The TPB is one of the most mature theories in the field of social psychology (Conner and Armitage, 1998). According to this theory, the most reliable predictor of an individual's behavior is his or her willingness to act, and an individual's willingness to act will be affected by psychological factors such as behavioral attitudes and subjective norms. For college students, their willingness to consume green food will directly determine their actual possibility of purchasing green food. Many studies have successfully used TPB to predict food choices. For example, Dowd and Burke (2013) applied the TPB in their study and found that attitudes and subjective norms accounted for a large portion of the variance in willingness to purchase sustainably sourced food. Kang et al. (2013) research found that behavioral attitudes and subjective norms have a positive impact on young consumers' willingness to purchase sustainable clothing in different countries. Specifically, behavioral attitude refers to an individual's positive or negative evaluation of a certain behavior. Regarding green food consumption, if college students believe that purchasing green food is beneficial, valuable, and consistent with their own values, then they will form a more positive behavioral attitude and be more inclined to have a willingness to purchase green food.

Subjective norms reflect an individual's perception of others' expectations and their motivation to meet these expectations. The consumption intention of college students reflects the emphasis on personal social identity, and is also affected by the online environment and peers, showing the characteristics of pursuing fashion trends and following the herd mentality (Chen, 2021). Subjective norms have a dominant influence on college students' consumption intentions, and consumption attitudes also have a certain influence (Qu et al., 2022). Studies by some scholars have shown that low-carbon awareness, environmental responsibility and consumption habits are the main factors affecting college students' low-carbon consumption intentions (Wang and Yao, 2023). For college students, their educational environment and peer groups may form a social norm of green consumption and have a greater impact on them. If college students perceive that important others such as family, friends, and school want them to buy green food, and they are personally willing to act to meet these expectations, their willingness to consume green food will increase accordingly. Therefore, subjective norms are also a key factor affecting college students' willingness to consume green food.

In summary, according to the TPB, the two key psychological factors, behavioral attitude and subjective norms, will have a significant positive impact on college students' willingness to consume green food. Accordingly, this article puts forward the following hypotheses:

H1: Behavioral attitude has a positive impact on college students' willingness to buy green food.

H2: Subjective norms have a positive impact on college students' willingness to purchase green food.

NAT is a behavioral theory proposed by Schwartz (1977) to explain and predict individual prosocial behavior. When applying NAT to explain a series of pro-social and pro-environmental behavioral intentions, it was found that the relationship between awareness of consequences, attribution of responsibility, individual norms and behavioral intentions is a chain mediation model, that is, perception of consequences will activate individuals through attribution of responsibility. Norms, individual norms lead to individual pro-social and pro-environmental intentions (Steg and De Groot, 2010). Zhou and Zhang (2020) research confirmed the positive relationship between perception of consequences, attribution of responsibility, individual norms and green purchase intention. College students are in a critical period of forming values and cultivating social responsibility awareness, and their behavior is more easily affected by normative factors (Li et al., 2024). When college students clearly perceive the environmental protection effects of purchasing green food, they are more likely to associate these positive consequences with their own responsibilities. In other words, college students will believe that they have the responsibility to contribute to environmental protection through green food consumption. This kind of consequence perception promotes responsibility attribution, which helps to explain the formation mechanism of college students' green food consumption behavior. When college students internalize this kind of responsibility attribution, they are more likely to internalize green food consumption as their own personal norms and moral standards, thereby forming individual norms that support green consumption. Therefore, college students' responsibility attribution will have a positive impact on their individual norms. Secondly, internalized individual norms will eventually be transformed into important motivations for behavioral intentions. For college students, if they have formed personal norms that support green food consumption, their willingness to consume green food will increase accordingly. This is because the personal norms and moral standards that college students internally identify with will directly lead to their willingness to buy green food.

Accordingly, this study proposes these hypotheses for the college student population:

H3: When consuming green food, consequence perception has a positive impact on responsibility attribution.

H4: In the process of green food consumption, perception of consequences has a positive impact on individual norms.

H5: In the process of green food consumption, responsibility attribution has a positive impact on individual norms.

H6: In the consumption and purchase of green food, individual norms have a positive impact on college students' willingness to purchase green food.

The TPB behavior takes the self-interest of individual behavior as the starting point and regards maximizing self-interest as the basic principle of individual behavioral decision-making. However, NAT emphasizes the altruistic tendency existing in individual behavior and uses the individual's sense of moral obligation as the motivation for altruistic behavior (Klöckner and Ohms, 2009). Panda et al. (2020) argues that an increased awareness of sustainability positively influences consumers' altruistic behavior, leading to higher purchase intention, loyalty to green brands, and communication about green brands. Therefore, studying green purchasing behavior only from a self-interested perspective will ignore the impact of individual altruistic motivations on consumers, and will not provide a comprehensive and in-depth understanding of consumers' willingness to purchase green food. According to research by Teisl et al. (2009), an individual's behavioral norms may be affected by subjective norms, which refers to an individual's perception of the opinions and evaluations of his or her behavior by important people and groups. Individuals will use these opinions and views to judge whether their behavior is in line with society's expectations, and thus decide whether it is consistent with their own inner expectations and values. A framework integrating the TPB and NAT can help improve the ability to predict and explain environmental behavior. Maichum et al. (2016) proposed that beliefs, societal norms, and perceived behavioral control strongly influence the inclination to purchase green items. Additionally, it was discovered that consumers' purchase intention was most highly impacted by their attitude. Environmental knowledge has a significant positive impact on attitudes and subjective norms about green products. Behavioral attitude is the consumer's view and evaluation of his or her purchasing behavior, which reflects the consumer's overall understanding of his or her purchasing behavior. Consequence awareness is the consumer's understanding of the impact of behavioral results, which will inevitably affect the attitude toward a specific behavior. As a result, the subsequent hypotheses are formulated for this study:

H7: In the consumption process of green food, the perception of consequences has a positive impact on behavioral attitudes.

H8: Regarding the consumption of green foods, individual norms are positively influenced by subjective norms.

The research paradigm in this article is depicted by the framework diagram displayed (Figure 2).

The article introduces a theoretical framework consisting of six variables, wherein each variable is assessed utilizing Likert scales featuring five levels—“strongly disagree” to “strongly agree.” The measuring questions utilized in this study of green food purchase intention have been extensively adapted from established scales utilized in prior research and are detailed in Table 2. All variables are assessed using these inquiries.

This study gathered data through both online and offline surveys targeting college students across multiple universities in China. The offline survey served as a preliminary study to verify the questionnaire's effectiveness, while the online version aimed to expand the size and diversity of the sample. Out of 420 collected questionnaires, 350 were deemed valid after excluding responses from those who had never purchased green food, as well as incomplete or improperly filled entries, achieving a validity rate of 0.833.

A pilot survey was conducted at a university in Xi'an, Shaanxi province, to assess the appropriateness of the question setup and respondents' comprehension following the initial questionnaire draft. Feedback indicated that while the main indicators and the questionnaire length were appropriate, the excessive use of technical terms led to confusion or vague answers. Consequently, modifications were made to clarify or simplify complex descriptions, enhancing the questionnaire's validity and scientific integrity.

The college student population is often regarded as a microcosm of society, and their economic behaviors are widely representative. A study published in the journal Nature suggests that student samples are appropriate for investigating many social behaviors, as they exhibit similar behavioral patterns to non-student groups in these aspects (Exadaktylos et al., 2013). Additionally, experimental evidence indicates that the neural activities and physiological responses displayed by college students during the food selection process are highly similar to those of other populations, reflecting a universal biological basis (Gómez-Carmona et al., 2021). This provides a typical sample group for studying the physiological mechanisms of consumer behavior. In terms of food preferences, visual attention, and other aspects of consumer decision-making, college student consumers also exhibit representative psychological characteristics in their physiological mechanisms of food behavior (Robinson et al., 2015). These findings collectively emphasize the importance of the college student population in studying consumer behavior, particularly in exploring the physiological and psychological mechanisms underlying food consumption decisions.

To ensure the data possessed extensive representatives, a geographically balanced distribution approach was adopted, thoroughly encompassing East, Central, and West China. The questionnaire was distributed across Zhejiang University, Huazhong University of Science and Technology, Shaanxi Normal University, and Xi'an Jiaotong University. This approach facilitates a thorough exploration of the ways in which cultural and economic contexts in various regions influence the consumption preferences for green food.

The questionnaires were distributed from November to December 2023. Considering the large number of questions in the formal survey questionnaire, in order to ensure the quality and authenticity of the responses, we first set screening questions about purchasing behavior before the questionnaire to exclude participants who did not have relevant consumption experiences. Secondly, in the formal questionnaire, we designed reverse questions and screening questions to identify respondents who did not answer seriously. Among the 350 respondents surveyed, 187 were male, while 163 were female. With regard to the purchasing habits of green foods, 209 respondents indicated that they frequently buy such products. As for expenditure on green foods, 13 respondents reported that they have no spending in this category, whereas another 44 respondents have expenditures on green foods that approach or exceed the majority of their total food spending.

Considering that the research data of this article comes from an online questionnaire survey platform and adopts a self-reported response method, which may lead to common method variance (CMV) in the test results and thus affect the validity of the research results, the sample data is therefore Common method variation is controlled and tested (Harris et al., 2016). In the process of collecting sample data, this study procedurally controlled common method bias by taking measures such as anonymity, topic selection, and some reverse projects. In testing the common method variation of the sample data, this article adopts Harman's single factor test method and uses SPSS25.0 to conduct factor analysis on all scale items. The results indicate that the eigenvalues obtained from unrotated exploratory factor analysis are higher than There are seven factors of one (more than 1), and the greatest factor explains 0.398 of the variance, suggesting minimal common technique.

