Neuromarketing: the Hope and Hype of Neuroimaging in Business

PersPectives

science and society information about their true preferences. Such hidden information could, in theory, be used to influence their buying behaviour, Neuromarketing: the hope and hype so that the cost of performing neuroimaging studies would be outweighed by the benefit of neuroimaging in business of improved product design and increased sales. In theory, at least, brain imaging could Dan Ariely and Gregory S. Berns illuminate not only what people like, but also what they will buy. Abstract | The application of neuroimaging methods to product marketing — Thus far, this approach to neuromarketing neuromarketing — has recently gained considerable popularity. We propose that has focused on this post-design application, there are two main reasons for this trend. First, the possibility that neuroimaging in particular on measuring the effectiveness of advertising campaigns. The general will become cheaper and faster than other marketing methods; and second, the approach has been to show participants a hope that neuroimaging will provide marketers with information that is not product advertisement, either in the form obtainable through conventional marketing methods. Although neuroimaging is of a print advertisement or commercial, and unlikely to be cheaper than other tools in the near future, there is growing evidence measure the brain’s response in the form of a that it may provide hidden information about the consumer experience. The most blood oxygenation level-dependent (BOLD) measurement, which is taken as a proxy for promising application of neuroimaging methods to marketing may come before a neural activation. product is even released — when it is just an idea being developed. The second reason why marketers are excited about brain imaging is that they hope it will provide an accurate marketing Despite many common beliefs about the As in all compromises, these approaches research method that can be implemented inherently evil nature of marketing, the main provide solutions with intermediate levels even before a product exists (FIG. 1). The objective of marketing is to help match prod- of cost, simplicity, realism and quality of assumption is that neuroimaging data ucts with people. Marketing serves the dual data (TABLE 1). would give a more accurate indication of goals of guiding the design and presentation The incorporation of neuroimaging into the underlying preferences than data from of products such that they are more compat- the decision-making sciences — for example, standard market research studies and would ible with consumer preferences and facili- neuroeconomics — has spread to the realm remain insensitive to the types of biases tating the choice process for the consumer. of marketing. As a result, there are high that are often a hallmark of subjective Marketers achieve these goals by providing hopes that neuroimaging technology could approaches to valuations. If this is indeed product designers with information about solve some of the problems that market- the case, product concepts could be tested what consumers value and want before a ers face. A prominent hope is that neuro- rapidly, and those that are not promising product is created. After a product emerges imaging will both streamline marketing eliminated early in the process. This would on the marketplace, marketers attempt to processes and save money. Another hope is allow more efficient allocation of resources maximize sales by guiding the menu of that neuroimaging will reveal information to develop only promising products. offerings, choices, pricing, advertising and about consumer preferences that is unob- Thus, the issue of whether neuroimaging promotions. tainable through conventional methods. Of can play a useful part in any aspect of market- In their attempts to provide these types course, with such high expectations, there ing depends on three fundamental questions, of inputs, marketers use a range of market is the accompanying hype. Several popular which we will address in this paper. First, can research techniques, from focus groups books and articles have been published that neuromarketing reveal hidden information and individual surveys to actual market push a neuromarketing agenda, and there that is not apparent in other approaches? tests — with many approaches in between are now a handful of companies that market Second, can neuromarketing provide a more (see Supplementary information S1 (box)). neuromarketing itself 5. In this Perspective, efficient cost–benefit trade-off than other In general, the simpler approaches (focus we aim to distinguish the legitimate hopes marketing research approaches? Third, can groups and surveys) are easy and cheap to from the marketing hype. As such, we hope neuromarketing provide early information implement but they provide data that can that this article serves the dual purpose of rec- about product design? include biases, and are therefore seen as ognizing the real potential of neuro imaging in not very accurate1–4. The approaches that business and providing a guide for potential Revealing hidden information are more complex and therefore harder to buyers and sellers of such services. Brain activity and preference measurement. implement, such as market tests, provide Allowing for the assumption in neuro- more accurate data but incur a higher cost, Why use brain imaging for marketing? marketing that the brain contains hidden and the product, production and distribu- Marketers are excited about brain imaging information about preferences, it is reason- tion systems have to be in place for market for two main reasons. First, marketers hope able to set aside, for the moment, the issue tests to be conducted. There are some that neuroimaging will provide a more effi- of ‘hidden’ and ask what relationships are compromise approaches between these two cient trade-off between costs and benefits. known to exist between brain activity and extremes, which include simulated markets, This hope is based on the assumptions that expressed (that is, not hidden) preference. conjoint analyses, markets for information people cannot fully articulate their prefer- As it turns out, different methods of and incentive-compatible pricing studies ences when asked to express them explicitly, eliciting a person’s preference often result in (see Supplementary information S1 (box)). and that consumers’ brains contain hidden different estimations of that preference3,4,6,7.

