Transcript
Page 1: Neuroeconomics || Experimental Methods in Cognitive Neuroscience

C H A P T E R

6

Experimental Methods in CognitiveNeuroscience

Christian C. Ruff and Scott A. Huettel

O U T L I N E

Introduction 77Measurement Versus Manipulation 78Strengths and Limitations of Different Methods 78

Measurement Techniques 79Invasive Neurophysiology: Single-Unit Recording 79Non-Invasive Neurophysiology 81Metabolic Neuroimaging 84

Manipulation Techniques 92Brain Stimulation 92Lesion Studies 101

Conclusion: Convergence across Methods 105

References 107

INTRODUCTION

The growth of cognitive neuroscience as an aca-demic discipline has been inextricably tied to thedevelopment of its research methods. These methodsnow provide unprecedented access to brain structureand function. When combined with theoretical per-spectives from psychology, economics, and other disci-plines, they allow us to generate new models offunctions like memory, attention, and decision making.As these methods have become more refined, theyhave also become more accessible to the research com-munity. Experiments that would have been impossiblea decade ago are now readily conducted as graduatestudent projects. The increased power, flexibility, andaccessibility of these techniques have had unques-tioned benefits for scientific progress: each year,several thousand scientific articles are published usingthe core methods of cognitive neuroscience.

However, the growth of cognitive neuroscience hascarried an unexpected cost. It has become possible forinexperienced researchers to design and carry out cog-nitive neuroscience experiments without having adeep understanding of the underlying brain functionor of what they are recording. This accessibility canhave undesirable consequences. When data are being

collected or analyzed, errors may go undetected, lead-ing to inaccurate results. Research results may lead tooverstated or implausible claims or may be reinter-preted to fit a previously held view. A poor under-standing of methods can alter the very direction ofnew research. Researchers who become too focused ona single technique may apply that technique indiscrim-inately, regardless of whether it is appropriate for theirspecific research question. Paradoxically, the advancesin cognitive neuroscience methods have made it easierfor researchers to make mistakes!

Any consideration of neuroscientific methodsshould begin with a fundamental observation: differ-ent techniques address different aspects of neural func-tion. This simple fact is often lost in populardescriptions of neuroscience, which often refer generi-cally to “activity in a brain region” that predicts somebehavior or trait. However, the interpretation of agiven result may strongly depend on what is beingmeasured: neuronal firing, brain metabolism, neuro-transmitter levels, or some other brain property.

Providing a comprehensive introduction to all of thediverse methods of cognitive neuroscience would gowell beyond the scope of this chapter. An in-depthunderstanding of any particular method would requirebackground knowledge of the neurophysiological

77Neuroeconomics. DOI: http://dx.doi.org/10.1016/B978-0-12-416008-8.00006-1 © 2014 Elsevier Inc. All rights reserved.

Page 2: Neuroeconomics || Experimental Methods in Cognitive Neuroscience

processes underlying the measured signals, the con-nections between brain structure and functionrecorded by the method, the biophysics and signalprocessing associated with the corresponding experi-mental hardware, and the statistical methods used totranslate raw data into inferences. This chapter intro-duces these topics at an elementary level and refersthe reader to excellent textbooks and primary researcharticles for more in-depth coverage. It focuses primar-ily on the conceptual issues involved in selecting aresearch technique and evaluating the data obtainedusing each technique. As such, it is primarily intendedfor those who are new to cognitive neuroscience andwho seek guidance on how to evaluate the strengthsand limitations of published work. Accordingly, eachtechnique is introduced in conjunction with specificexamples drawn from recent neuroeconomic studies.

Measurement Versus Manipulation

Cognitive neuroscience techniques can be dividedinto two main categories. Measurement techniques, asthe name implies, measure changes in brain functionwhile a research participant (human or animal)engages in some cognitive activity. A typical neuroeco-nomic experiment using a measurement techniquemight require the participant to make a series of sim-ple decisions while the researchers record changes inneuronal firing or metabolic activity that might differbetween, say, higher-value or lower-value choices.Measurement techniques are often described (some-times derisively) as being “correlational” because theycan show that signals from a brain region co-occurwith a function of interest, but they cannot show that aregion is necessary for that function.

Manipulation techniques, in contrast, examine howperturbations of the brain’s function � either bytransiently changing neuronal firing rates or neurotrans-mitter levels or by permanently damaging tissue �change cognitive functions or behavior. Accordingly,manipulation techniques are sometimes called causalapproaches. Neuroeconomists have used manipulationtechniques to disrupt processing in specific regions,which in turn alters the choices people make (e.g., ininteractive games).

This chapter follows this basic division, first intro-ducing techniques that measure changes in brain func-tion which track the variables within decision models,then considering techniques that change neural proces-sing and also decision behavior. It is important to rec-ognize that measurement and manipulation techniquesprovide distinct and complementary information aboutbrain function. Cognitive neuroscience research pro-gresses more quickly when measurement techniquesestablish links between brain structure and cognitive

function and then manipulation techniques probe thatrelationship to improve inferences and models.

Strengths and Limitations of Different Methods

How do neuroscientists determine which researchmethod to apply to a given research question? Broadlyconsidered, three factors have primary importance:temporal resolution, spatial resolution, and invasive-ness (Figure 6.1). Temporal resolution refers to the fre-quency in time with which measurements ormanipulations can be made. Techniques that recordneuronal activity directly through electrophysiologicalmeans tend to have very good temporal resolution(e.g., millisecond precision); techniques that measureindirect metabolic correlates of neuronal activity tendto have intermediate temporal resolution (e.g., secondsto minutes); and techniques that manipulate brainfunction through drug effects or brain lesions tend tohave the poorest temporal resolution (e.g., minutes todays). Spatial resolution refers to the ability to distin-guish adjacent brain regions that differ in function.Techniques that position electrode sensors directlywithin the brain have the highest spatial resolution(e.g., individual neurons or better); techniques of func-tional neuroimaging have intermediate spatial resolu-tion (e.g., millimeters to centimeters); and techniquesthat measure electrical signals that spread diffuselytend to have the lowest spatial resolution (e.g., centi-meters to the entire brain).

Finally, neuroscience techniques differ with respectto whether they can make measurements withoutdamage to or disruption of the brain (or other body tis-sue). Non-invasive techniques record endogenous brainsignals using sensors outside the body. Thus, thesetechniques can be conducted repeatedly in human vol-unteer participants, with no appreciable risk in partici-pation. Invasive techniques introduce a chemical orrecording device into the body. While some such tech-niques can be used in human volunteers (albeit withsignificant attention paid to issues of participantsafety), other invasive techniques can only be used inhuman patients (e.g., prior to neurosurgery) and/ornon-human animals.

This brief summary conveys the critical point that nosingle technique provides a comprehensive account ofbrain function. Different techniques provide comple-mentary information, some giving detailed spatial mapsof functions and others indexing very rapid changes inactivity when those functions are engaged. Every deci-sion process identified in this book has been exploredusing a range of neuroscience techniques, and converg-ing evidence from different techniques and researchparadigms has enabled more powerful conclusions thancould be obtained from any one approach in isolation.

78 6. EXPERIMENTAL METHODS IN COGNITIVE NEUROSCIENCE

NEUROECONOMICS

Page 3: Neuroeconomics || Experimental Methods in Cognitive Neuroscience

MEASUREMENT TECHNIQUES

The measurement techniques used within cognitiveneuroscience measure information transmission by neu-rons, either directly or indirectly. This section will considerfive such techniques that are organized by the aspect ofneural function they measure: single-unit recording,electroencephalography (EEG), magnetoencephalography(MEG), positron emission tomography (PET), and func-tional magnetic resonance imaging (fMRI).

As introduced in the previous chapter (Chapter 5),there are two types of neuronal information proces-sing: axonal signaling and dendritic integration. Whena neuron fires, it sends a signal called an action poten-tial down its axon to one or more other neurons. Theaction potential is evident as a small change in thevoltage of the axon’s membrane, and thus it can bemeasured with electrodes that are positioned immedi-ately adjacent to that neuron � a technique known assingle-neuron or single-unit recording.

The action potential evokes the release of neurotrans-mitters at the synapse; when those neurotransmittersbind to receptors on the dendrites of a post-synapticneuron, they cause its membrane potential to becomemore positive or negative. These changes in membrane

potential tend to be relatively synchronized over manyneurons within a given brain region. Thus, they gener-ate coherent changes in electrical potential (and thusthe associated magnetic fields) that can be measuredvia detectors on the scalp, forming the signal measuredin EEG and MEG experiments.

Both sorts of neuronal information processing � axo-nal signaling and dendritic integration � require sub-stantial energy. In particular, the restoration ofmembrane potentials requires glucose and oxygen to bedelivered through the cerebrovascular system. Thosemetabolites themselves are not involved in neuronalsignaling, but they serve as important markers that sig-naling activity has increased within a brain region.

Invasive Neurophysiology: Single-UnitRecording

How Single-Unit Recording Works

To many neuroscientists, the most basic element ofnervous system function is the action potential. Asdescribed in the preceding chapter, action potentials(or “spikes”) arise when the voltage of a neuron’s cellbody rises above a particular threshold (e.g., around

1m

10 cm

1 cm

1mm

100µm

10µm

1 µm

0.1µm

1ms 10ms 100ms 1s 1min 1hr 1day 1wk 1yr

Axon(diameter)

Neuron

Corticalcolumn

Voxel(fMRI)

Gyrus

Brain

Synapse

Scalp ERPs

MEGHuman optical

Human intracranial EPRs

Animal optical techniques

Single-unit recording

Patch-clamp recording

fMRI

PET

TMSEEG

Drugmanipulations

Lesion(human)

Lesion (animal)

FIGURE 6.1 Neuroscience techniques differ in their spatial and temporal resolution. The vertical axes illustrate spatial resolution in termsof distance (left) and the corresponding brain structures (right). The horizontal axis illustrates temporal resolution. This graph includes themost common techniques used in current cognitive neuroscience research, many of which are discussed in this chapter. Techniques thatinvolve data collection from human participants tend to operate at relatively coarser spatial scales than those that record from non-human ani-mals. Electrophysiological techniques that provide excellent temporal resolution in human participants (e.g., scalp ERPs) have the disadvan-tage of relatively low spatial resolution as compared to neuroimaging techniques (e.g., fMRI). Because of the differing strengths andlimitations of each technique, cognitive neuroscience research often applies a range of techniques to a single experimental question. ERPs,event-related potentials; MEG, magnetoencephalography; TMS, transcranial magnetic stimulation; EEG, electroencephalography; PET, positronemission tomography. Figure and caption adapted from Huettel et al. (2004) with permission.

79MEASUREMENT TECHNIQUES

NEUROECONOMICS

Page 4: Neuroeconomics || Experimental Methods in Cognitive Neuroscience

250 microvolts), typically as a result of input fromother neurons to its dendrites. Because action poten-tials have a stereotyped amplitude and waveform for agiven neuron, a large change in cell-body voltage doesnot alter the properties of individual action potentials,but increases the rate at which they are emitted. Thus,neuroscientists use changes in firing rate of a neuronas an index of whether a stimulus (or motor action,etc.) changes the ongoing information processing withwhich that neuron is associated. Baseline firing ratesvary considerably across types of neurons, with ratestypically ranging from a few spikes per second toabout a hundred spikes per second.

Technology

Measurement of action potentials requires the inser-tion of very fine electrodes � often made of a metalwire that is sensitive to relatively high-frequency elec-trical signals, surrounded by a protective insulatingsheath � into the neural tissue immediately adjacent tothe neurons of interest. The electrode itself does notcause appreciable damage to the brain, but openingthe skull to gain access to the brain is an invasive sur-gical procedure that carries significant risk. Thus, thevast majority of neuroeconomics experiments usingthis technique to date have involved non-human pri-mates (e.g., rhesus macaques, macaca mulatta) � oftenwith only a few subjects in each experiment. A fewhigh-profile studies have been conducted with humanparticipants, all involving patients who have electrodesimplanted for clinical reasons (e.g., located at the siteof ongoing epileptic seizures for treatment purposes).Such studies are necessarily rare but nevertheless canprovide unique information about the functioning ofneurons in the human brain.

Cognitive neuroscientists cannot target a specificneuron in humans or non-human primates; neuronsare simply too small and organized in too idiosyncratica fashion. Instead, researchers mount high-precisionmicrodrives on the surface of the skull and then slowlylower electrodes into a brain region of interest, as iden-tified using stereotaxic coordinates (i.e., standard map-ping systems for the positions of structures in a typicalbrain). Experimental localizer tasks (i.e., a task that reli-ably evokes a particular form of neuronal activity)may be used to evoke activity in that brain region sothat the experimenters know when their electrode iscorrectly positioned. Following an experiment, struc-tural MRI or another method may be used to verifythe track taken by the electrode. It can be difficult todistinguish the firing of a single neuron from the col-lective firing of several neurons in close proximity.Thus, this technique is sometimes called “single-unit”recording to emphasize the fact that the data reflectthe activity of a single functional unit that may or may

not contain multiple neurons � although it is now oftenpossible to distinguish single from multiple neurons.

Procedures

Once an electrode is positioned in the desiredregion, the experiment begins. Data can usually be col-lected from a single unit for a period of minutes tohours, until the position of the electrode shifts or theexperimental subject loses interest in the task. Duringthis time, the experimenter may collect data from hun-dreds of experimental trials; due to the extraordinarytemporal resolution of this recording technique, trialscan be packed very densely in time. Results fromsingle-unit recordings can be displayed in both rela-tively raw and averaged forms; by convention,researchers often show raw data from a single “repre-sentative neuron,” along with the average activityfrom all neurons that have met some criterion forinclusion (e.g., an increase in firing rate to the stimuliof interest). Box 6.1 shows recordings from a study inwhich monkeys learned about cues that predicted bothpositive and negative outcomes.

Advantages and Limitations

The fundamental advantage of single-unit record-ing is that it provides direct information about therate and timing of action potentials within a region.More than any other technique � save perhaps stud-ies with lesion patients � single-unit recording hasbeen critical in identifying the core functions of brainregions (e.g., responses of occipital neurons to fea-tures of visual stimuli). Data from single-unit studiesprovide the grounds for many computational modelsof brain function, both by identifying the processingassociated with individual neurons and by helping tomap out the supporting local circuitry. Analysis ofsingle neurons also helps to reveal the diversity ofprocesses within a brain region. Neuroscientists oftenfind several populations of intermixed neurons thathave qualitatively different response properties. Forexample, the neuron described in the figure in Box6.1 increased its firing rate in response to both posi-tive and negative outcomes, whereas other neuronsreported in the same paper showed a different pat-tern: increased firing rates in response to positive out-comes, but decreased rates in response to negativeoutcomes. Such differences would be largely invisibleto the other techniques discussed in this section,which combine data across a much larger set ofneurons.

