In consensus we trust? Persuasive effects of scientific consensus communication

Abstract

Scholars have recently suggested that communicating levels of scientific consensus (e.g. the percentage of scientists who agree about human-caused climate change) can shift public opinion toward the dominant scientific opinion. Initial research suggested that consensus communication effectively reduces public skepticism. However, other research failed to find a persuasive effect for those with conflicting prior beliefs. This study enters this contested space by experimentally testing how different levels of consensus shape perceptions of scientific certainty. We further examine how perceptions of certainty influence personal agreement and policy support. Findings indicate that communicating higher levels of consensus increases perceptions of scientific certainty, which is associated with greater personal agreement and policy support for non-political issues. We find some suggestive evidence that this mediated effect is moderated by participants’ overall trust in science, such that those with low trust in science fail to perceive higher agreement as indicative of greater scientific certainty.

Keywords

consensus communication scientific uncertainty trust in science

1. Introduction

Although scientific inquiry offers the most systematic method of producing knowledge about our world, segments of the public often hold beliefs and policy attitudes that are inconsistent with the views of the scientific community (Pew Research Center, 2015). Therefore, one aim of science communication is to shift public attitudes to better align with those of scientists. Recently, scientific consensus information (i.e. the percentage of scientists that agree on a given position) has been identified as influential in moving public attitudes toward positions supported by scientific evidence (Lewandowsky et al., 2012; van der Linden et al., 2015a, 2015b). Across a range of subjects, perception of scientific consensus is associated with acceptance of scientific positions (Ding et al., 2011; Lewandowsky et al., 2012), suggesting that the belief that scientists are certain acts as a “gateway belief” for greater agreement with scientists’ positions (van der Linden et al., 2015b, 2016). This research paints an optimistic picture of the ability of consensus communication strategies to persuade individuals to hold beliefs supported by scientific evidence.

However, two areas remain under-researched. First, most work has focused on the presence or absence of consensus information (e.g. Lewandowsky et al., 2012; van der Linden et al., 2016). Few studies have examined how exposure to different levels of scientific consensus impacts individuals’ perceptions of scientific certainty. Second, despite initial evidence that consensus communication strategies may be effective at persuading skeptics, there are inconsistent findings concerning how individuals’ prior attitudes toward science interact with consensus information. This study addresses these two under-researched areas by investigating how exposure to different levels of scientific consensus affects perceptions of certainty, which in turn influences personal agreement and funding support. We further examine how this mediated relationship may be moderated by individuals’ overall trust in science.

2. Literature review

Scientific agreement and perceptions of certainty

Although the complexity of science is often an obstacle to laypeople’s understanding, a more fundamental roadblock is that everyday citizens may be unsure what constitutes findings that are “certain” enough to inform their personal views. The Western production of scientific knowledge is, paradoxically, based on uncertainty. Popper (2005) argued that science is less a “body of knowledge,” and more “a system of guesses or anticipations” (p. 318). As a result, scientific research entails a certain degree of uncertainty, particularly on issues that are complex or relatively under-researched. The public, however, has often demonstrated a lower tolerance for scientific uncertainty. When told that scientists disagree, people may be more likely to attribute the disagreement to incompetence or self-interest, rather than insufficient data or methodological limitations (Dieckmann et al., 2015a; Johnson and Slovic, 1998). Growing distrust in science may be partly attributable to public frustration with scientific uncertainty (Maddox, 1995).

Therefore, we begin by examining how scientific consensus information influences laypeople’s perceptions of scientific certainty. The limited research on responses to scientific consensus suggests that higher levels of consensus are more likely to be persuasive. Yet, which levels of consensus are considered certain or uncertain remains unclear. Consensus communication has been predominantly evaluated within the context of climate change (Lewandowsky et al., 2012; van der Linden et al., 2015b, 2016), whose anthropogenic source enjoys a near-unanimous consensus within the scientific community (Cook et al., 2013). As a result, our understanding of the effects of communicating scientific consensuses has largely been limited to comparisons between 97% consensus and no consensus information (Lewandowsky et al., 2012; van der Linden et al., 2016).

Determining how the public perceives different levels of scientific consensus is vital for practitioners’ application and theoretical understanding. There are few studies that have examined this outside the context of climate change. Johnson (2017) showed that information about the proportion of scientists that endorse a given position affected judgments of scientific agreement, with respondents finding near-consensus proportions (e.g. 95% vs 5%) more indicative of scientific agreement than other proportions (e.g. 40% vs 60%). Aklin and Urpelainen (2014) found a negative effect on policy support when scientific agreement dropped below 98%, compared to a baseline in which no consensus information was provided. Such work suggests that in cases where agreement is not near-consensus, laypeople perceive the position as less certain. To better determine what levels of consensus laypeople perceive as certain, we examine differences between five discrete levels of scientific consensus (25%, 45%, 65%, 85%, and 95%) as well as the absence of consensus information. Based on the extant literature, we hypothesize that

H1. Messages that communicate higher levels of scientific consensus will lead to stronger perceptions of scientific certainty.

