The Stimulus–Organism–Behavior–Consequence (SOBC) model, proposed by Davis and Luthans (1980), constitutes an evolution of Mehrabian and Russell’s classic S–O–R framework (Mehrabian and Russell, 1974), extending its scope by incorporating an explicit behavioral stage between cognitive processing and observable consequences. This model has been widely applied to explain a range of individual behaviors, particularly in organizational and consumer contexts (Chakraborty, 2025; Talwar et al., 2021). The SOBC model posits that stimuli (S), understood as interactive environmental events, influence the organism (O), that is, an individual’s cognitive processes. This internal evaluation leads to an observable behavior (B), which in turn generates a consequence (C) derived from the action (Davis and Luthans, 1980). This sequence has been validated in diverse contexts, including ethical consumption behavior (Dhir et al., 2020), intention to adopt smart technologies (Saheb et al., 2022), and consumer responses to digital overload (Whelan et al., 2019).

In the context of digital food delivery platforms, the SOBC model provides a robust framework for analyzing how consumers simultaneously process functional and ethical-social stimuli. In this study, these are conceptualized as service quality and perceived working conditions. These dimensions operate as external influences that activate internal evaluative processes within the organism, represented by socially conscious consumption (SCC), a cognitive disposition to integrate ethical and social values into decision-making. This evaluation, in turn, leads to a specific behavior: the intention to use the application, which ultimately results in a socially observable consequence: intention to recommend the service to others.

From a functional perspective, service quality has been extensively studied as a key driver of consumer behavior in digital environments. The E-S-QUAL model, which includes the dimensions of efficiency, fulfillment, system availability, and privacy, has proven effective in explaining satisfaction, loyalty, and intention to use delivery service apps (Parasuraman et al., 2005; Wang et al., 2024). For example, attributes such as efficient navigation and loading times, timely and error-free deliveries, and secure transactions have been found to contribute directly to repeated usage and service recommendations (Belanche et al., 2021; Vu et al., 2023).

In parallel, the perceived working conditions of delivery riders represent a salient ethical-social stimulus. Recent studies show that consumers, particularly those with high levels of social consciousness, tend to penalize platforms associated with precarious labor practices, such as a lack of social protection or limited worker autonomy (Goods et al., 2024; Kim et al., 2023). These stimuli can activate critical cognitive processes (organisms) that shape consumer behavior.

Socially conscious consumption, as a central cognitive component of the model, reflects consumers’ critical evaluations of environmental stimuli. Prior research has shown that this variable influences both usage behavior and subsequent social actions such as recommending or criticizing the service (Ali et al., 2023; Belanche et al., 2021). However, this consciousness may be more strongly expressed through symbolic or relational actions, such as public recommendations or disapproval, than through immediate behavioral decisions, such as using or avoiding a platform (Vu et al., 2023).

Finally, intention to recommend is understood as a downstream behavioral consequence that socially validates or sanctions the consumption experience. Several studies have demonstrated that consumers who perceive a service to be efficient, ethical, and reliable often act as informal promoters, thereby amplifying a platform’s reputational value in digital environments (Fan et al., 2022; Kim et al., 2023). Taken together, the SOBC model offers an integrated perspective to capture the influence of environmental factors, cognitive evaluations, resulting behaviors, and social consequences. This framework is especially useful for understanding how functional assessments and perceptions of labor justice interact within the platform economy.

Within the Stimulus–Organism–Behavior–Consequence (SOBC) framework, we conceptualize intention to use as the behavioral stage (B) and intention to recommend as the consequence stage (C). This mapping follows prior SOBC applications in consumer contexts where a proximal behavioral response precedes a downstream, socially visible consequence (Awal et al., 2024; Talwar et al., 2021). In our case, usage reflects an immediate behavioral act, whereas recommendation represents a relational validation, consistent with SOBC’s broader scope from environmental stimulus to reputational consequence.

Although the canonical SOBC sequence is S → O → B → C, digital service environments often display direct “shortcut” effects from stimuli or organism to consequences. For example, consumers may recommend a platform based on efficiency, fulfillment, or fairness without altering their own usage behavior (Fan et al., 2022; Vu et al., 2023). Retaining these links is theoretically justified, as prior work on service quality shows that E-S-QUAL dimensions can drive both loyalty intentions and advocacy directly (Parasuraman et al., 2005; Wang et al., 2024). In the gig-economy setting, perceptions of labor practices also operate as salient stimuli that can influence use and recommendation differentially, while socially conscious consumption functions as a cognitive–affective filter of these evaluations (Belanche et al., 2021; Goods et al., 2024).

