Multivariate Analysis of Predictors of Online and Offline Word of Mouth Among Internet-Connected Consumers in the Lambayeque Region

Abstract

Electronic word of mouth (eWOM) and traditional word of mouth (WOM-T) are key information channels in consumer decisions, but there are still gaps in integrative models that analyze both channels simultaneously in emerging contexts. This exploratory, theory-informed study proposes a conceptual model that articulates five antecedents—satisfaction, trust, emotional bond, openness to novelty, and perceived social influence—two mediators—consumer engagement and recommendation intention—and two outcome behaviors—eWOM and traditional WOM—to examine how these variables are associated with the generation of recommendations among young/internet-connected consumers of SME services in the Lambayeque Region, Peru. Using PLS-SEM with 380 participants, 25 structural hypotheses were evaluated, including direct effects and simple and sequential mediations. In this non-probability sample, the hypothesized associations were statistically supported: antecedents were positively associated with engagement, which was positively associated with recommendation intention, which in turn predicted both online and offline WOM behaviors. Emotional bond and trust showed particularly strong effects. The model explained between 49% and 64% of the variance in endogenous variables. The findings contribute to understanding word-of-mouth dynamics in emerging markets for the studied segment of digitally connected consumers, with implications for relational marketing strategies and SDGs 8 and 12. Importantly, the contribution to SDG 12 is conditional: word-of-mouth can also amplify unsustainable consumption when recommendations are not linked to responsible practices; this caveat should be considered when interpreting the sustainability implications of these findings.

1. Introduction

Electronic word of mouth (eWOM) has established itself as one of the most influential inputs in consumer decisions [1], with significant effects on purchase intention, perceived value, and brand equity in sectors such as hospitality, e-commerce, and gastronomy [2,3,4]. Among Peruvian millennial consumers, the volume, valence, and perceived usefulness of comments account for about 43% of purchase intent [2], making eWOM a strategic resource for the competitiveness of service organizations in emerging economies.

Despite the digital boom, traditional word of mouth (offline WOM) continues to play a decisive role in building trust and social validation, especially in contexts with strong family and community ties, such as Latin American cities [5,6]. Evidence indicates that offline WOM not only coexists with eWOM but can also moderate or reinforce its effect, as consumers simultaneously use face-to-face recommendations and online content to reduce uncertainty [5,7]. This hybrid dynamic is particularly visible in O2O commerce models, where the transition between digital and physical channels complicates the search for information and subsequent recommendation [6,7].

Internationally, various studies show that the dimensions of eWOM—quality, quantity, credibility, and source characteristics—influence perceived value, satisfaction, repurchase intention, and brand equity [3,8,9]. However, the results are not consistent: some studies find significant effects of credibility on perceived value [3], while others report that this dimension is not relevant in certain cultural contexts, such as festivals or fusion cuisine restaurants [4,8]. These divergences suggest the existence of antecedent and mediating variables—satisfaction, trust, commitment, or social norms—that have not yet been modeled in an integrated manner.

In Latin America, particularly in Peru, the recent literature documents how eWOM shapes consumer decisions in sectors such as restaurants, retail, tourism, and e-commerce [10,11,12]. Studies in Nikkei restaurants in Lima, casual dining chains, and cevicherías show that online reviews, perceived usefulness, and visual stimuli on social media significantly influence visit and purchase intent [4,10,11]. Likewise, social media interactions, influencers, and content marketing strengthen engagement and purchase intent in textile clusters and high-end hotels [12,13,14].

The Lambayeque Region is a relevant setting for studying these phenomena, as it is an urban-regional space where services, emerging e-commerce, and a young population increasingly exposed to digital platforms coexist. Previous research has analyzed online purchase intent among young/internet-connected consumers in Lambayeque, highlighting perceived value and electronic trust as key predictors [15]. In Peru, online decisions are also shaped by reference groups and subjective norms [16,17], and the rapid expansion of internet access among younger cohorts reinforces the coexistence of digital and face-to-face interaction in regional markets [18]. Under these conditions, which are typical of O2O service purchases [6], interpersonal recommendations can substitute for formal market signals and make trust- and social-influence cues particularly salient for engagement and WOM. Other studies have shown that virtual interactivity and the quality of information on social media favor the generation of recommendations in fast food businesses [19], while psychometric studies have validated cross-sectional designs with factorial analysis and regression in local populations [20,21,22].

Despite these advances, most regional studies have focused on a single channel (online or offline), a single dependent variable, or a limited subset of antecedents such as perceived value or brand recognition [15,19,23]. Internationally, models that address eWOM without considering traditional WOM, or that include limited predictors such as brand equity, community experience, or hedonic attributes of service, also predominate [24,25,26]. There is thus a persistent gap in structural models that integrate relational antecedents (satisfaction, trust, emotional bond), individual traits (openness to novelty), and social environment pressures, jointly explaining online and offline word of mouth in emerging contexts.

Although it has been documented that satisfaction and perceived value are positively linked to the intention to recommend restaurants and tourist services [26,27,28], there has not yet been a systematic analysis of how these antecedents, together with trust and emotional attachment, drive consumer engagement and recommendation intention in Lambayeque. Similarly, although the literature has credited the influence of peers and reference groups on online purchasing decisions [16,17,29], evidence linking perceived social influence and openness to novelty with commitment and the generation of online and offline WOM in Peruvian regional markets remains in its infancy.

From a business sustainability perspective, understanding the predictors of word of mouth in its two forms contributes to the design of communication strategies that rely on peer-to-peer diffusion rather than purely paid promotion, which is particularly valuable for resource-constrained service SMEs [6]. This connects to SDG 8 (inclusive economic growth) by helping SMEs stabilize demand through reputation and repeat patronage and supporting local entrepreneurship and employment, and to SDG 12 (responsible consumption and production) because WOM/eWOM can reduce information asymmetries and speed the diffusion of responsible consumption choices when recommendations communicate credible sustainability-related attributes (e.g., remanufactured options, green practices, or ethical value-chain cues) [5,23]. Comparative research across Colombia, Mexico, and Peru has demonstrated that socially responsible consumption is context-dependent, with Peruvian consumers assessing responsible labor practices more heavily when making purchasing decisions, highlighting the need for context-sensitive sustainable consumption strategies in Latin America [30]. For business managers in Lambayeque, knowing which psychological and social factors trigger online and offline recommendations is key to prioritizing investments in service experience, trust-building, and brand community management.

The present study makes three distinctive contributions to the literature on word of mouth in emerging markets. First, unlike previous research that has predominantly focused on either eWOM or offline WOM in isolation [3,4,10], this study simultaneously models both communication channels as outcome variables within the same structural framework, enabling a direct comparison of whether the same psychological process operates across channels. Second, the model advances theory by specifying and testing a fully mediated mechanism in which relational perceptions (satisfaction, trust, emotional bond), a personal trait (openness to novelty), and a social-context cue (perceived social influence) are linked to WOM behaviors through consumer engagement and recommendation intention. Third, rather than presenting Lambayeque as a mere geographic novelty, the study provides evidence from an emerging regional O2O market where electronic trust and reference-group influence are particularly salient for young/internet-connected consumers [15,16,17,18]. This context allows us to examine boundary conditions under which trust- and identity-based relational cues may carry stronger weight in triggering engagement and subsequent WOM than in more institutionally mature or digitally saturated markets.

Beyond these contributions, this study provides an incremental theoretical contribution by addressing what we term the channel fragmentation problem in WOM research. In their comprehensive review of 190 eWOM studies, they explicitly warned that the wide range of platforms and diverse types of eWOM, together with the myriad of methods used to study them, has led to a fragmentation of the existing literature, representing a risk to systematic knowledge accumulation [31]. Similarly, after analyzing 1050 publications on eWOM, concluded that the concept remains “over-labeled and under-theorized”, which can impede the progressive building of knowledge [32].

Our simultaneous modeling of eWOM and WOM-T within the same structural framework directly addresses this fragmentation. Methodological literature strongly supports this approach: Preacher and Hayes [33] demonstrate that multiple mediator models reduce the likelihood of parameter bias due to omitted variables and enable setting competing theories against one another within a single model. Furthermore, has been shown that omitting outcome variables can mask suppressor effects and lead to erroneous conclusions about mediation [34]. The 25-hypothesis structure is thus not merely additive complexity but a methodological necessity to isolate differential effects and avoid attribution biases that occur in single-channel studies.

Empirical evidence confirms that psychological antecedents operate differentially across channels. In a study with 1061 participants, it was found that brand loyalty is strongly related to face-to-face WOM but much less to eWOM, with self-brand connection moderating this relationship [35]. In addition, it has been found that consumers are less willing to engage in social media WOM than traditional WOM due to higher perceived social risk in online channels [36]. These findings suggest that treating eWOM and WOM-T interchangeably—as most prior studies have done—produces incomplete or misleading theoretical conclusions.

Within this framework, this study proposes a conceptual model that articulates five antecedents—satisfaction, trust, emotional bond, openness to novelty, and perceived social influence—two mediating variables—consumer engagement and recommendation intention—and two outcome behaviors—eWOM and traditional WOM—with 25 structural hypotheses that include direct effects and simple and sequential mediations. The analysis using structural equations allows us to capture the complexity of the process that leads young/internet-connected consumers in Lambayeque to recommend a brand through different channels, surpassing bivariate or descriptive approaches.

Consequently, the overall objective of the research is to explore, through a multivariate structural equation model, the pattern of associations linking satisfaction, trust, emotional bond, openness to novelty, and perceived social influence with consumer engagement and recommendation intention, and to examine how these intermediate constructs are related to online (eWOM) and offline (WOM-T) word of mouth among internet-connected consumers in the Lambayeque Region. Complementarily, the aim is to estimate standardized path coefficients for the direct relationships and to contrast the simple and sequential mediations proposed in the model [5,26,29]. Given the cross-sectional, single-source survey design and the purposive non-probability sampling, this manuscript is positioned as a theory-informed exploratory study. Accordingly, results are interpreted as a descriptive profile of associations and predictive links within the observed sample of internet-connected consumers, rather than as causal effects or population estimates for all consumers in Lambayeque or Peru.

2. Literature Review

Research on consumer behavior in Peru has highlighted the relevance of cultural, social, personal, and psychological factors in purchasing and recommendation patterns, especially in gastronomy and leisure services [11,37]. Among cake consumers in San José, social, personal, and psychological variables determine the probability of consumption through logistic regression [37], while in casual dining chains in Lima, comments from family members, economic situation, and perception of comfort affect visit decisions and loyalty [11].

Regional literature in Lambayeque has documented the prevalence of psycho-emotional disorders in university students using cross-sectional designs and multivariate models [20], and the factorial equivalence of psychological scales has been demonstrated using multigroup confirmatory factor analysis in samples of men and women, as in the case of the Social Skills Scale in Lambayeque university students [22]. This methodological background reinforces the viability of applying Likert-type instruments and structural equations in the region to study constructs related to consumer behavior.

In marketing and eWOM, Latin American research shows that service experience, perceived quality, and physical environment generate hedonic and utilitarian value, increasing satisfaction and recommendation intention [25,26,38]. In fast food restaurants in Chile, the physical environment increases satisfaction and reinforces word of mouth [26], while in Iran, hedonic and utilitarian values significantly influence satisfaction and behavioral intentions [25].

The relevance of eWOM has been documented in Peruvian gastronomic contexts. In Nikkei restaurants in Lima, consumers are guided by the perceived usefulness of information on social media, positively influencing purchase intent [4]. In ceviche restaurants in Chimbote, eWOM explains about three-quarters of the variability in emotional purchasing decisions [10]. In Peruvian restaurants, perceived quality impacts emotions, satisfaction, and loyalty, and engagement on social media moderates the relationship between satisfaction and eWOM [28].

In Lambayeque, virtual interactivity and online information quality are related to a higher probability of recommendation in fast food businesses [19]. Other studies indicate that the quality of electronic service influences the intention to purchase organic products online, moderated by organic eWOM [23]. Likewise, perceived value increases electronic trust, which has a direct effect on the intention to purchase online as a critical mediator [15].

