It occurs to me that maybe we're heavily invested in finding answers to which we don't know the corresponding questions. Maybe the availability of the new technology is running the science rather than the other way round.
(Fodor Reference Fodor1999)I contribute a series of (numbered and partially ordered) notes and statements. For good measure, I also throw in a question and a modest suggestion. Some statements attempt to encode my reading of what we know, some statements are opinionated and speculative. NE, BE, and EE are abbreviations for neuro-, behavioural, and experimental economics. Disclaimer: I'm all in favour of prospecting the brain . . . at the right time . . . with the right questions . . . and the right techniques.Footnote 1
0. A definition
prospect (from: www.m-w.com)
intransitive verb : to explore an area especially for mineral deposits
transitive verb : to inspect (a region) for mineral deposits; broadly: explore
1. Our synaptic self, how it comes about, and where it ends.
Environmental stimuli (visual, auditory, olefactory, etc.) determine to quite some extent who we become (are): “You are your synapses”Footnote 2 (LeDoux Reference LeDoux2002: ix, 323, 324). “They are who you are” (LeDoux Reference LeDoux2002: 324).
1.1. The qualifier “to some extent” is important: We may be our synapses, and they may be who we are, but that does not mean necessarily that our neurons make us do it. (See Murphy and Brown Reference Murphy and Brown2007, and LeDoux Reference LeDoux2002: specifically ch. 11.) For the most part, however, I shall not pursue issues of free will here any further as they are tangential for the thrust of my contribution (but see 2; see also Harrison Reference Harrison2008).
1.2. (Important) environmental stimuli are encoded in neural networks that constitute our memory of them. Similar stimuli are encoded by similar networks that can be thought of as patterns of neural networks, or attractors, or generic memories, i.e. memories that encode the essence of similar constructs or experiences such as a hat, or having (been) cheated in a principal-agent situation. A store of generic memories is assembled over time. The neuronal substrates of external stimuli, or outer worlds, make possible the neuronal substrates of internal stimuli, or inner worlds, which are a prerequisite for consciousness and free will.Footnote 3 (See Murphy and Brown Reference Murphy and Brown2007, and LeDoux Reference LeDoux2002: specifically ch. 11, see also Goldberg Reference Goldberg2001, Reference Goldberg2005.) The inner worlds strike me as a likely contributor to the (unsteady) ground level of activations that neuroscientists seem to have started paying attention to only recently (Wang et al. Reference Wang, Jiang, Yu, Tian, Li, Liu, Zhou, Xu, Song and Li2007). See below note 15.3.
2. The unbearable lightness of being and its potential costs
As John Lennon once put it, “Life is what happens to you while you are busy making other plans.” (“Beautiful Boy”): Stimuli that we encounter are – rather automatically – matched with the store of generic memories. How we (manage to) snap out of routine processing is a fascinating problem without, at this point, a clear answer.
2.1. If a match is found then automatic, routine (and parallel) processing of the stimulus (and its various facets: sights, sounds, smells) is triggered that often is “subconscious” and hence beyond our control.
2.2. Up to 95% of our daily behaviour is prompted by automatic, routine processing (e.g., Baumeister and Sommer Reference Baumeister, Sommer and Wyer1997). Whatever the exact number, the automaticity of everyday life is a well-established and undisputed fact (e.g. Bargh and Chartrand Reference Bargh and Chartrand1999).
2.3 Because it economizes on our cognitive resources, automatic, routine processing has important benefits. Automatic, routine processing has also potentially important costs: By matching stimuli routinely with a store of generic memories, it might make us susceptible to stereotyping, “cognitive illusions”, and misreadings of laboratory one-shot scenarios for what they are (Ortmann and Hertwig Reference Ortmann and Hertwig2000; Binmore Reference Binmore2007a, Reference Binmore2007b; Levitt and List Reference Levitt and List2007).
2.4. If a match is not found, then less automatic, less routine, and quite deliberate (and sequential) processing of the stimulus (and its various facets) might get triggered.
2.5. The process of switching from automatic, routine processing to deliberate processing of stimuli seems not well understood, but it is likely to be triggered by mismatches of stimuli with the store of generic memories, the expectations that store creates, and the self-monitoring that we continuously do (Goldberg Reference Goldberg2005: especially 201–35; see also Anderson forthcoming and Glimcher et al. Reference Glimcher, Dorris and Bayer2005).
