Working Paper Series 2017/04/MKT

A Working Paper is the author’s intellectual property. It is intended as a means to promote research to interested readers. Its content should not be copied or hosted on any server without written permission from publications.fb@insead.edu Find more INSEAD papers at https://www.insead.edu/faculty-research/research

Red Bull Gives You Incentive Motivation: Understanding Placebo Effects of Energy Drinks on Human Cognitive

Performance

Liane Schmidt INSEAD, liane.schmidt@insead.edu

Corresponding author

Pierre Chandon INSEAD, pierre.chandon@insead.edu

Mathias Pessiglione

INSERM, mathias.pessiglione@gmail.com

Hilke Plassmann INSEAD, hilke.plassmann@insead.edu

The consumption of cognitive enhancers like energy drinks (EnD) is on the rise, but do they really improve cognitive performance, and, if yes, why? We examined two novel psychological mechanisms. First, we dissociated the role of expectations and actual consumption by crossing what people consumed—Red Bull Silver Edition or a similar-tasting Sprite soda—and what they thought they consumed. We found that participants performed better in a numerical Stroop task when they believed that they had consumed an EnD, irrespective of what they had actually consume, Second, we investigated the role of motivation for such a placebo effect of EnD. We found that expected, but not actual, consumption of EnD increased the effects of incentives on cognitive performance. Our results suggest that believing that one consumes an EnD increased participants’ motivation to perform and thus enhanced their performance. Keywords: Placebo Effects; Incentive Motivation; Mental Effort Significance statement Academic doping has become increasingly popular on campuses around the globe over the last years. However, it remains unknown if and how commercially available cognitive enhancers suchas energy drinks (EnD) impact cognitive performance. We varied actual and believed consumption of an EnD and measured cognitive performance and its allocation according to the magnitude of points to earn. We found that the belief that one has consumed an EnD but not its actual consumption, increased cognitive performance specifically in high point trials compared to low point trials. These findings provide novel insights into motivational processes underpinning marketing-led expectancy effects, and contribute to the current public policy debate on the use and abuse of increased alertness and concentration claims for EnD.

3

Introduction

Academic doping—the consumption of stimulants, energy drinks, and other commercial products

that are advertised as cognitive enhancers—has become very popular in recent years (1-4).

However, research investigating the impact of prescription stimulants (4) and energy drinks (5, 6)

on cognitive performance and alertness, and their psychophysiological mechanisms, has been

mostly inconclusive.

The goal of this paper is to improve our understanding of whether—and if yes, how—popular

commercial cognitive enhancers (i.e., energy drinks, or EnD) affect cognitive performance.

Previous studies have shown that the sugar in unlabeled EnD can improve some measures of

cognitive and physical performance (5, 6). However, because the drinks in these studies were

administered without a label, these studies cannot disentangle to what extent mere expectations

about the effects of consuming a drink labeled as EnD influence cognitive performance. This is

important given the distinct taste and color of the EnD that was used in all of these studies and

that thus might have revealed the experimental condition to the participants.

Interestingly, researchers have speculated that some of the effects of EnD on cognitive

performance may indeed be driven by marketing-led expectancies about their effects, rather than

by the effects of their actual ingredients (7, 8). Advertising for Red Bull®, for example, claims

that it “improves performance and concentration, so you can tackle everything from physics to

workouts.” Against this background, our first hypothesis was that the label of the drink (i.e.,

whether people think they consume and EnD vs. not) would increase cognitive performance.

Second, if this holds, we aimed at investigating how such positive expectancies, associated to an

EnD label, would increase cognitive performance. Based on the vast interdisciplinary literature

on expectancy-based placebo effects across domains (9), we focused on the role of motivation.

4

Evidence from functional magnetic resonance imaging (10, 11), patient (12-15) and animal (16)

studies showed that across domains positive expectancies linked to the administration of a sham

treatment can trigger activity in dopaminergic pathways and their targets — the ventral striatum

and the ventromedial prefrontal (18). These brain systems have been linked to incentive

motivation (19) — a term from behavioral neuroscience that designates the ability to allocate

effort according to the magnitude of incentives (20). However, the role of motivation for placebo

effects has never been directly tested in past studies. Building on this convergence, we

hypothesized that positive expectancies linked to thinking that one consumes an EnD (i.e., linked

to an EnD label) should enhance incentive motivation and through it better cognitive performance

when incentives to perform are high versus when they are low.

To test these hypotheses we dissociated the role of expectancies linked to drinking an EnD from

those of its ingredients by using a balanced between-participants placebo design, which has not

been done previously. In other words, we varied not only the actual drink consumed (EnD vs.

placebo, a soda of similar flavor and color) but also what participants thought they consumed

(EnD label vs. soda label). Participants performed a numerical Stroop task, that offered different

levels of incentives to perform well, and has previously been shown to depend upon the

activation of the ventral striatum (17). In addition, after training on the task and on the first trial

they indicated how well they expected to perform on the task. This measure of anticipated

performance allowed us to infer how participant’s expectancies varied as a function of label and

drink. Taken together our design and behavioral task assessed the relative effects of label and

drink on (1) cognitive performance (i.e. effort), (2) effort allocation according to the magnitude

of incentives (i.e. incentive motivation), and (3) performance expectancy (i.e. anticipated

performance).

5

Results

Effects of actual and believed consumption on incentive motivation.

In line with our hypothesis, we found a significant main effect of EnD label on cognitive

performance (ß = 0.21, SE = 0.1, p = .03, 95% CI [.01 – .41]) that stand in contrast to a null effect

of drink (ß = -0.002, SE = 0.1, p = .98, 95% CI [-.2 – .19]). Importantly, the main effect of EnD

label was entirely driven by an interaction of EnD label with incentives (ß = 0.05, SE = 0.02, p =

.002, 95% CI [0.02 – 0.09]; see Table 1): As shown in Figure 1a, participants who consumed a

drink labeled as an EnD showed an enhanced sensitivity to incentives. Regardless of what they

had actually consumed, participants assigned to the EnD-label groups allocated their cognitive

performance according to the magnitude of incentives. This enhanced incentive motivation

implied that labeling the drink as an EnD facilitated cognitive performances only in trials with

high incentives at stake.

In addition, we applied Bayesian statistics to test the robustness of the effect of label on incentive

motivation versus the null effect of drink consumed (21, 22). As shown in Figure 1b, the

Bayesian estimation of the posterior distribution of credible values for the difference in mean

incentive motivation between the EnD and soda label group revealed an exclusion of zero

difference (mode=0.11, 95% HDI [0.047 – 0.175]). The posterior distribution of credible values

for effect sizes (mode=0.735, 95% (HDI): [0.314 – 1.218]) also rejected zero (i.e. the region of

practical equivalence (ROPE) counted 0% of credible mean differences). In contrast, the

posterior distributions of possible differences in mean incentive motivation between the EnD and

soda drink group accepted the null hypothesis of zero difference (mode=-0.001, 95% HDI: [-

0.073 – 0.064]) with 34% of possible effect sizes falling within the ROPE (mode=-0.015, 95%

HDI: [-0.423 – 0.435], ROPE=34%). Taken together, our results provide convergent evidence for

6

a credible non-zero positive effect of label on motivation to perform contrasted by a possible null

effect of drink.