Table 1 | comparison of selected marketing research approaches Focus groups Preference Simulated choice Market tests questionnaires methods What is measured Open-ended answers, Importance weighting for Choices among products Decision to buy and body language and various product attributes choice among products behaviour; not suitable for statistical analysis Type of response process Speculative, except when The respondent must try A hypothetical choice, An actual choice, with used to assess prototypes to determine his decision so the same process as customers’ own money, weightings through the actual purchase — and therefore fully introspection, then map but without monetary consequential those weightings into the consequences response scale Typical use in new-product Early on to aid general Design phase, when Design phase, when End of process, to forecast development processes product design; at user determining customer determining customer sales and measure interface design for trade-offs is important trade-offs is important; the response to other usability studies may also be used as a elements of marketing, forecasting tool such as price Cost and competitive risk Low cost; risk comes only Moderate cost and Moderate cost (higher High cost and high risk of from misuse of data by the some risk of alerting if using prototypes alerting competitors, plus seller competitors instead of descriptions) the risk of the product and some risk of alerting being reverse engineered competitors before launch Technical skill required Moderation skills for Questionnaire design and Experiment design Running an instrumented inside the group and statistical analysis and statistical analysis market and forecasting ethnographic skills for (including choice (highly specialized) observers and analysts modelling)

This makes it difficult to know which to both of these questions is positive, However, such similarities do not necessarily method provides the truest measure of neuromarketing could become useful for mean that brain activation is the same ‘decision utility’ (that is, the expected utility, measuring preferences. across different elicitation methods, and which would ultimately drive choice in the Measurements such as willingness to there are differences between the neural marketplace). It is clear that market tests give pay (WTP) have only recently come under activation representing decision utility and the most accurate answer, but having to run functional MrI (fMrI) examination. In one that representing experienced utility14,22,23. a market test on every product would defeat experiment, subjects bid on the right to eat This caveat aside, the generally close corre- the purpose of market research — namely, snacks during the experiment. The amount spondence does suggest that neural activity to provide early and cheap information. they were willing to pay (a measure of deci- might be used as a proxy for WTP Similarly, we suspect (and economists are sion utility) correlated with activity levels in in situations in which WTP cannot easily certain) that methods that are incentive the medial orbitofrontal cortex (OFC) and be determined — although this has yet to be compatible are better than methods that prefrontal cortex (PFC)9,10. Interestingly, demonstrated. are not. Incentive-compatible elicitation similar activation in the OFC has been methods are methods that encourage the observed when subjects anticipate a pleasant Reverse inference and reward. The practice participant to truthfully reveal what is taste11, look at pretty faces12, hear pleasant of measuring an increase in BOLD activity being asked of him because to do so would music13, receive money14,15 and experience a in a region such as the ventral striatum or maximize the participant’s satisfaction (for social reward16,17. Such generally close corre- OFC and then concluding that a ‘reward- example, he would earn the most money spondence in regional brain activity related’ process was active has become or receive the product he likes the best). In between the anticipation of rewarding increasingly common. This form of deduc- other words, it is in the participant’s interest events, the consumption of enjoyable goods tive reasoning is known as ‘reverse infer- to answer product-related questions truth- and the willingness to pay for them ence’24,25. Given the readiness of many to fully. However, using such methods is not suggests that the representation of expected interpret brain activation as evidence of a always possible. utility may rely, in part, on the systems that specific mental process, it is worth examin- One important question for the evaluate the quality of the consumption ing this type of inference. using a Bayesian potential of neuromarketing is whether experience. The theme of common systems analysis, it is possible to estimate the spe- the neural signal at the time of, or slightly for expectation and experience also applies cificity of activation in a particular region of before, the decision (assumed to be a to things that are unpleasant or even painful the brain for a specific cognitive process. For measure of decision utility) can be a good (although this involves a different network example, Poldrack used the BrainMap data- predictor of the pleasure or reward at the including the insula)18–21. Such similarities base to analyse the frequency of activation of time of consumption (the ‘experienced suggest that neuroimaging can become a use- Broca’s area in language studies24. He found utility’)8. A second question is whether the ful tool in measuring preferences, particularly that activation of Broca’s area implied a link between these two signals holds even when incentive compatibility is important but Bayes factor of 2.3 for language involvement, when the preference elicitation methods there is no easy way to achieve it (for example, which means that taking brain activity into are not incentive compatible. If the answer when the products have not been created). account can make a small but significant

improvement to one’s prior estimate of whether a language process was involved. fMRI Many studies have shown that striatal activity correlates with hedonic rating scales26. neuromarketers have been quick to invert this finding and use ventral striatal activity as an indication that an individual Concept Design Testing likes something; but what is the evidence for Market analysis • Development this? using Poldrack’s method to analyse the feasibility • Prototyping BrainMap database, we estimated the posterior probability for a reward process given Delivery the observation of nucleus accumbens (nAc) 27 activation . The prior probability of engag- Release ing a reward-related process was assumed to be 0.5 (1:1 odds). According to this estima- Support tion, based on the number of fMrI papers reported in the BrainMap database with and without ‘reward’, and with and without Feedback nAc activation, nAc activation increases the probability of a reward-related process fMRI taking place to 0.90 (odds 9:1). This yields Advertising a Bayes factor of 9, which is considered moderate to strong evidence for a causal Figure 1 | Product development cycle. Neuromarketing applications of functional MRI (fMRI) can relationship (BOX 1). Although meaningful potentially enter into the product development cycle in two places. InNa thetur first,e Re viefMRIws can| Neur beoscienc used ase in a statistical sense, the assumptions behind part of the design process itself. Here, neural responses could be used to refine the product before it such a calculation are rather liberal and is released. In the second, fMRI can be used after the product is fully designed, typically to measure neural responses as part of an advertising campaign to increase sales. may suffer from a publication bias for positive results as well as differing definitions of reward. In real-world settings, the ability to infer whether an individual likes some- efforts and consumer choices depend on data have been at the forefront of analysis thing based on nAc activation alone may be an array of neurobiological processes, and techniques. These techniques treat sites of substantially less. that no single brain region is responsible brain activity agnostically — that is, without In the context of a product likeability for a consumer choice. But is it possible that reference to prior hypotheses. The primary experiment, Knutson et al. found signifi- some brain regions are more involved than assumption is that, regardless of how an cant correlations between nAc activity and others? Because the field of neuroeconomics individual’s brain represents information, product preferences in college students28. grew out of early brain-imaging studies of it does so consistently. The representa- However, in logistic regression (r2) calcula- the neurobiology of reward31,32, most of the tions may be spatially dispersed, and they tions aimed at predicting consumer choice, neuroeconomic data are about valuation may be distributed differently in different self-reported preferences outperformed mechanisms and the associated responses individuals, but they can still be reliably brain activation alone. Adding brain activa- of dopamine-rich brain regions. The OFC detected through multi-voxel pattern analy- tion to a logistic model improved predic- and striatum have been consistently impli- sis (MVPA). Because MVPA methods are tions, but only slightly (increasing r2 from cated in goal-directed action9,22,33–35. It is also not reliant on the activation of a small subset 0.528 to 0.533). re-analysis with more generally accepted that the insula has a key of brain regions, they have substantially sophisticated machine-learning algorithms role in physiological arousal, which is typi- increased sensitivity to detect activation36. further improved the predictive value of cally, although not exclusively, aversive in A crucial advantage of MVPA techniques brain activation29. nature21. But because of the reverse inference over approaches in which activation in a Although some have argued for the exist- problem, using striatal and OFC activity particular brain region of interest is meas- ence of a “buy button” in the brain5, current as a read-out of ‘liking’ and the insula as a ured is that MVPA has the statistical power evidence suggests that the cognitive proc- ‘disgust-meter’ is probably too simplistic to to predict the individual choices of a subject. esses associated with purchase decisions are be of use in a real-life setting. In the context Because MVPA involves statistical associa- multi factorial and cannot be reduced to a of neuromarketing, the statistical power of tions of complex activation patterns that single area of activation. Conversely, a given these single-region correlations may be too occur when an individual choice is being brain region may be involved in multiple low for the correlations to be of use as pre- made, it does not depend on the vagaries cognitive processes. A recent review of value- dictors of consumption unless, perhaps, the of an experimenter interpreting the mean- based decision making divided the process neuroimaging data is combined with other ing of an activation map. Some of the most of making a choice into five categories: rep- measures of preference. impressive demonstrations of MVPA have resentation of the decision; assignment of been in decoding visual responses to simple value to different actions; action selection; fMRI as a brain decoder. Given the limited stimuli37–39 and subsequently, to watching outcome evaluation; and learning30. even power of reverse inference from single- films40, the meanings of nouns41, event within this simplified framework, current region brain activations, more data-driven boundaries of written narratives42 and city data suggest that responses to marketing methods for interpreting brain imaging navigation43,44.