3. Routinization
A process of routinization condenses our experiences (e.g. for some fascinating evidence, Goldberg Reference Goldberg2005: fig. 14 and 15, 204–13 and also ch. 11).Footnote 4 Deliberation and the analysis of challenges (cognitive and emotional, on this pair of descriptors more below) seem to involve the right hemisphere's prefrontal cortex significantly; as we experience similar situations again and again, we see the activations that build up generic memories drift to the left hemisphere's subcortical limbic structures.
4. Intuition is the result of routinization
Intuition is “the condensation of vast prior analytic experience; it is analysis compressed and crystallized. . . . intuitive decision making is postanalytic, rather than preanalytic or nonanalytic. It is the product of analytic processes being condensed to such a degree that its internal structure may elude even the person benefiting from it. . . . The intuitive decision-making of an expert bypasses orderly, logical steps precisely because it is a condensation of extensive use of such orderly steps in the past. It is the luxury of mental economy conferred by vast prior experience” (Goldberg Reference Goldberg2005:150, 152; see also ch. 8). That vast prior experience is, of course, the store of generic memories that gets consulted for guidance.
5. Emotions are the result of routinization
What is true for intuition is true for some emotions (Murphy and Brown Reference Murphy and Brown2007: 24–7; see also Goldberg Reference Goldberg2005: 185–235; and, for that matter, Glimcher et al. Reference Glimcher, Dorris and Bayer2005 and Pessoa Reference Pessoa2008). As it is, there is a long-standing debate in psychology on the cognitive foundation of emotions (e.g. Ortony et al. Reference Ortony, Clore and Collins1988) of which neuroeconomists seem curiously unaware. Overall, it seems that little progress has been made since Adam Smith lucidly discussed the nature and causes, and proposed a classification, of emotions (see Meardon and Ortmann Reference Meardon and Ortmann1996a, Reference Meardon and Ortmann1996b).
6. Affective–cognitive divide? Not!
An important implication of 5. and 4. is: The alleged dichotomy between automatic and non-automatic processes prominently on display in Camerer et al. (Reference Camerer, Loewenstein and Prelec2005: 9), and in much of the current NE literature, seems a mis-specification. Glimcher et al. (Reference Glimcher, Dorris and Bayer2005: 251–2) are quite adamant about this issue, as is Pessoa (Reference Pessoa2008). Emotional experience and emotional expression (whatever that means), for example, involve the neocortex as well (Goldberg Reference Goldberg2005: 220–1). In sum, the attempts to construct an affective–cognitive divide (e.g. Camerer et al. Reference Camerer, Loewenstein and Prelec2005) seem thoroughly misguided.
7. “Module madness”Footnote 5
More generally, the notion that humans have specialized and tightly localized modules for almost everything – another view prominently on display in Camerer et al. (Reference Camerer, Loewenstein and Prelec2005) and in much of the current NE literature – is a modern version of phrenology (e.g., Cabeza and Nyberg Reference Cabeza and Nyberg2000; Fodor Reference Fodor2000;Footnote 6 Uttal Reference Uttal2001, no date; Goldberg Reference Goldberg2001, Reference Goldberg2005; Glimcher et al. Reference Glimcher, Dorris and Bayer2005; Anderson Reference Anderson2007a, Reference Anderson2007b, Reference Anderson2007c, and forthcoming; Pessoa Reference Pessoa2008; see also Posner Reference Posner2003 and Prinz Reference Prinz and Stainton2006).
7.1. Cabeza and Nyberg (Reference Cabeza and Nyberg2000) demonstrate convincingly by way of a meta-analysis of activations in 275 PET and fMRI studies that most brain regionsFootnote 7 are engaged by a variety of cognitive challenges. This is so even though control/comparison tasks have been subtracted out and reported activations are therefore a lower limit of the true activations. In a recent set of articles that draws on the analysis in Cabeza and Nyberg (Reference Cabeza and Nyberg2000), Anderson (Reference Anderson2007a, Reference Anderson2007b, Reference Anderson2007c) has quantified the number of activations for selected task categories. He finds that on average each of the 135 tasks in his sample activivated 5.97 brain regions (SD = 4.80) and that perceptual tasks and language activated 4.88 and 7.81 brain regions, respectively. The resultant brainmaps (e.g. Anderson 2007: 5) are not at all of the modular kind that the interested public is typically sold (e.g. Zak Reference Zak2004: fig. 2).