Effects of actual consumption and expected consumption on anticipated performance

In order to investigate whether our placebo manipulation indeed induced positive expectancies

about the efficacy of the consumed drink to enhance performance, we used the very first

anticipated performance rating as a measure of expectancy. A linear mixed model found a main

effect of label on anticipated performance that trended toward significance (ß = 0.42, p = .06 95%

CI [-0.02 – 0.86]) (SI Table 3). No fixed effect of drink and no significant interaction label by

drink were observed. Bayesian estimation of the posterior distribution of possible mean

differences in anticipated performance between the EnD and soda label group revealed that the

95% HDI included zero as a possible value (mode=0.079, 95 % HDI [-0.009 – 0.176]) with 6%

of possible effects sizes falling within the ROPE (mode=0.372, 95% HDI [-0.042 – 0.818]). The

posterior distributions for possible values for mean differences between the EnD and Soda drink

group also included zero as a possible value (mode=0.028, 95% HDI [-0.062 – 0.127]). Yet, the

distribution of possible non-zero effect sizes was narrower with 29% of possible values falling

within the ROPE (mode=0.121, 95% HDI [-0.299 – 0.56]).

Further exploration of the data revealed that the trending difference in anticipated performance

between the EnD and Soda label group was driven by a significant difference between

participants, who consumed soda in a EnD can (i.e., the placebo group) and participants, who

consumed soda in a soda can (i.e., the control group) (placebo > control group: M = 71 ± 0.4%

vs. M = 57 ± 0.5%, t(42) = 2.03, p = .04 two-tailed, Cohen’s d = 2.8). Bayesian estimation of

posterior distributions of the mean difference values between the placebo and control group

revealed zero still as a possible value (mode=0.125, 95% HDI: [-0.007 – 0.272]) with a larger

7

range of possible non-zero effect sizes involving only 5% of possible values within the ROPE

(mode=0.59, 95% HDI: [-0.042 – 1.236]).

Discussion

We investigated whether and why both actual and believed consumption of energy drinks

improve cognitive performance. Our findings provide first behavioral evidence that motivational

processes play a key role in placebo effects of EnD labels on performance.

Interestingly, our credible null effect of the actual drink consumed seems to be contrary to

previous findings that investigated the role of EnD’s content on mental and physical performance

(23, 24). Our study differs from previous work in that both the EnD and the soda contained

similar amounts of sugar (Sprite: 7 g per 100 ml; Red Bull: 11 g). Our setting therefore matches

what happens when people choose an energy drink rather than a soda in the hope that it will

enhance their cognitive performance.

In comparison to prior work on incentive motivation for motor effort (17, 20), we observed no

effects of incentives on performance in participants assigned to the soda-label groups. We cannot

disentangle whether these null effects were driven by the enhanced motivation in the EnD-label

groups, a decrease in motivation in the control groups, or both. Future studies measuring baseline

incentive motivation prior to group assignments are needed to address this issue.

More broadly, our findings extend the previous literature on expectancy-based placebo effects in

important ways. Previous studies could link dopamine functioning and reward-related behavior to

placebo effects, suggesting that the activation of dopamine pathways and supposedly linked

motivational processes might be the common mechanism that implements placebo effects across

domains (18). Here, we provide behavioral evidence for the importance of such motivational

processes for placebo effects on cognitive performance and predict that motivation might play a

8

similarly important role for expectancy effects in other domains. We call for future

neuroscientific work to causally establish the link between expectancy, motivation, and activation

of dopaminergic pathways as a common mechanism underlying expectancy-based placebo

effects. In turn, if motivation indeed plays an important role more generally for expectancy-based

placebo effects, this could be an important insight for clinical placebo studies and, for example,

shed light on new ways to improve patient well-being through motivational placebo interventions

instead of mere pharmacological ones.

Material and Methods

Participants and experimental procedure

To test our hypotheses, we applied a 2 (drink: Red Bull Silver Edition Lime® EnD or Sprite®

soda) x 2 (label: “Red Bull Silver Edition Lime” or “Sprite”) between-participants design:

Participants were randomly assigned to one of four experimental groups (see Figure 1a): a

placebo group (i.e., they drank soda labeled as EnD), an open EnD group (i.e., they drank an EnD

labeled as EnD), a hidden EnD group (i.e., they drank an EnD labeled as soda), and a control

group (i.e., they drank soda labeled as soda). This design allowed us to systematically disentangle

the effects of expectations about consuming an EnD from the effects of actually consuming it.

The placebo and open EnD groups were told that they were part of a test group for investigating

the effects of an EnD on various cognitive performance tasks. The hidden EnD and control group

were told that they were included in a control group for comparison to a group that had received

an EnD.

The study was approved by INSEAD’s institutional review board. All participants gave written

and informed consent. Eighty-eight French college students (37 males, mean age 23.6 ± 2.6

years, 22 per group) were recruited via public advertisement. They were screened for normal to

9

corrected-to-normal vision, no history of substance abuse, any neurological or psychiatric

disorder, and any medication..During the recruitment, participants were told that the study aimed

to test the effects of EnDs on cognitive performance.

All participants watched the experimenter open either a can of Red Bull Silver Edition Lime (the

placebo and open EnD groups) or a can of Sprite soda (the control and hidden EnD groups). The

drinks were chosen based on their similarity in taste, odor, and color, verified by a blind pretest.

Importantly, the administration of the placebo and hidden EnD were identical to the

administration of the open EnD and control except for the following step: After opening the cans,

and while out of the participants’ sight, the experimenter switched the real EnD can with another

EnD can that contained soda (placebo group) or switched the soda can with another soda can

containing EnD (hidden EnD group). All participants then watched the experimenter pour 250 ml

of the contents into a glass and were asked to drink all of it.

A poster highlighting the benefits of the ingredients of the EnD for mental performance was

displayed in the room for all four groups; the attention of the participants assigned to the placebo

and open EnD groups was drawn to this poster (see SI Figure 3 for the detailed instructions,

adapted from (25)). For these groups, the experimenter further highlighted that the ingredients of

one can of the EnD (80 mg of caffeine, 25.5 g of sugar, and taurine) begin to stimulate the

nervous system and accelerate the heart rate about 10 minutes after ingestion in order to energize

mental performance. After administration, participants were accompanied to an isolated test

booth to perform the various cognitive tasks. During the 10 minutes of waiting time, participants

received instructions and went through several practice trials for the main task, described below.

Participants were told that they would undertake two different cognitive tasks similar to graduate

10

exams in finance and accounting. The first task was our main task of interest; it tested whether

beliefs about the drink that participants consumed affected their performance and whether these

effects would be amplified by different monetary incentives at stake on a trial-by-trial basis. The

second task was, in line with our cover story, similar to graduate exams in finance and accounting

(see supplements for a more detailed description). Participants were compensated €12 for their

participation and received an additional bonus based on their performance in the first task.

