*4.1. Measurement Model*

In the PLS-SEM measurement model analysis, the main focus is on testing its reliability and validity. Reliability is the trustworthiness or stability of the values obtained from a test and is an indicator of the consistency of the measurement. The reliability tests were measured using three indicators: Cronbach's alpha, Composite reliability and Eigenvalues (Eig). The Cronbach's alpha indicator was introduced by Cronbach [98] in 1951 to respond to the internal consistency between measurement questions. According to the criteria proposed by Nunnally [99], a value of 0.7 or above for Cronbach's alpha is considered to be of high reliability in general exploratory studies. Composite reliability is another metric to assess the reliability of each measurement model. Composite reliability greater than 0.7 [100] is generally acceptable. As shown in Table 3, the Cronbach's alpha values for the measurement model range from 0.702 to 0.886, and the composite reliability values range from 0.835 to 0.930, all of which are above 0.7, which indicates that the internal consistency reliability of the scale indicators of the measurement model is acceptable. The eigenvalues (Eig) of the correlation matrix are another criterion indicating the appropriateness of the measured variable in reflecting the corresponding latent variable [95,101]. According to Sanchez [95], if the first eigenvalue in the correlation matrix is higher than 1 and the second eigenvalue is lower than 1, each group of indicators is in a unidimensional space. The measured variables in the measurement model all reflect the latent variables well. According to the results in Table 3, the first eigenvalues of the measurement models ranged from 1.886 to 2.448, and the second eigenvalues ranged from 0.419 to 0.686, indicating that the measured variables in all eight measurement models were a good reflection of these relevant latent variables.

Validity assessments focus on the measurement model's convergent validity and discriminant validity. As suggested by Anderson and Gerbing [96], convergent validity can be tested with validating factor analysis to determine whether each question item converges to the variable to be measured. According to Rumanti et al. [102], a measured variable is considered to have considerable explanatory power for a latent variable if the standardized factor loadings of the measured variable all exceed 0.7 or more. Furthermore, according to Ringle et al. [103] and Sanchez [95], the average variance extracted (AVE) value for all constructs needs to be above a threshold of 0.5 to satisfy the criterion of convergent validity of the latent variable. As shown in Table 3, the loadings for all measured variables ranged from 0.745 to 0.946, indicating a high degree of correlation between all measured variables and their associated latent variables; the average variance extracted for all constructs ranged from 0.628 to 0.816. For example, the AVE value for AT is 0.816, which means that AT1, AT2 and AT3 explain 81.6% of the variance in the AT variable. Therefore, the convergent validity of this measurement model is acceptable.


**Table 3.** Reliability and convergent validity test of the measurement model.

As suggested by Hair et al. [93], discriminant validity was mainly assessed through cross-loadings and the Fornell–Larcker criterion. Cross-loadings refer to the contribution of a question item to other latent variables, and discriminant validity is considered acceptable when the indicator's loadings on the relevant constructs are more significant than all its loadings on the other constructs. As shown in Table 4, using the PC to construct an example, the loading values of PC1, PC2 and PC3 with the latent variable positive context (CC) were 0.849, 0.782 and 0.785, which were all greater than 0.5 and significantly exceeded the cross-loadings values with other latent variables such as negative context (NC) and perceived behavioral control (PBC). This indicates that the model has good discriminant validity among the latent variables. Meanwhile, according to the Fornell–Larcker criterion [104], when the square root of the AVE of each construct is greater than the correlation coefficient between the latent variable and the other variables, the discriminant validity of the measurement model distinction is acceptable. In this model, the square root of AVE on the main diagonal of Table 5 is much higher than the non-diagonal values. Therefore, the discriminant validity between the latent variables in this model is good.


**Table 4.** Cross-loadings values for each block of indicators.

**Table 5.** Discriminant validity matrix (Fornell–Larcker criterion).


Note: Values (bold) on the diagonal represent the square root of the AVE while the off-diagonals are correlations.

#### *4.2. Structural Model*

In the structural model analysis section, first, the validity of the structural model was assessed using *R*<sup>2</sup> (predicted effect value) and *Q*<sup>2</sup> (predicted correlation) [96]. In terms of the overall model fit, the *R<sup>2</sup>* of the general model was 0.731, indicating that the latent variables explained 73.1% of consumers' sustainable consumption behavior, which was greater than 50%, proving that the model assumptions were reasonable and the model fit was good. The *Q*<sup>2</sup> values for the four endogenous latent variables ranged from 0.084 to 0.157, which met the criterion of >0, indicating that the structural model was valid.