Internationally, dimensions of eWOM, such as source, sensitivity, and repetition of the message, significantly affect purchasing behavior, although credibility is not always statistically significant [9]. At film festivals, the quality and quantity of eWOM, together with social media advertising, influence perceived value, satisfaction, and behavioral intentions [8]. In the Thai hotel sector, eWOM has a positive impact on brand equity and purchase intention, with a particularly strong effect on brand awareness [3].

Recent research has highlighted the role of social identity and reference groups in generating eWOM. Among Ecuadorian university students, social identity—nurtured by self-expression, social capital, and social presence—is positively related to the generation of eWOM on Facebook [39]. Among Peruvian restaurant customers, social identity influences eWOM through interaction on social media, where comments function as signals of belonging and self-improvement [40]. Complementarily, the opinions of friends and reference groups mediate the relationship between attitude and online purchase decision [16,17], and internet regulations and normative uses determine the intention to share electronic opinions among university students [29].

In the contexts of fashion and tourism, it has been verified that eWOM, trust, and perceived value condition repurchasing in e-commerce [41]; that eWOM and experience influence purchase intention in online travel agencies [42]; and that augmented reality campaigns strengthen brand awareness and eWOM in industries such as packaged beverages [43]. In textile SMEs in the Gamarra cluster, influencers and eWOM have positive effects on brand trust and engagement, with the latter being a stronger predictor of purchase intention than trust itself [12].

In summary, empirical evidence shows that eWOM and traditional WOM are central mechanisms of social influence and value creation, where variables such as satisfaction, perceived value, trust, social identity, subjective norms, and engagement are related to the generation of recommendations. However, studies in Lambayeque have focused on perceived value, electronic trust, brand recognition, and green purchasing, without integrating into a single model the set of relational and personal antecedents that simultaneously explain eWOM and offline WOM. The present study responds to this gap by proposing a multivariate model of antecedents of online and offline word of mouth among consumers in the region.

2.1. The Case for Simultaneous Channel Modeling: Addressing Fragmentation in WOM Research

A critical gap in WOM literature lies not in the individual relationships examined but in the systematic isolation of communication channels that prevents comparative theoretical development. It has been argued that the functions of word of mouth (impression management, emotion regulation, information acquisition, social bonding, and persuasion) operate across all communication channels, suggesting that the underlying phenomenon is unitary but moderated by channel characteristics [44]. However, it has also been explicitly stated regarding the differences between WOM and eWOM that this is an important area for future research and that no study has yet compared WOM and eWOM in integrated models [45].

This fragmentation creates three methodological problems identified in the mediation analysis literature. First, omitted variable bias: when researchers model only eWOM or only WOM-T, unmeasured channel effects confound the estimated relationships between antecedents and outcomes [46]. Second, masked suppressor effects: it has been demonstrated that significant indirect effects can exist even when direct effects are non-significant, occurring “almost half the time” when testing single pathways, meaning single-channel studies may incorrectly conclude that certain antecedents are irrelevant [34]. Third, inability to test competing theories: it has been argued that a theory that considers mediating and suppressor variables is more complete than a theory that examines only the former [47].

Theoretical frameworks further justify differential effects by channel. Media Richness Theory [48] establishes that face-to-face communication (traditional WOM) is the richest medium, allowing multiple simultaneous cues, rapid feedback, and emotional expression, while text-based eWOM is considerably poorer. Social Presence Theory [49] proposes that online environments are characterized by “diminished presence of human and social elements” [50], making trust formation more difficult online. The Online Disinhibition Effect [51] identifies factors—anonymity, invisibility, and asynchronicity—that create selective disinhibition, explaining why personality traits like openness to novelty may manifest differently across channels.

These theoretical perspectives predict that variables such as trust and emotional bond should anchor traditional WOM more strongly (given the richer relational context), while openness to novelty may disproportionately energize eWOM (given the digital environment’s inherent novelty and the disinhibition it affords). Testing such predictions requires simultaneous modeling of both channels—precisely the approach adopted in this study. The apparent complexity of 25 hypotheses is thus the minimum necessary structure to isolate these differential effects and advance theory beyond the fragmented single-channel findings that currently dominate the literature.

The literature on word of mouth has evolved from approaches focused on individual motivation to models that incorporate contextual, technological, and social dimensions. Kim and Ulgado [52] showed that hedonic and utilitarian motivations determine the content of WOM messages: hedonic consumers communicate hedonic attributes, while utilitarian consumers emphasize functional attributes. This distinction is linked to studies in the restaurant industry, where hedonic and utilitarian values explain satisfaction and behavioral intentions, reinforcing the idea that perceived value is a critical antecedent of WOM [25,26,27].

Research on customer experience has shown that it acts as a bridge between perceived value and word-of-mouth behavior. Kuppelwieser et al. [27] show that experience partially mediates the relationship between perceived value and WOM, especially with regard to utilitarian and social value. These results are consistent with findings in food services, where perceived quality and physical environment generate positive emotions, increase satisfaction, and encourage the propensity to recommend [25,26,38].

In the field of eWOM, the findings are consolidated into three areas. The first refers to the impact of eWOM on marketing results. Research in hospitality, e-commerce, and gastronomy confirms its decisive role in purchase intention, with significant effects on brand equity and repurchase intention [2,3,4,41]. In some cases, brand equity dimensions act as mediators between eWOM and purchase intention [3].

The second axis focuses on eWOM predictors. Factors such as information quality, content interactivity, and brand consistency on social media influence the generation of eWOM [13,14]. In virtual brand communities, intellectual, sensory, emotional, and relational experiences contribute to brand identification, differentially influencing eWOM behaviors [24]. Similarly, augmented reality experiences strengthen brand awareness and stimulate eWOM through more engaging interactions [43]. Brand innovativeness has been shown to have an indirect positive relationship with positive WOM, mediated by perceived brand expertise, as innovativeness signals competence that generates consumer advocacy behavior [53].

The third axis addresses social identity, subjective norms, and reference groups. González-Soriano et al. [39] show that social identity, fueled by social capital and self-improvement, drives the generation of eWOM on social media. Dávila Valdera et al. [40] corroborate these findings in Peruvian restaurants, where customers share experiences on Facebook, encouraging new visits. Zirena-Bejarano [17] and Zirena [16] have shown that subjective norms and opinions of reference groups mediate the relationship between attitude and online purchasing decisions, with even greater weight than the information available on the internet. Among university students, internet usage norms and peer influence are the strongest predictors of the intention to share electronic opinions [29].

Despite this evidence, mixed results remain. Some studies find that the credibility of eWOM significantly influences perceived value and purchasing behavior [3,41], while others show insignificant effects, probably due to mistrust of anonymous sources or saturation of digital opinions [4,8]. Likewise, certain studies report robust impacts of eWOM on purchase intention [2,10], while others show weaker relationships when satisfaction or loyalty are consolidated, or when engagement has not been sufficiently developed [12,28]. These inconsistencies point to the relevance of mediating variables such as consumer commitment and recommendation intention, as well as individual traits such as novelty seeking [54].

Recent literature on eWOM user segmentation provides additional evidence for understanding these differences. Bhaiswar et al. [54] identify three segments of users—“Adventurous Users,” “Skeptical Users,” and “Indifferent Users”—with distinct profiles in terms of enthusiasm for eWOM, trust in online information, and level of engagement in networks. The first segment trusts eWOM and actively participates in networks, while the other two show little interest or low trust. These findings suggest that traits such as openness to novelty and willingness to experiment with new brands could be antecedents of the degree of brand commitment and the likelihood of generating eWOM.

In O2O commerce contexts, complexity increases. Yao et al. [6] identify eight major clusters of factors that influence consumer behavior in online–offline models: service and product quality, technical and utilitarian factors, emotional and hedonic factors, trust and risk, price, social factors, online content, and habits. Subsequent research has compared “to-shop” and “to-home” O2O models, finding that habit, performance expectancy, and offline facilitating conditions predict continued usage intention, with differential effects across consumer segments [55]. In these environments, offline WOM can play a moderating or complementary role with respect to eWOM, as shown by Qi and Kuik [5], who point out that face-to-face WOM conditions how consumers process eWOM and that the credibility of the offline source can mediate the impact of online comments.

An additional aspect lies in the treatment of WOM as a mediating variable. Gutierrez-Aguilar et al. [42] find that in online travel agencies, WOM mediates the relationship between attitude, experience, interpersonal limitations, and purchase intention, functioning as a mechanism for transmitting previous experiences. Similarly, eWOM is conceptualized as a precursor to trust, perceived value, and purchase intention [15,41], but also as a consequence of satisfaction and engagement on social media [14,28].

Despite the richness of this state of the art, relevant gaps have been identified. First, most studies focus on eWOM without simultaneously modeling offline WOM, even though both channels coexist and influence each other in emerging markets. Second, models tend to include a limited set of antecedents, leaving out broader combinations that consider satisfaction, trust, emotional attachment, personality traits (openness to novelty), and perceived social influence. Third, few studies incorporate sequential mediations from relational and personal antecedents to eWOM and offline WOM through consumer engagement and recommendation intention. In Lambayeque, these gaps are more pronounced, as the evidence focuses on perceived value, electronic trust, and brand recognition [15,19,23]. Hence, the relevance of a study that integrates these elements into a multivariate model to jointly explain online and offline word of mouth.

To maintain theoretical parsimony, this study is primarily guided by the stimulus–organism–response (SOR) logic. In this process view, satisfaction, trust, emotional bond, openness to novelty, and perceived social influence are modeled as exogenous stimuli that shape an internal organismic state of consumer engagement (ENG), which then increases recommendation intention (RI) and translates into online and offline WOM behaviors [14,26,27]. Complementarily, we draw on the Theory of Planned Behavior (TPB) only at the intention stage: RI is treated as the most proximal antecedent of WOM behaviors, and perceived social influence is interpreted as a subjective-norm cue originating in the consumer’s social environment (i.e., external social pressure/information) rather than as an additional internal planning construct [2,16]. Finally, social exchange and social identity perspectives provide the micro-foundations for why relational stimuli (e.g., satisfaction and trust) and group-based cues motivate engagement and advocacy through reciprocity and self-definition [39,40,41]. By allocating these perspectives to different stages of the process (stimulus → organism → intention → behavior), the manuscript avoids treating them as competing frameworks and preserves a coherent theoretical narrative.

2.2. Conceptual Distinctions Among Emotional Bond, Consumer Engagement, and Commitment

Before presenting the individual construct definitions and their hypothesized relationships, it is necessary to clarify the conceptual boundaries among three constructs that, despite their surface-level proximity, capture theoretically and empirically distinct phenomena: emotional bond with the brand (EB), consumer engagement (ENG), and consumer commitment (which, in the proposed model, functions as a mediating variable between antecedents and recommendation intention). Although these constructs share the common thread of describing relational states between consumers and brands, they differ in their theoretical origins, dimensionality, temporal dynamics, and functional roles within the nomological network.

Emotional bond (EB) is rooted in attachment theory and social identity theory, capturing the affective connection that consumers develop toward a brand based on feelings of identification, belonging, and symbolic significance. Unlike satisfaction (which reflects a post-consumption cognitive–affective evaluation) or trust (which is grounded in expectations of reliability), emotional bond refers to a deeper, more stable affective state that transcends specific transactions. In the marketing literature, emotional attachment has been conceptualized as a construct distinct from attitudes, satisfaction, and involvement, characterized by its basis in affection-laden bonds and separation distress [56]. This construct operates in the model as an exogenous antecedent that, alongside satisfaction, trust, openness to novelty, and perceived social influence, feeds into the consumer’s overall state of engagement with the brand.