Numerical Stroop task

The main task consisted of a modified version of a previously used numerical Stroop task (17)

that allowed us to sample (1) participants’ cognitive performance and (2) how monetary

incentives affected their cognitive performance (i.e., incentive motivation) (20, 26) in a within-

participant and repeated-measure design. More specifically, the task goal involved earning as

much points as possible by detecting the numerically greater figure within figure pairs that

differed in their physical and numerical size. Importantly, the amount of points to earn according

to their performance on a given trial could vary between 1, 5, or 10 points. The points were

translated into a monetary payoff at the end of the experiment. We used this task because in

addition to being a reliable task to measure cognitive performance and incentive motivation, it

has also been shown to engage the dopaminergic system (17, 20, 27). This is important given that

our hypothesis about the role of motivation for placebo effects stems from findings about the role

of the dopaminergic system to mediate placebo effects (11, 13, 14, 18).

The task was set up as follows (see Figure 1b): Participants were first asked to indicate their

anticipated performance on the upcoming trials. We asked this question to measure their

expectancy about the efficacy of the drink they thought they had consumed. Participants saw a

11

graduated line representing a ladder with 10 steps, one for each figure pair, and were asked to

move the cursor onto the step they expected to reach in the upcoming trials. Because participants

underwent a brief training phase on this task during the waiting time for the EnD to be “active,”

they were able to form expectancies about their upcoming performance. Participants also had to

indicate their anticipated performance every sixth trial, for a total of 12 measures. Note that only

the first response was used as a measure of the strength of participants’ expectancies about the

efficacy of the drink they had consumed. The later responses were affected by their actual

performance on prior trials and thus were used as an additional measure of participants’

confidence reported in the supplements.

Subsequently, the amount of points to be earned (1, 5, or 10) was displayed in a yellow circle for

500 ms. After the point display, the same ladder used during the anticipated performance rating

appeared on the right side, next to the points to earn, for a response period of 3,500 ms. Each of

the 10 steps of the ladder corresponded to a fraction (10%) of the points at stake. The goal in each

trial was to earn as many points as possible—that is, to move the cursor as high as possible on the

ladder. To climb up one step of the ladder, participants had to indicate on which side the

numerically greater figure was shown by pressing the left or right keyboard button. A white box

highlighted the pair of figures to be considered. If the correct button was pressed, the white

cursor moved one step up, indicating that 10% of the points at stake were earned and highlighting

the next figure pair to consider. If the incorrect button was pressed, the cursor could not be

moved for 1 second of the response period (3,500 ms).

12

Importantly, the figures varied in both numerical (between 0 and 9) and physical size (between

two possible fonts).1 In some pairs, the numerically greater figure was also physically greater

(i.e., congruent pairs), whereas in other pairs it was smaller (i.e., incongruent pairs). Incongruent

pairs are known to generate a Stroop effect that requires more mental effort to inhibit interference

from the irrelevant information (i.e., font size). Half of the trials consisted of a 50% mix of

congruent and incongruent pairs (i.e., difficult trials), whereas the other half consisted of

congruent pairs only (i.e., easy trials) to vary cognitive demand (i.e., difficulty). At the end of the

response window, the cumulative total of points earned was displayed to give feedback to the

participants. The feedback was shown for random time intervals (3,000 ms + 100 to 1,000 ms) in

order to keep participants’ attention.

In all, we applied a 2 (difficulty: 100% and 50% congruent pairs) x 3 (number of points: 1, 5, and

10) factorial within-participant task design. Each condition was repeated 12 times, randomly

distributed over a series of 72 trials with a total task duration of about 10 minutes.

Statistical analysis approach

All statistical tests were conducted with the Statistics and Machine Learning Toolbox (MATLAB

2015a, from MathWorks). A linear mixed effects model (lme) using the fitlme function in

MATLAB was conducted for cognitive performance (i.e., percentage of the number of steps

climbed up the ladder on each trial) as dependent variable, with fixed effects for label (coded 1

for Sprite, 2 for Red Bull), drink (coded 1 for Sprite, 2 for Red Bull), trial number (coded 1 to 72,

to correct for possible learning effects), incentives (points to earn, coded 1, 2, or 3), difficulty

(coded 1 for easy, 2 for difficult), and all possible two-way interactions. The model also included

1The difference in physical size (i.e., font size) was the same for all pairs, but the numerical difference varied from 1 to 5 (with two pairs of each in all trials). New number pairs were presented in each trial so that participants could not anticipate them. �

13

three uncorrelated random effects. We nested the intercept and incentives by participants (coded

1 to 88), to assess whether performance and incentive motivation varied between participants.

The third random effect included the intercept nested by label (coded 1 for Sprite, 2 for Red Bull)

in order to assess effects of label on performance intercept irrespective of the fixed effects. All

regressors were z-scored. Results are reported in Table 1. A similar analysis was performed for

mean reaction times (see SI Table 1).

In addition we used a Bayesian approach to estimate posterior distributions of parameters

describing tendencies in the comparison of incentive motivation in the EnD versus Soda label;

and EnD versus Soda drink groups (22) (28). This analysis was conducted using the Matlab

Toolbox for Bayesian Estimation (BEST) (https://github.com/NilsWinter/matlab-bayesian-

estimation/blob/master/README.md). BEST provides similar results compared to the standard

linear mixed effects model, but it allows for different interpretations. Using flat priors and a

Monte-Carlo-Markov-Chain (MCMC) algorithm it generates parameter values (e.g. mean

differences between groups, effect sizes) from a posterior distribution of parameter values that

are jointly credible given the data. Parameter values that fall within a high-density interval (HDI,

95% of the possible parameters) are considered as more credible than parameter values falling

outside the HDI. Thus, this method allows consideration of the null, for example zero difference

in incentive motivation between the EnD and soda drink or label groups. Specifically, we used

BEST to calculate posterior distributions of two parameters: (1) the mean differences in the effect

of incentives on cognitive performance between the labels (EnD vs. soda), and the drinks (EnD

vs. soda), and (2) the corresponding effect sizes of these differences. To this aim we pulled

together: the placebo and open EnD groups into one EnD label group; the control and hidden

EnD groups into one soda label group; the open EnD and hidden EnD group into one EnD drink

14

group; and the placebo and control group into one soda drink group. Thus, each label and drink

group counted 44 participants. The effect of incentives on cognitive performance (i.e. measure

for incentive motivation) corresponded to the regression coefficient (i.e. beta) of incentives

obtained for each participant using a general liner model of cognitive performance predicted by

incentives, cognitive demand, trial number and all possible interactions (SI paragraph 1.2.

Alternative general linear model analysis, SI table 2).

Acknowledgements: We thank Daphna Shohamy and Vasilisa Skortskova for insightful

discussion, Sebastien Robin and the staff of the Centre Multidisciplinaire des Sciences

Comportementales Sorbonne-Universités-INSEAD for help with data collection.

15

Figure legends Fig. 1. (A) Experimental design: In a 2x2 between-participant double-crossed balanced placebo

design, participants were assigned to a control, a placebo, a hidden EnD, or an open EnD group.

