Assessing the impacts of normative and efficacy‐based messaging on the social diffusion of conservation science

Abstract

Encouraging people to share science‐based information on environmental issues with others is critical for enhancing the speed and scale of conservation action. However, little experimental research exists that examines how message framing can facilitate the social diffusion of scientific information. We report the results of a series of studies conducted to test the effects of normative and efficacy‐based messaging on Colorado residents' willingness to share scientific information about the state's wolf reintroduction initiative. We distributed messaging using mailings to the general public, surveys of online participant panels, and personal messages from undergraduate students to their own contacts. We then measured participants' individual engagement with the scientific information we provided and their engagement in social behaviors that would encourage further social diffusion of the information. While we find some evidence of increased engagement by people contacted through social diffusion, we do not find any evidence that normative and efficacy‐based messaging encourages people to engage in social diffusion, nor do we find that such messaging enhances individual engagement with the scientific information. We identify several barriers to the sharing of scientific information, which could inform the development of future interventions to scale up biodiversity conservation efforts beyond individual engagement through widespread social diffusion.

INTRODUCTION

Conservation organizations are faced with the challenge of designing effective messaging that motivates public engagement with conservation science and conservation initiatives (Kidd et al., 2019; Kusmanoff et al., 2020). A growing body of literature has begun to identify effective strategies for message framing to promote engagement with conservation issues, such as reducing cognitive separation from the issue, evoking strong positive or negative emotions, or appealing to the values of a specific target audience (Kidd et al., 2019; Kolandai-Matchett & Armoudian, 2020; Kusmanoff et al., 2016; Kusmanoff et al., 2020; van der Linden et al., 2015; van Vugt et al., 2014). Much of the existing literature has focused on enhancing public awareness of environmental issues and motivating individuals to engage in conservation action. However, individual action alone may prove insufficient to address ongoing and future conservation crises, given that the potential impacts of individual behavior are constrained by broader, collective social systems over which individuals have little control (Amel et al., 2017). To enhance the speed and scale of conservation action, people must be motivated not only to engage with conservation science themselves, but also to reach out to others to spread science-based information and motivate changes in behavior (Swim et al., 2014). Understanding how conservation behavior spreads through social networks is critical to engaging with environmental issues, because both individual behavior change and collective institutional change are needed to build sustainable cultural practices for the future (Brooks et al. 2018).

Motivating people who engage with a conservation cause to reach out to others is crucial to facilitate widespread action, because people are more likely to believe and to engage with new information when they learn about it from individuals they know personally or perceive similarity with (Burger et al., 2004; Goldberg et al., 2019; Gootee et al., 2010; Ma et al., 2012). A review of 20 experimental studies has shown that a peer-to-peer social influence approach (also called a “block leader approach” and hereafter referred to as “social diffusion”) can be highly effective at motivating behavior change in conservation campaigns, with an average increase of 0.82 standard deviations over control (Hedges' g, 95% CI [0.49, 1.16]: Abrahamse & Steg, 2013). One reason this approach may be more effective in promoting behavior change is because it facilitates the diffusion of information through existing social networks (Matous & Todo, 2015; Mesoudi, 2009; Salpeteur et al., 2017; van Vugt et al., 2014). Humans are highly social and are evolutionarily adapted to learn through social interaction (Csibra & Gergely, 2009). Furthermore, people often learn behaviors by directly copying from others, particularly family or those they regard as knowledgeable, prestigious, or successful, for example (Kendal et al., 2018), or simply by adopting the most common behavior among the majority of their peers (Boyd & Richerson, 1985; Richerson & Boyd, 2005).

Yet, despite the promise of a social diffusion approach, research suggests that people who are personally engaged with conservation causes often fail to follow through with reaching out to discuss the issue with others (Barnes-Mauthe et al., 2015; Geiger & Swim, 2016; Niemiec et al., 2018). This reluctance to engage in social diffusion appears to be the result of cognitive biases that influence how people receive and learn information from others (Berl et al., 2021; van Vugt et al., 2014) as well as a variety of psychosocial barriers (Amel et al., 2017), including: (a) the often-incorrect normative perception that others do not care about and are not engaged with the conservation issue (Geiger & Swim, 2016; Jachimowicz et al., 2018; Mildenberger & Tingley, 2019; van der Linden et al., 2015); and (b) low expectations of efficacy in reaching out to others—in other words, the belief that their efforts to share information would not make a difference (Bandura, 1998; Sekar, 2020; Niemiec et al., 2019; Jones & Niemiec, 2020; Niemiec et al., 2021; Mead et al., 2012; Swim & Fraser, 2014; Geiger et al., 2017). We posit that these two barriers to the social diffusion of scientific communication are—in part—issues of insufficient information presented in messaging that, if addressed through intentional design of message content, can encourage more widespread diffusion. While a growing number of studies have identified how cognitive biases and other psychosocial barriers influence the diffusion of information through social networks and people's reactions to that information (van Vugt et al., 2014), few studies have examined experimentally how the content of messaging influences social diffusion (Berl et al., 2021).

