Introduction

Social influence and sustainability

Social influence is one of the core tools for scaling up engagement in the kinds of sustainability behaviors that will help achieve urgent global social and ecological goals (Amel et al. 2017; Sparkman et al. 2021a). Social influence occurs when people’s thoughts and actions are changed by what they observe others around them are doing and saying, which activates innate human needs to conform and belong (Festinger 1954; Cialdini and Goldstein 2004). A recent meta-analysis accounting for publication bias found that social influence had a generally positive effect on pro-environmental behavior through processes such as social comparative feedback (Nguyen-Van et al. 2021). This confirmed results of a previous meta-analysis that found social influence interventions were effective, compared to a control, at shifting resource conservation behaviors (Abrahamse and Steg 2013). A third meta-analysis of social marketing campaigns identified interpersonal communication among peers as a critical pre-condition for conservation behavior change. It found, moreover, that mean campaign effect size was highest for interpersonal communication compared to other social marketing campaigns (Green et al. 2019). This suggests social influence is an environmentally important behavior in its own right.

Social influence can be harnessed for sustainability when ordinary people are encouraged to engage in relational organizing. Relational organizing is defined as motivated individuals encouraging others they know to engage in a desired behavior (Jones and Niemiec 2023). This encouragement can take many forms, including sharing information that a behavior is possible, actively seeking to persuade someone else to try the desired behavior, reminding others when it is time for a behavior to occur (e.g., voting in an election), creating and reinforcing a new norm about the desired behavior, and more. The term relational organizing emerged from political activism, but the concept has been studied in the sustainability space as social environmentalism (Larson et al. 2015), diffusion behavior (Jones and Niemiec 2020), the block leader approach (Abrahamse and Steg 2013), peer persuasion (Maki and Raimi 2017), peer learning (Ma et al. 2012), and peer effects or peer influence (Wolske et al. 2020).

Across these different studies, what characterizes relational organizing is (a) an attempt by the individual to influence others to adopt a behavior, and (b) the persuasion or encouragement occurs within existing social networks. As Ma et al. (2012) point out, sometimes “peers” have been used to include “relative strangers who just happen to be involved in similar activities under similar conditions,” who come into contact with each other through organizational efforts or happenstance. However, in this study, we are most interested in social influence targeting close others with whom people already have social bonds. These existing close ties are often a more salient comparative group (Festinger 1954), and norms among this group have been more predictive of personal behaviors such as alcohol consumption (Cho 2006; Cox and Bates 2011), as well as environmental beliefs like support for climate change mitigation among political conservatives (Goldberg et al. 2020). As Wolske et al. (2020) state, much of the research on peer effects has focused on the process of influence rather than the relational organizer’s intentions or approach, about which Maki and Raimi (2017) note we know “shockingly little.”

Relational organizing can be a powerful tool for a wide variety of political and social causes. In a political field experiment, for example, relational organizing led to the largest effect in an experimental get-out-the vote study in 2 decades (Green and McClellan 2020). A 2-year health study comparing a traditional outreach intervention to a peer-driven approach found that peers recruited 36% more participants to an HIV prevention program than the control, was more representative of the target population, and led to more learning among participants (Broadhead et al. 1998). Over another 32-year study, smokers’ chance of smoking decreased by 67%, 36% and 34%, respectively, if their spouse, friend, or coworker quit smoking, suggesting strong diffusion effects from close ties within a social network (Christakis and Fowler 2008).

Encouraging deliberate social influence has also been shown to be effective in the environmental realm. A meta-analysis of 29 experimental studies found that relational organizing (specifically a “block leader approach”) was the social influence approach most effective at encouraging resource conservation (Abrahamse and Steg 2013). An experimental study of sustainable consumer purchasing found that friends and family sharing information about their own adoption of sustainable products had a stronger and more enduring effect than simply being informed by researchers that their colleagues had chosen these products (Salazar et al. 2013). Hearing neighbors and acquaintances talk about their experience installing solar panels has been linked to a greater likelihood of people installing solar panels themselves (Palm 2017).

Relational organizing and climate action

Relational organizing may be particularly important for large-scale challenges like climate change. Climate action is often framed as either personal or political action. Amidst such discourse, relational organizing is a neglected tier in the middle, where individuals seek to promote personal pro-environmental behaviors to others in their social networks. Often these climate actions have direct benefits to the potential adoptee (the target of relational organizing), who might benefit from a healthier plant-forward diet, reduced wildfire risk, or cost savings from energy and water conservation, in addition to their broader benefits to the wider social–ecological community. Given the global nature of climate change, individuals may feel helpless to influence political and economic systems but still perceive that they can have impact at the scale of social networks (Meyer et al. 2020).

One arena of climate action where deliberate social influence is still underexplored is people’s participation in food systems. Reducing food waste and switching to low-carbon diets are widely recognized as meaningful climate actions (Abrahamse 2020). Beyond its climate benefits, a global shift from meat-forward to plant-forward diets can help reduce the negative ecological impacts of land conversion for intensive animal agriculture (Schiermeier 2019; Ritchie 2020) and reduce the harm to animals and humans associated with meat supply chain (Fraser 2008; Slade and Alleyne 2021). There are, furthermore, tangible human health benefits from limiting consumption of animal products, particularly red and processed meats (Godfray et al. 2018).

