Introduction

Due to the affordances of social media platforms and mobile connectivity, online consumer activism has become an increasingly popular way for consumers to take action against corporate wrongdoings (Braunsberger and Buckler 2011; Heldman 2017; Makarem and Jae 2016; Yousaf et al. 2021). Social media connectivity creates a more democratic space that empowers consumers to express their grievances and demands in highly personalized ways via microlevel action frames (Gray et al. 2015; Bennett 2012; Bennett and Segerberg 2013). Microlevel action frames (MAFs – or simply, microframes) are defined as highly-individualized discourses that can coexist within an activism campaign or social movement (Gray et al. 2015). Yet, there is limited research that has examined online consumer activism MAFs and their interplay in eliciting consumer support in the digital space.

Recent studies on online boycotting (Makarem and Jae 2016; Yousaf et al. 2021) found that consumer activism causes (issues) and targets (an offending brand) may play a motivational role in galvanizing participation on social media. Boycott supporters may express themselves on social media through posts oriented around a self-relevant cause to outwardly signal to their network the importance of taking action to defend an issue of public interest, such as human rights, animal rights, or environmental protection (Makarem and Jae 2016) – a cause-oriented MAF, as we call it. Alternatively, consumers can post on social media with an emphasis on targeting the offending brand (Makarem and Jae 2016; Yousaf et al. 2021) – denoted in this study as a brand-oriented MAF. Moreover, additional MAFs, such as information sharing and case contextualization microframes, may also emerge during online activism and supplement online participation (Makarem and Jae 2016; Legocki et al. 2020). Makarem and Jae (2016) suggest it is imperative to further examine (1) how activism causes and targets can thematically orient social media content production and influence the emergence of MAFs, (2) how different MAFs can create and sustain consumer mobilization over time, and (3) how MAFs are curated and amplified in the digital space via likes and shares (Bennett et al. 2014).

In this study, we aim to further consumer activism research by theoretically conceptualizing and empirically examining the MAFs consumer activists use to influence public awareness and perceived egregiousness of a brand transgression episode through user-generated content on social media (Klein et al. 2004). Empirically, we examine a real-world online consumer activism campaign against Carrefour, a French supermarket chain in Brazil, which occurred after a store employee killed a dog on company property. We theorize how online MAFs can influence online participation and the longevity of an online consumer activism anti-brand campaign. Additionally, based on our results, we propose the Online Consumer Activism Microframing (OCAM) model, presented in the Discussion section.

Drawing from marketing, political science, and social movement studies, our findings suggest that the Carrefour action was distinctively sustained by brand and cause-oriented MAFs, defined by social media posts focused on the cause (animal rights) and public fury against the consumer activism target (the Carrefour brand), respectively. Additionally, we identified information sharing and two case contextualization MAFs that broadened the scope of activists’ peer-content production and infused the online campaign with personal reflections concerning the Carrefour case.

Moreover, we show that brand-oriented MAF is most prominent in the initial days of the movement. These early expressions of brand fury trail off over time, having generated a spark that attracted attention from a broader base of social media users. In contrast, the cause-oriented MAF offered stable support for sustained action over time. From a practitioner standpoint, we conceptualize how various MAFs influence the longevity of an online consumer activism campaign and discuss how corporations can strategically react and respond to each MAF. To our knowledge, this is the first study to longitudinally examine the complex interplay between consumer activism causes and targets to influence online content production and curation. Several opportunities for future research are discussed.

Collective action framing

“Fight for the things that you care about, but do it in a way that will lead others to join you,” said Supreme Court Justice Ruth Bader Ginsburg concerning her legacy in the protection and promotion of women’s rights (Harvard Radcliffe Institute 2015). Ginsburg’s statement captures an important premise of online consumer activism campaigns: For people to join in, it is paramount that individuals are able to express their support for things that they care about and do so on their own terms (Micheletti 2003). Therefore, activism campaigns’ success likelihood, in terms of online public support, hinges on the ability to create a platform for consumers to express the personal meaning of a problem or a cause (Bennett and Segerberg 2013).

This is often accomplished via personalized collective action frames, i.e., microlevel framing (Gray et al. 2015). Microlevel framing is a communicative microprocess of signification about a problem through which individuals interact to create new meanings of “what happened” or “why it matters” (Gray et al. 2015). By enabling activists to express their personal viewpoints, frames play a vital role in motivating people to take action, identifying and attributing blame to the source of the problem, and orchestrating a plan of action (Benford and Snow 2000; Bennett and Segerberg 2013).

In the context of online activism, frames can be easily personalized through peer-generated content in the form of social media posts (Bennett and Segerberg 2013; Bennett et al. 2014). Additionally, posts on social media can be amplified through likes and shares, which are curation mechanisms to virally spread personally relevant frames (Bennett et al. 2014). Therefore, social media content production and curation represent two important mechanisms in online activism that enables consumers to keep an online campaign alive (Bennett and Segerberg 2013).

Collective action framing is a theoretical perspective that has enabled scholars to understand the signification mechanisms through which individuals are energized to act (Benford and Snow 2000). This has been extensively studied in the social movement literature in the context of different movements, such as Occupy Wall Street (Bennett 2003; Ganesh and Stohl 2013), Black Lives Matter (Bonilla and Tillery 2020), Me Too (Xiong et al. 2019), and Anti-Fracking (Williams and Sovacool 2019). Although not always explicitly adopted as a theoretical framework in consumer behavior studies, scholars have shown the importance of frames for the emergence of anti-brand online communities (Hollenbeck and Zinkhan 2010; Kozinets and Handelman 2004) and ethical consumerism movements (Valor et al. 2017).

