Theoretical foundation

The intention to use any new technology is a prerequisite for adopting it. Several earlier studies focused on three primary types of determinants, namely technological factors, human factors, and environmental variables. Most of these determinants were drawn from the “unified theory of acceptance and use of technology (UTAUT/UTAUT2)” and “Technology Acceptance Model (TAM)” [15, 19, 22–24]. Despite the tremendous recognition of UTAUT [25], Venkatesh et al. [26] expanded the UTAUT model with the addition of three new factors, namely "price value", "hedonic motivation", and "habit", resulting in the establishment of the UTAUT2 theory. Previous studies revealed that compared to the original UTAUT, the extensions offered in the UTAUT2 resulted in a substantial increase in the explained variance (e.g., 56% to 74%) of the intention to adopt new technologies [27]. Taking all these facts into account, the UTAUT2 model was used in this study to examine the factors influencing the consumers’ intentions to embrace WPD in the context of China.

The majority of earlier research studies [20, 23, 24, 28] found that price value was not a significant predictor of adoption intention in the context of an upgraded technology with contactless payments. Following the aforementioned literature results and judgments, price value was not considered in this study. Customers should acquire high experience in adopting technology to investigate the function of habit [29]. Furthermore, a relatively new technology that has not acquired significant market acceptance may be unable to establish a habit [28, 30]. Notably, many of the participants in this study were potential users who had yet to use WPD or had made attempts to use it. As a result, the researcher decided not to include the price value and habit in the current study model.

UTAUT and UTAUT2 were expanded by several studies [18, 19, 27] through the incorporation of perceived trust as an influencing factor of the intention to adopt contactless payments. According to other earlier studies, lifestyle compatibility was among the most important indicators of the intention to adopt contactless payment technologies [18, 19, 23, 31]. Therefore, this study evaluated the impacts of perceived trust and lifestyle compatibility on the intention to adopt WPD and UTAUT2 model components to explore the intention to adopt WPD among Chinese consumers. The research model is offered in Fig 1.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Research framework.

https://doi.org/10.1371/journal.pone.0273849.g001

Hypothesis development

Population and sample

This research employed a cross-sectional research design to observe the adoption intention of WPD in the context of China. Based on the studies incorporated new technological developments in contactless payments, the younger generation was deemed the ideal population to obtain an accurate image [39], given that youths are the advanced level adopters of new technologies, the internet, and mobile devices [40]. According to the International Labor Organisation [41], individuals aged 18 to 40 are considered a youth and working adults in the formal sector. Therefore, this study focused on youths in China who are 20 to 40 years old, are financially competent and have the decision-making ability to become users of WPD. In this quantitative study, a convenience sampling technique was applied to select the respondents as the sampling method was found to be economical. Thus, respondents could be addressed from any easily reachable part of the population. The minimal sample size in this study was calculated using G-Power Version 3.1 (with an effect size of 0.15, a power of 0.95, and eight predictors), while the sampling size amounted to 74 units. Data were obtained from 298 respondents through an online survey to avoid the drawbacks of a small sample size. Participants were assured that all information was kept confidential and their participation was voluntary. An online survey was conducted by posting the survey form at WJX site (http://www.wjx.cn/) from May 2021 to June 2021.

The local ethics committee (Universiti Malaysia Kelantan, Malaysia) ruled that no formal ethics approval was required in this particular case because this research did not collect any medical information, there was no known risk involved, did not intend to publish anyone’s personal information, and did not collect data from underaged respondents. This study has been performed in accordance with the Declaration of Helsinki. Informed consent for participation was obtained from respondents who participated in the survey. The respondents who participated in the survey online (using google form) were asked to read the ethical statement posted on the top of the form (There is no compensation for responding nor is there any known risk. In order to ensure that all information will remain confidential, please do not include your name. Participation is strictly voluntary, and you may refuse to participate at any time) and proceed only if they agree. No data was collected from anyone under 18 years old.

