Climate change as a social issue

Achieving global carbon emission reductions is a collective action problem, requiring changes to all levels of the social-ecological system. Actions such as recycling or switching off light bulbs are four to eight times less effective (respectively) than plant-based eating at lowering a person’s carbon emissions [1]. Indeed, behaviors that contribute to sustainable food systems, and in particular reduced consumption of animal-based products, are recognized as an integral part of climate action [2,3,1]. Therefore, there is a pressing need for research into how to scale up participation in healthy, accessible plant-based (i.e., vegan) and plant-forward (e.g., flexitarian, reducetarian, vegetarian, or other diets leaning heavily on plant-based options) diets [4,5].

There has been a recent effort to analyze and address climate action as a social process [6,7,4]. Scaling beyond the individual is necessary not only because approaches to climate change require collective action [5,4] but also because a focus on individual drivers has limited explanatory value in the psychological barriers theory of climate action or inaction [8]. Recent research has found that both action and inaction in the face of climate change are driven by social processes such as social norms (an individual’s perception of what is prototypical for the group) and collective efficacy beliefs (an individual’s perception of the group’s ability to meet its goals; [9,10,11,12,13]).

While people can be encouraged to engage in climate actions such as meat reduction through a variety of ways, such as active reflection and nudges [14], social norms are one area of particular salience [15]. Social norms reinforced by influential or close contacts may be particularly effective as encouraging climate action [16,15,17,18]. Meta-analyses and reviews highlight that pro-environmental norm perceptions generally predict pro-environmental behavior or intentions [19]. Nonetheless, interventions attempting to harness social influence via norms to reduce meat consumption may struggle to be effective for a variety of reasons, some of which may be due to poor intervention execution, and ultimately have a range of successes and failures [20]. The relationship between message and messenger is also complicated – even trusted messengers can be subject to a “backlash” [21].

As with social norms, efficacy beliefs have a substantial impact on climate action. Two main forms of efficacy beliefs [22] are self-efficacy (the perception that one is capable of taking action) and response efficacy (the notion that one’s actions will have a positive impact and/or reduce the threat). In the context of climate change, high self-efficacy leads people to take an approach (rather than an avoidance) orientation towards the topic, which encourages problem-solving [23]. Further, both self-efficacy and response efficacy are linked to willingness to have, and the frequency of, climate conversations [24,25]. Further, efficacy beliefs can function at the level of the collective. This is known as collective efficacy, and refers to the belief that the group one is part of is capable of achieving its goals, or that the group’s collective action makes a difference even if individual drivers do not. Perceived collective efficacy has been shown to increase intended participation in environmental actions [4] and support for pro-climate policies [26]. Since climate action is a social behavior, it is possible that a person’s social contacts might be effective conduits for involvement in pro-climate actions, and for the encouragement of plant-based food in particular [27].

Relational organizing

Relational organizing, also known as the block leader approach, is one form of social influence with a novel focus on increasing the motivation of potentially powerful messengers. It involves enlisting motivated volunteers to share information about and encourage others they know to engage in a desired behavior. Innovative experimentation shows it can be very effective because it harnesses the power of people’s social connections, which people are generally motivated to maintain, to help spread new information and create new social norms [28,29]. As such, relational organizing is a promising approach for achieving rapid, widespread community action for environmental conservation and social change [30].

Social relationships are important in food-related behavior because they can provide social influence, support and encouragement, shared resources and information, and a sense of community [31,32]. People often look to their peers and social networks for guidance on what to eat and how to live, and having supportive relationships can be helpful in maintaining a plant-based diet [33], though vegans may also be subject to stigma from friends and family [34]. Social relationships can also be a source of information and resources for individuals interested in following a plant-forward diet, and building relationships with others who eat in similar ways can help to reaffirm a shared purpose and feeling of community [35].