7.2. It is noteworthy that regions associated with some sensory or motor activities (such as perception and language) show tighter localization than those associated with higher cognitive processes such as “working memory” or “problem solving” (e.g., Cabeza and Nyberg Reference Cabeza and Nyberg2000: Table 5), where the data literally are all over (half of) the map. Narrow localizations, at least for higher cognitive processes, are a myth (Uttal no date: 12, see also 2001; Anderson 2007, Reference Anderson2007a, Reference Anderson2007b).
7.3. Anderson (2007, Reference Anderson2007a, Reference Anderson2007b) proposes an intriguing argument (the “massive redeployment hypothesis”) for the empirical observation that cognitive challenges of a evolutionarily more recent nature are correlated with more dispersed activation patterns including activations in evolutionarily older brain areas (e.g. Anderson 2007: fig. 4).
7.4. A major problem, also mentioned by Cabeza and Nyberg (Reference Cabeza and Nyberg2000: 31, see also 35–6) and Pessoa (Reference Pessoa2008), is the deplorable fact that the authors of most functional neuro-imaging studies interpret activations within their own domain, ignoring the multiple functions that the same brain region might have, and hence the more general function it might fulfil.
7.5. If something is not black, it is not necessarily white. In other words, questioning extreme conceptions of the modularity of the mind (as they are reflected in almost all neuroeconomics studies and reviews) does not mean that, especially for evolutionarily earlier (“lower”) cognitive challenges, one does not find narrowly localized regions of activations (Anderson Reference Anderson2007a, Reference Anderson2007b, Reference Anderson2007c; Pessoa Reference Pessoa2008). Surely one does. By the same token, for evolutionarily later (“higher”) cognitive challenges, activation patterns tend to be highly interactive and distributed. The interesting issue is how much of the brain is localized in the sense of having only one clearly defined function. And whether indeed necessary activations are also sufficient.
8. The surprising, or at least ironic, (neuro)economics of decisions and games
An answer also seems to be outstanding to the interesting question whether the rational choice explanations of seemingly well established neuroscience results, both for simple decisions and simple games, generalize to more complicated contexts. In fact, this issue seems a major bone of contention between NE and those more critical of the NE enterprise.
8.1. Glimcher et al. (Reference Glimcher, Dorris and Bayer2005; see also Schultz Reference Schultz2006; Lee Reference Lee2008; Doya Reference Doya2008) make, to my mind, a rather persuasive case that a simple unified framework drawing heavily on standard expected utility theory has considerable power to explain the data for simple decisions and simple games (namely, the inspection game which is essentially a gift or trust game under a different name). The brain, in essence, is in these contexts an economic evaluation machine. The work by authors such as Glimcher or Schultz is particularly fascinating because it seems to suggest that desirabilities, whether strategic or not, have a physiological basis that involves LIP neurons.
8.2. The irony that a unified framework seems to draw heavily on exactly those formulations that certain representatives of BE and NE have worked hard to critique/undermine, has not been lost on some of its more open-minded representatives: “An irony of neuroeconomics is that neuroscientists [such as Glimcher or Schultz] often find the most basic principles of rationality useful in explaining human choice, while neuroeconomists like ourselves hope to use neuroscience to help us understand limits of rationality in complex decision making” (Bhatt and Camerer Reference Bhatt and Camerer2005: 449). In the footnote following this passage, Bhatt and Camerer suggest that “a way to reconcile these views is to accept that simple rationality principles guide highly-evolved pan-species systems necessary for survival (. . .) but that complex modern choices are made by a pastiche of previously-evolved systems and are unlikely to have evolved to satisfy rationality axioms only discovered in recent decades” (Bhatt and Camerer Reference Bhatt and Camerer2005: 449). Maybe, maybe not. For starters, it strikes me as problematic to rationalize this way the results that Glimcher and his colleagues have produced on the inspection game, a prominent workhorse of the social preferences crowd. More importantly, a flurry of recent articles and books (e.g. Cherry et al. Reference Cherry, Frykblom and Shogren2002; List Reference List2006; Levitt and List Reference Levitt and List2007; Binmore Reference Binmore2007a, Reference Binmore2007b) provided significant evidence questioning the reality of social preferences that cannot be explained through basic formulations of decision and game theory and appropriate experimental controls. Footnote 8
9. Neurogenesis and plasticity
The adult brain produces new neurons well into adulthood (for references see Murphy and Brown Reference Murphy and Brown2007: 117) and, albeit at a decreasing rate, throughout one's life (Goldberg Reference Goldberg2005: especially ch. 14). The adult brain is furthermore rather malleable, or “plastic” (e.g. Rumpel et al. Reference Rumpel, LeDoux, Zador and Malinow2005; Goldberg Reference Goldberg2001, Reference Goldberg2005) and to quite some extent idiosyncratic in detail (e.g. Goldberg Reference Goldberg2001: especially ch. 7, Goldberg Reference Goldberg2005: ch. 14).
