Volume 23, Issue 2 p. 393-417
Open Access

Gains and losses for humans and the environment: Effects of social identity and message prospect framing on pro-environmental behaviors

Riley Dedman

Riley Dedman

School of Health and Biomedical Sciences, RMIT University, Melbourne, Australia

Search for more papers by this author
Eunro Lee

Corresponding Author

Eunro Lee

School of Health and Biomedical Sciences, RMIT University, Melbourne, Australia


Eunro Lee, RMIT University, PO Box 71 Bundoora VIC 3083, Melbourne, Australia.

Email: [email protected]

Search for more papers by this author
First published: 08 June 2023


Addressing the global climate emergency and an urgent need for psychological research, the present study drew on two major psychological perspectives: social identity theory's notion of a socially constructed sense of self, and prospect theory's cognitive heuristics on the asymmetric effects of gain and loss framed messaging. A 2 (Human vs. Environmental Identity) × 2 (Gain vs. Loss framing) factorial experiment (N = 160) sought causal evidence for superordinate identities, gain versus loss framed messaging, and their interactions upon motivations for pro-environmental behavior. Results suggested interaction effects between social identity and message frames on activism, support for environmental policy, and enactment measures. Challenging prospect theory's original findings, gain framed messages were dominant in enhancing private sustainable behaviors, while loss frames were dominant in enhancing political behaviors. Whereas income was a significant demographic predictor, the overall social psychological findings inform campaign strategies for pro-environmental behavior.


In November 2019, over 11,000 scientists signed a declaration that unequivocally stated Earth was in the midst of a climate emergency (Amadeo 2018; Ripple et al., 2020). Since then, over 1700 local, state and federal governments across 31 countries have declared a state of emergency regarding the environment (Climate Emergency Declaration and Mobilisation In Action, 2020). Facing this significant global challenge, psychological theories and research are urgently called upon to understand the mechanisms involved in explaining and cultivating human pro-environmental behavior.

By demonstrating that environmental degradation is caused by collective behaviors rather than personal ones, and that any efforts to mitigate this degradation are dependent on collective action, Fritsche et al. (2018) proposed a social identity Model of Pro-Environmental Action (SIMPEA). Such a model is based on two conjunctive social-cognitive theories. Social identity theory illustrates an individual's self-concept as a function of their membership with salient social groups (Turner et al., 1979). While Self-categorization theory describes how group phenomena and interaction are explained through individual self-cognitions that class oneself with differing subordinate (family, organizational, local) and superordinate (national, human-race) social groups (Turner et al. 1994; Turner & Oakes, 1986). Products of social identification that have shown to influence pro-environmental behavior include increases in empathy and a sense of shared responsibility (Schultz, 2001), increased collective efficacy (Greenaway et al., 2015; Hornsey et al., 2015; Stollberg et al., 2015), and conforming to in-group norms that favor pro-environmental behavior (Rabinovich & Morton, 2012).

The process of social identification also gives rise to the Minimal Group Paradigm: in which simply being a member of an arbitrary experimental group appeared to be the minimum requirement for in-group favoritism and out-group discrimination (Tajfel et al., 1971). The salience of social identification that evokes these cognitive biases has indeed shown to mediate pro-environmental behavior across ostensibly created experimental groups (Masson et al., 2016), community collectives (Sanguinetti, 2014), university in-groups (Smith et al., 2012), political parties (Corner et al., 2012; McCright et al., 2016; Unsworth & Fielding, 2014; Ziegler, 2017), and groups identifying with their country (Jang, 2013; Rabinovich & Morton, 2012; Sheldon et al., 2011). A key strength of the social identity Approach however is its ability to supersede lower levels of social categorization to encourage pro-environmental behavior in large populations (Batalha & Reynolds, 2012).

Superordinate identities provide the ability to unite these lower level social collectives. Human Identity - where a person feels a connection to the entire human in-group - has shown consistent correlations with generalized measures of environmental sustainability (McFarland et al., 2019; Reysen & Katzarska-Miller, 2013) and intentions to engage in environmental activism (Renger & Reese, 2017). Running's (2013) study found that across 57 countries differing on their pro-environmental commitment, participants with a global identity were 1.26 times more likely to view global warming as a very serious problem compared to those without global identity, even when controlling for left-wing political ideology. Broader, Environmental Identity- where one's self-concept is informed by a connection to the whole natural world (Clayton, 2003)- has also been consistently associated with actions that positively serve the environment (Balundė et al., 2019; Tam, 2013). This identification with nature transcends all lower regional, political, national, and human levels of social categorization, where the in-group is composed of all living things. Meta-analysis of 37 independent samples testing the relationship between environmental identity/ connection-to-nature scales and measures of pro-environmental behavior found a consistent relationship across these studies (r = .42) (Whitburn et al., 2020).

Currently however the literature lacks evidence as to whether these superordinate level social identities can be experimentally primed to enhance pro-environmental behavior. Studies analyzing the link between Human and Environmental Identity and sustainable behaviors have been largely correlational (Mayer & Frantz, 2004; Nisbet et al., 2009; Renger & Reese, 2017; Reysen & Katzarska-Miller, 2013; Running, 2013; Whitburn et al., 2020). Further, the only attempts to prime Human and Environmental Identity have not found significant differences amongst groups on manipulation checks (Davis et al., 2009; Sheldon et al., 2011). The ability to prime these identities and compare the subsequent effects upon a range of pro-environmental behaviors is necessary to validate the causation of a social identity approach to environmental behavior (Fritsche et al., 2018).

