Risk and benefit perceptions of human enhancement technologies: The effects of Facebook comments on the acceptance of nanodesigned food

The introduction of a new technology, such as a human enhancement technology, may induce apprehension and concern among the general public. Social media enable individuals to find information and share their insights and concerns regarding new technologies. This results in an abundance of viewpoints that guides the individual's acceptance and decision-making. A relevant question for this special issue is to what extent attitudes toward human enhancement technologies are influenced by online cues that signal the views of other people without obvious relevant expertise, such as online comments (social proof). An online experiment focusing on the enhancement of human health and the functioning of the human body through the application of nanotechnology in food was conducted. The study investigated to what extent social proof impacted views on the application of nanotechnology in food. The valence of comments on a fake Facebook image with four comments was manipulated (positive, negative, mixed). A representative sample of Dutch Internet users ( n = 289) completed the study. Perceptions, feelings, behavior, and information need were measured. Results showed that comment valence had a significant effect on risk perception, benefit perception and attitude: the more positive the comments read by the participants, the lower risk perception, the higher benefit perception and the more positive the attitude toward nanodesigned food. Significant interaction effects of initial feelings of dread and comment valence were further found for risk perception and willingness to buy. In contrast, there were no significant interactions of initial feelings of optimism and comment valence. Implications for risk communication regarding human enhancement technologies are discussed.

technology refers to a range of different processes, materials and applications with the common theme of the manufacture and use of materials on a nanometer-size scale (Chaudhry, Watkins, & Castle, 2017). Examples are nanosized ingredients and additives, and nanoscale carriers for the delivery of nutrients and supplements (Chaudhry et al., 2017). 1 Public health agencies expect that the application of nanotechnology in the food sector will contribute to a safe, healthy and sustainable diet (Zantinge, van Bakel, van Loon, & Ocke, 2017) and hence to a better health and enhanced functioning of the body. Yet, as with all technologies, risks accompany benefits. In the case of the application of nanotechnology in food (nanodesigned food), the most relevant one is the possibility that very small insoluble and bio-persistent particles may cross the gut wall (Chaudhry et al., 2017).
Nanodesigned food can only be effective in enhancing human health and the functioning of the human body, if the technology is accepted, and if the created products are purchased and consumed.
If this apprehension leads individuals to abstain from consuming the respective product, individuals will also miss out on their benefits (Frewer, 2017). This makes it extremely relevant to understand how individuals form their opinion and take decisions regarding food in which nanotechnology has been applied.
With the dominance of the Internet and social media, individuals are nowadays not only exposed to communications by experts, journalists and their organizations, but also to views by people on social media without an ostensible expertise or background. Evidence on the effect of such social media expressions on the risks and benefits of human enhancement technologies is scarce. This article aims to fill this gap by reporting the results of an online experiment examining the effect social media expressions by ordinary people without specific expertise on the individual's perceptions, attitudes and willingness to buy nanodesigned food. The topic of the study thus relates to the indirect enhancement of human health and the functioning of the human body through the enhancement of food by nanotechnology.
In the acceptance of nanotechnology in foods, risk and benefit perceptions play an important role, as do perceived controllability, ethical concerns regarding environmental impact and animal welfare.

| Information processing and decision-making regarding food choice
Introducing new foods on the market involves providing consumers with information on their nutritional value and other qualities. In their decision-making, consumers need to make sense of this information to generate meaning and understanding. This includes thoughts, emotions and actions (Dervin, 1998;Pirolli & Russell, 2011;Weick, Sutcliffe, & Obstfeld, 2005). Fundamental processes that contribute to consumer decision-making are information seeking, processing and sharing (Berger, 2014;Caughron et al., 2013;Hilverda, Kuttschreuter, & Giebels, 2017;Rimal & Real, 2003).
Heuristic processing is defined by the use of cues to arrive more easily at a judgment such as the source of the information and other non-content characteristics of a message. It is more likely to take place with low issue involvement. Systematic processing, on the other hand, involves the effortful scrutiny and comparison of information. It takes place when an individual encounters information of significant personal importance. Information often contains contradictory elements.
If individuals focus on these contradictory elements, they are engaging in systematic information processing. In such cases, a need for further information may arise. An emerging information need may thus point to the systematically processing of information (Griffin et al., 1999).