We therefore take a multi-faceted experimental approach to assess the effects of messaging that addresses these normative and efficacy-based barriers to the social diffusion of scientific information on biodiversity conservation. Along with social diffusion effects, we also examine the effectiveness of this messaging on individual engagement with scientific information. In the design and conduct of our experiments, we focus on the case of wolf restoration in the state of Colorado. In November 2020, Colorado residents voted on Proposition 114, which mandated that the state wildlife management agency develop and implement a plan to reintroduce wolves to Colorado. The ballot initiative sparked fierce debate and commentary among the public and in the media leading up to the election (Niemiec et al., 2022; Niemiec, Berl, et al., 2020), and the proposition was passed by a narrow margin of less than one percentage point (Colorado Secretary of State, 2020). The public discourse included the spread of false information, such as the misleading claim that wolves are a primary carrier of coronavirus and their reintroduction would lead to an increase in zoonotic infections in humans (Lambert, 2020; Walcher, 2020; Washburn, 2020). Encouraging public engagement with accurate scientific information and the social diffusion of science-based messaging is emerging as an increasingly important issue for conservation science and management, as it is for scientists and educators generally (Hopf et al., 2019; Scheufele & Krause, 2019). Thus, the studies we conducted have relevance to message framing, engagement, and behavior change in conservation, as well as having broader implications for science communication across all fields.

We approached this project driven by two primary research questions, both of which we frame using scientific information about the controversial conservation issue of wolf reintroduction:

• Research Question #1: Does the use of messaging with normative and efficacy-based content increase individual engagement with scientific information?
• Research Question #2: Does the use of messaging with normative and efficacy-based content increase social diffusion of scientific information?

Additionally, given the relative scarcity of literature on the social diffusion of information related to biodiversity conservation, we examine three exploratory research questions to better understand the effectiveness of a diffusion-based messaging strategy and the factors influencing social diffusion, using a combination of quantitative and qualitative methods:

• Research Question #3: To what extent are people contacted through social diffusion more likely to engage with scientific information than people contacted directly by conservation researchers?
• Research Question #4: How do people's demographic and psychographic characteristics affect individual engagement with and the social diffusion of scientific information?
• Research Question #5: What barriers and facilitators do people perceive to the sharing of scientific information?

In our final exploratory question, we examine the effectiveness of the materials we used at changing participants' attitudes and intentions, in order to assess whether the materials themselves served as barriers or facilitators to social diffusion:

• Research Question #6: How does exposure to scientific information about a controversial conservation issue affect people's attitudes and intentions about that issue?

METHODS

In a series of experiments, we distributed messaging to a total of 9091 Colorado residents using a combination of physical mailers, online surveys, and personal messages from close contacts. We used a varied set of participant recruitment methods that tested messaging effects across a range of naturalistic and controlled contexts. Our first experiment utilized the distribution of physical mailers, given conservation organizations' frequent use of mailers to increase awareness and engagement and to motivate conservation action. However, the mailers resulted in low response rates, so we designed our additional two experiments with online and undergraduate participants to recruit up to our target sample size, which we predetermined to have sufficient power to detect potential differences between messaging conditions. Further, the online and student sampling enabled us to gather responses to survey questions to examine additional barriers to sharing behavior and changes in attitudes resulting from our messaging. For each experiment, we tracked participants' individual engagement with the scientific information and their engagement in behaviors that encourage peer-to-peer sharing of the information to assess the effects of the messaging on these two categories of behavioral responses. The overall design of our study and data collection is depicted in Figure 1. We preregistered the initial methodology for our mailer study on OSF Registries, available at: https://osf.io/pf8zh, and expanded upon those methods for our additional studies.

FIGURE 1. Conceptual model of study design. Sample sizes within boxes refer to the total number of individuals contacted and presented with the messaging in each study (left) or through participant referrals (right), or that visited the data collection website (center). Sample sizes on arrows represent counts of participants that followed through to access and engage with the data collection website and the behavioral outcomes, with sources indicated by color: total counts of participants engaging in a behavior are shown in black, participants from Study 1 are in red, participants from Study 2 are in green, participants from Study 3 are in blue, and referrals from other participants are in yellow