Another important cognitive perspective that contributes to a behavioral-economic framework of environmental intentions is prospect theory. In developing this theory, Tversky and Kahneman (1981) demonstrated the asymmetrical nature of human-reasoning. Controlled experiments showed that when the same scenario was framed with an emphasis on either gains or losses, respondents were more likely to take risks to prevent against losses than to acquire gains (Tversky & Kahneman, 1981; Tversky & Kahneman 1992). Inconsistent findings, however, exist when this effect has been tested in an environmental context.

Experimental findings show that framing the losses involved with climate change have slightly more often increased intentions or enactment of environmental behaviors (Blose et al., 2015; Connor et al., 2016; Grazzini et al., 2018; Morton et al., 2011). However, other studies have found framing the gains associated with climate change mitigation to increase intentions toward recycling and energy consumption (Kim & Kim, 2014), as well as attitudes toward climate-change mitigation generally (Spence & Pidgeon, 2010). Additionally, Tversky and Kahneman (1981) originally developed prospect theory in the context of individual financial outcomes. However, all current investigations of environmental gain and loss framing conceive outcomes at an abstract environmental level, such as to the “health of the planet” (Kim & Kim, 2014). More recently, loss framing with more concrete message facilitated hotel guests’ more pro-environmental behaviors than loss framing and abstract messages (Grazzini et al., 2018). The significant mechanism was enhanced self-efficacy that can also be further extended to social identity that defines one's self at the group level. As the literature of environmental prospect framing effects is still in its infancy, there is a need for further investigation of whether personally framed gain or loss messages in social campaigns are superior in encouraging pro-environmental behaviors.

The lack of integration between these social-cognitive and individual heuristic processes may also limit the understanding of antecedents of human pro-environmental behavior. Studies have alluded to the effect that social identification and message frames have upon one-another (Bain et al., 2016; Blose et al., 2015). Specifically, they showed place attachment manipulated with location proximity for hotel guests, that is, mentioning the region name in the message, increased the loss aversion effect on their pro-environmental behavior. In contrast, gain framing was more effective than loss framing when the societal benefits were emphasized. Yet this interaction has never been formally experimentally or rigorously tested. A more recent study on pro-environmental hotel consumer behaviors has shown that normative gain framing was more effective than normative loss framing (Do et al., 2021). A novel insight from the findings was that anthropomorphism with using “Mr. Nature, happy earth face” further amplified the effect. Whereas self-efficacy was deemed as an important moderator, the anthropomorphism suggests human identity to be explored as a social group. Analysis of the comparative strength and mediating nature of the social identity and prospect theory frameworks is thus required to further understand the configuration of cognitive processes that influence behavior in an environmental context.

Another limitation pervasive across all environmental psychological literature is the superficial operationalization of pro-environmental behaviors. The majority of studies have only measured participant's intentions to engage in small domestic behaviors such as recycling and household energy consumption, yet these have long been considered to represent only a small portion of human impact on the environment (Stern, 2000). Multi-dimensional structures of environmentalism that include higher level social and political behaviors have been validated (Larson et al., 2015), yet are scarcely included in the literature (Restall & Conrad, 2015). Further, the gap between intention and behavior has been evidenced in environmental contexts (Carrington et al., 2010; Grimmer & Miles, 2017), yet studies persistently rely on outcome measures of environmental intentions and neglect the analysis of behavioral enactment.

Present study

Promoting pro-environmental behavior both at individual and socio-political levels presents urgent research agenda for policy making and social campaigns. Theoretically, we expect the integration of two major social psychological theories in a single study design will significantly advance our explanation of environmental behavior with examining individuals’ cognitive perspectives and social self as core factors of behavioral drive and change. Experimental methodology is critical in testing causal effects while we also attempted novel measurement of political voting intentions and enactment of actual behavior in addition to the creation of manipulation stimuli and the use of multiple measures existing in the field.

In light of the gaps aforementioned, the current study aims to experimentally compare the effects of priming superordinate Human and Environmental identities as well as Gain and Loss framing upon personal and socio-political environmental behaviors. Interaction effects between social identity and message framing will also be analyzed through a 2 (Human vs. Environmental Identity) × 2 (Gain vs. Loss framing) factorial design, controlling for political orientation and prior environmental values.

Based on correlations from human-nature connection and pro-environmental behavior (Balundė et al., 2019; Tam, 2013; Whitburn et al., 2020) and original findings from prospect theory experiments along with recent findings (Do et al., 2021; Grazzini et al., 2018; Tversky & Kahneman, 1981; Tversky & Kahneman 1992). Because of the strong loss aversive tendency and asymmetry leaning toward loss aversion over gain pursuit, loss framing is expected to enhance pro-environmental behavior. Furthermore, social identity leads to in-group favoritism and the categorization of the environment as belonging to one's superordinate social identity will foster more pro-environment behavior to pursue group goals than narrower human identity that excludes the environment from their significant group. Accordingly, it is predicted that:
  • Interaction effects will be significant with participants assigned to the Environmental Identity condition to show a larger effect of Loss framing on pro-environmental behavior compared to participants exposed to Human Identity priming.

Whether enhancing identification can transcend subgroups that may diverge on their stance towards environmental behaviors is important for cooperative efforts toward widespread shifts toward sustainability (Fielding and Hornsey, 2016; Reese, 2016). Social campaigning must understand whether people are more reactive to messages framing the gains or losses of environmental (in)action to effectively influence people towards pro-environmental behaviors (Nisbet, 2009). Further, the moderating effect that message framing and social identity have upon one another must be understood to clarify effective research and practical environmental movement strategies. It is also important that the validity of these approaches is tested upon both personal and socio-political behaviors that represent substantial environmental impact and are necessary for social and governmental reform (Millfont & Duckitt, 2010).


A pilot study was conducted to validate the 26 manipulation materials developed to prime social identity and gain/loss framing in the main study. Specifically pilot study participants rated the efficacy of priming images and text messages.

