Literature review

Conformity Feedback in an Online Review Helpfulness Evaluation Task Leads to Less Negative Feedback-Related Negativity

Amplitudes and More Positive P300 Amplitudes

Daomeng Guo Wuhan University and Hubei

Engineering University

Yang Zhao, Liyi Zhang, and Xuan Wen Wuhan University

Cong Yin Chongqing University of Technology

Compared with an offline context, the sources of online review are typically unknown, with more variable opinions from multiple individuals. This variability can make it difficult for consumers to judge the consensus of others’ views using only direct social clues. However, few studies have focused on the brain’s processing of mixed opinions in an online context. In this study, an experiment that involved voting on the helpful- ness of online reviews was designed to investigate how participants processed their personal views alongside others’ views. A total of 32 participants were asked to decide whether each online review was helpful and were then given feedback regarding how many people found each review helpful. Participants’ voting behaviors and conformity feedback-related event-related brain potentials (ERPs) were recorded and analyzed. Participants rated positive reviews as more helpful than negative reviews. Response times were longer when participants evaluated negative reviews. Therefore, the nega- tivity bias of reviews may not result from the review’s helpfulness but rather from the cognitive processing involved in the evaluation of the reviews. Further ERP analysis showed that the incongruence of participants’ choices with the relative majority opinion generated from a ranking of a review’s helpfulness elicited more negative-going feedback-related negativity and less positive-going P300 than did the condition of their choices’ congruence with the relative majority opinion. This finding suggests that incongruence with the relative majority opinion was processed as negative feedback due to expectation violation, whereas congruence with the relative majority opinion was processed as positive feedback for conformity. Furthermore, the feedback-related negativity response elicited by the trials of inconsistency with relative majority opin- ions during the early period was smaller than that in the later period, whereas the P300 response elicited by the trials of consistency with relative majority opinions in the early period was greater than that in the later period. The ERP results suggest that even in an online context, the brain can automatically encode the relative majority opinion by learning from a comparison of other visible social cues, and automatically categorize whether one’s personal views are consistent with those of the relative majority.

This article was published Online First February 25, 2019.

Daomeng Guo, School of Information Management, Wu- han University, and School of Economics and Management, Hubei Engineering University and Hubei Micro and Small Businesses Development Research Center; Yang Zhao, Liyi Zhang, and Xuan Wen, School of Information Management, Wuhan University; Cong Yin, Chongqing Intellectual Prop- erty School, Chongqing University of Technology.

This work was supported by grant 71373192, 71874126 from the National Natural Science Foundation of China; Daomeng Guo and Yang Zhao contributed equally to this work and should be considered cofirst authors.

Correspondence concerning this article should be ad- dressed to Liyi Zhang, School of Information Manage- ment, Wuhan University, 299# Bayi Road, Wuhan 430072, People’s Republic of China. E-mail: lyzhang@whu .edu.cn

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

Journal of Neuroscience, Psychology, and Economics © 2019 American Psychological Association 2019, Vol. 12, No. 2, 73– 87 1937-321X/19/$12.00 http://dx.doi.org/10.1037/npe0000102

73

Keywords: online review helpfulness, social judgment, negativity bias, relative majority, event-related potential

Amazon.com asks, “Was this review helpful to you?” after each online review and then shows the votes of review helpfulness alongside the review (e.g., “30 people found this help- ful”), and positions the most helpful reviews more prominently on the product information page. These methods are expected to help con- sumers overcome information overload and im- prove the effective use of online reviews. How- ever, these methods may influence consumers’ perception of online review helpfulness. Previ- ous studies have primarily focused on extract- ing features from vote-upon reviews to explain or predict the helpfulness of reviews (Baek, Lee, Oh & Ahn, 2015; Cao, Duan, & Gan, 2011; Hong, Xu, Wang, & Fan, 2017; Karimi & Wang, 2017; Lee, Jeong, & Lee, 2017; Mudambi & Schuff, 2010; Schindler & Bickart, 2012; Sen & Lerman, 2007; Wang, Li, & Sun, 2016; Wu, 2013; Yang, Chen, & Bao, 2017; Zhao, Ni, & Zhou, 2018). Those studies utilized the hypothesis that consumers’ usefulness eval- uation was based on their real experience and thoughts. In reality, consumers may post public opinions that are inconsistent with their private views due to social pressure. For example, con- sumers tend to post negative comments when they see negative comments from others, de- spite a positive personal experience with the product (Schlosser, 2005). In addition, the vast majority of individuals in an online context are lurkers (those not posting their opinions). Some studies have found differences in online reviews between lurkers (their private reviews) and posters as a result of less social pressure on lurkers (Moe & Schweidel, 2012; Schlosser, 2005). This finding may explain why the re- views predicted to be helpful by models are often not recognized by most consumers (Wang et al., 2016). However, the question remains as to how people process a vote for a review and perceive the social pressure of mixed opinions represented by the number of votes in an online context.

A substantial amount of research has shown that people’s opinions or behaviors are suscep- tible to the influence of others. People are mo- tivated by an accurate perception of reality or by

social identity to conform to others (Cialdini & Goldstein, 2004). In an online review context, many studies have revealed the existence of social influence. The closer the score is to the average product score, the higher the helpful- ness of the review (Baek et al., 2015); this finding suggests an effect of social influence on the perception of the helpfulness of online re- views. However, other individuals show anti- conformity behavior to generate a sense of uniqueness and personal identity. For example, consumers can post negative comments as a way to differentiate themselves from other members of a group (Moe & Schweidel, 2012). Current neuroimaging studies have provided more direct evidence and profound explanations for herd behavior. When people’s responses conflict with the group opinion, the ventral striatum is deactivated, and the regions of the posterior medial frontal cortex and anterior in- sula are activated. The activation of posterior medial frontal cortices can predict subsequent behavior conformity (Wu, Luo, & Feng, 2016). However, most studies have focused on social behavior in face-to-face environments (Thomas & Vinuales, 2017), whereas normative influ- ence is much weaker in a virtual environment (Perfumi, Cardelli, Bagnoli, & Guazzini, 2016). Processing the consensus of others’ views is the basis on which people decide what kind of information strategies to adopt next (to remain independent or to conform). There are clear social clues in a face-to-face environment that influence the social judgment of the consensus of others, such as group structure and informa- tion source. However, in an online context, the information sources are usually unknown (Nay- lor, Lamberton, & Norton, 2011), with groups of varying size. Taking the feedback “30 people found this helpful” as an example, consumers do not know who these 30 people are, how many lurkers are behind the 30 people, or what their views are. In addition, the highly dispersed opinion in an online context can exacerbate the difficulty of relying on experience in the forma- tion social judgments. Consumers in an online environment face the challenge of learning from multiple individuals whose choices emerge

74 GUO, ZHAO, ZHANG, WEN, AND YIN

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

from unobservable, latent social groups (Gersh- man, Pouncy, & Gweon, 2017). Although Per- fumi et al. (2016) studied social influence in the virtual environment, this study focused only on the neuroresponse differences between partici- pants’ behaviors of conforming to group views and insisting on their own independence in a series of Asch, cultural, and apperceptive tasks. Furthermore, the subjects in the experiment had seen the answers of the other group members before making their own choices, which makes it difficult to distinguish between their personal private views and the views of the others. Thus, we still do not know how users process the views of others with their personal views in an online environment. Furthermore, the subjects in their experiment were studied in relation to a fixed membership of six. Unlike face-to-face environments, online environments are full of digital user views and behaviors, rendering comparisons of the behaviors of net users more convenient and extensive. Exploring how users process and perceive the number representing the views or behaviors of others can help us to understand users’ herd or nonherd behavior in an online environment.