Multiple studies have demonstrated that consumers evaluate service quality based on their operational experience with the platform, which directly influences their cognitive processes (organism), shapes their behavioral intentions (application usage), and eventually leads to social consequences, such as service recommendation (Belanche et al., 2021; Erkmen and Turegun, 2022). One of the most widely used frameworks for operationalizing this construct in digital environments is the E-S-QUAL model, proposed by Parasuraman et al. (2005), which has been extensively adapted to the context of delivery applications. This model conceptualizes electronic service quality as a multidimensional construct comprising efficiency, fulfillment, system availability, and privacy. Each of these dimensions has significant effects on satisfaction, intention to use, and loyalty in digital settings (Fan et al., 2022; Vu et al., 2023).

Efficiency, defined as the ease and speed with which users can navigate the application and complete transactions, has been identified as a key dimension that influences user trust (Parasuraman et al., 2005). Fulfillment refers to the platform’s ability to deliver accurately on promises regarding time, product, and condition and is consistently cited as one of the most valued attributes by consumers (Kim et al., 2023). System availability, which relates to technical reliability and the absence of errors or crashes, is directly associated with the continued usage of the service (Vu et al., 2023). Finally, privacy, though less visible to users, is critical for fostering a perception of security in digital environments (Goods et al., 2024).

Beyond its direct influence on consumer behavior, service quality can also act as a cognitive trigger that facilitates socially conscious consumption as it constitutes the initial interpretive framework through which users assess the legitimacy and values of a platform (Ali et al., 2023). When the service is perceived as efficient, reliable, and aligned with user expectations, consumers may be more inclined to reflect critically on other dimensions such as labor conditions or the company’s social responsibility commitments (Erkmen and Turegun, 2022; Vu et al., 2023). In line with the reviewed literature and following the SOBC model, the following hypotheses are proposed:

In the context of food delivery platforms, working conditions are often shaped by subcontracting arrangements, absence of social protection, uncertain remuneration, and algorithmic control, which have been widely documented as constituting a precarious work environment (Belanche et al., 2021; Goods et al., 2024). These conditions have gained increasing media and public visibility through reports of workplace accidents, legal disputes, and worker protests, which have raised awareness among certain segments of the consumer public regarding labor justice on these platforms (Belanche et al., 2021; Tsarenko et al., 2019). Empirical studies have shown that negative perceptions of working conditions can directly influence consumer behavioral intentions, particularly among ethically sensitive users who may reduce their intention to recommend the service (Belanche et al., 2021; Kim and Jang, 2019).

Robust evidence also indicates that some consumers are willing to accept certain utilitarian sacrifices, such as longer wait times or modest price premiums, when they perceive that a platform is actively improving the working conditions of delivery workers (Belanche et al., 2021). However, this willingness is not uniform; while socially conscious consumers tend to penalize platforms perceived as unfair, other users prioritize convenience, price, or service speed (Audrain-Pontevia et al., 2013; Goods et al., 2024).

Moreover, many platforms operate under schemes that shift the operational costs to workers. This includes paying for their own vehicles, lacking insurance coverage, facing pressure to accept orders under rating-based systems, and being subject to the algorithmic management of tasks (Jabagi et al., 2019). Such organizational logic creates labor conditions that are intensive, risky, and opaque, and conditions that some consumers perceive as incompatible with their personal values, thereby activating internal evaluative processes.

From the perspective of the SOBC model, these perceptions function as environmental stimuli that shape an individual’s cognition (organism), which, in turn, affects their behavioral intention (service usage) and social consequences (recommendation). In other words, consumers filter these stimuli through their values and beliefs, which may be reflected in decisions that either favor or reject certain platforms (Belanche et al., 2021). Based on this evidence, the following hypotheses are proposed:

Socially conscious consumption reflects consumers’ willingness to consider the social impact of their purchasing decisions, particularly in precarious labor contexts, such as those of food delivery platforms (Belanche et al., 2021). This concern may lead to reduced intention to recommend such services. Recent studies have shown that consumers with high levels of social consciousness are especially sensitive to the working conditions of delivery workers, expressing lower intention to use and intention to recommend platforms perceived as unfair, even when these platforms offer functional benefits such as speed or low prices (Kim et al., 2023). This ethical evaluation not only shapes behavioral intentions but also influences consumers’ overall perception of the service by incorporating normative concerns into their functional judgment (Ali et al., 2023).