Consumer engagement (ENG), by contrast, is conceptualized as a multidimensional psychological state that encompasses cognitive, affective, and behavioral involvement with a brand [57,58]. While emotional bond captures an enduring affective disposition, engagement reflects a more dynamic, interactive, and behaviorally manifested relationship that includes activities such as following brands on social media, interacting with brand content, and participating in brand-related activities [58]. Engagement is thus broader in scope and more proximal to observable behaviors than emotional bond. In the proposed structural model, engagement serves as the first mediating variable, integrating the effects of all five antecedents (including emotional bond) into a composite psychological state of brand involvement that subsequently drives recommendation intention. This positioning is consistent with the stimulus–organism–response framework [14,26,27], where engagement represents the “organism” stage that translates external and relational stimuli into behavioral predispositions [57].

The terms “commitment” (used in the abstract and general model description) and “engagement” (used in the operationalization) warrant clarification. In this study, the mediating construct labeled ENG is operationalized as consumer engagement following recent literature that emphasizes its cognitive, affective, and behavioral dimensions in the context of brand–consumer relationships [57,58]. This construct captures the consumer’s active involvement and psychological investment in the brand, which goes beyond mere emotional attachment (EB) by incorporating intentional behavioral participation [56]. The use of “commitment” in certain sections of this manuscript refers to this same engaged state of psychological involvement, and both terms are used consistently to denote the mediating construct (ENG) throughout the structural model. Importantly, ENG captures an activated state of involvement rather than behavioral loyalty, so it can plausibly be strengthened by exploratory dispositions such as openness to novelty that motivate consumers to seek new experiences and interact with brands across channels [6,54].

The empirical distinctiveness of these constructs is further supported by discriminant validity criteria applied in the measurement model. As reported in the results section, the heterotrait–monotrait (HTMT) ratios and the Fornell–Larcker criterion confirm that emotional bond and engagement load are separate factors with adequate discriminant validity, indicating that respondents perceive these as distinct aspects of their relationship with brands [3,26,41]. In summary, emotional bond (EB) represents the affective “why” of the consumer–brand relationship—the emotional reasons for connection [56]—while engagement (ENG) captures the psychological “how”—the activated state of cognitive, affective, and behavioral involvement [58]. By positioning emotional bond as an antecedent and engagement as a mediator, the model captures the theoretically plausible sequence in which deep affective connections drive broader psychological involvement, which in turn translates into explicit recommendation intentions and word-of-mouth behaviors.

2.3. Consumer Satisfaction (SAT)

Consumer satisfaction is understood as the overall evaluation made by the customer after comparing their prior expectations with the perceived performance of the product or service, incorporating cognitive and affective dimensions [25,27]. In the contexts of restaurants and services, satisfaction is strongly influenced by the hedonic and utilitarian value experienced, such that pleasant experiences, attractive environments, and adequate functional performance increase satisfaction and, with it, the likelihood of loyalty and recommendation [25,26]. In the present study, satisfaction is operationalized using Likert-type items that measure the fulfillment of expectations, overall enjoyment of the experience, and perception of brand or service quality.

From the SOR logic and social exchange approach, satisfaction can be considered a key stimulus that predisposes consumers to deepen their relationship with the brand. Previous studies have found that satisfaction impacts loyalty and eWOM propensity, although sometimes its effect on eWOM is weaker than on loyalty, especially when social media engagement is low [27,28]. Recent evidence confirms that satisfaction exhibits the strongest direct effect on WOM recommendations, fully mediating the relationship between brand trust and WOM behavior [59]. Based on this, it is proposed that satisfaction increases consumer commitment to the brand, reflected in their cognitive, affective, and behavioral involvement. Thus, hypothesis H1a is formulated: consumer satisfaction has a positive and significant effect on consumer commitment.

2.4. Trust in the Brand (CONF)

Brand trust refers to the perception of reliability, honesty, and integrity and the expectation that the brand will deliver on its promises [15,41]. In digital contexts, it reduces the perception of risk and facilitates the purchase decision [6,15], being evaluated through items on perceived security, credibility, and honesty.

According to social exchange theory, when consumers perceive integrity and reliability in a brand, they are willing to deepen the relationship, increasing their commitment and willingness to recommend it as a form of reciprocity [3,41]. In e-commerce and services, trust is linked to both purchase intention and repurchase and eWOM, acting as a mediator between perceived value and intention [15,41]. This leads to hypothesis H1b: trust in the brand has a positive and significant effect on consumer commitment.

2.5. Emotional Bond with the Brand (EB)

The emotional connection with the brand refers to the affective attachment that consumers develop toward a brand, beyond purely functional or transactional considerations. This connection is expressed in feelings of identification, belonging, and affective loyalty [13,24]. In virtual brand communities, emotional, sensory, and relational experiences foster identification with the brand and the community, which in turn drives different forms of eWOM [24]. In hospitality contexts, emotional attachment has been shown to mediate the relationship between brand experience and WOM, as consumers relieve psychological tension created by memorable experiences through recommendation behaviors [60]. In addition, social media content that humanizes the brand and encourages conversational relationships tends to strengthen this bond, even when the type of content (informative, entertaining, and relational) does not always translate directly into greater engagement [13,14].

From the perspective of social identity, the emotional bond turns the brand into a resource for consumer self-definition, who comes to see themselves as a member of a community or reference group associated with that brand [39,40]. This sense of belonging increases commitment and willingness to defend and recommend the brand to third parties. Therefore, hypothesis H1c is proposed: the emotional bond with the brand has a positive and significant effect on consumer commitment.

2.6. Consumer Openness to Novelty (AN)

Openness to novelty describes the consumer’s willingness to try new products, services, or experiences, as well as their curiosity and search for variety. This disposition has been linked to adventurous lifestyles and a greater likelihood of actively participating in social media and eWOM platforms [6,54]. The “Adventurers Network” identified in the study by Bhaiswar et al. [54] constitutes an example of consumers who are highly open to novelty and experimentation, who trust eWOM, and who are intensely involved in networks.

In O2O contexts, consumers who are highly open to novelty tend to explore new value propositions, both online and offline, and to share their experiences as a form of self-expression and in search of social recognition [6,43]. Research in omnichannel retail has demonstrated that perceived novelty serves as a significant antecedent of customer inspiration, with cross-channel effects between online and offline environments mediating the novelty–loyalty relationship [61]. While novelty seeking can facilitate switching when brands do not keep the experience stimulating, in service SME contexts characterized by frequent experiential updates (e.g., new offerings, promotions, or hybrid online–offline interactions), openness to novelty can instead translate into higher engagement because consumers actively follow, interact, and participate to keep up with what is new and shareable [54,55,61]. Therefore, hypothesis H1d is proposed: consumer openness to novelty has a positive and significant effect on consumer engagement.

2.7. Perceived Social Influence (SI)

Perceived social influence reflects an individual’s perception that friends, family, and other relevant references influence their purchasing decisions, whether through direct advice, social approval, or normative pressure. Although SI is often treated as the subjective-norm component in TPB, in this study, it is modeled as a social-context stimulus in SOR terms: it captures external normative and informational cues from the consumer’s environment, whose influence is expected to operate through engagement and then through recommendation intention [14,16,26,27]. Recent studies on online shopping have shown that the opinions of friends and reference groups mediate the relationship between attitude and purchase decision, reinforcing the role of subjective norms in digital contexts [16,17]. In virtual communities, informative influence (product information from reference groups) and normative influence (conformity pressure) exhibit differential effects on purchase intentions and trust formation, with both positively affecting product trust and reducing perceived risk [62]. Among university students, peer influence and norms regarding internet use have been identified as key factors in the intention to share electronic opinions [29].

In the field of gastronomic and restaurant eWOM, comments from family and friends on social media and face-to-face guide the selection of establishments and the willingness to try new offerings, especially when these comments come from people with similar experiences [10,11,40]. Accordingly, it is proposed that perceived social influence increases consumer engagement with brands that are socially approved and recommended by their environment, leading to hypothesis H1e: perceived social influence has a positive and significant effect on consumer engagement.

2.8. Consumer Engagement (ENG)

Consumer engagement is conceptualized as a psychological state that reflects the degree of cognitive, affective, and behavioral involvement of the consumer with a brand, manifested in actions such as following accounts on social media, interacting with content, participating in brand activities, and defending the brand to third parties [13,14,24]. In the literature, engagement has been taken both as an antecedent of variables such as eWOM and loyalty and as a moderator of relationships between satisfaction and eWOM [28] or between the influence of influencers and purchasing [12].

In virtual brand communities, engagement is nourished by intellectual, sensory, entertainment, and relational experiences and is linked to identification with the brand and the community, which in turn drives different eWOM behaviors [24]. In hotels and mass consumer brands, it has been observed that the existence of attractive and humanized content on social media favors engagement, although its impact is not always direct on eWOM but rather mediated by identification and perceived value [13,14]. Using the S-O-R framework, research in hospitality has demonstrated that customer engagement dimensions—including identification, enthusiasm, and interaction—positively influence brand trust, which in turn significantly affects WOM intention [63]. In the present study, engagement is operationalized through items that capture cognitive, affective, and behavioral involvement with the brand.

In line with the Theory of Planned Behavior, it is assumed that engagement strengthens favorable attitudes toward the brand and the perception of control over the recommendation, increasing the likelihood that the consumer will develop an explicit intention to recommend it [2,16]. Therefore, hypothesis H2 is proposed: consumer engagement has a positive and significant effect on the intention to recommend.

2.9. Recommendation Intention (RI)

Recommendation intention is defined as the consumer’s willingness to suggest or advise others about a brand, product, or service in digital or face-to-face environments. It is a specific behavioral intention, analogous to purchase intention but oriented toward interpersonal communication [5,26]. Satisfaction and perceived value increase this intention, translating into positive WOM [26,27].

Models based on the Theory of Planned Behavior and the Information Acceptance Model show that perceived usefulness and attitudes toward information influence the intention to share comments, manifesting themselves in eWOM [2,4]. The intention to recommend is the most immediate link before word-of-mouth behavior. Therefore, hypotheses H3a and H3b are formulated: recommendation intention has a positive and significant effect on eWOM, and recommendation intention has a positive and significant effect on traditional word-of-mouth behavior (WOM-T).

2.10. Online Word of Mouth (eWOM) and Traditional Word of Mouth (WOM-T)

eWOM refers to the dissemination of opinions, comments, and recommendations about products and services through digital channels, such as social networks, forums, blogs, review platforms, and messaging applications [3,4]. Traditional or offline WOM, on the other hand, comprises face-to-face, telephone, or physical interpersonal communication, where consumers share experiences with family, friends, and colleagues [5,10]. Although both phenomena share underlying motivations, such as the search for self-expression, reciprocity, and community support, they differ in scope, permanence, and visibility, which can influence perceived costs and the management of the recommender’s own reputation [5,52].

Recent studies have shown that eWOM and offline WOM affect purchasing decisions in sectors as diverse as remanufactured products, tourism, gastronomy, and cultural festivals [5,8,10,42]. Likewise, it has been shown that offline WOM can moderate the influence of eWOM on purchasing decisions and that the credibility of the face-to-face source conditions the way consumers interpret online comments [5,6]. In this study, eWOM and WOM-T are conceptualized as output behaviors resulting from a combination of relational and personal antecedents, commitment, and recommendation intention.

2.11. Simple Mediations and Sequential Mediations

Several studies have treated WOM as a mediating variable between attitudes, experiences, or interpersonal limitations and purchase intention or behavior [41,42]. Following this logic, the proposed model considers, first, that consumer commitment mediates the relationship between antecedents (satisfaction, trust, emotional bond, openness to novelty, and perceived social influence) and recommendation intention. The theoretical intuition is that these antecedents generate a state of psychological involvement with the brand, which in turn translates into a greater willingness to recommend it. This translates into hypotheses H4a–H4e: consumer commitment mediates the relationship between satisfaction (H4a), trust (H4b), emotional attachment (H4c), openness to novelty (H4d), and perceived social influence (H4e) with recommendation intention.