Box 1 | NAc activation in studies of tasks with and without reward BOLD responses are influenced by so-called ‘expectation’ effects, which include The BrainMap database was searched for functional MRI studies with and without a reward task pricing effects, biases in the way the choice and with and without nucleus accumbens (NAc) activation. The NAc was defined as a bilateral is presented50 and placebo responses. This region of interest with vertices from MNI (Montreal Neurological Institute) coordinates (–12, 0, –12) suggests that neuromarketing could be to (12, 12, 0). The frequencies that were obtained are shown in the table below. helpful in identifying individual differences Assuming that the prior probability of engaging in a reward-related process is 0.5, calculations showed that NAc activation increases the probability of a reward-related process taking place to in consumer reactions to different types of 0.90, yielding a Bayes factor of 9: inputs. In a study of neural responses to sips of wine, medial OFC responses were higher Probability of NAc activation given a reward task = 27/68 = 0.397 when subjects were told that the wine was Probability of NAc activation given no reward task = 59/1283 = 0.046 expensive ($90 per bottle) versus inexpen- Assuming the prior probability of reward = 0.5, then sive ($5 per bottle)23. Activity in this region also correlated with self-report ratings of Probability of a reward task given NAc activation = how much participants liked the wine, even though all wines were actually the same. reward task No reward task These results suggest that the instantaneous experience of pleasure from a product — NAc activated 27 59 that is, experienced utility — is influenced NAc not activated 41 1,224 by pricing, and that this effect may be medi- ated by the medial OFC9. This result paral- lels a similar, behavioural finding that the It is possible, even likely, that such meth- marketing approaches is in the post-design strength of the placebo effect for analgesia is ods will soon be able to handle almost any phase, the goal of which is to increase sales greater for more expensive ‘medications’51. circumstance that can be created in an MrI of an existing product — for example, Subjects’ expectations also play an impor- environment. With increasing stimulus com- through advertisements and other types tant part in how the experimenter should plexity, simple interpretations of brain acti- of framing effects. early neuromarketing interpret striatal responses. Many studies vation will become more difficult. However, studies therefore used imaging approaches have shown that the reward-related signals for real-world marketing applications, it to evaluate consumer responses to adver- in the ventral striatum and nAc can be may be more important to predict future tisements. At this point, it is important more accurately linked to prediction errors behaviour than to understand the ‘why’ to distinguish between neural responses to for reward than to reward itself 22,52,53. of behaviour. Such a data-driven application the consumption of a product (that is, Placebo responses are an interesting of imaging (perhaps even lacking an under- experienced utility) and neural responses aspect of neuromarketing. The mecha- lying theory) is analogous to identifying a to representations of the product that may nism of the placebo response has been genetic polymorphism associated with a par- lead to future consumption. Only certain debated for decades54, but ultimately it can ticular cancer without understanding what types of products can be consumed in be considered an effect of marketing (that that gene does — which is likely to yield an MrI scanner. Therefore, much of the is, the actions of a doctor, pharmaceutical specific but not general insights. post-design neuromarketing literature has company or experimenter). The neural focused on brain responses to visual repre- correlates of the analgesic placebo effect costs and benefits sentations of products, such as pictures28,45 are widespread but generally point to a As noted above, it is not yet clear whether or advertisements for the product46–48; modulation of the cortical pain matrix neuroimaging provides better data than however, these advertisement studies, in the brain55,56. Because consumers can- other marketing methods (TABLE 1), but which used magneto encephalography not consciously report placebo effects, through the use of MVPA methods it might and electroencephalography (BOX 2), did the demonstration of neural correlates of be possible to reveal the ‘holy grail’ of hid- not link imaging data to actual purchase these effects suggests that having access to den information. Assuming that this is the decisions or other ratings, so it is not yet hidden brain information could enable a case, will using expensive neuroimaging possible to determine the value of this marketer to measure the effectiveness of ultimately be more efficient than using approach. a placebo marketing strategy in a particular cheaper methods? Typical charges for scan- individual. How well this type of informa- ning in a university research setting average The role of expectations. It has long been tion generalizes to a larger population will about uS$500 per hour. In a commercial known that the manner in which choices determine the cost–benefit ratio of doing setting, these will be higher. However, are presented can have a dramatic effect on neuroimaging. actual scan charges account for a small por- decisions49. This is where advertisements The aforementioned manipulations of tion of the total cost, with personnel and and product placement come into play. To expectations are simple and direct. For overhead expenses accounting for at least date, experiments have examined fairly example, the experimenter can manipulate 75% of the costs of an imaging project. If simple choices and responses to things that a single dimension of expectation, such as neuromarketing is to compete with conven- can be presented in an MrI scanner. Before price or descriptive words (for example, tional marketing approaches on the basis neuroimaging can be used to predict con- “ultra” and “new and improved”), and of efficiency, then the costs of labour and sumer choice, a greater understanding of measure the effect on the consumer behav- overheads will have to be reduced. the interplay between the decision maker, the iourally and neurally. More cognitively com- One area in which the cost of neuro- elicitation method and the decision context plex forms of expectations can be created imaging can be compared with conventional is needed. through advertisements and commercials.