Participants and procedure

Participants N = 21 (a sample size based on in accordance with previous studies; Greaves et al., 2013, 65% 18–24 years old, 35% 25–34 years old; 55% female; 76% undergraduate degree, 19% postgraduate degree) were recruited through convenience sampling to validate the effectiveness of all social identity and prospect frame priming materials that were developed for this study. Participants were randomized into one of the four experimental conditions to validate either Human or Environmental Identity materials and Gain or Loss materials. After completing the priming tasks, participants rated each item's effectiveness in priming the relevant construct (1 “not appropriate” to 5 “very appropriate”). Options for written feedback were also included. An a priori minimum mean efficacy rating of ≥3 was set to validate each material. Materials that did not satisfy this requirement were substituted based on features from high-rating materials as well as written comments.


Priming materials are described in the later sections, and presented in Appendix A.


Results of the Pilot Analysis showed generally high ratings for intended stimuli for both Environmental and Human Identity, as well as Gain and Loss conditions. Only 3 out of 26 stimuli tested did not meet the a priori set minimum effectiveness rating.

The statement “I am proud of being a human being” received a Mean efficacy rating = 2.45. Despite being rated as “very effective” “extremely effective” by 36.4% of participants. This was replaced with an item relating to the theme of human helping (M = 4.45).

Participant comments regarding the Human Identity writing task included statements such as “…the statement referring to personality was most effective because I believe that human identity is defined by our ability to connect…” Accordingly, a statement regarding the lack of genetic trademarks (M = 2.8) was replaced by a statement referring to the intrinsic sociality of humans.

The Highest rated image in the Human Identity Condition (M = 3.91, Figure 1) was used as a reference to replace an image that contained the world map on someone's hands (M = 2.55, Figure A2) (Tables 1 and 2).

Details are in the caption following the image

Example images from human identity priming [Colour figure can be viewed at wileyonlinelibrary.com]

note. (a) “Dancing 7”, from OASIS, Kurdi et al. (2016), (https://www.dropbox.com/sh/4qaoqs77c9e5muh/AABozzDlISkxyBgy0byCwAdHa/images?dl=0&preview=Dancing±7.jpg&subfolder_nav_tracking = 1), CC BY-SA 4.0

(b) Untitled, Stock Image from Pixabay (n.d.) (https://www.pexels.com/photo/community-connection-cooperation-diversity-275698/), CC BY-SA 4.0.

TABLE 1. Examples of the textual components within social identity priming.
Task Example
Human I would like to help others, no matter what country they are from
Statement Ranking/Image-Matching Task I feel a sense of belonging to a human or
world community
Summary Writing Task Among humans, 99.9% of the bases in the entire genome are remarkably similar
Environmental I feel very connected to all living things and the earth
Statement Ranking/Image-Matching Task I think a lot about the suffering of animals and plants
Summary Writing Task Humans are still only one species among many, a species that shares the earth with countless animals and plants
TABLE 2. Examples of gain and loss framed messages in the personal writing task.
Domain Example
Gain Decreasing the impact of climate change will decrease the amount and severity of natural disasters
Natural Disasters
Job Security Efforts to stop climate change would create 24 million new jobs by 2030.
Loss The amount and severity of natural disasters will increase if global temperatures continue to rise
Natural Disasters
Job Security The World Employment and Social Outlook (2018) estimated that climate change threatens 1.2 billion jobs


Participants and procedure

International adult participants from five English speaking countries (Australia, Canada, New Zealand, UK, & USA) were recruited using the online Microworkers crowd-sourcing platform (Microworkers, n.d, RMIT Ethics approval, No. 22959). This platform has been widely used and validated in previous research (Buhrmester et al., 2011; Gardlo et al., 2012; Hirth et al., 2011). Participants were required to use a non-mobile device to complete the study. After providing online consent, participants completed baseline measures before being randomly allocated to one of four conditions: Human/Gain, Environmental/Gain, Human/Loss, Environmental/Loss. Post manipulation, participants completed outcome measures, manipulation checks and demographic questions in that order. An instantaneous reimbursement of $2USD credit (Liu & Sundar, 2018; Pittman & Sheehan, 2016) was provided for completing the 30-min experiment (Crone & Williams, 2017).

A priori power analysis was conducted using group-based means and standard deviations from studies comparing environmental behavior of low and high identity groups (Kashima et al., 2014) and perceived risks and benefits of physical activity based on message framing (Latimer et al., 2008). Separate sample size calculation from two one-way between-subject designs1 were conducted using G*Power 3 (Faul et al., 2007). Results indicated equal sized groups of n = 17 were required to achieve a power of .80 (alpha = .05). However, as there are no direct reference parameters due to the current study's novel design and the nature of the current two-way factorial design with planned interaction effect testing, a more conservative sampling plan will be employed. Thus, four groups of n = 40 aimed to ensure adequate statistical power. After 237 initial responses, participants with <99% completion (n = 47) and responses completed in <5 min (n = 21) or >80 min (n = 9) were excluded from analysis. The exclusion criteria were based on the inspection of the quality of the responses and subsequent judgment for genuine and credible responses. Furthermore, this measure intended to preserve the priming of state social identity, which may have been compromised if participants completed the study in a very short or long timeframe. The final sample2 (N = 160, Table 3) was male dominant (60.6%), with ages ranging from 18 – 68 years (M = 32.67, SD = 10). We have reported all measures, conditions, and data exclusions in the following.