More and more attention is being given to the role of food in combating climate change, with a focus on the benefits of meat reduction and more plant-based diets. While food systems, food subsidies, and national dietary standards play a major role in a population’s dietary choices, so do many personal, social, and cultural factors (Markowski and Roxburgh 2019; Kurz et al. 2020). In the present work, we focus on the important role of social factors, honing in specifically on factors that lead people to promote plant-based diets to others as a case study of the social diffusion of climate action.

Social–psychological drivers of relational organizing

A key challenge to implementing a relational organizing approach is scaling it up, or encouraging a larger number of individuals to reach out to others in their social network. Environmental organizations often have limited reach and risk preaching to the choir of people who are already interested in a topic. Those interested individuals, in turn, have opportunities to reach audiences in their social networks that environmental organizations cannot reach, and opportunities to integrate behavior change conversations into daily interactions. Yet, what data that do exist suggest that often people who are committed to participating in a sustainability behavior in their personal lives may be reluctant to reach out to others about this behavior. For instance, people who care about climate change have been found to be reluctant to discuss it with others (Geiger and Swim 2016). Similar patterns emerge for people engaging in other pro-environmental behaviors, such as encouraging others to manage invasive species on their property (Niemiec et al. 2018) or to recycle (Nolan 2013).

A drop-off in engagement between personal and outreach behavior may be because relational organizing is a distinctly social type of behavior: relational organizers are deliberately trying to prompt their friends, family, colleagues or other acquaintances to try a new, likely counter-normative, sustainable behavior. As such, relational organizing evokes a wide range of social–psychological motivations and barriers, which we cluster into four categories here (Fig. 1). First, there are beliefs related to the sustainable behavior itself. These include attitudes about the sustainable behavior, individuals’ self-efficacy in doing the sustainable behavior (i.e., their belief in their ability to engage in the task successfully; Bandura 1998), and moral hypocrisy beliefs, i.e., a concern about perceived consistency or inconsistency between personal and relational organizing behaviors (Mullen and Monin 2016).

Fig. 1
figure 1

Four categories of social–psychological beliefs related to relational organizing, derived from previous research

Full size image

Preliminary research suggests people’s willingness to reach out to others is influenced by the strength of their attitudes and beliefs related to the personal behavior (category 1, Fig. 1). Howell et al. (2015), for example, found that the strength of bait shop owners’ attitudes towards invasive species were associated with their intentions to communicate with customers about preventing the spread of invasive species. Champine et al. (2022) found that attitudes towards native plant gardening were strong predictors of intentions to reach out to others about native plant gardening. Relational organizing self-efficacy has been found to predict reaching out to others, both in self-reports (Jones and Niemiec 2020) and in a field experiment (Niemiec et al. 2021a). Concerns about moral hypocrisy might also inhibit relational organizing about sustainability behaviors, as moral hypocrisy can be triggered by a perceived misalignment between speech and action that is accompanied by a claim to undeserved moral benefits (Effron et al. 2018).

Second, there are an individual’s beliefs about themselves as a relational organizer (category 2, Fig. 1). This includes personal norms, which are feelings of moral obligation to keep to certain standards of behavior that derive from their values (Schwartz 1977). Personal norms related to relational organizing capture whether people believe it is their role to encourage others or that convincing others is the right thing to do, and significantly predicts social diffusion intentions (Champine et al. 2022). This category also includes moral exporting, i.e., people’s willingness to try to persuade others to adopt their own values, which has been associated with willingness to engage in two-way conversations about environmental issues and to confront transgressors (Maki and Raimi 2017). In addition, people may vary in their level of respect for autonomy, defined as a belief in others’ intrinsic right to control their own decisions, which in one study was the most common response participants gave for not encouraging others their friends or family to try plant-based alternatives to animal products (Niemiec et al. 2021b). Lastly, individuals’ social identity is a powerful predictor of peer influence behavior. In the social identity model of collective action, people’s subjective sense of identification with a group (e.g., political, vocational, or stigmatized) predicted political activism and moderated the feelings of affective injustice and group efficacy beliefs, such that only those who identified with the group were moved to organizing (van Zomeren et al. 2008). In the case of relational organizing around food choices, social identity either as an activist or as a vegan/vegetarian may both be salient (Kurz et al. 2020).

Third are interpersonal beliefs, which we define as perceptions of others’ beliefs about the issue and about relational organizing related to the issue (category 3, Fig. 1). While there is substantial research on the role of interpersonal beliefs, especially social norms, in driving personal behavior, only recently have studies begun examining the effect of interpersonal beliefs on the peer influence behaviors needed for relational organizing (Geiger and Swim 2016; Niemiec et al. 2019; Jones and Niemiec 2020). For instance, Nolan (2013) suggests a key barrier to individuals reaching out to environmental transgressors may be a perceived lack of supportive social norms about approaching others. Other research suggests that people are often concerned they will be perceived by others as unlikeable or incompetent (i.e., reputational concerns; Geiger and Swim 2016; Jones and Niemiec 2020) or be faced with social sanctions (Niemiec et al. 2018) if they encourage others to change. People may also refrain from relational organizing if they anticipate psychological reactance from their audience, with others perceiving relational organizing as a threat to their freedom and acting against it (Reynolds-Tylus 2019).