Critically, although past research on sustainable diets has identified many motivations for (and barriers to) individuals adopting plant-forward eating habits, there has only been limited investigation of relational organizing in this context [36]. This provides an opportunity for experimental social scientists to leverage this existing body of work to design and test new messaging frames that might motivate individuals to reach out to their social networks about this topic. While social influence has been documented to affect dietary behavior specifically, and climate action more broadly, little research has studied what factors prompt motivated individuals to deliberately encourage others to change (e.g., engage in relational organizing). However, recent studies have pointed to this as an important area of research. Divakaran and Nerbonne [37] identify relational organizing as essential to building a collective climate movement. Relational organizing can help correct pluralistic ignorance (the inaccurate notion that others don’t care about climate change), which can further increase willingness to engage in climate conversations [23] – although this relationship might not hold true for people with no social connections to those who hold the majority opinion [38]. Climate conversations based around evidence-based communication strategies, such as tailored communication, can transform dialogue into action [39,40]. In this way, relational climate conversations are essential for closing the climate concern-action gap [41].

Further, promoting plant-forward diets is a rich context that may be particularly illuminating for relational organizing: dietary choice is both deeply personal and a societal issue, one that intersects with our close interpersonal relationships on a daily basis as well as long term global causes like the mitigation of climate change. As such, this multifaceted issue is ripe for analysis as a social behavior [42,43,44]. Social norms play a significant role in food choices [45,46], particularly in the choice to consume or not consume animal products [31]. Thus, research in this domain can speak to a particularly wide variety of possible motivations and constraining factors for relational organizing.

Further, in particular, conversations with family and friends can influence views and behaviors related to healthy eating. Congruent with climate change research, which finds a dialogue-based approach (rather than didactic instruction) as a way to encourage climate engagement [47], a conversation orientation with family (as opposed to a conformity orientation, where health decisions are forced) is positively associated with pro-health orientations and behavior [48].

Socio-psychological drivers of dietary relational organizing

As in prior work on this topic [30,36], we define relational organizing as motivated individuals encouraging others they know to engage in a desired behavior (in this case, plant-based eating). Relational organizing could be an effective means of scaling up participation in plant-forward eating, which is urgently needed. Relational organizing around plant-based food can be influenced by a variety of social-psychological variables. One is a person’s perceived identity as an activist and/or as a vegetarian or vegan [49,27]. For those who identify in this way, relational organizing about plant-based food can be viewed as a form of maintaining and reinforcing self-congruence, or acting in congruence with one’s identity [49]. Another potential variable is anticipated psychological reactance or “backlash” to this conversation, as conversations about plant-based eating can trigger anger and perceived threats to one’s freedom, which can limit receptivity to the message [50]. Another relevant variable is one’s own attitudes towards plant-based food; if respondents have confidence in the behavior themselves, e.g., believe that a plant-based diet is enjoyable and affordable, they are more likely to tell others about it [36]. Finally, relational organizing can be impacted by personal norms [51,52] towards both plant-based food and the process of relational organizing. For example, respondents may feel that advocating for plant-based eating is hypocritical given their own behavior [53].

As with other behaviors related to health and dietary change, relational organizing around plant-based food is complex and likely requires multiple stages of change [54]. Given evidence that people at different stages of change often need different forms of support, we conceptualize the process of relational organizing on plant-based food as progressing along multiple stages of change: precontemplation, contemplation, preparation, action, and maintenance [55]. In precontemplation, there is no acknowledgment of a problem that requires behavioral change. In contemplation, a person becomes aware of a problem or situation that requires a behavior, but is not yet ready to perform the behavior. In preparation, the person gets ready to change. Action entails engaging in the behavior and maintenance is continuing the behavior. At each of these stages, the decisional balance of perceived benefits and detriments of the behavior gradually shifts in favor of the behavior [54]. In the realm of health behavior (which includes smoking cessation, plant-based eating, exercise, etc.), message-based interventions can help people progress through the stages of change. For example, gain-framed messages (that focus on benefits of the behavioral change) can move people from the contemplation to the preparation stage in smoking cessation [56].