Each group counted 22 participants. (B) Task timeline: Successive screenshots displayed are

shown from left to right with durations in ms. Participants rated every sixth trial how well they

expected to perform during the next block of trials. The goal in each trial was to move a white

cursor up the scale as high as possible in order to earn as many points as possible. Each step on

the scale represented 10% of the points to earn in a given trial. To move the cursor up step by

step participants had to select the keyboard button on the side of the numerically greater figure in

the white box.

Fig. 2. Placebo effects of Red Bull label on incentive motivation. (A) The graph depicts cognitive

performance during low- (1-point), medium- (5-point), and high-incentive trials (10-point) by

drink label and actual consumption condition. Error bars correspond to standard errors of the

mean. (B) Bayesian estimation of the effect of EnD and soda label on incentive motivation. The

upper and middle left panels depict histrograms of the observed beta estimates for the effect of

incentives on cognitive performances in the EnD label group (green), Soda label group (blue)

with representative examples of posterior predictive distributions superimposed. The upper and

middle right panels depict the estimated posterior distributions of these effects given the observed

data with mean (mode) and marginal minimal (min) and maximal (max) values. The lower panels

depict the posterior distribution of mean differences and effect sizes for this mean difference.

ROPE designates the region of practical equivalence (red broken bars). Note, the green bar

indicates the localization of zero mean difference, which falls outside the 95% high density

interval (HDI).

16

SI Fig. 1. Mediation analysis diagram for the direct effect of label on cognitive performance

(paths c and c'), the effect of label on expectancy (i.e., anticipated performance) (path a), the

effect of expectancy on cognitive performance (path b), and the mediating effect of expectancy

(path ab). * p < .05 two-tailed, + p < .05 one-tailed. Bar graphs depict effects of drink and label on

belief and cognitive performance, respectively. Error bars correspond to standard errors of the

mean.

SI Fig. 2. Effect of label on incentive motivation during difficult and easy trials. The bars

represent the difference in cognitive performance in 10- versus 1-point trials averaged across

participants in each group for difficult (left) and easy (right) trials. Error bars correspond to

standard errors of the mean.

SI Fig. 3. Poster highlighting the benefits of EnDs such as Red Bull for performance under

pressure. Note that all participants saw this poster (in French), because it was taped on the wall of

the room in which they were administered the drinks. Participants assigned to the placebo or open

EnD groups were explicitly asked to read it.

References

1. Pettit ML & DeBarr KA (2011) Perceived stress, energy drink consumption, and academic performance among college students. (Translated from eng) J Am Coll Health 59(5):335-341.

2. Pettit ML & DeBarr KA (2011) Perceived Stress, Energy Drink Consumption, and Academic Performance Among College Students. J. Am. Coll. Health 59(5):335-341.

3. Sahakian B & Morein-Zamir S (2007) Professor's little helper. (Translated from eng) Nature 450(7173):1157-1159.

4. Smith M, Farah, E., Farah, M.J. (2011) Are prescription stimulants "smart pills"? The epidemiology and cognitive neuroscience of prescription stimulant use by normal healthy individuals. Psychological Bulletin 137(5):714-741.

5. Scholey AB & Kennedy DO (2004) Cognitive and physiological effects of an “energy drink”: an evaluation of the whole drink and of glucose, caffeine and herbal flavouring fractions. Psychopharmacology (Berl). 176(3):320-330.

17

6. Kennedy DO & Scholey AB (2004) A glucose-caffeine ‘energy drink’ ameliorates subjective and performance deficits during prolonged cognitive demand. Appetite 42(3):331-333.

7. Brasel SA & Gips J (2011) Red Bull "Gives You Wings" for better or worse: A double-edged impact of brand exposure on consumer performance. (Translated from English) J Cons. Psychol. 21(1):57-64.

8. Irmak C, Block LG, & Fitzsimons GJ (2005) The placebo effect in marketing: Sometimes you just have to want it to work. (Translated from English) J. Marketing Res. 42(4):406-409.

9. Plassmann H, Wager, T.D. (2014) How expectanies shape consumption experiences. . The interdisciplinary science of consumption., ed Stephanie D. Preston MLK, Brian Knutson (MIT Press).

10. Schweinhardt P, Seminowicz DA, Jaeger E, Duncan GH, & Bushnell MC (2009) The anatomy of the mesolimbic reward system: a link between personality and the placebo analgesic response. (Translated from eng) J Neurosci 29(15):4882-4887.

11. Scott DJ, et al. (2008) Placebo and nocebo effects are defined by opposite opioid and dopaminergic responses. (Translated from eng) Arch Gen Psychiatry 65(2):220-231.

12. Benedetti F, et al. (2004) Placebo-responsive Parkinson patients show decreased activity in single neurons of subthalamic nucleus. Nature Neuroscience 7(6):587-588.

13. de la Fuente-Fernandez R, et al. (2001) Expectation and dopamine release: mechanism of the placebo effect in Parkinson's disease. (Translated from eng) Science 293(5532):1164-1166.

14. Lidstone SC, et al. (2010) Effects of expectation on placebo-induced dopamine release in Parkinson disease. (Translated from eng) Arch Gen Psychiatry 67(8):857-865.

15. Schmidt L, Braun EK, Wager TD, & Shohamy D (2014) Mind matters: placebo enhances reward learning in Parkinson's disease. (Translated from Eng) Nat Neurosci.

16. Ben-Shaanan TL, Azulay-Debby, H., Dubovik, T., Starosvetsky, E., Korin, B., Schiller, M., Green, N.L., Admon, Y., Hakim, F., Shen-Orr, S.S., Rolls, A. (2016) Activation of the reward system boosts innate and adaptive immunity. Nature Medicine 22:940-944.

17. Schmidt L, Lebreton M, Clery-Melin ML, Daunizeau J, & Pessiglione M (2012) Neural mechanisms underlying motivation of mental versus physical effort. (Translated from eng) PLoS Biol 10(2):e1001266.

18. Wager TD & Atlas LY (2015) The neuroscience of placebo effects: connecting context, learning and health. (Translated from eng) Nat Rev Neurosci 16(7):403-418.

19. Schmidt L, et al. (2008) Disconnecting force from money: effects of basal ganglia damage on incentive motivation. (Translated from eng) Brain 131(Pt 5):1303-1310.

20. Berridge KC, Robinson TE, & Aldridge JW (2009) Dissecting components of reward: 'liking', 'wanting', and learning. (Translated from eng) Curr Opin Pharmacol 9(1):65-73.

21. Cumming G (2014) The new statistics: why and how. Psychological Science 25(7):7-29. 22. Kruschke JK (2013) Bayesian estimation supersedes the t test. Journal of Experimental

Psychology: General 142(2):1-33. 23. Howard MA, Marczinski, C.A. (2010) Acute effects of a glucose energy drink on

behavioral control. Experimental and Clinical Psychopharmacology 18(6):553-561. 24. Scholey AB, Kennedy, D.O. (2004) Cognitive and physiological effects of an energy

drink: an evaluation of the whole drink and of glucose, caffeine and herbal flavouring fractions. Psychopharmacology 176:220-330.

18

25. Irmak C, Block, L.G., Fitzsimons, G.J. (2005) The placebo effect in marketing: sometimes you just have to want it to work. Journal of Marketing Research 42:406-409.

