The effects of message framing in CSR advertising on consumers’ emotions, attitudes, and behavioral intentions

ABSTRACT While recent research on sustainability communication demonstrates the relevance of message framing, research on the effects of message framing on consumers’ emotions is scant. Using the Stimulus-Organism-Response (S-O-R) framework, this paper examines the impact of environmental advertisements (stimuli) on two discrete emotions – hope and guilt – (organism) and how these emotions influence consumers’ behavioral intentions (responses). Relying on the prospect theory, this study focuses on positive (gain) and negative (loss) frames. Study 1 shows that, in the context of Corporate Social Responsibility (CSR), a gain message elicits hope while a loss-message triggers guilt. Study 2 shows that both emotions positively influence consumers’ attitudes toward the cause; however, only hope affects attitude toward the company. Attitudes toward the cause and the company, in turn, influence consumers’ behavioral intentions. 虽然最近关于可持续性传播的研究证明了信息框架的相关性，但关于信息框架对消费者情绪的影响的研究却很少. 利用刺激-有机体反应 (S-O-R) 框架, 本文考察了环境广告 (刺激) 对两种离散情绪——希望和内疚 (有机体) 的影响, 以及这些情绪如何影响消费者的行为意图 (反应). 基于前景理论, 本研究主要关注正 (增益) 和负 (损耗) 帧. 研究1表明, 在企业社会责任 (CSR) 的背景下, 收益信息引发希望, 而损失信息引发内疚. 研究2表明, 两种情绪都会积极影响消费者对原因的态度; 然而, 只有希望会影响人们对公司的态度. 对事业和公司的态度反过来会影响消费者的行为意图.


Introduction
Individuals and companies are increasingly coming under pressure to exhibit responsible behaviors (Ettinger et al., 2021;White et al., 2019), thus making Corporate Social Responsibility (CSR) more important than ever (DiRusso & Myrick, 2021;Vitell, 2015). Not surprisingly, research on CSR in the hospitality context has received considerable attention over the last decade (García de Los Salmones et al., 2021;Grazzini et al., 2018;Guzzo et al., 2020;Qian et al., 2021;Rhou & Singal, 2020;Xu & Jeong, 2019). Environmental Hotel green message effectiveness is influenced not by the positive or negative aspects but the general or specific aspect of the message.
(Continued) recycling intentions (Grazzini et al., 2018); iv) food waste reduction intentions (Huang et al., 2021); and v) donation intentions (Huang & Liu, 2020). Dhanesh and Nekmat (2019, p. 30) state that: Most research on the effect of message frames on attitudes and behavioral intentions has focused on cognitive processes rather than affective processes. Findings of these studies bolster the emergent body of research that examines the role of affect in evaluating messageframing outcomes.
In summary, we identify the following gaps in CSR communication studies in the hospitality industry: i) a lack of clarity on which is better, positive or negative message framing, and ii) a lack of research on consumers' emotional responses to CSR communication.

Research framework: stimulus-organism-response (S-O-R)
In this study, we use the Stimulus-Organism-Response (S-O-R) framework as a theoretical lens (Mehrabian & Russell, 1974). We posit that exposure to a CSR advertisement induces discrete emotions (i.e., of hope or guilt), which in turn lead to attitudinal and behavioral responses, i.e., consistent with the framework, an advert (stimulus) leads to an emotion (organism), which leads to a behavior (response). Specifically, drawing on the prospect theory (Kahneman & Tversky, 1979), this study investigates the role of loss or gain-framing in CSR messages and its effect on consumers' responses. Our conceptual model is shown in Figure 1.

Stimulus: message framing in CSR advertising
Message framing refers to a technique whereby objectively equivalent information is described in different ways to elicit distinct responses and choices from audiences by making a specific aspect of a perceived reality more pertinent (Segev et al., 2015). Framing has the potential to influence behavioral outcomes via emotions, yet message framing in CSR communications remains under-researched (Han et al., 2019;Overton, 2018). Messages can be positively or negatively framed (Meyers-Levy & Maheswaran, 2004;  Segev et al., 2015). Those that are positively framed stress either the benefits to be gained or the negative consequences avoided (e.g., "If you reuse towels, you conserve natural resources."). Negatively framed messages stress either the negative consequences to be incurred or the benefits foregone (e.g., "If you do not reuse towels, natural resources will not be conserved.").
According to the prospect theory (Kahneman & Tversky, 1979), loss-framed (negative) messages are more effective at encouraging behaviors that involve risk, while gain-framed (positive) messages are more effective at encouraging cautious behaviors (Loroz, 2007;Segev et al., 2015). As most sustainable behaviors (for example, using reusable coffee cups) can be classified as low-risk and preventive, the prospect theory suggests that gainframed messages might be more effective in environmental communications (Loroz, 2007).