Message conditions

We developed two message conditions for use across our series of studies. Both the experimental message and the control message included a brief description of the proposed wolf reintroduction in Colorado, identified the goals of our research group in promoting a science-based understanding of the issue, and called on residents to engage with scientific information about wolves by visiting our website, downloading and sharing informational materials, signing up for a mailing list, and learning about being an advocate for science. The website offered a free sticker promoting human-carnivore coexistence as incentive to visit. The experimental message also presented normative and efficacy-based information related to the social diffusion of scientific information about wolves to convince respondents that: (a) many people were reaching out to others about this issue (i.e., that engaging with others is normative); and (b) reaching out to others would make an impact (i.e., that their engagement with others can be efficacious). Specifically, the normative section of the experimental message reported that 63% of Coloradans had engaged with acquaintances about wolves (drawn from the results of a previous survey: Niemiec, Berl, et al., 2020), and the efficacy section described findings that information retention is enhanced by a factor of 2 and behavior change by a factor of 10 when communicated by an acquaintance (Bollinger et al., 2020; Medley et al., 2009). The order of the normative and efficacy sections in the experimental message was randomized between participants to control for primacy and recency effects. The full text of both conditions is provided in Table S1.

Data collection

Data were collected using a website hosted on Colorado State University (CSU) servers that we designed for this purpose, and that we directed participants to visit in our messaging. The website represented an online outreach and education page for the Center for Human-Carnivore Coexistence (CHCC) at CSU, offering information synthesizing the available ecological and social science associated with wolf restoration. Upon arriving at the website, participants were asked to enter a unique access code provided with their materials, which was used to link their identifying information and the message condition they received to data on their engagement with the website. Participants were presented with an array of behavioral choices to engage with the scientific information on wolf restoration in Colorado. The layout of the data collection website is shown in Figure S1.

The outcome variables for our studies consisted of the behaviors engaged in by participants that visited the website (Table 1). We divided the measured behaviors into two types: “individual behaviors,” which individuals could engage in themselves but had no direct impact on others' engagement or the further diffusion of information; and “social behaviors”, which encouraged or facilitated engagement with others and the spread of information through social diffusion. Participants each had the option to refer up to 10 contacts to the website who were Colorado residents and at least 18 years old by entering their email addresses, which triggered an automated, anonymous email to the contact containing the same message condition initially presented to the referring participant, as well as a new access code. Referrals were linked to their referrer, so that the diffusion of the message conditions could be tracked along chains of participants.

TABLE 1 Behavioral outcome variables collected from study participants

Behavioral outcome variable	Variable description	Variable type	Behavior type
Access information sheets	Followed link to access CSU Extension information sheets with scientific information about wolf management and restoration	Dichotomous	Individual
Sign up for mailing list	Signed up for CHCC mailing list to receive information on wolves and human-carnivore coexistence	Dichotomous	Individual
Follow link to CHCC website	Followed link to CHCC main website	Dichotomous	Individual
Follow link to CHCC Twitter	Followed CHCC social media link	Dichotomous	Individual
Request CHCC sticker	Requested to receive free CHCC-branded sticker advocating for human-carnivore coexistence	Dichotomous	Social
Receive information on outreach	Requested to receive email with guidance on sharing information with others and on engaging in outreach	Dichotomous	Social
Refer others to website	Referred any friends or family to receive an anonymous email with a link to the webpage	Dichotomous	Social
Number of referrals to website	Total number of friends and family referred	Discrete	Social

Note: The type of behavior represented by each variable is listed as either “individual” for personal engagement with the information or “social” for behavior that encouraged the social diffusion of the information to other people. Two composite (discrete) variables were also created that summed all of the dichotomous variables within a behavior type for each participant (see Section 2).

Study 1: mailers

We contracted a local direct mail marketing service to distribute 8000 mailers to Colorado residents across the Front Range and Western Slope regions in July 2020. Our target sample size was established using power analyses of a set of pilot data assuming a 1% follow-through rate and a desired power of 0.8 and a significance level of 0.05 for a logistic regression model. Two different regions of Colorado, the Front Range and the Western Slope, were targeted because of demographic and ideological differences that could potentially lead to differences in the rates and types of engagement with the experiment (see Figure 1 of Niemiec, Berl, et al., 2020 for a map of the Colorado counties included in each region). Specifically, prior surveys have found that support for wolf reintroduction is lower in the more rural Western Slope of Colorado where wolves would likely be reintroduced (Niemiec et al., 2022; Niemiec, Berl, et al., 2020). Mailer recipients were identified using prospect street address lists (i.e., lists of individuals that have not been contacted prior and have not signed up for a specific service) provided by the direct mail marketing service. The samples consisted of randomly selected postal routes within each of the two regions, with a mixture of urban and rural routes.

We presented the messaging described above in Message Conditions (section 2.1), and other relevant information such as the access code for the website, on the back of the mailer (designs shown in Figure S2). Message conditions were randomly assigned such that, within each of the two regions, 2000 residents received the control condition, 1000 received the experimental condition with the normative section first, and 1000 received the experimental condition with the efficacy section first. Three mailers (all Western Slope region with the experimental condition and the efficacy section first) were returned as undeliverable, for a total sample of 7997 residents.

