*retain control*) or let the Agent toss (*delegate*) - Retain contol → Lottery A - Delegate → Lottery B - Outcomes of Lottery A and B may be different (see treatments below) --- class: top, left ## Treatments: lotteries (within) .pull-left[ <center> <img src="lott.png" width=600> </center> ] .pull-right[ - Outcomes of *Lottery A* (the one for the Principal) are *fixed* across rounds - Outcomes of *Lottery B* (the one for the Agent) are *manipulated* across rounds - `\(B+ > B = > B-\)` - Random order at matching group level ] --- ## Treatments (between) - *Human Agent* - *Fee* - A fixed amount is paid to the Agent - In addition, before rolling the die, Agents can ask to the Principal a fee between 0 and 1 Euro - Agents know the lottery B - Principals learn the fee and choose between A and B - *No Fee* - A fixed amount is paid to the Agent - *CPU Agent* - As in No Fee condition but agent explicitly replaced by a fair algorithm (50/50) --- ## Treatments (recap) | | B+ | B- | B= | |---|:--:|:--:|:--:| |Fee | `\(\circ\)` | `\(\circ\)` | `\(\circ\)` | |No Fee | `\(\circ\)` | `\(\circ\)` | `\(\circ\)` | |CPU | `\(\circ\)` | `\(\circ\)` | `\(\circ\)` | - Fee/No Fee CPU - between-subjects - B+/B-/B= - within-subjects (randomized) --- ## Additional controls - We control for - Beliefs of Principal about behavior of Agent - Likelihood that a report is favorable (even) - Incentivized (quadratic scoring rule) <br><img src="bel.png" width=500> - Other-regarding attitudes - SVO <br> <img src="SVO.png" width=500> - Classify participants into *Prosocial* (44.7%) and *Individualist* (55.3%) --- ## Participants and Procedures - Administered and programmed in Z-Tree - 12 Sessions run @ CEEL - Total number of participants: 226 <table> <thead> <tr> <th style="text-align:left;"> treatment </th> <th style="text-align:right;"> N </th> </tr> </thead> <tbody> <tr> <td style="text-align:left;"> cpu </td> <td style="text-align:right;"> 46 </td> </tr> <tr> <td style="text-align:left;"> fee </td> <td style="text-align:right;"> 90 </td> </tr> <tr> <td style="text-align:left;"> nofee </td> <td style="text-align:right;"> 90 </td> </tr> </tbody> </table> - Total number of observations: 678 - Earnings of Agents are fixed and announced at the end of the experiment - Choices of Agents are collected with strategy method --- class: inverse, center, middle # Behavioral Predictions --- ## Potential Outcomes .pull-left[ - .green[Truthful Principal] - .green[Truthful Agent] - *B+ → Delegate (€ ~6.5)* - *B= → Retain control (€ ~5.0)* - *B- → Retain control (€ ~4.0)* - .red[Deceitful Agent] - B+ → Delegate (€ 9.0) - B= → Delegate (€ 7.0) - B- → Delegate (€ 6.0) ] .pull-right[ - .red[Deceitful Principal] - .green[Truthful Agent] - B+ → Retain control(€ 7.0) - B= → Retain control(€ 7.0) - B- → Retain control(€ 7.0) - .red[Deceitful Agent] - *B+ → Delegate (€ 9.0)* - *B= → Retain control (€ 7.0)* - *B- → Retain control (€ 7.0)* ] - When are Agents and Principals more likely to be truthful? --- ## Theoretical Framework - We rely on the general framework of belief-dependent motivations - Principal (*Image concerns*) - The Principal wants to appear honest (even to herself) and experiences a pychological cost when lying <a name=cite-abeler_preferences_2019></a>([Abeler, Nosenzo, and Raymond, 2019a](#bib-abeler_preferences_2019)) - Agent (*Guilt aversion*) - The Agent experiences a pychological cost when lying but also dislikes letting down the Principal <a name=cite-battigalli_guilt_2007></a>([Battigalli and Dufwenberg, 2007](https://www.jstor.org/stable/30034441)) --- ## Principals: (Moral) Image concerns - The utility of a Principal `\(j\)` is given by `$$u_j(r_j,c(r_j,\omega),\theta)=\pi(r_j)- \theta_j c(r,\omega)$$` where `\(\pi(r_j)\)` is the payoff associated with the reported value by P, with `\(\pi(H)=H\)` and `\(\pi(L)=L\)`; - `\(c(r,\omega)\)` captures the psychological cost of lying (function of the difference between true and reporte dsignal) - with `\(c(H,L)>0\)` and `\(c=0\)` in all other configurations of `\(\omega\)` and `\(r\)`; - `\(\theta_j\geq 0\)` is the weight given to the cost of lying - → Principal `\(j\)` will report `\(r=H| \omega=L\)` (misreport) when `\(\theta < \frac{H-L}{c(H,L)}\)` - An assumption here is that `\(c=0\)` when the choice is delegated - To lie to be a cost must be under the direct control of the decisionmaker - The principal will delegate when `\(E[u(r_i)]>u_j(r_j,c(r_j,\omega),\theta)\)` --- ## Agents: Guilt aversion - The utility of an Agent `\(i\)` is given by `$$u_i(c(r,\omega),\theta,\gamma, \alpha_p)=\pi_i-\theta_i c(r,\omega)-\gamma_i \bigl[ E[\pi_j; \alpha_j]-\pi_j(r) \bigr]^+$$` - `\(E[\pi_j; \alpha_j]\)` is the payoff Principal `\(j\)` expects from the interaction, given her first-order beliefs `\(\alpha_j\)` - `\(\pi_j(r)\)` is the payoff `\(j\)` actually gets, given `\(i\)`'s report `\(r\)` - the parameter `\(\gamma \geq 0\)` captures `\(i\)`'s sensitivity to *let-down* `\(j\)`, i.e. to deliver a payoff that is lower than what `\(j\)` expects. - The Agent wil minimize `\(\theta_i c(r,\omega)+\gamma_i \bigl[ E[\pi_j; \alpha_j]-\pi_j(r) \bigr]^+\)` - A Principal *who delegates implicitly signals* his/her expectations - In half of the cases, there is a tension between the expectations of the principal and truth-telling --- class: inverse, center, middle # Results --- class: top, left ## Agents: Reports by Lottery .pull-left[ <img src="FIG_Agent_Lotteries.png" width=400> ] .pull-right[ <img src="reg_1.png" width=450> ] .bg-near-white.b--moon-gray.ba.bw2.br3.shadow-5.ph4.mt5[ - B- → advantageous misreports by Agents are significantly more frequent than in honesty. - B- → advantageous misreports are significantly more frequent than in baseline `\(B\)`. ] --- ## Agents: Reports by fee .pull-left[ <img src="FIG_Agent_Fee_Lott.png" width=450> ] .pull-right[ <img src="reg_2.png" width=400> ] .bg-near-white.b--moon-gray.ba.bw2.br3.shadow-5.ph4.mt5[ Misreports marginally more likely in the No Fee condition than in the Fee condition, especially for lottery `\(B-\)`. ] --- ## Agents: Reports by type .pull-left[ <img src="FIG_Agent_Type_Lott.png" width=450> ] .pull-right[ <img src="reg_3.png" width=400> ] .bg-near-white.b--moon-gray.ba.bw2.br3.shadow-5.ph4.mt5[ Overall, different social types do not differ in the likelihood of misreporting. How- ever, Prosocial agents display a higher propensity to misreport for the dominated lottery. In contrast, Individualist agents tend to misreport more in the dominating lottery ] --- ## Principals: Delegation .pull-left[ <img src="FIG_Principal_delegate.png" width=600> ] .pull-right[ <img src="reg_4.png" width=400> ] .bg-near-white.b--moon-gray.ba.bw2.br3.shadow-5.ph4.mt5[ Principals delegate only for lottery B+ ] --- <!--------------------------- --> ## Principals: Beliefs .pull-left[ <img src="FIG_Principal_beliefs.png" width=600> ] .pull-right[ .bg-near-white.b--moon-gray.ba.bw2.br3.shadow-5.ph4.mt5[ Overall, Principals have correct beliefs about honesty of Agents ] ] --- <!--------------------------- --> ## Principas: Reports <br> | Report | Lottery | x | n | rel.freq | LL | UL | |--------|-----------|-----|-----|----------|-----------|-----------| | Even | B= | 78 | 106 | 0.736 | 0.640 | 0.815 | | Even | B- | 103 | 126 | 0.817 | 0.737 | 0.878 | | Even | B+ | 20 | 21 | 0.952 | 0.741 | 0.998 | .bg-near-white.b--moon-gray.ba.bw2.br3.shadow-5.ph4.mt5[ Principals favourably misreport, in all lotteries ] --- class: inverse, center, middle # Conclusions --- ## Discussion <div style="float:right; margin:0 0 0px 0px"> <img src="QR.jpeg" width=300> </div> - *Agents* are generally not willing