Communicating science: Sending the right message to the right audience

All analyses were conducted in R (v. 3.1.1, R Core Development Team). Our final dataset for mixed models consisted of 334 participants. Sample sizes for each “group” were as follows: academic scientists = 59, natural resource managers = 23, non-academic scientists = 56, policymakers = 5, students = 93, and general public = 98. We binned the age of participants into five-year intervals to provide relatively balanced sample sizes per age group (minimum 13 participants per interval) while not losing information regarding the effects of participant age on information source use or trust, except for ages 18–20 (youngest age of participants), and 66+ (due to low number of participants in each five-year bin above this age). Thus, sample sizes for each age category were 18–20 = 24, 21–25 = 40, 26–30 = 52, 31–35 = 50, 36–40 = 37, 41–45 = 29, 46–50 = 24, 51–55 = 26, 56–60 = 13, 61–65 = 17, and 66+ = 22. For our models of information source trust and use, we excluded education as a predictor as this variable was highly correlated with both age and group. We did not expect survey medium (i.e., paper or online) to influence survey responses (Denscombe 2006); however, we included survey medium as a random effect in all models to be conservative, given the narrow geographical region sampled via paper versus the global distribution of online responses.

We used an information theoretic (IT) approach (Burnham and Anderson 2002), linear mixed-effects models, and model averaging to determine the relative influences of age, group, and general trust on both the use of three main sources of information (academic journals, social media, and personal experience) and the trust in these sources. These three sources were chosen as journals represent the traditional standard of science communication, social media is an emerging method of general communication, and personal experience is common to all groups and influences levels of trust. For each response, we scaled the dependent variable to reduce skew and then constructed a global model with country and community as random effects, and age, group, and general trust as fixed effects. We used a conservative approach to estimate the “best possible” (i.e., most parsimonious) model for each response variable. To do so, we fit all possible combinations of our fixed effects (eight models, or 2 ⁿ where n = number of predictors), selected those with ΔAICc (AICc = Akaike information criterion (corrected for small sample sizes)) ≤ 2 from the top model, and estimated parameters for each fixed effect included in these ΔAICc ≤ 2 subsets by model averaging following Burnham and Anderson (2002).

To compare trust in and use of information sources by groups, we used a Kruskal–Wallis analysis of variance (ANOVA) with multiple comparisons for “trust” and “use” scores for each information source (values binned between 0 and 100) by group. We also used this method to compare the perceived frequency of communication within and among groups.

Of our six series of linear mixed-effects models examining the trust and use of social media, academic journals, and personal experience by groups, global models investigating academic journals fit our data best for both trust (R ² = 0.19) and use (R ² = 0.35), whereas global models investigating the trust and use of personal experience had the lowest fit (R ² = 0.04, 0.16, respectively; Table 1). Full details of model comparison for each suite of models fitted to each response variable are presented in Table S1, whereas parameter estimates from the final averaged model for each response variable are presented in Table 1. Model results suggested that members of all groups trust peer review over other sources, but of all groups, the general public had the lowest trust of peer-reviewed journals, and used this information source the least, on average (Table 1). Participants between the ages of 26 and 30 exhibited higher trust of social media, whereas non-academic scientists and the general public exhibited significantly less trust of social media (Table 1). Participants over age 50 and the general public also used social media less to access scientific information (Table 1).

Each group ranked peer-reviewed scientific journals the highest in terms of trust followed by scientific conferences (84.9 ± 18.3 [SD] vs 80.8 ± 21.1, P < 0.001, n = 385; Fig. 1, Table S2). The public used peer-reviewed journals (36.6 ± 28.5) significantly less than news outlets (61.7 ± 33.5, P < 0.001, n = 89) and personal experience (59.8 ± 28.0, P < 0.001, n = 89), though not significantly less than information from colleagues (45.7 ± 31.8, P = 0.08, n = 89). Personal experience and colleagues were both widely used and trusted sources of scientific information across all groups (mean use: 59.2 ± 29.1, 68.5 ± 29.3; mean trust: 60.9 ± 24.0, 70.7 ± 23.6, respectively). News outlets and social media were also widely used to access scientific information (56.7 ± 34.0 and 50.9 ± 36.9), but were among the least-trusted sources of scientific information (44.4 ± 24.2 and 33.5 ± 24.9, respectively).

As expected, participants interacted more often with members from their own group than others (82.5 ± 27.9 within group vs 45.7 ± 35.1 out of group, P < 0.001, n = 273; Fig. 2). However, the public had the weakest within-group interaction (66.4 ± 37.4) compared to other groups (90.3 ± 18.4, P < 0.001, n = 366; Table S3). The public also estimated their frequency of interaction with all other groups as much lower than it was perceived by the other group in the interaction (29.6 ± 29.9 vs 77.8 ± 30.7, P < 0.001, n = 354; Fig. 2).

Peer review has long been the gold standard by which scientists measure information and communicate results (Neylon and Wu 2009). Thus, the finding that each group trusted peer-reviewed scientific journals the most is not surprising. Although scientists commonly rely on peer-reviewed journals, the general public used them the least of all groups and instead tended to learn from personal experiences, news outlets, or colleagues. Personal experience and colleagues were both widely used and trusted sources of scientific information among all groups, a trend that suggests ease of accessibility and the desire for discovery-based learning (Kolb and Kolb 2009). Interestingly, although news outlets and social media were commonly used to access scientific information, they were among the least-trusted sources, perhaps stemming from fears over sensationalism and misreporting (Kohring and Matthes 2007).