Study 2: online panel

We designed an online survey using Qualtrics (see Appendix S1) that we presented to Colorado residents recruited through the Amazon Mechanical Turk platform directly (n = 32) and through online participant pools sourced by CloudResearch (n = 384) in October 2020 and November 2020 up to and including election day (November 3rd). Strict qualifying requirements and a mid-survey attention check were used to ensure adequate data quality (Chandler et al., 2019) and that participants were Colorado residents of at least 18 years of age. Participants were asked about their attitudes toward wolves and their voting intentions on the ballot proposition mandating wolf reintroduction (adapted from Niemiec, Berl, et al., 2020). They were then presented with one of the two randomly assigned message conditions that were previously used in the mailers (as described above). Participants were instructed to visit the data collection website using a provided unique access code, but were informed that engaging with anything on the website was optional. After visiting the website, the attitude and voting intention questions were repeated for a post-exposure assessment, and participants were asked about their demographics, self-identification with interest groups, and perceptions about sharing information on wolves with others (also adapted from Niemiec, Berl, et al., 2020).

Study 3: student contacts

We designed a separate online survey using the Qualtrics platform (see Appendix S2) that was administered to 238 undergraduates at Colorado State University enrolled in introductory courses in psychology or natural resources during September and October 2020. This study differed substantively from Study 1 and Study 2 in that our goal was for the student participants to deliver the messaging to their close contacts whose engagement with the data collection website would be measured, rather than the students themselves being the subjects of the study. The online survey presented each student with a randomly assigned message condition (same as above) and asked them to send short messages about the website to three personal contacts that were at least 18 years old and also currently lived in Colorado. For these messages, students were required to include provided text drawn from the message condition that they received, a link to the website, and a new unique access code provided to them for each recipient. They were asked to personalize each message for the recipient and to indicate on the survey how they knew each contact (friend, family member, coworker, etc.) and how they intended to contact them (email, text, social media, etc.). The 238 students contacted a total of 678 Colorado residents with the messaging.

Analysis Research Questions #1 and #2: effects of messaging on engagement with scientific information

To test the questions of whether the use of messaging with normative and efficacy-based content increased individual or social behavioral engagement with the scientific information, we use binomial logistic regressions to model each behavioral outcome variable with a dichotomous response (see Table 1) as a function of message condition. Because we expect a large number of non-responses to create a “rare events” problem for traditional logistic regression (King & Zeng, 2001), we use Firth's bias-reduced penalized-likelihood logistic regression (Heinze & Schemper, 2002), implemented in the “logistf” package for the R statistical environment. Due to the limitations of our methodology, we assume in our models that all participants viewed the materials and were exposed to the messaging, so non-response is treated as a behavioral choice not to engage.

To model the only measured behavioral outcome variable with a discrete or count response (“Number of referrals to website”), we use a zero-inflated negative binomial model to account for the excess non-responses in the data, using the “pscl” R package. This model allows us to specify separate sources of zero values: first, from non-responses, modeled as a binomial process; and second, from visiting the website but not engaging in the social behavior of referring others, modeled as a negative binomial process. We use negative binomial models rather than Poisson due to observed overdispersion in the count data. Both model stages include message condition as a predictor.

For Research Question #2, we planned to conduct an additional test of referral count by message condition, geographic region, and the position of that participant in the diffusion chain (“generation” in Mesoudi et al., 2006) using a zero-inflated negative binomial model. However, no individuals that were referred to the website by participants carried through with referring additional participants themselves, leading to no chains having more than one step of diffusion.

To examine the effect of messaging on the overall amounts of individual and social behaviors rather than on the likelihood of a single behavior, we created composite variables for all individual behaviors and all social behaviors. For each of the two behavior types, we sum the number of behaviors engaged in by each participant as a count and fit a zero-inflated negative binomial model (as above), using message condition as a predictor.

In all models of the mailer data from Study 1, we include geographic region and its interaction with message condition as additional predictors, due to the potential for differential responses to the messaging and the topic of wolf reintroduction based on political and demographic differences (refer to Section 2.2).

For each group of models fit to the same data set (e.g., the set of models of every behavioral outcome for the mailer data from Study 1), we adjusted p-values using Holm's method to correct for the family-wise error rate (Holm, 1979). We considered p-values <.05 to be statistically significant.

Research Question #3: effects of information source on engagement with scientific information

Using the composite individual and social variables above, we fit zero-inflated negative binomial models of each behavior type with study sample as a predictor, to test for differences between Study 3, where participants were contacted by an acquaintance, and Study 1 and Study 2, where participants were presented with the information by the researchers.