TABLE 3. Participant descriptives.
Demographic N (%)
▒Environmental/Gain 33
▒Environmental/Loss 40
▒Human/Gain 43
▒Human/Loss 44
▒Age M = 32.7 (SD = 10)
▒Male 97 (60.6)
▒Female 58 (36.6)
▒Non-binary 0
▒Prefer not to say 5 (3.1)
▒Didn't finish high school 4 (2.5)
▒Finished high school 32 (20)
▒Vocational education 24 (15)
▒University degree 73 (45.6)
▒Postgraduate degree 27 (16.9)
▒United States 101 (63.1)
▒United Kingdom 28 (17.5)
▒Canada 15 (9.4)
▒Australia 11 (6.9)
▒New Zealand 3 (1.9)
▒Below average 53 (33.1)
▒Similar to average 74 (46.3)
▒Higher than average 18 (11.3)
▒Prefer not to say 15 (9.4)
  • Note. Uneven condition sizes are due to data cleaning after random allocation.


Social identity priming

Differing hierarchical superordinate social identities were primed by (1) completing either the Environmental Identity subscale (Olivos & Aragonés, 2011) of Clayton's (2003) Environmental Identity Scale or the Human Identity Scale (Albarello & Rubini, 2012), (2) ranking statements reflecting Environmental or Human Identity in order of personal relevance—ranking statements were all adapted from existing scales (Table A1 & A2 in Appendix A)—(3) subsequently matching these statements with images reflective of each social identity (Figure 1 & 2; refer to Figure A1 & Figure A2 for the full lists), and (4) a writing task of summarizing three pieces of information relative to each identity condition. All included components were designed as best as possible to prime only the relevant elements of each social identity construct, without inferring a need to conserve the natural environment (Table 1, Appendix A).

Details are in the caption following the image

Example images from environmental identity priming [Colour figure can be viewed at wileyonlinelibrary.com]

note. (a) “two koala bears”, by A. Pereira, (n.d) (https://unsplash.com/photos/8_ZwJSze_IM), CC BY 4.0 and “Mother and baby cassava farming”, by A. Spratt (n.d.), (https://unsplash.com/photos/yrzBgqapG1I), CC BY 4.0.

(b) Image of ‘Biodiversity’, from https://www.worldbank.org/en/topic/biodiversity, World Bank. CC BY-SA 4.

Gain-loss frame priming

Participants completed a writing task to illustrate their predicted personal gains or losses relating to five separate outcomes of environmental (in)action (Table 2, Appendix B). The five statements covered domains of mental health, physical health, employment, financial impact, and natural disasters. As prospect theory's predictions are related to personal framing effects (Tversky & Kahneman, 1992), the study required participants to conceptualize these outcomes at a personal level.

Baseline measures

Political Ideology (Authoritarianism) was assessed using the six-item Very Short Authoritarianism Scale (VSA; Bizumic & Duckitt, 2018, 𝛼 = .73) on a 9-point agreement scale. Example item was “What our country needs most is discipline, with everyone following our leaders in unity (Conservatism or Authoritarian Submission).”

Pro-Environmental Values were assessed using six items from the "Balance of Nature" and "Anti-Anthropocentrism" subscales of The New Ecological Paradigm (NEP; Dunlap et al., 2000). Example item was “When humans interfere with nature it often produces disastrous consequences” measured on a five-point agreement scale (𝛼 = .69).

Outcome measures

For all outcome measures with subfactors, the subscales were analyzed in the main models. Upholding the unidimensionality assumption (Gerbing & Anderson, 1988; Skogen et al., 2019) in the measurement of latent construct, each subscale was used as the analysis variable to warrant psychometric rigor.

Pro-Environmental Behavioral Intentions were measured using a green behavior frequency scale (Schultz & Zelezny, 1998). The 12-item scale (𝛼 = .88) measures behaviors covering recycling, green consumerism, waste avoidance, energy use, and socio-political engagement (e.g., Looking for ways to reuse things). A three-item Transportation subscale from the Pro-Environmental Behavior Scale (PEBS; Markle, 2013) was also used (e.g., In the following month, how likely are you to use public transport?). Both scales asked participants to indicate the frequency in which they intended to perform behaviors in the following month (1 = “Never”; 5 = “Very Often”).

Socio-Political Pro-Environmental Behavior was measured using the Environmental Movement Activism (e.g., I would like to join and actively participate in an environmentalist group. 𝛼 = .92) and Support for Interventionist Conservation Policies (𝛼 = .87) subscales the Environmental Attitudes Inventory (EAI; Milfont & Duckitt, 2010, e.g., Controls should be placed on industry to protect the environment from pollution, even if it means things will cost more). Both contained 10-items using a seven-point agreement response scale.

An Environmental Voting Intentions Scale was developed to measure the likelihood participants would vote for political candidates based on slogans that represented differing views of environmental policy. Participants were asked to indicate on a five-point scale (1 = “Very Unlikely to Vote For”; 5 = “Very Likely to Vote for”) their attitude toward eight political slogans (Table 4). Slogans represented stances ranging from outright commitment toward environmental policy to prioritizing economic stability over environmental protection. Example slogans included “We are at risk of climate catastrophe if we do not bring the world's biggest polluters to account” versus “We believe any environmental policy must ensure that our economy and our power prices are not jeopardised.”

TABLE 4. Factor analysis results: Environmental voting intentions.
Items 1a 2b
5. “We are at risk of climate catastrophe if we do not bring the world's biggest polluters to account.” .75 .04
7. “We believe the science that climate change is one of the greatest threats to our way of life. And we believe it is everyone's responsibility to help protect the environment.” .75 -.03
8. “Only an economy based around the environment, instead of profit – will save our future. Through our policies we will protect the natural environment, it's species and its beauty by moving away from fossil fuels and investing in renewable energy systems.” .69 -.11
1. “Regulations mean fossil fuels can be extracted with less damage to the environment.” .23 .60
2. “We believe any environmental policy must ensure that our economy and our power prices are not jeopardised.” -.22 .54
6. “Unchecked immigration will cause catastrophic overpopulation in our country – leading to increased waste, emissions and damage to our ecosystems.” -.38 .52
3. “A smooth transition for all- we are protecting our environment and our strong economy by ensuring working families aren't taking pay cuts for the environment.” -.02 .35
4. “Regulation not revolution – we must ensure fossil fuel companies pay their fair share of tax, clean up their mess, and invest in local communities.”c .44 .47
  • Note. Unweighted Least Squares extraction with Promax rotation and Kaiser normalization. The larger loading of each item in the two factor structure are presented in bold. The factor with larger bold values was estimated as the underlying construct of the item.
  • a “Environmental Voting Support.”
  • b “Resistance to Environmental Policy.”
  • c Not included in either factor due to the cross-loading.