Single-unit recording has important limitations,however. The invasive nature of single-unit recordinglimits its use to non-human animals, except in the rarecases of human patients with clinically indicated elec-trodes, as discussed previously. Moreover, data

80 6. EXPERIMENTAL METHODS IN COGNITIVE NEUROSCIENCE

NEUROECONOMICS

Page 5: Neuroeconomics || Experimental Methods in Cognitive Neuroscience

collection is generally slow and labor-intensive, withdata often collected serially from one neuron/unit at atime. A published study might describe data from doz-ens of neurons in two monkeys � and that study maytake months to years to complete. Due to the cost ofmaintaining an animal colony and the relatively slowpace of research, single-unit recording can be an expen-sive technique. Most published articles focus on neu-rons in a single brain region, which limits the inferencesthat can be drawn about complex cognitive processes,most of which involve interactions among sets of brainregions. Data from single-unit recording, therefore, areoften highly complementary to data from techniqueswith broader spatial coverage but more limited spatialand temporal resolution (e.g., fMRI).

Non-Invasive Neurophysiology

Electroencephalography (EEG)

HOW EEG WORKS

The EEG signal arises from synchronous changes inthe membrane potentials of the dendrites of manyneighboring neurons. Recall from the previous chapterthat input to a neuron � typically at synapses on den-drites or on the cell body � changes the electricalpotential of its cell membrane. These changes in elec-trical potential are manifest in the flow of ions likesodium across the membrane, leading to focused elec-trical currents within the cell and much more diffuseelectrical currents throughout the cell. If many neuronsevince similar changes in their membrane potential,

BOX 6.1

AN EXAMPLE OF A S INGLE -UN IT RECORD ING STUDY

In this study, monkeys learned that specific cues (col-

ored shapes) predicted either desirable fluid rewards

(top row) or aversive puffs of air to the face (bottom

row). Panel A shows data collected from a single neu-

ron. At the top of each panel, there is a “raster plot”

(from the Latin for “rake”) of neuronal activity, so

termed because the individual trials of the experiment

are stacked in parallel rows. Each dot represents a single

action potential, and areas in which the dots are rela-

tively dense indicate that the neuron’s firing rate was

high. Below the raster plots, there are histograms that

cumulate the raw data into a single estimate of firing

rate at each point in time across trials. Note that the

time window shown in these plots is relatively short,

only 1500 ms, reflecting the high temporal resolution of

this technique. Panel B shows the average response from

38 neurons that evinced a relatively similar firing pat-

tern. It is evident from these data that these neurons

increased their firing rate to cues (conditioned stimuli,

CS) of both the fluid rewards and the aversive puffs of

air, with greater increase for certain outcomes than for

probabilistic outcomes.

100% reward CS

40

20

0

100% airpuff CS 50% airpuff CS 0% airpuff CS

40

20

0

–500

–500

–500500

500 0

1,00

0

1,00

0

5000

1,00

00

50% reward CS 0% reward CS

20

(B)(A)

Reward CS

Spi

kes

per

seco

nd

Spi

kes

per

seco

nd

Airpuff CS

10

00 200 400

Time (ms)

Time (ms) Time (ms) Time (ms)

600 0 200 400Time (ms)

600

FIGURE BOX 6.1 (A) Data from a single neuron. Each panel represents individual trials (rows at top) and mean firing rates overtime (histograms at bottom) for the six cues used in the experiment. (B) Data aggregated over neurons that responded to positive andnegative cues, shown as changes in mean firing rate over time. For positive cues, reward probability ranges from 100% (red) to 50%(pink) to 0% (gray). For negative cues, reward probability ranges from 100% (dark blue) to 50% (light blue) to 0% (gray). Adapted fromMatsumoto and Hikosaka (2009) with permission.

81MEASUREMENT TECHNIQUES

NEUROECONOMICS

Page 6: Neuroeconomics || Experimental Methods in Cognitive Neuroscience

and if those neurons share a similar spatial locationand orientation, then the collective electrical currentthey generate can spread long distances within thebrain � and can be detected by electrodes positionedwithin the brain or even on the scalp.

The first demonstrations of the EEG signal werereported by the physiologist Hans Berger in the 1920s.Using electrodes positioned on the scalp, Berger noticedthat there were regular oscillations in the electricalpotential in the brain that rose and fell approximately 10times per second in an individual whose eyes wereclosed. The frequency of these waves changed with theparticipant’s arousal, speeding up in individuals whowere alert with eyes open, but slowing in drowsy indivi-duals. EEG recording soon became an importantresearch tool within neuroscience, particularly whenresearchers were investigating changes in participants’cognitive states. By the 1960s, researchers started investi-gating components of the EEG signal that were not oscil-latory, but instead time-locked to specific stimuli orevents. These components became known as event-related potentials (ERPs). Researchers soon identifiedERP components associated with various aspects of per-ception, motor preparation, and executive function.A common research approach in modern cognitive neuro-science is to investigate how an experimental manipula-tion or behavior influences the amplitude of a particularcomponent of the ERP signal (e.g., how attention shapescomponents associated with visual perception).

TECHNOLOGY

Most EEG studies record changes in electricalpotential using electrodes positioned on the scalp.While EEG signals can be recorded with as few as twoelectrodes, modern high-density electrode arrays posi-tion 64, 128, or more electrodes on the scalp, toimprove inferences about the spatial distribution of theelectrical activity. Typical electrodes consist of an elec-trically conductive disk connected to a long, light wire.Conductive gels or pastes can be applied between theelectrode and scalp to improve the quality of the elec-trical connection. Some hardware systems embed theelectrodes into a flexible cap to improve the consis-tency of electrode positioning and to decrease the timerequired to prepare the participant for the experiment.The electrode wires feed into a hardware amplifierthat allows very rapid sampling of the electrical signal(e.g., 250�1000 Hz). Data quality is improved by hard-ware- and software-based noise reduction and elimina-tion of signal artifacts (e.g., transient signals associatedwith eye blinks). The amplifiers feed into computerrecording systems that perform initial quality-assurance processing and link the recorded EEG datato the experimental paradigm. Source localization soft-ware combines data from multiple EEG channels to

estimate the location of the likely neural generators ofthe observed signal.

PROCEDURES

In a typical experiment, the participant completespaperwork and then practices the experimental taskwhile the electrodes are put into place. The use of capswith pre-positioned electrodes can greatly speed upthis step. The participant then performs an experimen-tal task repeatedly, often within a session that lastsfrom 30 to 120 minutes. Most uses of EEG within neu-roeconomics take advantage of the good temporal res-olution it provides. Experimental trials can be run veryrapidly, one after the other. Moreover, brain activitydata can be extracted during very precise time win-dows within the experimental task. One influentialexample comes from research on responses in prefron-tal cortex that occurred within 200�300 ms followingfeedback concerning monetary gains and losses (Box6.2). After data collection, the experimenter appliesprocessing algorithms to minimize data quality issuesand extracts trial-by-trial responses in each electrodefor subsequent analyses. Most EEG studies combinedata from 10�40 participants to improve statisticalpower.

ADVANTAGES AND LIMITATIONS

As mentioned above, EEG provides non-invasiveand high-temporal-resolution access to the electricalactivity of the brain. It can be used to separate changesin brain function that occur over several hundredmilliseconds; for example, studies of perception andattention might identify four or more separate compo-nents that arise in the first 500 ms following stimuluspresentation. This temporal resolution allows research-ers to create models of ongoing dynamic processing �such as the change from relatively posterior to frontalprocessing during perceptual decisions � using datanot obtainable by other techniques (save MEG, whichhas similar properties). EEG is also relatively inexpen-sive. The cost of acquiring a new EEG system is lessthan a tenth of the cost of a new MRI scanner, and theincremental costs of running the system are minimal(e.g., for consumables like replacement electrodes, gel).Thus, EEG systems are popular choices for institutionsthat do not have the resources for an MRI or MEGscanner or the facilities for invasive animal neurosci-ence. EEG has also become a primary technique forcommercial applications of neuroscience research (e.g.,neuromarketing). This accessibility has led to aremarkable diversity of research: More studies havebeen published using EEG methods than any othermethod discussed in this chapter.

The primary limitation of EEG comes from itsimprecise spatial localization. This limitation is more

82 6. EXPERIMENTAL METHODS IN COGNITIVE NEUROSCIENCE

NEUROECONOMICS

Page 7: Neuroeconomics || Experimental Methods in Cognitive Neuroscience

severe than just “poor spatial resolution.” The EEG sig-nal propagates readily throughout the brain but isattenuated greatly by the skull and scalp; this meansthat the activity recorded by scalp electrodes couldhave been generated by any of an infinite numberof potential sources. For example, the data shown inBox 6.2 could have resulted from a single source in themedial prefrontal cortex (as argued by its authors),from two sources in deeper bilateral frontal cortex, orfrom some other more complex pattern. This limitationis known as the “inverse problem,” reflecting the factthat researchers must trace back from an underspeci-fied signal measured at the scalp to a single postulatedset of neural generators. As a result, EEG researchers

tend to focus on understanding the properties of par-ticular well-studied components (i.e., changes in activ-ity over time) rather than differentiating the functionsof brain regions (i.e., changes in activity over space).EEG experiments can also require considerable set-uptime to ensure that electrodes are positioned properly.

Magnetoencephalography (MEG)

HOW MEG WORKS

Electrical currents, like those generated by dendriticactivity of neurons, also give rise to magnetic fields.Suppose that a task evokes coherent activity in a popu-lation of neurons, say, in the bank of the central sulcus.

BOX 6.2

AN EXAMPLE OF AN EEG STUDY

Participants in this EEG experiment played a simple

betting game and then received or lost small amounts

of money based on the accuracy of their bets. Each trial

was presented very rapidly; as for single-unit record-

ing, the temporal resolution of this technique allows

individuation of the responses from trial to trial. The

researchers combined data from many trials to generate

ERP components that were time-locked to the monetary

feedback. By doing so, the oscillatory EEG signal

averages out, as seen in the flat baseline during the

pre-feedback period. They found that monetary losses

evoked a larger negative ERP response, compared to

monetary gains, within 250 ms of feedback delivery.

(Note that, by convention, most ERP studies reverse the

y-axis so that negative polarity is upward.) Source anal-

yses indicated that the most likely generator of these

ERP components was a population of neurons within

the medial prefrontal cortex, along the anterior cingu-

late gyrus. These results proved to be important for

connecting research on feedback learning with a grow-

ing literature on medial prefrontal contributions to cog-

nitive control.

–100

(A) (B)

100 200 300 400 500

2 SEM

Loss

Gain

0

Gain-Loss3.8

2.8

1.8

0.8

–0.2µV

5µV

+

FIGURE BOX 6.2 (A) Shown are the mean ERPs in electrodes over the medial prefrontal cortex following monetary losses(red line) and monetary gains (green line). Analyses of ERP data usually compare differences between two (or more) conditions ofinterest to identify components of the ERP signal that are modulated by the experimental task. The maximum difference betweenconditions (arrow) is labeled “medial frontal negativity” (MFN) or “feedback-related negativity” (FRN). (B) On the surface of theskull, the distribution of the difference in activity between the gain and loss conditions is plotted, with color bar at right. Sourcelocalization algorithms estimated that the likely neural generator of this distribution was in the anterior cingulate gyrus, at theapproximate locations shown by the red sphere. Adapted from Gehring and Willoughby (2002) with permission.

83MEASUREMENT TECHNIQUES

NEUROECONOMICS

Page 8: Neuroeconomics || Experimental Methods in Cognitive Neuroscience

The electrical currents generated by that activity wouldrun parallel to the surface of the scalp, whereas theassociated magnetic fields would be oriented perpen-dicular to the scalp � and would extend into the spacearound the head. Neuroscientists have sought to mea-sure those magnetic fields using external sensors witha technique known as magnetoencephalography(MEG). In principle, measuring brain activity via mag-netic field changes would provide some very impor-tant advantages. Fluctuations in the magnetic fieldhave the same temporal properties as the generatingelectrical currents, so they can be analyzed either asoscillatory activity (like traditional EEG) or as changesin activity time-locked to stimulus events (like ERPs),both with millisecond resolution. Moreover, magneticfields do not suffer from significant attenuation as theypass through the skull and scalp. The very small sizeof the magnetic field changes produced by neuronalactivity (about 100 femtoTesla, or 10215 Tesla), how-ever, poses challenges for detection.

TECHNOLOGY

To measure such weak magnetic fields, MEG sys-tems use specialized electrical coils called supercon-ducting quantum interference devices, or SQUIDs.When cooled to very low temperatures, these coilsbecome superconductors that are extraordinarily sensi-tive to changes in magnetic field. The MEG systemarranges many SQUIDs, often several hundred, in alarge helmet-like device that surrounds the partici-pant’s head. Each coil records magnetic field changessimultaneously, at very high temporal resolution. Thepattern of activity across these many coils can be usedto draw inferences about the timing and spatial distri-bution of the underlying neuronal generators. Of note,SQUIDs are sensitive not only to magnetic fields com-ing from the brain, but also to magnetic fields gener-ated by any other source (e.g., nearby motors,electronic equipment, and even the earth itself), whichmay be many orders of magnitude greater. Thus, MEGsystems are installed in rooms surrounded by mag-netic shielding, to attenuate the contribution of exter-nal magnetic fields.

PROCEDURES

In a typical MEG experiment, the participant sitsupright in the MEG system � the experience is rela-tively open and natural compared to that of the con-strained environment of an MRI scanner. Theparticipant views the experimental stimuli on a screenin front of her (with the projector placed outside theroom to minimize magnetic interference) and respondsusing a button box or keyboard positioned on her lap.Compared to the other techniques considered in thischapter, MEG has been less commonly used in

neuroeconomic research. Data from onenotable example, however, are shown in Box 6.3.

ADVANTAGES AND LIMITATIONS

MEG provides many advantages for cognitive neu-roscience research. It is non-invasive, well-tolerated by(human) research participants, can be used with awide range of experimental paradigms, records datafrom the entire brain simultaneously, and can provideinsight into the combined location and timing of corti-cal activity with precision unmatched by any othertechnique. What then can explain the relative paucityof MEG studies, particularly within neuroeconomicresearch?

A primary limitation of MEG comes from its inac-cessibility. Purchasing a new scanner and settingit up in a shielded laboratory facility can require$2�3 million (or more), along with ongoing mainte-nance and personnel costs for the upkeep of the facil-ity. While substantial, these costs are roughly similarto those of fMRI, for example, which arose aroundthe same time but has become much more prevalent.There is an important difference between the twotechniques, however. Functional MRI research can beconducted using a standard clinical MRI scanner,meaning that an institution can recover much of thecost of the scanner purchase through scans ofpatients. Moreover, since MRI scanners have becomea workhorse diagnostic device for many clinical con-ditions, they have become very common; in contrast,there are only a few hundred research MEG systemsin the world.

MEG data also have some limits in their spatial sensi-tivity. Neurons oriented radially (i.e., those generatingelectric currents that flow perpendicular to the scalp) gen-erate magnetic fields that are oriented parallel to thescalp, which become very difficult to detect using MEG.Thus, MEG is often considered to be sensitive only to neu-rons oriented parallel to the skull (e.g., those in the foldsof sulci). Finally, while MEG has better source localizationthan EEG, in part because magnetic fields more readilypass through the skull and scalp, the inverse problem stillholds � researchers cannot unambiguously identify thegenerating neural sources from a MEG recording.