Still, there is limited knowledge about how MAFs influence the emergence and longevity of online consumer activism. Legocki et al. (2020) investigated a case of citizen digital vigilantism on Twitter related to the violent episodes surrounding the 2017 Charlottesville Unite the Right rally. They identified five distinct content categories that represented the different ways citizens framed their grievances and sustained the mobilization over 19 days following the rally. Makarem and Jae (2016), with a focus on online consumer activism, cross-sectionally analyzed tweets from multiple boycott campaigns to understand how boycott adherents differently expressed their participation motives. The authors proposed that online boycotting can be motivated by causes and brands, which may also stimulate information sharing and personal reflections. Similarly, Yousaf et al. (2021) qualitatively analyzed social media posts supporting the #BoycottMurree campaign and found that public support for consumer protection (the cause) and the local hospitality industry (the target) were highly influential in online boycott participation.

Together, these studies indicate that causes and targets can motivate consumer activism, but more research is needed to understand how they may thematically orient social media content production. Moreover, it is unclear how brand, cause-oriented, information sharing, and case contextualization MAFs emerge, evolve, and are amplified (or silenced) during an online consumer activism campaign. We begin to unravel this lack of clarity by formally theorizing why brand and cause-oriented MAFs are pivotal in spurring and sustaining consumer  activism. After that, we formally define information sharing and case contextualization MAFs.

Consumer activism causes and targets

New social movements (NSM) theorists (Kozinets and Handelman 2004; Melucci 1995) suggest that self-relevant causes can inform the development of shared collective action goals, which in turn facilitate sustained action over time. NSM theory views social mobilizations as unstructured, informal networks of citizens with a common perspective about an issue of public interest, from social to environmental causes (Buechler 1995; Copeland 2014; Diani and Porta 2006).

The literatures on lifestyle politics and political consumption emphasize that personal connection to a cause plays a vital role in the choice to participate in  consumer activism. Lifestyle politics is conceptualized as the adoption of a lifestyle congruent with one's moral values (de Moor 2017), and proposes that people may deliberately avoid certain products and brands because of their incongruence with an adopted lifestyle (Haenfler et al. 2012). Political consumption is defined as the use of consumption as a tool to advocate for and promote changes related to self-relevance issues (Micheletti 2003). Political consumers are understood to be motivated to take action against a company because of their existing concerns about a self-relevant cause (Micheletti and Stolle 2007).

The consumer activism target also plays an important role in spurring consumer activism participation. Consumers can establish human-like relationships with brands (Fournier 1998), which serve as symbolic platforms for self-concept expression (Belk 1988; Sinha and Lu 2016). Brand transgressions can be perceived as a violation of subjective relational norms, detrimental to the longevity of the consumer-brand relationship (Aaker et al. 2004; Aggarwal 2004) and consumers’ self-connection to the brand (Sinha and Lu 2016; Trump 2014). Therefore, we theorize that both the cause and the target brand jointly serve to bring people together and mobilize them to act. In doing so, cause and target may play a fundamental role in online content production by differently influencing MAFs.

Brand-oriented MAF

We borrow from the brand management literature to articulate a concept of MAF with a brand orientation in the context of consumer activism. Consumer-focused brand orientation has been termed “perceived brand orientation” (PBO). It refers to “respondents’ perception of the extent to which an organization engages in brand-oriented activities and behavior” (Mulyanegara 2011, p. 230). In other words, PBO answers the question: how do consumers view the firm’s brand? PBO has been linked to a variety of outcomes such as perceived product quality and category spending (Huddleston and Cassill 1990), perceived social, spiritual, and purpose-in-life benefits (Casidy 2013), service quality and WOM, and loyalty (Casidy 2014).

We propose that when consumers detect a perceived egregious brand transgression episode, brand-oriented online discourses may emerge. Perceived transgression egregiousness (Klein et al. 2004) can evoke strong negative emotions toward the brand (Grappi et al. 2013; Grégoire et al. 2009; Johnson et al. 2011; Romani et al. 2015; Xie and Bagozzi 2019), especially when the brand is strongly associated with consumers’ self-identity (Johnson et al. 2011). Congruent with the online consumer mobilization literature (Legocki et al. 2020; Makarem and Jae 2016; Yousaf et al. 2021; Østergaard et al. 2015), a brand-oriented MAF would contain high-intensity, negative emotional expressions and punitive calls to action against the target brand.

Cause-oriented MAF

To define a MAF with a cause orientation, we draw heavily from the political science and social movements literatures. The political science tradition focuses on theorizing consumer activism as a component of lifestyle politics. When individuals identify with a certain cause, they may change or adjust their lifestyle to promote social or environmental transformations at two levels: the individual, such as making a personal decision not to buy products tested on animals, or the collective, such as supporting a boycott against a company that conducts animal tests (Haenfler et al. 2012; de Moor 2017). NSM sees the collective level of lifestyle politics as an identity-based collective action created around shared support for a cause (Haenfler et al. 2012).