Box 2 | Neuromarketing technologies minds of consumers. However, one fMrI study that compared brain responses to Functional Mri (fMri) persons and brands found that activation The technique uses an MRI scanner to measure the blood oxygenation level-dependent patterns for brands differed from those (BOLD) signal. The BOLD changes are generally correlated with the underlying synaptic activity. Spatial resolution is 1–10 mm, and temporal resolution is 1–10 s. In general, the higher the spatial for people — even for brands with which resolution, the lower the temporal resolution. Of the three imaging technologies described in this subjects are identified — suggesting that Box, fMRI has a substantial advantage in resolving small structures and those that are deep in the brands are not perceived in the same way brain. However, some important brain regions, especially the orbitofrontal cortex, are affected by as people59. Another possibility is that spe- signal artefacts that may reduce the ability to obtain useful information. State of the art MRI cific emotions can be elicited in response scanners cost approximately US$1 million per Tesla and have annual operating costs of to advertisements, although whether neuro- $100,000–$300,000. imaging will help to reveal these emo- electroencephalography (eeG) tions may ultimately be limited by reverse EEG uses electrodes applied to the scalp and measures changes in the electrical field in the brain inference constraints, especially if tied to region underneath. EEG has very high temporal resolution (milliseconds) and can therefore detect specific regions. brief neuronal events. Because the skull disperses the electrical field, EEG has low spatial resolution The issue of how culturally derived iden- (~1 cm) that depends on how many electrodes are used. The number of electrodes can be as few as tities become embedded in the brain is of two or range up to hundreds in high-density arrays. The greater the number of electrodes, the great interest, not only from a marketing better the spatial resolution. Apart from the low spatial resolution, EEG has poor sensitivity for deep brain structures. Equipment costs can be low (<$10,000) but increase with high-density arrays and perspective. Although neoclassical economic the concomitant resources needed to process the data. A common technique is to measure the theory describes a framework in which indi- left–right asymmetry of the frontal EEG78. This is typically measured by the power in the alpha band viduals assess costs and benefits during their (8–13 Hz). This research has suggested that relatively greater activity in the left frontal region is decision-making processes, it is clear that associated with either positive emotional states or the motivational drive to approach an object79. people base many decisions on sociocultural Although there are strong correlations between frontal EEG asymmetry and personality traits, the rules and identities. Some are in a commer- degree to which the asymmetry changes from moment to moment is still debated. Some have cial context (for example, “I am a PC” or “I 80 suggested a minimum of 60 s to reliably estimate power asymmetry , in which case the temporal am a Mac”) but many are not (for example, advantage of EEG over fMRI is lost. Although some have used this approach to measure momentary “I am a Democrat” or “I am a republican”). fluctuations in emotion in response to advertisements81, without accounting for autocorrelations in These issues extend beyond the mundane time or multiple statistical comparisons, the validity of such approaches is dubious. questions of advertisement effectiveness and Magnetoencephalography (MeG) raise the more profound question of how the An expensive cousin of EEG, MEG measures changes in the magnetic fields induced by neuronal marketing of ideas affects decision making. activity. Thus, MEG has the same advantage of high temporal resolution and, because the magnetic field is less distorted by the skull than is the electrical field, it has better spatial resolution than EEG. But whether neuroimaging provides an Like EEG, MEG is most sensitive to superficial cortical signals (primarily in the sulci). MEG requires a efficient tool to answer this question has yet magnetically shielded room and superconducting quantum interference detectors to measure the to be shown. weak magnetic signals in the brain. An MEG set-up costs approximately $2 million. early product design Transcranial magnetic stimulation (TMS) TMS uses an iron core, often in the shape of a toroid wrapped in electrical wire, to create a As the ability of neuroimaging to predict or magnetic field strong enough to induce electrical currents in underlying neurons when placed on influence post-design purchase decisions the head82. TMS can be used as a single pulse, paired pulse or repetitive stimulation, and the seems to be limited (see above), neuro- neuronal effects range from facilitation to inhibition of synaptic transmission. As a research tool, imaging may be better suited to gauging TMS has been used to study the causal role of specific brain regions in particular tasks by responses before products are marketed. temporarily taking them ‘offline’. The primary reason is that neuroimaging may yield insights into the product experience itself. Post-design applications of neuroimaging neuroimaging studies57,58, it is still unknown have, for the most part, confirmed what whether neuroimaging can prospectively Food products. Various food products and was known about the behavioural effects of reveal whether an advertisement will be beverages have been administered in the product placement, which bypass the coun- effective. In a famous Coke–Pepsi study, par- MrI scanner, from simple sugar solutions ter-arguments in which people naturally ticipants who described themselves as Coke to chocolate, wine, sports drinks and colas. engage when facing advertisements. The drinkers showed significant activation in the Beverages are particularly easy to administer, imaging studies confirm that there are neu- hippocampus and right dorsolateral PFC with the usual route through a computer- ral correlates of exposure to advertisements when they were cued about the upcoming controlled pump attached to a tube that but do not directly suggest that maximizing drink of Coke45. Self-described Pepsi drink- delivers controlled amounts of fluid into activity in a particular brain region results ers did not have this response. In the absence the participant’s mouth. The perception in more sales. of brand information, there was no signifi- of flavour is a multisensory integration cant difference in preference during a taste process and thus provides several oppor- Culture and advertising. neuroimaging test. The study suggested that any differences tunities for neuroimaging to disentangle a is often hyped as an exciting new tool in the response (behavioural and neural) to complex perception that subjects might not for advertisers. Despite its enormous the two brands must be culturally derived. be able to articulate; taste, odour, texture, cost, advertising effectiveness is a poorly One possibility is that brands achieve a life appearance and even sound all contribute understood area of marketing. Although of their own by becoming animate objects, to the gustatory experience. These differ- advertising has been investigated in a few sometimes with human attributes, in the ent dimensions have been mapped onto