Research Question #4: effects of individual characteristics on engagement with scientific information

For Study 2, where we were able to collect demographic and psychographic information from all participants, we fit an additional binomial logistic model or zero-inflated negative binomial model of each outcome behavior and composite variable (as above), using the following variables as predictors: age, gender, education, income, community size, and whether the participant identifies as a wildlife advocate, animal rights advocate, gun rights advocate, hunter, or conservationist. We chose these variables given prior work (Niemiec, Berl, et al., 2020; Niemiec et al., 2022) which found that support for wolf reintroduction in Colorado varied by these factors.

Research Question #5: perceptions of facilitators and barriers to social diffusion of scientific information

In Study 3, we asked the student participants that distributed the scientific information to their contacts: “What is something that you found interesting, difficult, or effective about reaching out to others to share science-based information?” To evaluate responses to this question for factors that could have either facilitated or hindered the social diffusion of scientific information on wolves, we use inductive thematic analysis (Braun & Clarke, 2006; Clarke & Braun, 2014), a method for identifying and describing themes in qualitative data. Together, two coders used a semantic approach to thoroughly evaluate all responses and generate initial codes, grouped these codes into common themes, and generated additional codes to cover the participants' and the recipients' experiences, for a total of 16 codes (see Table S2 for descriptions and examples). A third coder charted all responses on a spreadsheet, indicating whether a response exhibited a presence or absence of each developed code. To confirm the reliability of the coding system, a fourth coder independently coded a random sample of 25 responses. A comparison of the third and fourth coders' responses yielded an intercoder agreement of 90.1%. The third and fourth coders reviewed the non-concurrent responses until 100% agreement was reached. In our analyses, we examine the frequencies of these themes.

Research Question #6: effects of messaging and scientific information on attitudes and intentions

In Study 2, we asked participants a set of questions prior to their exposure to the messaging and to the scientific information on wolves, and again following exposure, on: their general attitude regarding wolves (“dislike a great deal” to “like a great deal”); their perceptions of the likely overall impact of wolf reintroduction (“extremely negative” to “extremely positive”); their voting intention on the ballot proposition (“vote for,” “vote against,” or “not vote for or against”); and their degree of certainty in how they would vote (“not at all certain” to “extremely certain”). We conduct paired t-tests on individuals to examine whether these perceptions changed in the short term pre- and post-exposure to the information.

To assess whether message condition or gender (since gender proportions were found to differ between conditions) are predictive of changes in attitudes, we construct individual-level multiple linear regression models using the amount of change in general attitude regarding wolves and in perceptions of the likely overall impact of wolf reintroduction as outcomes, with message condition, gender, and the interaction between these two terms as predictors. To assess the impact of these variables on individual voting intention, we use a Cochran–Mantel–Haenszel test on contingency tables of the amount of change (integers ranging from −2, representing a change of voting for the ballot proposition to voting against it, to +2, representing a change of voting against to voting for, with intermediate values representing changes to or from the intention not to vote) crossed by message condition and gender. For post-hoc groupwise testing of voting intention, we use Fisher's exact test with Holm's method for p-value adjustments.

RESULTS Response characteristics

Across all three studies and including participant referrals, our final sample consisted of 9116 Coloradans contacted with messaging materials (Figure 1 and Table S3). Of these, 649 accessed the data collection website, for an overall follow-through rate of 7.1%. Follow-through rates were considerably lower for physical mailers (0.71%), consistent with the 0.5%–1% range predicted for direct mail marketing services (Direct Marketing Association, 2018), and highest for the online sample (96.9%), where participants were specifically instructed to visit the website as part of their participation. Follow-through for people contacted by students was intermediate between the other two samples, at 27.6%. Of the individuals that were sent referrals by existing participants, 8.0% accessed the website.

Across both message conditions, the most frequent behaviors among participants that visited the website were to access the information sheets with scientific information on wolves (45.5% overall) and to request a free sticker (31.0%; Table 2). Other behaviors were much less common, with the fewest participants following the links at the bottom of the website to the CHCC Twitter page (0.46%) or the main CHCC homepage (2.2%). Overall, 427 participants (211 experimental, 216 control) engaged in one or more outcome behaviors on the website beyond visitation.