26. Pessiglione M, et al. (2007) How the brain translates money into force: a neuroimaging study of subliminal motivation. Science 316(5826):904-906.

27. Aarts E, van Holstein M, & Cools R (2011) Striatal Dopamine and the Interface between Motivation and Cognition. (Translated from eng) Front Psychol 2:163.

28. Kruschke JK (2014) Doeing Bayesian data analysis. A tutorial zith R, JAGS and Stan. 2nd Edition. Academic Press, Amsterdam.

Figure1

a

b

500 ms

ITI#

500 ms

Incen've#

3000 ms

Effort# Outcome#

####### #######10#P# #######10#P#

9 7 6 2 1 6 5 2 4 5 7 2 4 8 3 6 2 0 5 6

TOTAL#2.5#points# ####### +

3500 ms

Ra'ng#

500 ms

How#many#steps#do#

you#expect#to#climb#up#in#the#next#trials?#

label: Sprite, drink: Sprite label: Red Bull, drink: Sprite

label: Sprite, drink: Red Bull label: Red Bull, drink: Red Bull

Open energy drink

Placebo

Hidden energy drink

Control

a

b

Figure2

% P

erfo

rman

ce Label: Red Bull

Label: Sprite

Incentive (points to earn)

Drink: Red Bull Drink: Sprite

43%

45%

47%

49%

51%

53%

55%

1 5 10 1 5 10

Posterior distribution differences of means

µ Red Bull label - µ Sprite label

mode=0.107

Min=0.047 Max=0.1750 0.05 0.1 0.15 0.2

Posterior distribution of effect sizes

mode=0.735

Min=0.314 Max=1.218

0%inROPE

(µ Red Bull label - µ Sprite label)/(Sqrt (σ Red Bull label + σ Sprite label)/2)

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6

Red Bull Label Group

Sprite Label Group

Observed Data with Posterior Predictions

N=44

P(In

cent

ive

mot

ivat

ion)

-0.4 -0.2 0 0.2 0.4 0.6

3 2.5 2 1.5 1 0.5 0 -0.1 -0.05 0 0.1 0.05 0.15

Posterior distribution of means

mode=0.089

Min=0.044 Max=0.136

µ Red Bull label

-0.1 -0.05 0 0.1 0.05 0.15

Posterior distribution of means

µ Sprite label

mode=-0.024

Min=-0.068 Max=0.023

Incentive motivation -0.4 -0.2 0 0.2 0.4 0.6

N=44 3 2.5 2 1.5 1 0.5 0

Observed Data with Posterior Predictions

P(In

cent

ive

mot

ivat

ion)

Incentive motivation

Red Bull versus Sprite Label Group

S1

a

ab=0.005, SE=0.004*

c=0.021, SE=0.01*

c’=0.017, SE=0.01+

Belief

Label Perfor- mance

% a

ntic

ipat

ed p

erfo

rman

ce

% p

erfo

rman

ce

Label Red Bull

Label Sprite

54%

59%

64%

69%

74%

Red Bull Sprite

drink

44%

46%

48%

50%

52%

54%

Red Bull Sprite

drink

0

1

2

Red Bull Sprite

drink

S2

-3%

-1%

1%

3%

5%

Red Bull Sprite

drink

-3%

-1%

1%

3%

5%

Red Bull Sprite

drink

Effect of label on incentive motivation %

per

form

ance

(10

– 1

poin

ts)

Difficult trials

Label Red Bull Label Sprite

Easy trials

Drink: Red Bull

Drink: Sprite

Drink: Red Bull

Drink: Sprite

S3

A can of Red Bull contains natural and healthy ingredients like taurine, vitamin B and caffeine. Taurine is a natural substance produced by the body under stress, and mental or physical effort. A scientific study from 2001 published in the European journal Amino Acids has found that 10 minutes after consumption of Red Bull, taurine and caffeine peak in your blood stream. Your cardiac rhythm gets faster, your blood pressure increases, and you feel awake as the stimulant kicks in. Moreover, caffeine is an ergogenic substance—it increases performance and wakefulness—via stimulation of the central nervous system. This is beneficial for mental effort during exams.

Red Bull improves performance and concentration, so you can tackle everything from physics to workouts. It also increases endurance, which means you can enjoy the more important things afterward.

Red Bull

Coffee

Tea

Soda

Performance per 25 cl:

Table 1

Linear Mixed Effects Model Predicting Cognitive Performance During the Numerical Stroop Task

Note: CI = confidence interval, LL = lower limit, UL = upper limit, DF = degrees of freedom, SD = standard deviation

Model informationNumber of

Observations

Fixed effects coefficients

Random effects

coefficients

Covariance

parameters

6336 16 264 5

AIC BIC Log Likelihood

Deviance

22836 22978 −11397 22794

Fixed effects predictors β SD t DF p LL UL

Intercept 4.84 0.10 47.67 6320 <.001 4.64 5.04Trial number 0.19 0.02 10.40 6320 <.001 0.15 0.22Incentives 0.03 0.02 1.85 6320 .06 −0.002 0.07Difficulty −1.05 0.02 −58.32 6320 <.001 −1.08 −1.02Label 0.21 0.10 2.13 6320 .04 0.01 0.41Drink −0.002 0.10 −0.02 6320 .98 −0.20 0.19Trial number by Incentives 0.01 0.02 0.48 6320 .63 −0.03 0.04Trial number by Difficulty −0.04 0.02 −2.11 6320 .03 −0.07 −0.003Incentives by Difficulty 0.02 0.02 1.34 6320 .18 −0.01 0.06Trial number by Label 0.01 0.02 0.61 6320 .55 −0.02 0.04Incentives by Label 0.05 0.02 3.02 6320 .002 0.02 0.09Difficulty by Label −0.07 0.02 −4.13 6320 <.001 −0.11 −0.04Trial number by Drink 0.01 0.02 0.70 6320 .49 −0.02 0.05Incentives by Drink −0.003 0.02 −0.19 6320 .83 −0.04 0.03Difficulty by Drink −0.01 0.02 −0.63 6320 .53 −0.05 0.02Label by Drink −0.03 0.10 −0.33 6320 .74 −0.23 0.16

Random effects covariance parameters

Subject (88 Levels) β LL UL

Intercept 0.93 0.81 1.1Incentives 0.02 0.003 0.13

Error β LL UL

Residual 1.43 1.41 1.45

95% CI

95% CI

1

Red Bull Gives You Incentive Motivation: Understanding Placebo Effects of Energy Drinks on Cognitive Performance Liane Schmidt,1,3 Pierre Chandon,2 Mathias Pessiglione,4 and Hilke Plassmann2,3 1Centre Multidisciplinaire des Sciences Comportementales Sorbonne Universités-INSEAD

2 INSEAD, Fontainebleau, France 3Laboratoire de Neurosciences Cognitives, INSERM U960, Département de Sciences Cognitives, Ecole Normale Supérieure, Paris, France 4 Motivation, Brain, Behavior Team, INSERM UMRS975, Institute du Cerveau et de la Moelle Epinière, Paris, France