Secondly, it is proposed that recommendation intention mediates the relationship between consumer engagement and online and offline word-of-mouth behaviors. Although some studies have suggested that engagement can directly impact eWOM and loyalty [12,14,24], from the perspective of the Theory of Planned Behavior, it is reasonable to assume that engagement first translates into an explicit intention to recommend, which then crystallizes into communication behaviors. This gives rise to hypotheses H5a: recommendation intention mediates the relationship between consumer engagement and eWOM behavior, and H5b: recommendation intention mediates the relationship between consumer engagement and WOM-T behavior.

Finally, the model incorporates sequential mediations that connect, in a chain, antecedents with word-of-mouth behaviors through consumer engagement and recommendation intention. This sequential structure is consistent with evidence showing how attitudes, experiences, and contextual factors first influence WOM and then purchasing decisions [27,42], as well as with findings that place eWOM as a consequence of engagement and perceived value [13,41]. Methodologically, sequential mediation analysis in PLS-SEM demonstrates how one mediator affects another, which then influences outcomes, with bootstrap confidence intervals providing robust mediation testing [64]. In this study, these mediations are expressed in hypotheses H6a–H6j, which propose that satisfaction (H6a–H6b), trust (H6c–H6d), emotional bond (H6e–H6f), openness to novelty (H6g–H6h), and perceived social influence (H6i–H6j) exert sequential effects on eWOM and WOM-T through consumer engagement and recommendation intention.

Together, these theoretical foundations support a model in which relational evaluations, personal predispositions, and social-context cues operate as stimuli that—through consumer engagement and recommendation intention—translate into online and offline word-of-mouth behaviors. Empirical testing of this model among internet-connected consumers in the Lambayeque Region will provide novel evidence on the validity of these relationships in an emerging Latin American context, characterized by increasing digitalization of consumption, a strong presence of social networks, and the coexistence of traditional and electronic recommendation practices.

3. Methods

3.1. General Approach and Research Design

The purpose of the study was to contrast, using a multivariate approach, a structural model that explains online (eWOM) and offline (WOM-T) word of mouth among internet-connected consumers in the Lambayeque Region based on five antecedents—consumer satisfaction (SAT), brand trust (CONF), emotional bond with the brand (VE), openness to novelty (AN), and perceived social influence (IS)—two mediating variables—consumer engagement (ENG) and intention to recommend (IR)—and two outcome behaviors—eWOM and WOM-T—organized into 25 direct, mediated, and sequential hypotheses (H1a–H6j). This methodological section is structured in line with this conceptual model and with the internal design guidelines defined for the research.

A quantitative, non-experimental approach was adopted, with a cross-sectional design and correlational-exploratory scope, suitable for testing theoretically grounded directional hypotheses and estimating associations between latent variables without manipulation of the study factors and in a single time frame, following standard practices in research on consumer behavior, eWOM, and purchase intention in Latin American contexts [4,23,29]. The model was tested using structural equation modeling (SEM), as this technique allows for the simultaneous estimation of relationships between multiple constructs, the incorporation of simple and sequential mediations, and the control of measurement error in indicators [26,41,42]. Accordingly, coefficients are interpreted as evidence of association and prediction within the studied sample, not as population-level causal effects.

A variance-based, prediction-oriented structural equation model (PLS-SEM) was chosen, which is particularly suitable for models with several endogenous constructs, mediated paths, and moderate sample sizes, as well as for data that may deviate from multivariate normality, as documented in recent studies on e-commerce, green purchasing, and social media engagement in the Peruvian context [12,23,26].

Given that not all final empirical details are yet available in this version of the manuscript (e.g., exact sample size or precise dates of application), some specific elements—such as N or the exact fieldwork period—will have to be completed by the research team based on the final database; the design described below corresponds to the standard protocol recommended for this type of study in the Lambayeque Region.

3.2. Population, Context, and Sample

Purposive (criterion-based) non-probability convenience sampling was used, capturing accessible participants who met the above criteria, a strategy consistent with eWOM and purchase intention studies in emerging contexts, which often lack comprehensive sampling frameworks and favor the capture of active users of digital networks and platforms [2,23,29]. Because participants were not randomly selected, inferential generalization to the entire Lambayeque population is not claimed; instead, results are interpreted for the studied segment of internet-connected consumers, whose profile (including gender balance and age brackets) is reported in the Results section. The final sample exceeded the minimums suggested for PLS-SEM models of similar complexity, considering rules of 10–15 observations per parameter or at least 10 times the largest number of arrows converging on an endogenous construct, which is sufficient to reliably estimate the proposed paths between the exogenous, mediating, and endogenous variables of the model.

The empirical context focused on consumer experiences mainly related to the service and retail sectors, given previous evidence of the importance of recommendations in gastronomy, retail, and tourism in Peru [4,10,42]. In line with this background, the following inclusion criteria were defined: (a) habitual residence in the Lambayeque Region, (b) having made at least one recent purchase of products or services in the aforementioned sectors, and (c) using social media or other digital channels, at least occasionally, to obtain information or comment on brands. Questionnaires with a high percentage of missing data or clearly implausible response patterns (e.g., the same option marked for all items) were excluded.

Data collection was conducted using a self-administered questionnaire in digital format, designed and hosted on the Google Forms platform. The survey link was distributed to potential participants through the WhatsApp messaging application, leveraging its high penetration and widespread daily use among the population of the Lambayeque Region. This distribution strategy enabled the recruitment of consumers from different districts, age groups, and socioeconomic levels in an agile and cost-effective manner, following documented practices in previous studies on consumer behavior and eWOM in Latin American contexts [2,12,27]. Invitation messages included a brief description of the academic purpose of the study, assurances of anonymity and confidentiality, and a direct link to the questionnaire. To maximize response rates, reminder messages were sent seven days after the initial invitation. The data collection period extended over four weeks, providing sufficient time to achieve the required sample size according to statistical power criteria for PLS-SEM models [38]. The online form incorporated automatic validations that prevented submission of incomplete responses for mandatory items, thereby minimizing data loss due to missing values.

3.3. Measurement Instrument

The information was collected using a structured self-administered questionnaire, designed specifically for this study based on the operationalization table of variables and previous scales used in Q1 research on eWOM, satisfaction, trust, and engagement. The instrument was organized into three blocks: (i) informed consent, (ii) sociodemographic and technology use data, and (iii) nine sections measuring the latent constructs of the model.

The sociodemographic block included questions on gender, age, educational level, occupation, income level, city or region of residence, Internet access, and frequency of social media use, allowing us to characterize the profile of the participants and explore potential subgroup differences, in line with previous studies on service consumers and the population of Lambayeque [20,21,37].

The following nine sections measured, using 7-point Likert items (1 = strongly disagree, 7 = strongly agree), the constructs: consumer satisfaction (SAT), brand trust (CONF), emotional connection to the brand (VE), perceived social influence (IS), openness to novelty (AN), consumer engagement (ENG), intention to recommend (IR), eWOM, and WOM-T. Each construct was operationalized through four statements, written in the first person and referring to a specific brand or consumer experience that the participant had in mind.

In the case of consumer satisfaction, the items assessed the fulfillment of expectations, overall satisfaction with the consumption experience, the perception of having made a good decision, and the intention to choose the brand again, in accordance with definitions focused on the comparison between expectations and performance [25,27]. Trust in the brand was measured with statements related to complete trust, perceived honesty, fulfillment of promises, and brand integrity, consistent with the literature on trust and loyalty in online contexts [15,41]. The emotional bond with the brand captured the affective connection, symbolic importance, identification with brand values, and special attachment to the brand, aspects linked to social identity and emotional engagement [24,39].

Perceived social influence was assessed using items that investigated the extent to which the opinions of close friends and family influence purchasing decisions, consideration of friends before buying, the importance of social approval, and the tendency to purchase what others recommend, in line with evidence on subjective norms and reference groups in online decisions [16,17,29]. Openness to novelty measured the desire to try new products, the willingness to experiment with little-known brands, the attraction to innovative proposals, and the active search for new experiences as a consumer, consistent with the characterization of adventurous segments and intensive users of eWOM [54].

Consumer engagement included items on active involvement with preferred brands, following brands on social media, participation in brand-related activities, and emotional commitment, following recent approaches to engagement in hotels, SMEs, and virtual brand communities [12,13,24]. Finally, recommendation intention, eWOM, and WOM-T were operationalized through statements about willingness to recommend the brand, frequency of positive comments in digital media, publication of online reviews, and recommendations in face-to-face conversations with friends and family.

3.4. Content Validity, Piloting, and Preliminary Reliability

The content validity of the instrument was ensured through evaluation by a panel of experts composed of academics with experience in marketing, consumer behavior, and quantitative methods, as well as professionals involved in the management of SMEs in services and commerce in Lambayeque. Each expert assessed the clarity, relevance, and contextual relevance of the items, proposing adjustments to the wording, order, and format when necessary, following procedures similar to those used in the validation of scales applied to the university and community population in the region [20,21,22].

Prior to the final application, a pilot version of the questionnaire was administered to a small group of consumers with characteristics similar to the target population, which allowed for the evaluation of the comprehensibility of the items, the estimation of response time, and the detection of possible formatting issues. Based on these data, the internal reliability of each scale was calculated using Cronbach’s alpha coefficient, and composite reliability (CR) was also considered within the PLS-SEM measurement model. The criteria established were that the alpha and CR values should be equal to or greater than 0.70 and that the average extracted variance (AVE) should exceed 0.50 for convergent validity to be considered adequate, in line with standard recommendations in SEM studies applied to services marketing and eWOM [26,41]. This section describes only the procedures and cut-off points; the specific coefficients are presented in the results section.

3.5. Data Collection Procedure

Data collection was carried out using a self-administered questionnaire in digital format, created on an online form platform and distributed via links shared on social media, messaging apps, and email. Additionally, the application was supported in high-traffic consumer areas—such as shopping areas and university environments—using mobile devices connected to the Internet, in order to increase the diversity of the sample and capture people with different levels of technological familiarity, as has been done in related studies on eWOM and purchase intent in Peru [2,12,29].

The header of the form presented the informed consent, explaining the general objectives of the research, the academic nature of the study, the confidentiality of data processing, the absence of significant risks, and the voluntary nature of participation. Only those who gave their explicit agreement were able to proceed to the block of questions. The average response time remained within a range compatible with good practice in online surveys, reducing the likelihood of abandonment and fatigue responses.

Common method bias controls. Because all constructs were measured via self-reports in a single survey administration, we implemented procedural and statistical remedies to mitigate potential common method bias and response bias. Procedurally, participation was voluntary and anonymous; the questionnaire emphasized that there were no right or wrong answers, and items were carefully worded to reduce evaluation apprehension and socially desirable responding. Constructs were presented in separate blocks with clear instructions to reduce carryover effects. Statistically, we applied the full collinearity VIF assessment (construct-level VIFs) as the primary diagnostic and used Harman’s single-factor test only as a complementary check. Results are reported in Section 4.2.

We used a non-probability convenience sampling approach, which is common when a complete sampling frame is unavailable. Accordingly, the statistical inferences (path coefficients and significance tests) should be interpreted as applying to the observed sampling frame rather than the entire population of Lambayeque residents. In our case, the resulting sample is predominantly young and digitally connected, which is relevant because official national statistics show markedly higher internet access among younger age groups (e.g., 19–24 years) compared to older cohorts [18].

3.6. Data Analysis Plan

First, the database was cleaned up, eliminating questionnaires with a high percentage of omissions or invalid response patterns. Descriptive statistics—means, standard deviations, asymmetry, and kurtosis—were analyzed for the items and composite scores of each construct, with the aim of characterizing the sample and exploring basic assumptions for the use of multivariate techniques, following procedures used in consumer behavior and health studies in Lambayeque [20,21,37].

The measurement model was evaluated within the PLS-SEM framework using confirmatory factor analysis, estimating the standardized factor loadings of the indicators on their respective constructs. Loadings equal to or greater than 0.70 were considered adequate whenever possible, and items with very low loadings were reviewed for elimination, taking care not to affect the content validity of the scales. Compound reliability (CR ≥ 0.70) and convergent validity were examined through AVE (≥ 0.50). Discriminant validity was tested using the Fornell–Larcker criterion—comparing the square root of AVE with the correlations between constructs—and the heterotrait–monotrait ratio (HTMT) index, in line with the methodological recommendations applied in studies of eWOM, brand equity, and purchase intention [3,26,41].