Therefore, this study attempts to use event- related brain potentials (ERPs) to reveal how people process their personal views with others’ views, as represented by a series of numbers in an online context. Social comparison is a com- mon phenomenon in social life (Festinger, 1954). Uncertainty often leads to relevant social comparisons. Previous studies have shown that people working in a virtual environment tend to look for an objective comparison criterion for their personal job performance because it is difficult to find directly comparable objects (such as colleagues; Conner, 2003). In addition, an individual occupying a given social space is more likely to be influenced by the local numer- ical majority than by either the local numerical minority or less proximate persons (Latané, 1996). Therefore, we propose the following hy- pothesis: The uncertainty about others’ opinion arising from a single feedback source of review helpfulness will inspire people to compare the votes of other visible reviews to build an eval- uation criterion of numerical relative majority, which will then be used to evaluate their per- sonal opinion in relation to other opinions. We developed a task in which participants were asked to judge the helpfulness of reviews while

their brain potentials were recorded. After a participant made his or her initial choice, he or she was provided with feedback regarding how many people found each review helpful. The participant’s choice may be consistent or incon- sistent with the relative majority opinion, which allows us to examine how the brain processes this social comparison information. In this con- text, the relative majority opinion means the following: If a review receives a higher number of helpfulness votes than do other visible re- views, the relative majority opinion is that the review is helpful; otherwise the relative major- ity opinion is that the review is unhelpful. Un- like the experiments previously noted, the par- ticipants in this study did not know how many people were involved in the online helpfulness voting or who was involved in the vote.

We focused on feedback-related negativity (FRN) and P300. FRN, a negative-going ERP component that peaks at approximately 250 to 300 ms, with the largest amplitude at the frontocentral recording sites, has been asso- ciated with outcome evaluation and perfor- mance monitoring (Hajcak, Moser, Holroyd, & Simons, 2006; Hajcak, Moser, Holroyd, & Simons, 2007; Kimura, Murayama, Miura, & Katayama, 2013; Masaki, Takeuchi, Gehring, Takasawa, & Yamazaki, 2006). Previous stud- ies have suggested that the FRN amplitude is more pronounced for negative feedback, such as incorrect responses, monetary losses, and viola- tions of expectancy, than for positive feedback (Wu & Zhou, 2009). Importantly, recent studies have found that social norms violations, such as conflict with the group opinion, also elicit more negative-going FRN (Chen, Wu, Tong, Guan, & Zhou, 2012). Based on these studies, we predict that when compared with a participant’s choice being congruent with the relative major- ity opinion in the helpfulness judgment task, a participant’s choice that is incongruent with the relative majority opinion will elicit a greater negative-going FRN response. P300 has been reported as another important ERP component related to outcome evaluation and reward pro- cessing and typically peaks at approximately 300 to 400 ms. Previous studies have shown that P300 was sensitive to the valence and mag- nitude of reward (Wu & Zhou, 2009; Yeung & Sanfey, 2004). Recent studies have extended the role of P300 to the social domain and found that winning more than others (Wu, Zhang,

75ERPS IN ONLINE REVIEW HELPFULNESS EVALUATION TASK

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

Elieson, & Zhou, 2012) or being more attractive than other faces resulted in a greater P300 am- plitude (Werheid, Schacht, & Sommer, 2007). Based on these studies, we predict that a partic- ipant’s congruence with the relative majority opinion in the helpfulness judgment task will be processed as a positive outcome and therefore elicit a greater positive-going P300 response. In addition, previous studies showed a decrease in FRN and P300 amplitudes following learning (Bellebaum & Daum, 2008; Sailer, Fischmeis- ter, & Bauer, 2010). The decline in FRN and P300 amplitudes was thought to be the result of learning that reduced the motivational signifi- cance and attentive processing of the feedback (Sailer et al., 2010). However, unlike the above- mentioned experiments, the participants in our experiment did not know the relative majority of the evaluation criteria before the experiment but learned from the number of votes cast in the experiment from the comparative study. Based on reinforcement learning theory (Berridge, 2012), this evaluation criterion is defined and strengthened in the brain of subjects as the number of trials accumulates. In addition, neg- ative feedback always has more impact than positive ones (Baumeister, Bratslavsky, Finke- nauer, & Vohs, 2001), which indicates that the effect of inconsistency with the relative major- ity does not decline as the effect of consistency with the relative majority does. Therefore, we predicted that the amplitudes of FRN in the early period of the experiment was lower than that in the later period of the experiment, whereas the amplitudes of P300 in the early period of the experiment would be greater than those in the later stage of the experiment.

Method

General Experimental Design

This experiment adopted a one-factor within- subject with a two-level (C1: initial rating of review helpfulness congruent with the relative majority opinion vs. C2: initial rating of review helpfulness incongruent with the relative major- ity opinion) repeated-measure design. The par- ticipants were asked to rate the helpfulness of online reviews (helpful or unhelpful) and were then provided with feedback on the reviews’ helpfulness (how many people found this re- view helpful).

Participants

A total of 32 right-handed students from uni- versities in Wuhan, China (18 women and 14 men, age range 18 –28 years) volunteered and were paid 50 RMB (approximately $8) to par- ticipate in this experiment. All the participants were native Chinese speakers who reported that they had normal or corrected-to-normal vision with no history of neurological or mental dis- ease. In addition, all the participants had more than 1 year of online shopping experience. In- formed consent was obtained from each partic- ipant before the test.