Furthermore, perceptions of scientific certainty when no consensus information is provided are not well-understood. Aklin and Urpelainen (2014) found that withholding consensus information and asserting that 98% of scientists agree resulted in comparable levels of policy support, while lower levels of consensus (80%, 60%) resulted in weaker support. In the absence of consensus information, individuals may assume that the scientific information they receive is endorsed by most scientists. It is important to note that this assumption may vary by issue. While Aklin and Urpelainen (2014) investigated perceptions toward a fabricated issue, studies exploring politicized issues, like climate change and genetically modified organisms (GMOs), found that perceived consensus levels are affected by prior beliefs (Dixon, 2016; Kahan et al., 2011; van der Linden et al., 2015b, 2016). In addition, claims about laypeople’s evaluations of scientific certainty in the absence of consensus information are often drawn from findings concerning outcomes like policy support (Aklin and Urpelainen, 2014) or reported knowledge like estimated level of consensus (Dixon, 2016; van der Linden et al., 2015b, 2016). To the best of our knowledge, no previous research has investigated how laypeople perceive scientific certainty in the absence of consensus information. To better understand advantages and disadvantages of providing and withholding consensus information, we ask:

RQ1. How will messages that communicate no scientific consensus information influence perceptions of scientific certainty compared to messages that communicate different specific levels of consensus information?

The moderating role of trust in science

Another goal of this study is to examine how the relationship between scientific consensus information and perceived scientific certainty might be moderated by individuals’ trust in science, an attitude known to influence how people evaluate scientific information. Trust is a heuristic that individuals use to form beliefs and make decisions about risky or complex subjects (Brewer and Ley, 2013; Cummings, 2014; Johnson, 2003), particularly when they are unequipped or unmotivated to critically interrogate information (Petty and Cacioppo, 1984). Those with higher trust in science evaluate new scientific information more favorably than those with lower trust (Druckman and Bolsen, 2011; Lee, 2005; Siegrist, 2000; Siegrist et al., 2007) and as trust decreases, so too do perceptions of the certainty of scientists’ assertions (Hmielowski et al., 2014).

While consensus communication is often deployed to persuade those who hold counter-scientific beliefs to adopt dominant scientific positions, it is unclear if those who are skeptical of science are influenced by consensus information. Dixon (2016) found that, when accounting for prior beliefs and deference to scientific authority, consensus communication increased public acceptance of scientific claims solely among those predisposed to believe scientists’ positions. Those disinclined to support scientists’ positions were less affected by consensus information, raising questions about the efficacy of consensus communication strategies for audiences who are predisposed to mistrust scientific information.

On the other hand, there is also some evidence that consensus messages may be able to persuade skeptics. van der Linden et al. (2016) report no differences between Republicans and Democrats in recalled consensus level on climate change after exposure to a consensus message, even though the two groups have different prior attitudes (Dunlap and McCright, 2008) and trust in science (Gauchat, 2012). This suggests that prior attitudes toward science may not moderate the proposed gateway belief: scientific certainty (Johnson, 2017).

Given the inconsistent findings concerning how evaluations of scientific certainty may be affected by skepticism toward science, we ask:

RQ2. How does trust in science moderate the relationship between levels of scientific consensus and perceptions of scientific certainty?

Perceptions of certainty and attitudinal outcomes

Next, we explore how perceptions of scientific certainty are related to attitudinal outcomes. As the goal of consensus communication is to shift public beliefs and policy attitudes to be in line with scientific evidence, these are central outcomes of interest. When forming opinions, individuals rely more on certain information compared with uncertain information (Kahneman and Tversky, 1979). This bias suggests that scientific information that is perceived to be certain is more likely to be utilized in attitude formation and evaluative processes than information that is perceived to be uncertain (Kahneman and Tversky, 1979). Perceiving scientific certainty is also positively associated with policy support (Ding et al., 2011). Correspondingly, one’s estimate of the percent of scientists who agree is positively associated with personal agreement, concern, and public engagement (van der Linden et al., 2015b, 2016). We therefore hypothesize that the perceptions of greater scientific certainty will be positively associated with personal agreement and funding support in line with scientists’ positions. Formally stated:

H3a. Higher levels of perceived scientific certainty will be positively associated with higher levels of personal agreement with scientists’ positions.

H3b. Higher levels of perceived scientific certainty will be positively associated with higher levels of support for policy to increase funding in line with scientists’ positions.

Effects of scientific agreement on outcomes through perceptions of certainty

Finally, we examine the indirect effects of scientific consensus information on attitudinal outcomes through perceptions of certainty. Thus far, we have offered predictions regarding (a) how levels of scientific consensus may influence perceptions of scientific certainty, (b) how this relationship may be moderated by trust in science, and (c) how perceptions of scientific certainty may, in turn, influence personal beliefs and funding support. Putting these predictions together, we examine the indirect effects of consensus level on attitudinal outcomes through perceptions of certainty and test whether these indirect effects vary across levels of trust in science.

In testing these relationships, this study addresses the incomplete findings of previous research in two important ways. First, we investigate different levels of scientific consensus to determine what the public perceives as certain and how one’s perceptions of certainty are associated with personal agreement and funding support. Second, we examine how one’s trust in science may moderate the persuasive effects of consensus information to explore whether this strategy is capable of influencing skeptics.

3. Methods

Participants

A total of 1177 adults located in the United States were recruited using Amazon Mechanical Turk (MTurk)¹ for a study about how people respond to media messages. In all, 13 participants who failed to complete the study or who answered a simple attention check question incorrectly were removed from the sample. Four participants who reported their gender as “other” were also removed from the sample because the size of this group was too small to allow for meaningful comparison to other groups.² The final sample had 1160 participants, who ranged in age from 18 to 73 (M_age = 35.08, standard deviation (SD) = 10.62). About 47% of the sample self-identified as male (N = 541) and 53% self-identified as female (N = 619). The majority of participants self-identified as White or Caucasian (73.19%), followed by Black or African American (8.28%), Asian/Asian American (7.41%), Hispanic (6.03%), and other racial groups (5.09%).