Additional findings suggest that people are influenced by inaccurate perceptions of others’ beliefs when making decisions about whether to discuss environmental issues with others (Geiger and Swim 2016). Some research demonstrates that people believe public support for environmental causes is lower than it actually is (i.e., pluralistic ignorance), which can lead to self-silencing (Geiger and Swim 2016; Sparkman et al. 2022). Social norms have been categorized in many ways; here, we divide norms across three axes. First, whether they are descriptive (beliefs about what others are actually doing) or injunctive (beliefs about what others would approve or disapprove) (Cialdini 2003); second, whether they are static (belief that the norm is unchanging) or dynamic (belief that the norm is becoming more or less common) (Sparkman and Walton 2017); and third, whether they are about the personal sustainable behavior or about relational organizing.

Finally, a fourth contributing factor may be beliefs about the efficacy of relational organizing (category 4, Fig. 1). Research suggests that people are influenced by two types of efficacy beliefs when considering relational organizing: self-efficacy and response efficacy. When deciding whether to relationally organize, people may be influenced by their perceived ability to convey a message to others effectively and competently, i.e., relational organizing self-efficacy (Geiger et al. 2017; Jones and Niemiec 2020; Champine et al. 2022). Indeed, Howell et al. (2015) found that bait shop owners’ intentions to reach out to their customers about invasive species management was influenced by their perceived self-efficacy to effectively inform customers. Geiger et al. (2017) found that interventions that bolstered people’s self-efficacy in discussing climate change increased their willingness to do so.

Studies suggest people are also influenced by response efficacy beliefs, or perceptions of the effectiveness of reaching out to others at influencing both others’ behavior as well as the desired societal outcomes (Nolan 2013; Jones and Niemiec 2020). Champine et al. (2022) and Jones and Niemiec (2020), for example, found that when reaching out to others about native plant gardening, residents were motivated by both social response efficacy, or their perceived ability to affect others’ behaviors (studied elsewhere as expected reciprocity; Lubell et al. 2007), and their environmental response efficacy, or their perceived ability to achieve conservation impacts by affecting others. Similarly, Guckian et al. (2018) found that the perceived efficacy of social sanctioning was the biggest predictor of catch-and-release anglers’ intent to sanction other anglers who were engaging in harmful fishing practices.

The present research

While past research has touched upon many possible contributing factors to relational organizing as synthesized above, what remains unclear is the degree to which these different social–psychological beliefs have differing levels of discrete influence on engagement in relational organizing. The coherence of these categories has not yet been tested, nor has there been integrative research comparing their predictive power on relational organizing behavior. There is additional uncertainty about whether these higher level categories are more or less predictive of relational organizing compared to discrete social–psychological beliefs. Lastly, it may be that certain combinations of beliefs are necessary to adequately predict relational organizing behavior. Mapping the full scope of relational organizing predictors and prioritizing among them is a critical first step for rigorously evaluating what types of interventions are most useful for scaling up from individual to community-wide engagement in sustainable behaviors. Given this need, in this study, we employ an integrative approach to assess the relative influence of a wide range of social–psychological drivers of participation in relational organizing for sustainability. To achieve this, we set out to answer the following research questions:

RQ1: To what extent do different social–psychological beliefs related to relational organizing coalesce into unifying social–psychological constructs?

RQ2: What is the relative predictive power of these different social–psychological constructs on relational organizing behavior?

RQ3: Which discrete social–psychological beliefs emerge as the most powerful predictors of relational organizing behavior?

We also conducted an exploratory assessment of whether these effects differ by specific sub-populations in a preliminary analysis of possible heterogeneity in research questions two and three.

Methods

Participants

We focused our study on the United States, which has some of the highest per capita meat consumption globally (Ritchie et al. 2017) and where average meat consumption exceeds nutrition recommendations (Grotto and Zied 2010). We used a cross-sectional study design with convenience sampling, surveying two subsamples of Americans interested in animal welfare and environmental issues along with one more disinterested subsample. We deliberately targeted an engaged audience to ensure that we had an adequate number of respondents who had engaged with the sustainability issues under consideration previously, and so who might be more likely to be interested in relational organizing. In November–December 2021, we invited over 23,500 people to participate in an online survey, with two subsamples: (1) an engaged audience with demonstrated interest in animal welfare who had opted into emails from the nonprofit Mercy For Animals, (2) residents of Boulder, Colorado, including (a) an engaged audience with demonstrated interest in environmental issues, who had opted into newsletters and social media communication from environmental groups, and (b) a random sample of Boulder residents, to increase our sample size.

For subsample one, an online survey link was emailed to 19,321 individuals selected from the Mercy For Animals (MFA) email list. These individuals are those who have either opened an MFA email in the previous 12 months, clicked on a link in an MFA email in the previous 6 months, or donated $1000–5000 to MFA in the previous 7 months. Two reminder emails were sent to this sample. For subsample two, an online survey link was shared via newsletter and social media pages through the following groups: City of Boulder Open Space and Mountain Parks, City of Boulder Climate Initiatives, Boulder County, Eco-Cycle, and Growing Gardens. Reminders to participate were sent out at least once. The total number of people who received a link to the survey in one of these ways was impossible to calculate, but is likely to be several hundred based on a rough estimate of the total number newsletter and social media subscribers. Lastly, for subsample three, a postcard inviting participation in the survey was mailed to a random selection of 3500 Boulder, Colorado residents. The postcard included a shortened web link and a QR code connecting to the survey. Across all samples, initial invitations came from the partner organization (e.g., MFA, City of Boulder Climate Initiatives) with the research team listed as a collaborator.