Lifestyle politics has infused consumption with other-regarding virtues and promoted individualized responsibility-taking for world problems. This blurs the line between the public and the private (Copeland 2014; Micheletti and Stolle 2012) and changes the meaning of being a good citizen (Thorson 2012). We propose that when consumer activism occurs based on shared consumer support for a cause, it gives rise to cause-oriented MAF. Therefore, a brand transgression would be more strongly expressed as a violation of the cause and consumers’ notion of justice and fairness.

Based on the theorization above, we propose that brand and cause can thematically orient social media content production through brand and cause-oriented MAFs that coexist within the same online  consumer activism anti-brand campaign. To the best of our knowledge, this inquiry that has not been directly investigated in the literature. Thus, we pose the question:

RQ1 Can cause and brand-oriented MAFs co-occur in an online consumer activism anti-brand campaign?

Information sharing and case contextualization MAFs

After defining cause and brand-oriented MAFs, in this subsection, we focus on the conceptualization of information sharing and case contextualization MAFs. Makarem and Jae (2016) found that information sharing was the third most frequent theme in online boycott campaigns on Twitter. Legocki et al. (2020) found that information sharing and opinion posts were the most prevalent themes among citizen-activists. Still, in their study, this topical focus only gained momentum after shaming and punishing content dominated the online conversation in the first two days of online mobilization.

An information sharing MAF could represent movement adherents’ efforts to increase the perceived legitimization of an anti-brand digital campaign by disseminating media news coverage (Friedman 1991; Tilly and Tarrow 2015). Also, opinion sharing may emerge as a case contextualization MAF representing consumer activists’ effort to contextualize and reflect about the broad implications or consequences of a brand transgression (Østergaard et al. 2015; Baumann et al. 2015; Kozinets and Handelman 2004).

Jointly, this limited literature indicates that information sharing and case contextualization MAFs may emerge in episodes of online activism. However, more research is required to determine what types of additional MAFs are likely to emerge and can co-exist with brand and cause-oriented MAFs. Therefore, we ask the following question:

RQ2 What other MAFs might emerge and co-occur with brand and cause-oriented MAFs in an online consumer activism anti-brand campaign?

Consumer activism MAFs evolution

Studying consumer activism from a longitudinal perspective is indispensable in a field dominated by cross-sectional studies (Makarem and Jae 2016). Based on the conceptualizations discussed above, MAFs are likely to follow different longitudinal patterns.

The service failure and brand betrayal literatures demonstrate that consumers’ desire for revenge against a brand may be a knee-jerk reaction that, while strong in the first moment, monotonically decreases over time (Grégoire et al. 2009, 2018). Thus, just like consumer anger seems to be short-lived in the context of a service failure, the brand-oriented MAF may dominate social media conversations in the early days of an online boycott and taper off as the boycott progresses. Additionally, because brand-oriented MAF may be highly-arousing due to strong negative sentiment against the brand, social media users may be more inclined to virally spread this type of MAF through likes and shares (Berger and Milkman 2012).

On the other hand, the cause-oriented MAF may behave differently as it may reflect a more enduring form of support for a cause. Cause support seems to be a durable individual trait connected to one’s self-identity and values (Micheletti 2003). Therefore, cause-oriented MAF may be a further expression of enduring commitment to a cause than a fleeting negative emotional response to a brand transgression. Thus, cause-oriented MAF might have a somewhat steady presence throughout a boycott campaign.

It is pivotal to examine the longitudinal patterns of online consumer activism as they may differently contribute to the longevity of the campaign. We ask:

RQ3 (a) How do MAFs change over time? (b) Does brand-oriented MAF monotonically decrease over time similarly to consumers’ desire for revenge? (c) Does cause-oriented MAF remain relatively stable over time?

MAFs amplification through content curation

Thus far, we have mostly focused on how MAFs are generated via peer-content production. We now turn to peer-content curation (Bennett et al. 2014), another mechanism responsible for sustaining and amplifying MAFs. Curation can be accomplished through online viral behaviors, such as likes and shares. Curation enables online activists to “sort, distribute, and draw timely attention” (Bennett et al. 2014, p. 245) to personally relevant content, thereby highlighting and sustaining MAFs of interest. Content curation represents consumer activists’ proactive intentions to virally propagate a message to their network (Alhabash and McAlister 2015). However, research is needed to understand what MAFs are more likely to be virally amplified and sustained by activists in an episode of online consumer activism.

One important feature that may facilitate MAF amplification is the negative sentiment evoked by activists’ online posts. Understanding how message features elicit viral behaviors in online consumer activism is a ripe area for investigation (Makarem and Jae 2016). Research has produced mixed findings suggesting that negative emotions may go viral more quickly (Berger and Milkman 2012; Tsugawa and Ohsaki 2015 –however, not supported by Tellis et al. 2019) because negatively-valenced posts can be more arousing (Berger and Milkman 2012). Therefore, it is possible that negative sentiment may be an important feature that generates MAF amplification, in particular for brand-oriented MAF. Given our earlier theorization, brand-oriented MAF may function as a conveyor of negative brand sentiment, and may be potentially higher than that of contained in other MAFs. This could be due to aggregated feelings of brand betrayal (Reimann et al. 2018) about a self-relevant cause violation (Trump 2014), which may incentivize higher amplification. Therefore, we pose our last set of research questions:

RQ4: (a) How are MAFs differently amplified through likes and shares? (b) Do posts related to brand-oriented MAF display higher levels of negative sentiment than other MAFs and, therefore, are these posts amplified at a higher rate?