Metabolic Neuroimaging

Positron Emission Tomography (PET)

HOW PET WORKS

The growth of cognitive neuroscience during thefinal decades of the 20th century was sparked, inmany ways, by the development of methods for func-tional neuroimaging. The term functional neuroimagingtypically refers to neuroscience techniques that create

84 6. EXPERIMENTAL METHODS IN COGNITIVE NEUROSCIENCE

NEUROECONOMICS

Page 9: Neuroeconomics || Experimental Methods in Cognitive Neuroscience

two- or three-dimensional maps that show the distri-bution of some aspect of neural activity. The idea ofcreating maps of brain function has a long history.Two centuries ago, the phrenologists attempted to cre-ate maps that divided the brain into a large number ofdistinct faculties (e.g., “wisdom”) based on flawed the-ories about bumps and depressions on the surface ofthe skull. Modern brain mapping has had muchgreater success, for a reason that might seem paradoxi-cal: Many of the functions of greatest interest to cogni-tive neuroscientists (e.g., memory, decision making)involve networks of brain regions, and thus techniques

that provide insight into the functioning of the entirebrain at once are particularly relevant.

The first neuroimaging technique to gain wide-spread acceptance was positron emission tomography(PET) (Figure 6.2). The basic principles of PET arereflected in its name. Researchers inject a quantity of aradioactive isotope � itself attached to a metabolicallyrelevant molecule like glucose or to a neurotransmitterthat binds to a particular type of neuron � into thevenous system of a participant. Depending on thenature of ongoing brain metabolism, that isotopewill be differentially distributed throughout the brain

BOX 6.3

AN EXAMPLE OF AN MEG STUDY

MEG is noted for its ability to provide good spatial

localization (albeit with caveats, as discussed in the text)

along with excellent temporal resolution. In this example

study, participants repeatedly chose between two risky

options that had unknown probabilities and reward

magnitudes. The researchers used MEG to track changes

in brain activity across time, with a specific focus on the

time periods after the presentation of the decision stimu-

lus and before choice execution. Using source localiza-

tion, they showed that early activity in visual cortex

(about 100 ms following stimulus presentation) was

followed by activation at the frontal pole, extending into

ventromedial prefrontal cortex (Panel A). Such rapid

changes would be much more difficult to identify using

neuroimaging techniques like PET or fMRI. Using time-

frequency analyses, the researchers were able to track

the power associated with different frequencies of oscil-

lation in the MEG signal. Within ventromedial prefron-

tal cortex (Panel B), power at relatively low frequencies

(2�6 Hz) ramped up over a period of several hundred

milliseconds, consistent with some computational mod-

els of information integration in this region.

100 ms(A) (B)

1,000 ms

R10

8

6

4

20 500 1,000

Time (ms)

Freq

uenc

y (H

z)L7.28

–7.28

4.75

–4.75

t29

1,000 ms

FIGURE BOX 6.3 (A) Using source localization methods, researchers can estimate the likely neural generators of the currentpattern of magnetic fields. Changes in those generators can be tracked over time with millisecond-level resolution. The color mapon the right indicates the statistical thresholds used for determining significant activation in the brain images. (B) Analysis of datarecorded from a single brain region can reveal changes in specific frequency bands within the MEG signal. Shown here are datafrom the ventromedial prefrontal cortex after presentation of the decision stimulus (0 ms). This region showed an initial broad-based increase in power across a range of frequencies, followed by increased power specifically at lower frequencies (2�6 Hz).Adapted from Hunt et al. (2012) with permission.

85MEASUREMENT TECHNIQUES

NEUROECONOMICS

Page 10: Neuroeconomics || Experimental Methods in Cognitive Neuroscience

(e.g., radioactive glucose will become more prevalentin areas with increased glucose metabolism). As theisotope decays, it emits radioactive particles, positrons,that travel through the brain until they encounter anelectron (on average, within a few millimeters); that col-lision annihilates both particles and releases twogamma rays that travel in opposite directions awayfrom the impact site. By detecting the coincidentarrival of gamma rays in detectors around the head,the PET scanner can compute the likely location atwhich the positron was emitted. If the brain is moni-tored for an extended period of time (typically min-utes), enough of these emission events will accumulateto allow analysis software to estimate the roughdistribution of the isotope throughout the brain. Thatdistribution, when converted to a statistical map,becomes a PET image.

TECHNOLOGY

PET imaging requires a complex array of equip-ment. Generating the radioactive isotopes requires thepresence of a type of particle accelerator called a cyclo-tron. The cyclotron must be located fairly near to thePET scanner, so that the radioactive isotopes can bedelivered to the participant before appreciable decayoccurs. The PET scanner itself is a large device thatlooks superficially similar to the more common MRIscanner; that is, it contains a large cylindrical bore intowhich the participant slides on a moving table. (PETstudies conducted in non-human animals are

sometimes conducted in specialized scanners withsmaller bore sizes, or MicroPET scanners.) Within thescanner, a large ring of scintillation crystals surroundthe bore. When such a crystal is struck by a gammaray, it generates a burst of light that can be measuredin adjacent electronic hardware; if two such events aredetected simultaneously, they are assumed to arisefrom emission events in the brain. Data from the PETscanner is then fed into computer systems for proces-sing and construction of the statistical images used inresearch.

PROCEDURES

Within cognitive neuroscience, a substantial major-ity of PET research has been conducted in human par-ticipants. Participants come to the laboratory, completesafety and consent forms, and then enter the PET scan-ner for the experiment. Depending on the isotopebeing used, the researchers may inject the isotopebefore scanning (e.g., in the case of radioactive glucose,an hour or so before) or during scanning (e.g., withsome chemicals that bind to neurotransmitter recep-tors). PET can be used to study a variety of neurobio-logical processes; for example, in the study whose dataare described in Box 6.4, researchers investigateddopamine binding in the striatum. There is one majordifference between the experimental designs used forPET and those used for other techniques: Because PETaggregates emission events over long time windows,typically several minutes, experiments are organizedinto long blocks of time.

ADVANTAGES AND LIMITATIONS

Unlike the other techniques considered in thischapter, PET imaging can provide information aboutdifferent aspects of neural metabolism or neurotrans-mission. As the example above indicates, PET can pro-vide information about specific chemical brain systems(e.g., dopamine function) that goes well beyond themore general measures of total metabolic demand pro-vided by fMRI. Researchers can create customizedmolecules that bind to particular receptors or that sub-stitute for particular metabolites to provide very pre-cise chemical information, and substantial ongoingresearch is directed at the creation of new radioiso-topes for both clinical and research purposes. Theimages PET creates cover the entire brain with moder-ate spatial resolution; different elements of a statisticalmap can be distinguished if they are separated byabout a centimeter or so. In addition, it can be con-ducted in human volunteer participants, humanpatients, and non-human animals.

The most salient disadvantage of PET is its invasive-ness: It requires injecting radioactive material into par-ticipants. Safety guidelines restrict how that

Fluorine-18 nucleusDetectors

Positron-electroncollision

Gammarays created

Gammaray

Positron

Electron

Gammaray

(A) (B)

FIGURE 6.2 Positron emission tomography (PET). PET imagingrelies on the injection of a radiolabeled tracer (e.g., a radioactive fluo-rine nucleus) that is embedded within a molecule that is relevant forsome biological process (e.g., glucose, oxygen). (A) When that radio-active atom decays, it emits a positron that travels for a few milli-meters before colliding with an electron. That collision annihilatesboth particles, leading to the release of two gamma rays, which canbe detected through their simultaneous arrival at different detectorssurrounding the brain (B). Figure adapted from Huettel et al. (2009)with permission.

86 6. EXPERIMENTAL METHODS IN COGNITIVE NEUROSCIENCE

NEUROECONOMICS

Page 11: Neuroeconomics || Experimental Methods in Cognitive Neuroscience

radioactive material can be created, handled, andadministered, making PET studies much more logisti-cally complex than those using the other measurementapproaches described in this chapter. There are restric-tions, for example, on the number of PET scans inwhich a given individual can participate. Also,planned sample sizes may be scrutinized both byfunding agencies and institutional review boards, aspart of evaluations of the risks and benefits of aresearch protocol. Because of these concerns, and thesignificant cost associated with each participant (oftenmore than $1000), sample sizes in PET studies are typi-cally smaller than those using other techniques.Nevertheless, PET remains a safe, effective, and com-mon technique for modern cognitive neuroscienceresearch.

PET imaging also has very limited temporal resolu-tion. For most studies, data is aggregated over anentire experimental condition, collapsed over the dif-ferent parts of a complex task. In the example shown

in Box 6.4, the change in dopamine transmission mightbe associated with any of the different parts of theexperiment: watching the spinning wheel, receivingthe outcome, or reading about the total amount ofmoney earned so far. Functional changes associatedwith all of these processes would contribute to theoverall PET activity observed.

Functional Magnetic Resonance Imaging (fMRI)

HOW fMRI WORKS

Since its development in the early 1990s, fMRI hasgrown to become the dominant measurement tech-nique in cognitive neuroscience. Its success comesfrom the intertwining of the image creation processfrom MRI with new insights into the metabolicchanges associated with brain activity. Accordingly,understanding this technique requires consideration ofhow MRI images are created as well as of what thoseimages measure. Note that this section necessarily only

BOX 6.4

AN EXAMPLE OF A PET STUDY

The primary goal of this study was to investigate

whether the processing of novel events evoked similar

changes in dopamine binding as did the evaluation of

the rewards themselves. Ten participants watched spin-

ning roulette wheels that delivered different outcomes

in each of three conditions: unpredictable amounts of

money (reward condition), unexpected visual images

and sounds (novelty condition), and blank screens

with no outcomes (control condition). Each condition

was presented repeatedly within its own 30-minute

block, during which time [11C] raclopride was injected, a

radiolabeled dopamine receptor ligand. The concentra-

tion of this ligand can be used to estimate dopamine

release within individual brain regions. The researchers

found reduced transmission of dopamine in the puta-

men during the reward and the novelty condition, as

compared to the control condition.

5.0

2.5

FIGURE BOX 6.4 The results of PET studies are statistical maps that show the estimated distribution of the radioactive traceracross the brain. Here, the tracer was [11C] raclopride, a dopamine receptor ligand. Positive statistical values (color map shown atright) indicate increased concentrations of that ligand, which in turn indicate decreases in dopamine transmission. Note that PETimages are often displayed on high-resolution structural MRI images to facilitate identification of the regions of interest.However, the PET data themselves are of much lower spatial resolution, often about a centimeter or so. Adapted from Hakyemezet al. (2008) with permission.

87MEASUREMENT TECHNIQUES

NEUROECONOMICS

Page 12: Neuroeconomics || Experimental Methods in Cognitive Neuroscience

covers a small subset of what is known about fMRI.Thus, we refer interested readers to general-interesttextbooks that provide more comprehensive coverage(see chapter references and Huettel et al., 2009).

The image creation process in MRI relies on threebasic steps, which fortunately are represented in itsabbreviation. First, the MRI scanner (Figure 6.3) gener-ates a very strong static magnetic field, usually of about1.5 Tesla (T) to 7 T. For reference, the earth’s magneticfield is approximately 0.5 Gauss or 50-millionths of aTesla. When a person’s brain enters the MRI scanner’sstrong magnetic field, the hydrogen atoms therein �particularly the hydrogen atoms in water molecules �tend to become aligned along the axis of the magneticfield. Slightly more atoms are aligned in the samedirection as the field (i.e., the parallel or low-energystate) than in the opposite direction (i.e., the anti-parallel or high-energy state).

Second, specialized electrical coils deliver energy inthe form of radio waves to the brain. These radiowaves are calibrated to a particular frequency thatdepends on the atomic nucleus being imaged (e.g.,hydrogen) and the strength of the MRI scanner. Thisfrequency is called the resonant frequency becauseenergy delivered at that frequency can be absorbed bythe targeted atomic nuclei, causing some of them tojump from a low-energy to a high-energy state. Thesimplest analogy for this process is that of pushingsomeone on a swing set: If you push them repeatedlyeach time they swing past, then they gain energy and

will swing higher and higher; pushes at other times,however, would not be as effective. Importantly, oncethe delivery of the radio waves is turned off, theatomic nuclei return to the low-energy state. Theenergy they release in this process is known as the MRsignal.

Third, to create images, the MRI scanner usesanother set of specialized magnetic coils to create spa-tial gradients in the strength of the magnetic field, forexample, by making the magnetic field on the leftside of the brain stronger than that on the right side.This has the effect of increasing the MR signalrecorded from some spatial locations as compared toothers. Modern MRI scanners change gradient direc-tions very rapidly in computer-optimized patterns, sothat data about the spatial distribution of the MR sig-nal can be differentiated into a large number of spa-tial locations (e.g., a 643 64 matrix within a singleslice of the brain) using data collected in only a fewmilliseconds.

The image creation process described above under-lies nearly all forms of MRI, including the standardclinical imaging of body structure. It is important torecognize that standard forms of MRI do not, by them-selves, provide any insight into brain function � theMR signal they record tracks basic properties of tissuelike the number of protons or fat content. But, in thelate 1980s, Seiji Ogawa, a biophysicist then working atBell Laboratories, discovered that the oxygen contentof venous blood altered the MR signal recorded whena particular type of imaging (i.e., what is often calledT2� imaging) is used. Specifically, when rats werebreathing air that had relatively low oxygen content,the venous system showed up as dark black lines onthe MR images. This effect became known as blood-oxygenation-level-dependent (BOLD) contrast. Soonafter, it was shown that MRI could be calibrated todetect the naturally occurring changes in blood oxy-genation that occur in the brain following neuronalactivity. The use of MRI to pick up an endogenousmarker of brain function became known as functionalMRI.

TECHNOLOGY

Generally speaking, the basic hardware of the MRIscanner (Figure 6.4) comprises the three elementsintroduced in the previous section: a main magnetic coilthat generates the strong static magnetic field, smallergradient coils that modulate the strength of that mag-netic field over space, and a set of radiofrequency coilsthat deliver energy to the object being imaged andreceive the evoked MR signal. The main magnet andgradient coils are embedded in the body of the MRIscanner and thus out of view, while the radiofre-quency coils are often placed in close proximity to the

FIGURE 6.3 An MRI scanner. The modern MRI scanner hasbecome an indispensable part of both clinical practice and neurosci-ence research. Much of the growth of fMRI as a cognitive neurosci-ence technique has been facilitated by the prevalence of high-fieldscanners for clinical applications. Figure adapted from Huettel et al.(2004) with permission.

88 6. EXPERIMENTAL METHODS IN COGNITIVE NEUROSCIENCE

NEUROECONOMICS

Page 13: Neuroeconomics || Experimental Methods in Cognitive Neuroscience

target. For brain imaging, the radiofrequency coils canbe arranged around the participant’s head in a devicethat resembles a birdcage. For functional MRI experi-ments, additional elements are necessary. The experi-mental stimuli are displayed via an MR-compatiblemonitor, head-mounted display, or projection system,and the participant indicates his responses by movinga joystick or pressing a button. The scanner is locatedwithin a magnetically shielded room, both to minimizeunwanted signals from external sources and to attenu-ate the scanner’s field outside the room, for safety rea-sons. These elements have remained largelyunchanged since the 1990s.