Pro-Environmental Behavioral Enactment was measured by asking participants to sign up to (“Yes” or “No”) a pair of fictitious activities (Appendix C). The first activity asked participants whether they would like to RSVP to a tree-planting event “in their area.” The second involved signing up to receive information regarding sustainable practices, environmental organizations, and volunteering opportunities. Post-study debriefing explained the real purpose of these questions.

Manipulation check measures

Environmental Identity was measured using the 11 item Commitment to Nature Scale (COM; Davis et al., 2009). The COM scale (e.g., I feel strongly linked to the environment. 𝛼 = .92) has shown strong positive correlations (r = .85, p < .001) with Clayton's (2003) Environmental Identity Scale (Tam, 2013). Example item was “I think of myself as a part of nature, not separate from it” on an 8-point agreement scale.

Human Identity was measured using the nine-item Identification With All Humans Scale (IWAH; McFarland et al., 2012). Example item was “How close do you feel with people all over the world” measured on varying five-point scales (𝛼 = .92).

Gain/Loss Framing was measured using opposingly framed statements from Connor et al. (2016). Participants indicated their agreement (1 = “Strongly Disagree”; 7 = “Strongly Agree”, 𝛼 = .88) with one gain and one loss framed message from each domain (5 gain items, 5 loss items) Following a leading “If we don't act on climate change now, in the future we will…” either of the gain (e.g., “have learned”) or loss (e.g., “missed out on learning” statement component was provided to each experimental group followed by the next segments (e.g., “new skills that could help us solve other societal problems”). See Table C1for the whole list.

Statistical analysis

Two-way Analyses of Covariance (ANCOVA) were used to test any baseline differences between groups and main and interaction effects between groups upon outcome measures. For the binary enactment measure with “Yes” or “No” responses, Chi-Square tests of contingencies identified any differences in response patterns between experimental groups. Factor analysis tested the underlying dimensions of the voting intentions scale. Follow-up Multivariate Analysis of Covariance (MANCOVA) was used to identify profiles of the outcome measures by experimental condition.


Table 3 displays group sizes and demographics after data cleaning and random allocation. Kolmogorov–Smirnov and Shapiro–Wilk normality tests indicated that all variables other than green behavior frequency had distributions deviating from normal. To deal with this non-normality, all statistical analyses used Bootstrapping of 1000 samples. Where applicable, Levene's test of homogeneity of variances was conducted and no assumption violation occurred within following analyses.

Demographic variables, authoritarianism, and environmental values

There were no significant differences between groups on all demographic variables other than Income with a medium effect size (φ = .32), χ2 (5, N = 145) = 14.59, p = .024 while “Prefer not to say” income responses (n = 15, 9.4%) were excluded due to the ambiguity. Due to unbalanced proportions of income earners across conditions, income was included as a control variable in all analyses. Factorial between-groups ANOVAs assessed baseline variables amongst experimental conditions and income groups. No main effects of the Identity, Frame, or Income were significant for Authoritarianism. For Environmental values, ANOVA revealed no significant main effects for Identity and Frame but did reveal a significant main effect of Income, F(1, 139) = 4.580, p = .012, ηp = .081. Income was thus continuously included as a control variable factor in all main analyses.3

Manipulation checks


Despite trends of higher comparative endorsement of each identity condition with scales measuring coinciding levels of social identity, ANOVA revealed no significant main effects of Identity on both the IWAH and COM scales (p = .360).


ANOVA revealed a significant effect of Frame upon mean frame agreement difference scores (gain agreement subtracted from loss agreement), F(1, 141) = 6.44, p = .012, ηp = .121. The sample as a whole had higher mean agreement with gain framed messages than loss framed messages, however this difference was significantly greater (Cohen's d = .46) in the Gain Frame (MD = -.46, SD = .67) compared to the loss frame (MD = -.14, SD = .73).

Candidacy outcome variable: Factor analysis

The underlying structure of the 8-item environmental voting commitment measure was analyzed with exploratory factor analysis (EFA). Extraction method was unweighted least squares that warrants the most rigorous factor solution for variables on Likert scales.

Promax with Kaiser normalization was used as the rotation method because the factors were hypothesized as correlated (Allen et al., 2018). Two factors with eigenvalues larger than one as confirmed with parallel analysis (O'connor, 2000) was extracted and labelled as Environmental Voting Support and Resistance to Environmental Policy explaining 55% of the variance in the measure. Factor loadings ranged from .35 to .75, while item 4 was excluded due to cross-loading (Table 4).

Environmental behaviors: ANCOVAs of outcome measures

Two-way Factorial between-groups ANOVAs were run to test effects of priming upon pro-environmental behavioral measures when income was controlled.


No significant main effect of Identity was found upon all outcome variables.