Twitter and online consumer activism

Twitter data has been leveraged by researchers interested in online consumer activism because it allows the examination of users’ online discourses in a naturalistic setting via unobtrusive data collection (Makarem and Jae 2016; Yousaf et al. 2021; Rim et al. 2020; Legocki et al. 2020). Moreover, a growing body of literature has examined the role of Twitter as a “stitching mechanism” in a collective action, bringing together diverse MAFs that exist in otherwise disconnected networks (Agarwal et al. 2014). That is mainly because Twitter’s hashtag-based structure facilitates users to quickly discover, search, and engage with ongoing content of interest (Yang 2016). We propose that Twitter plays the same function for online activism—“stitching” together multiple MAFs in anti-brand consumer mobilizations.

The Carrefour case

To observe the distinct MAFs theorized above, we selected a case study in which the cause (animal rights) and the brand (Carrefour) are not explicitly related. This is expected to increase the distinction of unique words and terms associated with each MAF. Greater distinction facilitates the detection of underlying textual patterns, further explained in the methods section. Carrefour is a French food retail company and one of the largest supermarket chains in Brazil, with no explicit business activities or corporate responsibility practices geared toward animal rights (Carrefour 2021). Therefore, the researchers deemed this consumer activism campaign methodologically appropriate for further examination.

On November 28th, 2018, a dog named “Manchinha” was poisoned and beaten with a metal stick by an employee of Carrefour inside store premises. The store is located in the city of Osasco, in the state of São Paulo—Brazil. “Manchinha” is a Portuguese noun meaning “small spot,” in reference to the dog’s coat’s black spots. The dog died a few hours later at a local animal hospital due to severe bleeding (Tomaz 2018). According to the news, the employee attempted to scare the animal away, who had been living in the supermarket premises by the time of the incident.

Information about Manchinha’s death, including pictures and footage of her beating, rapidly went viral on social media. The case received intense news coverage. Activists created online petitions, organized demonstrations at Carrefour stores around the country, and orchestrated a boycott campaign against the brand. Brazilian celebrities and politicians joined the cause. Public opinion pressured congress to pass a bill with more strict penalties for animal abuse (Chagas 2018). On March 15th, 2019, Carrefour signed a plea deal proposed by the São Paulo Department of Justice to pay one million reais (Brazil’s currency) in fines (Ministério Público do Estado de São Paulo 2019).

Methods

We collected posts connected to the Carrefour action between November 28th, 2018, the date of the dog’s death, and March 20th, 2019, five days after Carrefour signed a plea deal with public prosecutors. The plea deal is important indicator of consumer activism campaign success that signals to participants that their effort paid off (Sen et al. 2001). We collected data from the immediate days following the plea deal signing to capture its influence on user-generated online content.

Twitter posts can be classified as tweets, users’ original posts and retweets, or replications of someone else’s tweet. Also, some tweets can be replies when users directly address or respond to another user in their posts. To identify a comprehensive collection of posts connected to the campaign, we created a lexicon of related keywords and hashtags. Search terms were employed individually or in combination with other terms for data extraction, as depicted in Fig. 1. We used a mix of "umbrella" and "event-specific" terms and hashtags, similar to Thorson and Wang (2020), in an iterative process in which the authors read multiple tweets from different points in time during the campaign to identify the optimal search terms. We then tested these search terms on Twitter multiple times to avoid any selection biases.

Fig. 1
figure 1

Data extraction search query. Note English translation presented between brackets when appropriate

Full size image

We collected tweets using Twitter’s application programming interface (API) via Kearney's (2019) rtweet R package (version 0.7.0). Twitter data were extracted retroactively using the platform full archive endpoint free-access process (Twitter Developer 2020). All Twitter data matching the search query were extracted, which resulted in 34,966 posts. After excluding non-Portuguese posts, the final dataset contained 32,637 posts, of which 5,975 were original tweets. All posts were computationally analyzed in their original language (Portuguese). One of the authors, a native Portuguese speaker fluent in English, was responsible for translating the selected posts presented in the manuscript to English.

Analytical strategy

Topic modeling to detect MAFs

To address RQ1 and 2, we used a topic modeling algorithm to detect MAFs. Topic modeling provides a powerful, commonly employed method to analyze social media textual data and discover patterns through a probabilistic approach based on the co-occurrence of words (Omar et al. 2015). Similar to other studies (e.g., Bian et al. 2016; Kaila 2020; Liu et al. 2017), topic modeling and subsequent analysis were carried out using original tweets only (N = 5,975), not retweets. Retweets are replications of original tweets and are typically viewed as a measure of content amplification (Bennett et al. 2014; Kim et al. 2014; Makarem and Jae 2016; Thorson and Wang 2020; Yang 2016).

The latent Dirichlet allocation (LDA) algorithm via Chang's (2015) lda R package (version 1.4.2) was used for conducting topic modeling. LDA uses a generative probabilistic algorithmic model to summarize and categorize large collections of text documents (Blei 2012; Blei et al. 2003). LDA assumes that tweets can be represented by a randomly generated mixture of hidden, latent topics, where a certain distribution of words characterizes each unobserved topic (Bian et al. 2016; Blei et al. 2003). Each tweet is treated as a vector of word counts resulting in a non-mutually exclusive classification indicating which topic a tweet is more likely to represent.