Furthermore, according to the Goodness of Fit (GoF) formula (*GoF* = *communality* × *R*<sup>2</sup> ) proposed by Tenenhaus et al. [105], this metric is used to indicate the degree of fit between the simulation results and the actual measurements. Studies have shown that GoF values above 0.26 are considered to have good applicability in areas such as social and behavioral sciences [106]. The GoF value for this model was calculated to be 0.356, indicating a good model fit.

In order to assess the coefficients and significance of each path proposed in the research model, the paths were recalculated after 5000 replicate samples were taken based

on the Bootstrapping method. The model validation results are shown in Table 6 and parameter paths of the hypotheses in the model are shown in Figure 3. Overall, 12 of the 17 hypotheses for direct effects were supported. There was a significant positive effect of AT on PUR (β = 0.312, *p* < 0.000), TRAN (β = 0.262, *p* < 0.000) and REC (β = 0.312, *p* < 0.001). Hypotheses H1a, H1b and H1c were all supported. SN has a significant positive effect on both PUR (β = 0.095, *p* < 0.017) and TRAN (β = 0.160, *p* < 0.000) but not on REC (β = −0.012, *p* < 0.793). Hypotheses H2a and H2b were supported, but H2c was not supported. PBC also had a significant positive effect on both PUR (β = 0.170, *p* < 0.001) and TRAN (β = 0.233, *p* < 0.000) and no significant effect on REC (β = 0.068, *p* < 0.157). Hypotheses H3a and H3b were supported but H3c was not. PC had a significant positive effect on PUR (β = 0.162, *p* < 0.001), REC (β = 0.159, *p* < 0.001) and AT (β = 0.226, *p* < 0.000), but no significant effect on TRAN (β = 0.057, *p* < 0.268). NC had a significant positive effect on REC (β = −0.237, *p* < 0.000) and AT (β = −0.212, *p* < 0.000) had a significant negative effect, but no significant effect on PUR (β = −0.040, *p* < 0.347) and TRAN (β = 0.083, *p* < 0.095). Therefore, H4a, H4c, H5c, H6 and H7 were all supported and H4b, H5a and H5b were not supported.

**Table 6.** Hypothesis testing of the structural model.


**Figure 3.** Parameter path of the structural equation model standardized path coefficient estimates (\* *p* < 0.05, \*\* *p* < 0.01, \*\*\* *p* < 0.001).

The mediation effect (indirect effect) was also calculated using the Bootstrapping method after 5000 replicate samples. As shown in Table 7, the results demonstrated that the mediation effect of PC has a significant positive impact on the relationships between AT and PUR (β = 0.071, *p* < 0.001), TRAN (β = 0.059, *p* < 0.003) and REC (β = 0.036, *p* < 0.012). NC has a significant negative impact on the relationships between AT and PUR (β = −0.066, *p* < 0.000), TRAN (β = −0.055, *p* < 0.000) and REC (β = −0.034, *p* < 0.008). Thus, all six hypotheses regarding mediation effects (H8a, H8b, H8c, H9a, H9b, H9c) were supported.


**Table 7.** The path coefficient result of mediating effect model.

#### **5. Discussion**

This study divides sustainable consumption behavior into three sectors: green purchase behavior, green transportation behavior and recycling and resource conservation behavior. The influence paths of internal motivations and external contexts are investigated on different sustainable consumption behaviors. Meanwhile, the validity of the TPB–ABC integrated model as a research model to explain consumers' sustainable consumption behavior is confirmed, which is in line with the conclusion of previous studies [31,32,34].

As mentioned earlier, these three sectors of sustainable consumption behaviors are all positively or negatively influenced by both internal motivation and external contexts, but the paths of influence are different. For green purchase behavior, SN, AT and PBC of internal motivation positively influence behavior, with AT considered the most critical determinant (β = 0.312). This indicates that consumers' subjective preferences for green purchase behavior directly influence their implementation of this behavior. This is in line with Tan's study [107]. PC in external contexts can directly contribute to the formation of green purchase behavior. However, the effect of NC on consumers' green purchase behavior was not statistically significant. Therefore, it can be inferred that advertising, government regulations and financial incentives are positively associated with green purchase behavior. Still, high cost, time consumption and lack of infrastructure are indirect constraints on green purchase behavior. The results also suggest that attitude plays a mediating role in the relationship between external context and behavior. In other words, PC indirectly motivates green purchase behavior by promoting the formation of consumer attitudes towards sustainable consumption. Although NC does not directly limit green purchase behavior, the attitude is susceptible to the negative influence of NC, thus limiting the occurrence of green purchases. A possible explanation is suggested by the low-cost hypothesis [108]. The smaller the perceived negative context in which the behavior is engaged in a given situation, the greater the likelihood that attitudes will be transformed into actual behavior.