Given academic journals were the most trusted source of information across all groups, increasing their use outside academia is desirable. Harris et al. (2014) found that over half of the staff members in US state health departments did not use peer-reviewed journals to find evidence, with access and cost identified as reasons. The problem of poor access to scientific research is likely magnified for the general public, who may, in turn, be forced to access scientific information from other, likely non-peer-reviewed sources. As scientific research continues to become more “open”, there is great potential for scientists to engage in meaningful interactions with other groups. However, there are still barriers for authors to make their work open access, including publication costs, limitations in the types of open access journals that are available (e.g., certain disciplines may have few open access journals available to publish in), and in some cases, perceived lower quality of open access journals (e.g., predatory publishers, newer publications having low citations if they are not adequately indexed and searchable). However, there is evidence when controlling for journal age and discipline citation rates for open access articles are nearly identical in biomedical fields (Björk and Solomon 2012), and certainly over time, as open access articles are more widely accessible, many will have increased readership and citations. However, a general lay audience may still have difficulties understanding the technical language of a scientific paper; to communicate the main points of a paper to a wider audience, journals could consider publishing lay abstracts (i.e., plain language summaries) with traditional abstracts (Kuehne and Olden 2015). In addition, publicizing these lay abstracts through social media, which is already widely used by scientists and non-scientists alike, might be effective for engaging a diversity of audiences (Bickford et al. 2012). Scientists are increasingly using social media to engage with target audiences (e.g., Darling et al. 2013). However, to build trust in this medium, there is a need for mechanisms that distinguish trustworthy science from non-peer-reviewed information. One possibility might be the creation of “verified” social media accounts of leading scientists as a means of evaluating the quality and integrity of shared information. Such a platform could function like some open access journals where both reviews and author responses are published alongside the original research (e.g., PeerJ) with a lay, rather than academic, readership as the target. Social media would also allow target audiences to engage directly with scientists through posting questions or comments. The relatively low use of peer-reviewed journals among non-scientists may also reflect that many peer-reviewed studies are inaccessible behind paywalls (Hayes 1992). The push toward open access journals may begin to reverse this trend. In addition, although we did provide examples of common peer-reviewed journals in our survey (Nature and PLoS ONE), it is very likely that many members of the general public may be unclear about whether a publication is actually a peer-reviewed journal or not. It would likely be beneficial for scientists to conduct outreach educating the public about what peer review means, and how to evaluate journals and peer-reviewed articles. Understanding how to critique science on their own would likely increase the trust the public has in the scientific process. However, this is also an area where current research is lacking and little empirical evidence is available.

We found considerable within-group interaction and little bilateral interaction among many groups. In particular, the public self-reported less communication with other groups than was perceived by those groups. In contrast, policymakers reported a higher level of interaction with other groups than was perceived by those groups. To our knowledge, this is the first evidence of such an asymmetry in science communication. This pattern might be a byproduct of the relative number of policymakers (few) and public (many). Policymakers were very difficult to reach in conducting this survey (see Materials and methods), and this difficulty in direct communication may also factor into the low perceived interaction with policymakers by other groups. Furthermore, the very few policymakers who took the time to respond to our survey are likely more interested in science communication, and our results from such a small sample size should be treated with caution.

Our finding that the public reported little communication with other groups suggests there is a greater need for public outreach in scientific discourse. Part of the challenge is that, too often, science communication is still seen as a one-way flow of information, which is not the best mind-set for engaging with an audience (Miller 2001). Even when scientists are encouraged to frame their research in an interesting narrative, scientists are still cast as the guardians of knowledge, generously sharing their findings with members of the public (e.g., Olson 2009). Taking an empirical approach to determine how people understand scientific information may be key to improving communication among groups (Varner 2014). As our results show, personal experience is a common and trusted way for people from various groups to receive scientific information. Citizen science is an obvious way to increase the public’s personal experience with science; however, previous studies have found that the amount of scientific information participants learn from participating in citizen science projects can be limited (Jordan et al. 2011; Land-Zandstra et al. 2016). If citizen science is successful, it may help to not only inform the target audience but also create a sense of appreciation for the scientific process and better inform lay decision making (Dickinson et al. 2012). There is a clear precedent that active engagement, dialogue, and discussion are crucial factors in promoting knowledge retention, enthusiasm, and analytical capacity (Varner 2014), which may, in turn, support the communication of scientific information. Targeted communications training for the next generation of scientists could facilitate the process of information flow among groups and enhance the practice of science communication. In addition, many universities and institutions have developed media relation centers, and conferences have begun to include workshops on improving science communication; we encourage the use of these tools which are currently available.

Although we surveyed a broad audience across the web, most responses from the general public came from library submissions in Vancouver. Although it is unlikely that response quality differed between paper and online surveys (Yun and Trumbo 2000; Denscombe 2006) and respondents’ country of origin was included in our models to account for any geographical bias, some caution should still be taken in generalizing the results from the surveyed public to the global public, as religious and political values might influence how science communication is viewed regionally.