This article uses AMOS26.0 to conduct confirmatory factor analysis on the measurement model of the model to test the validity of the scale. Scholars unanimously agree that the Chi-square degrees of freedom ratio is a key indicator for assessing the goodness-of-fit of structural equation models. A CMIN/DF value ranging between 1 and 3 is considered indicative of a satisfactory model fit (Wen et al., 2004). An RMSEA value below 0.05 is considered excellent for approximation accuracy. Other fitness indicators (GFI, etc.) preferably >0.9 (Wen et al., 2005). The actual measurement results of the model show that the various fitting indicators of the measurement model are shown in the figure (Table 3). The validation of the green food consumption intention scale for college students revealed an KMO measure of 0.925, exceeding the commonly accepted threshold of 0.7, indicating suitable factor analysis suitability. Furthermore, the Bartlett's test of spherical yielded an approximate chi-square value of 9,096.991, which was statistically significant at p < 0.001, suggesting the appropriateness of performing factor analysis on the correlation matrix. The CMIN/DF of this scale is 1.863, the RMSEA is 0.05, the GFI is 0.973, and the RFI is 0.936. The values all meet the labeling of the fitness test. Therefore, the model fit of confirmatory factor analysis is good and further factor analysis can be conducted.

Each dimension's internal consistency was assessed using the Cronbach's coefficient reliability test method. The Cronbach's coefficient of the total scale and the coefficients of behavioral attitude, subjective norms, consequence perception, responsibility attribution, individual norms, and purchase intention are 0.922, 0.944, 0.936, 0.941, 0.931, 0.928, and 0.935, respectively. The Cronbach's alpha coefficients for the scale employed in this study range from 0.922 to 0.944, which are well above the generally accepted threshold of 0.7, signifying an excellent level of internal consistency among the items in the scale.

The study will analyze the convergent validity (AVE) and composite reliability (CR) of each scale dimension to assess model fit and ensure consistency, as shown in Table 4. This assessment involves a detailed examination of the standardized factor loadings of the items corresponding to their constructs within the structural equation modeling (SEM) context. Following this, the convergent validity and composite reliability criteria for each construct will be determined by employing recognized methods to estimate CR and AVE for the underlying constructs. This meticulous process allows for a comprehensive evaluation of the measurement model's reliability and validity, providing insights into the robustness of the scale in capturing the underlying constructs accurately. The analysis findings from Table 5 demonstrate that the factor loadings of the variables associated with each question surpass the threshold of 0.7. This observation underscores the scale's robustness in capturing the intended constructs effectively, thereby affirming its validity. The consistency and strength of these factor loadings signify the scale's ability to accurately measure the underlying model variables, reinforcing the credibility of the measurement tool in assessing the targeted phenomena. Overall, these results validate the scale design and underscore its capacity to adequately reflect the specified model variables, thereby enhancing the trustworthiness and utility of the measurement instrument in empirical research. The CR values all reached above 0.7, indicating that the question options highlighted the characteristics of the dimensional construct and had good internal consistency. The AVE values for each dimension, as determined by convergent validity testing, varied between 0.75 and 0.802. This indicates that all values surpassed 0.5. This shows that the overall convergent validity is good and the inherent quality of the model presupposed in this article is ideal. Combined with the above data analysis and comprehensive explanation, each dimension of this model has good convergent validity and combined reliability.

The discriminant validity of the scale is further confirmed by comparing the square root of AVE for each latent variable with the absolute value of the correlation coefficient between its row and column in Table 5. The research shows that the AVE values are higher than the correlation coefficients, indicating that the scale efficiently differentiates between the latent variables and displays strong discriminant validity. The disparity between the AVE and correlation coefficients underscores the unique contribution of each latent variable to the overall measurement model, reinforcing the scale's capacity to differentiate between the underlying constructs accurately. These results strengthen the credibility of the scale in capturing distinct dimensions of the phenomenon under study and highlight its efficacy in ensuring the validity and reliability of the measurement instrument in empirical research.

Building upon the structural model, this paper examines the research hypotheses, with the results presented in Table 6. The analysis revealed that the impact of behavioral attitude on purchase intention, with an estimate of 0.314, a standard error of 0.301, and significant p-value, significantly supports the proposed positive influence, thus confirming hypothesis H1. Hypothesis H2 was confirmed by the significant positive effect of subjective norms on purchase intention, as indicated by an estimate of 0.281, a standard error of 0.299, and a significant p-value. Furthermore, the influence of consequence perception on responsibility attribution was significant, with an estimate of 0.425, a standard error of 0.447, and a significant p-value, validating hypothesis H3. The impact of seeing consequences on individual norms was statistically significant, with an estimated value of 0.266, a standard error of 0.267, and a significant p-value, confirming hypothesis H4. The estimated value of responsibility attribution on individual norms is 0.208, the standard error value is 0.198, and the p-value is significant, indicating that responsibility attribution has a significant positive impact on individual norms, verifying the hypothesis H5 path. The positive impact of individual norms on purchase intention was significant, with an estimate of 0.22, a standard error of 0.216, and a significant p-value, endorsing hypothesis H6. The relationship between consequence perception and behavioral attitude was significantly positive, as indicated by an estimate of 0.433, a standard error of 0.446, and a significant p-value, confirming hypothesis H7. Lastly, subjective norms were found to significantly influence individual norms, with an estimate of 0.254, a standard error of 0.276, and a significant p-value, supporting hypothesis H8. The overall structural equation path analysis results of this article are in Figure 3.