Data Preprocessing. We followed a sequence of commonly employed steps in the topic modeling literature, similar to Liu et al. (2017). First, we filtered out punctuation, numerals, and URLs and standardized text capitalization to lowercase. Next, using the tm text mining R package, version 0.7-7 (Feinerer et al. 2008), we deleted stop words, language connectors that add little value for topic meaningfulness, such as the equivalents in Portuguese to the, at, on, and which. Next, we grouped frequent words with the same origin and meaning; for example, the terms animals and animal were renamed as animal (Chen et al. 2018). Also, the brand name, Carrefour, was filtered out during the final topic classification because it appeared recurrently across tweets and did not contribute to topic differentiation. Finally, we tokenized tweets by breaking text strings into separate words, thus, rendering data clean and organized for topic modeling.

Implementation of latent Dirichlet allocation. Because LDA assumes a fixed number of topics (Blei 2012), it is recommended to evaluate model solutions with varied numbers of topics (Roberts et al. 2016). When the primary goal is to explore the underlying topics within a body of text, it is useful to couple a quantitative assessment of model performance with a qualitative evaluation of topic interpretability (Chang et al. 2009; Hannigan et al. 2019; Liu et al. 2017). Following Hannigan et al. (2019), we evaluated model perplexity across models until values began to trail off. Perplexity (D) is a log-likelihood-based measure of model accuracy, where lower values are indicators of better models (Bian et al. 2016; Blei et al. 2003). Perplexity of competing models is presented in Fig. 2, showing that the optimal solution was between four to ten topics.

Fig. 2
figure 2

Perplexity values by topic modeling solution

Full size image

Topic interpretability was assessed across model solutions by inspecting the top ten most probable words of each topic (Omar et al. 2015) and reading a sample of tweets (N = 100) within each topic (Reisenbichler and Reutterer 2019). This allowed us to qualitatively judge semantic coherency (Kuhn 2018; Omar et al. 2015), topic interpretation, and word exclusivity (Kuhn 2018). The final solution detected five MAFs and was retained for representing a good trade-off across performance indicators. The retained model considerably reduced perplexity (D = 872.18) and achieved optimal interpretability results with tweets associated with one MAF displaying a cohesive narrative substantially distinct from tweets associated with other MAFs (see Table 1). Moreover, the retained model displays good word exclusivity (see Fig. 3), and semantic coherence (see Fig. 4), with MAFs being negatively correlated with each other.

Table 1 Sample of tweets exemplifying microlevel action frames (MAFs)
Full size table
Fig. 3
figure 3

Probability membership for top 10 words by microlevel action frame (MAF)

Full size image
Fig. 4
figure 4

Microlevel action frames (MAFs) correlational matrices. Note Figure A displays MAF correlations across all original tweets (N = 5,975)

Full size image

Additional analyses

To answer RQ3(a,b,c), we analyzed the proportion of each MAF longitudinally during the time of the campaign. Data were smoothed via locally weighted least squares regressions to facilitate the identification of patterns (Ruppert and Wand 1994). To address RQ4(a,b), we employed negative binomial modeling to predict number of likes and shares (retweets), which is appropriate when predicting over-dispersed count variables (Heiberger and Holland 2009). Additionally, a sentiment analysis was implemented via Jockers’ (2022) R syuzhet package using Mohammad and Turney’s (2013) NRC emotional lexicon available in Portuguese. This analysis complements topic modeling (Hartmann et al. 2019) by detecting levels of negative sentiment and has been frequently used in computational marketing research, including studies of brand-consumer social media interaction (Rambocas and Pacheco 2018).

Results

Descriptive overview of the consumer activism campaign against Carrefour

We start with a descriptive overview of the online consumer activism against Carrefour, a common approach in big data research. A total of 26,398 users posted tweets matching our search criteria. On average, each user posted 1.24 times (SD = 2.06), with the top one percent posting an average of four times (max. 186 posts). As depicted in Fig. 5, the online collective action started on December 1st, 2018, three days after the dog’s death. It grew rapidly, with an initial spike in posting volume between December 3rd and 10th, 2018. In this period, a bill was sent to congress for more stringent penalties for animal abuse, and consumers organized several demonstrations across multiple Carrefour stores. We can also see a smaller spike around December 19th, 2018, which coincides with the conclusion of the police investigation, followed by a third spike around February 15th, 2019. The third spike coincided with a young Black man’s assassination by a security officer inside another supermarket store in Brazil named Extra and the opening of the first public animal hospital in the city of Osasco, Brazil (Correio Paulista 2019; Revista Exame 2019). A final peak is observed around March 15th, 2019, when Carrefour signs a plea deal.

Fig. 5
figure 5

Descriptive overview of  consumer activism campaign against Carrefour

Full size image

The Carrefour mobilization elicited the participation of a considerable number of users. While the rate of repeated participation by Twitter users is somewhat low (users, on average, posted between one or two times throughout the campaign), this finding is typical in studies of Twitter activism. In a longitudinal study of Twitter activism on one issue, Thorson and Wang (2020) found that 76 percent of users only tweeted once in the data set. The power of social media activism resides in the breadth of participating users, supported by the committed posting behaviors of a small subset of individual users.

