Neuroscience & Economics Neuroeconomics • Why? • How? • Two examples Sanfey et. al. on neural basis of rejections in UG de Quervain et. al. on neural basis of punishment Fehr – Experimental and Behavioral Economics 1 Neuroeconomics & Behavioral Economics • Behavioral economics developed alternative models of economic behavior. o Fairness and reciprocity models o Prospect theory, hyperbolic discounting, learning models. • These models are black box models. They aim to predict behavior better but there is no ambition to understand the mind’s internal processes that generate the behavior. • Questions o Are components of behavioral models represented in brain structures? o Can insights into how the brain works improve economic modeling o Can those insights discriminate between alternative models? Fehr – Experimental and Behavioral Economics 2 Why is Neuroeconomics so fascinating? • Brain research has made great progress during the past decade, largely due to noninvasive techniques that allow observing the brain while it is active. • Systematic study of the relation between behavior and brain processes in healthy human subjects is possible. Fehr – Experimental and Behavioral Economics 3 Neuroscience may suggest fundamental changes in our primitive concepts (e.g. cognition-emotion interactions) Controlled Processes serial effortful evoked deliberately good introspective access Automatic Processes parallel reflexive effortless no introspective access Cognitive Affective Refinancing your house Economics deals mainly with this quadrant Actor who can truthfully induce his own emotions Tennis serve Jumping when someone says “boo”! Table from Camerer, Loewenstein & Prelec, in press, J. Economic Literature Fehr – Experimental and Behavioral Economics 4 Why brain details matter I • Two twins reveal a distaste for peanuts (example due to P. Romer) • Monica has nauseous “taste-aversion” from years ago… …but knows peanuts are safe o Automatic affect overrides controlled cognition • Claudia loves the smell of roasted nuts… …but knows she is allergic o Controlled cognition can override affect • Revealed preferences are the same… … but there are predictable differences: o Monica’s demand is more price-elastic o Intertemporal patterns (if Monica eats them once, she will eat more) o “Cure” for Monica is cognitive therapy & exposure, for Claudia is allergy treatment Fehr – Experimental and Behavioral Economics 5 Why brain details matter II: Automaticity • Priming effects (“prime” affects judgment/choice) o Push-polling • S Carolina primary ’00: `Would it make a difference to your vote if you knew John McCain had a child with a black prostitute?’ o “Are you planning to buy a car this year?” raises car sales o Environmental “cues” trigger drug craving (Laibson QJE 01) • • • Unconscious racial bias (measured by the implicit association test) correlates significantly with brain activity in DLPFC suggesting increased attempts to suppress racial bias cognitively. Economic behavior is the result of interactions among mechanisms in cells I-IV Cell I does not necessarily take over for big economic decisions o o o o war (“This is the guy that tried to kill my daddy”- Bush Jr.) bankruptcy (fear) marriage (lust?) status quo bias (“automatic”) in asset allocation Fehr – Experimental and Behavioral Economics 6 Methods • “Animal Models” – Many brain areas in humans and animals have similar structures. o It‘s possible to “produce” addicted rats. o Addiction is created in that part of the brain which we share with other mammals. Tierversuche o Single cell recording (i.e. measuring the electrical potentials of single neurons) is possible in non-human primates but not in healthy humans. • Psychophysiological measurements (skin conductance, heart rate) • Brain Imaging (EEG, PET, fMRI) o Observing the brain “when it works” • Studying humans with lesions o Associated deficits provide information about the function of the lesioned brain area. • Transcranial Magnetic Stimulation TMS o Enables a controlled, spatially and temporally limited, stimulation or inhibition of cortical brain areas. Fehr – Experimental and Behavioral Economics 7 Electro-encephalogram (EEG) • Measures electrical potentials at the scull • Very good temporal resolution but bad spatial resolution. • Large number of repetitions of the same situation are necessary. 