Some models extend the understanding of sustainable consumption, such as the stimulus-organism-response model, which examines cognitive and affective influences on behavior as external stimuli that affect the internal state and, consequently, result in behavior (Mehrabian and Russell, 1974). Similarly to theory of planned behavior of Ajzen (1991), the stimulus-organism-response model seeks to explain an individual’s behavior. However, unlike said theory, in the stimulus-organism-response model, factors external to the individual are the predictors of the individual’s internal state, which is a predictor of behavior. While in the theory of planned behavior attitudes, subjective norms and the perceived control of behavior are the predictors of individual’s behavior.

In food consumption, external stimuli that affect the internal state of the consumer and consequently lead to food purchasing behavior have been considered in various ways (Lee and Yun, 2015; Liu and Zheng, 2019; Lee et al., 2020). For example, it is analyzed how objects and psychological stimuli affect the individual’s internal state, and the individual as a response has a food sustainable consumption (Lee and Yun, 2015). Liu and Zheng (2019) analyze how stimuli (food safety incidents, consumer environment orientation, and consumer health orientation) influence consumer cognition, influencing organic purchasing. Through the stimulus-organism-response model, Lee et al. (2020) explain the purchasing behavior of organic food through the stimulation of the intrinsic and extrinsic characteristics of the food in the consumer’s attitude and the effect of this on shopping behavior.

Manthiou et al. (2017) consider that the physical environment (stimulus) influences the cognitive and emotional perspectives of the consumer (organism) responding with the behavior towards the environment (response). Therefore, the COVID-19 pandemic can be considered the physical environment, the stimulus. In this sense, the stimulus-organism-response model is used as a theoretical framework to analyze the purchasing behavior of organic food during the health contingency period due to COVID-19 (Liu et al., 2021; Yin et al., 2021). However, not all studies analyze the possible impacts of this disease on consumption; they only analyze consumption in the context of the pandemic without quantifying its effect (Liu et al., 2021). Yin et al. (2021) consider COVID-19 as the external stimulus through the event force that the pandemic has on organic food consumption. However, unlike this, in this research, the external stimuli of COVID-19 are analyzed through the health belief model. This model is used in research related to healthcare behaviors (Wong et al., 2020; Guidry et al., 2021; Mercadante and Law, 2021).

The health belief model (HBM) explains preventive health behaviors through personal motivation to achieve goals in the area of health (Maiman and Becker, 1974). In addition, I HBM aims to analyze behaviors in conditions of uncertainty (Becker et al., 1974), such as what is happening with the COVID-19 pandemic.

The Health Belief Model postulates that for the individual to have behavior related to preventive health, they must have the disposition to act based on the perception of vulnerability to a health condition and the severity of the consequences of contracting the condition. Furthermore, assessment of the feasibility and efficacy of reducing their exposure by performing the behavior is better than the barriers and costs. A cue to action is triggered as their interpersonal interactions (Maiman and Becker, 1974).

Likewise, Rosenstock et al. (1988) argue that individuals must have sufficient motivation that the health condition is relevant and that they are susceptible to it. Moreover, a health recommendation will benefit them since it will reduce their susceptibility to an acceptable cost, the barriers, which are not necessarily only economic.

Since the HBM, some authors consider perceived severity, benefits, and perceived barriers predictors of health-related behavior (Myers and Goodwin, 2011; Guidry et al., 2021). These factors related to the characteristics and knowledge of the individual affect their beliefs and encourage behavior (Mercadante and Law, 2021). In addition, studies prove that external stimuli, such as perceived severity, affect the organism under the analysis of the SOR model (Wang et al., 2021).

It is also necessary to contemplate the signals for action (cue to action) included within the health belief model as stimuli of the organism, which can be interpersonal interactions or with the media that provide individuals with knowledge about the health condition (Maiman and Becker, 1974). These may be social networks because they provide information to those who interact with them without topics such as recommendations and are predictors of consumer behavior (De Valck et al., 2009). Zaglia (2013) confirms that social interactions between members of a social network and belonging to that network influence consumers’ social identity. Therefore, according to the stimulus-organism-response model, it is possible to consider the predictors of the health belief model as the external stimuli caused by COVID-19 that affect the organism (social identity). With this, the following hypotheses are postulated:

Identity is defined as a consumer association towards a label they choose and a clear image of how the person looks, thinks, and feels (Reed et al., 2012). Finally, in green consumption, pro-environmental self-identity is defined as the consumer morally obliged to carry out a green action that will bring satisfaction (Mutum et al., 2021).