Organism: emotions in response to CSR advertising
To better understand the effectiveness of CSR advertising, it is important to consider the role of emotions. Previous advertising research has shown that affect mediates the relationship between cognition and behavior (Holbrook & Batra, 1987;Vakratsas & Ambler, 1999), while Poels and Dewitte (2019) indicate that there is a paucity of research on discrete emotions.
In CSR advertising, understanding emotions is important as emotions can influence consumers' attitudes toward the cause and the company (Lu, 2016). Emotions clearly influence people's environmental attitudes and behaviors (Nabi et al., 2018), and they help generate a sense of urgency regarding the consequences of climate change, which are often distant and, therefore, somewhat abstract to the individual (Bilandzic et al., 2017).

Effects of message framing on hope and guilt
Previous research suggests that hope and guilt are particularly relevant within the context of CSR advertising (Bilandzic et al., 2017). Hope is a future-oriented, discrete emotion (Winterich & Haws, 2011) induced by cognitive appraisal of an uncertain, goal-congruent future result that is important to the individual (Chadwick, 2015). Guilt arises when a person believes himself to have engaged in morally deficient behaviors; even more so if such behaviors have harmed others (Kapoor et al., 2021;Lazarus, 1991).
A gain-framed message emphasizes the positive consequences of an (in)action while a loss-framed message highlights the negative consequences of an (in)action (Segev et al., 2015). As gain-framed messages highlight the potentially positive outcomes of an action, or omission to act, they are likely to evoke positive emotions, particularly hope. In the context of CSR communications, Bilandzic et al. (2017) show that a positive, gain-framed message (i.e., one that outlines the desirable effects of engaging in climate protection) generates hope, while a loss-framed message (i.e., one that outlines the undesirable effects of not engaging in climate protection) decreases feelings of hope while strengthening feelings of guilt and fear. Similarly, for climate change, Nabi et al. (2018) show that gain-framed (vs. loss-framed) messages induce hope, while loss-framed (vs. gain-framed) messages generate fear. Taken together, we posit the following predictions: Hypothesis 1a: Gain-framed messages elicit higher levels of hope than loss-framed messages.
Hypothesis 1b: Loss-framed messages elicit higher levels of guilt than gain-framed messages.

Effects of hope and guilt on attitudes
Empirical evidence is mixed regarding the effects of hope and guilt on consumers' attitudes. Ojala (2012) establishes that a constructive type of hope on climate change issues, which is not based on denial, significantly increases engagement in proenvironmental behavior. Similarly, Nabi et al. (2018) find hope to be a key mediator between gain-framed messages and consumers' attitudes toward climate change policies. Chadwick (2015) demonstrates that hope increases a person's interest in climate protection but does not have a significant effect on that person's behavioral intentions. Conversely, Bilandzic et al. (2017) conclude that hope has a negative effect on behavioral intentions as gain-framed messages reduce an individual's willingness to make sacrifices for a climate change cause. They found that loss-framed messages have a positive influence on an individual's willingness to make sacrifices, due to higher levels of fear and guilt. Considering the mixed results, based on the findings of Ojala (2012)