In this study, we sought to test whether variations in message framing were more effective at different stages of change (precontemplation vs. contemplation) in the context of relational organizing. We drew on past research on relational organizing on plant-based food [36] to develop the factors (i.e., latent constructs comprising multiple variables) targeted by our message-based interventions (Table 1). These were:

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Factors containing significant socio-psychological constructs that predict engagement in dietary relational organizing.

https://doi.org/10.1371/journal.pclm.0000753.t001

1. Self-efficacy in engaging in plant-based eating and vegetarian/vegan identity
2. Supportive social norms for plant-based eating and relational organizing
3. Personal relational organizing aptitude, activist identity, and response efficacy
4. Concerns about others’ negative reactions to relational organizing, and
5. Relational organizing self-efficacy and perceived hypocrisy.

In this study we tested messages specifically tailored to address these five factors. We conducted an online experiment, hypothesizing that the following five factor conditions would have greater rates of relational organizing than the control condition:

1. Vegetarian/vegan eating and identity (H1)
2. Social norms (H2)
3. Activism (H3)
4. Anticipated reactance (H4)
5. Relational organizing beliefs (H5)

Ethics statement

All subjects provided informed written consent at the beginning of the survey. This study was approved by the Institutional Review Board of Oregon State University (IRB number 2022–1599).

Sample

We recruited participants through three online websites, Cloud Research, Amazon Mechanical Turk, and Prolific Academic, using a pre-screener to ensure participants lived in the USA, were at least 18 years old, and were either vegetarian, vegan or currently reducing their intake of animal products. Participants were recruited between 7 June and 3 November 2023. We purposely sampled an audience more engaged with this topic (of plant-based eating) than the population as a whole. This is because one’s self-efficacy in carrying out the desired behavior (confidence in one’s ability to eat plant-based) is an important precursor to relational organizing [36]. We aimed to recruit 500 participants per condition (3000 total), based on a pairwise power analysis calculated using GPower. This would give us 80% power to detect an 8% change (or d = 0.18 effect size), assuming 5% of participants are dropped from all rejection criteria. We considered 8% a reasonable target that could be powered by relatively small effect. Relational organizing is subject to numerous barriers (and can be a difficult outcome to alter experimentally.

We collected data from 4253 participants (2760 from CloudResearch, 1098 from MTurk, and 394 from Prolific; Table 2). However, our final sample size was 2902 after removing 646 participants for not having the correct diet, 349 for failing the attention check, and 299 for taking less than ½ the median time (12 minutes 14 seconds). We had a low exclusion rate in the full survey because we ran pre-screeners to identify this subpopulation and invited only those eligible to the full survey. Presumably, those who were then excluded either changed their diet in the interim or answered the pre-screener incorrectly. For these reasons, our results are applicable to motivated individuals (on the topic of reducing animal consumption) rather than the US population as a whole.

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Overview of demographic variables for participants in the study.

https://doi.org/10.1371/journal.pclm.0000753.t002

Demographic comparisons between the treatment and control groups can be found in Table A in S1 Text.

Materials and procedure

We presented participants with one of the six factor messages (five treatments and a control). These six factors had been identified previously via exploratory factor analysis and multidimensional scaling that tested the associations between 22 social psychological beliefs ([36], especially Table 1 and Fig 2). Constructs associated with each factor were measured with a series of Likert-scale survey items derived from previous literature ([36] supplemental materials).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Effect of factor messages on the binary outcome of whether participants wrote a short script for relational organizing, compared to the control condition.

Note. ^*p < 0.05, ^**p < 0.01, ^***p < 0.001. Effect size is the regression coefficient, on a binary outcome (this can be interpreted as percentage point difference between treatment and control). Vertical bars represent the 95% confidence intervals (CI), while the dotted line represents the control condition.

https://doi.org/10.1371/journal.pclm.0000753.g001

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Effect of factor messages on the general intention to engage in relational organizing, compared to the control condition.

