Conceived and designed the experiments: JBE CMC CN GB. Performed the experiments: JBE. Analyzed the data: JBE CMC GB. Wrote the paper: JBE CMC CN GB.
The authors have declared that no competing interests exist.
Financial advice from experts is commonly sought during times of uncertainty. While the field of neuroeconomics has made considerable progress in understanding the neurobiological basis of risky decision-making, the neural mechanisms through which external information, such as advice, is integrated during decision-making are poorly understood. In the current experiment, we investigated the neurobiological basis of the influence of expert advice on financial decisions under risk.
While undergoing fMRI scanning, participants made a series of financial choices between a certain payment and a lottery. Choices were made in two conditions: 1) advice from a financial expert about which choice to make was displayed (MES condition); and 2) no advice was displayed (NOM condition). Behavioral results showed a significant effect of expert advice. Specifically, probability weighting functions changed in the direction of the expert's advice. This was paralleled by neural activation patterns. Brain activations showing significant correlations with valuation (parametric modulation by value of lottery/sure win) were obtained in the absence of the expert's advice (NOM) in intraparietal sulcus, posterior cingulate cortex, cuneus, precuneus, inferior frontal gyrus and middle temporal gyrus. Notably, no significant correlations with value were obtained in the presence of advice (MES). These findings were corroborated by region of interest analyses. Neural equivalents of probability weighting functions showed significant flattening in the MES compared to the NOM condition in regions associated with probability weighting, including anterior cingulate cortex, dorsolateral PFC, thalamus, medial occipital gyrus and anterior insula. Finally, during the MES condition, significant activations in temporoparietal junction and medial PFC were obtained.
These results support the hypothesis that one effect of expert advice is to “offload” the calculation of value of decision options from the individual's brain.
Seeking advice from experts is common practice. The most prominent situations in which people turn to experts for advice occur under conditions of enhanced uncertainty, such as an economic recession. During such times, people may feel unfit to predict the consequences of their choices, and may seek the counsel of experts to reduce the enhanced perception of risk. For instance, when making investment decisions in a market downturn, people often ask an expert, or a knowledgeable colleague or friend for advice on where to invest their money. While the field of neuroeconomics has made significant progress in understanding the neurobiological basis of risk in decision-making (for reviews see
On each trial, participants were asked to choose between a sure win and a lottery, either in the presence of advice from an expert (MESSAGE) or in its absence (NO MESSAGE). Advice from the expert economist was provided on half the trials by way of placing the words “ACCEPT” above the option that the expert would choose and “REJECT” above the option that the expert would not choose. In the NO MESSAGE condition, the expert's advice was hidden by placing the words “UNAVAILABLE” above both options. The probability of the lottery varied across seven probability conditions ranging from 1% to 99% and the amount of the sure win varied based on decision weights estimated in a behavioral pre-scanning session using the PEST procedure. The self-paced decision period was followed by a 1-second feedback period, which provided confirmatory information about which option was chosen by the participant. Finally, a jittered intertrial interval that varied between 3 and 10 seconds was presented.
There are several possible mechanisms through which expert advice could affect an individual's decision-making process
To investigate the behavioral effect of expert advice (question 1), we compared decisions made during trials in which advice was received (which we refer to as the MES condition) and not received (the NOM condition). We used prospect theory
The second question was addressed by comparing brain activation patterns observed between the two conditions, MES (advice present) and NOM (advice absent). While information about the expert's professional background and his decision strategy were provided prior to the experiment in order to create a basic level of trust towards the expert, the quality of such information is not comparable to actual interactions with the expert. The specific nature of the expert's advice, which was suboptimal, could only be inferred via repeated interactions throughout the experiment. This process requires mental perspective-taking to infer the intentions and beliefs of the expert. Previous research in the field of social cognitive neuroscience has repeatedly associated such mental prespective-taking with activations in temporoparietal junction (TPJ) and medial prefrontal cortex (MPFC) (e.g.