Mutum et al. (2021) find that identity explains green shopping; if the consumer considers himself concerned and respectful of the environment, it causes him pride and pleasure to be considered a compliant consregarded as consumer will make green purchases regularly. In addition, according to Liu et al. (2021), the SOR model allows for analyzing consumer behavior as it provides a structured framework to evaluate the environmental stimulus in the consumer’s psychological factors such as emotion, perception, and cognition and turn their effect on consumption. Therefore, in this research, the following hypothesis is proposed:

An individual’s perceived risk is a prospective subjective loss that could endanger their health and well-being (Paek and Hove, 2017; Chen and Wang, 2022). In a crisis, consumers respond to risk based on their subjective perception since their knowledge of risk factors lacks objectivity (Paek and Hove, 2017; Lejano and Stokols, 2021). According to Slovic et al. (1984) when there is an unknown risk, people perceive that the dangers are newcomer, and unobservable, similar to what happens in the context of the pandemic. Since the pandemic, some research find that perceived risk affects the intention to purchase food, whether online or in person (Leung and Cai, 2021; Chen and Wang, 2022).

While a person’s perception of the seriousness of a threat and how it will affect them is known as perceived severity (Milne et al., 2000; Baghiani-Moghadam et al., 2015). Therefore, the perceived severity denotes how much the perceived risk, in this study, COVID-19, can affect the person. In research in the area of health, the perceived severity affects the decisions to carry out behavior that brings benefits to health, as the health belief model proposes since the perceived risk affects the intention to vaccinate (Myers and Goodwin, 2011; Guidry et al., 2021). During the health contingency by COVID-19, perceived severity positively influences the intention to purchase organic food. The negative impact of the disease leads consumers to be willing to buy organic food when shopping (Wang et al., 2021). For that, the following hypothesis is proposed:

Socially responsible consumption guides actions toward improving people’s quality of life and caring for the environment. Therefore, for sustainable food consumption to exist, it is required that it be economically and ecologically viable, that is, that food is accessible to the consumer, has a fair price, and does not deteriorate the environment. A relevant variable is the perception of personal gain, that is, the perceived benefits. This is conceptualized as the people’s perception advantages and disadvantages of being socially responsible (Ellen et al., 1991; Ellen, 1994). The perception of personal benefit refers to the subjective assessment that the individual makes about the personal advantages and disadvantages unique has when acting in asocially responsibly (Lin and Hsu, 2015).

In this sense, the behavior of the socially responsible consumer is explained by certain beliefs of perceived personal benefit (Zhao et al., 2014; Lin and Hsu, 2015; Lin and Niu, 2018; Pawaskar et al., 2018; Testa et al., 2019; Yarimoglu and Binboga, 2019). When there are health benefits for certain foods or beverages, such as coffee, consumers are more inclined towards their consumption; this phenomenon occurs when it comes to female consumers (Samoggia and Riedel, 2019). Therefore, the following hypothesis is proposed:

Promote a healthier lifestyle, there are both benefits and barriers. The barriers consumers perceive can be economic when there are significant differences in food prices (The European Food Information Council, 2009). The barriers can also be personal when they attend to the lack of time both to travel to those places where to find this type of sustainable products and to prepare them, which has brought with it a restructuring of eating habits, due to the growing consumption fast food (NESI Forum on New Economy and Innovation, OCU, 2019). There are also systematic barriers that refer to the lack of reliable information on products, the lack of confidence in company social responsibility policies, the planned obsolescence of products that force them to be replaced by others, and due to the lack of legislation that acts as a boost to responsible consumption (Lima et al., 2021). Finally, it can find the barriers of eating habits that refer to resistance to change, since the patterns of adults have been formed for a long time and are difficult to change (Munárriz and De Luis, 2009; Leng et al., 2017). When there are health benefits to performing a behavior such as reducing meat consumption, the perceived benefits and barriers influence the intention to perform such behavior (Cheah et al., 2020).