A marginal main effect (at 𝛼 = .10) for Frame upon Transport intentions was approaching significance F(1, 139) = 2.733, p = .101, ω2 = .012, with those in the Gain frame (M = 2.64, SD = 1.01) showing higher sustainable transportation intentions (d = .28) than those in the Loss Frame (M = 2.36, SD = 1.0)

Interactions of identity and frame

While interactions were not significant for most outcomes, Support for Movement Activism did contain a significant interaction F(1, 139) = 4.783, p = .03, ω2 = .031. Follow up simple main effects revealed the effect of Identity was only significant when the Loss Frame was primed (at Bonferroni adjusted 𝛼 = .25) F(1, 140) = 7.426, p = .007, d = 3.25 (Figure 3), with Environmental Identity priming resulting in significantly greater Support for Movement Activism than priming Human Identity t(140) = 2.80, p = .006, d = 1.30. Further, an interaction upon Resistance to Environmental Policy was marginally significant, F(1,139) = 3.075, p = .082, ηp = .023. Follow up simple main effects revealed the effect of frame was only significant when Environmental Identity was primed (at Bonferroni adjusted 𝛼 = .25), F(1,139) = 6.222, p = .14, ηp = .023, with loss frame participants significantly less likely to vote for candidates resistant to environmental policy, t(1,139) = −2.48, p = .021, d = 2.62.

Details are in the caption following the image
Interaction effect of identity and frame upon support for movement activism. [Colour figure can be viewed at wileyonlinelibrary.com]

Income: A control variable

Multiple main effects of income are presented in Table 5.

TABLE 5. Significant effects of income upon outcome variables.
Variable Test statistic Group differences Difference statistic
Green behavior frequency F(2, 139) = 4.267, p = .016, ω2 = .046 Average income (M = 3.16, SD = .77) t(139) = 2.62, p = 018,
Low income (M = 2.77, SD = .73) d = .52
Environmental voting support F(2, 139) = 3.204, p = .044, ω2 = .030 Average income (M = 4.09, SD = .74) t(139) = 2.21, p = .025,
High income (M = 3.54, SD = 1.05) d = .61
Resistance to environmental policy F(2,139) = 4.341, p = .015 ω2 = .044 Low income (M = 3.13, SD = .69) t(139) = -2.81, p = .009,
Average income (M = 3.44, SD = .76) d = .43

Behavioral enactment

Chi-Square Test of Contingency analyses were conducted to test whether experimental conditions were related to the binary response patterns upon Environmental Enactment measures.

For the tree-planting enactment measure, Framing effect was statistically significant, χ2(1, N = 160) = 5.335, p = .021 (ϕ = .183), with a higher proportion of participants in the gain frame signing up to the enactment measure. An interaction with Identity was marginally significant, where larger proportions gain frame enactment were dependant on Human Identity priming, χ2(1, n = 87) = 2.981, p = .084, ϕ = .183. Further, adding income to the analysis showed a statistically significant interaction χ2(1, n = 53) = 7.526, p = .006, ϕ = .38, with the larger proportions of enactment in the gain frame only significant within the low-income group (Cohen's w = .38).

For the activism information enactment measure, no significant effect of Identity or Framing was found; however, tests revealed a marginally significant interaction effect χ2(1, n = 87) = 2.981, p = .084, ϕ = .19. Again, the larger proportion of gain frame participants signing up to receive environmental information was only found in the Human Identity condition.

Behavior profiles

MANCOVA was used to test any effects of Identity or Frame upon the multiple pro-environmental outcome measures, while controlling for income (Figure 4 & 5). Assumptions of no multicollinearity, no multivariate outliers, and homogeneity of variance-covariance matrices were all satisfied. MANCOVA showed that both the effect of Identity F(4, 137) = 2.107, p = .083, 𝜂2 = .058 and the interaction effect of Identity and Frame F(4, 137) = 2.078, p = .087 𝜂2 = .067, were marginally significant at 𝛼 = .10. Individual analysis of dependent variables showed no statistically significant effects at a Bonferroni adjusted 𝛼 = .0125. However, the interaction effect of Identity and Frame upon Support for Movement Activism was approaching significance at this level F(1, 140) = 6.233, p = .014, ω2 = .036.

Details are in the caption following the image
Profile of environmental outcome measures of frame conditions: MANCOVA results. [Colour figure can be viewed at wileyonlinelibrary.com]
Details are in the caption following the image
Profile of environmental outcome measures of identity conditions: MANCOVA results. [Colour figure can be viewed at wileyonlinelibrary.com]


Literature has demonstrated the ability for both social identity and contrasting message frames to influence the extent to which people engage in pro-environmental behaviors (Jang, 2013; Masson et al., 2016; Morton et al., 2011; Unsworth & Fielding, 2014). However, causal evidence toward these behaviors had been lacking at superordinate Human and Environmental Identity levels of social categorization. Further, the effects that messages highlighting either the gains or losses associated with climate (in)action have upon pro-environmental behavior have shown mixed results in their initial application (Morton et al., 2011). Importantly, measurement of the effects and interactions between these cognitive processes upon varying personal, social, and political environmental actions had yet to be conducted (Larson et al., 2015; Restall & Conrad, 2015).

The current study therefore aimed to investigate both the main and interaction effects of priming superordinate social identities and message frames upon a spate of pro-environmental behaviors. Using a 2 (Human and Environmental Identity) ×2 (Gain and Loss Frame) factorial design, the comparative effect of these constructs was tested upon private-sphere environmental behaviors, social behaviors such as activism, as well as support for governmental conservation efforts and environmental policy. Of the series of outcome variables and respective hypotheses on the interaction effects between framing and social identity, supported were two constructs, that is, environmental activism as a sub-scale of socio-political pro-environmental behavior and likelihood of voting for politicians who were reluctant to environmental policy as a subscale of environmental voting intentions. Despite the limited support, the exploratory nature of the current pioneer research suggests potential extensions of the implications with further elaborated future research.