The coexistence of brand, cause-oriented, information sharing, and case contextualization MAFs

Our results indicate brand- and cause-oriented MAFs (RQ1) and information sharing and case contextualization MAFs (RQ2) can co-occur within the same online consumer activism campaign. In total, our retained algorithmic model proposes five main MAFs.

MAF 1, labeled as brand orientation, is characterized by explicit attacks on the brand, organized under specific hashtags, e.g., #CarrefourMurderer. The collective fury against the brand is also vividly present in the language employed by users, such as “I will never set foot again in this slaughterhouse” (see Table 1). Also, brand-oriented posts are more likely to display attacks directed toward Carrefour through Twitter replies, which represent activists’ direct response to the brand. As displayed in Table 2, the odds of a tweet associated with a brand-oriented MAF containing a direct reply to the brand are estimated to be five to one, significantly higher than any other MAF. This indicates that the brand-oriented MAF is comprised of tweets representing the negative ways in which the incident has fractured consumers’ relationship with Carrefour, often with direct attacks on the brand, which equates to the definition of perceived brand orientation (PBO).

Table 2 Logistic regression predicting brand replies
Full size table

MAF 2 represents a cause orientation, characterized by tweets that aim to mobilize people to take action against Carrefour. However, distinct from the brand orientation, here, the activism campaign is framed around overarching political causes, including animal rights and veganism, with tweets calling for respect, love, and the humane treatment of animals. Hence, the brand violation is more closely framed around a transgression of a cause, which is congruent with the political consumerism view of consumer activism. For instance, users call for a Carrefour boycott to seek justice and fight “for animal liberation” and “for a world where people can learn to respect and love animals.”

MAF 3 was defined as information sharing due to its association with tweets sharing news and newsworthy information related to the cause and the brand. As displayed in Table 1, tweets in this group, for instance, shared information about picketing at Carrefour stores and the São Paulo Department of Justice’s work in prosecuting the case, and references to the city where the Carrefour store is located and the monetary amount of Carrefour’s fine.

The other two detected MAFs are congruent with the proposed conceptualization of case contextualization MAFs. These represent consumer activists’ efforts to contextualize and reflect on the broad implications of the Carrefour incident. MAF 4 was labeled as case broadening. It is characterized by substantial contextual expansion of the Carrefour incident via personal reflections and commentary related to broader political and social issues in Brazil. For example, some tweets alluded to the contradictory reality perception that it might be safer to buy groceries at small stores with no security guards than in larger stores because “Carrefour’s security guard kills a dog [while at Extra] security guards kill people.” MAF 5 was named as case comparison because it is characterized by tweets making explicit comparisons between two or more instances of ethical and moral brand transgressions. The predominant comparison is between the dog’s death at Carrefour and the murder of a young Black man at Extra. A collection of tweets questioned the perceived asymmetry of media coverage and public commotion between both cases, for example, that “the death of the dog at Carrefour reverberated more than the death of a strangled black boy at Extra.”

Longitudinal variability in MAFs

The relative volume of different MAFs is expected to change over time, and RQ3 (a,b,c) focuses on examining these longitudinal patterns. As proposed, it is plausible that brand-oriented posts become less prevalent over time and that cause-oriented posts remain relatively constant. Two approaches were jointly employed to shed light on this. First, we computed the average probability of MAF membership for tweets posted on the same 24-hour period, indicating daily changes in the proportion of MAFs. Locally weighted least squares regressions were used to facilitate the identification of fluctuation patterns (see Fig. 6A). Additionally, we replicated the analysis by assigning tweets to the MAF with the highest estimated membership probability, a mutually exclusive classification (see Fig. 6B). Both approaches yielded similar results, indicating that observed patterns are not an artifact of the classification approach.

Fig. 6
figure 6

Longitudinal patterns of microlevel action frames (MAFs). Note Figure A displays patterns using a probabilistic, non-mutually exclusive classification approach. Figure B shows patterns based on a mutually exclusive classification. N = 5975 original tweets

Full size image

Congruent with our theorization, results indicate that cause orientation has a relatively stable presence during the campaign. In contrast, brand orientation dominates the mobilization in the early stages, but its prevalence rapidly dwindles after 30 days. The pattern of information sharing tweets resembles a U shape, with greater activity timed around two relevant, real-world events: activists’ demonstrations at Carrefour stores, and, almost four months later, the application of fines against Carrefour. Case broadening is virtually non-existent in the first days and gains momentum after the first month. Case comparison more strongly appears around the days of the killing of a Black man at Extra.

MAFs amplification through content curation

RQ4a asks whether distinct MAFs are differently curated and amplified through likes and shares (or retweets). It also considers whether brand-oriented posts are more strongly amplified than other topics due to a potential prevalence of negative sentiment against the brand.

Table 3 presents the negative binomial regression models predicting the number of likes and retweets. Deviance tests for likes (χ2(6) = 871.9, p < 0.01) and retweets (χ2(6) = 386.9, p < 0.01) show that both models are significantly better than their respective null models. Overall, both models estimate that case comparison elicited significantly higher amplification than brand orientation. In contrast, tweets classified as information sharing, on average, were related to lower amplification relative to brand orientation. Case broadening generates more likes but is retweeted at the same rate as brand orientation. Cause orientation is estimated to yield lower amplification than brand orientation, but that is only significant for retweets. For instance, when the user has an average number of followers and friends, brand orientation tweets are estimated to be retweeted 2.1 times, and cause orientation tweets are estimated to be retweeted 58 percent less.