Procedure

After providing informed consent, participants viewed one news article attributed to the Associated Press about a scientific issue. As we were interested in isolating the relationship between scientific consensus and attitudinal outcomes, we chose issues that may be debated within the scientific community but are not politically or religiously salient in the contemporary US context. Participants viewed an article concerning one of three issues: (1) whether the gravitational pull of the moon causes stronger earthquakes, (2) whether repeated motions lead to bone damage, or (3) whether artificial sweeteners affect the composition of gut microbiota. All stimuli were constructed for this study. Participants were randomly assigned to one of six conditions in which the percentage of scientists who agreed was manipulated to read “25%” (n = 199), “45%” (n = 188), “65%” (n = 192), “85%” (n = 203), “95%” (n = 189) or in which no consensus information was provided (n = 189). Participants were told that the consensus levels were the results of a survey of scientists conducted by the Pew Research Center. Each article was between 115 and 140 words and contained no images or graphics.

Measures

Control variables

Age (M_age = 35.08, SD = 10.62) and gender (53% female, 47% male) were included as demographic controls. In addition, previous literature suggested that people differ in numeracy, the ability to apply numerical concepts (Peters et al., 2013). We measured numeracy with a measure from Hart (2013) which adds the number of correct responses to seven items drawn from Lipkus et al. (2001) and Frederick (2005). These items were combined into a mean index of numeracy (M = 5.09, SD = 1.70), which was also included as a control variable.³

Moderating variable

Our model explores the extent to which the mediated association is dependent upon participants’ trust in science. Trust in science was measured with a combined scale of three items. Two of the items—(1) “How much do you trust science in general?” and (2) “How credible is science in general?”—were measured on a 6-point scale from 0 (Not at all) to 5 (Completely), while the third item—(3) “Scientists know what is good for the public”—was measured on a 6-point scale from 0 (Strongly disagree) to 5 (Strongly agree). These items captured a respondent’s perceived trust (Gauchat, 2012), as well as credibility of scientific information (Liu and Priest, 2009) and deference to scientific actors (Anderson et al., 2012). The three items were combined into an index of trust in science (Cronbach’s α = .82; M = 3.45, SD = 0.79).

Mediating variable

To assess the mediating variable, we asked participants how certain they thought the science was regarding the issue position featured in the article they read on a 6-point scale from 0 (Very uncertain) to 5 (Very certain). Their response was labeled perception of certainty (M = 2.93, SD = 1.15).

Dependent variables

Participants were asked to report their personal agreement in response to the question, “In your opinion, do you agree or disagree” with the scientific position presented in the stimulus on a 6-point scale from 0 (Strongly disagree) to 5 (Strongly agree). This was used as a measure of personal agreement (M = 3.14, SD = 1.07) with the position to which the participant was exposed. Participants were also asked how much they opposed or supported an invented bill, “H.R.2142 [which] will redirect tax dollars to increase US government funding to address issues associated with” the issue featured in the article on a 6-point scale from 0 (Strongly oppose) to 5 (Strongly support). This was used as a measure of support for a funding policy, hereafter called funding support (M = 4.56, SD = 1.45).

Analysis

The first step in our analysis was to check whether control variables (age, gender, and numeracy) and the moderating variable (trust in science) differed by condition using chi-square and analysis of variance (ANOVA) tests. ANOVA tests were also used to perform a manipulation check and examine the direct effect of experimental condition on the mediating variable (perception of certainty) and dependent variables (personal agreement and funding support). All post hoc pairwise comparisons were performed using Tukey tests with a pre-determined alpha level of .05.

To test for indirect effects, we used the SPSS macro PROCESS (version 2.16.1), which adopts a path analytic approach using ordinary least squares (OLS) regression (Hayes, 2013). Indirect effects were bootstrapped using 10,000 iterations and bias corrected 95% confidence intervals (CIs). We also used the R package “mediate” (Tingley et al., 2014) to probe the moderated mediation model and plot the resulting indirect effects of experimental condition across levels of trust in science.

4. Results

Preliminary analyses

Gender did not significantly differ by condition, χ²(5, N = 1160) = 6.89, p = .23, nor did age, numeracy, or trust in science (all Fs < 1.37, ps > .23). In order to examine the success of the experimental manipulation in communicating levels of scientific consensus on the issues, participants were asked to recall the percentage of scientists who agreed in the article they read. For conditions where a percentage was provided (i.e. all conditions except the no consensus condition) an average of 90.72% of participants recalled the correct percentage within ±5 percentage points. This manipulation check did not significantly vary by condition, χ²(4, N = 971) = 6.39, p = .17, indicating that the experimental manipulation was successful across conditions. Those in the condition where no level of scientific agreement was provided reported scientists agreeing at M = 58.35%, SD = 24.02.

ANOVA models

To investigate H1 and RQ1, we examined the effect of the experimental manipulation on the mediating and dependent variables by conducting a series of ANOVAs in which condition, issue, and the interaction between the two were entered as independent variables.