TABLE 2 Behavioral engagement by message condition

Behavior	Number in experimental condition (and percent of visitors in condition)	Number in control condition (and percent of visitors in condition)	Total count (and percent of visitors in condition)
Visit website	326	323	649
Access information sheets	152 (46.6%)	143 (44.3%)	295 (45.5%)
Request CHCC sticker	93 (28.5%)	108 (33.4%)	201 (31.0%)
Sign up for mailing list	28 (8.6%)	39 (12.1%)	67 (10.3%)
Receive information on outreach	10 (3.1%)	21 (6.5%)	31 (4.8%)
Refer others to website	3 (0.92%)	21 (6.5%)	24 (3.7%)
Follow link to CHCC website	8 (2.5%)	6 (1.9%)	14 (2.2%)
Follow link to CHCC Twitter	1 (0.31%)	2 (0.62%)	3 (0.46%)

Note: Counts by message condition and across both conditions are shown, with the percentage of participants that engaged in the specified behavior out of those that visited the website shown in parentheses.

Demographic data were not available for the individuals contacted for Study 1 or Study 3 except on a voluntary basis on the data collection website, so we were unable to assess balance of demographic characteristics across message conditions or demographically-based nonresponse bias in these samples. In Study 2, we found no differences between samples by message condition in any characteristic measured (age, ethnicity, education, income, community size, or identity) except for gender, where a greater number of males were present in the experimental condition (logistic regression, excluding the three non-binary gender individuals in each condition: X² (1, N = 416) = 4.522, p = .033, odds ratio = 1.596 [95% CI: 1.040, 2.467]). This imbalance occurred solely due to chance, since there was no possibility for self-selection into conditions.

Research Questions #1 and #2: effects of messaging on engagement with scientific information

In regressions of the behavioral outcomes on message condition, we found no significant increases in behavioral response under the experimental (i.e., normative and efficacy-based) condition in any study sample, including analyses of each behavior separately and of combined counts of individual and social behaviors (Table S4). This finding remained consistent when the same analyses were performed on the complete data set combining participants' engagement across all study samples and referrals (n = 9116). Two models instead indicated a significant increase in behavioral response under the control condition: referring others in the online sample (odds ratio [OR] = 0.174 [95% CI: 0.034, 0.591], p = .047) and the total number of social behaviors in the online sample (incidence rate ratio [IRR] = 0.392 [95% CI: 0.244, 0.630], p = .002). We also found a significant interaction between message condition and geographic region on the total number of individual behaviors in the mailer sample (IRR = 0.003 [95% CI: 0.000, 0.023], p < .001), indicating that participants in the Western Slope engaged in more individual behaviors under the control condition.

Across all samples, three participants in the experimental message condition and 15 in the control message condition referred one or more other people to the website. Fewer individuals in total were referred to the website by participants in the experimental condition (n = 4) than in the control condition (n = 21). Among referred individuals, 0 followed through to visit the website in the experimental condition (0%) and 2 in the control condition (9.5%). No referred participants went on to refer any additional people.

Some additional individuals were referred by student participants visiting the website (experimental: n = 6; control: n = 16) or by online participants that failed to complete the survey (experimental: n = 36; control: n = 9). Of these, 1 referred individual in the experimental condition (2.38%) and 2 in the control condition (8.0%) followed through to visit the website. None referred additional people. Because their referrers were not considered valid participants in our studies, these referrals are also excluded from our analyses and from sample counts.

Research Question #3: effects of information source on engagement with scientific information

Total engagement across both message conditions differed significantly between studies for individual behaviors (X²[2, N = 9091] = 202.1, p < .001) and for social behaviors (X²[2, N = 9091] = 16.34, p < .001). Participants contacted personally by students engaged in more individual and social behaviors on average than participants contacted directly by researchers using mailers (Table S5). Online participants engaged in more of both types of behaviors than mailer participants or student contacts; however, online participants were explicitly asked and paid to visit the website (though were told they were not required to engage further in any of the measured behaviors tested here).

Research Question #4: effects of individual characteristics on engagement with scientific information

In the online study, we found no demographic or psychographic variables to be significant predictors of any behavior, except that participants identifying as hunters were less likely to access the information sheets with scientific information on wolves (OR = 0.525 [95% CI: 0.297, 0.919], p = .024). There was insufficient variation among demographics and psychographics in participant referrals, in the total number of individual behaviors, and in the total number of social behaviors in this sample to conduct tests of these associations.

Research Question #5: perceptions of facilitators and barriers to social diffusion of scientific information

Of the students who participated in Study 3 by sharing scientific information with their personal contacts, 172 responded to our open-ended qualitative question regarding their experiences doing so. Out of these respondents, only 3 (1.74%) felt comfortable reaching out to others about the information, with 144 (83.7%) expressing a neutral opinion and 25 (14.53%) that reported feeling awkward or uncomfortable. The concerns of respondents who expressed negative feelings about sharing information ranged from statements of general social discomfort, such as “this is not the kind of information that you send to friends or family,” to admissions of fear and anxiety, as in “it was a little difficult for me to share science-based information to be honest, since a lot of people I know are from a different political spectrum and would get angry if I texted them information that would share a different view.” Aside from social awkwardness or discomfort, the most frequently noted barrier for sending information was the text itself: respondents reported experiencing “difficulty rewording it so it sounded like me,” or so that it did not “sound too sciencey,” or so that their contact “didn't think it was spam.”