Box 3 | The ethics of neuromarketing issues related to neuromarketing). Is it possible that such a neuroimaging approach The introduction of neuroimaging into an environment in which the ultimate goal is to sell more could create a ‘super-heroin of food’ — a product to the consumer may raise ethical issues. product so delicious that all but the most • Businesses will be able to read the minds of consumers. This concern is about the privacy of ascetic individuals would find it irresistible? thoughts. Can neuroimaging be used to gauge a person’s preferences outside of the specific task It is an extreme but real possibility. However, being performed? Possibly. This concern may be mitigated through transparency of purpose: that does not mean that neuroimaging is subjects must know what kind of endeavour they are helping, and their data should only be used necessarily problematic for food product for that purpose. development. Indeed, the same techniques • Private versus public information about preferences. Individuals need to be able to exercise control could be applied to making nutritious foods over what they choose to reveal about their personal preferences. A privacy breach occurs if more appealing. neuroimaging reveals a private preference that is outside the scope of the neuromarketer’s research question. Entertainment. As a typical big-budget • Information will be used to discriminate against individuals or exploit particular neurological traits found in a subgroup of individuals. Neuroimaging data could potentially target marketing to Hollywood film costs over $100 million, specific people or groups. Many people would find this tactic repugnant because it exploits a with almost as much spent on marketing, it biological ‘weakness’ that only exists in some people. Similarly, this information could be used would be surprising if film producers were to time pricing moves to capitalize on individual weaknesses that are known to coincide with not interested in using neuroimaging to particular biological states (for example, raising beverage prices when someone is known to improve their product. After static images, be thirsty). films are probably the easiest product to • Central versus peripheral routes of influence. A central route aims to influence consumers’ present in the scanner. Moreover, an fMrI preferences about the functional aspects of the product (for example, fewer calories in a beer). measurement is time locked to the film A peripheral route attempts to manipulate preferences through things that are peripherally related timeline. A film presents the same basic vis- to the product (for example, sex appeal of people in advertisements). Neuroimaging could ual and auditory stimuli to everyone viewing potentially be used to enhance both types of influence, but some consider the attempts to it and thus should serve as a cognitive syn- optimize the peripheral route more ethically dubious. chronizer. Indeed, an fMrI study of subjects • Brain responses obtained from a small group of subjects will be used to generalize to a large viewing a segment of the classic Western population. Of course, this is done all the time in the scientific literature. If neuromarketing data The Good, the Bad and the Ugly 40 showed are used in product design and the product injures someone, neuroimaging will be partly to blame. that large extents of the cortex responded 83 • Abnormal findings. Approximately 1% of the population will have an abnormality on their MRI . In similarly in time across subjects, suggesting a population without clinical symptoms, the clinical significance of an MRI abnormality is unknown. that much of the cortical response is essen- Many will be false positives; others will be real and require referral. Currently, there is no standard for how to handle these situations. However, it is standard practice to have a written policy in tially stereotypical. In another study, the place for abnormal findings. Failure to do so opens both the neuromarketing firm and their clients ability to recall narrative content of the TV to medical liability. sitcom Curb Your Enthusiasm three weeks • A lack of regulation. Traditional marketing methods, because they are not typically viewed as later was correlated with the strength of experimentation, have not been subject to institutional review board (IRB) oversight. MRI scans are hippocampal and temporal lobe responses 63 approved by the US Food and Drug Administration (FDA) for clinical use but, because no diagnosis during viewing . is being made in a marketing setting, there is the potential to circumvent both FDA and IRB Such stereotypical responses suggest that requirements. The burgeoning neuromarketing industry would be well advised to adopt an fMrI could be used during the editing proc- industry standard of independent review. Clients should demand it. ess. For example, different cuts of a movie • Management of perceptions. How will the public react when they discover that neuroimaging has could be measured against these cortical been used to design or market a product? The public’s response to genetically modified food could responses, which could then be used to select provide an indication. the final cut for release. Although it seems • Companies might not be primarily concerned with the best interests of the consumer. Companies hopelessly complex to interpret such brain and consumers maintain complex relationships in which some of their goals are compatible while responses, it may not be necessary if the only others are in conflict. On the one hand, companies seek to design, manufacture and sell products goal is to release the most profitable movie. that consumers seek to buy, resulting in compatible goals that benefit both parties. On the other Provided there were a metric of quality (for hand, companies also aim to maximize their short- or long-term profits, sometimes to the example, box office returns or test audience detriment of their consumers. Much like marketing itself, understanding consumer preferences reports), brain activation patterns could can be used for goals that are in the best interests of both the company and their consumers or for objectives that are in the interests of the company and to the detriment of their consumers. Which be chosen to optimize outcomes without approaches neuromarketers choose is an open question. any knowledge of what the patterns meant. Several neuromarketing companies have tar- geted their efforts towards the entertainment distinct brain regions but with substantial dimension of gustation is to be studied (for industry but, as most of this work is unpub- overlap60,61. The OFC is consistently linked example, taste, odour or texture) and maxi- lished, it is difficult to evaluate the quality of to perceived pleasantness, whereas viscos- mize a brain response to variations in that the product. However, guidelines for general ity and fat content seem to be represented dimension. quality of scientific work can be formu- in the insula62. The use of neuroimaging The drawback to such an approach is the lated based on two decades of neuroscience by commercial manufacturers to design possibility of creating food products that are research. Thus, without passing judgment on a more appealing food product is both so highly tuned to neural responses that whether neuromarketing works, we can at feasible and likely. For this to work, how- individuals may over-eat and become obese least identify the items to look for in a quality ever, one would need to identify which (see BOX 3 for a discussion of some ethical operation (BOX 4).