Table 3 Negative binomial regression models predicting MAF curation and amplification via likes and retweets
Full size table

Additionally, we investigated whether MAFs with a more negative sentiment language were more likely to be amplified (RQ4b). Sentiment analysis results (see Fig. 7) show that case comparison was the most negative MAF, but not significantly different from brand orientation, which was the second most negative MAF. Cause orientation was less negative than brand orientation in our sample, followed by case broadening. Information sharing was significantly less negative than the first three MAFs. Thus, it seems that strong negative sentiment is distinctively presented across MAFs and tends to be more prevalent among MAFs more actively liked and shared.

Fig. 7
figure 7

Sentiment analysis. Note Vertical black lines indicate Bonferroni-adjusted 95% confidence intervals

Full size image

Discussion

In this exploratory study, we employed a computational approach to analyze a large volume of social media user-generated content. This created the opportunity to examine actual communicative behavior and the production of MAFs during an online collective action with high ecological validity (Margolin 2019). We identified five MAFs manifested by Twitter users during an online consumer activism campaign against Carrefour and examined the role of brand and cause in thematically orienting social media peer-content production and curation over time.

Consumer activism is increasingly reliant on social media platforms to gain visibility and attract media attention (Bennett and Segerberg 2013), which are imperative for anti-brand activism campaign success (Friedman 1991, 2002). The campaign against Carrefour was successful across several consumer activism parameters (Heldman 2017). The campaign effectively held the firm accountable, promoted public discussion around the cause, and influenced public policy, legislation, and governmental accountability. Although our analyses cannot establish a causal link between online and offline actions, it seems that in the initial stages of the campaign (Fig. 5), online activism may have contributed to the organization and promotion of store protests and that during the latter stages of the campaign, social media were instrumental in keeping the case in the spotlight.

Our analysis suggests that in the early moments, consumer activism may benefit from a brand-oriented MAF, which seems to elicit an upsurge in participation and engagement from activists adopting a personalized, negative expressive action toward the brand. Brand orientation was constructed around a brand vilification narrative that dominated the first few weeks of the campaign, with direct attacks (replies) to the brand, in which activists compared Carrefour to “garbage” and claimed to “never buy anything from” this brand ever again. This shows that online consumer activists not only want to express their negative feelings about the brand (Hoffmann and Müller 2009; Klein et al. 2004; Makarem and Jae 2016), they also want to talk to and lash out at the brand. This initial upsurge in anger-expressive content bears some similarities with the longitudinal patterns of online citizen-activism posting observed by Legocki et al. (2020). The authors found that posts designed to publicly shame or punish white supremacists prevailed in the first two days of online action. Our findings suggest that negative expressions against a brand might last even longer (for at least two weeks in our data), in the form of brand-oriented MAF.

Additionally, our findings show that negative sentiment is an important mechanism for explaining MAF curation and amplification. MAFs with the most negative sentiment, case comparison and brand orientation, were retweeted at a higher rate than cause-oriented and information sharing MAFs, which were less negative. Also, the two case contextualization MAFs, case comparison and case broadening, were more liked than brand orientation. Negative sentiment could explain the greater number of likes for case comparison—the MAF with the highest negative sentiment score, but not for case broadening. Likes “express affective responses to online messages” (p. 1319) while shares represent a proactive behavior of amplifiying content to one’s network (Alhabash and McAlister 2015). Thus, our results indicate that negative sentiment is an important MAF feature to explain what posts are amplified in the crowd via retweets. A complementary explanation comes from the consumer-brand relationship literature. A brand transgression can result in feelings of betrayal (Reimann et al. 2018) and lead to a rapid brand relationship dissolution, especially when this violation is against a self-relevant cause (Trump 2014). When faced with a brand betrayal due to unethical or irresponsible corporate behavior, consumers may seek out ways to retaliate (Karaosmanoglu et al. 2018; Sweetin et al. 2013) in an attempt to “get even” with the brand (Grégoire and Fisher 2008). One of lowest effort activist-like behavior to punish a brand is through likes and shares (Barry et al. 2022). Therefore, the online promotion of brand orientation and case comparison MAFs are easy ways to retaliate against brands through a low-threshold, feel-good form of activism (Skoric 2012). Less negative (or more neutral) MAFs against the brand are less likely to serve the function of brand retaliation.

Still, if our findings hold across other consumer activism cases – an important test for future research – a brand-oriented action frame is unlikely to provide enough support to sustain a campaign in the long term. We find that posts emphasizing brand orientation rapidly decrease, thus demonstrating that consumer anger is short-lived (Grégoire et al. 2009). The cause orientation seems to provide more stable support to keep the campaign active and users engaged. Cause-oriented MAF had a relatively smaller but constant presence throughout months of campaigning and focused on animal rights and veganism causes. This pattern corroborates NSM theory (Diani and Porta, 2006; Melucci 1995), which proposes that a shared cause is vital to unify political consumers under an overarching collective identity that sustains an anti-brand mobilization. When consumers collectively identify themselves as part of a cause, it “ensures continuity and permanence of the movement over time” (Melucci 1995, p. 49).