Cue to action involves personal interactions and participation in social groups (Maiman and Becker, 1974; De Valck et al., 2009). Being a member of different groups in social networks causes consumers to acquire sustainable purchasing behavior. For example, they choose products with green packaging buy green products or verify the products’ ingredients to ensure that their purchase is sustainable (Cui et al., 2022). The above propose the following hypothesis:

Generally, identity is a predictor of sustainable consumption behaviors (Mutum et al., 2021), but it can act as mediator on this behavior. In socially responsible consumption, identity is considered a predictor of socially responsible purchasing behavior in young adults (Johnson and Chattaraman, 2021). According to the stimulus organism response model, the stimulus leads the organism to have a response that can be a behavior (Mehrabian and Russell, 1974). From an extension of the SOR, Talwar et al. (2021) find that the consumer’s identity as an ethical person predicts their willingness to purchase organic food. However, this study also seeks to know if the stimulus affects behavior through the organism, as evidenced by Liu et al. (2021). They find that the organism (cognition) acts as a mediators in the relation of stimulus and organic food purchasing behavior. Additionally, Wang et al. (2021) find that the organism (health consciousness) has a mediating effect between the stimulus (perceived severity) and the response (purchase intention to organic food). Therefore, social identity can take the role of mediator between perceived severity and socially responsible consumption and with this the following hypothesis is postulated:

To achieve a healthy and sustainable lifestyle, the consumer considers the benefits and barriers involved in their purchase as stated above, an assessment of the advantages and disadvantages of having a socially responsible behavior is made (Ellen et al., 1991), however, little is known about what affects or intervenes in these relationships. The barriers to socially responsible behavior can be economic, personal, or habitual (Munárriz and De Luis, 2009; The European Food Information Council, 2009; Leng et al., 2017), while the benefits are generally towards health (Samoggia and Riedel, 2019). In addition, through the SOR it is possible to consider mediation of the organism between the stimulus and the response (Liu et al., 2021; Wang et al., 2021).

For this reason, this research explores whether social identity has a role as a mediator between benefits and socially responsible consumption and barriers and socially responsible consumption, given that identity also explains ecological consumption and the SOR model provided the theoretical framework for their analysis (Liu et al., 2021; Mutum et al., 2021). Therefore, the following hypotheses are proposed. Identity mediates the relationships between stimuli (perceived benefits, perceived barriers) and socially responsible behavior.

Also, according to the health belief model, there must be a cue to action so that the individual can have a behavior that is good for him when there is a condition of risk to his/her health (Maiman and Becker, 1974), as in the case of COVID-19. Although, like the previous cases, little is known about the interactions that can affect this relationship, research explores whether social identity can mediate this relationship since this is also a variable that explains similar behavior (Mutum et al., 2021). Therefore, the following hypothesis is postulated:

Age is a variable that act as a predictor or moderator variable in the analysis of sustainable consumption (Chekima et al., 2016; Bulut et al., 2017). Bulut et al. (2017) find age to be a predictor of sustainable consumption in Turkey, while Chekima et al. (2016) find that age can act as a moderating variable of sustainable consumption in Malaysia. In addition, Quoquab and Mohammad (2020) in a sustainable consumption review from 2000 to 2020, propose age as a moderating variable in a conceptual model. In addition, age also acts as a moderating variable when analyzing the effect of COVID-19 on both sustainable consumption and social responsibility, as shown in the study by Ali et al. (2021). Their study confirms significant differences between generations X, Y, and baby boomers in the relationships between COVID-19 and sustainable consumption and COVID-19 and the social responsibility of Malaysian consumers. Therefore, when studying the effect of COVID-19 on socially responsible consumption, age can have a moderating effect on the relationships in the model, as proposed in the following hypothesis:

In this way, the stimulus-organism-response model and the health belief model allow us to analyze the effect of COVID-19 on socially responsible consumption, as shown in Figure 1.

Perceived severity has been defined as the perception of the seriousness of the consequences of contracting a condition that is detrimental to health (Maiman and Becker, 1974). For this study, this concept was adapted according to the research objective, so perceived severity was defined as the consumer’s perception of the seriousness of the consequences of contracting the COVID-19 disease. For its measurement, the items proposed by Myers and Goodwin (2011) were adapted to four items with a Likert scale from 1 totally disagree to 5 totally agree.