128 electrode array Fehr – Experimental and Behavioral Economics 8 Positron Emission Tomography (PET) • A radioactive substance is injected into the blood. • This substance emits positrons. • These positrons decay, together with electrons, into two photons. • PET detects the brain area where this decay occurs, i.e., it detects the areas into which the radiation went. • Variants: Glucose with radioactive fluorine. Water with radioactive oxygen. “Pet measures blood flow” • Advantage: Little repetition of situations of interest is necessary • Disadvantage: subjects exposed to radioactivity; bad spatial and temporal resolution. Fehr – Experimental and Behavioral Economics 9 fMRI (Functional Magnetic Resonance Imaging) • • • • MRI is based on the principle that protons in a magnetic field align with the field. If the magnetic field is perturbed the direction of the protons is disturbed. When the protons are redirected in the magnetic field electromagnetic radiation is emitted and is detected by the scanner. Depending on the environment of the protons (high fat or water content) the signals generated by the redirection of the protons is different. fMRI uses the fact that hemoglobin (red blood cells) have different magnetic properties depending on whether there is little or much oxygen in the blood. Increased neuronal activity in the brain uses up oxygen such that initially the oxygen level in the activated area fall; later on the fall in oxygen is overcompensated for when oxygen-rich blood moves to the activated area. BOLD-Signal (blood oxygen level dependent signal) Fehr – Experimental and Behavioral Economics 10 Temporal resolution of fMRI • Blood flow has a lagged response to neural activity. (Hemodynamic response function HRF) • Does still allow relatively good temporal resolution because HRF is known. • Shortest stimuli that have been detected with fMRI: • Blamire et al. (1992): 2 sec • Bandettini (1993): 0.5 sec • Savoy et al (1995): 34 msec Fehr – Experimental and Behavioral Economics 11 Übersicht Fehr – Experimental and Behavioral Economics 12 Brain regions that could play an important role in economic experiments Fehr – Experimental and Behavioral Economics 13 The Neural Basis of Responder Behavior in the UG (A. G. Sanfey, J. K. Rilling, J. A. Aronson, L. E. Nystrom, J. D. Cohen, Sci 13 March ’03) • Responder’s brain activations are measured by fMRI in a $10 UG. • A responder faces each of three conditions ten times. o Offers from a (supposed) human partner o Random offers from a computer partner o Money offer (there is no proposer here) • Research Questions: Which brain areas are more activated when subjects face… o fair offers (3-5) relative to unfair offers (1-2). o the offer of a human proposer relative to a random computer offer. • Method (very simplified): o Regression of activity in every voxel (i.e, 3D Pixel) in the brain on the treatment dummy (i.e., unfair offer dummy, human proposer dummy) Fehr – Experimental and Behavioral Economics 14 Details of the Experiment Fehr – Experimental and Behavioral Economics 15 Differences in brain activity between unfair and fair offers from a human proposer (t-Statistic) dorsolateral prefrontal cortex. Bilateral anterior insula and anterior cingulate cortex. Dorsolateral prefrontal cortex. Results • Regions have stronger activations if subjects face unfair human offers relative to fair human offers (the same regions also show more activation if the unfair human offer is compared to unfair random offers). o Bilateral anterior Insula, anterior cingulate Cortex • Emotion-related region • Insula also has been associated with negative emotions such as disgust and anger. • Dorsolateral prefrontal Cortex (DLPFC) • Cognition-related region • Associated with control of execution of actions • Associated with achievement of goals. • Unfair offers are more likely to be rejected if insula activation is stronger. Neurobiology of Trust, Social Preferences and GainLoss Asymmetry – Some Results o Kosfeld et al. (submitted, 2004): Subjects who receive a dose of the hormone oxytocin exhibit more trusting but not more trustworthy behavior. Trust has a neurobiological basis. o Breiter et al. (Neuron 2001): The same realized outcome (e.g. $0) activates reward circuits (e.g. NAc) positively if the counterfactual outcomes are worse (e.g. $-6) and negatively if the counterfactual outcomes are better (e.g. $10). Brain seems to process outcomes as gains and losses relative to reference points (counterfactual outcomes). o Rilling et al. (Neuron 2002): Mutual cooperation with a human partner activates reward related neural circuits relative to mutual cooperation with a computer that provides the same monetary return. Mutual cooperation with humans is extra rewarding. Fehr – Experimental and Behavioral Economics 18 The Sweet Taste of Revenge The Neural Basis of Altruistic Punishment Dominique J.