The literature has also conceptualized social consciousness as an individual trait that can mediate or moderate the relationship between perceived stimuli, such as service quality or working conditions, and behavioral intentions, such as using or recommending a service (Vu et al., 2023). For example, socially conscious consumers have been found to be more willing to make trade-offs in terms of convenience or price if they perceive that a company treats its workers fairly (Kim et al., 2023). Moreover, they tend to prefer platforms that actively communicate their commitment to labor well-being, even if this entails additional costs (Belanche et al., 2021).

Within the SOBC framework, socially conscious consumption translates environmental stimuli into concrete behavioral intentions and may also influence the nature of the relational consequences that users are willing to assume, such as recommending or actively promoting certain services. Based on this evidence, the following hypotheses are proposed:

In the context of food delivery platforms, the intention to recommend reflects not only customer satisfaction with the service received but also the evaluative judgment that consumers form regarding the functional and social legitimacy of the business model (Belanche et al., 2021). Various studies have shown that recommendation intentions are strongly associated with positive usage experiences in which users perceive operational efficiency and responsible business practices (Yeo et al., 2017). This relationship has been documented in both European and Latin American markets, where recommendation is understood not only as a utilitarian act, but also as a symbolic one: socially sensitive users often express their approval by actively promoting the service, reinforcing their identity as socially conscious consumers (Belanche et al., 2021).

Within the SOBC framework, intention to recommend is conceptualized as the consequence (C) of an observable behavior (B); in this case, intention to use, which is in turn influenced by internal organism processes (cognitive and affective evaluations) and environmental stimuli (Davis and Luthans, 1980; Talwar et al., 2021). Accordingly, consumers who report a high intention to use are more likely to act as informal ambassadors of the service (Vu et al., 2023).

Furthermore, the literature emphasizes that both functional and ethical-social stimuli affect this behavior. In particular, consumers with high levels of social consciousness are more likely to recommend platforms that ensure decent working conditions, even if doing so entails sacrificing functional aspects, such as price or speed (Goods et al., 2024). From this perspective, recommendations serve as a form of moral validation aimed at encouraging more just and sustainable business models. Based on this evidence, we propose the following hypothesis:

Figure 1 presents the conceptual model based on the Stimulus–Organism–Behavior–Consequence (S–O–B–C) framework. The model posits that perceived service quality and working conditions act as external stimuli (S) that influence the internal evaluative process of socially conscious consumption (O). This internal state, in turn, is associated with two key outcomes: the intention to use food delivery applications (behavior) and the intention to recommend such services to others (consequence). The model reflects a sequential logic whereby functional and ethical-social perceptions trigger value-based evaluations, which then lead to individual and relational behavioral responses. This structure enables the analysis of how users integrate operational performance and labor-related concerns into their decision-making processes, particularly distinguishing between personal usage and public endorsement of digital platforms.

This study employed a quantitative, cross-sectional design to analyze functional and ethical–social factors influencing intentions to use and recommend digital food delivery platforms. Data were collected between October and November 2023 across different regions of Chile via a structured questionnaire administered by professional interviewers. The final sample comprised 416 valid responses obtained through a non-probability intercept design. Participants were recruited through interviewer-administered intercepts at high-traffic public venues (e.g., shopping centers, parks, transit hubs) located across the northern, central, and southern macro-zones of Chile. To enhance coverage and reduce temporal biases, fieldwork was distributed across weekdays and weekends and included both daytime and evening sessions. Eligibility criteria required respondents to be residents of Chile, aged 18 or older, active users of food delivery apps with internet access, and to have placed at least one order in the month prior to the survey. Interviewers applied soft quotas for gender, age, and macro-zone based on official national statistics for digital service users (SUBTEL, 2017) to approximate the demographic distribution of the target population. Regarding the sample profile, 49.3% identified as women, 49.3% as men, and 1.4% as another gender. In terms of age, 50% were 21–30 years old, 18.3% were 31–40, and 14.4% were 18–20. Geographically, 56.5% resided in the central zone, 22.6% in the northern zone, and 20.9% in the southern zone. Most participants were single (75.7%). PedidosYa was the most frequently used application (44%), followed by Uber Eats (7.7%). On average, participants reported spending CLP $10,000–$20,000 per month on app-based orders (45.9%), with debit cards being the most common payment method (42.3%).