9.1. Neurogenesis, or neuronal growth, can keep us sharp until late age, can slow down progressive brain disorders such as dementias, or even reverse the consequences of head trauma, strokes, or plain underusage of the grey and white matter (Goldberg Reference Goldberg2005).
9.2. Neurogenesis can be enhanced through cognitive enhancement programmes (Goldberg Reference Goldberg2005).
9.3. Being an academic constitutes a naturally occurring cognitive enhancement programme.
9.4. Relatedly, a consensus seems to emerge that the organization of the brain (functional and anatomical) is highly plastic across one's lifetime (e.g. Rumpel et al. Reference Rumpel, LeDoux, Zador and Malinow2005). How tremendously plastic our brain is, even at an advanced age, has been demonstrated with striking success by Taub's and his collaborators’ work with stroke patients (e.g. Liepert et al. Reference Liepert, Bauder, Miltner, Taub and Weiller2000, Taub et al. Reference Taub, Uswatte, King, Morris, Crago and Chatterjee2006).
9.5. Neurogenesis and plasticity further question primitive notions of localization/modularity. In general, while people are pretty much the same in their basic make-up, they are idiosyncratically equipped through nature and nurture.
10. Lab environs, oh how artificial thou are
Lab environs for NE experiments qualify as exhibit A for the artificiality of lab environs. The artificiality of lab environs is likely to reduce the signal-to-noise ratio of techniques such as PET and fMRI.
10.1. At the minimum this artificiality, and the neural activations that it produces, adds to the ground noise (recall Wang et al. Reference Wang, Jiang, Yu, Tian, Li, Liu, Zhou, Xu, Song and Li2007; Fox et al. Reference Fox, Snyder, Zacks and Raichle2006, Reference Fox, Snyder, Vincent and Raichle2007; Fox and Raichle Reference Fox and Raichle2007). Part of the noise may have no functional purpose, part of the noise may come from subjects engaging in thoughts – not necessarily very conscious ones – like: Why am I lying here? What's that dizziness? What's that metallic taste in my mouth? Is this harmful to my health? Have they lied to me after all about the health risks? What if I think nasty/naughty thoughts? Can the experimenter decipher them?
10.2. “(Modern cognitive neuroscientists and neuropsychologists) have focused on how humans solve problems, but have largely ignored the issue of how they choose the problem to be solved” (Goldberg and Podell Reference Goldberg and Podell1999: 376; see also Goldberg Reference Goldberg2001: 77–85). To the best of my knowledge, this issue remains terra incognita for neuroeconomists, too. I believe this to be one area where neuroscience could contribute to our understanding of the brain's working.
10.3. More generally, all the motivations that have made experimental economists so eager to leave the lab and venture out in various kinds of field settings (Harrison and List Reference Harrison and List2004), apply to current neuroeconomics experiments. Of particular note here are the issues – such as the question of the representativeness of the stimuli, or the appropriate form of information presentation) – that have been the bone of contention in the debate between the heuristics and biases school (e.g. Kahneman and Tversky Reference Kahneman and Tversky1996) and those arguing from an ecological rationality point of view (e.g. Gigerenzer Reference Gigerenzer1996; see also Hertwig and Ortmann Reference Hertwig, Ortmann, Zwick and Rapoport2005).