-F. de Quervain*, Urs Fischbacher*, Valerie Treyer, Melanie Schellhammer, Ulrich Schnyder Alfred Buck & Ernst Fehr University of Zurich Science 305 (2004), 1254-1258 *the first two authors contributed equally Fehr – Experimental and Behavioral Economics 19 The neural basis of punishment hypothesis • Evolutionary models (Boyd et al. 2003) and experimental evidence (Fehr&Gächter 2002) suggest that altruistic punishment is a key element in human cooperation. • Evidence from a host of experiments that many people incur costs to punish others for unfair behavior and for norm violations. • Theories of social preferences (Rabin 1993, Fehr& Schmidt 1999) capture this by assuming a taste for punishment of unfair behavior. o Subjects are better off (in terms of revealed preference) by incurring cost to punish others although the costs are not offset by economic rewards. Fehr – Experimental and Behavioral Economics 20 Hypotheses • Hypothesis 1: The possibility for punishing unfair behavior activates reward-related neural circuits. (Nucleus Accumbens, Nucleus Caudate). • Hypothesis 2: If punishment is costly for the punisher there is activation in neural circuits (PFC, OFC) that are related to the integration of separate cognitive operations (weighing of benefits and costs). Fehr – Experimental and Behavioral Economics 21 Striatum and Reward • Strong evidence from single cell recording in non-human primates that ventral striatum (nucleus accumbens) and dorsal striatum (caudate nucleus) are important in processing rewards. o Schultz et al. etc. • Strong evidence from neuroimaging studies with humans – using money as a reward medium – that ventral and dorsal striatum is a key component of reward circuits. o Delgado et al. 2004, Knutson et. al. 2000 &2001 • Dorsal striatum seems particularly involved in rewards that accrue as a result of purposeful behavior. o Schultz et al., O’Doherty et al 2004. Fehr – Experimental and Behavioral Economics 22 The Experimental Design • Two subjects, A and B, have the opportunity for a social exchange that benefits both parties. • However, to complete the exchange, A has to move first while B moves second, i.e., A has to trust B, and B can cheat A. • We inform A about B’s action and give A the opportunity to punish B. • A’s brain is scanned when he is informed about B’s action (i.e., whether B cheated him or not) and when A deliberates whether, and how much, he wants to punish B. Fehr – Experimental and Behavioral Economics 23 The three stage experiment (baseline treatment) • Stage 1: o Both A and B receive an endowment of 10 MUs (1 MU = CHF 0.1). o A decides whether to keep or to transfer his endowment to B. In case of a transfer B receives 4 times A’s endowment (40 MUs), in addition to his own endowment of 10 MUs. • Stage 2: o B has the option to keep all the points he possesses or to give half of his MUs to A • If A trusts and B is trustworthy both earn 25. • If A trusts and B cheats, A has nothing and B has 50. • Both have 10 if A does not trust and B keeps everything. Fehr – Experimental and Behavioral Economics 24 The three stage experiment (baseline treatment) • Stage 3: o Both players receive an additional endowment of 20 Mus o A has the opportunity to use this money to punish B or to keep the money. o Every MU invested into the punishment of B decreases B’s payoff by 2 MUs. Fehr – Experimental and Behavioral Economics 25 The three stage experiment (baseline treatment) • Parameters were chosen such that A has a strong incentive to transfer his endowment (quadrupling of transfer) while B has a strong incentive to defect (1:2 punishment technology). Pretesting ensured that A’s almost always transfer while the B‘s cheat in 50 – 60% of the cases. Fehr – Experimental and Behavioral Economics 26 The three stage experiment (treatment conditions) • Player A sequentially faced 7 different players B. o In 3 cases B cooperated (transferred half of his