Perceived benefits were defined as the perceived feasibility and efficacy of reducing the consumer’s vulnerability to contracting the COVID-19 disease by engaging in socially responsible food purchasing behavior (Maiman and Becker, 1974). The variable was adapted from the scale used by Myers and Goodwin (2011) measured by three items from 1 totally disagree to 5 totally agree.

The perceived barriers were defined as the limitations to carrying out socially responsible food purchase behavior due to time and ignorance. For their measurement, the items proposed by Nguyen et al. (2016) were adapted to 2 items with a Likert scale of 5 points (1 totally disagree to 5 totally agree). Initially, 3 items were considered to measure this variable, however, since one of them during the pilot test had a factor loading of less than 0.7, it was discarded from the model. This item measured the financial barrier. Some authors have validated using 2 items measuring variables (Baumert et al., 2014; Forsell et al., 2019). Also, it is recommended improve scale items to remove ambiguity as procedural remedies to prevent common method bias (Podsakoff et al., 2012). Therefore, to measure this variable only 2 items were used.

The cue to action was defined as the interpersonal interactions that consumers have within groups where they obtain behavioral information (Maiman and Becker, 1974), which, in the case of this study, these groups are social networks. This construct was adapted from Cui et al. (2022) with an item with a Likert scale from 1 totally disagree to 5 totally agree.

Social identity was conceptualized as the set of attributes perceived by the individual that represents their way of thinking, feeling, and being (Stets and Biga, 2003). The Johnson and Chattaraman (2021) scale was adapted to 4 items measured at 5 points (1 totally disagree to 5 totally agree).

The measurement of the socially responsible food consumption construct was adapted and conceptualized by Villa Castaño et al. (2016) as the recognition of the consumer that the company is responsible for the effects caused by the production of its food towards the environment or vulnerable groups. This construct was measured by three items with a Likert scale from 1 totally disagree to 5 totally agree. Table 2 shows the constructs of the research model with their measurements.

The sample size (834) met the minimum required for the PLS-SEM analysis, which was obtained by a statistical power analysis using Cohen’s statistical power tables suggested by Benitez et al. (2020) when using PLS-SEM. That consisted of determining (1) the level of significance of the acceptable study, which was 0.05; (2) the number of predictors, considered as the most significant number of structural paths of the endogenous construct, which was 5; and (3) the effect size, which was small to have a conservative approach to the study (Cohen, 1988; Nitzl, 2016; Benitez et al., 2020). With these values, according to the statistical power tables, the minimum size required was 647 (Nitzl, 2016), the study sample size being greater than that calculated.

Data analysis was performed using PLS-SEM because it is recommended to use this method of analysis when a theoretical framework is tested, and there is a complex structural model with several constructs and indicators (Hair et al., 2019). Before analyzing the model with the total sample, a pilot test was carried out to validate the measurement scales and determine their reliability, convergent, and discriminant validity, for which items with low factor loads were eliminated, after which the data was collected total sample number. Before data analysis, it was confirmed that the data did not have an excessively abnormal distribution with kurtosis and asymmetry values outside the range of −1 to 1 (Hair et al., 2017). Therefore, first, the assessment of the measurement model was carried out, and second, the assessment of the structural model was carried out as postulated by Hair et al. (2019). The fit of the model was also evaluated (Benitez et al., 2020), and advanced analyzes were performed to complete the hypothesis tests involving mediation analysis and multigroup analysis to assess the moderating effect of age on the model of research (Nitzl et al., 2016; Wong, 2016). Analyzes made in this research are shown in Table 3.

When more than one measurement of the same or different traits is taken using the same method, it is known as common method bias (CMB). It is thought to cause the discrepancy between the trait and measured scores (Podsakoff et al., 2012). Therefore, given that bias may alter outcomes due to systematic errors, CMB could signify a hazard in social scientific research (Schwarz et al., 2017). When the same method is used to measure different constructs, they could share some of the observed covariation (Podsakoff et al., 2012). Therefore, it is recommended to use procedural remedies to prevent CMB during the research design phase when analyzes the data by PLS-SEM like improving scale items to eliminate ambiguity (Podsakoff et al., 2012; Felipe et al., 2017).