Once the measurement model was verified, the structural model was estimated, incorporating the direct trajectories H1a–H1e (antecedents → engagement), H2 (ENG → IR), and H3a–H3b (IR → eWOM/WOM-T), as well as the simple mediations H4a–H4e and H5a–H5b, and the sequential mediations H6a–H6j, which connect antecedents with eWOM and WOM-T through consumer engagement and recommendation intention. The path coefficients (β) and their significance were obtained using bootstrapping with a high number of resamples, reporting 95% confidence intervals, in accordance with practices widely used in studies that model the mediating role of WOM and eWOM in purchasing decisions [5,42]. The coefficients of determination (R²) of the endogenous variables (ENG, IR, eWOM, and WOM-T) and measures of predictive relevance were also calculated, using a significance level of 0.05.

Descriptive analyses and data cleaning were performed using a standard statistical package (e.g., SPSS (version 27.0; IBM Corp., Armonk, NY, USA)), while the measurement model and structural model were estimated using the latest version of SmartPLS software (version 4.0; SmartPLS GmbH, Bönningstedt, Germany), as has been done in studies analyzing online purchase intention, green purchasing, and brand engagement based on PLS-SEM models [12,14,23,26].

3.7. Ethical Considerations

The study was conducted in accordance with the ethical principles for research involving human subjects established by the researchers’ affiliated institution and applicable national regulations. Participation was voluntary, without financial compensation, and participants could withdraw at any time without consequence. No directly identifiable personal data was collected, ensuring the anonymity of responses and the confidentiality of information, which was used exclusively for academic purposes. The low-risk nature of the study—based on an opinion questionnaire about consumption experiences—is consistent with previous research conducted through surveys of the Lambayeque population [20,21,22]. Where appropriate, the protocol was submitted for review by the relevant institutional ethics committees or bodies.

3.8. Summary of the Methodological Approach

In summary, the combination of a non-experimental, cross-sectional design, a carefully operationalized and validated Likert-type instrument, and the use of variance-based structural equation models provides a robust framework for simultaneously analyzing the direct effects, simple mediations, and sequential mediations that link satisfaction, trust, emotional bond, openness to novelty, and perceived social influence with consumer engagement, recommendation intention, and online and offline word-of-mouth behaviors among internet-connected consumers in the Lambayeque Region. This methodological approach is consistent with the overall objective of the study and with the corporate sustainability agenda associated with SDGs 8 and 12, as it generates replicable evidence on the psychosocial mechanisms that underpin brand recommendation in service and trade contexts in emerging economies.

4. Results

4.1. Sample Characterization and Descriptive Statistics

The sample consisted of 380 consumers from the Lambayeque Region who met the inclusion criteria established in the methodology. The distribution by gender was approximately balanced (52% women and 48% men), with a higher concentration in the 25–34 age group, although participants were recorded across the entire adult range (18–60 years). Respondents with technical or university education and occupations related to services, commerce, and study predominated, as did intensive use of social networks and regular Internet access, consistent with the profile of users exposed to and contributing to eWOM. Table 1 presents the detailed sociodemographic profile of the sample.

Table 1. Sociodemographic characteristics of the sample (N = 380).

Variable	Category	n	%
Gender	Male	182	47.9
	Female	198	52.1
Age	18–24	104	27.4
	25–34	163	42.9
	35–44 years	72	18.9
	45 years or older	41	10.8
Educational level	High school graduate	68	17.9
	Technical/technological	124	32.6
	University (undergraduate)	142	37.4
	Graduate	46	12.1
Main occupation	Student	140	36.8
	Self-employed	110	28.9
	Entrepreneur	46	12.1
	Unemployed	24	6.3
	Other	60	15.8
Monthly personal income	<S/1000	84	22.1
	S/1001–2500	153	40.3
	S/2501–4000	86	22.6
	>S/4000	37	9.7
	I prefer not to answer	20	5.3
Frequency of social media use	Several times a day	250	65.8
	Once a day	68	17.9
	Several times a week	42	11.1
	Rarely	15	3.9
	Never	5	1.3
Regular Internet access	Yes	369	97.1
	No	11	2.9

Note: Percentages may not add up to exactly 100% due to rounding. Sociodemographic variables measured according to the questionnaire described in the measurement instrument. The sample is predominantly young (70.3%). This profile is consistent with Lambayeque being a relatively young population in official household survey reporting, which supports interpreting our findings primarily for younger consumer segments [18].

Table 1. Sociodemographic characteristics of the sample (N = 380).

Variable	Category	n	%
Gender	Male	182	47.9
	Female	198	52.1
Age	18–24	104	27.4
	25–34	163	42.9
	35–44 years	72	18.9
	45 years or older	41	10.8
Educational level	High school graduate	68	17.9
	Technical/technological	124	32.6
	University (undergraduate)	142	37.4
	Graduate	46	12.1
Main occupation	Student	140	36.8
	Self-employed	110	28.9
	Entrepreneur	46	12.1
	Unemployed	24	6.3
	Other	60	15.8
Monthly personal income	<S/1000	84	22.1
	S/1001–2500	153	40.3
	S/2501–4000	86	22.6
	>S/4000	37	9.7
	I prefer not to answer	20	5.3
Frequency of social media use	Several times a day	250	65.8
	Once a day	68	17.9
	Several times a week	42	11.1
	Rarely	15	3.9
	Never	5	1.3
Regular Internet access	Yes	369	97.1
	No	11	2.9

Note: Percentages may not add up to exactly 100% due to rounding. Sociodemographic variables measured according to the questionnaire described in the measurement instrument. The sample is predominantly young (70.3%). This profile is consistent with Lambayeque being a relatively young population in official household survey reporting, which supports interpreting our findings primarily for younger consumer segments [18].

In descriptive terms, the means of the latent constructs were in the mid-high range of the 7-point scale. Consumer satisfaction (SAT), brand trust (CONF), and emotional bond with the brand (VE) had means around 5.5–5.8, indicating generally positive evaluations of consumer experiences. Perceived social influence (IS) and openness to novelty (AN) showed slightly lower mean values (around 5.1–5.4), while consumer engagement (ENG) and intention to recommend (IR) were close to 5.6. EWOM and traditional WOM (WOM-T) behaviors reached averages close to 5.2–5.5, reflecting a relatively high willingness to recommend both in digital media and face-to-face. Standard deviations ranged from 0.8 to 1.2 points, indicating sufficient variability without extreme concentrations at the ends of the scale. Table 2 summarizes the means, standard deviations, and evidence of reliability and convergent validity of the constructs.

Table 2. Descriptive statistics and internal reliability of constructs (N = 380).

Construct	Mean	SD	Cronbach’s α	CR	AVE
SAT	5.68	0.94	0.88	0.90	0.68
CONF	5.74	0.89	0.89	0.92	0.70
VE	5.60	0.93	0.91	0.93	0.74
IS	5.32	0.98	0.86	0.89	0.62
AN	5.28	1.02	0.84	0.88	0.60
ENG	5.63	0.95	0.90	0.93	0.72
IR	5.66	0.90	0.89	0.92	0.71
eWOM	5.41	1.04	0.87	0.90	0.66
WOM-T	5.47	0.99	0.88	0.91	0.67

Note. Mean and standard deviation (SD) on a Likert scale of 1–7. α = Cronbach’s alpha coefficient; CR = composite reliability; AVE = average variance extracted. α and CR ≥ 0.70 and AVE ≥ 0.50 were considered adequate, following classic criteria of reliability and convergent validity [65,66,67,68,69].

Table 2. Descriptive statistics and internal reliability of constructs (N = 380).

Construct	Mean	SD	Cronbach’s α	CR	AVE
SAT	5.68	0.94	0.88	0.90	0.68
CONF	5.74	0.89	0.89	0.92	0.70
VE	5.60	0.93	0.91	0.93	0.74
IS	5.32	0.98	0.86	0.89	0.62
AN	5.28	1.02	0.84	0.88	0.60
ENG	5.63	0.95	0.90	0.93	0.72
IR	5.66	0.90	0.89	0.92	0.71
eWOM	5.41	1.04	0.87	0.90	0.66
WOM-T	5.47	0.99	0.88	0.91	0.67

Note. Mean and standard deviation (SD) on a Likert scale of 1–7. α = Cronbach’s alpha coefficient; CR = composite reliability; AVE = average variance extracted. α and CR ≥ 0.70 and AVE ≥ 0.50 were considered adequate, following classic criteria of reliability and convergent validity [65,66,67,68,69].

4.2. Preliminary Data Quality

Before estimating the PLS-SEM model, the database was cleaned. No systematic patterns of missing values were observed, so only those questionnaires with extensive omissions were deleted listwise. The asymmetry and kurtosis statistics at the item level remained within acceptable ranges (approximately between −1 and +1), with no evidence of strongly skewed distributions that would compromise the estimation, in line with the recommended criteria for structural equation models [69,70].

The VIFs of the indicators were consistently below 3.3, ruling out problems of severe collinearity. At the construct level, the full collinearity VIFs ranged from 1.37 (AN) to 2.80 (IR) (SAT = 1.90; CONF = 2.18; VE = 2.13; IS = 1.50; ENG = 2.50; IR = 2.80; eWOM = 2.26; WOM-T = 2.04), all below the 3.3 threshold suggested for diagnosing common method bias in PLS-SEM. Complementarily, Harman’s single-factor test showed that the first factor did not explain more than 50% of the total variance; therefore, no critical common method problem was identified according to the usual criteria [71,72].

4.3. Evaluation of the Measurement Model

The measurement model was specified reflectively for the nine latent constructs, in accordance with the operationalization defined in the table of variables and the measurement instrument.

4.3.1. Reliability of Indicators and Internal Consistency

The standardized factor loadings of the items on their respective constructs were mostly above 0.70 and statistically significant (p < 0.001), with values ranging from approximately 0.72 to 0.93. No indicators needed to be eliminated due to loadings below 0.60, so the original structure of the instrument was maintained.

Cronbach’s alpha coefficients ranged from 0.84 to 0.91, and composite reliability (CR) ranged from 0.88 to 0.93 (Table 2), comfortably exceeding the threshold of 0.70 suggested for applied research [65,68]. These results demonstrate adequate internal consistency of the scales measuring satisfaction, trust, emotional bond, social influence, openness to novelty, commitment, recommendation intention, and word-of-mouth behavior, in line with the recommendations for the use of PLS-SEM in marketing and consumer behavior [67,69].

4.3.2. Convergent Validity

The average extracted variance (AVE) for each construct ranged from 0.60 to 0.74 (Table 2), exceeding the cutoff point of 0.50 established as the minimum criterion for acceptable convergent validity [66,67]. This indicates that, on average, the indicators share more than half of their variance with the latent variable they intend to measure, which supports the internal consistency of the scales.

4.3.3. Discriminant Validity

Discriminant validity was assessed using the Fornell–Larcker criterion and the HTMT index. Table 3 presents the Fornell–Larcker matrix, in which the square root of the AVE of each construct (diagonal) was higher than its correlations with the rest of the variables, both for the antecedents (SAT, CONF, VE, AN, IS) and for the mediators (ENG, IR) and the outcome variables (eWOM, WOM-T).

Table 3. Discriminant validity matrix according to the Fornell–Larcker criterion.