The data collection instrument consisted of a structured questionnaire administered by professional interviewers using a digital form. The first section gathered basic sociodemographic information (e.g., gender, age, educational level, marital status). The second section comprised items designed to measure the variables included in the analytical model: perceived service quality, perceived working conditions, socially conscious consumption, intention to use, and intention to recommend food delivery applications. Perceived service quality was operationalized using the E-S-QUAL model (Parasuraman et al., 2005), which includes four dimensions: efficiency, fulfillment, system availability, and privacy. In line with E-S-QUAL, the Privacy dimension captures users’ perceptions about the protection of personal and financial data. Respondents evaluate whether apps protect their purchase information, safeguard credit card details, and avoid sharing personal data with other parties. These judgments are based on observable cues such as privacy notices, payment security features, or app permissions, rather than direct verification of back-end practices. Perceived working conditions were measured with scales related to internal social responsibility and labor ethics adapted from Öberseder et al. (2013), assessing perceptions of organizational justice and labor rights protection. Socially conscious consumption was evaluated using a scale adapted from Pepper et al. (2010). Intention to use was measured with items based on Belanche et al. (2011), and intention to recommend with items derived from Harrison-Walker (2001). All variables were measured on a 7-point Likert scale (1 = “Strongly disagree” to 7 = “Strongly agree”).

To ensure linguistic and cultural equivalence for the Spanish-speaking context, we followed a forward–back translation procedure (Brislin, 1980). Two bilingual researchers independently translated the original English items into Spanish, and a third researcher—blind to the originals—back-translated the Spanish version into English. Discrepancies were discussed and resolved by consensus to preserve semantic and conceptual equivalence. Prior to full deployment, the instrument underwent a two-stage pretesting process: (i) a pilot with a student subsample to refine wording and ensure semantic clarity, and (ii) a field pretest with 25 respondents from the target population to assess cultural adequacy and comprehension. Minor wording adjustments were incorporated based on feedback before launching the full survey.

Data analysis was conducted using Partial Least Squares Structural Equation Modeling (PLS-SEM), a statistical technique widely used in social and behavioral sciences because of its ability to model complex relationships among latent variables, control for measurement errors, and test comprehensive theoretical frameworks (Gudergan et al., 2008; Henseler et al., 2016). This approach is particularly suitable for exploratory research or models involving a large number of constructs and indicators, as is the case in this present study. Moreover, PLS-SEM offers flexibility regarding distributional assumptions and sample size requirements, making it especially well suited for applied research in digital marketing and consumer behavior on platform-based services. Following the methodological guidelines of Hair et al. (2022), the analysis was structured in two stages: first, the evaluation of the measurement model to assess the reliability and validity of the constructs, and second, the evaluation of the structural model to test the proposed hypotheses. Both stages were conducted using the SmartPLS 4 software (Ringle et al., 2022).

Three key criteria of the measurement model were assessed to ensure the psychometric quality of the constructs: internal reliability, convergent validity, and discriminant validity. This evaluation followed established methodological recommendations (Fornell and Larcker, 1981; Henseler et al., 2016; Sarstedt et al., 2017) using the Partial Least Squares Structural Equation Modeling (PLS-SEM) approach. Cronbach’s Alpha (CA) and Composite Reliability (CR) were used to assess the internal consistency of the constructs. As shown in Table 1, all the CA and CR values exceeded the recommended threshold of 0.70, indicating adequate internal reliability for each construct (Hair et al., 2022). Convergent validity was assessed using two indicators: individual item factor loadings and Average Variance Extracted (AVE). All item loadings exceeded the minimum accepted threshold of 0.70, and the AVE values were greater than 0.50, thus supporting the convergent validity of the constructs (Chin, 1998; Fornell and Larcker, 1981). System Availability, measured with two items after indicator refinement, showed high reliability and convergent validity, confirming its adequacy within the measurement model. Table 1 presents the detailed results for loadings, AVE, CR, and CA by construct.