11. Selection biases
The artificiality of environs is likely to produce important selection biases, a problem whose importance is enhanced through the typically exceedingly low number of subjects. Harrison (Reference Harrison2008) gives a number of examples. I'm not aware of any NE study that addresses this problem which is of some concern to experimental economists, even under the best of circumstances.
12. The need for better controls and higher numbers of subjects.
The between-subjects designs typically used in NE do not control for cognitive-sociodemographic characteristics that have been shown to be of significant importance in other experimental contexts (e.g. Holt and Laury Reference Holt and Laury2002 and Harrison et al. Reference Harrison, Johnson, McInnes and Rutström2005). Again, this problem gains in importance because of the exceedingly low number of subjects per treatment.
12.1. Wei et al. (Reference Wei, Yoo, Dickey, Zou, Guttmann and Panych2004), for example, argue that inter-subject variability is typically higher than intra-subject variability. While there is a host of antidotes (spatial normalization procedures, fancier tests) recruiting healthy right-handed, or normal subjects seems not enough of a control by a wide margin. Even controlling for cognitive-sociodemographic characteristics may be too little (e.g. Thirion et al. Reference Thirion, Pinel, Meriaux, Roche, Dehaene and Poline2007).
12.2. Thirion and his colleagues (Reference Thirion, Pinel, Meriaux, Roche, Dehaene and Poline2007) propose explicitly that fMRI inter-subject variability ought to be related to “behavioral differences and individual or psychological characteristics of the subjects” (115). They find empirically that 20 subjects or more (preferably 27) should be included in functional neuro-imaging studies to reach sufficient reliability. They identify as a key problem that activation detection on the basis of smaller numbers might overlook subtle (differential) activation patterns and argue that the proposed number of subjects is the minimum for auditory and motor activations. So, numbers ought to be higher for more subtle functional activation patterns. It seems to me that such a study is urgently needed for a couple of NE poster childs.
13. Standards and practices of experimentation in NE.
Standards and practices of experimentation in NE are problematic.
13.1. Deception, for example, seems to be accepted by some in a rather cavalier manner. Rilling et al. (Reference Rilling, Gutman, Zeh, Pagnoni, Berns and Kilts2002), Sanfey et al. (Reference Sanfey, Rilling, Aronson, Nystrom and Cohen2003), Coricelli et al. (Reference Coricelli, Critchley, Joffily, O'Doherty, Sirigu and Dolan2005), and Knoch et al. (Reference Knoch, Pascual-Leone, Meyer, Treyer and Fehr2006) are prominent examples but see also the pointed observations in Jaffe (Reference Jaffe2006) whose default assumption was only the experimenter that greeted him would be in the control-room and was “unsettled” when he realized that in fact there were four people watching him. Unfortunately, there is presently no such thing as “a guarantee of no deception” (Zak Reference Zak2004: 1738). And, apparently, subjects seem well advised to abandon even elementary default assumptions at the lab door. The empirical consequences of deception by commission or omission are presented in Ortmann and Hertwig (Reference Ortmann and Hertwig2002) and Hertwig and Ortmann (Reference Hertwig and Ortmann2008) for the behavioural level; it can only be guessed what ground noise (see note 15.3.) these practices provoke on the neural level.
13.2. Relatedly, it is a problematic (albeit, maybe, unavoidable) procedure in neuroeconomics labs that subjects are instructed individually. Apparently, however, they are typically not instructed by researchers blind to the research hypothesis – a procedure that is likely to make an experimenter's wishful thinking a reality, as Ortmann (Reference Ortmann, Carpenter, Harrison and List2005), based on readily available and persuasive evidence from psychology and taking aim at Henrich (Reference Henrich2000), argues.
13.3. Last but not least, a well-established (albeit occasionally ignored) practice of experimental economists of concatenating new experiments through replication of old results seems almost completely amiss in the NE literature. I do understand that the strategy of concatenating new experiments in this manner is very expensive, but there is evidence that the inter-subject and -laboratory variability of results is high so that the strategy of systematically trying to build on earlier studies ought not to be just dismissed.