Materials

Online review stimuli were extracted from the top five best-selling lists of computer mice and chocolate on Amazon.cn. The subjects were familiar with the abovementioned product types and had experience with buying the abovemen- tioned products online more than once; thus, they were able to easily complete the helpful- ness rating task. Computer mice and chocolates were used for this online review helpfulness study as typical representatives of search and experience goods (Yu, Zu, & Sun, 2016). Re- views that have been posted recently and con- tained more than 15 words were chosen (10 positive and 10 negative for each item) as the stimuli material set (total 200). To eliminate the influence of social information on the percep- tion of review helpfulness, rating (score), help- fulness votes, and brand information were re- moved from each review. In addition, some content was removed from longer reviews to ensure the relative consistency of review lengths without breaking the complete seman- tics of a given reserved topic. Furthermore, two postgraduate students in e-commerce reclassi- fied the valence of the reviews independently, and only the reviews with consistent categori- zation were included in the study. We obtained 128 final reviews. There were 64 positive and negative reviews each, and the mean length of the reviews was 25 words (SD � 5.05). A total of 10 product pictures (computer mice: 5; choc- olate: 5) were chosen from Amazon.cn and dig- itized at 500 � 500 pixels with the same gray background to match the reviews.

The feedback on review helpfulness adopted the style of Amazon.cn: “n people found this

76 GUO, ZHAO, ZHANG, WEN, AND YIN

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

helpful,” where the “n” was manipulated to be between 30 and 40 (numerical relative majority) or 10 to 20 (numerical relative minority) by the software E-Prime (PST, Psychology Software Tools, Inc., Sharpsburg, Maryland). Therefore, the stimuli consisted of 10 pictures (S1) � 128 reviews (S2) � 2 categories of review helpful- ness feedback (S3). The “n” was chosen for the following reasons: (a) We focused on the rela- tive size of votes compared with the votes of other visible reviews, and it was therefore nec- essary to ensure that “n” would not be processed directly as a numerical majority or minority; (b) there should be visible differences in the help- fulness votes between the numerical relative majority and the relative minority; (c) in refer- ence to the mean helpfulness of 15,095 reviews (76.98 � 25.63; Baek et al., 2015), selecting an “n” value that is half of the mean helpfulness was appropriate for a numerical relative major- ity; and (d) a pretest in 10 students showed that they did not directly take the two groups of numbers as the numerical majority or minority and that there was a good degree of identifica- tion between the two groups of numbers.

Procedure

During the experiment, the participant sat in a comfortable chair in a shielded room. The par- ticipant was then instructed to read the intro- duction to the experiment and focus on the stimuli while avoiding eyeblinks or movement of the eyes and head. Stimulus presentation and behavioral response collection were controlled by the E-Prime software. The stimulus (black on a white background) was presented in the center of a 22-in. computer monitor, with a visual angle of 2.58 � 2.4.

The experiment procedure is depicted in Fig- ure 1. Each trial began with a red cross (“�”), which appeared for 1,000 ms, as a fixation point. After a 500-ms blank screen, a picture of the product (S1) was shown for 1,000 ms be- cause we wanted the participant to focus on the review evaluation rather than the product. After another 500 ms blank screen, a review of the product (S2) was randomly presented and then disappeared upon the participant’s decision. The participant was asked to decide as quickly as possible whether the review was helpful. After showing the blank screen for another 500 ms, feedback on review helpfulness (S3) was presented for 2,000 ms. The helpfulness feed- back was predetermined by a program without the participant’s knowledge, and the two feed- back categories were randomly assigned.

The experiment consisted of 200 trials di- vided into four blocks, and the sequence of trials in each block was randomly assigned. There was a 2-min interval after each block for participants to rest. A practice block of 10 trials was assigned to familiarize the participants with the experiment before the formal test. The entire experiment lasted approximately 30 to 40 min.

Electroencephalogram Recordings and ERP Data Processing

Electroencephalogram (EEG) was continu- ously recorded (bandpass 0.05–100 Hz; sam- pling rate 1,000 Hz (Chen et al., 2012; Liu et al., 2013) with an Eego Amplifier (ANT Neuro, Inc., Hengelo, the Netherlands) using an elec- trode cap with 32 Ag/AgCl electrodes mounted according to the extended International 10 –20 system and referenced to the mastoids. Elec- trode impedance was kept below 10 k�

Figure 1. Sequence of events in a single trial. See the online article for the color version of this figure.

77ERPS IN ONLINE REVIEW HELPFULNESS EVALUATION TASK

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

throughout the experiment (Liu et al., 2013). E-Prime was used to collect all the behavioral responses including participant choices and re- sponse times.

Offline data processing was performed using ASA software (ANT Software BV, Enschede, the Netherlands). The continuous EEG was re- referenced to the average of the right and left mastoids and then digitally filtered with a high- pass filter at 0.1 Hz and a low-pass filter at 30 Hz (24 dB/octave; Chen et al., 2012; Liu et al., 2013). Subsequently, electrooculogram artifacts were corrected using the ASA software. After artifact correction, the data were segmented into 1,000-ms stimulus-locked epochs from �200 ms (before S3 onset) to 800 ms (after S3 onset), with the first 200 ms of prestimulus as a base- line. Epochs with a deflection exceeding � 80 �V were excluded from the averaging (Chen et al., 2010, 2012). The remaining epochs were then averaged for each participant and each condition (C1, initial rating on review helpful- ness congruent with the relative majority opin- ion; C2, initial rating on review helpfulness

incongruent with the relative majority opinion) to produce the ERP waveforms. Subsequently, the ERP waveforms were corrected to baseline (�200 ms-0) and then grand-averaged across all the participants in each condition to produce grand-averaged ERP waveforms. To investigate the neurophysiologic factors correlating with the processing of different review helpfulness evaluation consistency categories, we compared the amplitudes of the FRN and P300 using a within-subject repeated-measure analysis of variance (ANOVA).

For the FRN, the largest amplitude was typ- ically located at the frontocentral recording sites (Hajcak et al., 2006, 2007; Masaki et al., 2006); thus, the electrode sites of Fz, FC1, FC2, and Cz were selected for further analysis. According to the FRN latency and upon waveform visual inspection (Figure 2), the mean amplitudes of Fz, FC1, FC2, and Cz in the 250 to 350 ms time window were analyzed. For P300, the maxi- mum amplitude was reported at parietal sites (Wu & Zhou, 2009; Yeung & Sanfey, 2004); therefore, the electrode sites of CP1, CP2, and

Figure 2. Grand-averaged event-related brain potential waveforms for two conditions: C1 and C2 at electrode sites: Fz, FC1, FC2, and Cz (�V/ms). See the online article for the color version of this figure.

78 GUO, ZHAO, ZHANG, WEN, AND YIN

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

Pz were selected for further analysis. According to the P300 latency and waveform visual in- spection (Figure 3), the mean amplitudes of CP1, CP2, and Pz in the 300 to 400 ms time window were analyzed. Then, we performed a within-subject repeated-measure ANOVA of the mean amplitudes of FRN and P300 using the software SPSS 22.0 (SPSS Inc., Chicago, Illi- nois). The Greenhouse-Geisser correction for violation of the assumption of sphericity was applied, and the Bonferroni correction was used for multiple comparisons.