Effects of attitude on behavioral intentions
Behavioral intentions are an individual's readiness to perform the behavior and, as such, encompass an individual's motivations (Ajzen, 1991). Previous studies in the CSR context (Dhanesh & Nekmat, 2019;Grau & Folse, 2007;Inoue et al., 2017;Johnson-Young & Magee, 2019;Overton, 2018) have called for further research into demonstrating a direct link between attitudinal and behavioral intention variables.
In a meta-analysis of behavioral intentions toward environmentally friendly initiatives in the hospitality sector, attitudes are considered to be one of the consumers' internalized perceptions (Gao et al., 2016). García de Los Salmones et al. (2021) find that an individual's attitude toward a social media post (e.g., attitude toward the cause) has a double influence: first, it improves the individual's attitude toward the company and, second, it increases their behavioral intentions to act (e.g., their intention to share the post). Ertz et al. (2017) show that pro-environmental attitudes have a significant influence on behavioral intentions toward reusable containers. Wang and Anderson (2011) find that brand attitude acts as a mediator between an individual's assessment of CSR practices and their purchase intentions. Dhanesh and Nekmat (2019) demonstrate that attitude toward an organization positively influences an individual's behavioral intentions toward that organization. Based on the outlined empirical findings, it is proposed that: Hypothesis 4: A favorable attitude toward the cause has a positive impact on attitude toward the company.

Method
Given the numerous empirical challenges in CSR research, it was crucial to conduct experiments to establish a causal relationship between message framing, emotions, attitudes, and behaviors (Shadish et al., 2002). We used two sequential studies, as discussed later. In Study 1, we analyzed the direct effects of message framing (gain vs. loss) on hope and guilt (Hypotheses 1a and 1b). In Study 2, we examined the impacts of hope and guilt on consumer attitudes toward the environmental cause and toward the company (Hypotheses 2a, 2b, 3a and 3b). In addition, we examined the effect of attitude toward the cause on attitude toward the company (Hypothesis 4). Lastly, we analyzed the impact of the two attitudinal measures (toward the cause and toward the company) on behavioral intentions toward the cause and toward the company (Hypotheses 5a, 5b, 6a and 6b).

Design and sampling
This study adopted a single-factor, between-subjects design (message framing: gain vs. loss). Participants were randomly assigned to one of the two experimental conditions. Participants were exposed to a social media post ( Figure 2) that promoted reusable coffee cups (RCCs). We used a fictitious company to mitigate any potentially confounding effects of previous brand experiences (Geuens & De Pelsmacker, 2017) and we also controlled for gender, age, and level of education (Ribeiro et al., 2022).
Respondents were recruited online via Amazon Mechanical Turk (MTurk) and offered a small monetary compensation for completing a survey. Various CSR studies, published in reputable journals, have used this platform (e.g., Xu & Jeong, 2019;Zhang & Yang, 2019). Prior research suggests that online panels allow for fast data collection and participants tend to be more diverse than other sampling methods, such as college student samples (Buhrmester et al., 2011).
A total of 978 participants took part in Study 1. We removed responses of 20 participants who failed the attention check (i.e., "The Facebook post I just read was primarily about the following topic: Disposable and reusable coffee cups; Advertisement for a new product from Atlas; None of the above"). A one-way ANOVA showed that there were no significant differences between the two conditions (loss-or gain-framed message) in terms of age (M age = 38.1), gender (male: 50.7%), education (undergraduate degree: 26.7%) and ethnicity (non-Hispanic white: 73.4%). Thus, subsample equivalence was met as the two samples were homogeneous and comparable (Geuens & De Pelsmacker, 2017).

Measures
Following their exposure to the CSR message, the participants were asked to rate their emotional responses to the ad. Hope was measured using three items (optimistic, encouraged, and hopeful), adapted from Richins (1997). Guilt was measured using three items (remorseful, guilty, and conscience-stricken), adapted from Bilandzic and Sukalla (2019). All the responses were rated on seven-point Likert-type scales (1 = strongly disagree; 7 = strongly agree). To assess perceived message framing (i.e., potential loss or potential gain of the message), the 958 remaining participants were then asked to rate the extent to which they believed: "The Facebook post I just saw included information that primarily focused on gains or losses (1 = potential losses; 7 = potential gains)." Lastly, demographic information was collected.