The third question was addressed by investigating activations when participants followed, compared to when they ignored, advice provided by the expert economist. Ignoring the advice of an expert may lead to enhanced conflict during the decision process and increase the perceived risk associated with choice. On the other hand, following the expert's advice can be considered a much safer option involving less emotional and cognitive conflict. We therefore expected non-conformity with the expert's advice to lead to enhanced levels of conflict and arousal. To investigate the neurobiological basis of this effect, we probed for differential activation patterns as a function of whether participants followed vs. ignored advice. Based on previous research
Finally, to investigate the neurobiological basis of the influence of expert advice during risky decision-making, we focused on activations in regions showing correlations with valuation and probability computations. In agreement with the
Our behavioral results indicated that the expert's advice significantly influenced behavior. To investigate the extent to which expert advice affected individuals' estimated probability weighting parameters, we employed nonlinear logistic regression in combination with Prelec's compound invariant form
This figure shows probability weighting functions,
Followed | Ignored | ||||
Percentage | RT | Percentage | RT | ||
MES | Mean | 72.81 | 3.169 | 27.19 | 4.561 |
Std Error | 4.07 | 0.244 | 3.45 | 0.883 | |
NOM | Mean | 64.10 | 3.231 | 35.90 | 4.092 |
Std Error | 4.07 | 0.226 | 3.45 | 0.468 |
To isolate areas activated during the message condition, we investigated the main effect of the message, contrasting presence and absence of the expert's advice (MES-NOM). As shown in
Brain regions showing significant activations during the MESSAGE (
L/R | Structure | BA | Volume | RL | AP | IS | Max t |
R | Superior Frontal Gyrus / DMPFC | 8 | 101 | 3.5 | 38.4 | 51.2 | 6.654 |
L | SupramarginalGyrus / TPJ | 40 | 37 | −51.7 | −57 | 31.2 | 4.564 |
R | SupramarginalGyrus / TPJ | 40 | 31 | 50.9 | −50.2 | 31.6 | 5.43 |
L | Insula | 47 | 29 | −29.4 | 18.2 | −11.9 | 6.894 |
R | Insula | 47 | 24 | 35.7 | 25 | −11.3 | 4.8 |
R | Inferior Frontal Gyrus | 47 | 8 | 55.5 | 25.5 | 7.5 | 4.028 |
R | Caudate | 6 | 7.9 | 9.5 | 11.4 | 4.456 | |
R | Caudate | 5 | 9 | 8.4 | 1.2 | 4.552 | |
L | Inferior Frontal Gyrus | 47 | 5 | −51.6 | 25.8 | 9 | 4.037 |
R | Lingual Gyrus | 18 | 263 | 1.2 | −81.6 | −3.2 | −10.558 |
L | Cuneus | 18 | 9 | −9 | −66 | 4 | −4.266 |
R | Mid Cingulate Cortex | 32 | 9 | 10.6 | 6.7 | 36.3 | −4.952 |
L | FusiformGyrus | 19 | 7 | −19.7 | −69.9 | −9.4 | −4.428 |
R | Posterior Cingulate Cortex | 30 | 7 | 12.4 | −50.9 | 10.3 | −4.31 |
R | Middle Temporal Gyrus | 22 | 6 | 58.5 | −35.5 | −3.5 | 4.421 |
L | Posterior Insula | 22 | 5 | −41.4 | −13.8 | −1.8 | −4.402 |
To probe for the effects of conformity, we contrasted blood oxygenation-level-dependent (BOLD) responses during trials in which participants chose to ignore versus follow the expert's advice in the message condition only. As shown in
A. Brain regions responding when participants ignored the expert's advice included left anterior insula and right globus pallidus. B. Brain regions showing significant activation when participants conformed to the expert's advice included posterior insula, frontal eye fields and posterior cingulate, as well as two activation clusters in superior frontal gyrus (not shown here). C. The time course in the anterior insula shows a significant increase in activation when participants decided to ignore the expert's advice (
L/R | Structure | BA | Volume | RL | AP | IS | Max t |
R | Posterior Insula | 40 | 15 | 49.8 | −31.5 | 28.4 | −4.889 |
R | PrecentralGyrus / FEF | 4 | 18 | 49.2 | −7.4 | 48.7 | −4.9174 |
R | Posterior Insula | 5 | 15 | 17.6 | −35.1 | 50.3 | −5.3276 |
R | Superior Frontal Gyrus | 6 | 5 | 8.4 | −7.8 | 60 | −4.3113 |
R | Superior Frontal Gyrus | 6 | 6 | 14.5 | −2 | 69.5 | −4.0575 |