Results

Manipulation Check

A paired t test was used to compare the trial numbers among positive reviews (M � 100.28 � 2.29) and negative reviews (M � 99.72 � 2.29) for each participant. The difference between positive and negative reviews was not signifi- cant (t � .695, p .05). The trial numbers of each condition (conformity feedback, M � 92.03 � 6.34, vs. nonconformity feedback, M � 107.97 � 6.34) for each participant were counted; all the trial numbers of each condition

for each participant were more than 30, which met the requirements for an ERP experiment (Luck, 2005).

Behavioral Data

A total of 6,400 behavioral data (200 items per subject, including reviews helpfulness rat- ing and response times) were recorded by E- Prime. For each participant, response times greater than � 2 SDs from the mean in each condition were excluded from the helpfulness review rating and response time analyses (Liu et al., 2013). Finally, 6,120 behavioral data were retained (280 items were removed).

A paired t test on the helpful review rates between positive (M � 0.77 � 0.11) and neg- ative reviews (M � 0.69 � 0.13) for each participant showed that the helpful review rat- ings for positive reviews were significantly higher than were those for negative reviews (t � 3.455, p .01), indicating that participants more often rated positive reviews to be helpful than they did negative reviews. Interestingly, a paired t test on response times for positive re- views (M � 2518.52 � 840) and negative re- views (M � 2771.55 � 1154.88) found that the

Figure 3. Grand-averaged event-related brain potential waveforms for two conditions: C1 and C2 at electrode sites: CP1, CP2, and Pz (�V/ms). See the online article for the color version of this figure.

79ERPS IN ONLINE REVIEW HELPFULNESS EVALUATION TASK

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

response time for negative reviews was signif- icantly higher than that for positive reviews (t � 3.619, p .01), indicating that participants responded faster to positive reviews (Table 1).

To further compare the differences in re- sponse times for the early and late phase con- ditions (Bellebaum & Daum, 2008; Kraus & Horowitz-Kraus, 2014; Sailer et al., 2010), we divided each subject’s trials into two parts ac- cording to the presentation sequence of the trials (T1 and T2, each part for 100 trials). The results of the paired t test on response time for T1 (M � 2642.38 � 905.96) and T2 (M � 2589.63 � 1117.09) showed that the differences between the early phase and the late phase were not significant.

Feedback-Related Negativity

As shown in Figure 2, an obvious negative deflection was elicited with a peak at approxi- mately 300 ms. A two-factor 2 (condition: con- formity feedback and nonconformity feed- back) � 4 (frontocentral: Fz, FC1, FC2, and Cz) within-subjects repeated-measure ANOVA of the mean ERP amplitude between 250 and 350 ms showed that the main effect of condition

(conformity feedback and nonconformity feed- back) was significant, F(1, 31) � 18.112, p .001. This result indicates that the mean ERP amplitude across the four electrodes in the time window for nonconformity feedback was sig- nificantly different from that for conformity feedback. A paired t test on the mean ERP amplitude between the conditions of conformity feedback and nonconformity feedback found that the negative amplitude of the conformity feedback condition was significantly smaller than that of the nonconformity feedback condi- tion (t � 4.256, p .001), indicating that the nonconformity feedback condition elicited more negative-going FRN than did the confor- mity feedback condition. The main effect of electrode location was also significant, F(3, 93) � 9.671, p .001, but the interaction effect between the two factors was not significant, F(3, 93) � 1.224, p .05; this finding indi- cates that the amplitudes were significantly different among the four electrodes but that the main effect of condition (conformity feed- back, nonconformity feedback) was not af- fected by the electrode position. Subsequent paired t tests on mean ERP amplitude among

Table 1 Mean Helpful Reviews Rate and Mean Response Time Across Two Conditions

Condition Helpful reviews rate (%) SD t value Response time (ms) SD t value

Positive review 76.75 .11 3.455�� 2,518.52 840.63 �3.619��

Negative review 69.10 .13 2,771.55 1,154.88

Note. Helpful reviews rate is the percentage of “helpful” choices among all valid choices. �� p .01.

Table 2 Mean Feedback-Related Negativity and Planned Contrasts in Time Window of 250 to 350 ms (�V)

Electrode site

C1: Congruent with the relative majority

opinion

C2: Incongruent with the relative majority opinion

t valueM SD M SD

Fz �0.39 2.31 �1.00 2.14 2.729�

FC1 �0.51 2.20 �1.26 2.01 3.693��

FC2 1.00 2.64 0.09 2.44 5.581���

Cz �0.14 2.59 �0.97 2.07 3.524��

Frontocentral �0.01 2.20 �0.79 1.90 4.256���

|C1| |C2|, F � 18.112, p � .000

� p .05. �� p .01. ��� p .001.

80 GUO, ZHAO, ZHANG, WEN, AND YIN

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

the abovementioned four electrodes showed that the greatest main effect of condition (conformity feedback vs. nonconformity feedback) was on FC2 (Table 2).

To further compare the differences in the FRN between the early and late periods (Belle- baum & Daum, 2008; Kraus & Horowitz- Kraus, 2014; Sailer et al., 2010), we divided each subject’s trials into two parts according to the presentation sequence of the trials (T1 and T2, each part for 100 trials). A three-factor 2 (condition: conformity feedback and noncon- formity feedback) � 2 (period: T1 and T2) � 4 (frontocentral: Fz, FC1, FC2, and Cz) within- subjects repeated-measure ANOVA of the mean ERP amplitude in the interval 250 to 350 ms showed that the main effect of period (T1 and T2) was significant, F(1, 31) � 4.472, p .05. This result indicates that the mean ERP amplitude across the four electrodes in the early period was significantly different from that in the later period. A further paired t test on the mean ERP amplitude between T1 and T2 found that, except for site FC2, the negative ampli- tudes of T1 for nonconformity feedback at sites Fz (t � 3.255, p .01), FC1 (t � 2.092, p .05), and Cz (t � 2.272, p .05) were signif- icantly smaller than those for T2. However, all differences for conformity feedback between

T1 and T2 across the four sites were not signif- icant, indicating that the nonconformity feed- back condition in the later period elicited more negative-going FRN than did the nonconfor- mity feedback condition in the early period at sites Fz, FC1, and Cz (Table 3).