Contrary to popular conception, advances in MRItechnology have not been through stronger scanners �the standard field strength for research scanning was

about 1.5 T in the early 1990s and is still only about 3.0 Ttoday. What has changed, instead, are the hardware andprocedures for collecting fast and high-signal images.Most influential has been the development of what iscalled parallel or multi-channel imaging. Rather than onlyrecording signals from a single coil around the sampleobject, new multi-channel scanners record MR signalsfrom a larger number of sensors (16 or more) at differentpoints in space. The resulting images can be combinedusing sophisticated algorithms to improve the image’sresolution, signal-to-noise ratio, and/or speed of collec-tion. Other advances have been made in the customizedinstructions to the radiofrequency coils for MRI dataacquisition � called pulse sequences � that can improvedifferent features of data quality (e.g., sensitivity in aparticular brain region). These real advances in data

Scanner room

Console room

Laboratory

Scanner controlMRI signal returnExperimental control

Computer room

Shim control

x-gradient amplifier

y-gradient amplifier

z-gradient amplifier

Radiofrequency amplifier/transmitter

Radiofrequencypreamplifier

Digitizer

Waveformgenerator

Workstation

WorkstationStorageserver

Reconstructioncomputer Scanner controlconsoleReal-timeanalysis

Stimuluscontrol

ControlCPU

Magnet(static field)

Gradientcoils

RF (head)coil

LCDgoggles

Joystick

Patienttable

FIGURE 6.4 Schematic organization of the fMRI scanner and computer control systems. Two systems are important for fMRI studies. Thefirst is the hardware used for image acquisition. In addition to the scanner itself, this hardware consists of a series of amplifiers and transmit-ters responsible for creating the gradients and pulse sequences (shown in black) and the recorders of the MR signal from the head coil (shownin red). The second system is responsible for controlling the experiment and for recording behavioral and physiological data (shown in green).Figure adapted from Huettel et al. (2009) with permission.

89MEASUREMENT TECHNIQUES

NEUROECONOMICS

Page 14: Neuroeconomics || Experimental Methods in Cognitive Neuroscience

collection notwithstanding, most of the exciting newdevelopments in fMRI research have come from creativenew experimental designs and increasingly sophisticatedmethods of data analysis.

PROCEDURES IN A TYPICAL fMRI EXPERIMENT

Each year, over 2000 new fMRI studies enter thecognitive neuroscience literature. Given the remarkablediversity of topics these studies investigate � coveringeverything from memory and perception to altruismand moral decision making � there is no common setof characteristics that defines the typical fMRI study.These studies tend to involve many repeated trials,like the other methods considered so far, to improvethe signal-to-noise ratio associated with the effects ofinterest. What participants can perceive and do is lim-ited by the physical environment of the scanner, whichcan be both loud and confining for research partici-pants. However, within those broad constraints, nearlyany sort of experimental design can be introduced forfMRI research.

That said, consideration of the steps of one studycan illustrate the general procedures that shape many,if not all, fMRI studies. A core challenge for neuroeco-nomic research has been to elucidate the neuralmechanisms that support decision making in the faceof economic uncertainty. In one early study, partici-pants made a series of decisions about economic gam-bles that involved economic risk (i.e., outcomes withknown probabilities) or ambiguity (i.e., outcomes withunknown probabilities). Based on a half-century ofbehavioral research, it was clear that people tend to bemuch more averse to ambiguous options than to riskyoptions, but the differences in processing associatedwith these two types of decisions remained largelyunidentified. Beginning in the mid-2000s, severalgroups began exploring the neural bases of decisionmaking under ambiguity, and we highlight one suchstudy in Box 6.5.

ADVANTAGES AND LIMITATIONS

The advantages of fMRI are evident in its wide-spread acceptance among researchers and its visibilityamong the general public. Stated simply, fMRI allowsus to map complex cognitive functions in the brains ofhuman volunteer participants with a good combina-tion of spatial and temporal resolution. It can be con-ducted on typical clinical MRI scanners � indeed,most new scanners already include basic fMRI proto-cols in their standard packages � but can also takeadvantage of cutting-edge hardware. The data it gener-ates can be subjected to a remarkably wide rangeof analyses. Researchers now use techniques fromsignal processing and computer science to examine

both the temporal interactions between regions (i.e.,functional connectivity and effective connectivity) and thelocal spatial patterns within a single brain region(i.e., multivariate pattern analysis or MVPA). Perhapsmost critically, there now exists a large and activeworldwide community of fMRI researchers who con-tinually develop new experimental designs and dataanalysis methods.

Still, despite all of its advantages, fMRI is hardly apanacea. At a logistical level, it remains expensive toconduct, with scanner charges typically around$500�1000 per hour. While its static magnetic fieldand radiofrequency signals do not pose risks in them-selves � radio waves are low-energy, non-ionizingradiation � there are potential safety risks associatedwith any strong magnet (e.g., ferrous metal will movewithin the magnetic field). Some participants willthus be excluded based on issues related to safety(e.g., implanted devices) or comfort (e.g., claustropho-bia). Some kinds of experiments may be difficult toconduct in the confined, loud bore of the MRI scanner.Moreover, even very small physiological variation �like head movements of only a few millimeters, breath-ing, or heartbeats � introduces noise into the BOLDsignal.

Probably the greatest limitation of fMRI comes,paradoxically, from its greatest strength � the flexibil-ity it affords for experimental designs and analyses.Because fMRI is amenable to so many different kindsof experiments, there has been an explosion of differ-ent approaches to studying the brain. No one experi-ment can provide definitive evidence for the mappingof a given cognitive function to a specific brainregion. Instead, evidence builds over time as a seriesof different experiments converge on a common con-clusion. Descriptions of fMRI research in the mediaoften overlook this limitation, reasoning instead fromthe conclusions of individual studies. One high-profile example described (unpublished) experimentsin which people saw images of the iPhone and othersimilar devices and found activation in the insularcortex. Based on previous work linking the insularcortex to romantic attachment, the authors inferredthat the participants felt the emotion of love whenviewing the iPhone. This exemplifies the fallacy ofreverse inference (Poldrack, 2006), or reasoning froma pattern of activation to the mental state that evokedthat pattern. When activation in a given region isevoked by a wide range of cognitive functions, as isthe case for the insular cortex, the activation aloneprovides little insight into the participants’ mentalstate. New methods of combining data across manystudies are being explored (Yarkoni et al., 2010), andthese hold promise for improving the specificity ofconclusions from fMRI research.

90 6. EXPERIMENTAL METHODS IN COGNITIVE NEUROSCIENCE

NEUROECONOMICS

Page 15: Neuroeconomics || Experimental Methods in Cognitive Neuroscience

BOX 6.5

AN EXAMPLE OF AN fMR I STUDY

Participants made decisions about simple economic

gambles that involved risk (e.g., a 50% chance of receiv-

ing $20), ambiguity (e.g., an unknown chance of receiv-

ing $30), or certainty (e.g., a sure $15). fMRI data were

collected when the participants were considering what

gamble to select. Three aspects of the results of this

study, each of which represents a way of presenting

fMRI data, are shown in the figure below. Maps of acti-

vation (Panel A) seem intuitive, but are often misinter-

preted. When a map of activation is shown in an fMRI

paper, it nearly always represents the outcome of many

statistical tests; color indicates brain regions in which

the statistical test was passed; the absence of color indi-

cates regions in which the test was not passed (or where

no test was conducted). Such an image is not a snapshot

of brain activity, or even a map of brain function. It sim-

ply indicates the results of a particular set of statistical

tests. In almost all fMRI experiments, the threshold for

significance is corrected for the number of tests con-

ducted (i.e., for the number of independent spatial loca-

tions in the brain). This means that the tests are typically

conservative, so it is impossible to claim that a brain

region “is not active” based on a given experiment.

The time course of activation (Panel B) shows

how the BOLD contrast MR signal � here in the region of

interest in the posterior inferior frontal sulcus (pIFS) �changed over the duration of the experimental trials.

Within the first phase of the trial, in which the partici-

pant is making a decision, there is a rise in BOLD signal

in this region for each of the trial types involving ambi-

guity. The pattern of changes in BOLD signal over time

is called a hemodynamic response.

Finally, fMRI data are often shown as parameter esti-

mates, or the estimated effects of the experimental condi-

tion obtained from a regression analysis (Panel C). This

typically involves creating a hypothesized model for the

changes in brain activation that would be observed if

there was an effect of the experimental condition. This

model is then used as a factor in a regression analysis.

Here, the parameter estimates were calculated for the

decision phase for each of the three trial types and were

significantly greater for decisions involving ambiguity

than for decisions involving only risk. Analysis of

parameter estimates underlies nearly all current fMRI

research. Its main advantage is that it provides a

hypothesis-based statistical framework that can be

adapted to any experimental design. However, because

it calculates statistics with respect to whatever model

the experimenter creates, poorly chosen statistical mod-

els can lead to absent or misleading results.

AC

BO

LD p

aram

eter

(a.

u.)

AR RC RR

Sig

nal c

hang

e (%

)

0.12

20

(A) (B) (C)

pIF

S

FIGURE BOX 6.5 (A) Maps of activation highlight the regions associated with the function of interest � in the present study,a greater response to ambiguity than to risk. The color of each volume element (voxel) reflects the outcome of a statistical test thatcompares its fMRI signal to that associated with an experimental hypothesis. Note that voxels not shown in color are not necessar-ily “inactive”; instead, the experiment may not have the power to draw a conclusion about their relationship to the hypothesis.(B) The average time courses of BOLD fMRI signal associated with different types of decisions: Orange, Ambiguity vs Certainty;Red, Ambiguity vs Risk; Green, Risk vs Certainty; and Blue, Risk vs Risk. The time intervals associated with the different phasesof each trial are roughly shown below the graph: presentation of the choice stimulus, waiting for the outcome of the trial to berevealed, and viewing the outcome of the trial. Because there is a lag of about 4�6 s between the BOLD signal and neuronal activ-ity, the gray rectangles indicate the windows of time in which changes in BOLD signal would be expected for each of those threephases. (C) Parameter estimates of fMRI activation (colors are the same as in Panel B). Adapted from Huettel et al. (2006) withpermission.

91MEASUREMENT TECHNIQUES

NEUROECONOMICS

Page 16: Neuroeconomics || Experimental Methods in Cognitive Neuroscience

MANIPULATION TECHNIQUES

The previous section described techniques thatreveal correlations between behavioral variables andmeasures of brain activity. These methods areimmensely important for neuroeconomics as theyallow researchers to identify whether, where, andwhen decision-relevant variables (such as values, risk,social contexts, etc.) are represented in the brain.However, none of these correlative techniques candetermine whether these neural representations areindeed involved in controlling behavior. In otherwords, do a given person’s choices really depend onthese neural computations? Would this person behavethe same way if a given region in her brain was pre-vented from computing the choice-relevant variables �or if computation was facilitated? Answers to thesequestions are fundamentally important for neuroeco-nomics, as they are prerequisites for truly mechanisticneural models of decision making. Models like thesewould allow us to predict choice behavior on the basisof brain activity and to identify the neural mechanismsthat causally underlie pathological disruptions of deci-sion making in brain disorders. Moreover, they wouldindicate potential treatment options for behavioral def-icits via manipulation of the underlying neuralprocesses.

In order to address the impact of neural processeson behavior, neuroscientists have developed severalresearch techniques to experimentally manipulate neu-ral processing in specific brain areas. Researchersusing these methods resemble engineers trying tounderstand the function of a specific part of a machine(e.g., the brakes of a car) by directly controlling thefunction of this part (e.g., pressing and releasing thebrake pedal) and examining the resulting changes inmachine operation (e.g., brake function) and output(e.g., wheel motion). In the case of neuroscience, themachine of interest is the brain, its output is behavior,and the manipulated parts are spatially and tempo-rally localized neural computations. The manipulationtechniques most commonly used for this purpose willbe presented in the following sections. These techni-ques can be grouped into two classes: brain stimula-tion techniques and techniques that study theconsequences of brain lesions. A third important classof manipulation techniques � neuropharmacologicalinterventions � are covered in a separate chapter (seeChapter 14) and will not be discussed here.

Brain Stimulation

Communication between connected neuronsdepends on the flow of electric charges. Neurons

maintain an electric resting potential of about 270 mV;when this potential rises above a fixed threshold,voltage-gated ion channels open and trigger actionpotentials. As described in the preceding chapter, fluc-tuations in dendritic potentials are normally caused bysynaptic input from other neurons. However, an exter-nally applied electrical current within brain tissue canalso affect membrane voltages and thus generateor inhibit action potentials. This general principle isused by brain stimulation techniques that produceelectrical currents in the brain in a controlled andlocally specific fashion (Clark et al., 2011; Deisseroth,2011; Wassermann et al., 2008).

By the 19th century, animal physiologists had estab-lished that an electrical pulse applied to a wire placedin an animal’s brain could reproducibly trigger veryspecific limb movements (Fritsch and Hitzig, 1870).Neurosurgeons in the early 20th century began toemploy this general technique in studies of humans.For instance, Wilder Penfield and colleagues (Penfieldand Rasmussen, 1950) attached electrodes to the corti-cal surface of human patients who were about toundergo neurosurgery and applied electrical current atvarious parts of the cerebral cortex. The behaviors andsensations elicited by stimulation of each area weredocumented in one of the first empirical maps of vari-ous motor, sensory, and cognitive functions in thehuman cortex.

Nowadays, while direct electrical stimulation ofneurons via intracranial electrodes remains a routinetechnique in animal research (see the section onInvasive Stimulation Methods in Animals), most neuros-cientists use non-invasive brain stimulation techniquesin human research as these techniques do not requiresurgery and can thus be employed routinely in healthyparticipants. The two most popular techniques aretranscranial magnetic stimulation (TMS) and transcra-nial direct current stimulation (tDCS), which will beintroduced in the following section

Transcranial Magnetic Stimulation (TMS)

HOW TMS WORKS

TMS stimulates neurons by means of electromagneticinduction. Simply put, the technique involves placing alooped copper coil against the part of the scalp overly-ing the site to be stimulated and running a strong, rap-idly changing electrical current through the coil. Likeany pulsing electric field, this one produces a magneticpulse perpendicular to the coil that permeates the skulland brain tissue without attenuation. The rapid changeof the magnetic pulse generates a complementary elec-tric field in any conductive material within the fielditself � this is how electrical transformers work, but inthis case the object in which the secondary electric field

92 6. EXPERIMENTAL METHODS IN COGNITIVE NEUROSCIENCE

NEUROECONOMICS

Page 17: Neuroeconomics || Experimental Methods in Cognitive Neuroscience

is induced is the neural tissue immediately beneath thecoil. TMS thus uses a magnetic field, which can passeasily through the skull, to generate an electrical cur-rent inside the skull. This electric current acts on theunderlying neurons and triggers action potentials.Different types of TMS protocols apply different num-bers and temporal patterns of TMS pulses, which havedifferent effects on neural processing underneath thecoil (see below for details). Importantly, the likelihoodthat an action potential will be generated at any locationdepends on the orientation of these neurons withregard to the induced electrical field (see Figure 6.5Afor a schematic summary). This means that some loca-tions in the cortex are easier to stimulate than othersusing this technique.