Measures of constructs

The survey questionnaire was adapted from previously tested and validated surveys, with minor changes made to fulfill this research’s purposes. During the questionnaire development, clear, understandable, and unbiased wordings were used to ensure that the respondents were attracted to the questionnaire and concretely elaborated on their answers and thoughts. Six items were adopted from Gupta and Arora [24] and Rahman et al. [42] to measure performance expectancy. Four items under effort expectancy were derived from Gupta and Arora [24] and Talukder et al. [43]. To estimate the social influence, five items were adopted from Gupta and Arora [24], Rahman et al. [42], and Talukder et al. [43]. Furthermore, facilitating condition was evaluated with five items adopted from Gupta and Arora [24] and Rahman et al. [42], while five items were used to assess hedonic motivation, which was adopted from Gupta and Arora [24], Rahman et al. [42], and Talukder et al. [43]. Five items were derived from Talukder et al. [43] to measure lifestyle compatibility. Moreover, perceiver trust was obtained by adopting five items from Loh et al. [44] and Hussain et al. [23]. Four items were extracted from Gupta and Arora [24] and Rahman et al. [42] to measure the intention to adopt WPD. A five-point Likert scale was used for all items, which ranged from ‘strongly disagree’ to ‘strongly agree’. All items adapted in this study are presented in S1 Appendix. Survey Instrument. The descriptive characteristics of all varieties are presented in Table 1 below.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Measurements.

https://doi.org/10.1371/journal.pone.0273849.t001

Common Method Variance (CMV)

In this study, the single factor accounted for 40.281% (below the recommended threshold of 50%), which approved the inconsequential influence of CMV [45]. CMV evaluated the current study by testing the full collinearity of all constructs [46]. All constructs regressed on the common variable. The recorded values of the variance inflation factor (as presented in Table 2) were less than 3.3, suggesting the absence of bias from the single-source data [46].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Full collinearity test.

https://doi.org/10.1371/journal.pone.0273849.t002

Method of data analysis

Partial least squares structural equation modelling (PLS-SEM) and artificial neural networks (ANN) were employed in the study as a part of the two-staged hybrid analytical approach. The PLS-SEM was used in the first stage to assess the model reliability, validity, and significant associations among the predictors of WPD adoption intention. In the second stage, ANN was applied to determine the significant importance of the factors of the adoption intention of WPD. The two approaches were compared to determine whether any variances were present in the factors used to predict WPD adoption intention. According to Chan and Chong [47], despite the frequent use of PLS-SEM in social and behavioural science to validate the relationship between independent and dependent variables, it is rarely incorporated with other artificial intelligence algorithms. Moreover, SEM could effectively manage complicated models, small samples, and non-normal data [48, 49]. Given that SEM is confined to the evaluation of linear models, the complexity may be oversimplified, making it unpredictable. The ANN method was used to establish non-linear relationships between the predictors in this study to mitigate this drawback. According to Chan and Chong [47], ANN improves understanding of complex non-linear and linear associations among components that affect technology adoption intentions. Compared to the traditional regression methods, ANN improves the accuracy of predictions [50]. The ANN analysis was also conducted for multi-layer perception to create normalised importance.

Data analysis

Demographic characteristics.

Demographic details, including respondents’ age, gender, average monthly income, marital status, education and resident city, are presented in Table 3.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Respondents demographic details.

https://doi.org/10.1371/journal.pone.0273849.t003

Reliability and validity

The construct reliability and validity were evaluated, as shown in Table 4. The minimum value of Cronbach’s alpha value amounted to 0.860, which was higher than the minimum threshold value of 0.7 as per the suggestion by Hair et al. [51]. Overall, the factors were proven reliable. Based on the assessment of internal consistency using Composite Reliability (CR) with the values ranging from 0.898 to 0.966, all values were found to be higher than the minimum threshold value of 0.7 as per Hair et al.’s [51] recommendation. Therefore, the CR test indicated that all of the variables in this study dataset were highly reliable. Fornell and Larcker [52] suggested that convergent validity should be assessed through AVE using a threshold value higher than 0.50. The AVE values for all of the components in this study dataset ranged from 0.639 to 0.850, which was higher than the minimum suggested threshold. This result proved that convergence validity had a positive effect that supported the uni-dimensionality for each factor.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Reliability and validity.