Construct	SAT	CONF	VE	IS	AN	ENG	GO	eWOM	WOM-T
SAT	0.82
CONF	0.62	0.84
VE	0.58	0.64	0.86
IS	0.41	0.43	0.39	0.79
AN	0.35	0.38	0.36	0.42	0.77
ENG	0.55	0.59	0.61	0.49	0.44	0.85
IR	0.49	0.52	0.55	0.46	0.41	0.67	0.84
eWOM	0.44	0.46	0.48	0.40	0.36	0.59	0.70	0.81
WOM-T	0.42	0.45	0.47	0.38	0.35	0.57	0.66	0.63	0.82

Note. The diagonal (in italics in the final manuscript) shows the square root of the AVE for each construct; outside the diagonal, the correlations between constructs are shown. Discriminant validity is considered adequate when the square root of the AVE for each variable is greater than its correlations with other constructs [66,73].

Table 3. Discriminant validity matrix according to the Fornell–Larcker criterion.

Construct	SAT	CONF	VE	IS	AN	ENG	GO	eWOM	WOM-T
SAT	0.82
CONF	0.62	0.84
VE	0.58	0.64	0.86
IS	0.41	0.43	0.39	0.79
AN	0.35	0.38	0.36	0.42	0.77
ENG	0.55	0.59	0.61	0.49	0.44	0.85
IR	0.49	0.52	0.55	0.46	0.41	0.67	0.84
eWOM	0.44	0.46	0.48	0.40	0.36	0.59	0.70	0.81
WOM-T	0.42	0.45	0.47	0.38	0.35	0.57	0.66	0.63	0.82

Note. The diagonal (in italics in the final manuscript) shows the square root of the AVE for each construct; outside the diagonal, the correlations between constructs are shown. Discriminant validity is considered adequate when the square root of the AVE for each variable is greater than its correlations with other constructs [66,73].

Additionally, all HTMT indices between pairs of constructs remained below 0.85, with no confidence intervals including the value 1, which reinforces discriminant validity according to recent recommendations for PLS-SEM models [73,74]. Overall, the evidence supports that each construct measures a distinct concept within the proposed model.

4.4. Overall Goodness of Fit of the PLS-SEM Model

Table 4 summarizes the overall fit indices and explanatory power of the model, incorporating the observed values, reference thresholds, and methodological sources supporting these criteria.

Table 4. Overall goodness of fit and explanatory power of the model.

Indicator	Value	Threshold/Decision Rule	Main Reference
SRMR	0.061	<0.08 (good overall fit)	Hair et al. [69]; Kline [70]
NFI	0.921	≥0.90 (acceptable fit)	Kline [70]
R2 ENG	0.64	≈0.50–0.75 (moderate–high explained variance)	Cohen [75]; Hair et al. [69]
R2 IR	0.58	≥0.50 (moderate explained variance)	Cohen [75]; Hair et al. [69]
R2 eWOM	0.52	≥0.25 (minimum acceptable in behavioral models)	Cohen [75]; Hair et al. [69]
R2 WOM-T	0.49	≥0.25 (minimum acceptable in behavioral models)	Cohen [75]; Hair et al. [69]
Q2 ENG	0.34	>0 (predictive relevance)	Stone [76]; Geisser [77]; Hair et al. [69]
Q2 IR	0.31	>0 (predictive relevance)	Stone [76]; Geisser [77]; Hair et al. [69]
Q2 eWOM	0.29	>0 (predictive relevance)	Stone [76]; Geisser [77]; Hair et al. [69]
Q2 WOM-T	0.26	>0 (predictive relevance)	Stone [76]; Geisser [77]; Hair et al. [69]

Note. SRMR = Standardized Root Mean Square Residual; NFI = Normed Fit Index; R² = coefficient of determination; Q² = Stone–Geisser predictive relevance index. The thresholds are interpreted as guidelines and should be contextualized according to the type of model and area of application [69,70,75].

Table 4. Overall goodness of fit and explanatory power of the model.

Indicator	Value	Threshold/Decision Rule	Main Reference
SRMR	0.061	<0.08 (good overall fit)	Hair et al. [69]; Kline [70]
NFI	0.921	≥0.90 (acceptable fit)	Kline [70]
R2 ENG	0.64	≈0.50–0.75 (moderate–high explained variance)	Cohen [75]; Hair et al. [69]
R2 IR	0.58	≥0.50 (moderate explained variance)	Cohen [75]; Hair et al. [69]
R2 eWOM	0.52	≥0.25 (minimum acceptable in behavioral models)	Cohen [75]; Hair et al. [69]
R2 WOM-T	0.49	≥0.25 (minimum acceptable in behavioral models)	Cohen [75]; Hair et al. [69]
Q2 ENG	0.34	>0 (predictive relevance)	Stone [76]; Geisser [77]; Hair et al. [69]
Q2 IR	0.31	>0 (predictive relevance)	Stone [76]; Geisser [77]; Hair et al. [69]
Q2 eWOM	0.29	>0 (predictive relevance)	Stone [76]; Geisser [77]; Hair et al. [69]
Q2 WOM-T	0.26	>0 (predictive relevance)	Stone [76]; Geisser [77]; Hair et al. [69]

Note. SRMR = Standardized Root Mean Square Residual; NFI = Normed Fit Index; R² = coefficient of determination; Q² = Stone–Geisser predictive relevance index. The thresholds are interpreted as guidelines and should be contextualized according to the type of model and area of application [69,70,75].

The SRMR value of 0.061 is below the threshold of 0.08, indicating a satisfactory overall fit. The NFI of 0.921 suggests an adequate fit between the empirical matrix and the one reproduced by the model. The R² values for ENG (0.64), IR (0.58), eWOM (0.52), and WOM-T (0.49) are in the moderate to high range, while the positive Q² values show predictive relevance for all endogenous variables, in line with the criteria proposed for the evaluation of PLS-SEM models [38,43].

4.5. Evaluation of the Structural Model

4.5.1. Collinearity Between Constructs and Explanatory Power

The internal VIFs of the paths converging on ENG, IR, eWOM, and WOM-T remained below 3.0, ruling out severe collinearity problems between exogenous and mediating constructs, in accordance with the thresholds suggested in the literature [69,72]. The explanatory power of the model, reflected in the R² and Q² presented in Table 4, can be considered adequate for consumer behavior studies with attitudinal and behavioral intention variables [70].

4.5.2. Direct Paths: Contrast of H1a–H3b

Table 5 presents the standardized path coefficients (β), standard errors, t-values, and significance levels for the direct hypotheses H1a–H1e, H2, H3a, and H3b.

Table 5. Structural model results: direct effects.

Hyp.	Path	β	EE	t	p	f2	Result
H1a	SAT → ENG	0.21	0.06	3.50	0.001	0.06	Supported
H1b	CONF → ENG	0.24	0.06	4.00	<0.001	0.08	Supported
H1c	VE → ENG	0.27	0.06	4.46	<0.001	0.10	Supported
H1d	AN → ENG	0.16	0.05	3.02	0.003	0.04	Supported
H1e	IS → ENG	0.19	0.06	3.23	0.001	0.05	Supported
H2	ENG → IR	0.76	0.04	18.90	<0.001	0.59	Supported
H3a	IR → eWOM	0.72	0.05	14.40	<0.001	0.47	Supported
H3b	IR → WOM-T	0.68	0.05	13.20	<0.001	0.42	Supported

Note. β = standardized coefficient; SE = standard error; f² = local effect size. Values of f² ≈ 0.02, 0.15, and 0.35 are considered small, medium, and large, respectively [70,75].

Table 5. Structural model results: direct effects.

Hyp.	Path	β	EE	t	p	f2	Result
H1a	SAT → ENG	0.21	0.06	3.50	0.001	0.06	Supported
H1b	CONF → ENG	0.24	0.06	4.00	<0.001	0.08	Supported
H1c	VE → ENG	0.27	0.06	4.46	<0.001	0.10	Supported
H1d	AN → ENG	0.16	0.05	3.02	0.003	0.04	Supported
H1e	IS → ENG	0.19	0.06	3.23	0.001	0.05	Supported
H2	ENG → IR	0.76	0.04	18.90	<0.001	0.59	Supported
H3a	IR → eWOM	0.72	0.05	14.40	<0.001	0.47	Supported
H3b	IR → WOM-T	0.68	0.05	13.20	<0.001	0.42	Supported

Note. β = standardized coefficient; SE = standard error; f² = local effect size. Values of f² ≈ 0.02, 0.15, and 0.35 are considered small, medium, and large, respectively [70,75].

All direct paths were positive and statistically significant (p < 0.01). SAT, CONF, VE, AN, and IS antecedents contributed complementarily to consumer engagement, with VE (β = 0.27) and CONF (β = 0.24) having particularly strong effects. The effect of ENG on IR was high (β = 0.76) and had a large effect size (f² = 0.59), while recommendation intention showed strong effects on eWOM (β = 0.72) and WOM-T (β = 0.68). Consequently, hypotheses H1a–H1e, H2, H3a, and H3b are supported by the data. Given the single-source design, we assessed common method bias (full collinearity VIFs < 3.3 and Harman’s single-factor test; Section 4.2), and the results suggest that CMB is unlikely to be a critical threat to these effects [71,72].

4.6. Indirect Effects and Mediations

Indirect effects were estimated using bootstrapping with 5000 resamples and 95% confidence intervals. Table 6 summarizes the results of simple mediations (H4a–H4e, H5a–H5b) and sequential mediations (H6a–H6j).

Table 6. Results of simple and sequential mediations.

Hypothesis	Path	Type of Mediation	Indirect Effect	95% CI (LI; LS)	Result
H4a	SAT → ENG → IR	Simple (via ENG)	0.16	[0.11; 0.22]	Supported
H4b	CONF → ENG → IR	Simple (via ENG)	0.18	[0.13; 0.24]	Supported
H4c	VE → ENG → IR	Simple (via ENG)	0.20	[0.15; 0.27]	Supported
H4d	AN → ENG → IR	Simple (via ENG)	0.12	[0.07; 0.18]	Supported
H4e	IS → ENG → IR	Simple (via ENG)	0.14	[0.09; 0.20]	Supported
H5a	ENG → IR → eWOM	Simple (via IR)	0.55	[0.47; 0.62]	Supported
H5b	ENG → IR → WOM-T	Simple (via IR)	0.52	[0.44; 0.59]	Supported
H6a	SAT → ENG → IR → eWOM	Sequential	0.12	[0.08; 0.18]	Supported
H6b	SAT → ENG → IR → WOM-T	Sequential	0.11	[0.07; 0.17]	Supported
H6c	CONF → ENG → IR → eWOM	Sequential	0.13	[0.09; 0.19]	Supported
H6d	CONF → ENG → IR → WOM-T	Sequential	0.12	[0.08; 0.18]	Supported
H6e	VE → ENG → IR → eWOM	Sequential	0.15	[0.10; 0.21]	Supported
H6f	VE → ENG → IR → WOM-T	Sequential	0.14	[0.09; 0.20]	Supported
H6g	AN → ENG → IR → eWOM	Sequential	0.09	[0.05; 0.14]	Supported
H6h	AN → ENG → IR → WOM-T	Sequential	0.08	[0.04; 0.13]	Supported
H6i	IS → ENG → IR → eWOM	Sequential	0.10	[0.06; 0.15]	Supported
H6j	IS → ENG → IR → WOM-T	Sequential	0.09	[0.05; 0.14]	Supported

Note. LI = lower limit; LS = upper limit of the 95% confidence interval (CI). Indirect effects are considered significant when the CI does not include zero [67,78].

Table 6. Results of simple and sequential mediations.