P300

In Figure 3, an obvious positive deflection can be observed near 350 ms. A two-factor 2 (condition: conformity feedback and noncon- formity feedback) � 3 (parietal: CP1, CP2, and Pz) within-subjects repeated-measure ANOVA on the mean ERP amplitude between 300 and 400 ms was performed. The results showed that the main effect of condition (conformity feed- back and nonconformity feedback) was signifi- cant, F(1, 31) � 17.776, p .001, along with a significant main effect of electrode location, F(2, 62) � 4.368, p .05, but the interaction effect between condition and location was not significant, F(2, 62) � 0.193, p .05. This result indicates that the main effect of condition (conformity feedback vs. nonconformity feed- back) was significant and was not affected by electrode location. A paired t test on the mean ERP amplitude between the conformity condi- tion and the nonconformity feedback condition revealed that the amplitude of the conformity feedback condition was significantly greater than that of the nonconformity feedback condi- tion (t � 4.216, p .001), indicating that the conformity feedback condition elicited more positive-going P300 than did the nonconformity feedback condition. Further paired t tests on the mean ERP amplitude among the above- mentioned three electrodes showed that the largest main effect of condition (conformity feedback vs. nonconformity feedback) was on CP2 (Table 4).

To further compare the differences of FRN between the experimental periods, a three-factor 2 (condition: conformity feedback and noncon- formity feedback) � 2 (period: T1 and T2) � 3 (parietal: CP1, CP2, and Pz) within-subjects repeated-measure ANOVA of the mean ERP amplitude between 300 and 400 ms was con- ducted; the results showed that the main effect of period (T1 vs. T2) was significant, F(1, 31) � 4.619, p .05. This result indicates that the mean ERP amplitude across the four elec- trodes in the early period was significantly dif-

Table 3 Mean Feedback-Related Negativity and Planned Contrasts Between T1 and T2 in Time Window of 250 to 350 ms (�V)

Electrode site

T1: First half trials

T2: Second half trials

t valueM SD M SD

Fz C1 �.36 2.70 �0.31 2.17 �0.152 C2 �.55 2.43 �1.42 2.11 3.255��

FC1 C1 �.24 2.31 �0.72 2.49 1.364 C2 �.96 2.18 �1.57 2.13 2.092�

FC2 C1 .93 2.92 1.23 2.58 �0.985 C2 .27 2.60 0.03 2.30 1.011

Cz C1 .15 2.69 �0.31 2.83 1.257 C2 �.58 1.96 �1.25 2.29 2.272�

Note. C1 � congruent with the relative majority opinion; C2 � incongruent with the relative majority opinion. � p .05. �� p .01.

81ERPS IN ONLINE REVIEW HELPFULNESS EVALUATION TASK

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

ferent from that in the later period. Further paired t test on the mean ERP amplitude be- tween T1 and T2 found that, except for site CP2, the amplitudes of T1 for the conformity feedback condition at sites CP1(t � 2.395, p .05) and Pz (t � 2.362, p .05) were signifi- cantly larger than those of T2; however, all differences for conformity feedback between T1 and T2 across the three sites were not sig- nificant, indicating that the conformity feedback condition in the early period elicited more pos- itive-going P300 than did the conformity feed- back condition in the later period at sites CP1 and Pz (Table 5).

Discussion

The behavioral results of the helpfulness re- view rating task indicated that participants rated

positive reviews as more helpful than negative reviews. Negativity bias is a widespread and widely recognized phenomenon. However, re- search conclusions on this topic in the context of online reviews have been inconsistent. Sev- eral previous studies found that negative re- views were more helpful than positive reviews (Cao et al., 2011; Lee et al., 2017), whereas other studies suggested that negative reviews were not more helpful than positive reviews (Mudambi & Schuff, 2010; Sen & Lerman, 2007; Wu, 2013). Interestingly, participants re- sponded significantly more rapidly to positive reviews than to negative reviews. Previous stud- ies have shown that response time was posi- tively associated with cognitive load (Cowen, Ball, & Delin, 2002; Sweller, 1988). The un- derlying causes of negativity bias on reviews include the findings that negative reviews car- ried greater surprise value and the ability to avoid losses (Yin, Mitra, & Zhang, 2012). The surprise value of negative reviews may more easily capture people’s attention (Carretié, Mer- cado, Tapia, & Hinojosa, 2001) but cannot guarantee the perceived value of negative re- views (Chen et al., 2010). The positive and negative reviews were manipulated to be ran- domly presented with the same probability in our experiment; thus, the influence novelty of negative reviews would be weakened. Negative reviews are typically associated with risks, which may cause the participants to spend more time assessing potential risks. In a laboratory environment, it is difficult to guarantee the par- ticipants’ purchase motivation; thus, the value of negative reviews on avoiding losses might be discounted. In contrast, the relatively low cog- nitive load of positive reviews means that sub-

Table 4 Mean P300 and Planned Contrasts in Time Window of 300 to 400 ms (�V)

Electrode site

C1: Congruent with the relative majority opinion

C2: Incongruent with the relative majority opinion

t valueM SD M SD

CP1 1.06 2.94 �.07 2.66 3.970���

CP2 1.75 3.05 .58 2.33 4.673���

Pz 1.14 3.01 .05 2.69 3.750��

Across three sites 1.32 2.88 .19 2.43 4.216���

|C1| |C2|, F � 17.776, p � .000

�� p .01. ��� p .001.

Table 5 Mean P300 and Planned Contrasts Between T1 and T2 in Time Window of 300 to 400 ms (�V)

Electrode site

T1: First half trials

T2: Second half trials

t valueM SD M SD

CP1 C1 1.57 2.95 0.71 3.27 2.395�

C2 0.08 2.75 �0.33 2.85 1.455 CP2

C1 2.06 3.13 1.63 3.15 1.450 C2 0.72 2.16 0.50 2.51 0.987

Pz C1 1.62 2.90 0.81 3.41 2.362�

C2 0.03 2.76 �0.09 2.84 0.445

Note. C1 � congruent with the relative majority opinion; C2 � incongruent with the relative majority opinion. � p .05.

82 GUO, ZHAO, ZHANG, WEN, AND YIN

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

jects do not require excessive cognitive effort to make decisions about a review’s helpfulness, which may explain why positive reviews re- ceived a higher rate of “helpful” votes. In addi- tion, the lack of other social clues in the online review helpfulness task might have caused the participants to perceive less social pressure. Previous studies have shown that negativity bias emerges in the context of public opinions but not in private opinions or thoughts (Schlosser, 2005). Consequently, the response time together with the behavioral results of the review helpfulness ratings indicates that nega- tivity bias in the context of reviews may not lie in the perception of a review’s helpfulness but rather in the cognitive processing of such re- views. This issue clearly requires further inves- tigation.