Participation in all three subsamples was incentivized through $5 online gift cards for respondents who chose to claim it. Both based on survey response rates in our previous research and because the survey recruiting entity was familiar to the survey respondents, we anticipated a response rate of 5–15% depending on the subsample (Yan et al. 2018; Jones and Niemiec 2020; Champine et al. 2022). After removing responses that failed to meet data quality checks (see Appendix S2), 1784 people participated in our survey across all three subsamples, 1415 from MFA (response rate of 7.5%), 209 from Boulder online newsletters and social media, and 160 from Boulder via the mailer. The survey took a median of 17.6 min to complete. We excluded responses that were missing 30% or more of the predictor variables, i.e., completed less than 50% of the survey, leading to an adjusted total sample of 1529 (1166 MFA, 203 Boulder online, and 160 Boulder mailer). Compared to the US population as a whole, our participants were older, more likely to be women, white, and have a college degree, and had a similar median household income (Table S2). Roughly, 35% report eating more plant-based meals in the last 12 months compared to the previous year, and roughly 41% report eating less meat compared to the previous year (Table S4). Most respondents (87.3%) reported they had previously encouraged someone they know to eat a more plant-based diet (Table S3).

Measures

We measured 22 social–psychological beliefs related to relational organizing, which reflect the four a priori categories derived from the literature (Fig. 1; see Table S1 for measures and their derivation). Category 1 included attitudes about the personal behavior, personal behavior self-efficacy, and moral hypocrisy. Category 2 contained personal norms of relational organizing, moral exporting, respect for others’ autonomy, activist social identity, and vegan/vegetarian social identity. Category 3 comprised relational concern, anticipatory reactance, descriptive norms about the personal behavior (static and dynamic), descriptive norms about relational organizing (static and dynamic), injunctive norms of relational organizing (approval and disapproval), and perceived receptivity of others to relational organizing (dynamic). Category 4 included relational organizing self-efficacy as well as social, environmental, health, and animal welfare response efficacy related to relational organizing.

We measured two types of outcomes: self-reported behavioral intentions and proxies of actual behavior. Indicators of actual behavior were operationalized as two actions respondents could take at the end of the survey. First, early in the survey, respondents were asked to write a 1-sentence script of what they would say to encourage someone else to eat a more plant-based diet, which included what they would say, to whom, and through what communication channel (all open-ended responses). At the end of the survey, respondents could opt in to receive their script via email to remind them to contact the other person they chose. Second, respondents could opt in to write a separate public pledge of something they commit to do to encourage others to eat a more plant-based diet, which they were told would be shared anonymously with the wider animal welfare (MFA sample) or climate change (Boulder sample) volunteer community. We coded responses to the public pledge question and removed any answers that were about personal behavior (changing one’s own diet) rather than relational organizing. We measured behavior categorically as a binary variable (did the respondent do either post-survey action) because there was insufficient variation for a three-point scale (respondents did neither, did one, or did both).

We measured future behavioral intention via a 7-point Likert scale response to the question, “How likely or unlikely are you to encourage someone you know to eat a more plant-based diet in the next month?” As additional covariates, we included two measures of self-reported past behavior: frequency (“Approximately how frequently do you typically encourage others to eat a more plant-based diet?”) and proportion of social network (“Approximately how many people have you encouraged to eat a more plant-based diet?”). We defined a more plant-based diet as a diet lower in meat and other animal products and higher in fruits, vegetables, nuts, legumes, and other whole foods, including some consumption of processed plant-based products such as dairy-free milks and cheeses or vegan and vegetarian meats. We used this language to keep participants’ focus on small behavioral shifts and avoid triggering categorical thinking about diet-related social identities (Kurz et al. 2020; Sparkman et al. 2021b).

Analysis

To answer our first research question about the existence of latent social–psychological constructs, we ran Pearson’s correlations to determine the correlations between different predictor variables (Table 1), as well as a Kaiser–Meyer–Olkin (KMO) test to measure sampling adequacy. Since many items were at least moderately correlated with each other (≥ 0.30) and the overall KMO value was sufficiently high (≥ 0.60), we then ran exploratory factor analysis (EFA) on the predictor variables. To better visualize the relationship between these predictor variables, we also used forced distance plotting based on the distances from multidimensional scaling (MDS) using dissimilarity scores derived from the correlations between our independent variables (see methods in Jones et al. 2018). Specifically, we plotted a graphical lasso network, using zero-order MDS configuration to plot node position and running a graphical lasso to regularize the partial correlations, so that only the strongest partial correlations remained visible.

Table 1 Results of factor loadings from exploratory factor analysis (n = 1529)
Full size table

To answer our second research question about which overarching social–psychological constructs best predicted relational organizing behavior, we ran a series of ordinal logistic regressions to determine whether adding the different factors improves the model’s ability to predict actual relational organizing behavior above and beyond a baseline including demographics only. We used the ‘stats’ package in R for this analysis. We calculated Akaike information criterion (AIC) and conducted likelihood ratio tests using the ‘lmtest’ package in R to compare model fit. We repeated this process with OLS regressions to determine the effect of different factors on behavioral intention. To help visualize outcomes’ associations with independent measures, we created a forced distance plot based on the MDS from dissimilarity scores derived from the correlations between the independent variables and outcomes (Jones et al. 2018).