TECHNOLOGY

By the 19th century, brave TMS pioneers had begunto produce nerve stimulation using electromagneticinduction by showing that participants who stuck theirheads inside large metal cylinders perceived briefflashes of light when a pulsed current was run through

the cylinder. Nowadays, TMS is performed with muchmore practical and less frightening devices(Figure 6.5B). The centerpiece of this setup is the stimu-lator, which contains a high-voltage power supply andassociated electronics that can produce briefly pulsed,and highly precise, strong electrical currents in a TMScoil. Activating this circuit leads to a rapid current pulseflowing through the TMS coil on a time scale of lessthan 1 ms. Most modern stimulators then re-absorb apart of the stimulation current as it passes out of thecoil for reuse; such stimulators are thus capable ofvery short recharge times before the next pulse is gener-ated. This makes it possible to apply repetitive TMS(rTMS) pulses with a temporal separation of only a fewmilliseconds.

The TMS coils connected to the stimulator are plas-tic-encased, looped metal rings (often made of copper)with low electrical resistance. The size and spatialarrangement of the looped coil elements determinethe maximum depth and degree of focus of themagnetic field and hence the induced current. Thetwo most commonly used coil shapes are circular and

FIGURE 6.5 Transcranial magnetic stimulation (TMS). (A) Schematic of the biophysics of TMS. The electric current flowing through thecoil induces an electric field in the neural tissue below the coil. The field results in local depolarization and thus action potentials in axonsthat cross the field at appropriate orientations (e.g., perpendicular). Adapted with permission from Ruohonen (1998). (B) TMS stimulator and aconnected coil (left photo). Application of TMS to a volunteer (right photo). The reflective balls mounted on the coil and on the participant’shead are used for neuronavigation. (C) Estimated electric field strength in a plane 20 mm below a standard circular (left graph) or figure-eight(right graph) TMS coil. The figure-eight coil yields a much more focal field with a peak under the intersection of the two windings.Figure created by Anthony Barker and adapted from Walsh and Pascual-Leone (2003). (D) Strength and time course of neural excitability reductionfollowing continuous theta burst stimulation (cTBS) of motor cortex. The connected markers represent the strength of motor-evoked potentials(MEPs) elicited by single pulses of TMS over motor cortex. Such MEPs are direct measures of the excitability of motor cortex neurons to exter-nal input. Preceding cTBS reduced the size of MEPs by about 30�50% for a period of 20 minutes. A matched control condition (square mar-kers) with the same number of pulses given at 15-Hz frequency did not have this effect. Adapted from Huang et al. (2005) with permission fromElsevier.

93MANIPULATION TECHNIQUES

NEUROECONOMICS

Page 18: Neuroeconomics || Experimental Methods in Cognitive Neuroscience

figure-eight-shaped (two adjacent windings in oppo-site directions on a horizontal plane). Circular coilsgenerate powerful but more diffuse fields, whereasfigure-eight coils result in more focal fields that pro-duce the maximum current at the intersection of thetwo windings (Figure 6.5C). For conventional coils andTMS intensities, the functional resolution is estimated

to be around 1 cm, as indicated empirically by the factthat moving the coil this distance on the scalp over themotor cortex (the topographically organized brain areamost directly controlling the musculature) results inobservable changes in hand muscle activation.

Apart from TMS stimulators and coils, most TMS labo-ratories nowadays possess some sort of neuronavigation

BOX 6.6

AN EXAMPLE OF A TMS STUDY

An rTMS study (Knoch et al., 2006) examined the

functional role of the dorsolateral prefrontal cortex

(dlPFC) in reciprocal fairness, as studied by the ultima-

tum game. In this game, participants are paired up.

Player 1 of each pair divides an initial amount of money

between himself and Player 2. These offers can be fair or

unfair, and Player 2 can accept or reject the offer. If

Player 2 accepts, the money is paid out; if he rejects,

both players receive nothing. Player 2 can thus punish

Player 1 at his own cost for the unfair offer. The right

dlPFC has been found to be particularly activated when

Player 2 receives an unfair offer (Panel A). Whether this

dlPFC activation is indeed necessary for fairness-related

behavior in response to the unfair offer was then investi-

gated by comparing the offer acceptance rates of three

groups of participants who had been randomly assigned

to one of three rTMS conditions: rTMS over the right

dlPFC coordinates displayed in Panel A, over the corre-

sponding dlPFC region in the left hemisphere, or sham

rTMS without neural stimulation over either right or left

dlPFC. For each of these sites, rTMS pulses of fixed

intensity were given offline at 1-Hz temporal frequency

for 15 minutes before participants played several rounds

of the ultimatum game. The crucial effect observed in

this study was a change in behavior following rTMS to

the stimulation site compared to the baseline control

conditions: Participants with rTMS to the right dlPFC

accepted unfair offers significantly more often than par-

ticipants with rTMS to the left dlPFC or sham rTMS

(who did not differ in acceptance rates; Panel B). This

rTMS effect did not reflect changes in fairness percep-

tion, as participants in all three groups judged the offers

to be equally unfair (Panel C). Thus, reductions in neu-

ral excitability in the right dlPFC indeed caused partici-

pants to implement fairness-related punishments less

often, even though they knew the offers were unfair.

This demonstrates how TMS studies can provide evi-

dence that activity in specific brain areas is necessary for

distinct aspects of behavior.

100(A) (B) (C)

Acc

epta

nce

rate

(%

)

Fairn

ess

(1 =

ver

y un

fair,

7=

very

fair)90

80

70

60

50

40

30

20

10

0Left TMS Left TMSRight TMS Right TMSSham Sham

7

6

5

4

3

2

1

FIGURE BOX 6.6 (A) A region in the right dlPFC that showed greater activation when Player 2 received an unfair offer in anfMRI study using the ultimatum game. Adapted from Sanfey et al. (2006) with permission from Elsevier. (B) Acceptance rates for unfairoffers in the three rTMS conditions. (C) Fairness judgments about the unfair offers in the three rTMS conditions. (B) and (C) rep-rinted from Knoch et al. (2006) with permission from AAAS.

94 6. EXPERIMENTAL METHODS IN COGNITIVE NEUROSCIENCE

NEUROECONOMICS

Page 19: Neuroeconomics || Experimental Methods in Cognitive Neuroscience

system, which enables individual stereotactic localizationof brain areas based on MR images of each participant’sbrain that have been gathered prior to the TMS session.These systems usually use infrared cameras or ultrasounddetectors to accurately measure in real time the spatialpositions and orientations of the participant’s head and ofthe TMS coil. By dynamically aligning a pre-recorded MRimage of the participant’s brain with her head, the systemallows the investigator to determine which position andorientation of the coil on the scalp overlies the neuroana-tomical area that is to be stimulated.

PROCEDURES

The first step of any TMS experiment involves local-izing the scalp area overlying the cortical area that isto be stimulated. Primary motor cortex and earlyvisual cortex can be identified easily by the motortwitches or visual sensations (brief flashes of lightcalled “phosphenes”) resulting from TMS pulses deliv-ered to these brain areas. For other brain regions, how-ever, the experimenter needs to estimate where on thescalp the TMS coil needs to be placed in order toinduce currents in the target area. This can be achievedwith neuronavigated stereotactic localization (seeabove) based on neuroanatomical criteria or coordi-nates of task-related activations found in previousfMRI studies. The stimulation area can also be identi-fied as the site at which TMS has maximal behavioraleffects in a separate task performed before the actualexperiment begins. The optimal TMS intensity is usu-ally determined for each participant individually as afixed percentage of the motor threshold (MT), that is,the minimum intensity at which TMS applied over themotor cortex elicits hand twitches.

After these preliminary procedures, TMS can beapplied to influence brain activity and behavior.Neuroscience experiments have used TMS in at leasttwo different ways. First, repeated TMS pulses can beapplied online during task performance at a temporalfrequency of about 5�20 Hz. The rTMS pulses elicitunspecific neural activity in the targeted area that dis-rupts cortical computations at that location. Second,rTMS can also be applied just prior to task performance,in a so-called offline fashion. In this mode, rTMS isapplied either for several minutes at low temporal fre-quency (1 Hz) or for less than a minute in what is calleda theta burst pattern (theta burst patterns are typically3�5 pulses at 100 Hz repeated at 5 Hz). Both types ofoffline rTMS produce neural after-effects, lowered corti-cal excitability in the stimulated area that persists for10�30 minutes. These after-effects thus offer a temporalwindow in which the normal functional contributionsof the stimulated area and possibly interconnectedbrain areas are markedly reduced (Figure 6.5D).

To assess the behavioral consequences of online oroffline rTMS to a given area, behavioral performanceduring or after rTMS is compared with that during abaseline control condition. This is necessary to controlfor unspecific side effects of TMS, such as the associ-ated clicking noises and tactile sensations at the scalpproduced by the stimulation of scalp nerves.Suitable control conditions involve TMS over anotherscalp position, TMS over the same site but at differenttime periods during task performance, or neurallyineffective “sham” TMS with special coils that producesimilar sounds but no magnetic field. This latter strat-egy, however, does not control for the tactile sensationof TMS at the scalp or any spatially unspecific effect ofneural stimulation.

ADVANTAGES AND LIMITATIONS

TMS allows non-invasive manipulation of neuralprocessing with high spatial resolution (about one cen-timeter) and exceptional temporal resolution (millise-conds). It can be employed very flexibly with respectto temporal profiles and patterns of stimulation thatcan have markedly different effects on neural proces-sing and behavior. Finally, TMS can be employed inalmost any healthy volunteer who meets a few basichealth-related criteria (e.g., absence of proneness toepilepsy and of previous brain damage or brain illness;see Rossi et al. (2009) for a detailed list of thesecriteria).

Like any research technique, however, TMS also hassome disadvantages. Due to the drop-off of the mag-netic field with increasing distance from the coil, it ispresently only possible to target brain areas on the cor-tical surface, not deeper brain areas that would be ofconsiderable interest to neuroeconomists (e.g., striatum,medial prefrontal cortex). Moreover, the noise and tac-tile sensations produced by TMS can be experienced asdistracting or painful by some participants, althoughthis can be partially avoided by the use of ear protectionand suitable coil positioning. These side effects of TMSmay complicate comparisons of behavioral effects fordifferent stimulation sites and can make it difficult toconduct blind or double-blind studies in which partici-pants and/or experimenters are unaware of the specificstimulation condition. Finally, for offline studies, thereis some uncertainty about the precise duration of thetime window of TMS after-effects during which behav-ioral tests can be conducted.

Transcranial Direct Current Stimulation (tDCS)

HOW tDCS WORKS

From a technical point of view, tDCS is straightfor-ward. It involves attaching two electrodes to the scalpand applying a constant electric potential difference,

95MANIPULATION TECHNIQUES

NEUROECONOMICS

Page 20: Neuroeconomics || Experimental Methods in Cognitive Neuroscience

thus running a weak but constant electrical currentbetween them. This affects the neurons along the pathof the current, slightly changing their membrane vol-tages and thus their spontaneous firing. These effectsare, of course, strongest directly beneath the electrodeswhere the current density is highest. For tDCS withconventional intensities, these effects on neural func-tion have opposite polarity for the positively (anode)and negatively (cathode) charged electrodes: Neuralexcitability and spontaneous firing is increased underthe anode, but decreased under the cathode. Thisallows tDCS to be used in two modes: anodal tDCS toupregulate and cathodal tDCS to downregulate neuralprocessing in a brain region (Figure 6.6D).

Importantly, as for TMS, the effects of tDCS can out-last the duration of stimulation. Neural excitability ofmotor cortex, as measured by motor evoked potentialsresulting from TMS, changes during stimulation andcontinues to be increased or decreased for up to 60minutes following cathodal or anodal tDCS, respec-tively. This means that the behavioral effects of activitymanipulations via tDCS can be studied both duringand after stimulation.

tDCS usually involves the application of constant cur-rents, but the electrode setup described above can alsobe used to deliver an alternating current that changes itspolarity at a specific frequency. This form of stimulationis referred to as transcranial alternating current stimulation(tACS). tACS is mostly used to study the functional roleof oscillatory neural activity in specific frequency bands(e.g., alpha/8�13 Hz, beta/13�35 Hz, or gamma/35�65 Hz bands). These oscillations have been found tocorrelate with specific cognitive states in EEG or MEGstudies (see the section on Non-Invasive Neurophysiology).By externally inducing (or “entraining”) oscillations inthe membrane voltages of the underlying neurons, tACScan provide information on the importance of oscillatoryneural activity in specific frequency bands for behavior.

tDCS DEVICES

All that is needed for tDCS is a power source capa-ble of safely generating a weak (1�2 mA) constantdirect current and two electrodes that can be attachedto the scalp. tDCS stimulators are often battery-powered but may involve rechargeable units andmore sophisticated electronics both to generate

FIGURE 6.6 Transcranial direct current stimulation (tDCS). (A) Simultaneous testing of participants in a tDCS experiment. tDCS is admin-istered via electrodes that are wrapped in sponges soaked in saline solution and mounted to the head. Photo courtesy of Marc Latzel.(B) Schematic drawing of the components of a tDCS stimulator. The stimulator allows administration of current of variable strength and wave-form for a specified duration; actual current output is measured continuously. The current can be slowly ramped up and down at the transi-tion points of a stimulation sequence. The stimulators contain safety mechanisms that prevent excessive or uncontrolled stimulation. Adaptedfrom Nitsche et al. (2008) with permission from Elsevier. (C) Standard electrode montages for stimulation of visual cortex (left drawing) andmotor cortex (right drawing). The electrode positions are illustrated on a schematic head (nose on top, ears on left and right) in accordancewith the 10�20 electrode system. Adapted from Nitsche et al. (2008) with permission from Elsevier. (D) Strength and direction of neuromodulatoryeffects resulting from tDCS. The boxplots represent the distribution of MEP amplitudes elicited by TMS pulses to motor cortex immediatelyafter 4-second episodes of anodal (white) or cathodal (striped) tDCS. MEP amplitude is expressed relative to baseline MEP amplitude withouttDCS, thus demonstrating increases and decreases in neural excitability by anodal and cathodal tDCS, respectively. Adapted from Nitsche andPaulus (2000) with permission from Blackwell Publishing Ltd.

96 6. EXPERIMENTAL METHODS IN COGNITIVE NEUROSCIENCE

NEUROECONOMICS

Page 21: Neuroeconomics || Experimental Methods in Cognitive Neuroscience

reliable and customized current waveforms (e.g., asused for tACS) and to ensure subject safety. It isessential that the stimulators contain safety mechan-isms that prevent excessive or uncontrolled currentsto be delivered to the participants. The mostadvanced stimulators at present contain several out-put channels that can be independently controlled, sothat several participants can be stimulated with differ-ent current types simultaneously with the same stim-ulator (Figure 6.6A, B on p. 94).

tDCS electrodes are usually made of silicone orrubber and attached to the head with wide rubberbands and a medium that ensures good electricalconductivity, for example, a sponge soaked in saltwater or conductive electrode paste. The size of thetDCS electrodes determines the current density andhence the focality of the current delivered to thebrain.

PROCEDURES

As for TMS, the first step of a tDCS experimentinvolves localizing the scalp site at which the activeelectrode should be attached to stimulate the targetsite. A second important issue to be solved prior tostimulation concerns where the reference electrodeshould be attached. The reference electrode is theelectrode with effects that are not of interest in agiven study. Thus, in a study that seeks to increaseexcitability, the anode is referred to as the active elec-trode and the necessary but irrelevant cathode isreferred to as the reference electrode. The location ofthe reference electrode is very important as it deter-mines the direction of current flow and hence the pre-cise stimulation effects under the active electrode.Some standard electrode montages, or spatial configura-tions, have been validated for tDCS of the motor andvisual cortex (Figure 6.6C); for other stimulation sites,it is less clear which montage is optimal. In general,selecting a reference electrode that is much larger thanthe active electrode will reduce the effect of that elec-trode on local neural processing due to diffusion of thecurrent.