For perception of certainty, there was an effect of both scientific consensus, F(5, 1142) = 12.70, p < .001, and issue, F(2, 1142) = 31.83, p < .001, but no interaction between the two, F(10, 1142) = 0.76, p = .66. The first row of Table 1 presents mean levels of perception of certainty by manipulated consensus level and demonstrates a general trend in which higher levels of scientific consensus resulted in higher perceived certainty. Post hoc Tukey tests revealed that these differences were only significant between some levels of scientific consensus at p < .05 (see Table 1; first row). Those in the 25% condition (M = 2.51, SD = 1.30) were significantly lower in perceived certainty than those in the 65% (M = 2.83, SD = 1.07), 85% (M = 3.20, SD = 1.08), 95% (M = 3.27, SD = 1.11), and none (M = 3.08, SD = 1.05) conditions. Those in the 45% condition (M = 2.70, SD = 1.10) were significantly lower than those in the 85%, 95%, and none conditions. Those in the 65% condition were significantly higher than those in the 25% condition and significantly lower than those in the 85% and 95% conditions. There was no difference between the 85%, 95%, and none conditions. All other comparisons were not significant.

Table 1.

Means by manipulated level of scientific consensus.

Dependent variable	Condition
	25%	45%	65%	85%	95%	None
Perception of certainty	2.51 (1.30)a	2.70 (1.10)ab	2.83 (1.07)bc	3.20 (1.08)d	3.27 (1.11)d	3.08 (1.05)cd
Personal agreement	2.87 (1.17)a	3.07 (0.98)ab	3.06 (1.01)ab	3.30 (1.12)b	3.35 (1.00)b	3.21 (1.04)b
Funding support	4.48 (1.46)a	4.46 (1.41)a	4.55 (1.35)a	4.71 (1.48)a	4.59 (1.51)a	4.53 (1.51)a

Condition means reported with standard deviations in parentheses. Means within the same row with different subscripts were found significantly different at p < .05 using a Tukey test.

These results support H1 that messages communicating higher levels of consensus lead to stronger perceptions of scientific certainty than lower levels of consensus. In addition, they answer RQ1, regarding how perceptions of scientific certainty compare when one receives no consensus information vs a specified level of consensus. Withholding consensus information was associated with significantly higher perceptions of certainty than reporting 25% or 45% consensus and did not significantly differ from the 65%, 85%, and 95% conditions.⁴

In terms of issues, the bone damage issue (M = 3.27, SD = 1.06) was perceived as most certain, followed by the artificial sweetener issue (M = 2.94, SD = 1.12) and the gravity issue (M = 2.60, SD = 1.18; all significantly different at p < .05).

For personal agreement with the position presented, there was an effect of both scientific consensus, F(5, 1142) = 4.54, p < .001, and issue, F(2, 1142) = 42.97, p < .001, but no interaction between the two, F(10, 1142) = 1.18, p = .30. The second row of Table 1 presents mean levels of personal agreement by condition and again demonstrates a trend in which higher levels of scientific consensus resulted in higher personal agreement. Those in the 25% condition (M = 2.87, SD = 1.70) were significantly lower in personal agreement than those in the 85% (M = 3.30, SD = 1.12), 95% (M = 3.35, SD = 1.00), and none (M = 3.21, SD = 1.04) conditions; however, there were no other significant differences between the three latter conditions. Both 45% (M = 3.07, SD = 0.98) and 65% (M = 3.06, SD = 1.01) conditions were not significantly different than other conditions (see Table 1). The bone damage issue (M = 3.49, SD = 1.03) led to the highest personal agreement followed by the artificial sweetener issue (M = 3.17, SD = 0.94) and the gravity issue (M = 2.79, SD = 1.11; all significantly different at p < .05).

Finally, there was no significant effect of scientific consensus on funding support, F(5, 1142) = 0.46, p = .81 (see Table 1: third row), but there was an effect of issue, F(2, 1142) = 27.93, p < .001. Both the bone damage issue (M = 4.87, SD = 1.27) and artificial sweetener issue (M = 4.68, SD = 1.50) received significantly more funding support than the gravity issue (M = 4.12, SD = 1.47) at p < .05. There was no significant difference between bone damage and artificial sweetener issues.

Simple mediation models

To examine our third set of hypotheses, we next tested a simple mediation model in which different levels of scientific consensus on an issue indirectly influence personal agreement and funding support through the mediator. As our hypotheses predicted scientific certainty as mediator between consensus level and attitudinal outcomes, we investigated the mediated effects despite not observing a main effect of consensus level on funding support. Preacher and Hayes (2004) point out that it is possible to find a significant indirect effect in the absence of a main effect, critiquing the criteria for mediation laid out by Baron and Kenny (1986) and advocating direct, formal testing of mediation hypotheses (Preacher and Hayes, 2004). We followed the advice of Preacher and Hayes (2004) and formally tested the indirect effects on personal agreement and funding support with a bootstrapping approach to produce 95% CIs.

To do so, we used PROCESS model 4, which computes the unstandardized indirect effect of each condition dummy variable on each dependent variable as mediated through perceptions of certainty. Age, gender, numeracy, and issue are included as control variables in all analyses. We first examined the direct effects of experimental condition on perception of certainty (mediator) while also controlling for trust in science (moderator). Because the results of our ANOVA model suggested that most significant differences in perception of certainty occurred between very low levels and very high levels of scientific consensus, we selected 25% as the reference category. Results, which are presented in column 1 of Table 2, replicated our findings from the ANOVA model. We found that, with the exception of the 45% condition, moving from the 25% condition to any another condition increased perception of certainty (bs = .33–.71, ps < .01).

Table 2.

Regressions for moderated mediation models.