Box 4 | What to look for when hiring a neuromarketing firm Political candidates. Finally, neuromarketing might be applied to perhaps the greatest mar- We provide a list, which is by no means exhaustive, of what could be considered standard keting campaign of all: politics. According practice in the application of neuroimaging methods in cognitive neuroscience and related to the Federal election Commission (see fields. It is based on standard criteria for reviewing research proposals and adapted to a Further information), the cost of the 2008 uS business setting. Presidential race was approximately $1.6 • What is to be gained from neuroimaging? Good neuromarketers will begin by discussing the pros billion. It was also around that time that neu- and cons of the proposal in detail. For example: what will neuroimaging yield over traditional methods? Ask for data about the predictive value of neuroimaging findings in a real-world setting. roimaging made its way into politics, perhaps most prominently in the form of a New York • What are the dependent and independent measurements? Assessing brain activation is not Times op-ed piece67. Peer-reviewed studies generally useful without correlating it with some other measurement. It is necessary to have another behavioural measurement to anchor the interpretation of the brain activation. Be wary if have shown a complex pattern of activation someone claims to know what a person thinks based solely on brain activation. in response to statements about candidates; • How many subjects are needed? Apart from the simplest of tasks, any task invoking a response these patterns have been interpreted as evi- that is expected to vary across individuals demands a sample size of at least 30 (REF. 84). If groups dence that motivated reasoning involves acti- of individuals are being compared under different treatments or conditions, the sample size will vation in the ventromedial PFC, the anterior need to be much greater to detect differences between groups and between different treatments. cingulate cortex, the posterior cingulate cor- 68 • What is the nature of the stimuli? Simple stimuli are the easiest to analyse. Real-world images, as tex and the insula . Subsequent studies have might appear in an advertisement, become difficult to characterize unless one element at a time is suggested that activation of the medial PFC varied. For statistical power, a minimum of 10 repetitions within a stimulus category are required, might be associated with maintaining a although 20–30 would be more likely to achieve meaningful results. subject’s preference for a candidate in • What type of software will be used to analyse the neuroimaging data? Several software packages response to advertisements, whereas activity exist, and although these programmes make neuroimaging seem simple, it takes a minimum of in the lateral PFC might be associated with 69 1 year of training to be able to use them and 3 years to become fully competent. changing candidates . • How will motion correction be performed? In marketing terms, the political candidates are the products that must be sold to • Are conditions balanced in time? If not, how will subjects’ drifting attention be compensated for? the electorate. Therefore, like other products, • Is this a whole-brain analysis or is a specific part of the brain being examined? These necessitate candidates and their campaigns have pre- different thresholds of identifying activation. The chance of an activation appearing somewhere in and post-design phases. Political marketing the brain is high due to random noise. is aimed at selling an existing candidate but, • Will regions of interest be defined a priori? If so, what is the justification for this? Conclusions with more foresight, can also be used to based on activation of a single region will have relatively little predictive power over conventional ‘design’ a better candidate. The aforemen- behavioural methods. tioned neuroimaging studies have focused • If multi-voxel pattern analysis (MVPA) methods will be used, will they be completely data-driven on the post-design responses to advertise- (principal component analysis or independent component analyses) or will they be based on 68,69 classifier training of subject responses (support vector machine, relevance vector regression or ments for political candidates . Gaussian process regression)? How will the resulting activity maps be interpreted? Could neuroimaging also be used to design a candidate? Although potential • How robust are the results? Ask for a ‘bootstrap’ — for example, testing on a ‘fresh’ subsample of data. nominees already go through a ‘grooming’ process, it is worth examining this prospect. • What type of scanner will be used? Either 1.5 or 3 Tesla scanners can yield images of acceptable A candidate’s appearance, trustworthiness quality. Open MRIs do not have the field homogeneity or the gradient technology necessary for fMRI. What quality control checks are performed to make sure the scanner is operating optimally and message content might determine a and consistently from day to day? What steps will be taken to minimize signal artefacts in areas voter’s decision. Considerable neuroimag- with poor signal? ing work has been done on the perception of human faces70 and features such as facial symmetry, skin colour and attractiveness. Architecture. A growing number of neuro- has already been used to understand neural Key brain structures in visual processing scientists and architects have begun to activation during automobile driving65,66. include the fusiform face area for basic consider the relationships of the brain to the In spatial navigation tasks such as driving, face processing71, the superior temporal architectural experience64. The neuroscience and presumably navigating a building, the sulcus for gaze direction and intention of architecture could be considered from two hippo campus has a key role. These early and the nAc for attractiveness12. A recent perspectives: first, the neural activity associ- virtual reality experiments suggested that study on the effect of political candidates’ ated with seeing specific aspects of a build- the hippocampus is active when the subject appearance found that insula activation in ing; and second, the use of neural responses makes navigation decisions but not when response to seeing a picture of a candidate to guide the architectural design process. they are externally cued65. Perhaps taking into was associated with a greater likelihood Clearly, one would need to identify these neu- account ‘hippocampal load’ may be a useful of that candidate losing the election72. In ral responses before attempting to use them tool in architectural design — for example, to addition, dorsolateral PFC and anterior in architectural design, but it is precisely the make buildings easier to navigate. extending cingulate cortex activation occurred when application in design that places neuroimag- this idea by considering the neurobiological subjects viewed images of a candidate ing within the neuromarketing framework. changes associated with ageing, it might be of a political party different from their Virtual reality can provide a surprisingly possible to design buildings and retirement own73. The neurobiology of trust has also accurate simulation of an architectural experi- communities that mitigate the memory loss become quite popular to study with both ence and can be used in an MrI scanner. It associated with Alzheimer’s disease. fMrI and, more recently, pharmacological