Note. ^*p < 0.05, ^**p < 0.01, ^***p < 0.001. Effect size is the regression coefficient, on a 7-point Likert scale. Vertical bars represent the 95% confidence intervals (CI), while the dotted line represents the control condition.

https://doi.org/10.1371/journal.pclm.0000753.g002

The factor messages contained several paragraphs of text, each two-four sentences long, geared towards the specific psychological constructs (i.e., cognitive domain) targeted by that condition (Section B of S1 Text). For example, the relational organizing self-efficacy message read: “People often underestimate how persuasive our suggestions can be at guiding others. We regularly convince our friends to see the movie we want to see, or work colleagues to go in a different direction on a project. So too can we guide the dietary choices others make. A thoughtful request can often lead a person to change or at least try something new. You don’t have to be a charismatic salesperson to encourage someone to try a new plant-based meal. And after they do, there’s a good chance they’ll sometimes make similar choices in the future”.

After each paragraph, participants were asked to reflect and elaborate on that content (essentially asked how they may agree with it) by writing one to two sentences. Prior research has found that, when people are asked to reflect on a norm after its presentation, that can lead to more salient norm effects, perhaps because people are more persuaded by arguments that they generate themselves [57]. Further, reflection has been shown in experimental contexts to specifically increase support for pro-climate policies and actions [58]. In order to provide the control participants with an experience of similar length and completion time to that of the treatment, the control consisted of four similar tasks, on topics completely unrelated to plant-based diets (specifically, preferences of streaming services, social media platforms, colors, and seasons). The order of specific items within each factor message was randomized to avoid possible order effects. We refined these messages in an iterative manner by running 5 pilot studies, analyzing the qualitative responses, and rewriting specific messages, until over 70% of the participants agreed with all of the messages.

We had two outcomes, contemplation of (intended) relational organizing and preparation for relational organizing (both measured after treatment). Intentions are an important topic to analyze when looking at climate advocacy [59]. Understanding how different climate messages impact intentions to engage [60] can help support a person’s progression through the stages of change. We measured contemplation of relational organizing via a) a seven-point Likert index for general intention to encourage someone you know to eat a more plant-based diet in the next month, and b) a seven-point Likert scale for intentions to undertake six specific relational organizing behaviors in the next two weeks (specific intention). These were: 1) Share a plant-based recipe, 2) Share resources such as web links about the benefits of plant-based diets, 3) Suggest going to a vegetarian or vegan restaurant, 4) Initiate a conversation about the health benefits of eating a more plant-based diet, 5) Initiate a conversation about the environmental benefits of eating a more plant-based diet, and 6) Initiate a conversation about animal welfare benefits of eating a more plant-based diet.

To measure preparation for relational organizing, we asked participants if they were willing to “take the first steps in talking to people you know about adopting a more plant-based diet.” If they said yes, we asked them to write a short script for how they could approach a specific person they know, and classified this into a binary outcome (wrote something or skipped this optional section). For data privacy reasons (and since this was meant to measure contemplation rather than action), we didn’t ask respondents to share their script. The full survey can be found in Section C in S1 Text.

We included instrument manipulation checks for each of the specific beliefs in the factor messages (e.g., “I feel a moral obligation to encourage others to eat a more plant-based diet” on a seven-point Likert scale; see Section 4 of the survey in Section C in S1 Text). Other measured variables included current diet (on a spectrum from plant-forward diets to a fully plant-based or vegan diet), race, gender, age, education, and income. Although this was not a construct previously studied in relation to dietary relational organizing, we also measured the subjective norm of whether participants think it’s acceptable to influence others (generally or specifically what they eat). This was based on expert feedback when the pilot version of the experiment was presented at a conference. Finally, there was one attention check in the survey (asking what the majority of the survey questions were about; Section C in S1 Text). Participants who failed this were removed from the analysis, as were people who took less than ½ the median time.