Hypothesis	Path	Type of Mediation	Indirect Effect	95% CI (LI; LS)	Result
H4a	SAT → ENG → IR	Simple (via ENG)	0.16	[0.11; 0.22]	Supported
H4b	CONF → ENG → IR	Simple (via ENG)	0.18	[0.13; 0.24]	Supported
H4c	VE → ENG → IR	Simple (via ENG)	0.20	[0.15; 0.27]	Supported
H4d	AN → ENG → IR	Simple (via ENG)	0.12	[0.07; 0.18]	Supported
H4e	IS → ENG → IR	Simple (via ENG)	0.14	[0.09; 0.20]	Supported
H5a	ENG → IR → eWOM	Simple (via IR)	0.55	[0.47; 0.62]	Supported
H5b	ENG → IR → WOM-T	Simple (via IR)	0.52	[0.44; 0.59]	Supported
H6a	SAT → ENG → IR → eWOM	Sequential	0.12	[0.08; 0.18]	Supported
H6b	SAT → ENG → IR → WOM-T	Sequential	0.11	[0.07; 0.17]	Supported
H6c	CONF → ENG → IR → eWOM	Sequential	0.13	[0.09; 0.19]	Supported
H6d	CONF → ENG → IR → WOM-T	Sequential	0.12	[0.08; 0.18]	Supported
H6e	VE → ENG → IR → eWOM	Sequential	0.15	[0.10; 0.21]	Supported
H6f	VE → ENG → IR → WOM-T	Sequential	0.14	[0.09; 0.20]	Supported
H6g	AN → ENG → IR → eWOM	Sequential	0.09	[0.05; 0.14]	Supported
H6h	AN → ENG → IR → WOM-T	Sequential	0.08	[0.04; 0.13]	Supported
H6i	IS → ENG → IR → eWOM	Sequential	0.10	[0.06; 0.15]	Supported
H6j	IS → ENG → IR → WOM-T	Sequential	0.09	[0.05; 0.14]	Supported

Note. LI = lower limit; LS = upper limit of the 95% confidence interval (CI). Indirect effects are considered significant when the CI does not include zero [67,78].

The indirect effects of SAT, CONF, VE, AN, and IS on recommendation intention through consumer engagement (H4a–H4e) were positive and significant, indicating complete mediation by ENG, since no additional direct paths from these antecedents to IR were specified. Similarly, the indirect effects of ENG on eWOM and WOM-T through IR (H5a–H5b) were high and significant, also configuring complete mediation of recommendation intention in these relationships.

Regarding sequential mediations (H6a–H6j), all SAT/CONF/VE/AN/IS → ENG → IR → eWOM/WOM-T paths showed positive and statistically significant effects, with greater magnitudes for sequences starting from emotional bond with the brand and trust in the brand. These findings are consistent with the theorized sequential process proposed in the model, in which relational and personal antecedents are associated with WOM behaviors through a chain that involves higher engagement and, subsequently, stronger recommendation intention [67,70,79].

4.7. Summary of Quantitative Findings

In summary, the measurement model showed high levels of reliability and validity, in accordance with classic standards of psychometrics and structural equation modeling [65,66,68,69,73]. The structural model showed good overall fit, moderate–high explanatory power, and predictive relevance for engagement, recommendation intention, and eWOM and WOM-T behaviors.

All the hypotheses proposed in the conceptual model—direct paths (H1a–H1e, H2, H3a, H3b), simple mediations (H4a–H4e, H5a–H5b), and sequential mediations (H6a–H6j)—were statistically supported in this sample. Taken as an exploratory mapping exercise, this pattern provides a descriptive profile of how the studied relational, personal, and social variables are associated with engagement, recommendation intention, and WOM behaviors within the segment of internet-connected consumers in the Lambayeque Region, and it motivates confirmatory replication with independent samples and stronger designs.

5. Discussion

5.1. Overview of Findings

The main objective of this study was to explore, among internet-connected consumers in the Lambayeque Region, how satisfaction (SAT), brand trust (CONF), emotional bond (EB), openness to novelty (ON), and perceived social influence (SI) are associated with consumer engagement (ENG), intention to recommend (IR), and, finally, online word-of-mouth (eWOM) and traditional word-of-mouth (WOM-T) behaviors, incorporating simple and sequential mediations in a PLS-SEM model.

Within this purposive sample, all direct, simple mediation, and sequential mediation paths (H1a–H6j) were statistically significant and in the expected direction. Consumer engagement and recommendation intention showed the strongest associations with eWOM and WOM-T, while emotional bond, trust, and satisfaction were the most strongly associated predictors of engagement, complemented by the contribution of openness to novelty and perceived social influence. Taken as exploratory evidence, these results suggest that, in the Lambayeque context of services and commerce, word of mouth—both digital and face-to-face—is associated with a combination of satisfactory experiences, relational trust, emotional ties to the brand, openness to new proposals, and social pressures/stimuli.

Beyond statistical significance, the pattern of relative magnitudes is theoretically informative. The stronger influence of emotional bond and trust on engagement suggests that, in service and trade SMEs, recommendation processes are anchored primarily in relational and affective mechanisms rather than solely in evaluative satisfaction. This aligns with a relationship-marketing view in which bond and trust reduce perceived risk and increase psychological ownership, making engagement more resilient and more likely to translate into intention and subsequent WOM. By contrast, openness to novelty and social influence—while significant—appear to play a more contextual/accelerating role, energizing engagement and channel activation under specific situational and social conditions. This differential profile supports a stage-based interpretation of the model: relational quality provides the “base” for engagement, while dispositional and social cues modulate the speed and channel through which recommendations materialize.

5.2. Discussion of the Measurement Model and Construct Validity

Although recommendation intention (RI), eWOM, and offline WOM (WOM-T) are behaviorally proximate, they represent different levels of psychological proximity to action. RI captures a proximal motivational state—the willingness or plan to recommend—whereas eWOM and WOM-T are enacted communication behaviors that require additional situational triggers, effort, and perceived risk management. This stage-based separation is consistent with intention–behavior logic in consumer research, where intentions precede and partially translate into observable actions but do not collapse into them.

We conceptualize eWOM and WOM-T as two channel-specific manifestations of WOM that differ in structural affordances (audience reach, permanence, visibility), social presence, and perceived social risk. Offline WOM occurs in high-social-presence contexts and is typically ephemeral, while eWOM is more persistent, searchable, and publicly observable, increasing reputational and social-risk considerations and creating different behavioral thresholds. Therefore, even with a shared underlying recommendation propensity, the two behaviors remain theoretically distinct outcomes that can diverge in magnitude and antecedent strength across channels.

The results of the measurement model showed high levels of internal reliability (α and composite reliability) and convergent validity (factor loadings and AVE), in line with classical psychometric criteria and recommendations for variance-based structural equation models (PLS-SEM). The adequate internal consistency of the SAT, CONF, VE, AN, IS, ENG, IR, eWOM, and WOM-T scales confirms that the items used accurately capture the constructs defined in the variable table and the measurement instrument.

Discriminant validity verified using the Fornell–Larcker criterion and HTMT indices indicates that each construct maintains sufficient conceptual autonomy from the others. In particular, the distinction between engagement (ENG), intention to recommend (IR), and WOM behaviors in their online and offline modalities is empirically corroborated. This result is consistent with studies that differentiate between psychological states (identification, experience, engagement) and information dissemination behaviors in service environments and virtual brand communities (e.g., [24,27]).

The preservation of this multifactorial structure is particularly important in the context of Lambayeque, where consumers combine face-to-face interactions in food, retail, and service SMEs with a growing digital presence. Previous studies in the region have shown the feasibility of validating complex scales in Peruvian samples using SEM, which reinforces the psychometric robustness of the present model [15,22].

5.3. Discussion of Direct Effects (H1a–H3b)

5.3.1. Antecedents → Commitment (H1a–H1e)

The results confirm that SAT, CONF, VE, AN, and IS are positively associated with consumer commitment, supporting H1a–H1e. From a relational perspective, the combination of satisfaction and trust reinforces the satisfaction–loyalty and trust–commitment logic described in the relational marketing literature, in which consistent and reliable experiences translate into long-term bonds with the brand [80,81]. In service contexts, satisfaction and perceived value have been shown to directly influence recommendation intention and WOM, both in person and online [25,26,27].

The positive effect of emotional bonding (EB) on ENG suggests that engagement in Lambayeque is not only cognitive or transactional but deeply affective. This finding is consistent with evidence linking emotional attachment, brand identification, and willingness to share experiences on social media and in social settings [12,39,56]. In local restaurants and retail SMEs, where the brand is often associated with faces, stories, and traditions, the emotional dimension can be particularly salient.

The influence of openness to novelty (ON) on ENG suggests that consumers who are more willing to try new options are also those who tend to engage more actively with brands. This is consistent with studies that highlight the role of traits and lifestyles in the propensity to participate in online communities, generate eWOM, or explore new value propositions [6,54]. In the case of Lambayeque, where traditional offerings and innovative formats coexist (e.g., gastronomic fusions or hybrid physical–digital experiences), this openness can enhance interaction and engagement.

Finally, the positive effect of perceived social influence (SI) on ENG confirms that subjective norms and peer pressure (friends, family, reference groups) continue to be a relevant driver of engagement with brands and services [16,17]. In line with studies that show the impact of social identity and the opinions of reference groups on the generation of eWOM and online purchase intent [39,40], the results suggest that young/internet-connected consumers in Lambayeque are more engaged with brands that are socially validated in their close circles. Multi-group analysis has revealed that family influence exhibits stronger impacts on brand identification than peer influence, highlighting the differential moderating effects of reference group types on consumer–brand relationships [82].

5.3.2. Engagement → Recommendation Intention (H2)

The strong and positive effect of ENG on IR supports H2 and is consistent with theoretical frameworks that conceive engagement as a state of connection that precedes recommendation, loyalty, and other proactive customer behaviors [27,57]. Our results suggest that, to the extent that consumers in Lambayeque engage cognitively, affectively, and behaviorally with a brand—by following it on social media, participating in its activities, or integrating it into their symbolic identity—their willingness to explicitly recommend it to others increases.

This relationship coincides with recent evidence in service and hospitality contexts, where engagement is linked to a greater willingness to spread positive recommendations and sustain loyalty behaviors [13,14]. In restaurants and food services, the combination of satisfying and shareable (Instagrammable, unique) experiences reinforces this mechanism, encouraging committed customers to become brand “ambassadors” both online and offline [4,28].

5.3.3. Recommendation Intention → eWOM and WOM-T (H3a–H3b)

Evidence of positive and significant relationships between IR and eWOM (H3a) and WOM-T (H3b) behaviors confirms that recommendation intention effectively translates into communicative actions in both channels. The results suggest that the relationship could be slightly stronger toward eWOM, consistent with contexts in which social networks and digital platforms have established themselves as privileged spaces for sharing consumer experiences, especially among young and urban segments [2,5,43].

However, the simultaneous support of IR over WOM-T indicates that, in Lambayeque, face-to-face conversations and informal exchanges continue to play a key role in disseminating information about service and trade SMEs. This pattern coincides with studies in gastronomy and restaurants that show a shared prominence of face-to-face and digital recommendations, where the opinions of family and friends remain decisive in the choice of establishments [10,11,26].

Taken together, the “sequential core” ENG → IR → eWOM/WOM-T supports the idea that engagement operates as an attitudinal antecedent to the intention to recommend, which in turn acts as a gateway to effective information dissemination behaviors, both online and offline.

5.4. Discussion of Simple and Sequential Mediations (H4a–H6j)

5.4.1. Simple Mediations with ENG (H4a–H4e)

Simple mediations show that ENG channels the effects of SAT, CONF, VE, AN, and IS on IR, supporting H4a–H4e. In substantive terms, this implies that relational perceptions (satisfaction, trust, emotional attachment) and personal/social dispositions (openness to novelty, perceived social influence) do not automatically translate into recommendation intention but rather do so to the extent that they first generate a state of commitment to the brand.

This pattern is consistent with studies that have identified mediating roles of engagement, satisfaction, or trust in the relationship between perceptions of value and recommendation or loyalty behaviors [12,27,41]. Likewise, research on eWOM in gastronomic and entertainment contexts shows that the quality of the experience and perceived value influence the intention to share opinions mainly when they generate an affective and participatory bond with the brand or establishment [8,25].

5.4.2. IR Mediation Between ENG and WOM Behaviors (H5a–H5b)

The finding that IR mediates the relationship between ENG and eWOM/WOM-T (H5a–H5b) reinforces the distinction between attitudinal commitment and effective recommendation behavior. Committed consumers do not necessarily recommend automatically; rather, commitment increases the likelihood of forming an explicit intention to recommend, and it is this intention that translates into concrete messages, whether on social media or in face-to-face conversations.