MUs). o In 3 cases B cheated (kept all his MUs). o In 1 case a random device forced B to “cheat”. A knew that in this case B’s “decision” was determined by rolling a die. • We asked pre-test subjects in the role of B whether we can use their decisions a second time. These decisions were then used in the PET-experiment as an input. Fehr – Experimental and Behavioral Economics 27 The three stage experiment o Between each of the 7 trials there was a break of 10 minutes. o We scanned all cases in which player B cheated or was forced to cheat. o Scanning time was 1 min. A’s were instructed to make a decision during this time. At the end of the scanning time subjects were asked whether they punish B and, if so, how much. o After each punishment condition A answered a questionnaire. Among other things A was asked whether he experienced a desire to reward or to punish player B (answers given on a 7-point likert scale from -3 to +3). o Each subject received a flat fee of CHF 150 for participation in the experiment plus the earnings from the interaction with player B. Fehr – Experimental and Behavioral Economics 28 The three stage experiment (treatment conditions) • From the viewpoint of stage 3 we have 4 treatments of interest in all of which player B cheated: o Treatment where B decides himself and punishment is costly for both A and B (Intentional & Costly, IC). A is hypothesized to experience a desire to punish cheating and he can in fact punish. o Treatment where B decides himself and punishment is free for A but costly for B (Intentional & Free, IF). A is hypothesized to experience a desire to punish cheating and he can in fact punish – even without a cost. o Treatment where B decides himself and punishment is only symbolic, i.e., A and B have no costs of punishing (Intentional & Symbolic; IS). A is also hypothesized to experience a desire to punish cheating but he cannot punish. o Treatment where B’s decision is randomly imposed on B; punishment is costly for both A and B (Non-intentional & Costly, NC). A is to have no or a much less of a desire to punish B for “non-cooperation” because B is not responsible for his “action”. Fehr – Experimental and Behavioral Economics 29 Hypotheses regarding activation of neural circuits • The following contrasts are hypothesized to show activation of reward related (N Accumbens, N caudate) neural circuits: o o o o o • (IC – IS) because no punishment possible in S. (IF – IS) because no punishment possible in S. (IC – NC) because no desire to punish in R. (IF – NC) because no desire to punish in R. (IC + IF) – (IS + NC) because in IS & NC there is either no possibility or no desire to punish so that there cannot be any satisfaction from punishing. The following contrast is hypothesized to show activation of cognition related (PFC, OFC) neural circuits. o (IC – IF) because in IC subjects must weigh the costs and benefits of punishing while in IF there are no costs. o Related to the Sanfey et al. (2003) hypothesis. Fehr – Experimental and Behavioral Economics 30 Rationale behind the treatment conditions • A skeptic might argue that whatever is differentially activated in the (IC – IS) contrast is due to cost of punishing for the punisher. • However, if we see the same activation in the (IF – IS) contrast, the activation cannot be due to this cost. • The possibility of many contrasts provides a robustness check. If the same reward-related circuits are activated in each of the mentioned contrasts we have a strong result. Fehr – Experimental and Behavioral Economics 31 intentional & costly IC intentional & free IF intentional & symbolic IS non-intentional & costly NC Perceived unfairness 3 Behavioral and Questionnaire Results 2 1 0 -1 -2 Actual payoff reduction imposed on B -3 40 Desire to punish 35 3 30 2 1 25 20 15 0 10 -1 -2 5 0 -3 Fehr – Experimental and Behavioral Economics 32 (IC + IF) - (IS + NC) activates the nucleus caudate Fehr – Experimental and Behavioral Economics 33 Differential activation of Nucleus Caudate relative to mean brain activation (IC + IF) - (IS + NC) R 2 6 5 4 3 2 1 0 Size of effect at [6, 22, 4] 7 Z value L intentional & costly IC intentional & free IF intentional & symbolic IS non-intentional & costly NC 1.5 1 0.5 0 -0.5 -1 -1.5 -2 Fehr – Experimental and Behavioral Economics 34 Differential activation across contrasts Table 1. PET results Contrast Region (BA) Coordinates