After sharing the messaging, 15 (8.72%) student respondents were surprised by the positive reactions of the message recipients. These students attributed the positive reactions either to the fact that the information was shared by someone they knew or to the personalization of the message itself. Text messages were the preferred modality for contact at 77%, with respondents acknowledging its more casual nature.

Research Question #6: effects of messaging and scientific information on attitudes and intentions

Pre- and post-testing in the online sample indicated significant shifts within individuals toward positive general attitudes toward wolves (pre: M = 5.50, SD = 1.41; post: M = 5.62, SD = 1.33; t[415] = −3.05, p = .002), perceptions of positive impacts from wolf restoration (pre: M = 4.96, SD = 1.69; post: M = 5.14, SD = 1.64; t[415] = −4.51, p < .001), and intent to vote in favor of the ballot proposition to reintroduce wolves (pre: n = 284; post: n = 306; McNemar's X²[3, N = 416] = 17.64, p < .001) following engagement with the website.

We found that neither message condition nor gender (excluding the three non-binary gendered individuals in each condition) were significant predictors of the amount of change in attitude toward wolves (F[3, 406] = 0.709, p = .547) or of the amount of change in perceived impact of wolf reintroduction (F[3, 406] = 0.213, p = .887). We did find a significant difference in the amount of change in voting intention across message conditions and genders (X²[4, N = 416] = 9.983, p = .041). This difference appeared to suggest a greater number of men changing to more affirmative voting intentions under the experimental condition and a greater number of women changing to more affirmative voting intentions under the control condition; however, post-hoc tests grouped by message condition (within experimental and within control: p = .054) or by gender (within female: p = .128; within male: p = .621) failed to find a significant association when accounting for the other variable.

DISCUSSION

Encouraging people to share scientific information with others through their social networks is critical for enhancing the speed and scale of conservation action. However, little is known about how messaging can help facilitate this social diffusion. The results of our three studies, conducted across a diverse set of experimental contexts, do not provide any evidence of increased social diffusion of scientific information about a conservation issue from addressing normative and efficacy-based barriers through the content of messaging. Moreover, we observed that there were no significant effects from our messaging on individual-level engagement. Our results add to a growing list of studies in conservation psychology that have found the manipulation of normative or efficacy-based cues in messaging for biodiversity conservation to be less effective than anticipated (Byerly et al., 2019; Niemiec et al., 2021; Niemiec, Sekar, et al., 2020), given the impact of these strategies in other domains (Abrahamse & Steg, 2013; Witte & Allen, 2000).

There are numerous potential explanations for our observed null (and, in some cases, negative) effects of normative and efficacy-based messages on individual and sharing behavior. For example, it is possible that participants did not believe the messages, did not read the messages in depth, detected the manipulation of the experimental messaging, or reacted to a loss of freedom in being subjected to an experiment (Steindl et al., 2015). Furthermore, there are a number of factors that could affect the potency of normative and other psychosocial influences on behavior, including the type of norms being used (Niemiec, Champine, et al., 2020) and a myriad of other barriers, motivations, and environmental conditions influencing people's decision-making (Schultz, 2014). It is possible that beliefs about norms and efficacy were not the most powerful or salient barriers to participants' individual engagement or willingness to share scientific information with others. Additionally, we did find short-term changes in online participants' attitudes and intentions following exposure to the messaging, but the mixed results of groupwise tests on the amount of change in voting intention suggest that there could be gender-based differences in individual responses to different messaging strategies that merit further investigation.

In our qualitative analysis of students' experiences, we identified two particularly salient barriers to sharing scientific information with friends and family. First, many students expressed anxiety about sharing information because they believed this sort of sharing was not socially acceptable and that others would respond negatively. While we sought to address this barrier in our normative messaging by indicating that discussions on wolf reintroduction were common, it is possible that participants did not find this message or its presentation believable, or did not think that the findings applied to their own social circles. Second, participants expressed that the wording they were asked to share was too formal or did not sound natural, which made the experience of reaching out to others uncomfortable. More intensive and individualized interventions may be necessary to address these barriers. For example, workshops could be used to bring together groups of volunteers and facilitate discussions on how they could share scientific information with others in the course of natural conversation and help create a supportive community and generate new social norms around the sharing of scientific information (as described in Niemiec et al., 2019). Our findings suggest the need for more work in this area to design and test alternative messaging approaches to promote social diffusion under varying contexts.