Mediator	Mediator		Dependent variables
	Perception of certainty		Personal agreement	Finding support
Perception of certainty	–	–	.67*** (.02)	.40*** (.04)
Condition dummy variables
Condition (45% vs 25%)	.21 (.11)	.31 (.46)	.09 (.07)	−.07 (.13)
Condition (65% vs 25%)	.33** (.11)	.27 (.47)	−.03 (.07)	−.04 (.13)
Condition (85% vs 25%)	.65*** (.10)	−.18 (.47)	−.06 (.07)	−.08 (.13)
Condition (95% vs 25%)	.71*** (.11)	.38 (.46)	−.03 (.07)	−.22 (.13)
Condition (None vs 25%)	.50*** (.11)	.09 (.51)	−.07 (.07)	−.25 (.13)
Control variables and moderator
Age	.01** (.003)	.01** (.003)	.01*** (.002)	.01 (.004)
Gender	−.04 (.06)	−.04 (.06)	−.06 (.04)	−.01 (.08)
Numeracy	−.14*** (.02)	−.14*** (.02)	.03** (.01)	−.07** (.02)
Issue (bone damage vs gravity)	.62*** (.08)	.62*** (.08)	.25*** (.05)	.47*** (.10)
Issue (sweeteners vs gravity)	.37*** (.08)	.37*** (.08)	.15** (.05)	.45*** (.09)
Trust in science (TIS)	.33*** (.04)	.26** (.09)	.07* (.03)	.42*** (.05)
Interactions
Condition (45% vs 25%) × TIS		−.03 (.13)
Condition (65% vs 25%) × TIS		.02 (.13)
Condition (85% vs 25%) × TIS		.24^# (.13)
Condition (95% vs 25%) × TIS		.10 (.13)
Condition (None vs 25%) × TIS		.12 (.14)
Constant	1.48*** (.22)	1.75*** (.36)	.46** (.15)	1.88*** (.27)
Adjusted R²	.20	.20	.57	.22
Residual standard error	1.05	1.05	0.70	1.29
F statistic	23.73*** (df = 11; 1147)	16.68*** (df = 16; 1142)	129.50*** (df = 12; 1146)	28.20*** (df = 12; 1146)

Unstandardized coefficients reported. Standard errors in parentheses.

p < .10; *p < .05; **p < .01; ***p < .001 (two-tailed). N = 1159.

Looking to the influence of perception of certainty (mediator) on personal agreement and funding support (dependent variables), respectively, perception of certainty had a positive association with personal agreement, b = .67, standard error (SE) = 0.02, p = .001 (Table 2; column 3), and funding support, b = .40, SE = 0.04, p = .001 (Table 2; column 4), finding support for hypotheses 3a and 3b.

Concerning the indirect effects, Figure 1 (left panel) plots the indirect effects of each condition dummy variable (reference = 25%) on personal agreement. We find significant indirect effects on personal agreement for 65% vs 25% (point estimate (PE) = .223, 95% CI (.076, .337)), 85% vs 25% (PE = .438, 95% CI (.286, .587)), 95% vs 25% (PE = .475, 95% CI (.321, .638)), and none vs 25% (PE = .335, 95% CI (.194, .477)). There was no significant indirect effect for 45% vs 25%, (PE = .138, 95% CI (−.018, .285)). These results suggest that when compared to the lowest level of scientific agreement (25%), levels higher than 45% indirectly increase personal agreement by increasing perceptions of certainty (see Figure 1, left panel).

Figure 1.

Indirect effects of scientific consensus on personal agreement and funding support through the influence of perception of certainty (simple mediation model).

We conducted the same mediation analysis using “none” as the reference category. We found significant negative indirect effects on personal agreement (95% CI entirely below zero) for 25% vs none (PE = −.335, 95% CI (−.490, −.186)), and 45% vs none (PE = −.198, 95% CI (−.334, −.067)); however, there were no significant indirect effects for 65% vs none (PE = −.112, 95% CI (−.245, .150)) or 85% vs none (PE = .103, 95% CI (−.027, .233)). In contrast, there was a positive indirect effect found for 95% vs none (PE = .140, 95% CI (.001, .279)). These results suggest that when compared to no scientific consensus information, low levels of consensus (25% and 45%) indirectly decrease personal agreement, while the highest level of consensus (95%) indirectly increases personal agreement.

Next, we investigated the indirect effects of scientific consensus on funding support. Figure 1 (right panel) plots the indirect effects of each condition dummy variable on funding support. We find significant indirect effects on funding support (95% CI entirely above zero) for 65% vs 25% (PE = .133, 95% CI (.047, .234)), 85% vs 25% (PE = .262, 95% CI (.165, .375)), 95% vs 25%, (PE = .284, 95% CI (.184, .404)), and none vs 25% (PE = .200, 95% CI (.112, .305)). There was no significant indirect effect for 45% vs 25% (PE = .082, 95% CI (−.009, .176)). These results suggest a pattern of indirect effects similar to that found for personal agreement (see Figure 1, right panel).

Again, we conducted the same funding support mediation analysis using “none” as the reference category. We found significant negative indirect effects on funding support (95% CI entirely below zero) for 25% vs none (PE = −.200, 95% CI (−.305, −.112)) and 45% vs none (PE = −.118, 95% CI (−.206, −.040)); however, there were no significant indirect effects for 65% vs none (PE = −.067, 95% CI (−.150, .008)) or 85% vs none (PE = .062, 95% CI (−.015, .147)). In contrast, there was a positive indirect effect found for 95% vs none (PE = .084, 95% CI (.003, .172)). These results suggest that when compared to no scientific consensus information, low levels (25% and 45%) indirectly decrease funding support, while the highest level of consensus (95%) indirectly increases funding support.