Data analysis

We used a linear model to test the impact of message type on the binary outcome of relational organizing preparation: outcome ~ factor message. While using logistic regression was the norm for these kind of data in statistics in the past, more recent simulation studies have shown that linear models are as good and sometimes better than this approach [61]. Binary data in many real world settings tend to partially violate assumptions about homoskedasticity [62], but we also check if the results vary using robust SEs for binary outcomes. We made pairwise comparisons with each level of the factor message treatment to the control. This unadjusted regression was our primary analysis, and a regression with additional covariates was a secondary sensitivity analysis: outcome ~ factor message + current diet + income + gender + education. We repeated this procedure to analyze the Likert-scale intention outcomes for relational organizing contemplation, both the general intention and the scale for specific behaviors. For Likert scales that some could construe as ordinal, recent methodological work makes a compelling case that linear analyses are appropriate (e.g., [63]). However, we also ran an ordinal logistic regression for the general intentions outcome as a secondary sensitivity analysis. After observing heterogeneity in treatment effects by data source, we reran all regressions subset by data source (Fig A and Table E in S1 Text). All data analysis was conducted in R.

We performed instrument manipulation checks to see if each factor treatment moved the specific socio-psychological constructs they were intended to target. To do this we filtered the data just to participants exposed to a specific treatment and the control, then ran linear regressions on the relevant constructs. We also used linear regressions to check the reverse, namely whether factor treatments had a significant impact on constructs relating to other factors.

We also conducted mediation analyses to explore whether the social psychological constructs targeted by each treatment condition moved in the same direction as the outcome variable. For methodological reasons, we only performed mediation analyses when there was a significant main effect [64,65,66,67,68]. We report these results as correlations only, since our experimental design was not set up to fully test mediation [69]. To run these analyses we filtered the data to a binary treatment outcome for each factor and used the lavaan package in R with the relevant constructs [70].

Prior to gathering data, we preregistered our design (https://osf.io/r8ekp/). One deviation from our pre-registration was that we collected data from multiple sources (CloudResearch, MTurk, and Prolific Academic); this was due to the inability to draw an adequate sample size of participants that met our criteria (such as our dietary criteria) through Prolific Academic alone.

Manipulation checks

First, we checked whether each factor treatment moved the specific socio-psychological constructs they were intended to target (Fig F in S1 Text). The factor 1 treatment had a significant positive effect on each of its four component constructs (d = 0.23, p = 0.001; d = 0.31, p = 0.01; d = 0.33, p < 0.001; and d = 0.66, p < 0.001 respectively). The factor 2 treatment only had a significant positive effect on two of its component constructs - the plant-based dynamic norm and receptivity dynamic norm (d = 0.15, p < 0.001; and d = 0.16, p < 0.001 respectively). It had no effect on the plant-based static norm (d = -0.01, p = 0.94) or relational organizing dynamic norm (d = -0.01, p = 0.91). The factor 3 treatment had a significant positive effect on each of its five component constructs (d = 0.44, p < 0.001; d = 0.28, p = 0.01; d = 0.43, p < 0.001; d = 0.39, p < 0.001; and d = 0.69, p < 0.001 respectively). The factor 4 treatment also had a significant positive effect on its component construct (d = 0.19, p = 0.01). Finally, the factor 5 treatment also had a significant positive effect on each of its component constructs (d = 0.36, p < 0.001; and d = 0.5, p < 0.001 respectively).

Many constructs were affected by multiple factor treatments (Fig F and Tables I-M in S1 Text). For example, plant-based eating self-efficacy was significantly improved by exposure by not only both the factor 1 treatment (d = 0.29, p = 0.001), but also the factor 2 and factor 3 treatments (d = 0.15, p = 0.03; and d = 0.21, p = 0.002 respectively). However, in almost every case the biggest effect size came from the factor treatment that specifically targeted that construct. There two exceptions were anticipatory reactance and relational organizing self-efficacy. Anticipatory reactance was specifically targeted by the factor 4 treatment, but moved more in response to the factor 5 treatment which was focused on self-efficacy and hypocrisy (d = 0.19, p = 0.006; vs d = 0.22, p = 0.002 respectively). Relational organizing self-efficacy was specifically targeted by the factor 5 treatment, but moved more in response to the factor 3 treatment which was focused on personal norms, identity, and response efficacy (d = 0.36, p < 0.001; vs d = 0.45, p < 0.001 respectively).