14. Standards and practices of experimentation in NE, continued: Lack of social distance as the key confound.
Most importantly, issues of social distance and experimenter confounds that recent papers in economics have wrestled with (e.g. List Reference List2006; Levitt and List Reference Levitt and List2007) seem to be completely ignored in NE. That's deplorable because these articles provide, in my view, rather damaging results to the BE enterprise and, by extension, also to the NE enterprise.
14.1. The subjects in McClure et al. (Reference McClure, Laibson, Loewenstein and Cohen2004) were asked to make choices between an immediate reward and a distant reward and between a distant reward and an even more distant reward. All choices showed activation in the lateral prefrontal areas and associated parietal areas of the brain. Choices involving immediate rewards were correlated with BOLD responses in the limbic system. When the immediate reward was available, but subjects chose the delayed reward, the authors found greater BOLD responses in the lateral prefrontal and parietal areas of the brain compared to its limbic areas. This pattern strikes me as observationally equivalent to that of subjects being worried about the credibility of the experimenter and/or assessing the transaction costs. See Harrison and Lau (Reference Harrison and Lau2005) for the detailed arguments and empirical evidence supporting the credibility/transaction costs explanation. Roth (Reference Roth2007) offers the additional observation that, as a matter of fact, all rewards are delayed as long as the subject lies in the scanner and that “immediate versus delayed seems to be at least in part a verbal phenomenon” (slide 16).
14.2. McCabe et al. (Reference McCabe, Houser, Ryan, Smith and Trouard2001), using a binary choice version of the trust game, argue that their seven “cooperators” showed greater prefrontal activations than the five “defectors”. The observed activation pattern is also consistent with subjects' concerns about what the experimenter might think about them.
14.3. Sanfey et al. (Reference Sanfey, Rilling, Aronson, Nystrom and Cohen2003) find that their subjects reject roughly 50% of the unfair offers by humans (and only 20% of the unfair offers by computers). Is the observed activation consistent with negative emotions or with concerns about what the experimenter might think about them? (Both.) Could these explanations be distinguished? (No.) Should they be distinguished? (Yes.)
14.4. Xiao and Houser (Reference Xiao and Houser2005) find that subjects that are allowed to express anger to their counterpart are more likely to accept the offer. Is the observed behaviour consistent with lesser concerns about what the experimenter might think about their action? (I think so.) Could the competing explanations be distinguished? (No.) Should they be distinguished? (Yes.)
14.5. Spitzer et al. (Reference Spitzer, Fischbacher, Herrnberger, Gron and Fehr2007) identify a brain locus for norm compliance. Using that old workhorse, the dictator game, they study activations in a no-punishment and a punishment condition. Subjects transfer more money in the punishment condition. The authors also find increased activations in some parts of the prefrontal cortex and the caudeate nucleus in that condition. Again, the question is to what extent these activations could have been a reflection of the subjects' concern about the experimenters’ opinion of them.
14.6. Even in as seemingly innocent a context as the neuro-imaging study of Coricelli et al. (Reference Coricelli, Critchley, Joffily, O'Doherty, Sirigu and Dolan2005), the regret – and blame-taking – demonstrated there might have been directed towards the experimenter.
14.7. And so it goes. In their mini review of related literature on the occasion of the publication of Spitzer et al. (Reference Spitzer, Fischbacher, Herrnberger, Gron and Fehr2007), Montague and Lohrenz (Reference Montague and Lohrenz2007) not once bring up the issue of the social distance confound. Given the evidence, for example, in Cherry et al. (Reference Cherry, Frykblom and Shogren2002) and List (Reference List2006), that strikes me as remarkable. Surprisingly, even experimenters that should know better (since, in fact, they have laid the ground work for some of these kind of concerns; see e.g. Hoffman et al. Reference Hoffman, McCabe and Smith1996) rarely ever seem to discuss the social distance confound. It seems clear that, apart from adding inevitably ground noise to the activation patterns, this confound must interfere with all attempts trying to understand the neural correlates of social, risk and time preferences.
15. Data, blackboxed
The processing of data is too often a black box and something to be viewed with suspicion (e.g., Cabeza and Nyberg Reference Cabeza and Nyberg2000; Uttal no date, 2001; Dobbs Reference Dobbs2005; Thirion et al. Reference Thirion, Pinel, Meriaux, Roche, Dehaene and Poline2007); it destroys the key premise of replicability and reproducibility upon which the credibility of experimentation techniques is built. The colourful pictures suggest more precision than there is and hence invite overselling and hyperbole.