After setup, tDCS is applied by running the currentfor about 10�20 minutes. Participants often perceive abrief tingling/itching sensation on the scalp at theonset of stimulation, but this sensation quickly fadesaway and leaves participants unaware of whether theyare being stimulated or not. A perfect control conditionfor tDCS thus consists of a current that is switchedoff after about 30 s without the knowledge of theparticipants.

ADVANTAGES AND LIMITATIONS

It has been argued that tDCS is well suited forstudying subtle decision processes, in particular in

social situations, for several reasons. First, tDCS doesnot have any distracting side effects such as noise orpersisting tactile sensations. Second, it offers a verygood control condition that is perceptually indistin-guishable from active stimulation. This means thattDCS can be (double-)blinded, which may be essentialfor decision-making situations prone to expectationsand demand effects. Third, tDCS is inexpensive andeasy to use so a group of participants can be testedsimultaneously (Figure 6.6A), which may be essentialfor studies of social decision processes. Fourth, beingable to either up- or downregulate neural excitability(Figure 6.6D) allows the researcher to conduct interest-ing tests of the functional role of both enhancementsand reductions of neural function.

Despite all of its advantages, tDCS also has signifi-cant disadvantages as compared to TMS. For instance,its spatial resolution is much lower so it is difficult toassume that neural processing only changes in a veryfocal cortical region. Questions concerning the place-ment of the reference electrode may complicatethe interpretation of tDCS results. Finally, tDCS is nottemporally precise, as its effects are continuouslyexpressed throughout the stimulation period and per-sist in its aftermath. tDCS and TMS thus occupy some-what different niches in the neuroeconomist’s toolbox.The former is most often used to modulate ongoingtask-related neural activity in a manner that is virtuallyimperceptible to the participants, whereas the latter ismost often used to disrupt normally occurring patternsof neural activity in a spatially and temporally precisefashion.

Invasive Stimulation Methods in Animals

MICROSTIMULATION

In contrast to research methods employed in healthyhumans, neural manipulation techniques for animalstudies are often invasive (i.e., require neurosurgicalprocedures). As outlined in the section InvasiveNeurophysiology: Single-Unit Recording, microelectrodesare routinely inserted into the brains of non-humanprimates and rats to record electrical activity fromsmall populations of neurons. Intracortical electrodeslike these are not only used for recording neural activ-ity, but can also be employed for inducing it. This tech-nique � referred to as microstimulation � involves theapplication of weak electric currents to affect the activ-ity of neurons in the direct vicinity of the electrode. Inrare cases, invasive electrical stimulation by means ofimplanted electrodes is also used in humans to treatchronic and severe brain disorders such as Parkinson’sdisease, depression, or obsessive�compulsive disorder(see also the section Invasive Neurophysiology: Single-Unit Recording).

97MANIPULATION TECHNIQUES

NEUROECONOMICS

Page 22: Neuroeconomics || Experimental Methods in Cognitive Neuroscience

BOX 6.7

AN EXAMPLE OF A tDCS STUDY

A tDCS study (Fecteau et al., 2007) tested the func-

tional role of neural activity in the right dlPFC in the

control of risk-taking behavior. Previous studies had

repeatedly found that lesions in, or TMS to, right dlPFC

were associated with increased risk-taking, suggesting

that neural activity in this brain area may be necessary

for curbing the impulse to select more rewarding but

riskier behavioral options. In the tDCS study, the investi-

gators tested whether upregulating neural excitability

in the right dlPFC led to diminished risk-taking.

Participants performed a risk-taking task in which they

were presented with six boxes per trial. Each box was

either pink or blue and one of the six boxes contained a

financial reward. The ratio of pink to blue boxes deter-

mined the level of risk, as the majority color was associ-

ated with a higher probability of winning (lower risk)

but a lower financial reward, whereas the minority color

was associated with a lower chance of winning (higher

risk) but a higher reward. On each trial, participants

chose one of two colors. If the winning box turned out

to be of the same color, they won the associated amount

of points; if it did not, they lost the associated amount of

points. The chosen color thus indicated whether partici-

pants had chosen the high-risk, more rewarding option

or the low-risk, less rewarding option (Panel A). The

participants were assigned to one of three tDCS groups:

anodal tDCS over the right dlPFC paired with cathodal

tDCS over the left dlPFC; cathodal tDCS over the right

dlPFC paired with anodal tDCS over the left dlPFC; or a

sham control condition. In the two active conditions,

tDCS was performed during the whole duration of the

task, whereas in the sham condition, it was switched off

after 30 s. As hypothesized, the study found that anodal

tDCS to the right dlPFC led to a higher proportion of

low-risk choices as compared to both cathodal and sham

tDCS to the right dlPFC (Panel B). Anodal tDCS to the

right dlPFC also led to higher overall earnings for the

participants (Panel C), indicating that the diminished

risk-taking due to brain stimulation was a better strategy

for this game than choosing the high-risk, high-reward

option. This study demonstrates how tDCS can be used

to test hypotheses about the functional role of neural

activity upregulation in specific brain areas. However,

as the reference electrode was always placed over the

contralateral dlPFC, it is safest to conclude that the

behavioral effects reflect the interhemispheric balance of

activity across both dlPFCs rather than just effects

induced in one hemisphere.

100(A) (B) (C)

Level of risk

Balance ofreward

800

600

400

200

Poi

nts

earn

ed

Cho

ice

of lo

w-r

isk

pros

pect

(%

)

0

90

**

**

80

70R anodal/L cathodal

R anodal/L cathodalL anodal/

R cathodal

L anodal/R cathodalSham

Sham90

90

10

10

WIN

Points: 10

Points: 100

FIGURE BOX 6.7 (A) Schematic illustration of the task. The participant chose the color she expected the reward to appear in.The two colors were associated with different probabilities of winning (number of boxes) and numbers of points (the numbers inthe colored boxes). After the choice was made, the winning color was revealed and the corresponding number of points was added(for wins) or subtracted (for losses) from the participant’s total. (B) Proportion of choices in which the participant chose the low-risk option. This was significantly increased for right anodal/left cathodal dlPFC tDCS (black bars). (C) Number of points earnedduring the task. Right anodal/left cathodal dlPFC tDCS led to higher total earnings. Adapted from Fecteau et al. (2007) with permis-sion from Society for Neuroscience.

98 6. EXPERIMENTAL METHODS IN COGNITIVE NEUROSCIENCE

NEUROECONOMICS

Page 23: Neuroeconomics || Experimental Methods in Cognitive Neuroscience

In close analogy to the use of single TMS pulses,microstimulation is often used to induce “surrogate”neural activity in order to study how the animal’sbehavior is affected by action potentials generated inthe stimulated area. Classic examples of this approachare the self-stimulation studies pioneered by Olds andMilner (1954). In those studies, rats were given theoption to press a lever in order to electrically stimulateneurons in the septal area of their own brain via aninserted electrode (Figure 6.7A). The rats rapidlylearned to press the lever and did so repeatedly forlong periods of time, establishing that neural activityin the septal area is positively reinforcing.

From a technical point of view, microstimulationrequires the same general setup as the invasive neuro-physiology techniques described in the section InvasiveNeurophysiology: Single-Unit Recording. The only differ-ence is that a current generator (rather than a record-ing device) needs to be connected to the intracorticalelectrode. The current running through the electroderesults in a local electrical field that depolarizes neu-rons near the electrode and leads to the generation ofaction potentials (Figure 6.7B). It should be noted thatwhile most electrical stimulation studies in animalsuse the implanted electrode as the cathode, this excites

rather than inhibits neurons around the electrode tip.For technical reasons beyond the scope of this chapter,both anodal and cathodal deep brain stimulation resultin action potentials being generated, but more reliablyso in the cathodal mode. Irrespective of the precisestimulation mode, this method can be used to induceneural activity in a spatially and temporally precisefashion (millimeters and milliseconds, respectively).A major advantage of this method is that it can beapplied to any cortical or deeper-lying neural structureinto which electrodes can be inserted. Finally, it isadvantageous that the animal is not aware of beingstimulated, as the brain does not contain sensoryreceptors that perceive the electrical current as a directsensation. Microstimulation studies thus often do notrequire stimulation of a control site as behavior can becompared between conditions with different stimula-tion patterns or with and without stimulation at thesame electrode.

OPTOGENETICS

The cortex contains many different types of neuronsthat respond to different types of neurotransmittersand project to different local or more remote targets.All these different types of neurons are stimulated by

Guidecannula

Opticalfibre

Optogenetic excitationElectrical stimulation Optogenetic inhibition

(A) (C)

(B) (D)

FIGURE 6.7 Microstimulation and optogenetics. (A) A rat self-stimulates its brain by pressing a lever that releases electric current in animplanted microelectrode in the septal area. Olds and Milner (1954) used this setup to show that the lever pressing is strongly reinforced byelectric stimulation, suggesting that the septal area encodes signals that positively reinforce actions. Reprinted from Joseph (2000) with permissionfrom Elsevier. (B) Schematic of the biophysics of electrical stimulation. The microelectrode induces a local current (drawn in red) that locallydepolarizes all types of neurons in its vicinity and thus induces action potentials (red flash symbols). Adapted from Deisseroth (2011) with per-mission from Macmillan Publishers Ltd. (C) Optogenetic stimulation of a genetically prepared mouse via blue laser light shining on a specificneural structure though a surgically implanted glass fiber tube. Adapted from Airan et al. (2009) with permission from Macmillan Publishers Ltd.(D) Schematic drawings of the biophysical principles underlying optogenetic excitation or inhibition. Blue light selectively activates a neuronwith genetically engineered, blue light-activated channelrhodopsin ion channels (left drawing). Yellow light selectively inhibits neurons withyellow light-activated halorhodopsin ion channels (right drawing). Adapted from Deisseroth (2011) with permission from Macmillan Publishers Ltd.

99MANIPULATION TECHNIQUES

NEUROECONOMICS

Page 24: Neuroeconomics || Experimental Methods in Cognitive Neuroscience

the electrical currents induced via the electrical stimula-tion methods covered in the previous sections. It is thusdifficult to ascribe any behavioral changes during elec-trical stimulation to effects on one specific cell type orneurotransmitter system. This shortcoming of existingstimulation methods has recently triggered the develop-ment of optogenetic approaches, which allow research-ers to control the action potentials of specific cell typeswith a combination of genetic engineering and the intra-cranial application of light (Deisseroth, 2011).

Optogenetics was made possible by the discoveryand characterization of light-sensitive proteins in themembranes of microbes. These proteins act like ionchannels in response to light of a specific wavelength(e.g., blue): They open in response to such light, lead-ing to an increase in the inflow or outflow of specificcharged atoms (like sodium) which changes the mem-brane voltage. Advances in genetic engineering havemade it possible to insert DNA for these proteins intothe neurons of a living animal’s brain in such a way

BOX 6.8

AN EXAMPLE OF A M ICROST IMULAT ION STUDY

Classic microstimulation studies in non-human pri-

mates have identified the precise neural signals necessary

for perceptual decisions about somatosensory stimuli or

visual motion. In one such study (Romo and Salinas, 2001),

macaque monkeys were to report which of two vibrotactile

stimuli applied to their fingertip had the higher stimula-

tion frequency. The experimenters had inserted a micro-

electrode into the primary somatosensory cortex where

neuronal firing patterns correlate with the frequency of the

tactile stimulus. When the experimenter substituted

microstimulation of somatosensory cortex for one of the

tactile stimuli to the fingertip (Panel A), the monkeys

continued to perform the task with the same level of

accuracy � but they were now judging the frequency of

the action potentials induced artificially in somatosen-

sory cortex rather than the tactile stimulus on the skin

(Panel B). This study thus showed that microstimulation

can indeed induce specific patterns of neural activity in

a given brain area that the monkey uses to control his

choice behavior.

12

Com

paris

on fr

eque

ncy

judg

ed h

ighe

r (%

)

100

75

50

25

0

100

75

50

25

0

16

Comparison frequency (Hz)

2420 28

n = 8

n = 8

(A)

(B)

FIGURE BOX 6.8 (A) Schematic illustration of the vibrotactile task. The monkey is first presented with sinusoidal stimulationto the fingertip and then either with a second tactile stimulus or with cortical microstimulation � both with a different frequencythan that of the first tactile stimulus. The monkey’s task is to determine which of the two stimuli has the higher frequency.Different versions of this task present the second stimulus with different waveforms (top vs bottom drawing), which does notaffect performance. (B) The monkey’s performance is identical for tactile and microstimulation stimuli. Adapted from Romo andSalinas (2001) with permission of Annual Reviews, Inc.

100 6. EXPERIMENTAL METHODS IN COGNITIVE NEUROSCIENCE

NEUROECONOMICS

Page 25: Neuroeconomics || Experimental Methods in Cognitive Neuroscience

that these ion channels are produced only in very spe-cific cell types. The ion channels can then be experi-mentally opened by light of the respective wavelength,which is locally applied via glass fiber tubes insertedinto the neural structure of interest through the ani-mal’s skull in much the same way an electrode isinserted (Figure 6.7C). By switching the light on or off,the experimenter can thus induce action potentials inspecific cell types, a degree of specificity not possiblewith electrical stimulation.

Optogenetics has just recently been developed inthe last few years and thus has many unresolvedissues of practical importance. However, resultsacquired in mice and other small animals have beenspectacularly successful in demonstrating a linkbetween the activation of specific cell types and behav-ior (e.g., Box 6.9). Optogenetic approaches have alsomade it possible to empirically confirm long-heldassumptions about causal associations between differ-ent neurophysiological phenomena (e.g., action poten-tials and the BOLD effect measured in fMRI; Lee et al.(2010)). The coming years will show whether thesevery promising methods can also be routinelyemployed in neuroeconomics studies in animals orperhaps in therapeutic approaches to human braindisorders.

Lesion Studies

The earliest and most striking demonstrations thatspecific aspects of behavior depend on brain functionwere provided by studies that systematically investi-gated the behavioral deficits associated with braindamage (Kolb and Whishaw, 2009; Shallice, 1988).Historically, these studies were closely tied to the clini-cal fields of neurology and neurosurgery, as braindamage resulting in behavioral deficits usually occursin humans as a consequence of an accident or illness.Understanding these deficits is very important fortheir diagnosis, treatment, and rehabilitation. Thesepatients also, however, produce interesting basic sci-ence findings on brain�behavior relationships.Brain lesions in animals can also be experimentallyinduced in the laboratory, which enables scientiststo test anatomically specific hypotheses about therelevance of brain areas for specific behaviors. Thesetwo approaches will be discussed in the followingsections.

Lesion Studies in Humans

The study of behavioral deficits in patients withbrain damage, often referred to as neuropsychology,originated in the neurological clinic. Pioneers of thisapproach in the 19th and 20th centuries systematically

documented their observations of behavioral disrup-tions in individual patients (see Figure 6.8A for afamous example). The reports of some of theseresearchers even resulted in a neurological syndromeor brain area being named after them. For instance,Paul Broca and Carl Wernicke described distinct typesof speech disorders following stroke-related damage toregions in the left inferior frontal gyrus or the left pos-terior superior temporal gyrus; these regions are nowwidely referred to as Broca’s area and Wernicke’s area,respectively. Since the early days, numerous system-atic relationships between brain damage and behav-ioral disruptions have been documented and havethus shaped our understanding of brain�behaviorrelationships.