Results
Manipulation check: A one-way ANOVA on the manipulation check question indicated that the manipulation was successful; participants in the loss-framed message condition revealed having felt potential loss, compared to those in the gain-framed message condition (M potentialLoss = 1.61, SD = 0.727 vs. M potentialGain = 6.30, SD = .736; F = (1, 955) 9703, p < .001, η 2 = 0.910). This suggested the manipulation worked as intended. Hypothesis testing: We predicted that the gain-framed message would elicit hope (H1a), while the loss-framed message would induce guilt (H1b). To test H1a and H1b, an ordinary least square (OLS) regression was undertaken with message framing (gain vs. loss) as the independent variable and with hope and guilt as dependent variables, while controlling for age, gender and level of education. To test H1a we used a dummy coding message frame (gain = 1; loss = 0). To test H1b the dummy code was swapped (gain = 0; loss = 1; Hayes & Preacher, 2014). The results indicated that the gain-framed message elicited hope (β = 0.47, t = 26.63, p < .001), while the loss-framed message triggered feelings of guilt (β = 0.27, t = 17.03, p < .001). Taken together, these results provide support for H1a and H1b.

Study 2: effects of emotions on consumer responses
Design and sampling As in Study 1, a between-subjects design (message framing: gain vs. loss) was adopted, and participants were randomly assigned to one of the two experimental conditions. A total of 472 participants were recruited using MTurk and offered a small monetary compensation for completing a survey. Six responses were removed due to failed attention checks resulting in a final sample of 466 participants (Gain 234 and Loss 232). Similar to Study 1, a one-way ANOVA showed that there were no significant differences between the two groups in terms of age (M age = 39.9), gender (male: 50.6%; female: 48.7%; prefer not say: 0.7%), education (undergraduate degree: 43.8%; some college degree = 26.7%; high school degree or less = 11.8%; postgraduate degree = 10.2%, other: 7.5%) and ethnicity (non-Hispanic white: 73.6%, African or Black American = 11.3%, Asian = 6.7, Hispanics = 6%; other: 2.4%).

Measures
After their exposure to the same message frames as in Study 1 (see, Figure 2), participants were asked to indicate their emotional responses to six items (three related to hope and three to guilt) using the same Likert-type scales (see, Table 2). Attitude toward the environmental cause was measured on a 7-point Likert scale using three items adapted from Johnson-Young and Magee (2019). Attitude toward the company was measured using four items, following the semantic differential scale adapted from K. Kim et al. (2015). The items used to measure behavioral intention toward the cause were adapted from Price and Arnould (1999); the questions measured the likelihood of the participants to use RCCs. A 2-item scale was used to measure behavioral intention toward the company (Price & Arnould, 1999;Putrevu & Lord, 1994). Finally, the last section of the survey contained demographic questions.
As a cross-sectional survey was used to collect the data, common method bias (CMB) needed to be ruled out. We used Harman's single factor test, with all 17 of the items loaded into a single, unrotated, exploratory factor. The results demonstrated that 41.42% of the variance of the items could be explained through a single factor, suggesting that common method variance (CMV) was not a pervasive issue in our study (Jordan & Troth, 2020;Ribeiro et al., 2021).

Results
Before testing the structural model, a confirmatory factor analysis (CFA) was conducted following a two-step approach (Anderson & Gerbing, 1988). Results of the CFA (provided by IBM AMOS v27 software) indicated that the model showed a good fit to the data: chisquare (χ 2 = 243.032, df = 102, χ 2 /df = 2.383), comparative fit index (CFI) = 0.991, Tucker-Lewis index (TLI) = 0.998, root mean square error of approximation (RMSEA) = 0.038, standardized root mean square residual (SRMR) = 0.035 (Hu & Bentler, 1999), which demonstrates the accuracy of the model (Kline, 2016). As presented in Table 2, the factor loadings were higher than 0.80 and significant (p < .001). Table 3 shows that the average variance extracted (AVE) exceeded the 0.50 cutoff value for all constructs and was higher  Note: α = Cronbach alpha; CR = composite reliability; AVE = average variance extracted. The bold elements in the diagonal matrix are the square root of the average variance extracted (AVE). Interconstruct correlations are shown off-diagonal. All correlations are significant at p <0.001 level than the squared correlation between the constructs. These results provided evidence for both convergent and discriminant validity of all the constructs (Fornell & Larcker, 1981). Cronbach's alpha and composite reliability (CR) values for all the constructs were greater than the recommended threshold of 0.70 (Table 3), which evidenced the reliability of the construct (Hair et al., 2019). Overall, these results revealed that the measurement model was adequate to assess the latent variables used in the study's structural model. After assessing the validity and reliability of the measurement model, the structural model was tested. Fit indices showed that the model fit the data reasonably well (χ 2 = 285.131, df = 109, χ 2 /df = 2.616, CFI = 0.990, TLI = 0.975, RMSEA = 0.059, SRMR = 0.052; Hu & Bentler, 1999) and was strong enough to test the proposed hypotheses. Table 4 shows the results of the structural model with standardized paths of the proposed relationships. The results indicate that eight out of the nine hypotheses are supported. Specifically, hope positively influences both attitude toward the cause (β = 0.48, p < .001) and attitude toward the company (β = 0.40, p < .001); thus, H3a and H3b are supported. Similarly, guilt significantly influences attitude toward the cause (β = 0.42, p < .001) but does not influence attitude toward the company (β = 0.02, p > .05). Thus, H4a is supported while H4b is rejected. Additionally, attitude toward the cause positively affects attitude toward the company (β = 0.37, p < .001) as well as behavioral intentions toward the cause (β = 0.70, p < .001) and the company (β = 0.64, p < .001); therefore, H5, H6a and H6b are supported. Lastly, attitude toward the company positively influences behavioral intentions toward the cause (β = 0.15, p < .001) and the company (β = .45, p < .001), lending support to H7a and H7b.