| Decision-making under uncertainty: Social proof
Key aspects to the introduction of new technologies seem to be a lack of knowledge and a high level of uncertainty among the target audience regarding the weighing of the risks and benefits of the technology. A relevant theory in this respect is the principle of social proof.
This principle can be understood as a form of heuristic information processing where individuals assimilate the behaviors of others and rely on their judgments and behavior, in case they are uncertain about an appropriate course of action (Cialdini, 2001).
In line with recent studies (Amblee & Bui, 2011;Lee, Shi, Cheung, Lim, & Sia, 2011), we define social proof as any type of social information to infer a course of action. Defined in this way, social proof is not limited to behavior of others, but also includes collaboratively shared information and experiences of others that help individuals to form their opinion or decide upon an appropriate action. This type of social influence is also referred to as informational social influence. It differs from normative social influence, which occurs when individuals conform to social norms and expectations (Cialdini, 2001). Facebook comments that provide information about the use of a technology, such as nanotechnology in food products, that may help individuals form their opinion, and that do not express any expectations by other people on how to behave, can thus be viewed as informational social influence, and thus as social proof.
The reasoning behind the principle of social proof is that the likelihood of making an incorrect response is smaller, when one behaves in the same manner as other people who might be more knowledgeable in responding to the situation (Lee, Park, & Han, 2008;Okdie, Guadagno, Petrova, & Shreves, 2013). Based on this principle, one might reason that information on the risks and benefits of nanodesigned foods might generate uncertainty among consumers, which makes them susceptible to social proof.
In line with the affect heuristic, one might also assume that such information on risks or benefits affects risk perception as well as benefit perception (Finucane, Alhakami, Slovic, & Johnson, 2000;Siegrist et al., 2008). The risk as feelings perspective further suggests that the interaction of these cognitions and the feelings associated with the technology mutually influence each other and that their interplay determines the individual's attitude and willingness to buy (Loewenstein, Weber, Hsee, & Welch, 2001).
A key aspect of a statement or behavior that might lead to social proof is its valence: in favor or against. Most research into online social proof focused on Facebook as the most relevant social media platform. This research demonstrated effects of online social proof in a large variety of contexts, such as organic food (Hilverda, Kuttschreuter, & Giebels, 2018), breastfeeding attitudes (Jin, Phua, & Lee, 2015), marihuana legalization (Winter, Bruckner, & Kramer, 2015), brand engagement and sales (Kim & Johnson, 2016), and vaccination (Peter, Rossmann, & Keyling, 2014). There is also empirical evidence with respect to other social media, such as YouTube (Shi, Messaris, & Cappella, 2014;Walther, DeAndrea, Kim, & Anthony, 2010).
Most of these studies investigated the impact of viewing exclusively negative or exclusively positive expressions. Individuals are however most likely exposed to both positive and negative opinions from various sources at the same time (Lee et al., 2008). Research findings suggested that a higher percentage of narratives reporting adverse consequences led to a higher risk perception, which in turn led to a lower intention to vaccinate (Betsch, Ulshofer, Renkewitz, & Betsch, 2011). Similar results were found for a Facebook page with food safety information on restaurants (Seo, Almanza, Miao, & Behnke, 2015).
The evidence so far thus supports the idea that individuals may be influenced by the valence of what they read on social media. Evidence on the effects of online comments of mixed valence is scarce.

| Potential moderators: Initial attitudes
Initial attitudes affect cognitions, feelings, attitudes and behavior following provision of information in two ways. Firstly, there is the main effect of initial attitudes: the more positive the initial attitudes, the more positive cognitions, feelings, attitudes and behavior following information provision (Frewer, Howard, Hedderley, & Shepherd, 1999;Frewer, Scholderer, & Bredahl, 2003;Van Dijk, Fischer, de Jonge, Rowe, & Frewer, 2012).
Secondly, initial attitudes may affect the impact of the provided information. Cognitive dissonance theory suggests that individuals are likely to stick to their opinions, which might affect the way they seek, process and avoid new information (Deline & Kahlor, 2019;Gaspar et al., 2016;Kuhn, 2000;Narayan, Case, & Edwards, 2011). The empirical evidence on the effects of prior attitude in risk communication is fragmented (Frewer et al., 2016). There is however qualitative as well as quantitative experimental data to support an interaction effect of initial attitudes and provided information (McFadden & Lusk, 2015;Vainio, Irz, & Hartikainen, 2018;Vardeman & Aldoory, 2008).
Adapted to our context: exposure to positive information on a human enhancement technology might strengthen the benefit perception among individuals with a positive initial attitude, whereas exposure to negative information might strengthen the risk perceptions among individuals with a negative initial attitude toward the technology.
There is hardly any evidence on the interaction effects in case of information of mixed valence. Evidence by Van Dijk et al.
(2012) suggested that initial attitudes might become less strong when information on risks as well as benefits is given. Whether this also holds for information posted on social media by ordinary people is still unclear. Another reasoning is that information of mixed valence contains contradictory elements. As contradictory information leads to uncertainty (Boholm & Larsson, 2019) and as uncertainty increases risk perception, one might hypothesize that information of mixed valence would have a similar effect as negative information. To answer this question, we conducted an experiment involving a fictitious Facebook page on the application of nanodesigned food, that included four comments beneath a broad question that differed in valence (all positive, all negative, mixed [2 positive, 2 negative]). Dependent variables were risk perception, benefit perception, perceived retail safety, anxiety, positive emotions, attitude and willingness to buy. Information need was added to the dependent variables to examine whether observed effects could be ascribed to heuristic information processing as opposed to the systematic processing of the contents of the comments. Prior attitudes were taken into account, split into initial feelings of dread and initial feelings of optimism.