manipulations74–76. These studies have Finally, we return to the opening ques- everyday economic transactions. J. Neurosci. 27, 9984–9988 (2007). found that different dimensions of trust, tion: hope or hype? It is too early to tell but, 10. Hare, T. A., Camerer, C. F. & Rangel, A. Self-control such as reputation, fairness and uncertainty, optimists as we are, we think that there is in decision-making involves modulation of the vmPFC valuation system. Science 324, 646–648 (2009). correlate with activity in different brain much that neuromarketing can contrib- 11. O’Doherty, J. P., Deichmann, R., Critchley, H. D. & regions. Moreover, the hormone oxytocin ute to the interface between people and Dolan, R. J. Neural responses during anticipation of a primary taste reward. Neuron 33, 815–826 (2002). affects human behaviour in various eco- businesses and in doing so foster a more 12. Aharon, I. et al. Beautiful faces have variable reward nomic exchanges that depend on social human-compatible design of the prod- value: fMRI and behavioral evidence. Neuron 32, 537–551 (2001). 77 interactions . Finally, a candidate’s message ucts around us. At the same time, neuro- 13. Zatorre, R. J., Chen, J. L. & Penhume, V. B. When content could be viewed as an experiential marketing as an enterprise runs the risk of the brain plays music: auditory-motor interactions in music perception and production. Nature Rev. product. One could theoretically attempt quickly becoming yesterday’s fad. Seasoned Neurosci. 8, 547–558 (2007). to maximize striatal and OFC responses marketers still remember the hype around 14. Knutson, B., Adams, C. M., Fong, G. W. & Hommer, D. Anticipation of increasing monetary reward selectively to platform statements although, for the subliminal advertising, which quickly faded recruits nucleus accumbens. J. Neurosci. 21, RC159 reasons stated above, this is not necessarily and died despite the research interest that (2001). 15. O’Doherty, J., Kringelbach, M. L., Rolls, E. T., Hornak, J. predictive of success. surrounded it (and research on subliminal & Andrews, C. Abstract reward and punishment priming remains a large part of academic representations in the human orbitofrontal cortex. Nature Neurosci. 4, 95–102 (2001). conclusions and future directions research in social psychology). How can 16. Izuma, K., Saito, D. N. & Sadato, N. Processing of neuromarketing has received considerable we make sure that neuromarketing will not social and monetary rewards in the human striatum. Neuron 58, 284–294 (2008). attention in both the scientific community suffer a similar fate? For one, the academic 17. Rilling, J. K. et al. A neural basis for social and the media. Although few scientific community should take this topic seriously cooperation. Neuron 35, 1–20 (2002). 18. Ploghaus, A., Becerra, L., Borras, C. & Borsook, D. neuro marketing studies have been con- and not leave it to the neuromarketers and Neural circuitry underlying pain modulation: ducted, the existing evidence suggests that the op-ed page of the New York Times. expectation, hypnosis, placebo. Trends Cogn. Sci. 7, 197–200 (2003). neuroimaging could be used advanta- We should also ask deeper questions on 19. Ploghaus, A. et al. Dissociating pain from its geously in several domains of marketing. how marketing works — and not simply anticipation in the human brain. Science 284, 1979–1981 (1999). For a marketer, neuroimaging could be examine whether type X of advertising 20. Koyama, T., McHaffie, J. G., Laurienti, P. J. & attractive because it might be cheaper works better or worse than type Y. If we Coghill, R. C. The subjective experience of pain: where expectations become reality. Proc. Natl Acad. Sci. USA and faster than current marketing tools, and take neuromarketing as the examination of 102, 12950–12955 (2005). because it could provide hidden informa- the neural activities that underlie the daily 21. Craig, A. D. How do you feel? Interoception: the sense of the physiological condition of the body. Nature Rev. tion about products that would otherwise activities related to people, products and Neurosci. 3, 655–666 (2002). be unobtainable. We think it unlikely that marketing, this could become a useful and 22. Hare, T. A., O’Doherty, J., Camerer, C. F., Schultz, W. & Rangel, A. Dissociating the role of the orbitofrontal neuroimaging will be more cost-effective interesting path for academic research cortex and the striatum in the computation of goal than traditional marketing tools, and so and at the same time provide useful inputs values and prediction errors. J. Neurosci. 28, 5623–5630 (2008). the first point is mostly hype. However, to marketers. 23. Plassmann, H., O’Doherty, J., Shiv, B. & Rangel, A. continuing developments in analytical Marketing actions can modulate neural Dan Ariely is at the Fuqua School of Business, representations of experienced pleasantness. tools for neuroimaging data — for example, Center for Cognitive Neuroscience, Department of Proc. Natl Acad. Sci. USA 105, 1050–1054 (2008). 24. Poldrack, R. A. Can. cognitive processes be inferred MVPA — suggest that neuro imaging will Economics, and the Department of Psychiatry and from neuroimaging data? Trends Cogn. Sci. 10, 59–63 soon be able to reveal hidden information Behavioural Sciences, Duke University, Durham, (2006). about consumer preferences. Although North Carolina 2770, USA. 25. Poldrack, R. A. The role of fMRI in cognitive neuroscience: where do we stand? Curr. Opin. this information could boost post-design Gregory S. Berns is at the Department of Psychiatry Neurobiol. 18, 223–227 (2008). sales efforts, we think that the real pay-off and Behavioural Sciences, Economics Department, 26. Delgado, M. R. Reward-related responses in the human striatum. Ann. NY Acad. Sci. 1104, 70–88 will come during the design process. using Center for Neuropolicy, Emory University, Atlanta, Georgia 30322, USA. (2007). fMrI data during design could affect a 27. Fox, P. T. & Lancaster, J. L. Mapping context and content: the BrainMap model. Nature Rev. Neurosci. wide range of products, including food, Correspondence to G.S.B. e-mail: [email protected] 3, 319–321 (2002). entertainment, buildings and political 28. Knutson, B., Rick, S., Wimmer, G. E., Prelec, D. & doi:10.1038/nrn2795 Loewenstein, G. Neural predictors of purchases. candidates. Neuron 53 Published online 3 March 2010 , 147–156 (2007). There are two sides to the use of such 29. Grosenick, L., Greer, S. & Knutson, B. Interpretable 1. Beckwith, N. E. & Lehmann, D. R. The importance classifiers for FMRI improve prediction of purchases. information. Product manufacturers could of halo effects in multi-attribute attitude models. IEEE Trans. Neural Syst. Rehabil. Eng. 16, 539–548 use neural information to coerce the public J. Mark. Res. 12, 265–275 (1975). (2008). 2. Day, G. S. The threats to marketing research. 30. Rangel, A., Camerer, C. & Montague, P. R. into consuming products that they neither J. Mark. Res. 12, 462–467 (1975). A framework for studying the neurobiology of need nor want. However, we hope that 3. Griffin, A. & Hauser, J. R. The voice of the customer. value-based decision making. Nature Rev. Neurosci. 9, Mark. Sci. 12, 1–27 (1993). 545–556 (2008). future uses of neuromarketing will help 4. Green, P. E. & Srinivasan, V. Conjoint analysis in 31. Camerer, C., Loewenstein, G. & Prelec, D. companies to identify new and exciting marketing: new developments with implications for Neuroeconomics: how neuroscience can inform research and practice. J. Mark. 54, 3–19 (1990). economics. J. Econ. Lit. 43, 9–64 (2005). products that people want and find useful. 5. Lindstrom, M. Buyology. Truth and Lies About Why 32. Glimcher, P. W. Decisions, decisions, decisions: One example is a new trend in ‘user design’ We Buy (Doubleday, New York, 2008). choosing a biological science of choice. Neuron 36, 6. Hauser, J. R. & Shugan, S. M. Intensity measures 323–332 (2002). in which companies allow consumers to of consumer preference. Oper. Res. 28, 278–320 33. Yin, H. H., Ostlund, S. B. & Balleine, B. W. Reward- participate, through the internet, in the (1980). guided learning beyond dopamine in the nucleus 7. Buchanan, B. & Henderson, P. W. Assessing the bias accumbens: the integrative functions of cortico-basal design of new products and by doing so of preference, detection, and identification measures ganglia networks. Eur. J. Neurosci. 28, 1437–1448 create products that are more useful for the of discrimination ability in product design. Mark. Sci. (2008). 11, 64–75 (1992). 34. Padoa-Schioppa, C. & Assad, J. A. Neurons in the companies and for their customers. Perhaps 8. Kahneman, D., Wakker, P. P. & Sarin, R. Back to orbitofrontal cortex encode economic value. Nature a next phase in user design is one that incor- Bentham? Explorations of experienced utility. 441, 223–226 (2006). Q. J. Econ. 112, 375–405 (1997). 35. Schoenbaum, G. & Roesch, M. Orbitofrontal cortex, porates not only what consumers express, 9. Plassmann, H., O’Doherty, J. & Rangel, A. associative learning, and expectancies. Neuron 47, but also what they think. Orbitofrontal cortex encodes willingness to pay in 633–636 (2005).