The second section of sustainable consumption in this study is green transportation behavior. According to the measurement results, external contexts do not directly influence green transportation behavior, unlike green purchases. Still, PC and NC can have a modestly positive and negative influence on green travel, respectively, with attitude

as a mediator. This is consistent with behavioral reasoning theory (BRT) findings that consumers use positive or negative psychological processes or paths to make consumption decisions [18]. The reasonable reason is that the advertising, policy regulation and economic incentives for green transportation in China are developing reasonably with a better social atmosphere forming. At the same time, public transportation in China is cheap and timesaving, so the constraints on green transportation might not come from the external context. Therefore, the indirect effect of the external context on green transportation is mainly caused by subjective attitudes. In addition to AT, the other two internal motivations, SN and PBC, also positively impact green transportation.

The significant effects of PC (β = 0.159) and NC (β = −0.237) on recycling and resource conservation behavior suggest that, on the one hand, the policy context, advertising and economic incentives of recycling and resource conservation directly promote consumers to perform this behavior in China. On the other hand, waste separation facilities are not complete, and consumers still have to overcome many constraints, such as time and effort, when implementing waste separation. Unlike the first two sections, neither SN nor PBC significantly affects recycling and resource conservation among internal motivations. The above findings are corroborated by the relevant studies of Meng et al. [109].

#### **6. Conclusions**

Promoting the transformation of consumers' consumption patterns is a crucial breakthrough in promoting sustainable development. Unlike previous studies, this study constructs a theoretical framework that includes both internal motivations and external contexts to explain the decision-making mechanism of consumers' sustainable consumption behaviors. The conclusions show that, in general, attitudes significantly influence the implementation of sustainable consumption behaviors, while attitudes are positively and negatively affected by positive and negative contexts, respectively. The strength of the effects of each influencing factor on different sectors of sustainable consumption behaviors varied slightly. For green purchase and green transportation behaviors, the impact of internal motivations is higher than the external contexts' and becomes the most important influencing factor. In contrast, consumers' recycling and resource conservation behaviors are more influenced by external contexts. In addition, attitudes are partially mediated between external contexts and sustainable consumption behaviors.

The theoretical implications of this study are reflected as follows. First, two critical variables influencing sustainable consumption behavior, internal motivations and external contexts, are verified. Consumers' sustainable consumption behavior can be effectively motivated by internal motivations; at the same time, when consumers perceive that external contexts have positive or negative effects on their behavior, they will increase or decrease their sustainable consumption behavior accordingly. Second, this study extends the theory of planned behavior (TPB) and Attitude-Behavior-Context (ABC) theory to sustainable consumption. By developing to the micro and macro levels and systematically exploring the factors influencing sustainable consumption behavior, the theories become more effective in explaining the implementation process of consumers' sustainable consumption behavioral decision-making mechanisms.

The findings of this study have the following implications for urban authorities to develop measures to motivate citizens to implement sustainable consumption behaviors. Firstly, relevant authorities should work to reduce or eliminate negative contexts that prevent consumers from participating in sustainable consumption behaviors. For example, relevant authorities should strive to improve the appropriate infrastructure, such as increasing the number of sorting bins recycling service staff to make it more convenient for consumers to participate in sustainable consumption behavior. Secondly, the government and companies should provide incentives or promotional strategies to reduce the high costs of sustainable consumption, such as reducing taxes or lowering the prices of environmentally friendly products. Finally, the internal drivers of consumers cannot be ignored, as some negative contexts for sustainable consumption behavior are long-standing

and objective. Therefore, in such a realistic situation, policymakers should take some effective measures to foster consumers' attitudes toward sustainable consumption, such as strengthening publicity and education to enhance their level of perceived behavioral control. By improving sustainable consumption attitudes, the adverse effects of negative situations that are difficult to overcome in the short term can be reduced. An internal driving mechanism for sustainable consumption can be formed over a long time.

The limitations of the study are as follows. Firstly, because of the limited number of influencing factors identified in this study, the behavioral decision-making mechanism established can only partially explain the occurrence of sustainable consumption behavior, which can be discussed in more depth in future studies. Secondly, the research methodology also needs improvement. Although the sample was randomly selected from the residents of Dongying and the findings are close to the actual situation, as a study of Chinese consumers, this study's sample size and representativeness are still limited. Thirdly, due to the difficulty of obtaining actual observed data on individual sustainable consumption behavior, this study assesses consumers' sustainable consumption behavior based on the self-report measure commonly used in the previous literature. Thus, it cannot effectively avoid the measurement error caused by the inconsistency between the subjective reports of respondents' behavior and their actual objective behavior.

**Author Contributions:** B.Q. and G.S. contributed equally to the article. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Informed consent was obtained from all subjects involved in this study.

**Data Availability Statement:** Not applicable.

**Conflicts of Interest:** The authors declare no conflict of interest.
