3.1. Questionnaire Design and Data Collection
Data for this study were collected using an online questionnaire survey in consideration of cost and data availability. On the basis of theoretical model and hypothesis framework, we designed the questionnaire. All constructs were measured with reference to previous research on green consumption behavior and acceptance of technology. To measure the items of TAM, we adapted from the research of Choi and Totten [22
], and the items of TPB refer to Kim and Han [23
] and Chan and Lau [24
] (see Appendix
Except for the demographic variables, every item was used a 5-point Likert scale to match the degree to which respondents agreed with the items shown to them, ranging from 1, “strongly disagree,” to 5, “strongly agree.” In addition, we first asked the participants about their gender, age, educational level, and annual household income, and they were asked whether they knew the meaning and details of the China Energy Label.
] pointed out that the sample size needs to be more than five times the number of measurement items. Furthermore, in order to get better results, the sample size should be at least 10 times larger than the number of measurement items. Hence, since there were 25 measurement items, the sample size should be at least 125 and preferably more than 250. In this study, 331 people responded to the questionnaire. After removing unqualified or incomplete responses, the final sample was decreased to 280.
By the professional questionnaire platform Survey Star, we conducted the questionnaire survey, thus all the respondents have already registered on Survey Star. Generally speaking, the demographic characteristics of the respondents, such as gender and geographical location, roughly showed consistency with actual population profile of Chinese residents. Although there were some differences in age, educational level and annual household income from the demographic profile of Chinese residents, they were similar to the demographic data of Survey Star members [26
]. Compared with Chinese residents, members of Survey Stars have the following traits: younger, richer and more educated [26
], and internet users in China are mainly younger people. Hence, the sample of this survey had a better representativeness.
The respondents came from 20 provincial level administrative regions, covering all 7 geographical zones in China. Therefore, they represented almost the entire nation and varied hugely by background, profession, and income. Table 1
showed the demographic profile of the sample. The respondents mainly covered various consumer groups, which could be used for further analysis.
As shown in Table 1
, we obtained answers from 192 women and 88 men. Most of the respondents were between 18 and 25 years old, and about 55% had a bachelor’s degree. The annual household income was concentrated around 100,000 RMB (1 USD = 6.71 RMB).
The ratio of male to female in China is about 104:100, and women are the main household appliance shoppers. According to data from Suning, one of the three major e-commerce platforms in China, women accounted for 62% of users who ordered household appliances during the Double Eleven period (internet shopping festival).
Among the survey respondents, the youngest was 17 years old and the oldest was 38 years old. According to the Statistical Report on the Development of Internet in China issued by the China Internet Information Center, by 2016, 87.4% of Internet users were 10 to 49 years of age. Among them, people born in the 1990s accounted for 47%, the largest group of netizens, and it was predicted that over the next few years users born in the 2000s will grow substantially. As we can see, internet users in China are relatively young as a whole.
In addition, in 2010, China carried out the sixth census and the results suggested that about 40% of the population was between 15 and 35. According to the report “Accenture’s China Consumer Insight 2018: New Consumer Power,” it was supposed that by 2017 there were about 700 million people between 18–35 years of age, accounting for over 50% of the total population. China reached a fertility peak in the 1990s, thus when the younger generation starts to work and becomes of marriageable age, the demand for household appliances will increase. In 2013, Data of Consumption of China showed that young people who were born in the 1990s were becoming the mainstream consumer group in the household appliances market. Data from Baidu, Chinese top search engine, showed that post-90s consumers have become the main consumer force in the household appliances market and the target audience of household appliance manufacturers.
The majority of questionnaire respondents were aged from 18 to 25 years old and most of them were women, who make up the main force of the Chinese household appliances market. On this account, these respondents were representative and their intention to purchase energy-efficient household appliances is more in line with China’s national conditions and has more practical significance. Paying attention to the consumption orientation of the younger generation can help in understanding the direction of promoting energy-saving household appliances in the future.
3.2. Measurement Model
In this study, we used confirmatory factor analysis (CFA) to test whether the relationship between a construct and the corresponding measurement items was consistent with the theoretical relationship designed by the researcher. To make sure that the effectiveness of the study is reasonable, we tested the reliability and construct validity. Reliability represents the consistency and stability of items under the same structure. Convergent validity and discriminant validity are subsets of construct validity. Convergent validity reflects the relationship of items under the same structure; discriminant validity shows significant differences between the characteristics of more than one construct.
According to the initial CFA results, the value of factor loading for item PEU1 was lower than 0.4, so we removed this indicator [27
]. Then, with the help of SPSS v20.0 (IBM Corporation, New York, USA) statistical software, we tested the remaining items to determine the reliability of constructs. The Cronbach’s alpha value for the six variables was 0.838, 0.721, 0.887, 0.897, 0.829, and 0.771, respectively, indicating that these constructs had good reliability [28
]. Furthermore, the composite reliability (CR) of all constructs was greater than 0.7, meaning the questionnaire had good reliability [29
As we can see from Table 2
, all the factor loadings were higher than the standard o.4, and the average variance extracted (AVE) of each construct was also higher than the standard 0.5, which indicated good convergent validity for the questionnaire according to Fornell and Larcker [30
] Also, as Table 3
showed, the correlations between constructs of the model were less than the AVE square root values, which meant that the discriminant validity of all constructs was quite good [31
3.3. Structural Equation Model
The hypotheses were tested by means of structural equation modeling (SEM) with AMOS v22.0 (IBM Corporation, New York, USA). By using SEM, the unobservable variables can be explicated through some observable variables. Regarding sample size of SEM, Jackson [32
] thought that the N:q rule could be used to roughly determine the number of samples required, where N is the number of samples and p is the parameter that needs to be estimated in the model. The recommended proportion is 20:1, and it can be relaxed to 10:1. Bentler end Chou [33
] pointed out that the sample number and estimated parameters must have at least a 5:1 ratio to ensure that the estimated value of the parameter is credible, and the ratio should be at least 10:1 to ensure the validity of the significance test. Therefore, two studies recommended that 10:1 is the proportion to ensure good comparison. In a paper using SEM as an analytical method, Barrett [34
] suggested that a sample volume of more than 200 was more appropriate, although this was not an absolute criterion.
Scuotto et al. [35
] surveyed 175 companies using SEM to determine the key factors in their preference for informal inbound open innovation (OI). To explore the relationships among management, production and achievement of company, Yang et al. [36
] surveyed 309 manufacturing enterprises and used structural equation modeling for analysis. Here, with the aid of AMOS, we established a structural equation model based on TAM and TPB to do research on 280 respondents.
After calculating the survey data of the model, AMOS output the results, and we found that some fitting indicators did not reach the criterion, which meant that the goodness of fit was not high and the model needed to be modified. Hence in light of suggestions from AMOS, we modified the initial model so that it would have a better fitting degree. Goodness of fit indices are as follows: X2/df = 2.079, GFI = 0.891, NFI = 0.897, CFI = 0.943, IFI = 0.944, TLI = 0.924, RMSEA = 0.062 (GFI—goodness-of-fit index; NFI—normed fit index; CFI—comparative fit index; IFI—incremental fit index; TLI—Tucker-Lewis index; RMSEA—root-mean-square error of approximation). Most of the indicators meet the criterion, which means that the survey data match the theoretical framework well.
The results of hypothesis testing are listed in Table 4
. As we can see from Table 4
, the relationship between perceived usefulness and intention is not significant. In addition, the results of other relationships are all significant, that is, these hypotheses passed the test. In short, hypotheses H1, H2, H3, H5, H6, and H7 are supported and H4 is not supported.