L | Anterior Insula | 47 | 21 | −30.3 | 20.5 | −9 | 4.9924 |
R | Globus Pallidus | 6 | 15.5 | 2 | −5.1 | 4.9392 |
We probed for brain regions showing correlations with the weighted value of the lotteries (1000*
Brain regions sensitive to weighted value of either the lottery (
L/R | Structure | BA | Volume | RL | AP | IS | Max t | NPRR99 NOM>MES |
L | Anterior Insula | 47 | 15 | −31.4 | 20.8 | −2.1 | 5.058 | No |
L | Medial Frontal Gyrus | 8 | 14 | −6.2 | 27.7 | 42.9 | 4.7863 | Yes |
L | Thalamus | 9 | −0.7 | −13 | 0.6 | 4.3643 | Yes | |
L | Anterior Cingulate Cortex | 9 | 7 | −37.7 | 23.6 | 41.2 | 4.6137 | Yes |
R | Middle Occipital Gyrus | 18 | 6 | 28.1 | −89.5 | −3.5 | 4.7843 | Yes |
R | IntraparietalSulcus | 7 | 44 | 31.3 | −48.5 | 44.2 | 5.8676 | Yes |
L | Cuneus | 18 | 35 | −23.9 | −95.4 | −8.3 | 5.2682 | Yes |
R | Middle Occipital Gyrus | 18 | 31 | 32 | −92 | −1.1 | 4.9522 | Yes |
R | Precuneus | 19 | 26 | 29.7 | −64.8 | 40.5 | 4.4483 | Yes |
L | Cuneus | 30 | 21 | −11.3 | −73 | 6.4 | 4.4809 | Yes |
R | Posterior Cingulate Cortex | 30 | 13 | 19.6 | −68.8 | 4.4 | 4.3002 | Yes |
R | Middle Temporal Gyrus | 37 | 5 | 59.5 | −49.3 | −12 | 4.998 | Yes |
L | Middle Occipital Gyrus | 19 | 5 | −27 | −60.6 | 1.1 | 4.9919 | Yes |
R | Posterior Cingulate Cortex | 30 | 5 | 28.8 | −65.4 | 7.2 | 4.2667 | Yes |
R | Inferior Frontal Gyrus | 9 | 5 | 40.8 | 7.2 | 30 | 3.9422 | No |
L | Caudate | 4 | −9.0 | 7.5 | −6.0 | 4.2107 | No | |
L | Cuneus | 18 | 7 | −27 | −20.1 | 52.7 | −4.204 | Yes |
L | Middle Frontal Gyrus | 4 | 6 | −58.4 | −35.1 | −5.5 | −4.5856 | Yes |
To consider whether the profile of the relationship between brain activation and probability differed between MES and NOM conditions in areas showing significant correlations with value, we obtained separate neurobiological probability response ratios (NPRRs) for the MES and NOM conditions. Because we did not observe any significant correlations with w(p) in the MES condition, we expected to find relatively flat NPRRs in the MES condition compared to the NOM condition. This would be reflected by significant differences between NPRRs in the two treatments, particularly for high probabilities, during which the expert advice was maximally different from risk neutrality or expected value maximization. Our findings are consistent with this hypothesis. Significantly greater responses in the NOM compared to the MES condition were obtained in the majority of areas showing correlations with w(p), or equivalently, with weighted value. This indicates extensive recruitment of valuation mechanisms in the absence of the expert's advice, and attenuation (offloading) when the expert's advice is present.
The mean ratio for each subject was computed at each probability relative to the NOM baseline, and the median across subjects is plotted with error bars indicating the 95% confidence interval for the median. The NPRR curves are plotted with reference to the diagonal, which indicates linear probability weighting. Significant differences between NOM and MES function, denoted with “*”, were obtained consistently in the 99% probability condition (see
A simple financial decision-making task involving risk was employed in the current study to investigate the behavioral and neural mechanisms by which financial advice, provided by an expert economist, affected decisions under risk. Behavioral results showed a significant effect of expert advice on probability weighting, such that probability weighting functions changed in the direction of the expert's advice. The behavioral effect of expert messages was paralleled by changes in neural activation patterns. Of note, significant correlations with the value of the lottery were obtained in the absence of the expert's advice (NOM), but not during its presence (MES). These results support the hypothesis that one effect of expert advice is to “offload” calculations of the value of alternative behavioral options that underlie decision-making from the individual's brain.