| Main effect of comment valence (H1)
A main effect of comment valence was hypothesized: the more positive comments the individual read, the higher benefit perception, perceived retail safety, evoked positive emotions, attitude and willingness to buy, and the lower risk perception and anxiety. It was further hypothesized that the mixed set of comments would induce uncertainty and hence lead to a higher need for information than exclusively positive or exclusively negative comments.

| Main effect of initial dread (H2)
A main effect of initial dread was hypothesized: the higher initial dread, the higher risk perception and anxiety, and the lower benefit perception, perceived retail safety, evoked positive emotions, attitude and willingness to buy. It was further hypothesized that a higher initial dread would be associated with a higher level of uncertainty and hence a higher level of information need.

| Interaction effect of comment valence and initial dread (H3)
Departing from the idea that information that is congruent with the individual's initial attitude, carries more weight than incongruent information, a significant interaction between comment valence and initial dread was hypothesized. Risk perception and anxiety were expected to be highest among participants with a high initial dread who read the negative comments. In contrast, benefit perception, perceived retail safety, evoked positive emotions, attitude and willingness to buy were expected to be highest among participants with a low initial dread who read the positive comments. It was further hypothesized that a high number of negative comments would strengthen the effect of the initial dread on information need: information need was expected to the highest among participants with a high initial dread who read the negative comments.

| Main effect of initial optimism (H4)
A main effect of initial optimism was hypothesized: the higher the initial optimism, the higher benefit perception, perceived retail safety, evoked positive emotions, attitude and willingness to buy, and the lower risk perception and anxiety. It was further hypothesized that a higher initial optimism would be associated with a lower level of uncertainty and hence a lower level of information need.

| Interaction effect of comment valence and initial optimism (H5)
The reasoning that congruent information carries more weight also applies to positive feelings. It was therefore hypothesized that there was a significant interaction between comment valence and initial optimism. Benefit perception, perceived retail safety, evoked positive emotions, attitude and willingness to buy were expected to be highest among participants with a high initial optimism who read the positive comments. In contrast, risk perception and anxiety were expected to be highest among participants with a low initial optimism who read the negative comments. It was further hypothesized that a high number of positive comments would strengthen the effect of initial optimism on information need: information need was expected to the highest among participants with a low initial optimism who read the negative comments. The sample consisted of 139 males (48%) and 150 females (52%).
Age ranged from 18 to 77 with a mean age of 47 years. All participants were familiar with Facebook. Including themselves, they mostly lived in households of two (43%), three (16%) or four persons or more (19%), whereas 21% lived on their own. In the month prior to the study, almost all participants had been responsible for grocery shopping (96%) and for cooking the main meal of the day at least once a week (91%).
Sample selectivity was assessed by comparing the participants who answered the manipulation check question correctly to those who did not. There were no differences with respect to gender, age, education, daily occupation, income, number of household members, grocery shopping, cooking and Facebook use. The differences in initial dread and initial optimism and the perceived emotionality and helpfulness of the comments were also insignificant. Three significant differences were found. The participants perceived themselves significantly less well informed on nanodesigned food (M = 2.19, SD = 1.18) than those who filled out the manipulation questions incorrectly (M = 2.74, SD = 1.70, t [90.41] = −2.58, p ≤ .05). They further considered the comments to be clearer, Mann-Whitney U-test, Z = 4.21, p ≤ .0005, and more biased, Mann-Whitney U-test, Z = 2.16, p ≤ .05, than those who filled out the manipulation check incorrectly. The analyzed sample thus seemed to be selective with respect to their perceived knowledge on nanodesigned food and the appreciation of the comments, but not with regard to background characteristics and initial attitudes.
The participants were randomly assigned to three conditions. A randomization check showed a significant difference between conditions for only 1 out of 15 tested variables, gender. There were, relatively speaking, more females among the participants who read the positive comments (64%) than among those who read the negative comments (47%) or the mixed set of comments (44%). There were no differences between the three conditions with respect to age, education, daily occupation, income, number of household members, grocery shopping, cooking, online media use (Facebook, Twitter, Skype, fora and blogs), initial dread, initial optimism, and the perceived knowledge on nanodesigned food.