Preparation for relational organizing

The two outcomes “willingness to take first steps” and “writing a short script” were highly correlated (Cronbach’s alpha = 0.99). Forty percent of people exposed to the control wrote a short script, while 34–45% of people in the treatment conditions went on to write a script. We found no significant effects on preparation for relational organizing from any of the treatments compared to the control (Fig 1; Tables B and C in S1 Text). Indeed, the factor 1 treatment (veg eating and identity) actually had a small but significant dampening effect on preparation for relational organizing (d = -0.06, 95% CI = [-0.12, -0.004], p = 0.04). The same results were found with robust SEs (Table D in S1 Text).

We also unexpectedly observed heterogeneity in the performance of different messages relating to the source of the data (CloudResearch, MTurk, or Prolific Academic; see Fig A, Table E, and Table F in S1 Text for full details). However, if we exclude Prolific Academic due to the small sample size from this source, we only see this difference between CloudResearch and MTurk for factors 1 and 2 on writing a relational organizing script; in other instances, these two samples overlap at least in the significance of the effect. However, the treatments did have an effect on an earlier stage of the behavior change process: contemplation.

Contemplation of relational organizing

We found that the correlation between intentions to engage in (contemplation of) relational organizing and writing a script was low (Cronbach’s alpha = 0.33). However, general and specific intentions were highly correlated (Cronbach’s alpha = 0.76), so here we focus on general intentions. All of the treatments except for factor 4 (anticipated reactance) significantly increased intentions to engage in relational organizing (Fig 2; Table G in S1 Text). Factors 3 and 5 performed the best (d = 0.62, 95% CI = [0.43, 0.82], p < 0.01 and d = 0.49, 95% CI = [0.3, 0.68], p = < 0.01 respectively) so we mainly focus on these in the results section, but full results for the behavioral outcome of relational organizing are in Tables B and C in S1 Text. Similar results were found via ordinal logistic regression (Table H in S1 Text).

Mediation analysis

Our manipulation checks, above, found that factors 1, 3, and 5 significantly affected all component constructs, and factor 2 significantly affected two of four component constructs. Mediation models showed that the impact of the treatment on the general intentions outcome is mainly correlated with two constructs for factor 1 (veg eating and identity): vegetarian/vegan identity (d = 0.18, 95% SE = 0.13, p < 0.001) and personal behavior is enjoyable (d = 0.39, 95% SE = 0.08, p < 0.01). For factor 2 (social norms), the effect of the treatment on the general intentions outcome was primarily correlated with personal behavior dynamic norm (d = 0.21, 95% SE = 0.09, p < 0.05), relational organizing dynamic norm (d = 0.35, 95% SE = 0.04, p < 0.001), and receptivity dynamic norm (d = 0.19, 95% SE = 0.04, p < 0.05). For factor 3, the outcome was mainly correlated with the constructs personal norm of relational organizing (d = 0.4, 95% SE = 0.03, p < 0.01), social response efficacy (d = 0.34, 95% SE = 0.03, p < 0.01), and health response efficacy (d = 0.16, 95% SE = 0.03, p < 0.01). Lastly, for factor 5 (relational organizing beliefs), the mediation showed that the impact of the treatment on the general intentions outcome was mainly associated with the construct of relational organizing self-efficacy (d = 0.55, 95% SE = 0.04, p < 0.01). For full path analyses, see Fig B-E in S1 Text.

Other drivers of relational organizing

The full regression with demographic variables shows that general intention is also predicted by current diet, gender, and whether participants think it’s acceptable to influence others (generally, or specifically what they eat; Table 4). Education, age, and length of diet are not predictors. The full regression for the behavioral outcome of preparation (i.e., writing a short script) is in Table C in S1 Text.