15.1. Cabeza and Nyberg (Reference Cabeza and Nyberg2000: 1–3) discuss explicitly methodological concerns such as the inability of functional neuro-imaging techniques to determine the essentiality of various activations prompted by certain tasks; the “sluggishness” of haemodynamic techniques such as PET and fMRI; and the various problems of the subtraction method (including evidence that the insertion assumption is not always valid). Uttal (no date, Reference Uttal2001) covers much of the same ground but in much more detail and more critically, pointing out even if the brain were modular in a way that would allow haemodynamic techniques to succeed, and even if we were assured that the essentiality problem were reasonably addressed, that there remain important issues of the appropriate level of significance to define the boundaries of reported activations. All of this does not even question the quite questionable assumption that data of individuals can be averaged/aggregated in some fashion (see Dobbs Reference Dobbs2005, for a good illustration of how heroic that assumption seems to be; see Beauchamp Reference Beauchamp2005 or Wei et al. Reference Wei, Yoo, Dickey, Zou, Guttmann and Panych2004, for harder evidence).
15.2. More recently, Gerlach (Reference Gerlach2007) finds, comparing 20 fMRI studies that compared visual processing of natural objects and artefacts in normal subjects, that “not a single area is consistently activated for a given category across all studies” (296). Needless to say that his results – which he attributes to too few observations and a liberal adoption of statistical thresholds that led to false-positive activations, among others – are based on a rather simple task. Exploring social preferences is likely to produce variability of an order of magnitude higher.
15.3. Wang and his colleagues (Reference Uttal2007) add yet another complication. These authors find that the brain (in the primary visual areas of normal-sighted subjects) is “a system intrinsically operating on its own, and sensory information interacts with rather than determines the operation of the system” (697). See also Fox et al. (Reference Fox, Snyder, Zacks and Raichle2006, Reference Fox, Snyder, Vincent and Raichle2007) and Fox and Raichle (Reference Fox and Raichle2007) for related findings. Apparently this ground noise can easily swamp the significant effects that the subtraction method identifies as activation patterns. This emergent discussion is of major importance to the reality of NE findings, be it only for the simple reason that ground level activations are likely to be more pronounced (and unsteady) in typical NE settings. See also 10., 12.–14.
15.4. The colourful graphics that are the outcomes of all that intransparent processing of fMRI raw data imply much more precision than there actually is. One rarely finds caveats formulated (for a laudable but shy attempt, see Zak Reference Zak2004: 1745–6).
15.5. Which brings us to the marketing side of things: The ways most neuroeconomists go about the reality of their techniques reminds me of the ways certain behavioural economists go about the reality of cognitive illusions.Footnote 9 See Hertwig and Ortmann (Reference Hertwig, Ortmann, Zwick and Rapoport2005) for a primer on the Cognitive Illusions controversy. As far as I am concerned, NE is the continuation of the politics of BE (in economics) and the heuristics and biases (in psychology) approach with different means. The difference is: NE is more intransparent since the barriers to entry (in human capital acquisition and scanner time) are rather high. These barriers to entry notwithstanding, we are likely to see the same kind of broadening scepticism and debunking of behavioural truths in NE that we currently see in economics (regarding BE: e.g. Cherry et al. Reference Cherry, Frykblom and Shogren2002; List Reference List2006; Levitt and List Reference Levitt and List2007; Plott and Zeiler Reference Plott and Zeiler2005, Reference Plott and Zeiler2007) and even earlier in psychology (regarding heuristics and biases: see Gigerenzer Reference Gigerenzer1996; Hertwig and Ortmann Reference Hertwig, Ortmann, Zwick and Rapoport2005).
16. Availability of data as key condition for replicability and reproducibility
The reported non-availability of data (Harrison Reference Harrison2008) is scandalous by all standards to which experimental economists adhere (at least conceptually) and thoroughly discredits the NE enterprise.
16.1. Again, not making data available destroys the key premise of replicability and reproducibility on which the credibility of experimentation techniques is built. I do understand the large amounts of data involved and the logistics problems that they might bring about. But something ought to be done in NE if the enterprise wants to keep a modicum of credibility. For example, NE could make it a precondition for publication that data are submitted to the fMRI Data Center (http://www.fmridc.org/f/fmridc), which is designed to be a publicly accessible repository of peer-reviewed fMRI studies that aspires to provide all data necessary to interpret, analyse, and replicate fMRI studies.