| Design and manipulation
A randomized one factor between subjects experiment with two moderators was carried out. Participants viewed screenshots of an alleged Comparison of the ratings in the three conditions showed a significant difference with respect to partiality, Kruskall-Wallis, χ 2 (2, N = 289) = 30.91, p ≤ .0005: participants who read the set of mixed valence comments considered the comments to be less biased than those who read only positive or only negative comments. There were also significant differences in emotionality, Kruskall-Wallis χ 2 (2, N = 289) = 29.11, p ≤ .0005, and perceived clearness, Kruskall-Wallis χ 2 (2, N = 289) = 8.46, p ≤ .05: the participants who read the positive comments considered the comments to be less emotional and less clear than those who read the negative or set of mixed comments. There was no difference with respect to the helpfulness of the comments in advising a friend.

| Dependent variables
Risk perception. To measure risk perception, participants were requested to indicate to what extent they agreed with four statements regarding the hazardousness of nanotechnology in foods to their health (4 items, α = .94, 7-point-Likert scale, 1 = strongly disagree to 7 = strongly agree).
Benefit perception: We measured benefit perception with four statements about the advantages of the application of nanotechnology in foods to the participant's health (4 items, α = .94, 7-point-Likert scale, 1 = strongly disagree to 7 = strongly agree).
Perceived retail safety: Participants indicated to what extent they had confidence in the safety of food products that were sold in retail (3 items, α = .94, 7-point-Likert scale, 1 = strongly disagree to 7 = strongly agree).
Anxiety: Anxiety was measured by asking the participants to what extent they experienced anxiety when thinking about eating foods in which nanotechnology had been applied (4 items, α = .94, 7-point scale ranging from 1 = not at all to 7 = very much).
Positive emotions: Participants indicated to what extent they experienced positive emotions when thinking about eating foods in which nanotechnology had been applied (4 items, α = .96, 7-point scale ranging from 1 = not at all to 7 = very much).
Willingness to buy: As research attitudes are more reliably measured when focused on specific foods rather than food categories (Bredahl, 1999), willingness to buy was measured by asking the participants to what extent they were inclined to purchase a variety of seven nanodesigned food products (7 items, α = .94, 7-point scale ranging from 1 = not at all to 7 = very much).
Information need: Participants filled out to what extent they wanted to know more about nanodesigned foods (3 items, α = .95, 7-point scale, 1 = strongly disagree to 7 = strongly agree).

| Moderators
Initial dread: Initial dread was measured before the participants viewed the screenshot. Participants were asked to what extent they dreaded the application of nanodesigned food (4 items, α = .87, 7-point scale; 1 = strongly disagree to 7 = strongly agree).
Initial optimism: Initial optimism was also measured before the participants viewed the screenshot. They were asked to what extent they expected the application of nanodesigned food to have advantages (4 items, α = .90, 7-point scale; 1 = strongly disagree to 7 = strongly agree).

| Additional variables
Manipulation check: Participants were asked to indicate the valence of the majority of the comments on the Facebook page (positive, negative, about equal).
Evaluation of the comments: Participants were asked to rate the comments on the Facebook page for clearness, emotionality, partiality and helpfulness in advising a friend (individual items, 7-point scale, 1 = strongly disagree to 7 = strongly agree).
Perceived knowledge on nanodesigned food: Participants were asked to indicate to what extent they agreed with statements that their level of knowledge on the application of nanotechnology in food was satisfactory (3 items, α = .88, 7-point scale; 1 = strongly disagree to 7 = strongly agree).
Online media use: Participants were questioned on the frequency of their online media use: Facebook, Twitter, Skype, and fora and blogs (4 items, 7-point frequency measure, 1 = less than once a month to 7 = multiple times a day.
Sociodemographics: Gender, age, education, income and household composition were measured. The participants were further asked how often they did the grocery shopping and cooked the main meal of the day.

| Procedure
After indicating their consent, participants filled out an online questionnaire. They were first given a short description of nanotechnology and told that nanotechnology is to a greater or lesser extent applied in food products and food packages. The description of the benefits mentioned increased shelf life, but focused on the benefits for the human body: improved absorption of vitamins and stimulating the defense against germs and diseases. It was further indicated that little was known about the long-term effects of the use of nanotechnology.
To increase involvement, a request by a friend for advice about eating foods created on the bases of nanotechnology was used as a cover story. After filling out the questions about their knowledge and initial attitude regarding nanodesigned foods, the participants viewed the alleged Facebook screenshot. They then indicated their evaluation of the comments, and filled out the manipulation check question. The dependent variables were subsequently measured. At the end, the participants answered questions about their socio-demographics and online media use. They were then debriefed and thanked for their participation.