A full collinearity test using variance inflation factors (VIF) was conducted to identify a potential CMB scenario (Kock, 2015; Felipe et al., 2017). Collinearity occurs when two or more variables measure the same attribute and is measured in models with multiple variables to avoid redundancy (Kock and Lynn, 2012). Calculating the scores of the latent variables in PLS-SEM does not eliminate the collinearity between them, although they have passed validity and reliability tests, it only minimizes the collinearity (Chin et al., 2003; Haenlein and Kaplan, 2004; Kock and Lynn, 2012). Vertical and lateral collinearity were assessed (Kock and Lynn, 2012; Felipe et al., 2017). Vertical collinearity was evaluated among latent variable predictors (Kock and Lynn, 2012), the results are shown in Table 4. To assess the lateral collinearity a dummy variable obtained with random values was used as an endogenous variable and the other variables of the model were the exogenous variables as recommended by Kock and Lynn (2012) (Table 5). All VIF values were less than 3.3, which according to Kock (2015) and Felipe et al. (2017) indicates that there are no multicollinearity problems and no CMB.

The evaluation of the reflective measurement model was carried out, which involves the assessment of the reliability of the indicators and the reliability of internal consistency, and the convergent and discriminant validity of the measures of the constructs (Ali et al., 2018). The indicators were considered to have values greater than 0.7 so that each would explain at least 50% of the variance of their construct (Benitez et al., 2020). Composite reliability, Cronbach’s alpha, and Dijkstra and Henseler’s value were used to assess internal consistency (Dijkstra and Henseler, 2015; Ali et al., 2018; Hair et al., 2019). It was taken into account that the internal consistency values were between 0.6 and 0.95 since values greater than 0.6 are considered acceptable in exploratory research and values greater than 0.95 suggest multicollinearity (Hair et al., 2019). To evaluate the convergent validity values greater than 0.5 of the average variance extracted (AVE) were taken as valid since this indicates that at least the set of indicators of the construct explains it by 50% (Hair et al., 2017; Ali et al., 2018). Hair et al. (2019) indicate that the Heterotrait-Monotrait Ratio (HTMT) is used to evaluate and guarantee the constructs’ discriminant validity. The values must be less than 0.85 when the constructs are conceptually different, as in the case of the constructs of this study. The results of the evaluation of the measurement model are shown in Table 6.

In the structural model assessment, the determination coefficients (R²), the effect sizes (f²), and the path coefficients were determined (Benitez et al., 2020), with which the tests of the hypotheses 1a, 1b, 1c, 1d 2, 3a, 3b, 3c, and 3d were carried out. Figure 2 shows the values of R²; the value of socially responsible consumption was weak because it had a value of less than 0.5 and greater than 0.25 (Hair et al., 2011). However, the effect size of social identity in socially responsible consumption is large (f² = 0.474) because it is greater than 0.35 (Cohen, 1988) as shown in Table 7. Only perceived benefits have a small effect on social identity (f² = 0.27). According to Cohen (1988) values greater than 0.25 of effect size are considered small, while the other values do not represent any effect.

With the path coefficients, the tests of hypotheses 1a to 3d of the study were carried out. First, the effects of the stimuli on the organism (social identity) were evaluated, for which hypothesis tests 1a, 1b, 1c, and 1d were performed. Hypotheses 1a, 1b, and 1d are tested. Perceived severity significantly influences social identity (β = 0.118, p = 0.000). In turn, perceived benefits significantly influence social identity (β = 0.164, p = 0.000). Furthermore, the cue to action significantly influences social identity (β = 0.137, p = 0.000). Although hypothesis 1c is not tested, perceived barriers do not significantly influence social identity (β = 0.049, p = 0.162).

Second, the effect of the organism (social identity) on the response (socially responsible consumption) was evaluated using hypothesis 2. Finally, Hypothesis 2, which postulates that social identity significantly influences socially responsible consumption, is tested (β = 0.571, p = 0.000).

Third, the effects of the stimuli on the response (socially responsible consumption) were evaluated through hypotheses 3a, 3b, 3c, and 3d. In the case of hypotheses 3a, 3b, and 3d, there is no significant influence from perceived severity (β = 0.027, p = 0.349), perceived benefits (β = 0.028, p = 0.342), and cue to action (β = 0.049, p = 0.080) in socially responsible consumption, so these hypotheses are not proven. On the other hand, while hypothesis 3c is tested, although perceived barriers do not significantly influence social identity, they significantly influence socially responsible consumption (β = 0.085, p = 0.006).