Post hoc analysis
Further analyses were performed to estimate separate structural models for each of the message frame conditions (gain and loss) to see if there were any substantive differences in their structural relationships (Vandenberg & Lance, 2000). A structural invariance test was performed to determine whether structural invariance existed in the model (emotions → attitudes → behaviors) when exposed to outcome message framing (i.e., gain and loss) in CSR advertising. A baseline (unconstrained) model with maximum Likelihood (ML) estimation was generated. The results of the baseline model (where all the loadings across the two message frames were constrained to be equivalent) demonstrated that the model fit  Table 5, the baseline model, where paths were allowed to vary freely (i.e., gain and loss), was compared to nine constrained models in which the paths were equally constrained in sequence across the two message framing groups. As anticipated, results of the χ 2 difference test revealed that the two message framing groups varied at the model level (Δχ 2 (9) = 62.213, p < .001), indicating that there were differences in the path relationships between gain and loss outcome message framing in CSR advertising.
To understand which path estimates varied between the two message framing conditions, we calculated the statistical differences. The results, depicted in Table 5, demonstrate that the effect of hope on both attitude toward the cause and attitude toward the company was stronger in the gain (vs. loss) message condition. Conversely, the effect of guilt on attitude toward the cause was stronger in the loss (vs. gain) message condition. However, the effect of guilt on attitude toward the company did not differ across the two framing conditions. In addition, the positive effect of attitude toward the cause on company attitude was stronger in the loss (vs. gain) message condition. However, the effect of attitude toward the cause on behavioral intention toward the cause and toward the company did not differ between the two message framed conditions. Lastly, the effect of attitude toward the company on behavioral intention toward the cause was significant only in the loss message condition.

Conclusions and implications
The contributions of this paper are two-fold: first, our study demonstrates how message framing elicits two discrete emotions of hope and guilt; second, we examine how such emotions influence consumer attitudes and behavioral intentions toward the cause and the company. This study adds to the body of literature on the effects of discrete emotions on consumer responses to advertising stimuli (Poels & Dewitte, 2019), particularly in the context of message framing of CSR ads (Segev et al., 2015).