| Analysis
The analysis was conducted by means of SPSS (version 25). Descriptive statistics were established to describe the sample. To obtain an indication of the reliability of the instruments, Cronbach's alpha was calculated. Next, composite variables were computed, and means and correlations were calculated. The hypotheses were tested by means of multivariate analysis variance (GLM), followed by univariate analysis if applicable. Three main effects were included in the analysis (comment valence, initial dread and initial optimism) as well as two interaction terms (valence by initial dread, valence by initial optimism).
Regression analysis was applied to further examine the significant interaction effects. To correct for differences between the three conditions, gender was included as a covariate. Additional analyses were carried to examine the role of perceived knowledge on nanodesigned food and online media use.

| Means and correlations
Means, standard deviation and correlations of the main variables are reported in Table 1. Before reading the Facebook comments, the participants expressed some concern about nanodesigned food Note: All variables: 7-point scales, ranging from 1 to 7. ***p ≤ .001,**p ≤ .01, *p ≤ .05, two-tailed.
optimism, benefit perception, attitude, positive emotions and willingness to buy, and, to a lesser extent, perceived retail safety and perceived knowledge on nanodesigned food. The correlations between these two groups of variables were highly negative. Information need correlated significantly and positively with initial dread as well as initial optimism, but not significantly with any of the dependent variables, anxiety excepted.

| Hypotheses testing: Multivariate analysis of covariance
Multivariate analysis of covariance was applied to test the effect of valence (H1), initial dread (H2), the interaction of valence and initial dread (H3), initial optimism (H4) and the interaction of valence and initial optimism (H5). Given differences between the conditions, gender was included as a covariate.

| Hypotheses testing: Univariate analysis of covariance
Subsequent univariate analysis showed that these effects together explained 56% of the variance in risk perception, 56% in benefit perception, 21% in perceived retail safety, 24% in anxiety, 42% in positive emotions, 59% in attitude, 38% in willingness to buy and only .09% in information need (Table 2).

| Main effect of valence (H1)
Comment valence had a significant main effect on risk perception, benefit perception and attitude (Figure 2).
Correcting for the effects of gender, initial dread, initial optimism and both interaction effects, participants who read the negative comments scored highest on risk perception (M estimated = 4.32, SE = .07), followed by those who read the mixed set (M estimated = 4.14, SE = .08) and those who read the positive comments (M estimated = 4.02, SE = .07).
Similar effects, but in the opposite direction, were found for benefit perception. Correcting for gender, initial dread, initial optimism and both interaction effects, participants who read the negative comments scored lowest on benefit perception (M estimated = 3.52, SE = .07), followed by participants who read the mixed set of comments (M estimated = 3.61, SE = .08) and those who read the positive ones (M estimated = 3.86, SE = .07).
For attitude, there was a statistically significant main effect of comment valence, too. Correcting for gender, initial dread, initial T A B L E 2 Univariate results of testing the main effects for valence, initial dread and initial optimism, and the two-way interaction effects between valence and initial dread, and between valence and initial optimism, with gender as covariate (N = 289) optimism and both interaction effects, participants who read the negative comments scored lowest on attitude (M estimated = 3.34, SE = .08), followed by those who read the mixed set and (M estimated = 3.60, SE = .09) and those who read the positive comments (M estimated = 3.86, SE = .08).

Constructs
For risk perception, benefit perception and attitude there was thus evidence of an effect of comment valence in the expected direction.
The effect was medium-sized for attitude, and small to medium-sized for risk and benefit perception (Hedrick, Bickman, & Rog, 1993). For perceived retail safety, anxiety, positive emotions, willingness to buy and information need, the main effect of comment valence was not significant. H1 has thus partly been confirmed.

| Main effect of initial dread (H2)
The analysis showed highly significant main effects for initial dread on  (Hedrick et al., 1993). H2 has thus been confirmed.

| Interaction effect of initial dread and comment valence (H3)
The analysis showed significant interaction effects between comment valence and initial dread for risk perception and willingness to buy.
Both effects were small to medium-sized (Hedrick et al., 1993).

Risk perception
Subsequent examination showed that the positive relationship between initial dread and risk perception was strongest for the participants who read the mixed set of comments, β = .71, t (78) = 11.08, p ≤ .0005, followed by the participants who read the positive, β = .50, t(100) = 7.27, p ≤ .0005, and the negative comments, β = .24, t(105) = 2.93, p ≤ .01. The effect of the initial dread on risk perception was thus strongest for the participants who read the mixed set of comments, in the middle for the participants who read the positive comments and lowest for those who read the negative comments. The results further showed that risk perception was highest among the participants who scored high on initial dread and read the mixed set of comments, whereas it was lower among the participants who scored high on initial dread and read the negative comments (Figure 3).
H3 predicted a significant interaction of comment valence and initial dread, in the sense that information that is congruent with the individual's initial attitude, would carry more weight than incongruent information. We did find a significant interaction; risk perception was however highest among participants who scored high on initial dread and read the mixed set of comments. H3 is thus partly confirmed.