36. Norman, K. A., Polyn, S. M., Detre, G. J. & 55. Wager, T. D. et al. Placebo-induced changes in fMRI in 74. Fehr, E. & Camerer, C. Social neuroeconomics: Haxby, J. V. Beyond mind-reading: multi-voxel the anticipation and experience of pain. Science 303, the neural circuitry of social preferences. Trends pattern analysis of fMRI data. Trends Cogn. Sci. 10, 1162–1167 (2004). Cogn. Sci. 11, 419–427 (2007). 424–430 (2006). 56. Benedetti, F., Mayberg, H. S., Wager, T. D., Stohler, C. S. 75. Kosfeld, M., Heinrichs, M., Zak, P. J., Fischbacher, U. 37. Haynes, J.D. & Rees, G. Decoding mental states from & Zubieta, J. K. Neurobiological mechanisms of the & Fehr, E. Oxytocin increases trust in humans. activity in humans. Nature Rev. Neurosci. 7, 523–534 placebo effect. J. Neurosci. 25, 10390–10402 (2005). Nature 435, 673–676 (2005). (2006). 57. Kenning, P. H. & Plassmann, H. How neuroscience can 76. King-Casas, B. et al. Getting to know you: reputation 38. Kamitani, Y. & Tong, F. Decoding the visual and inform consumer research. IEEE Trans. Neural Syst. and trust in a two-person economic exchange. Science subjective contents of the human brain. Nature Rehabil. Eng. 16, 532–538 (2008). 308, 78–83 (2005). Neurosci. 8, 679–685 (2005). 58. Lee, N., Broderick, A. J. & Chamberlain, L. What is 77. Baumgartner, T., Heinrichs, M., Vonlanthen, A., 39. Kay, K. N., Naselaris, T., Prenger, R. J. & Gallant, J. L. neuromarketing? A discussion and agenda for future Fischbacher, U. & Fehr, E. Oxytocin shapes the Identifying natural images from human brain activity. research. Int. J. Psychophysiol. 63, 199–204 neural circuitry of trust and trust adaptation in Nature 452, 352–356 (2008). (2007). humans. Neuron 58, 639–650 (2008). 40. Hasson, U., Nir, Y., Levy, I., Fuhrmann, G. & 59. Yoon, C., Gutchess, A. H., Feinberg, F. & Polk, T. A. 78. Davidson, R. J., Ekman, P., Saron, C. D., Senulis, J. A. & Malach, R. Intersubject synchronization of A functional magnetic resonance imaging study of Friesen, W. V. Approach-withdrawal and cerebral cortical activity during natural vision. Science neural dissociations between brand and person asymmetry: emotional expression and brain physiology I. 303, 1634–1640 (2004). judgments. J. Consum. Res. 33, 31–40 (2006). J. Pers. Soc. Psychol. 58, 330–341 (1990). 41. Mitchell, T. M. et al. Predicting human brain activity 60. Small, D. M. & Prescott, J. Odor/taste integration 79. Harmon-Jones, E. Clarifying the emotive associated with the meanings of nouns. Science and the perception of flavor. Exp. Brain Res. 166, functions of asymmetrical frontal cortical activity. 320, 1191–1195 (2008). 345–357 (2005). Psychophysiology 40, 838–848 (2003). 42. Speer, N. K., Zacks, J. M. & Reynolds, J. R. 61. Rolls, E. T. Brain mechanisms underlying flavour and 80. Huster, R. J., Stevens, S., Gerlach, A. L. & Rist, F. Human brain activity time-locked to narrative even appetite. Philos. Trans. R. Soc. Lond. B Biol. Sci. 361, A spectroanalytic approach to emotional boundaries. Psychol. Sci. 18, 449–455 (2007). 1123–1136 (2006). responses evoked through picture presentation. 43. Spiers, H. J. & Maguire, E. A. Spontaneous mentalizing 62. De Araujo, I. E. T. & Rolls, E. T. The representation Int. J. Psychophysiol. 72, 212–216 (2008). during an interactive real world task: an fMRI study. in the human brain of food texture and oral fat. 81. Ohme, R., Reykowska, D., Wiener, D. & Choromanska, A. Neuropsychologia 44, 1674–1682 (2006). J. Neurosci. 24, 3086–3093 (2004). Analysis of neurophysiological reactions to advertising 44. Spiers, H. J. & Maguire, E. A. Decoding human brain 63. Hasson, U., Furman, O., Clark, D., Dudai, Y. & stimuli by means of EEG and galvanic skin response activity during real-world experiences. Trends Cogn. Davachi, L. Enhanced intersubject correlations measures. J. Neurosci. Psychol. Econ. 2, 21–31 Sci. 11, 356–365 (2007). during movie viewing correlate with successful (2009). 45. McClure, S. M. et al. Neural correlates of behavioral episodic encoding. Neuron 57, 452–462 (2008). 82. Kobayashi, M. & Pascual-Leone, A. Transcranial preference for culturally familiar drinks. Neuron 44, 64. Eberhard, J. P. Applying neuroscience to architecture. magnetic stimulation in neurology. Lancet Neurol. 379–387 (2004). Neuron 62, 753–756 (2009). 2, 145–156 (2003). 46. Ambler, T., Ioannides, A. & Rose, S. Brands on the 65. Spiers, H. J. & Maguire, E. A. Neural substrates 83. Illes, J. et al. Incidental findings in brain imaging brain: neuro-images of advertising. Bus. Strategy Rev. of driving behaviour. NeuroImage 36, 245–255 research. Science 311, 783–784 (2006). 11, 17–30 (2000). (2007). 84. Thirion, B. et al. Analysis of a large fMRI cohort: 47. Rossiter, J. R., Silberstein, R. B., Harris, P. G. & Nield, G. 66. Calhoun, V. D. et al. Different activation dynamics in statistical and methodological issues for group Brain-imaging detection of visual scene encoding in multiple neural systems during simulated driving. analyses. NeuroImage 35, 105–120 (2007). long-term memory for TV commercials. J. Advert. Res. Hum. Brain Map. 16, 158–167 (2002). 41, 13–22 (2001). 67. Freedman, J. This is your brain on politics. New York Acknowledgements 48. Astolfi, L. et al. Neural basis for brain responses to TV Times (18 Jan 2005). This work was supported by grants to G.S.B. from the commercials: a high-resolution EEG study. IEEE Trans. 68. Westen, D., Blagov, P. S., Harenski, K., Kilts, C. & National Institute on Drug Abuse (R01DA016434 and Neural Syst. Rehabil. Eng. 16, 522–531 (2008). Hamann, S. Neural bases of motivated reasoning: R01DA025045), the Office of Naval Research and Air Force 49. Tverksy, A. & Kahneman, D. The framing of decisions an FMRI study of emotional constraints on partisan Office of Scientific Research, and the National Science and the psychology of choice. Science 211, 453–458 political judgment in the 2004 US Presidential Foundation (BCS0827313). (1981). election. J. Cogn. Neurosci. 18, 1947–1958 (2006). 50. De Martino, B., Kumaran, D., Seymour, B. & 69. Kato, J. et al. Neural correlates of attitude change Competing interests statement Dolan, R. J. Frames, biases, and rational decision- following positive and negative advertisements. The authors declare competing financial interests: see web making in the human brain. Science 313, 684–687 Front. Behav. Neurosci. 3, 6 (2009). version for details. (2006). 70. Tsao, D. Y. & Livingstone, M. S. Mechanisms of face 51. Waber, R. L., Shiv, B., Carmon, Z. & Ariely, D. perception. Ann. Rev. Neurosci. 31, 411–437 (2008). Commercial features of placebo and therapeutic 71. Kanwisher, N. & Yovel, G. The fusiform face area: FURtHeR inFoRMation efficacy. JAMA 299, 1016–1017 (2008). a cortical region specialized for the perception of Dan Arieley’s homepage: 52. Montague, P. R. & Berns, G. S. Neural economics faces. Philos. Transact. Roy. Soc. B Biol. Sci. 361, http://www.predictablyirrational.com/ and the biological substrates of valuation. Neuron 2109–2128 (2006). Gregory s. Berns’s homepage: 36, 265–284 (2002). 72. Spezio, M. L. et al. A neural basis for the effect of http://www.neuropolicy.emory.edu/overview.html 53. Schultz, W., Dayan, P. & Montague, P. R. A neural candidate appearance on election outcomes. Soc. Federal election commission: http://www.fec.gov/ substrate of prediction and reward. Science 275, Cogn. Affect. Neurosci. 3, 344–352 (2008). 1593–1599 (1997). 73. Kaplan, J. T., Freedman, J. & Iacoboni, M. Us versus sUPPLeMentaRy inFoRMation 54. Colloca, L. & Benedetti, F. Placebos and painkillers: them: political attitudes and party affiliation influence See online article: S1 (box) is mind as real as matter? Nature Rev. Neurosci. 6, neural response to faces of presidential candidates. All liNkS Are ACTive iN The oNliNe PdF 545–552 (2005). Neuropsychologia 45, 55–64 (2007).