The significant behavioral effect of expert advice on probability weighting was paralleled by the fMRI results. At the behavioral level, we obtained a significant effect of the expert's advice on the curvature of the probability weighting function, indicated by a significant effect of the presence of the advice on α. Neurally, this was paralleled in two ways: 1) an attenuated activation of regions whose BOLD responses showed significant correlations with value in the NOM condition; and 2) significantly flattened NPRRs, reflecting an attenuated relationship between activation level and probabilities in regions that showed correlations with probability, in the MES condition relative to the NOM condition. Specifically, our results revealed that, in the absence of the expert's advice, participants engaged two largely separate networks involved in valuation mechanisms. These networks were composed of regions exhibiting correlations with two types of value, namely (a) payoff of the sure win and (b) weighted value of the lottery. These correlations were attenuated when the expert's advice was available.
These results implicate particular networks in evaluating the different behavioral options and underline their importance in the financial decision-making process. Regions that showed sensitivity to payoff magnitudes of prospects in the current study have previously been associated with decision-making under uncertainty, such as the parietal cortex, including precuneus
We also confirm the presence of nonlinear weighting of probability, both at the behavioral, as well as the neural level in a network of areas sensitive to probability. We show that the degree of probability weighting can be influenced by the presence of advice from a financial expert, both at the behavioral (
In previous work, we demonstrated the presence of nonlinear probability weighting functions in a network of areas when probabilistic information about an impending electrical shock was provided
Recruitment of valuation processes reflective of reward magnitude and probability weighting was greatly attenuated in the MES condition. Together with our behavioral results, these findings indicate that the presence of the expert's advice significantly altered the decision-making process. During the presence of the expert's advice, a network of brain regions was active that included areas associated with mentalizing others' intentions, such as TPJ and DMPFC (e.g.
Taken together, the activation pattern obtained in the presence of the expert's advice indicates an attenuated recruitment of valuation mechanisms that was accompanied by significant activations in regions associated with TOM reasoning. The TPJ, especially in the right hemisphere, has previously been associated with judgments of true and false beliefs that other people may hold
Activations in TOM areas during the presence of the expert's advice were accompanied by activations in regions associated with valuation in the context of decision-making tasks, including anterior insula and caudate nucleus. Previous research has implicated the caudate in feedback processes that guide future actions
To investigate the involvement of areas in conformity with, or independence of the expert, we probed for brain regions showing differential responses when following vs. ignoring the expert's advice. Regions showing increased responses when ignoring the expert included the anterior insula and globus pallidus, implicating these regions in nonconforming decisions that override the expert's advice. Interestingly, responses of a subset of globus pallidus neurons have recently been associated with negative reward-related signals
Little is known about the neurobiological basis of conformity. In a previous study, participants made binary perceptual decisions about rotated 3D objects in the presence of answers provided by either peers or a computer
These findings are corroborated by results from the current study demonstrating increased responses in the anterior insula when participants ignored the expert. The anterior insula has repeatedly been implicated in risky decision-making and is thought to encode the negatively valenced affective aspects involved in risk
In summary, our results demonstrate that financial advice from an expert economist, provided during decision-making under conditions of uncertainty, had a significant impact on both behavior and brain responses. Behavioral results showed a significant effect of expert advice, such that probability weighting functions changed in the direction of the expert's advice. The behavioral effect of expert messages was paralleled by neural activation patterns. Specifically, (1) significant correlations with the value of choice alternatives were obtained only in the absence of the expert's advice, but not during its presence. This indicates an attenuation in the engagement of valuation processes in the presence of expert advice; (2) during the message condition, areas associated with mentalizing, such as DMPFC and bilateral TPJ were recruited, and finally, (3) ROI analyses of regions associated with probability indicated a significant “flattening” of neurobiological probability response ratios (NPRR) in the message condition compared to the no-message condition. This lends further support to the hypothesis that in the presence of advice from an expert, recruitment of valuation mechanisms was attenuated. Taken together, these results provide significant support for the hypothesis that one effect of expert advice is to “offload” the calculation of expected utility from the individual's brain.