Willingness to buy
Subsequent examination showed that the negative relationship between initial dread and willingness to buy was strongest for the par- Results further showed that willingness to buy was highest among participants who scored low on initial dread and read the positive comments, where as it was lower among the participants who scored low on initial dread and read the negative or mixed set of comments ( Figure 4).
H3 predicted a significant interaction of comment valence and initial dread, in the sense that information that is congruent with the individual's initial attitude, would carry more weight than incongruent information. We did find a significant interaction. We also found that F I G U R E 2 Estimated means for risk perception, benefit perception and attitude, correcting for the effects of gender, initial dread, initial optimism, and both interaction effects willingness to buy was highest among those participants who scored low on initial dread and read the positive comments. This supports H3.

| Main effect of initial optimism (H4)
There were significant main effects of initial optimism on all the dependent variables. Following their viewing of the Facebook comments, the participants who already perceived nanodesigned food in a positive way, scored higher on benefit perception (β = .54), perceived retail safety (β = .32), positive emotions (β = .40), attitude (β = .51), willingness to buy (β = .18), and information need (β = .18), and lower on risk perception (β = −.45) and anxiety (β = −.28). There were strong effects for risk perception, benefit perception, positive emotions, attitude and willingness to buy, while those for perceived retail safety, anxiety and information need were medium-sized (Hedrick et al., 1993). H4 was thus confirmed with respect to perceptions, emotions, attitude and willingness to buy. For information need, H4 was rejected: instead of the hypothesized negative effect a positive one was found.

| Interaction effect of initial optimism and comment valence (H5)
In line with the insignificant multivariate result, all univariately tested interaction effects of initial optimism and comment valence were insignificant (all p's > .05). There was thus no empirical support for H5.

| Effect of the covariate
The covariate gender was had a small, significant effect on three variables: positive emotions, attitude and willingness to buy (Hedrick et al., 1993). Adjusting for all main and interaction effects, men were more positive about nanodesigned food than women.
F I G U R E 3 Interaction effect between initial dread and comment valence for risk perception F I G U R E 4 Interaction effect between initial dread and comment valence for willingness to buy 3.4 | Additional analyses

| Perceived knowledge on nanodesigned food
Social proof is assumed to take place in case individuals are uncertain about an appropriate course of action (Cialdini, 2001). This would imply that the effect of comment valence would be stronger among individuals who perceived themselves less informed about nanodesigned food. This hypothesis was tested by adding perceived knowledge on nanodesigned food, and its interaction with comment valence, to the analysis. Tested multivariately, neither the main effect of perceived knowledge, Wilks' λ = .95, F(8, 268) = 0.87, ns, η 2 = .03, nor the interaction effect of comment valence and perceived knowledge, Wilks' λ = .95, F(116, 536) = 0.95, ns, η 2 = .03, was significant.
There was thus no support for the hypothesis that the effect of comment valence was stronger among individuals who scored lower on perceived knowledge on nanodesigned food.

| Online media use
The frequency of Facebook use was not related to any of the dependent variables. Experience with Facebook thus did not moderate the relationship between comment valence and the dependent variables.

| Summary of the results and theoretical implications
This study examined the effects of online social proof with respect to nanodesigned food by manipulating the valence of four comments to a fictitious Facebook page (4 positive; 2 positive +2 negative; 4 negative). A randomized-one-factor-between-subjects-experiment with two moderators, initial dread and initial optimism, was carried out.
The study was conducted in the Netherlands on a sample that was a representative of the population of Dutch Internet users with respect to age and gender. Randomization was successful for all examined variables, except gender. Including gender as a covariate in the analysis corrected for this. Variables were reliably measured.