While we did not find evidence of a positive effect of normative and efficacy-based messaging in promoting social diffusion over control, we found suggestive evidence that social diffusion could be effective at promoting behavior change if people can be convinced to share information with others in their social network. Specifically, we found that participants contacted personally by students engaged in more individual and social behaviors on average than participants contacted directly by researchers via mail. Moreover, the follow-through rate for personal contacts was vastly improved above that of mailers (27.6% vs. 0.71%). These exploratory results follow previous studies and meta-analyses that have found powerful effects of peer-to-peer contact and social diffusion at promoting behavior change (Abrahamse & Steg, 2013; Burn, 1991; Hopper & Nielsen, 1991). However, there are important limitations that prevent drawing strong conclusions from our comparisons. Specifically, in our case, information source (researchers, in Study 1, vs. personal contacts, in Study 3) was heavily confounded with the population being sampled (general Colorado residents vs. the social networks of university undergraduates), the sampling strategy (random vs. intentional selection), and the mode of contact (mail vs. predominantly text messages). Additionally, students were requested to personalize their messages while the content of the mailers remained consistent. Thus, further studies are needed in this area to explore the potential impact of social diffusion for biodiversity conservation by comparing the effectiveness of messages delivered through different methods while controlling for other relevant factors.

Future research could also examine how engagement with and the sharing of scientific information could be enhanced through messaging approaches that build on strategies from other fields, such as community-based social marketing (McKenzie-Mohr & Schultz, 2014). This area of research investigates the importance of tailoring messages to the values of specific target audiences, identifying key opinion leaders of those audiences to deliver the message, and harnessing the power of cognitive biases, such as the scarcity heuristic, to promote behavior change and engagement (Kusmanoff et al., 2020). This literature has also highlighted the power of tailored outreach approaches and peer-to-peer communication in promoting conservation behavior change (Green et al., 2019; Wright et al., 2015).

Finding effective means to motivate human behavior change for conservation issues is one of the highest priorities for modern conservation science. Theory on social diffusion from across the social sciences, but especially from the largely untapped interdisciplinary field of cultural evolution, has not yet seen widespread application to environmental issues, but is beginning to be recognized as a vital component of efforts in sustainability science and conservation behavior change (Brooks et al., 2018; van Vugt et al., 2014; Waring, 2010). Because our messages were ineffective at encouraging participants to share information with others and did not spread beyond one step of diffusion through social networks, our data are insufficient to address the question of how best to design messaging that is maximally effective at propagating itself through social diffusion, so further work devoted to this topic is urgently needed. A better understanding of the potential influences of social diffusion on conservation behavior will allow the development of improved methods of utilizing messaging and public appeals for conservation action, which can serve as important tools for applied conservation efforts and the development of optimal policy (Kinzig et al. 2013; Barnes et al., 2016). Our studies highlight the critical need to develop reliable methods to scale up biodiversity conservation efforts from individual engagement to widespread social diffusion and collective action to confront present and future conservation challenges.

ACKNOWLEDGMENTS

We wish to thank the Center for Human-Carnivore Coexistence at Colorado State University, and especially Kevin Crooks, for valuable support and feedback on this project. We thank Stacy Ray and United Mailing for proofing, printing, and distributing the mailers, Stephen Oglesby, Dhruv Padalia, and Meetkumar Savaliya in the Warner College of Natural Resources for assistance in developing and implementing the data collection website, and Kristina Maldonado Bad Hand (https://linktr.ee/Kristina_Maldonado_Bad_Hand) for designing the sticker used as incentive for participants. We thank the National Geographic Society (NGS-59650R) and an anonymous donor to the CHCC for providing funding for the studies involved in this project. We thank the editor and two anonymous reviewers for their contributions to the improvement of this manuscript.

CONFLICT OF INTEREST

The authors declare that they have no conflicts of interest.

AUTHOR CONTRIBUTIONS

Richard E.W. Berl, Samantha Sekar, and Rebecca M. Niemiec conceived of and designed the research. Richard E.W. Berl acquired the data. Richard E.W. Berl, Anyll Markevich, Cassiopeia Camara, and Rebecca M. Niemiec analyzed and interpreted the data. Richard E.W. Berl drafted the manuscript. All authors contributed to critically revising the manuscript, gave final approval of the version to be published, and agree to be accountable for all aspects of the work.

DATA AVAILABILITY STATEMENT

All data and R scripts used for analysis are in a public OSF project titled “Scientific Messaging Experiment,” available at: https://osf.io/zn3vx/.

ETHICS STATEMENT

Prior approval for research protocols was obtained from the Colorado State University Institutional Review Board (protocol #19-8662H). Consent language describing the study was included on all materials distributed to individuals, including information on how to opt out of the study. We offered a free Center for Human-Carnivore Coexistence branded sticker as incentive for participating in the study. Online participants were compensated monetarily for their time and participating students received course credit or psychology research participation credit.