Moderated mediation models

Finally, we used PROCESS model 7 to test RQ2, how trust in science may moderate the effects of consensus level on perceptions of certainty and the indirect effects on attitudinal outcomes. The results, reported in Table 2 (column 2), show that all interaction terms are non-significant (ps > .05). However, as Brambor et al. (2006) note, the entry of multiple interactions terms into a regression model can increase multicollinearity and bias coefficients, potentially obscuring significant interactive effects. This possibility was supported by a test of the variance inflation factors for the interaction terms in our model, which were all >30, indicating high multicollinearity (Miles, 2005). In addition, Hayes (2013) and Brambor et al. (2006) argue that when the primary concern is how a variable might moderate a mediated effect, the statistical significance of individual coefficients is less interpretable than the indirect effects and bootstrapped CIs.

Accordingly, we probed for moderated mediation by visualizing the indirect effects found from the PROCESS analysis using the R package “mediate.” We plotted indirect effects for moderated mediation models specifying personal agreement and funding support as dependent variables (Figures 2 and 3, respectively).⁵ Each plot presents indirect effects and 95% CIs for a given dummy variable across levels of the trust in science. Points along each solid plotted line where the CIs do not cross zero suggest significant indirect effects. For both dependent variables, there appears to be no significant moderation for the 45% vs 25% dummy variable (the band depicting the 95% CI contains zero across the range of the moderator). However, for all other contrasts, a general pattern emerges in which significant indirect effects only occur at higher levels of trust in science. This pattern of findings is likely due in part to the fact that there were fewer respondents who had very low levels of trust in science, resulting in wider CIs for low levels of trust in science. Nonetheless, the pattern of the PEs for the comparisons of 85% vs 25%, 95% vs 25%, and none vs 25% still suggests that individuals with higher levels of trust in science may be more sensitive to the differences in scientific consensus messaging.

Figure 2.

Indirect effects of scientific consensus on personal agreement at levels of trust in science (moderated mediation model).

Figure 3.

Indirect effects of scientific consensus on funding support at levels of trust in science (moderated mediation model).

5. Discussion

The findings suggest that as scientific consensus increases, so do perceptions of scientific certainty, supporting H1. Our manipulation of consensus level revealed that, in general, people shift their perception of how certain the science is on a given issue to match the consensus information they receive. We note that while we observe differences in perceptions of certainty between moderate levels (e.g. 45% vs 65% or 65% vs 85%), there were not observed differences between lower levels (e.g. 24% vs 45%) or higher levels (e.g. 85% vs 95%). These findings are in line with Johnson’s (2017) study, which found a similar trend concerning judgments of scientific agreement.

In answer to RQ1, these results replicate previous studies which found no significant differences between reporting high levels of consensus (e.g. 85%, 95%) and withholding consensus information (Aklin and Urpelainen, 2014). Compared with providing no consensus information, any consensus below 65% decreased perceptions of certainty. For the issues investigated here, in the absence of information about consensus levels, individuals appear to think of scientific information as largely certain.

Perceptions of scientific certainty were in turn positively associated with personal agreement with scientists’ positions (H3a) and corresponding funding support (H3b). The results of these tests are in line with predictions that perceiving scientific certainty is a “gateway belief” for adopting attitudes in line with the consensus, as suggested by previous research (Ding et al., 2011; van der Linden et al., 2016).

We identified some suggestive evidence for RQ2 that trust in science moderates the association between scientific consensus and perceptions of certainty, as well as the indirect effects on personal agreement and funding support. Looking at the figures examining the moderated indirect effects (Figures 2 and 3), it appears that the influence of consensus information on scientific certainty, and thereby on personal agreement and funding support, is limited to those who have moderate to high trust in science. We note that formal tests for interaction with trust in science were not significant, and the suggestive patterns revealed in the figures are in need of additional research.

Overall, we found consistent patterns of association across three different topics, suggesting that individuals respond to consensus information in similar ways in cases of non-politicized scientific issues. While our issue conditions did not interact with consensus conditions, it is clear from the mean differences of issue conditions that individuals do consider issue characteristics when making evaluations.

Furthermore, this study identifies self-reported perceptions of scientific certainty as an important mediator between consensus information and attitudinal outcomes. Most previous work highlighting individuals’ recognition of consensus as an antecedent of persuasive outcomes conceptualizes participants’ recalled consensus level as the mediator between consensus information and attitudes (Dixon, 2016; van der Linden et al., 2015b, 2016). Participants’ recall of consensus level captures their attention to recently provided information but may not capture their acceptance of that information. Understanding how the public perceives scientific information is as important in strategic communication research as what information they receive. Our study measured individuals’ acceptance of a consensus, the proposed gateway belief, more directly than previous studies by explicitly assessing perceived scientific certainty. This provides a more detailed picture of the process through which individuals evaluate consensus information.

We chose to focus on non-politicized issues to help establish a fundamental understanding of how individuals process different levels of scientific consensus. However, a central interest of many exploring consensus communication strategies is the potential for consensus information to persuade skeptics of the validity of dominant scientific positions. It is possible that politically polarized science issues or issues which see substantial skeptic engagement may exhibit a different pattern of effects. When exposed to the same level of consensus, individuals who hold views that conflict with a set of scientific findings may be motivated to perceive less scientific certainty than those predisposed to agree with scientists’ positions. When individuals are motivated to reject scientific consensuses for different reasons (e.g. partisan identity, religious beliefs), individuals may exhibit a pattern of results similar to participants in this study who report little trust in science: they may be less sensitive to consensus information (Dixon, 2016). Examination of these relationships in the context of more politicized issues is an important next step in this line of research.