To answer our third research question about the effects of discrete social–psychological beliefs on behavior, we used penalized regression models using complete cases with elastic net-selected predictors from the full set of social–psychological predictor variables (Table S1). Elastic net is a form of regularization that penalizes a model for increasing complexity. It achieves this by setting penalty parameters known as alpha and lambda in order to reduce the absolute sum of coefficients while minimizing the sum of squared residuals. This is particularly useful for relatively small-n datasets with a large number of potential predictor variables (Yarkoni and Westfall 2017). We chose elastic net regression after initial correlation tests revealed that some predictor variables were highly correlated (|r| ≥ 0.70), since elastic net combines lasso and ridge regressions to better handle strong correlations (Zou and Hastie 2005). We used the ‘caret’ and ‘glmnet’ packages in R and tenfold cross-validation to calculate lambda and alpha values and select predictors. We then conducted binary logistic and OLS regressions on elastic net-selected predictors to obtain coefficients, standard errors, and p values, which are most commonly reported in behavioral science research.

As a sensitivity analysis for our elastic net regression, we used targeted maximum likelihood estimation (TMLE), estimated with flexible ensemble machine learning models, to rank-order variables based on the outcome probability ratio between high and low levels of predictors. This allowed us to estimate the relative probability of our primary behavioral outcome (actual post-survey behavior) between agree and disagree values for all Likert-type response variables and between categories of demographic variables, adjusting for all other predictors (demographics, all 22 social–psychological beliefs, and sample), regardless of whether they were retained in the elastic net. Targeted maximum likelihood estimation is a doubly robust method that allows for flexible estimation methods and theoretically better bias/variance tradeoffs than parametric methods (van der Laan and Rose 2011). We used SuperLearner ensemble machine learning for the estimation step with the ‘tlverse’ package in R, using cross-validation to weigh the individual algorithm predictions in the ensemble to maximize predictive accuracy. We included in the SuperLearner library simple means, generalized linear models, generalized additive models, LASSO regressions, and gradient boosted machines. As an additional sensitivity analysis, we re-ran the TMLE analysis twice, once controlling for demographics only and again controlling for all other predictors and other behavioral variables (behavioral intention and past self-reported frequency and amount of relational organizing). To explore potential differences between subsamples that were not captured by our survey questions, we created a binary dummy variable of subsample type (MFA vs. Boulder) and included this variable in our regression models and TMLE analyses. All analyses were conducted using R versions 4.0.4, and complete-case analysis was used to handle missing data.

Results

Underlying social–psychological constructs

Pearson’s correlation analyses revealed many large correlations of 0.5 or higher, and the KMO test result was high (0.9), and Bartlett’s test of sphericity returned a small p value, suggesting EFA was appropriate. Based on parallel analysis and Velicer’s minimum average partial test, we ran an EFA with five factors using oblique rotation (oblimin) to allow for correlations between the factors (see Table 1). We removed one cross-loaded item with a value of 0.32 or greater on two factors and three low-loading items with a value below 0.30 (Costello and Osborne 2019). This led to Factor 5 containing only two items, necessitating rerunning the EFA with four factors. The final EFA explained 44% of the variance in the data for the full sample.

In the final EFA (Table 1), Factor 1 captures confidence engaging in the personal behavior, combining self-efficacy in plant-based eating, vegan/vegetarian identity, attitudes about plant-based eating, and hypocrisy beliefs (negatively correlated). Factor 2 captures supportive social norms, including dynamic and static descriptive normative beliefs about the personal behavior and relational organizing, beliefs about others’ receptivity, and injunctive norms of approval. Factor 3 captures personal relational organizing aptitude through a combination of personal norms of relational organizing, moral exporting belief, activist identity, beliefs about others’ autonomy, and response efficacy beliefs about the impact of relational organizing on others’ engagement in the personal behavior, others’ health, the environment, and animal welfare. Factor 4 reflects concerns about others’ negative reactions to relational organizing, including anticipatory reactance, relational concern, and injunctive norms of disapproval. The MDS output largely aligns with these factors, with Factors 1 and 3 being most visibly clustered, and self-reported behavior and behavioral intention visually clustering within Factor 3 (Fig. 2).

Fig. 2
figure 2

MDS graphs visualizing EFA output a before and b after removal of cross-loaded or low-loaded items (white circles), and c with inclusion of self-reported past relational organizing behavior and relational organizing intention. See Table 1 for a key of which variable corresponds to which number

Full size image

Assessing these results against the first research question, our EFA suggested the existence of four social–psychological constructs that were somewhat different than the four we had a priori derived from previous research (Fig. 1). Supportive social norms (Factor 2) were separate from second-order beliefs about others’ disapproval (Factor 4). Vegetarian and vegan social identity loaded onto a different factor (Factor 1) than did activist social identity (Factor 3), and relational organizing response efficacy was most closely associated with respondents’ self-concept of themselves as relational organizers (Factor 3). Meanwhile relational organizing self-efficacy failed to load well onto any of the four factors and so was dropped in the EFA.

Relative predictive power of underlying social–psychological constructs

We used estimated factor scores, calculated using Bartlett scores produced by maximum likelihood estimates (DiStefano et al. 2009; McNeish and Wolf 2020), as predictors in a series of binary logistic regressions, which we summarize in Table 2. Each respondent’s factor score represents their placement on that factor. Individual models for each factor showed that when controlling for demographic variables and subsample, Factors 1 (confidence in engaging in plant-based eating), 2 (supportive social norms) and 3 (personal relational organizing aptitude) were statistically significant predictors of post-survey relational organizing (p < 0.01) and Factor 4 (concerns about others’ beliefs about relational organizing) was not. In a combined model containing all factors and demographics, Factors 1, 2 and 3 were significant.