to lie for "altruistic" reasons - They lie when their technology is worse than that of the Principal - The expectations of the delegating Principal are explicit - *Let-down aversion* may induce to lie - Mediated by individual social attitudes - Second order expectations create stronger commitment than the fee -- - *Principals* do not generally delegate others to lie and directly lie - They anticipate honesty of agents - They delegate only when the expected outcomes of an honest report by the agent are reasonably high (but still lower than those of a direct lie) - To *lie is a cost* but it is easily *overcome by economic incentives* -- <div style="display: flex; justify-content: flex-end; font-size:200%"> <div> Thank you! </div> </div> --- # References .compact[ <a name=bib-abeler_preferences_2019></a>[Abeler, J., D. Nosenzo, and C. Raymond](#cite-abeler_preferences_2019) (2019a). "Preferences for truth-telling". In: _Econometrica_ 87.4. Publisher: Wiley Online Library, pp. 1115-1153. <a name=bib-bartling_shifting_2011></a>[Bartling, B. and U. Fischbacher](#cite-bartling_shifting_2011) (2011). "Shifting the blame: On delegation and responsibility". In: _The Review of Economic Studies_ 79.1. Publisher: Oxford University Press, pp. 67-87. <a name=bib-battigalli_guilt_2007></a>[Battigalli, P. and M. Dufwenberg](#cite-battigalli_guilt_2007) (2007). "Guilt in Games". In: _The American Economic Review_ 97.2. Publisher: American Economic Association, pp. 170-176. ISSN: 0002-8282. URL: [https://www.jstor.org/stable/30034441](https://www.jstor.org/stable/30034441) (visited on Mar. 08, 2022). <a name=bib-buckle_lying_2021></a>[Buckle, G. E., S. Füllbrunn, and W. J. Luhan](#cite-buckle_lying_2021) (2021). "Lying for others: The impact of agency on misreporting". En. In: _Economics Letters_ 198, p. 109677. ISSN: 0165-1765. DOI: [10.1016/j.econlet.2020.109677](https://doi.org/10.1016%2Fj.econlet.2020.109677). URL: [https://www.sciencedirect.com/science/article/pii/S0165176520304377](https://www.sciencedirect.com/science/article/pii/S0165176520304377) (visited on Mar. 07, 2022). <a name=bib-erat_avoiding_2013></a>[Erat, S.](#cite-erat_avoiding_2013) (2013). "Avoiding lying: The case of delegated deception". In: _Journal of Economic Behavior & Organization_ 93. Publisher: Elsevier, pp. 273-278. <a name=bib-kandul2018care></a>[Kandul, S. and O. Kirchkamp](#cite-kandul2018care) (2018). "Do I care if others lie? Current and future effects when lies can be delegated". In: _Journal of Behavioral and Experimental Economics_ 74, pp. 70-78. <a name=bib-levine_are_2014></a>[Levine, E. E. and M. E. Schweitzer](#cite-levine_are_2014) (2014). "Are liars ethical? On the tension between benevolence and honesty". En. In: _Journal of Experimental Social Psychology_ 53, pp. 107-117. ISSN: 0022-1031. DOI: [10.1016/j.jesp.2014.03.005](https://doi.org/10.1016%2Fj.jesp.2014.03.005). URL: [https://www.sciencedirect.com/science/article/pii/S0022103114000328](https://www.sciencedirect.com/science/article/pii/S0022103114000328) (visited on May. 10, 2022). <a name=bib-michailidou2019d></a>[Michailidou, G. and V. Rotondi](#cite-michailidou2019d) (2019). "I'd lie for you". In: _European Economic Review_ 118, pp. 181-192. <a name=bib-weisel_collaborative_2015></a>[Weisel, O. and S. Shalvi](#cite-weisel_collaborative_2015) (2015). "The collaborative roots of corruption". In: _Proceedings of the National Academy of Sciences_ 112.34. Publisher: National Acad Sciences, pp. 10651-10656. ISSN: 1091-6490. DOI: [10.1073/pnas.1423035112](https://doi.org/10.1073%2Fpnas.1423035112). URL: [http://www.ncbi.nlm.nih.gov/pubmed/26261341 http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=PMC4553769](http://www.ncbi.nlm.nih.gov/pubmed/26261341 http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=PMC4553769). ]