16.2. The current practice, at least that's what economic theory suggests, is an invitation to play fast and loose with data processing as well as its interpretation. I do not easily trust results produced under these circumstances. Neither, apparently do others (albeit, predictably, these others are not behavioural economists).
17. Improved research methods? Not by a wide margin . . .
Because of the concerns formulated in 10 through 16, the same confounds that trouble EE and BE (Hertwig and Ortmann Reference Hertwig and Ortmann2001a, Reference Hertwig and Ortmann2001b) also trouble NE, only more so. I do not agree with the statement in Zak (Reference Zak2004: 1737) that neuroeconomics is improving research methods. Certainly not on the economics side.
18. So what?
Let us, for the sake of the argument, assume that all the problems discussed above can be solved: Let's specifically assume numbers and selection problems have been solved successfully, appropriate data processing strategies and tests have been used, experimental design and implementation issues have been addressed (or at least controlled to a reasonable degree), and we have therefore reason to believe that the neural hotspots that are superimposed on the brain scan ex post identify sufficient activations rather than necessary ones. What then could economists learn and do with their insights? Again, for the sake of the argument, let us assume that (the lack of) self-command on some issue (drinking, smoking, sweets, procrastination, norm violations) could be tied to some identifiable locale in the brain, what would be the consequence? Would it not be a behavioural one (e.g. make the proband play chess to learn to think forward) in ways similar to Taub's and his colleagues’ strategies to help the patient (putting the functional hand into a boxer glove to force the stroke victim to train the incapacitated hand)? But, then, do I need NE results to devise such strategies? While I am very sympathetic to the neuroscience enterprise per se, I do not see any persuasive evidence that the kind of NE experiments currently done can contribute the kind of significant insights that experiments in economics promised almost from the first day (well at least the early work of Vernon L. Smith). And I definitely can not see how studying colourful pictures of the brain has implications for economic growth.
19. Modest suggestion
In order to gain some respect (and not just the huge grants it has managed to attract), NE would be well advised to take serious the conceptual and methodological concerns mentioned above. There are too many open questions that have been pushed aside by the easy availability of funding for projects that too often are built upon problematic conceptions of how the mind works and that too often have used pre-confused designs and implementations. See 10. and 14. and also Harrison Reference Harrison2008.
Most importantly, NE needs to find ways to address successfully the question whether it wants to adhere to the key premise of replicability and reproducibility upon which the credibility of experimentation techniques is built. If it does not, it will become a doomed enterprise and will surely invite a sequel to Binmore and Shaked (Reference Binmore and Shaked2007).
Specifically, NE ought to stick to the standard conventions that experimental economists have accepted for good reasons: financial incentives whenever possible and a complete ban on deception, both by commission and omission. Most importantly, NE needs to address the social distance issue. Since NE, with few exceptions, seems to be about the various aspects of social, risk, and time preferences, understanding experimenter – subject interactions seems to be of extraordinary importance.
I believe that in the last few years people were rushing into NE to stake claims and find the kind of gold nuggets that would be able to set the stage (and would make them sit pretty). This rush is understandable, and individually rational, but came at a cost of questionable practices that are hardly trust-building. So far (and it seems that McCabe agrees, as surely does Harrison) that rushing to stake claims has produced few (no?) nuggets of note. It is hard to avoid the impression that the availability of new technology has indeed run the science rather than the other way round.
Of course, given all the money that has been sunk into the NE enterprise, something is going to come of this. A key question is at what cost this something comes: To argue, as Rustichini (Reference Rustichini2005) does, that “at the very least, neuroeconomics provides new data in addition to those we have available from theoretical, empirical, and experimental research on human behavior” (201) is an obviously flawed argument.
I agree with Harrison (Reference Harrison2008), and by way of reference Smith (Reference Smith2008), that NE has to go beyond the application of fancy new tools to old questions (based on a problematic reading of experimental evidence in psychology) and address new questions. Such questions might require completely new experiments but should take cues from well-established principles of design, implementation, and replicability in EE.