Discriminant validity was assessed using two complementary criteria: the Fornell–Larcker criterion and the heterotrait–monotrait ratio (HTMT). According to the Fornell–Larcker criterion, the square root of the AVE of each construct should be greater than its correlation with other constructs in the model. As shown in Table 2, this condition was satisfied in all cases (Fornell and Larcker, 1981). Regarding the HTMT criterion (Table 3), the values obtained for all pairs of constructs were below the threshold of 0.90, indicating adequate discriminant validity between latent variables (Henseler et al., 2016; Voorhees et al., 2016).

Service quality was modeled as a reflective higher-order construct (second-order) composed of four first-order dimensions: efficiency, fulfillment, system availability, and privacy. To estimate this hierarchical model, we applied the disjoint two-stage approach in PLS-SEM, whereby in the first stage the latent variable scores of the first-order constructs are obtained, and in the second stage these scores are used as manifest indicators of the second-order construct (Becker et al., 2012; Sarstedt et al., 2019). This procedure is consistent with the general two-step logic of structural equation modeling, which emphasizes establishing measurement quality prior to the assessment of structural relations (Anderson, 1988), and aligns with the literature on multidimensional constructs (Wright et al., 2012).

As shown in Table 4, all dimensions of the second-order construct exhibited loadings above 0.70 and satisfied the required thresholds for Average Variance Extracted (AVE), Composite Reliability (CR), and Cronbach’s alpha (CA), thereby confirming the validity of the hierarchical model. In addition, and consistent with best practice for higher-order models (Sarstedt et al., 2019), we examined collinearity diagnostics among the first-order dimensions. All VIF values ranged between 1.7 and 2.8, which are well below the recommended cut-off of 3.3 (Hair et al., 2022), confirming the absence of multicollinearity issues in the higher-order measurement model. For completeness, we also inspected item-level collinearity for the E-S-QUAL dimensions; all outer VIFs were below 5.0.

Overall, these results indicated that the hierarchical measurement model demonstrates adequate reliability, convergent validity, and discriminant validity, providing a solid foundation for the subsequent evaluation of the structural model.

We used bootstrapping with 5,000 resamples, two-tailed tests, and 95% confidence intervals to obtain standard errors, t-values, and confidence intervals for all direct and indirect effects. Before analyzing the structural relationships among the constructs, the overall model fit was assessed using the Standardized Root Mean Square Residual (SRMR) index. The model yielded a value of 0.057, indicating a good fit according to the established thresholds for PLS-SEM models (Henseler et al., 2016). For completeness, additional exact-fit indices were also inspected (dULS = 0.612; dG = 0.367), which are in line with values typically reported in prediction-oriented PLS-SEM studies.

To address the potential risk of common-method bias associated with the single-source cross-sectional design, both procedural and statistical remedies were applied. Procedurally, anonymity and confidentiality were ensured, and items were randomized within the online questionnaire to reduce evaluation apprehension and consistency artifacts. Statistically, we conducted the full collinearity assessment test as recommended by Kock (2015). The full collinearity VIFs were 2.89 for service quality, 2.11 for working conditions, 1.76 for socially conscious consumption, 2.54 for intention to use, and 2.42 for intention to recommend. All values were well below the conservative threshold of 3.3, indicating the absence of common-method variance issues and supporting the robustness of the observed relationships.

The results showed that the stimuli included in the model explained 30.5% of the variance in socially conscious consumption, 51.2% of the variance in intention to use, and 65.6% of the variance in intention to recommend (Figure 2). These values suggest a moderate to high explanatory capacity, particularly with respect to relational consumer behaviors such as service recommendations.

Table 5 reports standardized path coefficients (β), t-values, two-tailed 95% confidence intervals, and p-values obtained via bootstrapping. The hypotheses proposed within the SOBC framework are partially supported. Service quality exerts significant positive effects on socially conscious consumption, intention to use, and intention to recommend, reinforcing the central role of functional attributes (efficiency, system availability, fulfillment, privacy) in shaping both individual and relational behaviors on food delivery platforms. Working conditions significantly affect socially conscious consumption and intention to recommend but not intention to use. Socially conscious consumption significantly predicts intention to recommend but not intention to use. Finally, intention to use significantly predicts intention to recommend.