Table 2 Results from a series of binomial regressions using factor scores to estimate effects of underlying social–psychological constructs on actual relational organizing behavior (n = 1529)
Full size table

A comparison between the models using AIC scores showed that the combined model had the lowest AIC score. Likelihood ratio tests revealed that the combined model better predicted relational organizing behavior than all individual factor models and a regression model with demographics only (p < 0.0001). However, a likelihood ratio test comparing the full model to an intermediate model with only Factors 1, 2 and 3 (plus demographics) found no significant difference (p = 0.5933) suggesting Factors 1–3 predicted relational organizing behavior (and behavioral intentions) just as well as the combined model with Factor 4. Factor 3, relational organizing aptitude, had the largest odds ratio of 0.61 in the intermediate model, an effect size 2.9–3.8 times greater than Factors 1 (confidence in plant-based eating) or 2 (supportive social norms).

Linear regressions of the secondary outcome found that in individual and combined models with all factors, all four factors significantly predicted relational organizing intention (see Table 3). Gender and age were consistently significant in these models, and political orientation and subsample were occasionally significant. Again Factor 3 had the largest coefficient of 0.68 in the combined model, compared to Factors 1 and 2 (0.26 each) and Factor 4 (− 0.18).

Table 3 Results from a series of linear regressions using factor scores to estimate effects of underlying social–psychological constructs on relational organizing behavioral intention (n = 1529)
Full size table

Most powerful discrete social–psychological belief predictors

We used three different methods to identify the most powerful discrete predictors, and report the results of each here and in Fig. 4. First, we ran a binomial regression to predict engagement in any post-survey relational organizing behavior, using variables selected through elastic net regression. Seventeen variables were retained in predicting post-survey relational organizing behavior (Table 4), of which two social–psychological beliefs were significant at the p = 0.01 level: personal norms of relational organizing (OR = 0.183, SE = 0.05, p < 0.000) and the belief that plant-based eating is enjoyable (OR = 0.082, SE = 0.039, p = 0.034). Second, we ran an OLS regression to predict our secondary outcome variable of relational organizing intention using the 34 variables retained by elastic net (Table 4). Personal norms of relational organizing (β = 0.175, SE = 0.038, p < 0.000) and the belief that plant-based eating is enjoyable (β = 0.092, SE = 0.027, p = 0.001) were still significant social–psychological predictors, as were nine other beliefs, as well as age and gender (Table 4).

Table 4 Comparison of elastic net logistic (1) and linear (2) regression models of the predictive power of discrete social–psychological beliefs on two outcomes: post-survey relational organizing and self-reported relational organizing intention (n = 1529, predictor variables selected with elastic net penalization using complete cases, n = 1055)
Full size table

Third, the TMLE analysis found that, controlling for all other measured beliefs, demographics, and subsample, several social–psychological beliefs significantly increased or decreased the probability of post-survey relational organizing behavior (Fig. 3). Variables increasing the probability were animal welfare response efficacy (RR = 2.25, 95% CI [1.87, 2.71]), plant-based eating self-efficacy (two measures; RR = 2.03, 95% CI [1.55, 2.65] and RR = 1.26, 95% CI [1.05, 1.51]), dynamic descriptive norm of receptivity (RR = 1.54, 95% CI [1.29, 1.82]), activist identity (two measures; RR = 1.49, 95% CI [1.15, 1.93] and RR = 1.26, 95% CI [1.05, 1.51]), health response efficacy (RR = 1.45, 95% CI [1.22, 1.73]), and dynamic descriptive norm of plant-based eating (RR = 1.24, 95% CI [1.01, 1.53]). Variables that decreased the probability of relational organizing behavior were: anticipatory reactance (RR = 0.57, 95% CI [0.48, 0.69] and RR = 0.69, 95% CI [0.53, 0.90]), static descriptive norm of plant-based eating (RR = 0.69, 95% CI [0.54, 0.88]), social response efficacy (RR = 0.83, 95% CI [0.73, 0.95]), and plant-based eating appetizing attitude (RR = 0.87, 95% CI [0.76, 0.98]).

Fig. 3
figure 3

Targeted maximum likelihood estimates of the relative probabilities of post-survey relational organizing action between high and low categories of variables, adjusting for other discrete social–psychological beliefs, demographics, and subsample variables (n = 1529)

Full size image

Regarding research question three, therefore, four discrete beliefs related to personal aptitude predicted both relational organizing action and intention: personal norms, activist identity, animal welfare response efficacy and social response efficacy (Fig. 4). Two beliefs related to confidence in the personal behavior were also significant: plant-based eating self-efficacy and the belief plant-based eating is enjoyable. One supportive norm (static descriptive norm of relational organizing) and one type of concern about others’ negative reactions (anticipatory reactance) were also significant. None of the coefficients was particularly large in the elastic net regressions (OR and β all below |0.2|; Table 4). However, a shift from negative to positive animal welfare response efficacy or plant-based eating self-efficacy more than doubled the chance of taking action (Fig. 3).