FRN is an ERP component that is closely related to outcome evaluation. Similar to the results of a study on group opinion (Chen et al., 2012), a significant difference in FRN ampli- tude was observed in the present study. The stimulus condition of inconsistency with the relative majority opinion elicited a greater FRN response than did consistency with the relative majority opinion. In the studies of Chen et al. (2012) and Perfumi et al. (2016), all subjects were in a group with fixed members; thus they were able to judge the degree of the consistency of group opinion directly. However, in our study, it was difficult for participants to judge the consensus opinions of others directly from a vote on a review. However, significant differ- ences in FRN amplitude were also observed in the present experiment, indicating that conflicts between personal views and the views of others can be detected by the brain of the participant in this context. That is, participants can build an evaluation criterion for the number of votes that can represent the relative consensus of others by learning from a comparison among more than one visible voting number. Differences in par- ticipants’ performance between the early and late phases of a learning task were often used to test the learning effect (Bellebaum & Daum, 2008; Kraus & Horowitz-Kraus, 2014; Sailer et al., 2010). The results of a comparison of the FRN amplitude between the different periods of the trials showed that the nonconformity feed- back condition in the later period elicited more negative-going FRN than did the nonconfor- mity feedback condition in the early period,

confirming our inference. Because the definite evaluation criteria of the number of votes had not yet been fully formatted in the early period of the experiment, the stimulus condition of inconsistency with the relative majority opinion elicited a smaller negative-going FRN. How- ever, as the number of trials increased, this criterion was strengthened continuously; thus, a greater negative-going FRN was observed in the later period of the experiment. However, there was no significant difference in the re- sponse time between the different periods of the trials, indicating that the learning effect was not on the speed of processing. Therefore, we con- cluded that the brain establishes a criterion for judging the relative majority opinion by learn- ing from a comparison of the visible votes of other reviews. The results showed that the brain could automatically encode the relative major- ity opinion by comparing the visible votes and could detect whether one’s personal choice is consistent with the relative majority opinion. Because social pressure in online environments is much less than that in face-to-face environ- ments, the FRN response elicited by inconsis- tency with the relative majority opinion should be smaller in online environments than in face- to-face environments. However, it is interesting that the mean amplitudes of FRN in the current experiment, mean (Fz, Cz, Pz) � �4.41 � 1.62 for consistent with the relative majority opinion and �4.47 � 2.02 for inconsistent with the relative majority opinion, were far lower than those in the face-to-face experiment environ- ment of Chen et al. (2012), mean (Fz, FCz, Cz, CPz, Pz) � 3.98 � 1.13 for highly incongruent trials, 5.72 � 1.07 for moderately incongruent trials, and 8.56 � 1.13 for congruent trials, indicating that the FRN response was much greater in the online environment than in the face-to-face experiment environment of Chen et al. (2012). Differences in the experimental task and environment may be reasons for the above- mentioned phenomenon. Compared with the line judgment task of Chen et al. (2012), the task of judging review helpfulness was more subjective and ambiguous. Therefore, the sub- jects might have paid more attention to the views of others based on the need for accurate cognition (Cialdini & Goldstein, 2004), thus allowing them to be more influenced by the informational influence of others’ views (vote of helpfulness). Second, compared with the

83ERPS IN ONLINE REVIEW HELPFULNESS EVALUATION TASK

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

face-to-face environment, group sizes in the on- line environment are much larger, and the dig- ital visualization of users’ opinions or behaviors make comparing the views or behaviors of oth- ers more convenient and extensive. Together, these factors resulted in a greater FRN response when processing the votes of online reviews. At the same time, the normative influence of the online environment is much weaker than that of face-to-face environments (Perfumi et al., 2016); thus subjects in online environments will not worry more about whether their views are consistent with others’ views, as in face-to-face environments. Therefore, the differences in FRN amplitudes between stimuli conditions should not be greater than the differences in the face-to-face environments. In a comparison of the main effect of stimulus condition, F(1, 31) � 18.112, p .001, Chen et al. (2012), F(2, 36) � 64.57, p .001 showed that the effect of condition on differences in FRN response for the online environment was smaller than that for the face-to-face environment, consistent with our assumption.

In the current study, P300 was a slightly late ERP component following FRN and could elicit significantly enhanced deflection in the condi- tion of consistency with the plurality opinion. P300 has been associated with outcome evalu- ation, reward processing, and selective attention (Wu & Zhou, 2009; Yeung & Sanfey, 2004). As shown in a previous study, FRN was sensitive to feedback valence, whereas P300 was sensi- tive to reward magnitude (Schindler & Bickart, 2012). Therefore, FRN and P300 are thought to be responsible for encoding different aspects of outcome evaluation. In addition, other studies found that P300 was related to participants’ expectations (Hajcak et al., 2006, 2007; Kimura et al., 2013). Compared with being given ex- pected results, feedback beyond expectations elicited more positive P300 amplitudes. Be- cause the present study contained only two lev- els randomly presented for review helpfulness feedback, the probability of inconsistency with the relative majority opinion should be the same as the probability of consistency with the rela- tive majority opinion. Thus, the factors of ex- pectation and magnitude that can affect P300 amplitude could be ignored in this context. Therefore, the P300 effect in the current exper- iment is likely to be primarily related to the valence of feedback. Consistency with the rel-

ative majority opinion may be encoded as pos- itive outcomes (rewards, such as social ap- proval) by the nervous system. Furthermore, the results of comparing P300 amplitudes between the different periods of the trials show that the P300 response elicited by the trials of consis- tency with the relative majority opinion in the early period is significantly greater than that in the later period, indicating that the marginal effect of consistency with relative majority de- clines as the number of trials increases.

Together, the FRN and P300 results indicate that conflict with the relative majority opinion of online review helpfulness triggered a cascade of neuronal responses including an earlier FRN response monitoring the violation of relative majority opinions and a later P300 response differentiating positive from negative feedback. Furthermore, the FRN response elicited by the trials of inconsistency with relative majority opinions in the early period were smaller than those in the later period, whereas the P300 response elicited by the trials of consistency with relative majority opinions in the early pe- riod were greater than those in the later period. The results confirmed that in an online environ- ment in which it was difficult to clearly judge the consensus views of others, the brain could automatically encode relative majority opinions by learning from a comparison of other visible social cues and automatically categorize whether one’s personal views are consistent with those of the relative majority. In addition, although the normative influence of online en- vironments is weaker than that of face-to-face environments, the advantages of convenient comparison among users’ behaviors and the scale of net users strengthen the informational influence of net user behavior.