For hypothesis tests, 4a, 4b, 4c, and 4d, the mediation effect of social identity (organism) between the relationship of external stimuli and socially responsible consumption (response) was evaluated, see Table 8. Two steps were proposed by Nitzl et al. (2016) with a Bootstrapping analysis to assess effect of PLS-SEM mediation. The first step consisted in determining the significance of the indirect effect and the second step was determining the type of mediation by evaluating the significance of the direct effect. If the indirect and direct effects are significant, there is a partial mediation, and when only the indirect effect is significant, it is a total mediation (Zhao et al., 2010). Therefore, only hypotheses 4a, 4b, and 4d are tested. Social identity significantly mediates the relationship between perceived severity and socially responsible consumption (β = 0.067, p = 0.000). Likewise, social identity significantly mediates the relationship between perceived benefits and socially responsible consumption (β = 0.094, p = 0.000). Moreover, social identity significantly mediates the relationship between cue to action and socially responsible consumption (β = 0.078, p = 0.000). However, hypothesis 4c is not tested, social identity does not mediate the relationship between perceived barriers and socially responsible consumption (β = 0.028, p = 0.166).

The model’s fit was evaluated using the standardized mean square residual (SRMR) values considering values less than 0.08 to confirm a good fit (Benitez et al., 2020). The SRMR value of the model was 0.050, which was less than 0.08, so it is considered a good model fit.

To test hypothesis 5, it was necessary to carry out a multigroup analysis; however, before said analysis, the calculation of the measurement invariance of composite models (MICOM) was carried out to corroborate that the categorical variable causes the changes in the structural model, in this case, the age; this analysis tested the measurement invariances between groups (Hair et al., 2021). The MICOM calculation was integrated into three stages, the confirmation of the configural invariance, the compositional invariance, and the equality of means and variances of the composites (Henseler et al., 2016). First, configural invariance was confirmed since the same indicators and scales were used in the four age groups, the same data treatment, and the same algorithms (Hair et al., 2021). Second, for the confirmation of compositional invariance, a permutation analysis was carried out with 1,000 permutations, and it was confirmed p-value of the correlations of the constructs was greater than 0.05 to guarantee that there are no differences between the composites and thus prove the invariance of the composites (Hair et al., 2021). Table 9 shows the composite invariance of the constructs between the groups that had significant differences between their path coefficients. Third, since all the composites in the groups with significant differences had compositional invariance, equality of means and variances between the composites in the groups was confirmed. It was examined that the value of the difference between means and variances was greater than 0.05 to prove the equality of means and variances and with it full measurement invariance (Henseler et al., 2016). However, in some cases, equality of means and variances was not confirmed, so in some cases, there was only partial measurement invariance; this happens when only the second step is completed. If there is at least partial measurement invariance, it is possible to perform a multigroup analysis (Hair et al., 2021).

A bootstrapping analysis was performed to determine the path coefficients of the relationships of the research model between groups and their significance (Hair et al., 2017, 2021). Table 10 shows these results. Differences in path coefficients between groups were analyzed by multigroup analysis (MGA; Henseler et al., 2009). The results of the multigroup analysis are shown in Table 11. There is a significant difference in the cue to action and social identity relationship in age groups 1 and 2, which correspond to generation Z and generation Y. A significant difference was also found in groups 1 (generation Z) and 3 (generation X) in the relationships between cue to action and social identity and between perceived benefits and socially responsible consumption. Moreover, a significant difference was found in groups 2 (generation Y) and 3 (generation X) in the relationship between cue to action and social identity. Therefore, hypothesis 5 is partially tested because a difference was only found in four relationships between the generational groups.

An alternative conceptual model with an additional exogenous variable was analyzed as Stewart et al. (2010) did in constructing an alternative conceptual model. Like these authors, age was included as an exogenous variable of the model since there were differences in some relationships due to generational change. However, age did not significantly influence on socially responsible consumption and its effect was negative (β = −0.020, p = 0.479) as shown in Table 12. Furthermore, this variable did not affect the degree of explanation of the endogenous variable since the determination coefficient remained the same as in original model (R² = 0.375) as seen in Figures 2, 3.