| Main effect of comment valence
Results showed a significant, small to medium-sized effect of comment valence in the expected direction on risk perception, benefit perception and attitude. The more positive the comments read by the participants, the lower risk perception, the higher benefit perception and the more positive the attitude toward nanodesigned food. The effects on perceived retail safety, anxiety, positive emotions, willingness to buy and information need were not significant. This might perhaps be attributable to the focus of the comments on the risks, benefits and evaluation of nanodesigned food. Another explanation could be that the comments did not arouse an affective response because the participants perceived a high level of control over their exposure to the risk.
In line with previous research on Facebook comments (Betsch et al., 2011;Seo et al., 2015;Winterbottom et al., 2008), there was thus some evidence to support Hypothesis 1 and the principle of social proof as operationalized in the valence of comments below a Facebook post. This finding is the more remarkable, because the alleged authors were characterized by a name only, and neither background information nor picture was provided. The effect was thus not attributable to the author being presented as an expert or as someone to whom the participant was similar. Granted that it is the perceived expertise of or similarity with an author that often counts (Hilverda et al., 2017), the authority principle and the similarity principle do not seem to provide a plausible explanation of our results.
The principle of social proof specifies uncertainty about an appropriate course of action as a relevant condition (Cialdini, 2001). This suggests that the effects of the comments would depend on the individual's level of knowledge. We found no evidence for this. An explanation might be that anxiety related to nanodesigned food was not high (Kuttschreuter & Hilverda, 2019). To better understand the conditions that lead to social proof future research should focus on a human enhancement technology that generates a higher level of anxiety and concern.
The affect heuristic suggests another mechanism underlying the effect of comment valence. This heuristic states that a risk-related message may generate an affective response in the message's recipient, which would result in effects on the perception of the risks as well as the benefits of the message's topic (Finucane et al., 2000). As our comments were mostly risk-related, our result that comment valence affected risk and benefit perception lends some support for the affect heuristic. To further explore the mechanisms behind the effect of comment valence, future research might include an instrument to measure the affect aroused by the comments.

| Main effects of initial dread and initial optimism
As new technologies carry the expectation of potential benefits at the expense of potential risks, initial attitudes were split into feelings of dread and optimism. Results showed highly significant main effects for initial dread and initial optimism on all the dependent variables.
This is in line with the literature (Frewer et al., 1999;Frewer et al., 2003;Van Dijk et al., 2012) and confirmed hypotheses 2 and 4.
Interestingly, for all variables, both main effects were significant.
All dependent variables thus reflected both the potential benefits as well as the potential negative consequences of the technology. This suggests that in the case of new human enhancement technologies, individuals base their cognitions, feelings and behavior on the anticipated positive as well as on the potential negative consequences (Frewer, 2017).

| Interaction effects of comment valence and initial dread and initial optimism
Cognitive dissonance theory and empirical evidence demonstrated that initial attitudes may moderate the effect of risk-benefit messages (Frewer et al., 2016;Gaspar et al., 2016;Narayan et al., 2011). Distinguishing between initial feelings of dread and optimism, we studied whether this also held for nanodesigned food.
For two variables, a small to medium-sized, but significant interaction between comment valence and initial dread was observed. For risk perception, the relationship between initial dread and risk perception was strongest for the participants who read the set of comments of mixed valence, followed by those who read the positive, and the negative comments. This suggests that the participants who read comments of mixed valence relied the most on their initial feelings of dread. An explanation could be that comments of the mixed valence did not motivate participant high on initial dread to re-evaluate their ideas, whereas exclusively positive or exclusively negative comments did.
Among the participants high on initial dread, those who read the mixed set of comments scored highest on risk perception. This might perhaps be due to the uncertainty or negative affect evoked by the comments of mixed valence (Boholm & Larsson, 2019). Contrary to expectation, among participants high on initial dread, those who read the negative comments scored lowest. Following the Extended Parallel Processing model, this could be indicative of a coping process where individuals apply a fear-reduction strategy instead of riskreducing behavior (Witte, 1992).
Willingness to buy was highest for the participants low on initial dread who read the positive comments. This supported the idea that congruent information carried more weight than incongruent information. At the same time, the effect of initial dread on willingness to buy was strongest for the participants who read the positive comments.
An explanation could be that the information in the positive comments was already familiar to the participants and already included in their attitude toward nanodesigned food.
All in all, results partly confirmed H3: we did find two significant interactions, and one of these suggested that comments that were congruent to initial feelings of dread carried more weight. The explanation given above are however highly speculative and in need of corroboration by future research.
There was no significant interaction between comment valence and initial optimism. H5 was thus rejected. Perhaps processes based on positive attitudes or emotions are simpler than those that involve negative attitudes and emotions.

| Information need
Systematic information processing involves the effortful comparison of information and resolving inconsistencies as opposed to the use of peripheral cues that characterizes heuristic processing (Chaiken, 1980;Griffin et al., 1999;Kahlor et al., 2003;Trumbo, 1999). In our view, information need plays an important role in systematic processing, but not as much in heuristic processing. To find out whether our findings could be attributed to the systematic processing of the contents of the comments as opposed to the heuristic processing, we studied the effect of comment valence on information need. The participants expressed a high need for information, yet comment valence did not affect their information need.
Information need was further only very weakly related to just one of the other dependent variables. The effects of the comments should thus not be attributed to the participants' systematic processing of the comments' content, but rather to heuristic processes.
An unexpected and intriguing result regarding information need was that it was significantly positively related to initial dread as well as initial optimism. This suggests that feelings of dread and optimism may both motivate a need for information. While the former is in line with the Risk Information Seeking and Processing model (Griffin et al., 1999), the role of positive feelings in information need and information seeking has been given little attention (Savolainen, 2014