In addition to the analysis of path significance, we assessed the effect sizes (f²) of each predictor on its endogenous constructs by comparing the explained variance of the full model with models where each predictor was omitted in turn (Cohen, 1988). Table 6 reports the f² values, which indicate the relative contribution of each predictor beyond statistical significance. Following conventional thresholds (0.02 = small, 0.15 = medium, 0.35 = large), values below 0.02 are considered negligible or trivial (Cohen, 1988).

The results show that service quality exerts the strongest influence, with a large effect on Intention to Use (f² = 0.664), a medium effect on socially conscious consumption (f² = 0.214), and a small-to-medium effect on intention to recommend (f² = 0.121). Working conditions have only a small effect on socially conscious consumption (f² = 0.059) and intention to recommend (f² = 0.057), and a negligible effect on intention to use (f² = 0.001). Socially conscious consumption contributes negligible effects on both intention to use (f² = 0.007) and intention to recommend (f² = 0.041). Finally, Intention to use exerts a medium effect on intention to recommend (f² = 0.211), confirming its role in linking functional assessments to relational consequences. Overall, these results reinforce the theoretical expectation that functional quality strongly drives immediate behavioral responses, whereas ethical-social considerations have more limited predictive power, shaping evaluations and downstream recommendations only modestly.

We tested indirect paths using 5,000 bootstrap resamples (two-tailed 95% CIs). Results in Table 7 show that service quality has a significant total indirect effect on recommend (β = 0.338, p < 0.001), decomposed mainly via use [Service quality → Use → Recommend: β = 0.265, 95% CI (0.187, 0.341)] and, to a lesser extent, via SCC [Service quality → SCC → Recommend: β = 0.061, 95% CI (0.025, 0.099)]. The serial route Service quality → SCC → Use → Recommend is positive but not significant [β = 0.012, 95% CI (−0.005, 0.032)].

For working conditions, the specific indirect effect via SCC on recommend is significant [working conditions → SCC → Recommend: β = 0.032, 95% CI (0.011, 0.058)], whereas the total indirect effect is small and not significant (β = 0.027, 95% CI crosses zero) due to a negligible indirect component through Use. The indirect effect SCC → Use → Recommend is not significant (β = 0.028, 95% CI includes 0). These results align with the SOBC mapping that emphasizes Use as a proximal behavior (B) and recommend as a downstream consequence (C), with SCC operating as a value-laden cognitive mechanism (O) that channels ethical-social appraisals into relational consequences.

Taken together, these findings indicate that service quality exerts both direct and indirect effects on recommendation, with indirect channels accounting for a substantial share of the total effect, mainly via Use and secondarily via SCC. In contrast, working conditions influence recommendation primarily through SCC, underscoring that ethical appraisals translate into symbolic, relational consequences rather than transactional use. These results reinforce the theoretical mapping of the SOBC model and provide empirical evidence for the coexistence of sequential and shortcut paths in platform-based consumption.

To further evaluate the model’s predictive capability, we assessed predictive relevance (Q²) using the blindfolding procedure (omission distance = 7) and applied the PLSpredict procedure (Shmueli et al., 2019). Table 8 presents the Q² values for the three endogenous constructs. All Q² values were greater than zero, confirming the model’s predictive relevance (Chin, 2010; Hair et al., 2019).

The results indicate that the model exhibits medium predictive relevance for Socially Conscious Consumption and strong predictive relevance for both Intention to Use and Intention to Recommend, underscoring the robustness of the model in predicting key behavioral outcomes.

In addition, PLSpredict was performed to assess out-of-sample prediction accuracy. Table 9 compares the prediction errors (RMSE and MAE) of the PLS model against a linear regression benchmark (LM).

The PLSpredict results confirm that the model achieves lower prediction errors than the linear benchmark for socially conscious consumption, with significant improvements in predictive accuracy (p = 0.022). For intention to use and intention to recommend, PLS and LM perform similarly, which is consistent with the strong explanatory and predictive power already indicated by R² and Q². Taken together, these findings strengthen the empirical robustness of the model and provide additional support for its practical implications in digital platform contexts.