Supplementary Materials 1. Additional analyses and results

1.1. Effects on reaction times 1.2. Alternative general linear model analysis of cognitive performance 1.3. Mediation analysis for the effect of label on cognitive performance 1.4. Effects on confidence

2. Additional supplementary tables 1–9 3. Instructions

3.1. Recruitment e-mail 3.2. Experimental instructions and drink administration 3.3. Instructions for numerical Stroop task

2

1. Additional analysis and results

1.1. Effects on reaction times:

Analogous to performance, a linear mixed effects model (lme) using maximum log likelihood

was conducted to fit the fitlme function in MATLAB for average reaction times during the

numerical Stroop task, with fixed effects for label (coded 1 for soda or 2 for EnD), drink

(coded 1 for soda or 2 EnD), trial number (coded 1 to 72, to correct for possible learning

effects), points to win (coded 1, 2, or 3 to assess incentive motivation), difficulty (coded 1 for

easy, 2 for difficult), all possible two-way interactions, and uncorrelated random effects for

intercept and points to win nested by participant (coded 1 to 88) and intercept nested by label

(coded 1 for soda or 2 for EnD). All regressors were z-scored. Results are reported below in

SI Table 1. We found a significant main effect of trial number (β = –0.01, p = .03), suggesting

that the participants’ responses became faster across trials (SI Table 1), which is in line with

the positive effect of trial number on performance (Table 1). Moreover, reaction times slowed

down in difficult trials compared to easy trials, as indicated by a significant positive effect of

difficulty (β = 0.05, p < .001).

3

1.2. Alternative general linear model analysis of cognitive performances:

An alternative statistical analysis of the data was performed using a generalized linear model

(GLM) predicting each participant’s cognitive performance (i.e., % of steps climbed up the

scale on each trial) with the following regressors (z-scored): trial number to correct for

possible learning effects, points to earn to assess incentive motivation, difficulty, and all

possible interactions. Regression coefficients (i.e., beta estimates) for the intercept and slopes

(SI Table 2) were fit into a second-level random effects analysis using factorial 2x2 analysis

of variance (ANOVA) to obtain main effects of label, of drink, and of the interaction label by

drink on the intercept and all slopes.

a) Second-level random effects of label on the intercept:

A significant main effect of label was observed on the intercept of cognitive performance,

F(1,87) = 4.14, p = .04, η2 = .05. Participants assigned to an EnD-label group started off with

better performance than participants assigned to the soda-label groups. There was no

significant main effect of drink (F(1,87) = 0, p > .5) and no interaction between label and

drink (F(1,87) = 0.18, p > .5).

b) Second-level random effects of label on the slope of performance across trials:

To assess potential effects of learning (i.e., increase in performance across trials) or fatigue

(i.e., decrease in performance across trials), a second-level random effects ANOVA revealed

a significant interaction between label and drink on the slope of performance across trials,

F(1,87) = 9.5, p = .003, η2 = .10. No significant main effects of label (F(1,87) = 0.3, p > .5) or

drink (F(1,87) = 0.7, p = .4) were found. Participants who were drinking and EnD labeled as

an EnD (i.e., open EnD) improved in performance more across trials than those in the placebo

and hidden EnD groups (post hoc two-sampled, two-tailed t-tests on the beta estimates for

trial number: open EnD > placebo: t(42) = 2.5, p = .02, Cohen’s d = 3.5; open > hidden EnD:

4

t(42) = 3.1, p = .004, Cohen’s d = 4.3). Similarly, participants in the control group, who were

administered a soda labeled as a soda, tended to improve more in performance across trials

compared to the hidden EnD (control > hidden EnD: t(42) = 1.9, p = .03, one-tailed, Cohen’s

d = 2.6) and placebo (control > placebo: t(42) = 1.6, p = .06., one-tailed, Cohen’s d = 2.2)

groups. No significant differences were found between the open EnD and control group (open

EnD > control: t(42) = 1.2, p = .25). This finding suggests that participants who were not

deceived about the content of the drink they consumed improved in task performance across

trials compared to participants who were deceived about the drink’s identity.

c) Second-level random effects of label on the slope of performance across incentives:

An ANOVA revealed a main effect of label (F(1,87) = 11.9, p < .001, η2 =.13) on the slopes

of performance across incentives. Participants who were administered a drink labeled as an

EnD allocated their effort more according to the magnitude of incentives (i.e., points to earn)

(linear effect of incentives placebo group: r = 0.1 ± 0.03, p < .001; open EnD group: r = 0.06

± 0.03, p < .05) than participants who were administered a drink labeled as soda (control

group: r = –0.03 ± 0.03, p = .24; hidden EnD group: r = –0.01 ± 0.03, p = .76). There was no

significant main effect of drink (F(1,87) = 0.01, p > .5), nor was there an interaction between

label and drink (F(1,87) = 1.28, p > .25).

d) Second-level random effects of label on the slope of performance across difficulty

conditions: The average regression coefficient for the effect of difficulty (easy vs. difficult)

was negative in each group, suggesting that performance was limited in difficult trials (i.e.,

50% congruency of number pairs) compared to easy trials (i.e., 100% congruency of number

pairs). Yet the ANOVA found no significant main effects of label (F(1,87) = 1.96, p = .16) or

drink (F(1,87) = 0.05, p > .5), and no interaction between label and drink (F(1,87) = 0.06, p >

.5). This finding stands in contrast to the results from a more powerful linear mixed effects

model (reported in the main text) that indicated a significant interaction label by difficulty

5

(β = –0.07, p < .001) (Table 1). As shown in SI Figure 3, further exploration of the data

revealed that this interaction might be due to enhanced incentive motivation in the placebo

and open EnD groups during difficult trials, as suggested by a significant three-way

interaction between label, incentives, and difficulty (linear mixed effects model on cognitive

performance: β = –0.06, SE = 0.02, p < .001, 95% CI [–0.09 to –0.02]) (SI Figure 1).

1.3. Mediation analysis for the effect of label on cognitive performance:

To further investigate the role of participants’ expectancies, we conducted a formal mediation

analysis of label effects on cognitive performance by anticipated performance ratings. To this

aim we performed a mediation analysis using the Mediation Toolbox

(http://wagerlab.colorado.edu/tool). Mediation analysis extends standard analyses by jointly

considering two dynamic relationships: (1) between the label and the associated expectancy of

the EnD’s efficacy and (2) between the expectancy of the EnD’s efficacy and cognitive

performance. Testing these effects involves the following path regressions(1):

(1) path c: y = cx + ey

(2) path a: m = ax + em

(3) path b: y = bm + c'x + e'y

The variables x, y, and m correspond to participant-by-participant data vectors containing each

participant’s average cognitive performance (y), label (i.e., EnD label coded 1 and soda label

coded –1) (x), and the expectancy of the efficacy of the drink that participants thought they

had consumed given on the first of the 12 anticipated performance rating trials (m). The

variables ey, em, and e'y denote residual variance for each of the three regression analyses,

respectively. The first regression, called path c, tests the direct (or total) effect of the

experimental manipulation (x = label) on behavior (y = cognitive performance). The second

regression, called path a, tests the relationship between the experimental manipulation (x =

label) on the mediator variable (m = expectancy, i.e.; anticipated performance). The third

6

regression, called path b, assesses the relationship between the mediator (m = anticipated

performance) and behavior (y = actual cognitive performance), controlling for the

experimental manipulation (x = label). Finally, single-level mediation is assessed by the

product of path a and path b coefficients: a * b = c – c', with c' corresponding to the direct

effect of label on actual performance after controlling for mediation by anticipated

performance. In order to control for joint effects of the verum drink consumed, drink was

added as a covariate (coded 1 for soda and 2 for EnD drink).

