**3. Material and Method**

#### *3.1. Questionnaire Design*

Based on the hypothesis model, a questionnaire on the factors influencing sustainable consumption behavior was designed using focus group discussions, on-site collection, expert evaluation and representative interviews. The pre-survey was first conducted in a street (subdistrict) in Dongying and a university in Beijing. The questionnaire was adjusted and improved for the formal survey regarding the pre-survey results to accurately and fully reflect the subjective views and practices of the respondents. The formal questionnaire contains eight latent variables, including positive context (PC), negative context (NC), attitude (AT), subjective norm (SN), perceived behavioral control (PBC), green purchase behavior (PUR), green transportation behavior (TRAN) and recycling and resource conservation behavior (REC), with each latent variable consisting of three measurement items. The measurement items were adapted from existing literature to ensure the scale's content validity. The specific items and their sources are shown in Table 1. All items were measured using a three-level scale (1 = disagree, 2 = neither agree nor disagree, 3 = agree). In addition to the 24 items above, individual characteristics such as gender, age, education level and monthly income were also included in the questionnaire, which consisted of a total of 32 items.

**Table 1.** Measurement item design of latent variables and observed variables.


#### *3.2. Data Collection*

Dongying has a good policy environment and facility base for sustainable consumption and has committed to actively practicing sustainable development since joining the PAGE program of the UN in 2016. As a result, it was selected as a study area. In January 2018, our research team conducted a formal field survey among permanent residents in Dongying. The data was collected through a questionnaire survey. The formal survey adopted the Probability Proportionate to Size (PPS) Sampling method to select the sample. The specific sampling process was as follows. PPS was used to identify each district (two districts based on each district's share of the city's population), street (three per sample district, based on each street's share of each district's population) and community (two per sample street, based on each community's share of each street's population) drawn from the city. Communities were drawn using a random number table, resulting in the selection of 12 communities in the city. At the community level, systematic sampling was used to select household samples (sample size of no more than 45 per community, but calculated with a sample size of 55 in the systematic sampling process in case of blanking) and selected the members in the selected households whose birthday were closest to June 30th to answer the questionnaire face-to-face. A total of 586 questionnaires were collected, of which 552 were valid, with a return rate of 94.36%. After removing the missing values, a new database comprising 534 samples was obtained. The measurement software used in the data collation process for this study was R (version 4.2.0).

Four demographic variables commonly used in behavioral research—gender, age, education level and monthly income—were selected as sample characteristics (Table 2). The ratio of male to female respondents in the sample was approximately 4 to 6 (216:318). The age structure of the respondents shows a slightly aging trend, with 28.46% of the sample aged 55 and above, including 14.79% aged 65 and above, with most of the sample concentrated between 35 and 54 years old, with 28.84% aged 35 to 44 and 24.16% aged 45 to 54. 18.54% of the sample was under 34 years old. The age distribution of the sample is generally consistent with the distribution in Dongying. 17.23% of respondents had middle-high education or below (20%), more than one-third (38.58%) attended high school education, and 4 out of 10 respondents held college-level (21%) or university-level education (20%). A total of 0.94 percent attended post-graduate education or higher. In terms of monthly income, nearly three-quarters (74.16%) of the respondents had a monthly income between RMB 3000–8000, with only 2.43% of respondents earning more than RMB 10,000 per month. The overall distribution of education and monthly income is generally consistent with the actual situation of Dongying residents.


**Table 2.** Description of sample structure characteristic.

#### *3.3. Methodology*

Partial Least Squares Structural Equation Modeling (PLS-SEM) [93–95] was adopted in this study, which does not require the data to obey a multivariate normal distribution and has significant advantages in dealing with complex models with a large number of explanatory variables and multiple correlations between variables [93]. Compared to covariance-based structural equations modeling (CB-SEM), PLS-SEM offers considerable convenience and flexibility in forecasting and is more useful in practical fields where application and practical forecasting control are valued. Since the sample data obtained from this investigation does not strictly obey a normal distribution and many variables in the study model and the relative complexity of the relationships involved, this study uses PLS-SEM for an exploratory study.