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Summary of the LM to model the effect of factor treatments and other variables on the general intention to engage in relational organizing (contemplation).

https://doi.org/10.1371/journal.pclm.0000753.t004

Limitations

It is difficult to study behaviors performed by only a small subsection of the general population. Estimates of meat-free eaters (vegetarians and vegans), range from 5 to 10% of the US population [86,87]. In our research, 36–40% of participants in the pre-screeners were eligible for the full survey, but with only a 62–70% response rate we needed to screen approximately three times the target sample size. For this reason, we had to recruit participants from multiple platforms in order to have enough participants who met the dietary criteria. Unfortunately, the terms of service of the different platforms prevented us from collecting personally identifiable information that would allow us to determine whether there was any duplication across the platforms. While this is a limitation, we also recognize that, given the large numbers of people on each platform at the time data was collected, duplicate sampling is unlikely.

In order to study a subset of the population, this study purposefully obtained a biased sample; one limitation of this sampling is that it may have reduced the variance more than necessary. Further, while it is reasonable to remove participants who fail simple attention checks (8% in our study) or who complete the survey so fast they likely did not read it (7% in our study) implementing these standards could have some drawbacks. For instance, removing these participants could mean the study results would not generalize a portion of the population who (like the 15% dropped in our sample) would not pay attention to the messages if they encountered them in real life.

In this study, concerns about experimental demand are partially mitigated because all conditions, including the control group, read instructions that the survey is about discussing eating less meat with others and were given information about plant-based diets prior to the outcome measures. However, the experimental content, which comes immediately after the outcome measures, may make the topic even more top-of-mind for treatment participants than for those in the control group. Future research could benefit from having a greater time delay between the independent and dependent variable to further mitigate concerns about experimenter demand effects.

We found differences in behavior between messaging campaigns that were not statistically significant because we weren’t powered to detect differences that small. However, a difference in adoption of 6% might be practically significant for real-world messaging campaigns where participation is already fairly low. For example, if a program had only a base rate of 6% participation, then this difference (the effect of relational organizing) would double participation. Future research is needed to confirm whether these differences can be detected in real-world contexts.

Our experiment had more effect on intentions than on actual relational organizing. Unfortunately, it has been well established in the literature that intentions do not perfectly predict actual behaviors [88]. Indeed, a medium-to-large-sized change in intentions will typically lead to only a small-to-medium-sized change in behavior (d+ = .36; [89]). The correlation between intended and actual relational organizing behavior in our study (Cronbach’s alpha = 0.33), was also lower than might be expected from a meta-analysis of behaviors generally (r+ = 0.53; [90]). This could be an artifact of the “practice” behavior we chose to measure, but in a pilot study we tested eight different possible proxies for relational organizing, including clicking on a link to write a social media post, sending an “intro to plant-based eating” guide to a friend, or writing a pledge to engage in relational organizing. These were chosen based on the literature and consultations with practitioners. The practice behavior had the best combination of correlation with general intent (0.35) and a large proportion of people actually performing the measured behavior (35%) out of all the eight behaviors we tested. It is possible that either the correlation between intended and actual relational organizing behavior is lower than general behaviors, or simply that it is difficult to measure relational organizing in an online survey. Future research might consider the use and measurement of in-person relational organizing opportunities, such as offering people the chance to share information about plant-based food at an environmental or animal welfare conference, to determine how the online format might influence or constrain the willingness to engage in relational organizing about plant-based food.

Future research could continue investigating potential barriers (either motivational or structural) to relational organizing. There are numerous contextual factors, such as a network of amenable close contacts (e.g., close contacts resistant to plant-based food because of perceived social stigma, [91]), opportunities for conversations (e.g., vegans not creating opportunities to discuss plant-based food in order to “fit in,” [83]), or access to plant-based food to share with others [92] that could potentially explain why not all intentions to share information on plant-based food translate into preparation or action. This will offer additional information on the transition from relational organizing intention to behavior, which in turn can help climate policymakers and advocates scale the consumption of plant-based food from the individual to the social level.