This mechanism is consistent with studies that conceptualize the intention to recommend as an intermediate link between favorable attitudes or states of commitment and WOM behaviors [2,23,29]. In environments where the density of digital messages is high, the existence of a clear intention to recommend can act as a filter that selects which experiences are reflected in reviews, posts, or comments.

5.4.3. Sequential Mediations (H6a–H6j)

Sequential mediations support the complete chain SAT/CONF/VE/AN/IS → ENG → IR → eWOM/WOM-T (H6a–H6j), suggesting that relational, personal, and social antecedents are linked to WOM behaviors through a dual mechanism: higher engagement and, subsequently, stronger intention to recommend.

Within this sequence, VE and CONF tend to show particularly intense indirect effects, suggesting that emotional attachment and relational trust are central drivers of recommendation in the context of service SMEs. This pattern is consistent with studies that highlight the ability of perceived value, brand experience, and trust to drive WOM and repurchase intention in both physical and digital channels [5,27,41].

At the same time, the significance of the sequential routes associated with AN and IS reinforces the relevance of dispositional and social factors in understanding how consumer experiences spread in emerging markets. Research on social identity, subjective norms, and WOM in Latin America has shown that belonging to specific reference communities—whether peer groups, virtual communities, or business clusters—enhances information transmission and recommendation [12,39,40]. The results from Lambayeque are consistent with this line of research, providing evidence on how social and personal factors combine with brand experience to trigger recommendation behaviors.

5.5. Theoretical Contributions

Taken together, the findings offer several theoretical contributions. First, the study integrates relational antecedents (SAT, CONF, VE), personal dispositions (AN), social factors (IS), engagement (ENG), recommendation intention (IR), and WOM behaviors in their two modalities (eWOM and WOM-T) into a single model. This integration allows for a more complete understanding of the complexity of the process that leads from the consumer experience to the dissemination of recommendations, expanding on previous models that tend to focus on a subset of these variables [4,5,26].

Second, the study reinforces the role of consumer engagement as an integrating mechanism between various antecedents (satisfaction, trust, emotional attachment, openness to novelty, social influence) and the intention to recommend, which in turn precedes effective WOM. This sequential structure is consistent with the literature on engagement and planned behavior, but the present evidence consolidates it in a Latin American context of service and trade SMEs, where most previous studies on eWOM have focused on large global brands or specific platforms [2,12,14]. Notably, despite prior evidence that drivers can differ between online and offline WOM [35,36], our results suggest that the engagement → recommendation intention pathway provides a shared core mechanism that predicts both channels with comparable magnitudes in this O2O SME setting.

Third, the contribution is not the geographic label per se but the examination of boundary conditions typical of emerging regional markets: hybrid O2O consumption where face-to-face and social media interactions coexist [6], increasing but uneven Internet access among young consumers [18], and the documented salience of electronic trust and reference-group cues in Peruvian online decisions [15,16,17]. In such contexts, interpersonal communication can substitute for formal market information, making trust- and identity-based relational cues particularly relevant to explain WOM behaviors. Moreover, the strong and comparable effects of recommendation intention on both eWOM and WOM-T (H3a–H3b) support a channel-convergent mechanism, helping to reconcile fragmented WOM findings across online and offline settings [31,32].

5.6. Practical and Sustainability Implications (SDGs 8 and 12)

From a management perspective, the results suggest several lines of action for service and trade SMEs in Lambayeque. First, strengthening satisfaction and trust requires ensuring consistency between promise and performance, proactively managing complaints, and ensuring stable service quality standards. This aligns with evidence showing how perceived quality and customer experience influence satisfaction, repurchase intention, and positive word of mouth [8,25,26].

Second, the prominent role of emotional connection indicates that SMEs should cultivate narratives and experiences that reinforce brand identity: local stories, connection to regional culture, visibility of producers and suppliers, and creation of spaces for interaction that generate a sense of belonging. This affective dimension has proven to be key to driving both engagement and willingness to recommend in different service sectors [12,27,56].

Third, the relevance of NA and SI suggests the advisability of designing campaigns and programs that encourage the testing of new proposals (flavors, formats, promotions) while leveraging the weight of reference groups. Strategies such as “bring a friend” programs, referral benefits, and activations with local micro-influencers can reinforce positive social influence and trigger favorable WOM chains [12,43].

Our results do not measure sustainability outcomes directly; however, they help clarify behavioral mechanisms through which WOM can plausibly support SDG-relevant objectives. Regarding SDG 8 (Decent Work and Economic Growth) [83], the model explains how relational quality (trust and emotional bond) and engagement translate into recommendation intention and actual WOM in SMEs—behaviors that can stabilize demand through reputation and repeat patronage, thereby supporting revenue predictability and the conditions for sustained employment and local entrepreneurship. Regarding SDG 12 (Responsible Consumption and Production) [84], WOM can contribute when the content of recommendations communicates attributes linked to responsible consumption (e.g., local sourcing, waste reduction practices, ethical value-chain cues). Prior research shows that WOM and eWOM can amplify the adoption of more sustainable behaviors and reinforce the perceived value of proposals aligned with social and environmental responsibility [5,23]. Importantly, this contribution is conditional: WOM can also amplify unsustainable consumption if the experience and messaging are not aligned with responsible practices. Future studies should therefore link WOM pathways to observed sustainability outcomes (e.g., choice of responsible options, waste reduction behaviors, or SME environmental/social performance indicators) to test these mechanisms empirically.

6. Conclusions

This study analyzed, as a theory-informed exploratory assessment, the predictors in the model of electronic word of mouth (eWOM) and traditional word of mouth (WOM-T) among internet-connected consumers in the Lambayeque Region, Peru, using a variance-based structural equation model (PLS-SEM) with 380 participants. The findings show that satisfaction, brand trust, emotional bond, openness to novelty, and perceived social influence are significantly associated with consumer engagement, which is in turn associated with recommendation intention and, ultimately, recommendation behaviors in both digital and face-to-face channels. All 25 proposed hypotheses—direct paths, simple mediations, and sequential mediations—were statistically supported in this sample, highlighting the central role of emotional bond and trust as the most powerful predictors of engagement. However, this universal significance warrants critical interpretation: the funneling architecture of the model—with five antecedents converging on a single mediator (ENG)—combined with the characteristics of the non-probability sample, may jointly contribute to the absence of non-significant paths. Readers should therefore treat these results as exploratory and theory-consistent rather than confirmatory; replication with independent, probability-based samples is needed to confirm robustness (see Section 6.1 for extended methodological discussion).

The model explained between 49% and 64% of the variance in endogenous variables, demonstrating substantial predictive capacity for understanding how consumer experiences are linked to interpersonal communication. These results have direct implications for service and trade SMEs in emerging markets, suggesting that relational marketing strategies should prioritize building emotional connections, consistency in fulfilling promises, and leveraging the influence of reference groups to stimulate genuine recommendations.

6.1. Strengths and Limitations of the Study

Among the main strengths of the study are the use of PLS-SEM to estimate a complex multivariate model that includes multiple simple and sequential mediations, as well as the simultaneous integration of relational, personal, and social antecedents, consumer engagement, recommendation intention, and WOM behaviors. This approach allows us to capture latent processes that would be difficult to analyze using bivariate techniques or simpler models [69,74].

However, several limitations must be acknowledged. First, the cross-sectional design limits causal inference in temporal terms; while hypotheses were specified directionally based on theory, the sequence of events over time is not observed. Second, non-probabilistic convenience sampling limits inferential generalization: path coefficients and significance tests describe associations within the observed sample of internet-connected consumers rather than population parameters for all Lambayeque residents [4,28], limiting the statistical generalization of the results to the total population of young/internet-connected consumers in Lambayeque. Specifically, convenience-based recruitment via WhatsApp may introduce self-selection bias, as participants with greater digital engagement or stronger brand opinions may be overrepresented. This could inflate the observed relationships between engagement-related constructs and WOM behaviors. While this sampling strategy is consistent with established practices in eWOM research in Latin American emerging markets [4,17,28], the findings should be interpreted as describing the relational patterns within the studied sample rather than as population-level estimates. Future studies should employ probability-based sampling or quota stratification by age, gender, and socioeconomic level to strengthen external validity.

Third, the use of self-report questionnaires introduces the risk of common method bias, despite the methodological and statistical measures taken to mitigate it. Although the full collinearity analysis and single-factor tests suggest that there is no single dominant factor, it is always possible that certain response patterns may be influenced by social desirability or by the tendency to respond consistently [71,72]. Moreover, the exclusive reliance on procedural and statistical post hoc remedies for CMV—without ex-ante design controls such as temporal separation of predictor and criterion measures or the inclusion of marker variables [71]—means that common method variance cannot be definitively ruled out. This limitation is particularly relevant given the cross-sectional, single-source, single-occasion design of the study.

Fourth, the fact that all 25 hypothesized paths reached statistical significance warrants careful interpretation. While this outcome is consistent with the strong theoretical grounding of each hypothesis, it may also reflect characteristics inherent to the methodological design. PLS-SEM is a variance-maximizing technique that tends to produce higher path coefficients and lower standard errors compared to covariance-based SEM, particularly with moderate sample sizes [79], which may increase the likelihood of detecting significant effects. Additionally, the model’s structure—with five exogenous constructs converging on a single mediator (ENG), followed by a sequential chain to two outcomes—creates a funneling architecture that may amplify shared variance across paths. Although the full collinearity VIF assessment (all values < 3.3) and Harman’s single-factor test rule out a dominant common method factor [71,72], these statistical checks cannot entirely eliminate the possibility that residual common method variance inflates correlations among self-reported constructs measured in the same instrument and at the same point in time. Similarly, while the model is not statistically saturated (the number of free parameters is substantially lower than the number of available data points), the absence of non-significant paths suggests that the model may benefit from competing-model comparisons in future research—for example, testing alternative configurations that include direct effects from antecedents to WOM behaviors, bypassing engagement and recommendation intention. Readers should therefore interpret the universal significance of paths as supportive of the proposed theoretical structure, but with the caveat that replication with independent samples, alternative methods (e.g., CB-SEM), or experimental designs is needed to confirm the robustness of these effects.

Fifth, because the sample is skewed toward younger participants, the estimated relationships—particularly those involving openness to novelty and eWOM—may over-represent patterns typical of digital-native consumers. Therefore, generalization to the broader Lambayeque population should be made with caution. Future studies should employ quota/stratified sampling aligned with known regional demographics and include older and less digitally active consumers to test robustness across segments [18].

Finally, the focus on service and trade SMEs in the Lambayeque Region calls for caution when extrapolating the findings to other sectors (e.g., health, education, long-distance tourism) or to other regions and countries.

6.2. Future Lines of Research

These limitations give rise to several opportunities for future research. A first line of research consists of developing longitudinal studies or quasi-experimental designs that allow us to follow the evolution of engagement and WOM over time, for example, by observing how changes in service experience, communication campaigns, or economic conditions influence the ENG → IR → eWOM/WOM-T dynamic.

A second line of research aims to replicate and extend the model to other sectors and geographies. It would be relevant to analyze whether the relationship structure identified in Lambayeque holds true in contexts such as hotels, tourism, higher education, or health, as well as in other regions of Peru and Latin American countries, where the relative importance of antecedents or the intensity of mediations could vary [3,4,12].

Finally, future research could incorporate additional variables that have been shown to be relevant to WOM in the literature, such as perceived service quality, hedonic and utilitarian value, social identity, omnichannel experience, or type of digital platform [6,14,25]. It would also be valuable to explore the moderating role of cultural and generational factors or intensity of social media use, which could condition the transition from engagement to recommendation in different consumer segments.

Together, these future lines of research would allow us to deepen and refine the findings of this study, moving toward a more nuanced understanding of how consumer recommendations—both online and offline—are formed and disseminated in the context of SMEs focused on economic sustainability and responsible consumption.