A bootstrap test was conducted to estimate the significance of path coefficients (2, 3). This

involved estimating the distribution of individual path coefficients by randomly sampling

with replacement 10,000 observations from the matrix of [a, b, c, c', a*b] path coefficients.

Two-tailed p-values were calculated from the bootstrap confidence intervals.

As outlined in SI figure 1 this mediation framework found that (1) label positively predicted

the participants’ expectancy (ßa = 0.04, SE = 0.02, p = .06 two-tailed), with participants

expecting to perform better when administered a drink labeled as EnD compared to soda; (2)

expectancy significantly predicted cognitive performance, controlling for joint effects of label

(ßb = 0.11, SE = 0.04, p = .01 two-tailed); and (3) label significantly predicted cognitive

performance corresponding to the main effect of label on performance (ßc = 0.021, SE =

0.010, p =.035 two-tailed). Crucially, expectancy formally mediated the effect of label on

cognitive performance as indicated by a significant path ab coefficient (ßab = 0.005, SE =

0.004, p =.04 two-tailed). The mediating effect of expectancy was partial as the direct effect

of label on performance trended toward significance after controlling for belief strength (ßc' =

0.016, SE = 0.010, p =.10 two-tailed).

Taken together, these findings suggest that the expectancy of performing better linked to the

efficacy of the drink that participants thought they had consumed increased their cognitive

performance.

7

1.4. Effects on confidence in performance:

Participants rated their confidence every sixth trial during the numerical Stroop task, for a

total of 12 measures. A linear mixed effects model was fit to these ratings with the following

fixed effects regressors: label (coded 1 for soda or 2 for EnD), drink (coded 1 for soda or 2 for

EnD), trial number (coded 1 to 12, to correct for possible learning/habituation effects), all

possible two-way interactions, and an uncorrelated random effect for intercept grouped by

participant (coded 1 to 88). Results are reported in SI Table 4.

We found a significant negative effect of trial number (ß = –0.22, p < .001), suggesting that

confidence in performance decreased across trials during the numerical Stroop task. This

effect interacted with label (ß = –0.08, p = .04) as that confidence decreased faster over trials

in the placebo and open EnD groups than in the control or hidden EnD groups.

8

2. Supplementary tables 1 - 4

Supplementary Table 1 Linear Mixed Effects Model Predicting Reaction Time During the Numerical Stroop Task

Model informationNumber of

ObservationsFixed effects coefficients

Random effects coefficients

Covariance parameters

6336 16 264 5

AIC BIC Log Likelihood Deviance

6446.4 6588.2 −3202.2 6404.4

Fixed effects predictors β SD t DF p LL UL

Intercept 0.18 0.01 16.02 6320 .16 0.16 0.20Trial number −0.01 0.005 −2.15 6320 .03 −0.02 −0.0009Incentives −0.001 0.006 −0.17 6320 .86 −0.01 0.01Difficulty 0.05 0.005 9.78 6320 <.001 0.04 0.06Label −0.00004 0.01 −0.004 6320 .99 −0.02 0.02Drink −0.006 0.01 −0.51 6320 .61 −0.03 0.02Trial number by Incentives 0.0006 0.005 0.12 6320 .90 −0.01 0.01Trial number by Difficulty −0.01 0.005 −1.88 6320 .06 −0.02 0.0004Incentives by Difficulty −0.002 0.005 −0.41 6320 .68 −0.01 0.008Trial number by Label −0.001 0.005 −0.16 6320 .87 −0.01 0.009Incentives by Label −0.002 0.006 −0.28 6320 .77 −0.01 0.01Difficulty by Label 0.003 0.005 0.65 6320 .52 −0.006 0.01Trial number by Drink −0.008 0.005 −1.65 6320 .09 −0.02 0.001Incentives by Drink −0.001 0.006 −0.22 6320 .82 −0.01 0.01Difficulty by Drink 0.001 0.005 0.26 6320 .79 −0.008 0.01Label by Drink 0.01 0.01 1.31 6320 .19 −0.007 0.03Random effects covariance

parameters

Subject (88 Levels) β LL UL

Intercept 0.06 NaN NaNPoints to win 0.03 0.02 0.05

Error β LL UL

Residual 0.39 0.38 0.40

95% CI

95% CI

Note: CI = confidence interval, LL = lower limit, UL = upper limit, DF = degrees of freedom, SD = standard deviation

9

Supplementary Table 2 Multiple Linear Regression Models Predicting Performance During the Numerical Stroop Task for Each of the Four Experimental Groups

Placebo group: drink soda, label EnD β SD t Z p

Intercept 5.10 0.21 23.87 8.13 <.001Trial number 0.12 0.05 2.60 2.12 .02Incentives 0.10 0.03 2.95 2.42 <.001Difficulty -1.11 0.13 -8.74 -5.50 <.001Trial number by Incentives -0.01 0.03 -0.38 -0.55 .71Trial number by Difficulty -0.08 0.03 -2.99 -2.45 <.001Incentives by Difficulty 0.04 0.03 1.69 1.25 .11Trial number by Incentives by Difficulty -0.03 0.03 -1.11 -0.58 .28

Open energy drink group: drink EnD, label EnD

β SD t Z p

Intercept 5.01 0.22 23.27 8.13 <.001Trial number 0.27 0.03 9.25 5.62 <.001Incentives 0.06 0.03 2.02 1.58 .05Difficulty -1.11 0.10 -11.09 -6.18 <.001Trial number by Incentives 0.01 0.05 0.31 0.72 .76Trial number by Difficulty -0.06 0.05 -1.29 -0.80 .21Incentives by Difficulty 0.04 0.04 0.96 0.39 .35Trial number by Incentives by Difficulty 0.01 0.03 0.39 0.52 .70

Control group: drink soda, label soda

β SD t Z p

Intercept 4.59 0.25 18.46 7.57 <.001Trial number 0.22 0.03 7.82 5.12 <.001Incentives -0.03 0.03 -1.21 -0.71 .24Difficulty -0.93 0.09 -10.36 -5.99 <.001Trial number by Incentives 0.01 0.03 0.34 0.63 .74Trial number by Difficulty -0.01 0.05 -0.27 -0.80 .79Incentives by Difficulty -0.01 0.02 -0.30 -0.72 .77Trial number by Incentives by Difficulty -0.03 0.03 -0.97 -0.41 .34

Hidden energy drink group: drink EnD, label soda β SD t Z p

Intercept 4.68 0.14 33.68 8.13 <.001Trial number 0.15 0.03 4.99 3.82 <.001Incentives -0.01 0.03 -0.31 -0.71 .76Difficulty -0.99 0.09 -10.94 -6.15 <.001Trial number by Incentives 0.04 0.03 1.40 0.93 .18Trial number by Difficulty -0.00 0.03 -0.14 -1.22 .88Incentives by Difficulty 0.02 0.03 0.58 0.17 .60Trial number by Incentives by Difficulty 0.05 0.03 1.36 0.88 .19 Note: SD = standard deviation. Values describe effects of predicting regressors on the second level (i.e., population N = 22) for each group, respectively. Individual beta estimates for regressors of interest were used as dependent variables for a 2x2 [label by drink] analysis of variance, reported in sections 2.1 a–d.