HOW LESION STUDIES WORK

There are many ways in which people might sustaindamage to parts of their brain. For instance, vascularconditions (e.g., strokes) and head trauma (e.g., due tofalls or accidents; Figure 6.8A) often cause brain dam-age. Tumors and their surgical removal, infectious dis-eases (e.g., meningitis and encephalitis), and metabolicpathogens (e.g., neurotoxins such as alcohol) are alsofrequent causes of brain damage. Depending on thecircumstances, this damage can be more or less focaland can even selectively affect different types of neu-rons. One of the greatest challenges in neurologicalresearch is thus to determine the exact scope andextent of the neural damage associated with the givencondition.

PROCEDURES

To test a hypothesis about the functional role of agiven brain area using the lesion approach, researchersfirst identify a group of patients with more or lessselective damage to that brain area. An important stepis the reconstruction of the full extent and overlap ofthe lesions, ideally with MRI and possibly functionalmeasures of brain activity (Figure 6.8B). As it is nor-mally not possible to measure behavior prior to thebrain damage (because the injury cannot be antici-pated), it is necessary to identify a suitable controlgroup for behavioral comparison. To render such com-parisons meaningful, the control participants need tobe closely matched to the patients with respect tobehaviorally relevant factors such as age, intelligence,socioeconomic status, cultural background, etc.

Most lesion studies then measure and comparebehavioral performance across the two groups using aseries of tasks designed to isolate specific componentsof cognition or behavior. If the control group consistsof healthy participants, such comparisons can revealsingle dissociations, which involve selective impairmentrelative to the controls on some behavioral tests but

101MANIPULATION TECHNIQUES

NEUROECONOMICS

Page 26: Neuroeconomics || Experimental Methods in Cognitive Neuroscience

not others (Figure 6.8C). For the detection of double dis-sociations, patients are grouped according to brainlesion site and compared. Double dissociations arepresent if one lesion group shows deficits on Task 1but not on Task 2 and the other lesion group on Task 2but not on Task 1 (Figure 6.8C). As this pattern ofresults cannot be explained by general cognitive defi-cits potentially associated with any brain lesion, it isoften argued that a double dissociation constitutes thestrongest empirical support for the notion that two

behavioral or cognitive functions can be fully sepa-rated in terms of underlying neural computations.

ADVANTAGES AND LIMITATIONS

Behavioral deficits due to brain lesions can be veryprofound; in some cases, they are even evident tountrained observers (e.g., Figure 6.8D). Severe deficitsmay provide much stronger support for the behavioralnecessity of a brain area than the subtler changes intask performance found in brain stimulation studies.

BOX 6.9

AN EXAMPLE OF AN OPTOGENET IC S STUDY

Using optogenetic methods in mice, a study (Tsai

et al., 2009) was able to empirically confirm the long-

held hypothesis that behavioral conditioning directly

depends on the phasic firing of dopaminergic cells in

the ventral tegmental area (VTA). The investigators

genetically engineered mice with light-sensitive excit-

atory ion channels specifically expressed in the dopami-

nergic neurons of the VTA. The application of light

pulses to these neurons through a fiber optic cable

inserted into the brain reproducibly led to strong bursts

of action potentials in these neurons as ascertained by

various neural recording techniques. The experiment

then used a standard conditioned place preference par-

adigm in which a rat is allowed to roam freely between

two connected test chambers (Panel A) and the time

spent in each chamber is an index of the preference for

that location (Panel B). Numerous studies have

established that pairing one of the two locations with

appetitive stimuli (e.g., food) results in the rat spending

more time in that location, even if administration of the

appetitive stimulus ceases. In the optogenetic study, the

investigators paired one of the locations with light-

induced phasic (50-Hz) stimulation of dopaminergic

VTA neurons. In matched control conditions, the same

location was either paired with tonic (1-Hz) stimulation

or no stimulation. As hypothesized, only phasic stimu-

lation of the dopaminergic VTA neurons led to a strong

conditioned preference for the location paired with

stimulation (see Panel B). This finding demonstrates

that reward-related approach behavior is directly influ-

enced by phasic firing of dopaminergic cells in the

VTA. More generally, this study illustrates how optoge-

netic approaches can be used to show that specific pat-

terns of neural activity in a pre-defined cell type of a

given neural area are sufficient to elicit specific types of

behavior.

Pre-Test Post-Test(A) (B)

150 s

0 s

Toni

cP

hasi

c

FIGURE BOX 6.9 (A) Schematic display of the conditioned place paradigm. The mouse is free to roam between two intercon-nected chambers; see right drawing for an example of a running path. The time spent in each chamber is taken as an index of placepreference. Adapted from Airan et al. (2009) with permission from Macmillan Publishers Inc. (B) “Heat maps” displaying the time spentin the chamber associated with phasic or tonic optogenetic stimulation of dopaminergic VTA neurons. A clear preference (longerstaying times) for the chamber associated with tonic stimulation emerges in the post-test phase, after stimulation. Adapted fromTsai et al. (2009) with permission from Macmillan Publishers Inc.

102 6. EXPERIMENTAL METHODS IN COGNITIVE NEUROSCIENCE

NEUROECONOMICS

Page 27: Neuroeconomics || Experimental Methods in Cognitive Neuroscience

Moreover, the behavioral deficits resulting from natu-rally occurring illnesses or accidents can be very unex-pected; this can lead to entirely new hypotheses aboutbrain�behavior relationships that otherwise would nothave been considered. Finally, the knowledge gainedfrom lesion studies is always relevant for medical careas it specifies behavioral deficits in patients with spe-cific types of brain damage, which may help the diag-nosis and treatment of these disorders.

The most obvious disadvantage of the lesionapproach is that naturally occurring brain damage isoften spatially diffuse and seldom selective to specificbrain areas. This can make it very difficult to findpatients with overlapping damage in the structures ofinterest and to assign all of their deficits just to these

brain areas. Moreover, lesion studies offer no informa-tion about the timing of neural activity as the effects ofbrain lesions are constant and usually irreversible.Often little is known about the patients’ behavior priorto the accident or illness; there is thus uncertainty as towhether the deficits observed in a patient reflect theeffects of brain damage or simply independent behav-ioral idiosyncrasies. Finally, brain injuries and illnessesand their treatment can have nonspecific sequelae thatmay affect behavior, such as brain reorganization,medication effects, or an altered life situation. Some ofthese effects can be controlled by appropriate experi-mental designs (see above), but may nevertheless affectthe strength of the conclusions that can be drawn froma single study.

Anterior

Posterior

Number of subjects1 3

1 5

(A) (B)

(C) (D) (E)

FIGURE 6.8 Lesion studies. (A) One of the most famous brain lesion patients. A dynamite explosion drove an iron rod (the one shown inthe left photo) through railroad worker Phineas Gage’s skull (middle photo), thereby destroying his left eye and parts of his left prefrontal cor-tex (see right photo for a present-day simulation). Gage survived, but showed substantial behavioral changes from that day onwards, includ-ing irresponsibility, lack of foresight, bad temper, and impulsiveness. Doctors at the time suggested that the brain lesion led to a destructionof “the equilibrium or balance, so to speak, between his intellectual faculties and animal propensities.” This simplified notion of the effects ofprefrontal cortex damage has been considerably refined since then. Left photo courtesy of Wikimedia Commons; middle and right photos reprintedfrom Van Horn et al. (2012) with permission. (B) Lesion overlap analysis in a series of patients with damage to the ventromedial prefrontal cortex(vmPFC, top row) or dorsolateral prefrontal cortex (dlPFC, bottom row). The colors overlaid on the different MR image slices indicate thenumber of patients who show damage to the colored structure (see color legend at bottom of slice). Adapted from Fellows and Farah (2003) withpermission from Oxford University Press. (C) Schematic examples of a single dissociation (top) and a double dissociation (bottom). In a single dis-sociation, patients are impaired on one task, but not on another, as compared to other patients and healthy controls. In a double dissociation,patients of one group show deficits on one task, but not on another, while patients of another group show the opposite pattern of deficits.(D) Drawing of a patient with damage to the right parietal cortex. The patient exhibits hemispatial neglect, a syndrome in which patients failto represent one half (in this case, the left half) of space and thus only draw half (here, the right half) of a visual object shown to them despitefully intact vision. Adapted from Husain and Rorden (2003) with permission from Macmillan Publishers Ltd. (E) Post mortem lesion overlap analysisin three macaque monkeys with experimental lesions of the orbitofrontal cortex. The left column draws the regions that were meant to beaffected by the experimental lesions. The gray values overlaid on the brain slices in the right column indicate the number of monkeys thatactually showed damage in the respective region in the post mortem analysis. Adapted from Walton et al. (2010) with permission from Elsevier.

103MANIPULATION TECHNIQUES

NEUROECONOMICS

Page 28: Neuroeconomics || Experimental Methods in Cognitive Neuroscience

Experimental Lesions in Animals

The medical importance of understanding thepathophysiology and behavioral effects of brain lesionsin humans has triggered considerable interest in lesionmodels in experimental animals. In this kind ofresearch, lesions are generated in clearly defined brainregions by various means so that therapeutic measuresand the time course of recovery can be studied. Thishas opened the possibility for research onbrain�behavior relationships with full experimentalcontrol over the precise site and shape of the brainlesion.

HOW THE ANIMAL LESION APPROACH WORKS

After determination of the neural structures of inter-est in three-dimensional space using brain atlases andstereotactic devices (see the section on TranscranialMagnetic Stimulation), surgery is performed to produce

a lesion at the designated site. This can be achieved indifferent ways: by mechanical removal of the tissueusing surgical devices, by local injection of a neuro-toxin that binds selectively to specific receptors anddestroys the corresponding cells, by application of astrong local electric current to damage tissue, or byinsertion of a probe that can be cooled to a tempera-ture that prevents the cells from functioning normally.The effects of the first three methods are often irrevers-ible, whereas the effects of cooling can be reversed tostudy recovery.

PROCEDURES

Lesion experiments in animals usually involve anexperimental and a control group of animals thatundergo matched procedures to rule out any unspe-cific effects of training, surgery, etc. The two groupsare trained to a criterion in the task used to measure

BOX 6.10

AN EXAMPLE OF A NEUROPSYCHOLOG ICAL LE S ION STUDY

This study (Fellows and Farah, 2003) used a neuro-

psychological lesion approach to test the hypothesis that

the ventromedial prefrontal cortex (vmPFC) is necessary

for flexible updating of stimulus�outcome associations.

The authors compared patients with lesions in either the

vmPFC or the dlPFC (labelled VMF and DLF in

Figure 6.8B) with participants in a matched control

group. The computerized task used to measure stimu-

lus�outcome associations was simple: It required the

participants to pick one of two cards drawn from stacks

of different colors. Cards from one stack always resulted

in a play money win of $50, cards from the other

stack in a loss of $50. After eight consecutive picks

from the winning stack, the contingencies between

card color and win/loss were switched. Patients and

controls showed similar performance during the initial

learning of the color�outcome association (“learning

errors”; left bars in Panel A). However, after reversal of

the association, the patients with vmPFC lesions made

significantly more incorrect choices than both the dlPFC

patients and the controls (“reversal errors;” right bars in

Panel A). An image (Panel B) displaying the lesion over-

lap for the participants with severe behavioral impair-

ments revealed one particular region in the left vmPFC

(in green/yellow) in which structural damage was maxi-

mally associated with behavioral consequences in stimu-

lus�outcome reversal learning. Adapted from Fellows and

Farah (2003) by permission of Oxford University Press.

Learning errors Reversal errors

15(A) (B)

*

10

5

0

CTL

DLF

VMF

1 10

FIGURE BOX 6.10

104 6. EXPERIMENTAL METHODS IN COGNITIVE NEUROSCIENCE

NEUROECONOMICS

Page 29: Neuroeconomics || Experimental Methods in Cognitive Neuroscience

behavior and their preoperative performance level isrecorded. Then surgery is performed and the desig-nated lesions are made. The control group also under-goes surgery, but the procedures do not involve harmto the brain (e.g., no neural tissue is removed duringthe surgical procedure). Behavioral tests are then con-ducted to measure how task performance has changedas a result of the lesion. To demonstrate the behavioralrelevance of the lesioned brain area and to control forany unspecific side effects of the surgical procedures,the experimental group needs to show a significanteffect relative to the control group. Some studies alsofollow a more stringent double-dissociation logic(Figure 6.8C) by testing the hypothesis that lesions totwo different brain areas cause selective complemen-tary deficits in only one of two experimental tasks. Atthe end of testing, the extent of the lesions is documen-ted by detailed post mortem neuroanatomical and neu-rochemical examination of the brain tissue(Figure 6.8E).

ADVANTAGES AND LIMITATIONS

Experimental studies in animals allow full controlover many variables that vary randomly in the contextof pathological brain lesions in humans. For instance,the experimenter can determine the precise neuroana-tomical location and extent of the lesions. Moreover,animals can be randomly assigned to either lesion orcontrol group and can be perfectly matched in termsof experience, life situation, and presurgical task per-formance. Finally, the effects of medication and treat-ment (which patients unavoidably receive) cannotconfound the results, and the characteristics of theinduced lesions can be very precisely determined postmortem.

Animal lesion studies are, however, difficult to con-duct � particularly in non-human primates. The train-ing and keeping of experimental animals can be verylabor-intensive and costly and surgery and behavioraltesting require considerable infrastructure. Apart fromthese practical problems, it is generally difficult tocompare behavior across species, so good models ofspecific human behaviors may be hard or impossibleto identify in animals. This is less of a problem forexperiments concerning sensory brain function(e.g., vision, audition), but may strongly affect thestudy of more complex aspects of human behavior(e.g., decision making, social behavior, language).Creative research designs are needed to overcome thislimitation and train animals to exhibit potentiallyhomologue behaviors. However, in such cases, ques-tions always remain about the degree to which thebehavior under study reflects the animal’s naturalbehavior or simply involves over-trained artificialstrategies. Irrespective of these difficulties, studies of

animal lesion models have resulted in considerableknowledge about brain�behavior relationships thatwould be hard or impossible to derive through studiesof human brain lesions alone.

CONCLUSION: CONVERGENCE ACROSSMETHODS

The aim of this chapter was to introduce the mainresearch methods used by neuroeconomists to estab-lish brain�behavior relationships. These differentmethods have distinct and often complementarystrengths and weaknesses. For instance, the measure-ment techniques covered above are very useful forestablishing the anatomical location or timing of theneural computations underlying behavior. Metabolicimaging techniques (such as fMRI or PET) allow accu-rate spatial localization throughout the brain, but havelow temporal resolution and rely on correlative linksbetween neurometabolism and neural activity.Electrical imaging techniques (such as EEG or MEG)have exquisite temporal resolution but only measuresignals from superficial cortical regions with low spa-tial certainty about their anatomical origin. Invasiverecording techniques (such as single-unit recording)measure neural processing more directly with spatialand temporal precision, but use of these methods isrestricted to studies of animals and a small group ofhuman patients prior to neurosurgery.