24 healthy, right-handed participants (15 females) participated in the current study, which was approved by the Emory University Institutional Review Board. The average age was 23 with a standard deviation of 5.3 years. The majority of participants were undergraduate students (17), 6 participants had graduate-level education, 1 subject chose not to provide educational information. All participants gave written informed consent and reported good health with no history of psychiatric disorders.
Each session of the experiment consisted of two phases. The first phase, was conducted outside the scanner and subjects made choices between a sure win and lotteries providing ex-ante probabilities of winning a comparatively higher payoff, as shown in
In order to control for wealth effects, participants received a chance to win cash-rewards at the end of each session by randomly selecting one of the trials via a throw of three 10-sided dice. The decision made on the selected trial determined payment as follows: if the sure win was chosen on the selected trial, the respective amount was paid to the subject; if the lottery was chosen, a “computerized coin was tossed” giving subjects a chance to win 1000 laboratory Yen at the probability indicated in the lottery. An exchange rate of 1000 laboratory Yen = 16 USD was established at the beginning of the experiment, and subjects were informed of this rate. We used 1000 YEN as the lottery amount to facilitate the computation of the expected value of the lottery (should the subject wish to make such a calculation).
We optimized order and timing of our experimental design for adequate estimation of message- and probability-related responses
As shown in
It has previously been argued that subjects can exhibit inconsistent preference behaviors, show strong framing effects, and distort reward magnitude or reward probability
There are both advantages and limitations of using the PEST procedure. One advantage is that we can generate CEs without having to opt for an auction such as the Becker-deGroot-Marschak (BDM) procedure, which would be difficult to implement in the scanner. On the other hand, one disadvantage is that it is possible, at least in principle, for the procedure to be manipulated by highly sophisticated subjects. However, we did not observe any evidence indicating that participants employed strategies to manipulate the staircase procedure in order to increase their chances of winning more money. Such a strategy would have led to a substantial overweighting of small probabilities relative to typical levels, which we did not find. Furthermore, during debriefing at the end of the study, participants indicated that they had not identified any way to engage in strategic behavior.
Inside the scanner, subjects made choices between lotteries and sure wins in a similar fashion as in the behavioral pre-scanning session, except that an expert economist provided his suggestions during half of the trials. In order to make the economist trustworthy, participants were informed of the economist's credentials and achievements, as well as his preferred decision strategy, in detailed instructions, which read as follows: “An expert Economist (Professor Charles Noussair of the Department of Economics at Emory University) is going to tell you his preferred decisions on half the trials. Professor Charles Noussair, Ph.D., earned Bachelor's degrees in Economics and Psychology from the University of Pennsylvania, and Master's and Doctorate Degrees from the California Institute of Technology. He has taught at Purdue and Emory Universities, and been a visiting professor in Australia, Japan, France, and the Netherlands. He has consulted for NASA, the Federal Reserve and the French ministry of Agriculture and has published numerous articles in high-impact peer-reviewed scientific journals.” None of the participants had been acquainted with Charles Noussair previously.
The expert's suggestions followed approximately a satisficing rule, which were, in part, consistent with those of a decision-maker trying to maximize his probability of winning at least 200YEN. Specifically, in trials in which the sure win was below 200YEN, the expert's advice was the option that maximized expected value. In trials in which the sure win was greater or equal to 200YEN, the expert advised acceptance of the sure win. Suggestions were displayed at the top of the screen via placing the word “ACCEPT” above the recommended option, and “REJECT” above the option not recommended (see
During phase 2, the sure win magnitudes were based on CEs estimated during phase 1, and differed for each subject. Specifically, sure win magnitudes were selected randomly from the interval [CE−(0.4 * CE), CE+(0.4 * (1000-CE))], except for eight of the subjects, for which sure win magnitudes were provided by CE±0.2 * CE and CE±0.4 * CE (the discretized range was changed to a random sampling of the interval to provide better coverage of the offer space).