10

Supplementary Table 3 Linear Mixed Effects Model Predicting Anticipated Performance During the Numerical Stroop Task

Model informationNumber of

ObservationsFixed effects coefficients

Random effects coefficients

Covariance parameters

88 4 88 2AIC BIC Log Likelihood Deviance

392.51 407.38 −190.26 380.51

Fixed effects predictors β SD t DF p LL ULIntercept 6.58 0.22 29.36 84 <.001 6.13 7.02Label 0.42 0.22 1.87 84 .06 −0.02 0.86Drink 0.15 0.22 0.66 84 .51 −0.29 0.59Label by Drink −0.24 0.22 −1.06 84 .29 −0.68 0.20

Random effects covariance parameters 95% CI

Subject (88 Levels) β LL ULIntercept 1.48 0 Inf

Error β LL ULResidual 1.48 0 Inf

95% CI

Note: CI = confidence interval, LL = lower limit, UL = upper limit, DF = degrees of freedom, SD = standard deviation

Supplementary Table 4 Linear Mixed Effects Model Predicting Confidence in Performance During the Numerical Stroop Task

Model informationNumber of

ObservationsFixed effects coefficients

Random effects coefficients

Covariance parameters

1056 7 88 2AIC BIC Log Likelihood Deviance

3810 3854.6 −1896 3792 95% CI

Fixed effects predictors β SD t DF p LL ULIntercept 4.67 0.16 29.01 1049 <.001 4.36 4.99Trial number −0.22 0.04 −5.42 1049 <.001 −0.31 −0.14Label 0.15 0.16 0.91 1049 .36 −0.16 0.46Drink 0.04 0.16 0.27 1049 .78 −0.27 0.36Trial number by Label −0.08 0.04 −2.02 1049 .04 −0.16 −0.002Trial number by Drink −0.02 0.04 −0.38 1049 .69 −0.09 0.06

Label by Drink −0.16 0.16 −0.98 1049 .32 −0.47 0.15

Random effects covariance parameters

95% CI

Subject (88 Levels) β LL ULIntercept 1.46 1.25 1.71

Error β LL ULResidual 1.29 1.24 1.35 Note: CI = confidence interval, LL = lower limit, UL = upper limit, DF = degrees of freedom, SD = standard deviation

11

3. Instructions 3.1 Recruitment mail

Hi, You are receiving this email because you are on the mailing list of the Sorbonne Universités-INSEAD Behavioral Lab. We are currently recruiting participants for a new study called “Energy Drinks.” The study involves drinking a soda-like sweet drink that may contain caffeine prior to performing a series of tasks on a computer. It will take about 1 hour, and your participation will be remunerated €12 plus a bonus calculated according to performance.

Please respond here or at http://participant.inseadlab.com if you are interested in participating.

IMPORTANT: You must have no food allergies. Please do not drink any caffeinated drinks or alcohol at least 2 hours prior to the experiment. An alcohol test will be administered prior to enrollment.

Best,

Centre Multidisciplinaire des Sciences Comportementales Sorbonne Universités-INSEAD 6 rue Victor Cousin

75005 Paris

3.2 Experiment instructions and drink administration For participants assigned to the open energy drink group (i.e., drink Red Bull, label Red Bull) and placebo group (i.e., drink Sprite, label Red Bull): We invited you to this study to test the effect of energy drinks like Red Bull on your concentration, memory, and overall performance during exams. You were assigned to the group of participants that as a test group will have a drink of Red Bull before performing a couple of tests that are frequently part of prestigious accounting exams, and that assess some of the abilities of future employees in finance. College students usually buy Red Bull to be smarter, concentrate better, and decrease fatigue during learning for competitive exams such as national concourses. As a consequence, ads like the one below are frequently added to Amazon book orders:

12

For participants assigned to the control group (i.e., drink Sprite, label Sprite) and hidden energy drink group (i.e., drink Red Bull, label Sprite): We invited you to this study to test the effect of sodas and energy drinks like Red Bull on your concentration, memory, and overall performance during exams. You were assigned to the group of participants that as a control group will have a regular soda before performing a couple of tests that are frequently part of prestigious accounting exams, and that assess some of the abilities of future employees in finance. All participants were then told the following: You will perform two tasks. The first task consists of checking pairs of numbers as fast as possible and correctly identifying which one is the greatest The second task consists of learning stock market ticker symbols of some of the largest French companies. This is important for accountants and for anyone who is investing in the stock market. Your score on each task will determine your overall score, which will determine how much money you will be able to earn in the study. 3.3 Instructions for the numerical Stroop task The aim of the task is to score as high as possible in terms of total points during this exam. This first task comprises 72 trials and will last 10 minutes. To get started press the “S” key when the word “ready” is displayed on the screen. Each trial a coin, indicating the number of points you can score, will be displayed on the screen. After a few seconds, a scale will be displayed on the screen. The scale comprises 10 steps. Each step corresponds to a fraction of the number of points you can win. For instance, the top step corresponds to the whole number of points you can win, and the middle step to half of the number of points you can win.

13

To the left of each step a number pair will be displayed. The numbers within each pair vary in numerical and physical size. To win a fraction of the amount of points, you have to move a white cursor up the scale by selecting the numerically bigger number in each pair.

• To choose the right number, press the right arrow key (�) • To choose the left number, press the left arrow key (�)

If your choice is correct, the white cursor will move up one step to the next number pair. Thus, your score is proportional to the number of points at stake in each trial and your performance on the scale (the step the white cursor reaches on the scale). At the end of each trial a cumulative total will be displayed on the screen, indicting how many total points you have won so far. Before you start the task there will be a short practice session to get used to displays and response keys.

Good luck!

1. BaronRM&KennyDA(1986)Themoderator-mediatorvariabledistinction

insocialpsychologicalresearch:conceptual,strategic,andstatisticalconsiderations.(TranslatedfromEng)JPersSocPsychol51(6):1173-1182(inEng).

2. WooCW,RoyM,BuhleJT,&WagerTD(2015)Distinctbrainsystemsmediatetheeffectsofnociceptiveinputandself-regulationonpain.(Translatedfromeng)PLoSBiol13(1):e1002036(ineng).

3. AtlasLY,BolgerN,LindquistMA,&WagerTD(2010)Brainmediatorsofpredictivecueeffectsonperceivedpain.(Translatedfromeng)JNeurosci30(39):12964-12977(ineng).