Z value x y z Caudate nucleus 6 22 4 5.11* Thalamus 22 -24 10 4.43* Thalamus 22 -22 10 4.21 Caudate nucleus 6 22 4 3.55 Thalamus 22 -22 10 4.15 Caudate nucleus 6 24 2 3.70 IF-NC Caudate nucleus 6 22 4 4.18 IC-NC Caudate nucleus 6 22 4 4.23 IC-IF Ventromedial prefrontal / 2 54 -4 4.59 medial orbitofrontal cortex -4 52 -16 3.35 (IC+IF)-(IS+NC) IF-IS IC-IS The coordinates (x, y, z) locate the maxima of changes in blood flow. * indicate significant activations at a level < 0.05 corrected, otherwise p<0.001 uncorrected. BA, Brodmann area. I, intentional notransfer; N, non-intentional no-transfer; C, costly punishment; F, free punishment; S, symbolic punishment. Fehr – Experimental and Behavioral Economics 35 If caudate activity reflects satisfaction from punishment subjects with stronger caudate activation should punish more 0.08 Question: Does caudate activation cause punishment or does punishment cause caudate activation? Response at [10, 26, -2] in IC 0.06 0.04 0.02 0 -0.02 p < 0.001 -0.04 -0.06 0 2 4 6 8 10 12 14 16 18 20 Amount invested for punishing in IC Fehr – Experimental and Behavioral Economics 36 Are subjects with higher caudate activation in the IF condition willing to pay more for punishing in the IC condition? If caudate activation reflects satisfaction from punishment, subjects with stronger caudate activation in IF should be willing to pay more for punishment. 0.06 Response at [10, 26, 0] in IF 0.04 0.02 0 -0.02 -0.04 -0.06 0 2 4 6 8 10 12 14 16 18 20 Amount invested for punishing in IC Fehr – Experimental and Behavioral Economics 37 Decision-Making & PFC/OFC Activations • In IC and IS subjects face a decision problem o IC: Shall I pay money to punish the defector? o IS: Does it make sense to tell the defector that he is an bad person or that I dislike what he did? • In IF and NC subjects face less of a decision problem o IF: There is a desire to punish and subject can satisfy this desire without cost. Thus there is no trade off. o NC: There is no desire to punish, thus no trade off. • Hypotheses: In IC and IS we predict above average activation of PFC (BA 10) and OFC (BA 11) whereas in IF and NC we predict below average activation of these regions. Fehr – Experimental and Behavioral Economics 38 Does (IC – IF) activate the PFC/OFC? L R 7 6 4 3 Z value 5 2 1 0 Fehr – Experimental and Behavioral Economics 39 Ventromedial Prefrontal (BA 10) IC-IF & IC-NC are significant; IS-IF is weakly significant Medial OFC (BA 11) IC-IF IC-NC IS-IF All significant IC IF Fehr – Experimental and Behavioral Economics IS NC 40 Summary • Experimental evidence and social preference theories suggest that many people prefer to punish unfair acts. • Our study provides support for the view that satisfaction is associated with the punishment of unfair acts. • Correlation between caudate activity and punishment in IC, and between caudate activity in IF and punishment in IC supports this interpretation. • More difficult decisions lead to a stronger activation in the ventromedial PFC (BA 10) and the medial OFC (BA 11). Fehr – Experimental and Behavioral Economics 41 Summary • The activation of BA 10 and BA 11 in (IC-IF) supports our interpretation of the caudate activation because if punishment of defectors is not associated with satisfaction there are no offsetting benefits that need to be weighed against the cost of punishing. • Thus, revenge seems to be associated with “a sweet taste”. • Remark: This suggests that punishment of defectors is better modeled as a preference phenomenon rather than as a phenomenon of bounded rationality!!! Fehr – Experimental and Behavioral Economics 42 Other Questions and Results • Is ambiguity aversion better interpreted as a preference phenomenon or does it reflect bounded rationality? • Which brain mechanisms are associated with honesty and positive reciprocity? • Which brain mechanisms process trust, empathy, fear of punishment. • Is reputation a belief or a taste? Singer et al. (2004) show that just seeing the faces of people who previously cooperated as a 2nd mover in a trust game activates reward related brain regions. • What is the neural representation of beliefs about facts, beliefs about other people’s intentions, beliefs about other people’s actions? • Ultimate aim: Theories that have a sound basis in the insights of neurobiology and neuroscience. Fehr – Experimental and Behavioral Economics 43