Looking to trust in science, we note that our investigation may have been hampered by an insufficient sample of mistrusting individuals with which to make claims about their evaluations. Indeed, our index of trust in science is negatively skewed (−.74). Online panels, such as MTurk, are frequently younger, more educated, less religious, and more liberal than the US population (Paolacci and Chandler, 2014). Prior work suggests that Democrats (Gauchat, 2012) and more educated individuals (Achterberg et al., 2017) tend to be more trusting of science, thus we may expect to encounter fewer mistrustful individuals in our sample than exist in the population. Future work exploring moderating effects of trust in science may benefit from a more diverse sample in this regard.

Ultimately, our findings should be interpreted within the broader context of science communication. Although consensus communication strategies seek to persuade individuals to acknowledge a scientific consensus as a gateway belief for holding corresponding attitudes, it may not resolve broader social conflicts about the role of science in public decision-making. For those who believe that the values of science are out of step with social goods, or that scientific information is but one consideration among many (e.g. religious beliefs, civic values), a high degree of scientific agreement may count for little. This study also raises concerns about the ethical use of consensus information. While our results suggest that high levels of consensus can be persuasive, communicators should also consider the uncertainty inherent in scientific research. Historically, scientific findings which were widely accepted within the scientific community have been overturned. While some argue that exposing the public to the uncertainties inherent in the scientific process can depreciate the cultural authority of science (Maddox, 1995), others argue that transparency demonstrates the value of scientific knowledge for public decision-making (Lupia, 2017).

Research into consensus messages is timely given that such strategies are already employed by prominent figures and organizations (e.g. @BarackObama, 2013; NASA, 2018; “The Consensus Project,” n.d.). In addition, those with different media diets may be exposed to dramatically different presentations of scientific consensuses (Feldman et al., 2012). Our findings are in line with research that posits that differences in the presentation of scientific consensuses may be in part responsible for differences in viewers’ attitudes (Feldman et al., 2012; Hmielowski et al., 2014). The variability in prevalence and presentation of consensus information across media diets warrants further research into effects of different forms of consensus communication on diverse audiences.

In conclusion, our finding that individuals may use reported levels of scientific consensus to discern the certainty of the science, and that perceptions of scientific certainty are in turn related to personal beliefs and funding support, has important implications for public opinion on scientific issues. Although near-consensus agreement appears persuasive to most people, our suggestive findings caution practitioners that, even in cases of non-politicized issues, mistrust in science may restrict the efficacy of consensus messages.

Footnotes

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

Author biographies

Sedona Chinn is a PhD candidate in the Department of Communication Studies at the University of Michigan. Her research focuses on science communication, political communication, and media effects.

Daniel S. Lane is a PhD candidate in the Department of Communication Studies at the University of Michigan. His research focuses on the role that communication plays in engaging people in the world of political and social change.

Philip S. is an Associate Professor in Communication Studies and the Program in the Environment at the University of Michigan. Dr. Hart specializes in environmental, science, and risk communication.

References

@BarackObama (2013) Ninety-seven percent of scientists agree: #climate change is real, man-made and dangerous. Read more: [Twitter Post]. Available at: https://twitter.com/barackobama/status/335089477296988160?lang=en

Achterberg

de Koster

van der Waal

(2017) A science confidence gap: Education, trust in scientific methods, and trust in scientific institutions in the United States, 2014. Public Understanding of Science 26: 704–720.

Aklin

Urpelainen

(2014) Perceptions of scientific dissent undermine public support for environmental policy. Environmental Science & Policy 38: 173–177.

Anderson

Scheufele

Brossard

Corley

(2012) The role of media and deference to scientific authority in cultivating trust in sources of information about emerging technologies. International Journal of Public Opinion Research 24: 225–237.

Baron

Kenny

(1986) The moderator-mediator variable distinction in social the moderator-mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations. Journal of Personality and Social Psychology 51: 1173–1182.

Brambor

Clark

Golder

(2006) Understanding interaction models: Improving empirical analyses. Political Analysis 14: 63–82.

Brewer

Ley

(2013) Whose science do you believe? Explaining trust in sources of scientific information about the environment. Science Communication 35: 115–137.

Cook

Nuccitelli

Green

Richardson

Winkler

Painting

et al . (2013) Quantifying the consensus on anthropogenic global warming in the scientific literature. Environmental Research Letters 8: 1–7.

Cummings

(2014) The “trust” heuristic: Arguments from authority in public health. Health Communication 29: 1043–1056.

10.

Dieckmann

Johnson

Gregory

Maygora

Han

PKJ

Slovik

et al . (2015a) Public perceptions of expert disagreement: Bias and incompetence or a complex and random world? Public Understanding of Science 26: 325–338.

11.

Dieckmann

Peters

Gregory

(2015b) At home on the range? Lay interpretations of numerical uncertainty ranges. Risk Analysis 35: 1281–1295.

12.

Ding

Maibach

Zhao

Roser-Renouf

Leiserowitz

(2011) Support for climate policy and societal action are linked to perceptions about scientific agreement. Nature Climate Change 1: 462–466.

13.

Dixon

(2016) Applying the gateway belief model to genetically modified food perceptions: New insights and additional questions. Journal of Communication 66: 888–908.

14.