Analysis revealed no main effects of social identity upon any of the outcome measures. This may portray the balanced effect of both Human and Environmental Identity in encouraging pro-environmental behaviors, however non-significant manipulation checks confound such an interpretation. Contradicting Tversky and Kahneman's (1981, 1992) original proposition that people are more likely to respond to loss framed messages, some main effects of frame showed messages highlighting gains were superior in cultivating environmental behaviors. Overall however, the main hypothesis of the study was supported by the significant interaction effects between social identity and message frame on some of the outcome variables, suggesting these cognitive influences on behavior do moderate each other's effects and thus should not be tested in isolation of one-another.

With the novel emphasis on political environmental behavior, the study developed an 8-item measure on environmental voting intentions. Factor analysis supported two underlying distinctive dimensions. As these factors are referred to throughout the discussion of main and interaction effects, they are first discussed here. Items displaying a concentrated commitment to climate change mitigation were distinguished from environmental policy as conditional upon factors such as economic stability. This study was by no means a scale development study, thus more rigorous development of candidacy measures is required to better understand the antecedents of demand for sustainable policy.

Interactions of social identity and message framing

Providing support for the main hypothesis, interactions showed that Environmental Identity produced larger support for environmental activism than Human Identity, only in the loss framed condition. Similarly, a marginally significant interaction demonstrated that when primed with an Environmental Identity, those in the loss frame were less likely to vote for people who were noncommittal to environmental policy. However, against the prediction, enactment measures contained marginally significant interactions where greater enactment within the gain frame condition was only seen when Human Identity was primed. Despite their varied directions, these interactions point to the constellation of cognitive biases that precede decisions regarding sustainable actions. This study was the first to formally test interactions of social identity and prospect framing, and such results suggest this combined investigation is warranted. Evidently, analyzing the effect of message frames or social identities upon pro-environmental behavior should not be done exclusively.

These results provide the first formal test of gain and loss frame interactions with social identity, and indeed the interdependence of these influences resonates with the literature. Studies testing varying domain specific gain framed messages have found that highlighting the ability for climate change mitigation to create a more caring society has often increased pro-environmental behavior (Bain et al., 2012; Bain et al., 2016; Scannell & Gifford, 2013). Place attachment—which contains a social element of attachment alongside psychological connection to place (Lewicka, 2010, 2011; Scannell & Gifford, 2010)—has also shown to predict the power of message framing upon environmental actions (Scannell & Gifford, 2013). Additionally, presenting messages as appealing to ingroup values (Hurst & Stern, 2020) or coming from an ingroup source (Wolsko et al., 2016), has also been causal in enhancing the efficacy of climate change messaging. Many of these studies have however focused on moral based frames, overlooking the key prospect consideration that arises from cognitive biases related to gains and losses. The current findings indicate such fundamental human biases must be considered alongside the socially constructed identity of individuals when designing messages for pro-environmental behaviors.

Hierarchy and manipulability of superordinate social identity

In line with the significant interaction effects between social identity and prospect framing, no main effects of Identity were seen across all measures of pro-environmental behavior. This may be due to the interdependence of social identity and message framing; however, such an interpretation is blurred by non-significant manipulation checks. The fact that Human and Environmental Identity levels were not significantly different across groups may arise from the difficulties in priming distinguishable identities that are hierarchical. These senses of connection are also relatively void of clear descriptive and injunctive norms, which are essential to social categorization (Onorato & Turner, 2004). The challenge of priming these separate identities parallels non-significant manipulation checks in other studies priming Environmental and Human Identity (Davis et al., 2009; Sheldon et al., 2011). While the current study used a multi-modal priming approach, the difficulties experienced in priming these groups reflects a need to further advance priming methodologies. Future studies should focus on specific elements of Human Identity, such as sense of global community, intergroup helping, and empathy that are theorized to lead to pro-environmental behavior (Renger & Reese, 2017; Reysen & Katzarska-Miller, 2013). Refining the elements of Environmental Identity that enhance pro-environmental behavior may also increase manipulability, as studies have shown different effects on behavior across ecocentric, anthropocentric, and biospheric connections to nature (Amerigo et al., 2007; Gagnon Thompson & Barton, 1994; Schultz, 2001).

Effects of gain and loss framing

Regarding prospect theory, differences in the relative efficacy of message frames do provide for critical interpretation. Estimated marginal means (Figure 4) indicate the gain frame showed higher pro-environmental behavioral intentions across private-sphere measures. This finding was significant in one of the two enactment measures, while marginally significant upon transport intentions. Alongside these findings however was the observation that loss framed messaging decreased support for policy that was resistant to environmental protection, and showed trends toward increased support for conservationist policies. Despite these effects being qualified by marginally significant interaction effects, such patterns seem to point toward the likelihood that frame effects are dependent on the behavioral context to which they are attached.

Results showing gain framed messages were dominant in enhancing private-sphere behaviors, yet loss framed messages were dominant in enhancing political behaviors are reflective of context dependent findings in the literature. Findings demonstrate that while presenting the potential personal benefits of environmentalism may motivate personal behaviors that may conjure these benefits (Bruderer Enzler, 2015; Kim & Kim, 2014; Spence et al., 2010), highlighting the varying risks associated with climate inaction can lead people to have stronger political volition towards environmental protection (Bilandzic et al., 2017). Other studies have however found that framing the ability for environmental policy to achieve positive, growth related results increases the likelihood participants will vote for candidates who promote these policies (Bertolotti & Catellani, 2014).