Fig. 4
figure 4

A summary of significant predictors of post-survey relational organizing and relational organizing intention. The thickness of arrows depicts the model coefficients for Factors 1–4. Discrete beliefs are color-coded to show if they predict post-survey relational organizing in either the elastic net regression or SuperLearner (black), relational organizing intention in elastic net regression (blue), or both behavior and intention (black lined in blue)

Full size image

Heterogeneity between subsamples

In the factor score regressions for both outcome variables, subsample was a significant predictor of behavior in the individual models for Factors 1, 2 and 4, but not for Factor 3 or in the combined or intermediate models. In other words, subsample was a significant predictor of behavior and intention while controlling for confidence about plant-based eating, belief about supportive social norms, or concern about others’ negative reactions to relational organizing, but inclusion of personal relational organizing aptitude removed this effect. Exploratory factor analysis with each subsample in isolation returned slightly different factor assemblages (Appendix S4), but in both, concern about others’ negative reactions was insignificant for predicting behavior and significant for predicting intention.

Analysis of discrete social–psychological predictors within each subsample showed personal norms of relational organizing predicted behavior and intention in both samples, and activist identity predicted behavior in both samples and intention in the MFA sample (Appendix S6). Relational organizing behavior was also predicted in the MFA sample by dynamic plant-based descriptive norms. Intention was also predicted in this sample by the belief that plant-based eating is enjoyable, static relational organizing descriptive norm, relational organizing self-efficacy, social response efficacy, gender, and age. Vegetarian/vegan identity also predicted intention in the Boulder sample. Subsample was not significant in the SuperLearner output.

Discussion

Relational organizing for sustainability has been studied under a variety of terms, from social environmentalism to peer influence, but to date, the literature has lacked a holistic investigation of what social–psychological beliefs drive people’s engagement in this kind of behavior. We examined a suite of 22 social–psychological constructs from the literature in the context of the shift to more plant-based food choices, one of the more meaningful climate actions individuals can take (Abrahamse 2020). If relational organizing from close others can scale up engagement in this type of climate action, it may be a useful additional tool for promoting climate solutions.

In our survey of animal welfare and environmentally minded US residents, we found first and foremost that personal aptitude in relational organizing significantly predicted engagement in and behavioral intention about relational organizing to encourage others to eat more sustainable diets. This included personal norms of relational organizing and alignment with an activist identity. Personal norms, through which individuals internalize a sense of obligation to take a particular action, have been consistently associated with environmental behavioral intention (Niemiec et al. 2020) and can mediate the impact of social norms on intention to eat less meat or reduce food waste (de Groot et al. 2021). Social identities like activist identity, through which individuals define themselves as part of a broader social community of activists, can be important mechanisms for environmental action (Schulte et al. 2020; Mackay et al. 2021), although vegetarian and vegan identity—which in this study predicted relational organizing intention but not action—can also be polarizing (MacInnis and Hodson 2021). More intervention-based research attempting to shift personal norms and social identities related to sustainable diets may help elucidate the plasticity and strength of these relationships.

Different forms of response efficacy were also significant and meaningful in our models. In the TMLE results, a shift from negative to positive animal welfare response efficacy was linked to respondents being more than twice as likely to take a relational organizing action at the end of the survey. The same change in health response efficacy—that is, the belief that successfully encouraging others to eat more plant-based meals would improve their health—also increased likelihood of acting. Concerns about animal welfare and health implications of a meat-based diet have increasingly been linked to more plant-based eating habits (Godfray et al. 2018; Mathur et al. 2021). Our findings extend that previous research to suggest that these beliefs can help drive outreach behaviors.

Interestingly, environmental response efficacy—a belief about the impact on climate change of effectively encouraging others to eat more plant-based meals—was not significant on its own, although it was part of personal aptitude. Climate response efficacy has been identified as a driver of climate action (Swim et al. 2019; Bradley et al. 2020), but it is unclear if increasing levels of climate response efficacy can drive greater climate engagement (Castiglione et al. 2022). Since climate change is a collective action problem, with a global emissions reduction threshold needing to be reached before the collective impact of individual contributions can be felt (Ostrom 2010), response efficacy may be less connected to action than in other contexts. In situations where individuals are being encouraged to take climate actions that have personal benefits, our findings suggest it may be more effective to highlight those benefits. For instance, animal welfare and health may be more motivating ‘hooks’ than climate change for encouraging people to talk to their friends and family about sustainable diets.

These results offer a variety of insights for climate action groups who wish to scale up community efforts. For instance, messaging to remind people of their own past behavior and evoking a moral framing has been found to activate personal norms (Schultz 1999). In the context of relational organizing about food, public-facing communication might remind people how frequently they talk about and eat food with close others, and highlight this as an ethical opportunity to encourage others to consider more sustainable eating habits. Such appeals might also include response efficacy language, to inform or remind people of the potential benefits of promoting this behavior to people they know (Niemiec et al. 2021a). Conversely, campaigns focusing on social identity might evoke group membership that is particularly salient to the target audience (for instance, affiliation with a particular organization or place-based community). Such outreach could also try to make relational organizing a more visible—and thus more socially accountable—behavior by forming relational organizing teams or having a relational organizing leaderboard within a volunteer group (Brick et al. 2017).