We also uncovered the importance of information sharing and case contextualization MAFs. Case-related newsworthy information sharing was concentrated around relevant real-world events and has been detected in prior research on digital vigilantism on Twitter (Legocki et al. 2020) and online boycotting (Makarem and Jae 2016). Thus, information sharing is likely an important MAF across different cases of online consumer activism. Moreover, we identified the role of broader case contextualizations in connecting the Carrefour case to other social issues, such as urban violence and racism. These contextualizations had a negligible presence in the early weeks of the campaign and only flourished once the case gained public awareness through brand, cause, and information-sharing MAFs. This timing of awareness seems logical since it’s only through repeated attention that a subject can become part of the public agenda (Littlejohn et al. 2017). Case contextualization MAFs underscore that consumer activism does not happen in a void and that adjacent cases of brand transgressions and competing issues influence the signification process of what a brand transgression episode means and why it matters.

Practical implications

Our findings show that consumer activists can use their social media accounts to promote public awareness of a cause and a brand transgression through different MAFs that can distinctively influence the longevity of online consumer activism campaigns. We organize the results suggested by our data as the OCAM model, presented in Fig. 8, which aims to help corporations to strategically react and respond to different consumer activists’ MAFs. We proffer this model in the spirit of theory building in this area, as our case study approach is not well suited for generalizable tests of the model.

Fig. 8
figure 8

Online consumer activism microframing (OCAM) model

Full size image

The OCAM model proposes that when participants with a brand orientation enter online mobilization, the narrative can be quickly dominated by individual expressions of anger and direct attacks against brands, amplified through likes and shares on social media. Our data indicate that brand-oriented MAF is likely to dissipate faster than other MAFs, but that does not mean brand managers should disregard it since it may feed case contextualization MAFs with a broader scope and enough power to sustain the mobilization longer. We recommend that, when facing a brand-oriented collective action, brands quickly engage in brand forgiveness strategies and attempt to restore brand-consumer relationship quality to pre-transgression levels (Fetscherin and Sampedro 2019), which are more likely to succeed when perceived as authentic (Guèvremont and Grohmann 2018).

When cause-oriented MAF is detected, it suggests that a brand transgression has tapped into lifestyle politics territory (Bennett 2012), such that the consumer-cause connection becomes more salient. In turn, a cause orientation can nurture a shared collective identity around a common cause that gives steady support for online mobilization. In this context, upon identifying the cause, we recommend that brands publicly demonstrate their commitment to repairing their mistake and supporting the cause. Since corporate pro-cause actions, also known as corporate advocacy, can be met with skepticism or deemed inadequate (Vredenburg et al. 2020), it is vital to partner with experts, such as credible NGOs and policymakers, to develop and promote an action plan with greater perceived legitimacy.

Information sharing seems to be a recurring MAF in online mobilizations (see Legocki et al. 2020; Makarem and Jae 2016), where media coverage can spur flashes of engagement on social media, thereby keeping brands in the spotlight. Our data indicate that information sharing is relatively neutral and factual. Therefore, we recommend that corporations quickly communicate with the press to generate case-related newsworthy information, which can also be a platform to show accountability and present their efforts to the public.

The other two case contextualization MAFs, case broadening and case comparison, have a negligible presence in the early days of online consumer activism and seem to be a byproduct of the three MAFs mentioned earlier. Case contextualizations related to broader political and social issues (case broadening) and other cases of corporate misconduct (case comparison) are likely to only appear once the case has gained enough media attention and public awareness or once the case becomes part of the public repertoire, as seen in Fig. 6. Thus, the timely and effective implementation of prior strategies can help to control the emergence and growth of case contextualization. As the adjacent political and social issues deemed relevant for the recontextualization of activists’ online discourses are likely to differ across case contexts, it is likely that many case contexts will produce case contextualization MAFs (see Freelon et al. 2018; Legocki et al. 2020; Makarem and Jae 2016; Yousaf et al. 2021).

Conclusion

This paper examines the online consumer activism campaign against Carrefour, in Brazil, by analyzing Twitter data. Our findings reveal the complex interplay between cause support and brand relationship expressions to galvanize collective action through different MAFs. Rather than experimental or survey methods, where actual behavioral outcomes are commonly inferred based on proxy measures of behavioral intent or collected in an artificial condition, we used a computational approach to analyze large volumes of social media data and observe the actual communicative behavior as it unfolds in a naturalistic setting (Margolin 2019).

Our focus was in detecting online consumer activists’ MAFs and understanding how they can influence virality on social media. In doing so, our study proposes important new lines of inquiries for consumer activism research that can be explored further via traditional, deductive methods. We encourage surveys or experiments designed to capture individual psychographic information, to investigate if and why people vary in their predisposition to express themselves via a brand, cause orientation, or other MAFs. We have proposed several potential explanations for our findings and future research should be designed to formally test these mechanisms.

Our study is not without limitations. While our findings can be reasonably extended to other brands based on sound logical inferences, our method does not permit tests of generalizability. The observational nature of our data limits our ability to make causal claims. In addition, as much as online consumer activism carries significant weight for brand equity in a digital world, online communicative behaviors may not equate to offline consumer behaviors.