Despite equivocal consistency amongst findings, the fact framing effects were seen to be dependent upon specific environmental behaviors challenges Tversky & Kahneman's (1981) original finding of ubiquitous loss aversion. The current study also presented gain and loss frames upon a range of outcome domains. Presenting frames multi-dimensionally may help to reconcile the inconsistencies between environmental behavioral effects of financial based messaging (Wolske et al., 2018), geographical framing (Spence & Pidgeon, 2010), and framing presenting the impacts upon nature (Kim & Kim, 2014). The behavior dependent effect of message framing also speaks to the importance of measuring environmental behavior multi-dimensionally, as no effect of frame, identity or interaction was consistent across all behaviors.

Future studies may extend the current findings and insights with more intrapersonal variables such as self-efficacy and social norms (Do et al., 2021; Grazzini et al., 2018). While numerous psychological theories and factors would explain how individuals behave in various settings such as consumption, politics, and economic act, the interaction between perspectives and cognitive sense of self would be necessary factors in better understanding pro-environmental behavior as the current findings highlight. Meanwhile more social level framing may also work differently because the social nature of human identity and environmental identity would in line with the framing level. The gain and loss framing used in the present study remained at personal level whereas the manipulated identities scope societal and global dimensions. Therefore, future studies may cross validate the current findings with social level framing of gain and loss.

Multiple main effects of a covariate variable: Income

A secondary finding of the analysis was the pervasive effect of income across differing environmental behaviors. Generally, low- and average-income earners showed higher levels of pro-environmental behavior than high income earners. Especially interesting was the finding that low income earners even showed less support for policy that prioritized job security and a stable economy above climate change mitigation. Such a pattern may be explained by findings that lower income earners are among the most vulnerable to the impacts of climate change (Pearson & Schuldt, 2018). Another benefit of using a spread of environmental measures was this ability to see how income impacts performance across behaviors that have differing financial constraints. This is important as socio-demographic predictors of environmental behaviors have shown to be varied across behaviors with differing cost demands (Lavelle et al., 2015; Otto et al., 2016). Despite non-response and unstable group sizes often hindering the inclusion of income as a covariate (Ziegler, 2017), such findings demonstrate the necessity of considering socio-demographic variables upon a range of pro-environmental behaviors.

Limitations and future directions

The major limitation of this study has been previously discussed regarding the non-significant manipulation checks of social identity. This limitation may however be influenced by the low post-hoc power indicated in many of the analyses. Statistical power was reduced by the unforeseen need to include income as a fixed control factor in analyses, while the novelty of the study design impeded the ability to calculate accurate a priori power. This study does however now serve as a reference point for future studies to calculate appropriate sample sizes for effects across a range of behavioral domains.

Another limitation includes the lack of qualitative filtering related to participant engagement in priming tasks. While writing activities were designed to elicit cognitions related to either Human or Environmental Identity, or gains and losses, some written responses reflected a trivial level of participant engagement. Future studies may include data cleaning criteria that addresses the engagement of participants in priming tasks.

In terms of sampling bias, we controlled for income because of the significant differences in participant income levels between the experimental groups. Despite the non-significant differences in other demographics between groups from the online cloud platform, future studies may cross validate if the findings from other samples are consistent.

While the attempt to operationalize political behavior with the voting for candidates constituted an important component in the research design, the factor analysis to identify the dimensions in the voting responses had a limitation due to the smaller sample size than required. This limitation suggests future opportunities for refining the measurement methodology concerning political behaviors.


The current study provided evidence for the interaction between effects of gain and loss framed environmental messaging and social identification upon multiple pro-environmental behaviors. This enforces the need to consider both these moderating cognitive influences when designing campaigns to motivate such behavior. Varying main effects of frame also indicated that the comparative strength of gain and loss framed messages upon pro-environmental behavior is likely dependent on the behavior itself. Further, difficulties in effectively priming superordinate Human and Environmental identities highlights the need for further development of potent manipulation materials. Developing these superordinate social identities can help to form coalitions amongst groups that may have been previously apathetic, or in conflict over environmental action (Batalha & Reynolds, 2012). The variation of effects across behaviors, as well as the significant effect of income across outcomes also represents the need to measure pro-environmental behavior multi-dimensionally, while accounting for socio-demographic predictors. Considering these outcomes, psychological investigation must endeavor to inform social campaigning that appeals to socially constructed senses of self, as well as prospect biases, to best cultivate behavior that tackles the current climate crisis. Understanding such motivations may also enable campaigns to motivate political level environmental behaviors, consequently increasing the demand for policy that protects the environment.


Not applicable.


    No conflicts of interest.


    • Eunro Lee, PhD. Eunro is honorary principal researcher at RMIT University, Australia having received her undergraduate degree at Seoul National University and her Master and PhD at Chungbuk National University in South Korea. Her research spans applied statistics in psychology, educational and social psychology, and forensic psychology.

    • Riley Dedman, Riley is a mental health support professional at Aclarity Health, Australia. He received his undergraduate degree and Honours in psychology at RMIT University, Australia. His research interests include environmental behaviour, social psychological interventions for behavioural change, and policy for mental health.


    Not applicable.

    • 1 Kashima et al., 2014: The Green shopping intention means = .4, -.1 for high and low identity groups respectively with the same SD = .19, N = 1058; Latimer et al., 2008: Inactivity risks/activity benefits means 4.24 and 2.76 for Gain Loss groups respectively (SD = 1. 43, 1.83; n = 128, 165). The post hoc power of the models was .96 at maximum for some models but remained lower as .62 for the interaction models. While small effects may not have detected due to the low power in some models, future studies may improve power using the findings of the present study.
    • 2 The data is not available due to the research ethics approved for the research where the participants consented only to the current use of their data. However, group level statistics is available by request to the corresponding author.
    • 3 Main Analyses using Analysis of Covariance (ANCOVA) were also run with income included as a continuous variable ranging from 1 (low) to 3 (high income) rather than a fixed factor treated as a categorical group variable. No different results were found although post-hoc power was increased when using the covariate model.