To allow for potential changes in CE which might occur if participants followed the expert advice during the scanning session, an attenuated version of the staircase algorithm was employed during phase 2 as well. This was done to ensure an adequate number of offers both above and below the subject's CE, even if the CE changed during the course of the experiment. This staircase algorithm tracked extreme behaviors, such that CEs in a given probability condition were adjusted when subjects deviated from expected behavior. Specifically, CEs were decreased by ¼ of the difference between CE and sure win magnitude in a given probability condition when participants accepted a sure win that was lower than CE, while CEs were increased by ¼ of the difference between CE and sure win magnitude, when participants chose to reject a sure win that was larger than the CE of the current probability condition. This adaptive algorithm tracked changes in probability weighting during the scanning session, which (1) ensured that sure win magnitudes were based on current decision-making parameters by accounting for potential changes in probability weighting that might occur as a function of context change (such as scanner environment, presence of expert, experienced outcome after completion of PEST procedure) and (2) provided a method for estimating the effects of expert messages on probability weighting that is independent of our behavioral decision making model outlined below.
Empirical evidence suggests that decisions under risk are typically consistent with the transformation of objective probability,
We employed nonlinear logistic regression to estimate each participant's probability weighting and utility functions from their binary decisions (lottery or sure-win) using a modified version of Prelec's compound invariant form
Neuroimaging data were collected using a 3 Tesla Siemens Magnetron Trio whole body scanner (Siemens Medical Systems, Erlangen, Germany). A three dimensional, high-resolution anatomical data set was acquired using Siemens' magnetization prepared rapid acquisition gradient echo (MPRAGE) sequence (TR of 2300 ms, TE of 3.93 ms, TI of 1100 ms, 1 mm isotropic voxels and a 256 mm FOV). Functional data consisted of thirty-five axial slices that were sampled with a thickness of 3 mm and encompassing a field of view of 192 mm with an inplane resolution of 64×64 (T2* weighted, TR = 2500 ms, TE = 31 ms). The task was presented with Presentation software (Neurobehavioral Systems, Albany, CA) and visual stimuli were projected onto a frosted glass screen, which the subject viewed through an angled mirror mounted to the head coil. Inhomogeneities in the magnetic field introduced by the participant were minimized with a standard two-dimensional head shimming protocol before each run and the anatomical data acquisition. In our dataset, each participant completed 4 runs with 56 trials, whose length depended on participants' decision time.
Initial preprocessing of the data was conducted using Analysis of Functional Neuroimages (AFNI,
FMRI data were analyzed using the General Linear Model and a standard two-stage mixed effects analysis. Trials were classified according to type of decision made by participants. Specifically, responses were sorted according to whether participants
To localize regions involved in processing the presence of the expert (MES main effect), a random effects model was implemented at the second level as separate paired two-sample t-tests contrasting the beta images corresponding to 1) presence and the absence of the MES and 2) ignoring the expert and following the expert in the MES condition only. Finally, to probe for brain regions associated with valuation processes (probability weighting and magnitude of sure win), correlation contrasts were performed with separate one-sample t-tests on parametric modulators in MES and NOM conditions. All t-maps were thresholded at an uncorrected p-value of 0.001 with a cluster size threshold of k>5, except for striatal areas for which a cluster size threshold of k>4 was employed.
To extract the temporal dynamics within regions showing a significant response when subjects followed or ignored the expert's advice in the MES condition, a different first-level model was fitted to each participant's fMRI data. In this model, the hemodynamic response during MES and CHOICE was modeled with a basis set of seven cubic spline functions spaced one TR (2.5 s) apart and spanning the interval from 0 to 15 seconds post trial onset. In order to create reconstructed event-related responses on a 1 s temporal grid, the set of fitted spline functions was resampled at a temporal resolution of 1 second and averaged within each ROI as a function of the following condition of interest: MESfollowed, MESignored and NOM.
The above whole-brain analysis probed for regions encoding non-linear probability weighting and sure win magnitude. To illustrate fMRI responses within structures associated with valuation, and, importantly, differences in probability weighting as a function of the expert's advice, we analyzed the ROI activations using a previously developed method of transforming neural activations to a neural analog of the probability weighting function. Specifically, we were interested in how the NPRR, the relationship between neural activation and probability of a given outcome, was affected by the presence of the expert's advice. This model was only used in areas already identified by the aforementioned contrasts. We converted mean activations reflecting signal change to neurobiological probability response ratios (NPRR) following methods described in detail in Berns et al. (2008). According to prospect tgheory, the probability weighting function,