To meet the needs of a controlled experimen- tal, the editing of stimuli materials and their presentation resulted in the loss of some impor- tant information. In addition, it was difficult to guarantee the purchase motivation of the partic- ipants; thus, the personal relevance of a review may have affected participant attitudes toward the tasks and, in turn, their judgment and ERP responses to the reviews’ helpfulness. Further- more, the previous task may affect the FRN response of the latter task (Schmidt et al., 2017). These factors may weaken the generalizability of these results to the real world. In addition, the comparison of these results with those of Chen

84 GUO, ZHAO, ZHANG, WEN, AND YIN

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

et al. (2012) may be influenced by differences in the tasks and recording instruments used; thus, the results need to be further verified. Although the current study revealed how the brain processes personal views with others’ views in an online context, the information strategies (to remain independent or to con- form) that could be subsequently adopted war- rant further study. Furthermore, in contrast to helpfulness votes, the opinions presented by online reviews are more complex, with a single review often mixing positive and negative views from multiple dimensions. A previous study showed that people’s attribution for dis- persion in online reviews was affected by the perceived taste of the product (He & Bond, 2015). How we cognitively process the contra- dictory and complex views of others in an on- line environment remains a very important top- ic. In addition, there are many types of information available online, such as sales and total reviews. How we understand these social clues and use them for social judgment or de- cision making remains to be fully understood.

References

Baek, H., Lee, S., Oh, S., & Ahn, J. H. (2015). Normative social influence and online review help- fulness: Polynomial modeling and response sur- face analysis. Journal of Electronic Commerce Research, 16, 291–306.

Baumeister, R. F., Bratslavsky, E., Finkenauer, C., & Vohs, K. D. (2001). Bad is stronger than good. Review of General Psychology, 5, 323–370. http:// dx.doi.org/10.1037/1089-2680.5.4.323

Bellebaum, C., & Daum, I. (2008). Learning-related changes in reward expectancy are reflected in the feedback-related negativity. European Journal of Neuroscience, 27, 1823–1835. http://dx.doi.org/10 .1111/j.1460-9568.2008.06138.x

Berridge, K. C. (2012). From prediction error to incentive salience: Mesolimbic computation of re- ward motivation. European Journal of Neurosci- ence, 35, 1124–1143. http://dx.doi.org/10.1111/j .1460-9568.2012.07990.x

Cao, Q., Duan, W. J., & Gan, Q. W. (2011). Explor- ing determinants of voting for the “helpfulness” of online user reviews: A text mining approach. De- cision Support Systems, 50, 511–521. http://dx.doi .org/10.1016/j.dss.2010.11.009

Carretié, L., Mercado, F., Tapia, M., & Hinojosa, J. A. (2001). Emotion, attention, and the ‘negativity bias’, studied through event-related potentials. Interna-

tional Journal of Psychophysiology, 41, 75– 85. http://dx.doi.org/10.1016/S0167-8760(00)00195-1

Chen, J., Wu, Y., Tong, G., Guan, X., & Zhou, X. (2012). ERP correlates of social conformity in a line judgment task. BMC Neuroscience, 13, 43. http://dx.doi.org/10.1186/1471-2202-13-43

Chen, M., Ma, Q., Li, M., Dai, S., Wang, X., & Shu, L. (2010). The neural and psychological basis of herding in purchasing books online: An event- related potential study. Cyberpsychology, Behav- ior, and Social Networking, 13, 321–328. http://dx .doi.org/10.1089/cyber.2009.0142

Cialdini, R. B., & Goldstein, N. J. (2004). Social influence: Compliance and conformity. Annual Re- view of Psychology, 55, 591– 621. http://dx.doi .org/10.1146/annurev.psych.55.090902.142015

Conner, D. S. (2003). Social comparison in virtual work environments: An examination of contempo- rary referent selection. Journal of Occupational and Organizational Psychology, 76, 133–147. http://dx.doi.org/10.1348/096317903321208925

Cowen, L., Ball, L. J. S., & Delin, J. (2002). An eye movement analysis of web page usability. Inter- face, 28, 317–335.

Festinger, L. (1954). A theory of social comparison processes. Human Relations, 7, 117–140. http://dx .doi.org/10.1177/001872675400700202

Gershman, S. J., Pouncy, H. T., & Gweon, H. (2017). Learning the structure of social influence. Cogni- tive Science, 41, 545–575. http://dx.doi.org/10 .1111/cogs.12480

Hajcak, G., Moser, J. S., Holroyd, C. B., & Simons, R. F. (2006). The feedback-related negativity re- flects the binary evaluation of good versus bad outcomes. Biological Psychology, 71, 148–154. http://dx.doi.org/10.1016/j.biopsycho.2005.04.001

Hajcak, G., Moser, J. S., Holroyd, C. B., & Simons, R. F. (2007). It’s worse than you thought: The feedback negativity and violations of reward pre- diction in gambling tasks. Psychophysiology, 44, 905–912. http://dx.doi.org/10.1111/j.1469-8986 .2007.00567.x

He, S. X., & Bond, S. D. (2015). Why is the crowd divided? Attribution for dispersion in online word of mouth. Journal of Consumer Research, 41, 1509–1527. http://dx.doi.org/10.1086/680667

Hong, H., Xu, D., Wang, G. A., & Fan, W. (2017). Understanding the determinants of online review helpfulness: A meta-analytic investigation. Deci- sion Support Systems, 102, 1–11. http://dx.doi.org/ 10.1016/j.dss.2017.06.007

Karimi, S., & Wang, F. (2017). Online review help- fulness: Impact of reviewer profile image. Deci- sion Support Systems, 96, 39– 48. http://dx.doi.org/ 10.1016/j.dss.2017.02.001

Kimura, K., Murayama, A., Miura, A., & Katayama, J. (2013). Effect of decision confidence on the evaluation of conflicting decisions in a social con-

85ERPS IN ONLINE REVIEW HELPFULNESS EVALUATION TASK

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

text. Neuroscience Letters, 556, 176–180. http://dx .doi.org/10.1016/j.neulet.2013.09.020

Kraus, D., & Horowitz-Kraus, T. (2014). The effect of learning on feedback-related potentials in ado- lescents with dyslexia: An EEG-ERP study. PLoS ONE, 9, e100486. http://dx.doi.org/10.1371/ journal.pone.0100486

Latané, B. (1996). Dynamic social impact: The cre- ation of culture by communication. Journal of Communication, 46, 13–25. http://dx.doi.org/10 .1111/j.1460-2466.1996.tb01501.x

Lee, M., Jeong, M., & Lee, J. (2017). Roles of negative emotions in customers’ perceived help- fulness of hotel reviews on a user-generated re- view website: A text mining approach. Interna- tional Journal of Contemporary Hospitality Management, 29, 762–783. http://dx.doi.org/10 .1108/IJCHM-10-2015-0626

Liu, X., Liao, Y., Zhou, L., Sun, G., Li, M., & Zhao, L. (2013). Mapping the time course of the positive classification advantage: An ERP study. Cognitive, Affective and Behavioral Neuroscience, 13, 491– 500. http://dx.doi.org/10.3758/s13415-013-0158-6

Luck, S. J. (2005). An introduction to the event- related potential technique. Cambridge, MA: MIT Press.