Our findings also reinforced that people may need a certain level of confidence doing the personal behavior before they talk to others about it. Respondents in our study with high plant-based eating self-efficacy were twice as likely to take relational organizing action after the survey compared to respondents with low plant-based eating self-efficacy. Further, the more strongly respondents believed plant-based eating is enjoyable, the more likely they were to do relational organizing. This suggests that interventions to increase people’s self-efficacy in taking personal climate actions or making the action more enjoyable (e.g., tastier plant-based meals) may have behavioral spillover into increased relational organizing. However, longitudinal and experimental research would be needed to confirm whether this occurs, and there may be interaction effects with other social–psychological beliefs, such as elements of personal relational organizing aptitude. Communication campaigns to encourage sustainability-related relational organizing might also target groups with high self-efficacy for the personal behavior. In our study context that might be people who already eat plant-based diets, but this could also include people who have recently participated in some form of efficacy-building, such as a gardening class or a home energy efficiency consultation.

Of the norms we studied, several emerged as important. These included the dynamic normative belief that others are becoming more receptive to learning about plant-based diets and a dynamic normative belief that others were trying more plant-based foods, which predicted post-survey relational organizing in the SuperLearner output. Dynamic norms have been identified elsewhere as useful messaging frames when an environmental behavior is increasingly common but still marginal (Sparkman et al. 2021b). Moments where public attention is captured by new innovations, such as Impossible Burgers and Beyond Meat in this context, may provide opportunities to capitalize on a growing dynamic norm (Sparkman et al. 2021a). Climate action groups can also run campaigns to increase the saliency of shifting dynamic norms, or to help people recognize that they might be underestimating their neighbors’ or their nation’s support for climate action (Sparkman et al. 2022).

The impact of concerns about others’ negative reactions was more mixed. This factor (Factor 4) was neither predictive of post-survey relational organizing nor were the injunctive norms significant in any of our analyses of discrete beliefs. However, anticipatory reactance did predict both relational organizing intention and action, and concerns about others’ beliefs about relational organizing predicted relational organizing intention. This could suggest that among those with high levels of concern about making others feel pressured or manipulated, this barrier may need to be alleviated before they are willing to discuss the shift to more sustainable diets. For instance, a sustainability nonprofit might run a workshop with volunteers to help them discuss climate-friendly behaviors in such a way that reduces reactance to their message. It could also be that people are worried about others’ negative reactions but that does not actually inhibit them from acting (a perceived but not actual barrier).

Overall, however, it is noteworthy is that in the regressions in which we tested the influence of factor scores on behavior, the models that combined three or more factors were as or more predictive of relational organizing behavior than any model with just one factor. From this, we might conclude that it is the combination of many different enabling social–psychological beliefs that is necessary to bridge the gap between motivation and relational organizing action on sustainability topics like plant-based eating. From a practical standpoint, this means that campaigns to promote sustainability-related relational organizing might need to emphasize multiple factors to motivate recipients to attempt to influence close others in their social network.

Limitations and future research

While we use a variety of predictive methods to make thorough use of the data available, these data are correlational in nature, with a great deal of covariance. Both features pose a clear challenge to causal inference. As such, these results are best understood as an effort to organize and synthesize the variety of social–psychological beliefs suggested by past research to predict relational organizing. These results indicate which beliefs are more or less promising to pursue in future experimental research on relational organizing.

We did run these analyses with two relatively distinct subsamples within the United States, and found that subsample was not an important predictor in the elastic net regression, SuperLearner ensemble, or combined factor models. Nonetheless, research with more diverse samples is needed to test for consistency. This would also help assess which drivers may matter for which population segments in the (likely) event that some are more influenced by certain beliefs than others. For instance, social norms may be more powerful in more collectivist cultures like Japan, while personal attitudes may be more compelling in more individualistic cultures like the US (Eom et al. 2016).

Importantly, we deliberately selected for a sample made up predominantly of people who had opted in to some form of environmental or animal welfare community. Our sample had a skewed distribution for some predictor variables, particularly in the MFA subsample for variables such as plant-based self-efficacy and vegetarian/vegan identity. Highly engaged volunteer communities may be particularly fruitful for relational organizing campaigns because they already care about the target issue. However, it may be that less engaged individuals who care less about the cause have different psychological drivers of relational organizing, which could be explored in future. Practitioner organizations may need to conduct some form of audience analysis prior to a campaign to ensure their messaging is tailored to audience beliefs.

Finally, we note that we were limited to measuring proxies of relational organizing behavior, rather than people’s actual communication with friends, family and close others. The absolute levels of behavior participants reported are subject to the precise presentation and wording. It is also possible our method may have led to higher levels of willingness reported, which may have added some noise to the process of determining which factors are the best predictors of relational organizing.

Conclusions

The results of this exploratory study suggest that harnessing social influence for sustainability requires a multi-pronged approach. Even among people who are already interested in a climate-related topic, as in our sample, people may need to believe that they have a personal responsibility to act before they will encourage others to try shifting to more sustainable habits. They may also benefit from feeling like their relational organizing efforts are aligned with a social group they are a part of, like an activism community, or aligned with actions their social networks are already taking, like learning about and trying plant-based foods. Further, they may need to believe that the target climate action is achievable and enjoyable, and that their own persuasive communication will have tangible benefits for goals that matter to them, like animal welfare or personal health. By continuing to explore the different and combined impacts of these kinds of beliefs on relational organizing behavior, researchers can identify paths forward for scaling up towards more widespread adoption of new climate actions.