Druckman

Bolsen

(2011) Framing, motivated reasoning, and opinions about emergent technologies. Journal of Communication 61: 659–688.

15.

Dunlap

McCright

(2008) A widening gap: Republican and democratic views on climate change. Environment: Science and Policy for Sustainable Development 50: 26–35.

16.

Feldman

Maibach

Roser-Renouf

Leiserowitz

(2012) Climate on cable: The nature and impact of global warming coverage on Fox News, CNN, and MSNBC. The International Journal of Press/Politics 17: 3–31.

17.

Frederick

(2005) Cognitive reflection and decision making. Journal of Economic Perspectives 19(4): 25–42.

18.

Gauchat

(2012) Politicization of science in the public sphere: A study of public trust in the United States, 1974 to 2010. American Sociological Review 77: 167–187.

19.

Hart

(2013) The role of numeracy in moderating the influence of statistics in climate change messages. Public Understanding of Science 22: 785–798.

20.

Hayes

(2013) Introduction to Mediation, Moderation, and Conditional Process Analysis: A Regression-Based Approach. New York, NY: Guilford Press.

21.

Hmielowski

Feldman

Myers

Leiserowitz

Maibach

Hmielowski

et al . (2014) An attack on science? Media use, trust in scientists, and perceptions of global warming. Public Understanding of Science 23: 866–883.

22.

Johnson

(2003) Further notes on public response to uncertainty in risks and science. Risk Analysis 23: 781–789.

23.

Johnson

(2017) “Counting votes” in public responses to scientific disputes. Public Understanding of Science 1–17. DOI: 10.1177/0963662517706451.

24.

Johnson

Slovic

(1998) Lay views on uncertainty in environmental health risk assessment. Journal of Risk Research 1: 261–279.

25.

Kahan

Jenkins

Smith-H

Braman

(2011) Cultural cognition of scientific consensus. Journal of Risk Research 14: 147–174.

26.

Kahneman

Tversky

(1979) Prospect theory: An analysis of decision under risk. Econometrica 47: 263–292.

27.

Lee

C-J

(2005) Public attitudes toward emerging technologies: Examining the interactive effects of cognitions and affect on public attitudes toward nanotechnology. Science Communication 27: 240–267.

28.

Lewandowsky

Gignac

Vaughan

(2012) The pivotal role of perceived scientific consensus in acceptance of science. Nature Climate Change 3: 399–404.

29.

Lipkus

Samsa

Rimer

(2001) General performance on a numeracy scale among highly educated samples. Medical Decision Making 21: 37–44.

30.

Liu

Priest

(2009) Understanding public support for stem cell research: Media communication, interpersonal communication and trust in key actors. Public Understanding of Science 18: 704–718.

31.

Lupia

(2017) Now is the time : How to increase the value of social science. Social Research: An International Quarterly 84: 669–694.

32.

Maddox

(1995) The prevalent distrust of science. Nature 378: 435–437.

33.

Miles

(2005) Tolerance and variance inflation factor. In: Everitt

Howell

(eds) Encyclopedia of Statistics in Behavioral Science, vol. 4. Hoboken, NJ: John Wiley & Sons, pp. 2055–2056.

34.

NASA (2018) Global climate change: Facts. Available at: https://climate.nasa.gov/scientific-consensus/

35.

Paolacci

Chandler

(2014) Inside the Turk: Understanding Mechanical Turk as a participant pool. Current Directions in Psychological Science 23: 184–188.

36.

Peters

Västfjäll

Slovic

Mertz

Mazzocco

Dickert

(2013) Numeracy and decision making numeracy. Psychological Science 17: 407–413.

37.

Petty

Cacioppo

(1984) The effects of involvement on responses to argument quantity and quality: Central and peripheral routes to persuasion. Journal of Personality and Social Psychology 46: 69–81.

38.

Pew Research Center (2015) Public and scientists’ views on science and society. Available at: http://www.pewinternet.org/2015/01/29/public-and-scientists-views-on-science-and-society/

39.

Popper

(2005) The Logic of Scientific Discovery. London; New York, NY: Routledge.

40.

Preacher

Hayes

(2004) SPSS and SAS procedures for estimating indirect effects in simple mediation models. Behavior Research Methods, Instruments, & Computers 36: 717–731.

41.

Siegrist

(2000) The influence of trust and perceptions of risk and benefits on the acceptance of gene technology. Risk Analysis 20: 195–203.

42.

Siegrist

Cousin

Kastenholz

Wiek

(2007) Public acceptance of nanotechnology foods and food packaging: The influence of affect and trust. Appetite 49: 459–466.

43.

The Consensus Project (n.d.) The consensus project. Available at: http://theconsensusproject.com

44.

Tingley

Yamamoto

Hirose

Keele

Imai

(2014) Mediation: R Package for causal mediation analysis. Available at: https://imai.princeton.edu/research/files/mediationR2.pdf

45.

van der Linden

Clarke

Maibach

(2015a) Highlighting consensus among medical scientists increases public support for vaccines: Evidence from a randomized experiment. BMC Public Health 15: 1207.

46.

van der Linden

Leiserowitz

Maibach

(2016) Communicating the scientific consensus on human-caused climate change is an effective and depolarizing public engagement strategy: Experimental evidence from a large national replication study. SSRN Electronic Journal. Available at: http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2733956

47.

van der Linden

Leiserowitz

Feinberg

Maibach

(2015b) The scientific consensus on climate change as a gateway belief: Experimental evidence. PLoS ONE. Epub ahead of print 25 February. 10(2): e0118489.