Theoretical implications
Research on emotions, in the context of message framing, has only emerged relatively recently and remains rather atheoretical (Nabi et al., 2018). Prior research in climate change communication highlighted the relevance of message framing and emotions when examining perceived threats and people's willingness to make sacrifices for the environment (Bilandzic et al., 2017;Nabi et al., 2018). However, previous hospitality research on environmental CSR communication has largely ignored consumers' emotional responses.
In response to the call to examine the effectiveness of message framing strategies (Segev et al., 2015), there has been an increase in research on the impact of message framing on consumers' responses to CSR communication. However, the findings have been mixed. Several studies have suggested that negatively framed CSR messages are more effective (Grazzini et al., 2018;Randle et al., 2019), while others have concluded that positively framed messages lead to more positive evaluations (Kim & Kim, 2014). This research offers two main contributions. First, to reconcile the mixed findings on message framing, we examine the impact of message framing on consumers' emotional responses (particularly, hope and guilt). Second, we demonstrate the relevance of the S-O-R framework in the context of CSR communications, supporting the influence of advertising stimuli (i.e., message framing) on the organism (i.e., consumers' emotions), which in turn influences consumers' responses (i.e., their attitudes and behavioral intentions).
Congruent with prior research in climate change communication (Bilandzic et al., 2017;Nabi et al., 2018), our findings from Study 1 show that message framing (gain vs. loss) induces hope and guilt (respectively) in the context of pro-environmental communications. We further demonstrate that loss-framed messages generate higher levels of negative emotions, such as guilt (Nabi et al., 2018), and that guilt is linked to consumers' attitudes toward the cause; this finding is consistent with prior research that shows the power of guilt at enhancing consumer participation in CSR initiatives (Bilandzic et al., 2017). Feelings of hope also influence consumers' attitude toward the CSR cause (Ahn, 2021). However, only hope has a positive effect on attitude toward the company. Guilt is triggered when an individual perceives him/herself to have engaged in a moral transgression (Lazarus, 1991) and, therefore, it is not surprising that guilt fails to influence consumer attitudes toward the company.
The impact of framing on environmental attitudes and behaviors remains inconclusive (Homar & Cvelbar, 2021). The evidence is mixed in relation to whether CSR messaging influences attitudinal, as well as behavioral, outcomes (Dhanesh & Nekmat, 2019;Grau & Folse, 2007;Inoue et al., 2017;Johnson-Young & Magee, 2019;Overton, 2018). Prior research suggests that consumers' perceptions of a company's CSR activities have a halo effect (Albus & Ro, 2017). A positive attitude toward a cause improves a consumer's attitude toward the company. Consistent with prior studies (García de Los Salmones et al., 2021), we demonstrate that favorable attitudes toward a CSR cause lead to positive attitudes toward the company. Additionally, our research shows the direct effects of attitudinal responses to behavioral intentions in the context of CSR. Specifically, consumers' attitudes toward both the CSR cause and the company have a positive effect on consumers' behavioral intentions (Ojala, 2012).

Practical implications
The study provides practical insights for the hospitality industry to enhance the effectiveness of their CSR communication. As evidenced by National Geographic's Planet or Plastic campaign, it is relatively easy to incorporate cues inducing discrete emotions into CSR messages (DiRusso & Myrick, 2021). However, as only hope-inducing messages (i.e., gainframed messages) have a positive effect on consumers' attitudes toward a company, hospitality firms are advised to stay away from loss-framed CSR messaging. To induce feelings of hope, companies can either highlight the positive outcomes of their CSR initiative or highlight the negative consequences avoided, that have resulted from the target behavior. Headlines such as "one reusable cup in your hand = 1,000 single-use cups not in landfill" or "reusable coffee saves our sea life," supported by evidence, can be used by companies to reinforce the positive impacts that their customers are having, or the negative impacts they are avoiding, with their purchase.

Limitations and avenues for future research
The use of behavioral intention variables, instead of measuring actual behavior, is a limitation. There is a gap between intentions and actions in the CSR context, which is likely due to two reasons. First, actual purchase behaviors are more complex than intentions because numerous factors (e.g., price, situational circumstances) play a role in the decisionmaking process (Inoue et al., 2017). Second, topics such as responsible behaviors, which are loaded with societal expectations, create a social desirability bias whereby participants respond in accordance with general expectations (Babakhani et al., 2017). Consequently, future research in this realm should use behavioral data to validate previous findings on behavioral intentions. In addition, future research should investigate subjective norms and perceived behavioral controls, both of which are likely to have a major influence on behavioral intentions and actions. Lastly, this study used two online experiments to test the proposed hypotheses. Further research could combine lab experiments (i.e., electrodermal, facial reader recognition) with field experiments to examine the influence of CSR stimuli on consumers' emotions (e.g., Gómez-Carmona et al., 2021), attitudes, and behaviors.

Disclosure statement
No potential conflict of interest was reported by the author(s).

Funding
This work is partially financed by Portuguese Funds provided by FCT -Fundação para a Ciência e a Tecnologia (Foundation for Science and Technology, Portugal) through project UIDB/04020/2020.