Masaki, H., Takeuchi, S., Gehring, W. J., Takasawa, N., & Yamazaki, K. (2006). Affective-motiva- tional influences on feedback-related ERPs in a gambling task. Brain Research, 1105, 110–121. http://dx.doi.org/10.1016/j.brainres.2006.01.022

Moe, W. W., & Schweidel, D. A. (2012). Online product opinions: Incidence, evaluation, and evo- lution. Marketing Science, 31, 372–386. http://dx .doi.org/10.1287/mksc.1110.0662

Mudambi, S. M., & Schuff, D. (2010). What makes a helpful online review? A study of customer re- views on amazon.com. Social Science Electronic Publishing, 34, 185–200. http://dx.doi.org/10 .2307/20721420

Naylor, R. W., Lamberton, C. P., & Norton, D. A. (2011). Seeing ourselves in others: Reviewer am- biguity, egocentric anchoring, and persuasion. Journal of Marketing Research, 48, 617– 631. http://dx.doi.org/10.1509/jmkr.48.3.617

Perfumi, S. C., Cardelli, C., Bagnoli, F., & Guazzini, A. (2016). Conformity in virtual environments: A hybrid neurophysiological and psychosocial ap- proach. International Conference on Internet Sci- ence (pp. 148–157). Florence, Italy: Springer In- ternational Publishing.

Sailer, U., Fischmeister, F. P. S., & Bauer, H. (2010). Effects of learning on feedback-related brain po- tentials in a decision-making task. Brain Research, 1342, 85–93. http://dx.doi.org/10.1016/j.brainres .2010.04.051

Schindler, R. M., & Bickart, B. (2012). Perceived helpfulness of online consumer reviews: The role

of message content and style. Journal of Consumer Behaviour, 11, 234–243. http://dx.doi.org/10.1002/cb .1372

Schlosser, A. E. (2005). Posting versus lurking: Communicating in a multiple audience context. Journal of Consumer Research, 32, 260–265. http://dx.doi.org/10.1086/432235

Schmidt, B., Mussel, P., Osinsky, R., Rasch, B., Debener, S., & Hewig, J. (2017). Work first then play: Prior task difficulty increases motivation- related brain responses in a risk game. Biological Psychology, 126, 82– 88. http://dx.doi.org/10 .1016/j.biopsycho.2017.04.010

Sen, S., & Lerman, D. (2007). Why are you telling me this? An examination into negative consumer reviews on the web. Journal of Interactive Mar- keting, 21, 76–94. http://dx.doi.org/10.1002/dir .20090

Sweller, J. (1988). Cognitive load during problem solving: Effects on learning. Cognitive Science, 12, 257–285. http://dx.doi.org/10.1207/s15516709cog1202_4

Thomas, V. L., & Vinuales, G. (2017). Understand- ing the role of social influence in piquing curiosity and influencing attitudes and behaviors in a social network environment. Psychology and Marketing, 34, 884– 893. http://dx.doi.org/10.1002/mar.21029

Wang, Z. S., Li, H. Y., & Sun, R. (2016). Determi- nants of votes of helpfulness for Chinese online customer reviews: A moderating effect of product type. Management Review, 28, 143–153.

Werheid, K., Schacht, A., & Sommer, W. (2007). Facial attractiveness modulates early and late event-related brain potentials. Biological Psycho- logy, 76, 100–108. http://dx.doi.org/10.1016/j .biopsycho.2007.06.008

Wu, H., Luo, Y., & Feng, C. (2016). Neural signa- tures of social conformity: A coordinate-based ac- tivation likelihood estimation meta-analysis of functional brain imaging studies. Neuroscience and Biobehavioral Reviews, 71, 101–111. http://dx .doi.org/10.1016/j.neubiorev.2016.08.038

Wu, P. F. (2013). In search of negativity bias: An empirical study of perceived helpfulness of online reviews. Psychology and Marketing, 30, 971–984. http://dx.doi.org/10.1002/mar.20660

Wu, Y., Zhang, D., Elieson, B., & Zhou, X. (2012). Brain potentials in outcome evaluation: When so- cial comparison takes effect. International Journal of Psychophysiology, 85, 145–152. http://dx.doi .org/10.1016/j.ijpsycho.2012.06.004

Wu, Y., & Zhou, X. (2009). The P300 and reward valence, magnitude, and expectancy in outcome evaluation. Brain Research, 1286, 114–122. http:// dx.doi.org/10.1016/j.brainres.2009.06.032

Yang, Y., Chen, C., & Bao, F. S. (2017). Aspect- based helpfulness prediction for online product reviews. IEEE, International Conference on

86 GUO, ZHAO, ZHANG, WEN, AND YIN

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

TOOLS with Artificial Intelligence (pp. 836– 843). San Jose, CA: IEEE.

Yeung, N., & Sanfey, A. G. (2004). Independent coding of reward magnitude and valence in the human brain. The Journal of Neuroscience, 24, 6258– 6264. http:// dx.doi.org/10.1523/JNEUROSCI.4537-03.2004

Yin, D., Mitra, S., & Zhang, H. (2012). Mechanisms of negativity bias: An empirical exploration of app reviews in apple’s app store. In Proceedings of the 33rd International Conference on Information Sys- tems, Orlando, FL.

Yu, W. P., Zu, X., & Sun, Y. B. (2016). Online review impact on heterogeneous consumer pur-

chase intention based on product category. Journal of Dalian University of Technology, 37, 1–5.

Zhao, Y., Ni, Q., & Zhou, R. X. (2018). What factors influence the mobile health service adoption? A meta-analysis and the moderating role of age. In- ternational Journal of Information Management, 43, 342–350. http://dx.doi.org/10.1016/j.ijinfomgt .2017.08.006

Received January 28, 2018 Revision received December 3, 2018

Accepted December 17, 2018 �

Correction to Murphy (2016)

In the article, “Kissing Babies to Signal You Are Not a Psychopath,” by Ryan H. Murphy (Journal of Neuroscience, Psychology, and Economics, Vol. 9, Iss. 3– 4, pp. 217–225. http://dx.doi.org/10.1037/npe0000062), the following para- graph should appear as a quotation:

“. . .Cheater detection stands out in acuity from mere error detection and the assessment of altruistic intent on the part of others. It is furthermore triggered as a computation procedure only when the cost and benefits of a social contract are specified. More than error, more than good deeds, and more than even profit, the possibility of cheating by others attracts attention. It excites emotion and serves as the principal source of hostile gossip and moralistic aggression by which the integrity of the political economy is maintained (E. O. Wilson, 1998, pp. 186 –187) . . .”

http://dx.doi.org/10.1037/npe0000106

87ERPS IN ONLINE REVIEW HELPFULNESS EVALUATION TASK

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.