A common disadvantage of all measurement techni-ques is that they cannot conclusively determine thecausal role the identified neural computations mightplay in behavior. Neural manipulation techniques canbe used to address such questions. TMS (see the sec-tion Transcranial Magnetic Stimulation) changes brainfunction non-invasively by means of magnetic fieldsthat are focal in space and time, but this method hassome potentially distracting side effects. tDCS (see thesection Transcranial Direct Current Stimulation) is lessnoticeable to participants than TMS but also lessprecise with respect to both spatial and temporal reso-lution. Invasive stimulation techniques (such as micro-stimulation and optogenetics; see the section InvasiveStimulation Methods in Animals) have exquisite spatialand temporal resolution, and may even selectivelyaffect different cell types, but use of these methods isrestricted to animal studies. Finally, lesion methods(see the section Lesion Studies in Humans) can producestrikingly convincing findings on the necessity of abrain area for behavior. However, naturally occurringlesions in humans rarely affect only one specific ana-tomical structure and can be associated with sideeffects that complicate the interpretation of the results.Experimental lesions in animals can be more precisely

105CONCLUSION: CONVERGENCE ACROSS METHODS

NEUROECONOMICS

Page 30: Neuroeconomics || Experimental Methods in Cognitive Neuroscience

BOX 6.11

AN EXAMPLE OF AN EXPER IMENTAL LE S ION STUDY IN AN IMALS

This study (Rudebeck et al., 2006) used a lesion approach

in rats to demonstrate that the orbitofrontal cortex (OFC)

and the anterior cingulate cortex (ACC) show a double dis-

sociation in sensitivity to delay and effort costs in decision

making. Three groups of rats were trained on two tasks

that involved collecting food pellets in T-maze testing

environments. In the delay task (Panel A), the rat was

placed at the start of the T-maze and decided whether to

run into the left or right arm of the maze. In both cases, a

gate closed behind the rat and a second gate opened in

front of the rat to give it access to the food. Choosing the

left arm (low-reward arm, LRA) resulted in immediate

access to one food pellet (short delay1 low reward),

whereas choosing the right arm (high-reward arm, HRA)

resulted in the rat having to wait for 15 s to receive 10 pel-

lets of food (long delay1high reward). In the effort task

(Panel B), the left arm led to two pellets of food that were

easy to reach; in the right arm, the rat had to exert effort

and climb over a barrier to reach four pellets of food. Prior

to surgery, all rats had a strong preference for the options

yielding the highest rewards (A1�A3 and B1�B3 in Panel C

and Panel D), even though they required waiting on the

delay task or exerting effort in the effort task. After the

lesions (C1�C3 in Panel C and Panel D), the animals

with the OFC lesions became more “impatient” on the

delay task and chose the short-delay, low-reward option

more often than the other two groups, whereas their

behavior on the effort task did not differ from that of the

control group. Conversely, animals with ACC lesions

got “lazier” on the effort task and chose the low-effort,

low-reward option more often than the other groups,

but their behavior on the delay task resembled that of

the control group. Interestingly, none of these effects

reflected reward insensitivity: After surgery, all animals

chose the high-reward option when delay or effort was

matched (D1�D3 in Panel C and Panel D). This study

demonstrates how double dissociations can provide evi-

dence for fully separable contributions of different brain

areas to behavior. Panels A and B adapted from Rudebeck

et al. (2006) with permission from Macmillan Publishers

Ltd., Panels C and D adapted from Rushworth et al. (2007)

with permission from Elsevier.

1

1 �15 s

1 �30 cm 1 �30 cm 1 �30 cm 2 �30 cm

1 �15 s 1 �15 s 2 �15 s

2A

delay delay delay delay

B C D3 1 2 3 1 2 3 1 2 3

10(A) (C)

(B) (D)

8

6

4

OFCACC

Mea

n nu

mbe

r of

HR

Cho

ices

/10

Control

2

0

1 2A B C

barrierbarrierbarrierbarrier

3 1 2 3 1 2 3 1 2 3

10

8

6

4

Mea

n nu

mbe

r of

HR

cho

ices

/10

2

0

OFCACCControl

Gate A Gate B

1 pellet(LRA)

10 pellets(HRA)

10 pellets

4 pellets

4 pellets(HRA)

2 pellets(LRA)

Barrier

15 s delay

Start

Start

FIGURE BOX 6.11

106 6. EXPERIMENTAL METHODS IN COGNITIVE NEUROSCIENCE

NEUROECONOMICS

Page 31: Neuroeconomics || Experimental Methods in Cognitive Neuroscience

controlled but only allow the study of behaviors thatare displayed by the respective animal. Finally, alltechniques used for experimental neural manipulationsmay affect processing not only in the targeted brainarea but also in other interconnected regions (Driveret al., 2009; Lomber and Galuske, 2002). How suchpotential network effects relate to behavioral changesis unclear from neural manipulation studies alone.

The limitations of each research method curtail theexplanatory power of data obtained with only onetechnique. One way to address this problem is tosequentially combine several complementary researchtechniques in closely related experiments in order toprovide converging evidence for a given neural modelof decision making. For instance, studies may first spa-tially localize specific neural computations with fMRIand then test the specific role of activated brain areaswith brain stimulation techniques such as TMS ortDCS. Other studies may employ EEG or MEG toobtain a picture of the temporal dynamics of the neuralactivity that has been precisely localized in space inparallel fMRI studies. Yet other approaches may com-bine fMRI studies in humans with related lesion stud-ies in animals to investigate possibly homologouseffects in the monkey and the human brain. Such pro-cedures can harness the specific strengths of each ofthe methods while partially compensating for theirweaknesses, thus yielding a more complete model ofthe neural processes underlying behavior.

An even more ambitious attempt to overcome thelimitations of single neuroscience methods is the com-bination of several research methods within one set ofmeasurements. Such approaches are often referred toas multi-modal imaging, even though correlative neuro-imaging techniques may � strictly speaking � onlybe part of the methods employed. While multi-modalstudies are technically considerably more complicatedthan sequential applications of different researchtechniques in separate experiments, they offer moreexplanatory power and can highlight aspects of neu-ral processing that would be difficult to study other-wise. For instance, several experiments havecombined manipulation and measurement techniquesin one experiment to reveal how external influenceson activity in one brain area affect neural processingin other interconnected areas (Driver et al., 2009).Such network effects of causal activity manipulationsin one area can provide evidence for dynamic neuralcommunication between different brain areas thatmay underlie behavior. Examples of such studies arefMRI experiments in patients with brain lesions thatdocument changes in neural computations for non-lesioned interconnected brain areas relative to healthybrains. Related approaches in healthy participantshave applied TMS either directly before or during

fMRI scanning to detect where in the brain neuralactivity changes in response to activity disruptions inthe stimulated area. TMS can also be combined withEEG to investigate the temporal profiles of influencesfrom the stimulated area on other interconnectedregions. Such multimodal combinations of manipula-tion and measurement techniques offer a unique per-spective on potentially causal contributions of thestimulated/lesioned area in coordinating activitythroughout brain networks and how such directedcommunication in connected neural circuits mayunderlie the control of behavior.

We hope to have illustrated the main techniquesthat neuroeconomists can use to establishbrain�behavior relationships. None of these methodsby itself is perfect, but different techniques can be com-bined either sequentially or in parallel to provide con-verging evidence for a specific neural model ofbehavior. Such a close integration of research methodsmay ultimately prove essential for achieving the com-mon goal of all neuroeconomists, irrespective of theirmethodical background: the construction of detailedmechanistic models of how our brains allow us tomake decisions, learn from their outcomes, and inter-act with the world around us.

References

Airan, R.D., Thompson, K.R., Fenno, L.E., Bernstein, H., Deisseroth, K.,2009. Temporally precise in vivo control of intracellular signalling.Nature. 458, 1025�1029.

Clark, K.L., Armstrong, K.M., Moore, T., 2011. Probing neural cir-cuitry and function with electrical microstimulation. Proc. Biol.Sci. 278, 1121�1130.

Deisseroth, K., 2011. Optogenetics. Nat. Methods. 8, 26�29.Driver, J., Blankenburg, F., Bestmann, S., Vanduffel, W., Ruff, C.C.,

2009. Concurrent brain-stimulation and neuroimaging for studiesof cognition. Trends Cogn. Sci. 13, 319�327.

Fecteau, S., Knoch, D., Fregni, F., Sultani, N., Boggio, P., Pascual-Leone, A., 2007. Diminishing risk-taking behavior by modulatingactivity in the prefrontal cortex: a direct current stimulationstudy. J. Neurosci. 27, 12500�12505.

Fellows, L.K., Farah, M.J., 2003. Ventromedial frontal cortex mediatesaffective shifting in humans: evidence from a reversal learningparadigm. Brain. 126, 1830�1837.

Fritsch, G., Hitzig, E., 1870. Ueber die elektrische Erregbarkeit desGrosshirns [On the electrical excitability of the cerebrum]. Arch.Anat. Physiol. Wiss. Med. 37, 300�332.

Gehring, W.J., Willoughby, A.R., 2002. The medial frontal cortex andthe rapid processing of monetary gains and losses. Science. 295,2279�2282.

Hakyemez, H.S., Dagher, A., Smith, S.D., Zald, D.H., 2008. Striataldopamine transmission in healthy humans during a passive mon-etary reward task. Neuroimage. 39, 2058�2065.

Huang, Y.-Z., Edwards, M.J., Rounis, E., Bhatia, K.P., Rothwell, J.C.,2005. Theta burst stimulation of the human motor cortex. Neuron.45, 201�206.

Huettel, S.A., Song, A.W., McCarthy, G., 2004. Functional MagneticResonance Imaging. Sinauer Associates, Sunderland, MA.

107REFERENCES

NEUROECONOMICS

Page 32: Neuroeconomics || Experimental Methods in Cognitive Neuroscience

Huettel, S.A., Song, A.W., McCarthy, G., 2009. Functional MagneticResonance Imaging, second ed. Sinauer Associates, Sunderland,MA.

Huettel, S.A., Stowe, C.J., Gordon, E.M., Warner, B.T., Platt, M.L.,2006. Neural signatures of economic preferences for risk andambiguity. Neuron. 49, 765�775.

Hunt, L.T., Kolling, N., Soltani, A., Woolrich, M.W., Rushworth, M.F.,Behrens, T.E., 2012. Mechanisms underlying cortical activity dur-ing value-guided choice. Nat. Neurosci. 15, 470�476.

Husain, M., Rorden, C., 2003. Non-spatially lateralized mechanismsin hemispatial neglect. Nat. Rev. Neurosci. 4, 26�36.

Joseph, R., 2000. Neuropsychiatry, Neuropsychology, ClinicalNeuroscience. Academic Press, New York, NY.

Knoch, D., Pascual-Leone, A., Meyer, K., Treyer, V., Fehr, E., 2006.Diminishing reciprocal fairness by disrupting the right prefrontalcortex. Science. 314, 829�832.

Kolb, B., Whishaw, I.Q., 2009. Fundamentals of HumanNeuropsychology. Worth Publishers, New York, NY.

Lee, J.H., Durand, R., Gradinaru, V., Zhang, F., Goshen, I., Kim, D.S.,et al., 2010. Global and local fMRI signals driven by neuronsdefined optogenetically by type and wiring. Nature. 465, 788�792.

Lomber, F., Galuske, R.A.W., 2002. Virtual Lesions: ExaminingCortical Function with Reversible Deactivation. OxfordUniversity Press, Oxford.

Matsumoto, M., Hikosaka, O., 2009. Two types of dopamine neurondistinctly convey positive and negative motivational signals.Nature. 459, 837�841.

Nitsche, M.A., Cohen, L.G., Wassermann, E.M., Priori, A., Lang, N.,Antal, A., et al., 2008. Transcranial direct current stimulation:state of the art 2008. Brain Stimul. 1, 206�223.

Nitsche, M.A., Paulus, W., 2000. Excitability changes induced in thehuman motor cortex by weak transcranial direct current stimula-tion. J. Physiol. 527, 633�639.

Olds, J., Milner, P., 1954. Positive reinforcement produced by electri-cal stimulation of septal area and other regions of rat brain. J.Comp. Physiol. Psychol. 47, 419�427.

Penfield, W., Rasmussen, T., 1950. The Cerebral Cortex of Man: AClinical Study of Localization of Function. Macmillan, Oxford.

Poldrack, R.A., 2006. Can cognitive processes be inferred from neuro-imaging data? Trends Cogn. Sci. 10, 59�63.

Romo, R., Salinas, E., 2001. Touch and go: decision-making mechan-isms in somatosensation. Ann. Rev. Neurosci. 24, 107�137.

Rossi, S., Hallett, M., Rossini, P.M., Pascual-Leone, A., 2009. Safety ofTMS Consensus Group: safety, ethical considerations, and appli-cation guidelines for the use of transcranial magnetic stimulationin clinical practice and research. Clin. Neurophys. 120,2008�2039.

Rudebeck, P.H., Buckley, M.J., Walton, M.E., Rushworth, M.F.S.,2006. A role for the macaque anterior cingulate gyrus in socialvaluation. Science. 313, 1310�1312.

Ruohonen, J., 1998. Transcranial Magnetic Stimulation: Modellingand New Techniques, Doctoral dissertation. Helsinki Universityof Technology, Espoo, Finland.

Rushworth, M.F., Behrens, T.E., Rudebeck, P.H., Walton, M.E., 2007.Contrasting roles for cingulate and orbitofrontal cortex in deci-sions and social behaviour. Trends Cogn. Sci. 11, 168�176.

Sanfey, A.G., Loewenstein, G., McClure, S.M., Cohen, J.D., 2006.Neuroeconomics: cross-currents in research on decision-making.Trends Cogn. Sci. 10, 108�116.

Shallice, T., 1988. From Neuropsychology to Mental Structure.Cambridge University Press, New York, NY.

Tsai, H.C., Zhang, F., Adamantidis, A., Stuber, G.D., Bonci, A.,de Lecea, L., et al., 2009. Phasic firing in dopaminergic neurons issufficient for behavioral conditioning. Science. 324, 1080�1084.

Van Horn, J.D., Irimia, A., Torgerson, C.M., Chambers, M.C., Kikinis,R., Toga, A.W., 2012. Mapping connectivity damage in the case ofPhineas Gage. PLoS One. 7 (5), e37454.

Walsh, V., Pascual-Leone, A. (Eds.), 2003. Transcranial MagneticStimulation: A Neurochronometrics of Mind. MIT Press, Boston,MA.

Walton, M.E., Behrens, T.E., Buckley, M.J., Rudebeck, P.H.,Rushworth, M.F., 2010. Separable learning systems in themacaque brain and the role of orbitofrontal cortex in contingentlearning. Neuron. 65, 927�939.

Wassermann, E.M., Epstein, C.M., Ziemann, U., 2008. The OxfordHandbook of Transcranial Stimulation, first ed. OxfordUniversity Press, Oxford.

Yarkoni, T., Poldrack, R.A., Van Essen, D.C., Wager, T.D., 2010.Cognitive neuroscience 2.0: building a cumulative science ofhuman brain function. Trends Cogn. Sci. 14, 489�496.

108 6. EXPERIMENTAL METHODS IN COGNITIVE NEUROSCIENCE

NEUROECONOMICS


Top Related