| Other technologies
Statements on the generalizability of our results to other human enhancement technologies are highly speculative. Main reasons are the lack of empirical evidence and the broad range of existing human enhancement technologies (Dijkstra & Schuijff, 2016). Concerns about new technologies are, at least partly, issue-specific (Besley & McComas, 2015). This is further complicated by the inconsistent relationships between the determinant of the acceptance of a technology such as risk and benefit perception (Bearth & Siegrist, 2016 Our operationalization of social proof consisted of plain, short, riskrelated texts on Facebook with a single emoticon issued by an character identified by a name only; no images. Familiarity with an author affects a message's effect (Martensen et al., 2018). We might therefore assume that effects of Facebook comments will be stronger when they are issued by someone to whom the individual is familiar. One might further expect that an image draws the attention away from the text. That would imply that a text with an image would have a smaller effect than a text without one. Unless, of course, the image underlines the text. A related aspect is the clarity of the proof shown by others. Social proof in terms of the number of likes associated with a text seemed too subtle for platform users to notice . Finally, as the authors of the comments were presented as ordinary people without any obvious expertise on nanodesigned food, one should be careful to generalize the findings to social media platforms, such as LinkedIn and Twitter that are often used by professionals to disseminate professionrelated information. Further research seems indicated.

| Other countries
There are two important aspects to the generalization of our findings to other countries: social media usage and the level of knowledge and attitude related to nanodesigned food.
In the Netherlands, Internet penetration and the social media usage are both high (CBS, 2013(CBS, , 2015

| Implication for risk communication practice
Our study showed that social media expressions by other people without obvious expertise may affect the individual's views regarding a human enhancement technology. Risk communicators should be aware of this phenomenon. They should not ignore the discourse on social media, but monitor it and intervene, if applicable, by entering the conversation, spreading their arguments and views, and, if applicable, correct inaccurate or false information (Veil, Buehner, & Palenchar, 2011). This would provide platform users with information that might assist them in a better understanding the risks and benefits of the technology and would increase informed decision-making.
One scenario is that the social media discourse disproportionally focuses on the risks of a new human enhancement technology. Monitoring the discourse would enable risk communicators who want to avoid that the public misses out on the benefits of the technology to counter potential detrimental effects of negative expressions on social media.
Another scenario is that the social media discourse disproportionally focuses on the benefits of a new human enhancement technology. In that case, monitoring the discourse would alert risk communicators and enable those who want to increase public awareness of the risks of the technology, to respond timely.

| Final comments
This is one of the first studies that examined the effects of comments on

ACKNOWLEDGMENT
We would like to thank the Netherlands Food and Consumer Product Safety Authority for funding this study.
2 Randomization was successful (χ 2 (2, N = 362) = 0.02, p = .99). There were, however, more participants in the mixed condition who incorrectly answered the manipulation check question. It is unclear why this was the case. As these participants were excluded from the data analysis, the number of participants in the mixed condition decreased more than those in the positive and negative condition (χ2 (2, N = 263) = 23.89, p ≤ .0005). In the final data set, there was however no significant statistical difference in the number of participants in the three valence conditions (χ 2 (2, N = 289) = 4.28, p = .12).
likes version and a low-number-of-likes version. The manipulation check showed that this manipulation was not successful: participants just did not seem to pay attention to the number of likes. Analysis showed that there were no significant relationships between the number of likes beneath the comments and (a) comment valence, (b) the dependent variables, (c) the moderators initial dread and initial optimism, (d) the perceived knowledge on nano-designed food, and (e) socio-demographics. The distinction between the high-number-of-likes and the low-numberof-likes condition could therefore be dropped from the study. (1) the role of new information channels such as the Internet and more specifically social media in risk communication, (2) the concept of social influence in relation to health behavior, and

AUTHOR BIOGRAPHIES
(3) hard-to-reach target groups for health communication. What do you think about the application of nanotechnology in foods?
Positive comments • Comment 1: I saw once on TV how we can use nanotechnology and I am happy about the application of nanotechnology in foods!
• Comment 2: I think people get ill less often because of the nanotechnology in foods; therefore it is safe!
• Comment 3: I am very much convinced that nanotechnology in foods makes food products healthier.
• Comment 4: To the best of my knowledge, nanotechnology in food products is not harmful and I will just eat it!! Negative comments • Comment 1: I saw once on TV how we can use nanotechnology and I am unhappy about the application of nanotechnology in foods!
• Comment 2: I think people get ill more often because of the nanotechnology in foods; therefore it is dangerous!
• Comment 3: I am very much convinced that nanotechnology in foods makes food products unhealthier.
• Comment 4: To the best of my knowledge, nanotechnology in food products is harmful and I just won't eat it!!