https://doi.org/10.1371/journal.pone.0273849.t004

The Fornell-Larcker criterion, (Table 5) and cross-loading (Table 6) were employed in this study to provide a more thorough understanding of discriminant validity [53]. According to the Fornell-Larcker criterion, the square root of AVE for each construct should be higher than its strongest correlation with other constructs, which maintains the discriminant validity of the measurement model [48]. The Fornell-Larcker Criterion values of all of the components in this research dataset fulfilled the discriminant validity criteria. As a result, no lack of discriminant validity was found. Cross-loading is a technique for comparing the outer loadings of constructs, while previous research demonstrated that all factor loadings must be higher than 0.60 [51, 54]. An indicator’s loading with its associated latent construct should surpass the loadings of all the remaining latent variables in cross-loadings [55]. All the factor loadings in the study dataset are more than 0.5, with all positive values exceeding the threshold value. Additionally, all the factor loadings surpassed the remaining latent variables’ loading values, which proved the strong discriminant validity of the study dataset.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Discriminant validity (Fornell-Larcker Criterion).

https://doi.org/10.1371/journal.pone.0273849.t005

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 6. Loadings and cross-loading.

https://doi.org/10.1371/journal.pone.0273849.t006

Path analysis

The coefficient of determination (r²) reflects the degree of explained variance or the percentage of the variance in outcome variables that could be explained through a linear model. The range of 0.19 to 0.32 is considered weak explanatory power, while 0.33 to 0.66 is considered moderate explanatory power, and 0.67 and above denotes strong explanatory power of r² [55]. The r² value (0.680) indicates that all the predictors could explain a significant portion of the variation in IWPD (e.g., 68%), demonstrating a strong explanatory power. The f² measures the effect size, while the ranges of f² ≥ 0.02, f² ≥ 0.15, and f² ≥ 0.35 represent small, medium, and large effect sizes, respectively [56]. Based on the analysis of the study dataset, f² values of PE, SI, FC, HM, and LC indicated a small effect size of the constructs over IWPD. In contrast, the f² value of PT denoted a relatively higher effect size than other variables.

The path coefficients and p-values for all associations are presented in Table 7. Based on the Table 7, the effect of PE, SI, FC, HM, LC, and PT on IWPD amounted to 0.234, 0.195, 0.250, 0.101, 0,157 and 0.239, respectively. The findings highlighted the positive effect of PE, SI, FC, HM, LC, and PT on IWPD, which was also statistically significant at a 5% level of significance. The EE demonstrated an unprecedented negative effect on IWPD, although it was not statically significant.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 7. Path coefficients.

https://doi.org/10.1371/journal.pone.0273849.t007

Artificial neural network analysis

An artificial Neural Network is a robust and adaptable method. In contrast to other linear methods, this method does not require multivariate assumptions to be fulfilled (e.g., normality, homoscedasticity, linearity, and multicollinearity). Thus, ANN models are considered more precise and extensive compared to all other linear models [57]. At this phase of ANN, the root mean square of error (RMSE) values for both the training and testing datasets were used to determine the relative accuracy of the prediction (see Table 8). The small and close values in the test and training dataset indicate the study model’s high precision and robust predictive power [15]. In this study, the values of RMSE ranged from 0.321 to 0.440 in the training dataset. The values ranging from 0.340 to 0.461 in the testing dataset were found to have small and close values, which established the strong predictive power and relative prediction accuracy.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 8. RMSE values of artificial neural networks (N = 298).

https://doi.org/10.1371/journal.pone.0273849.t008

Typical model sensitivity evaluations include ‘one-at-a-time’ simulations that individually evaluate the impact of each independent variable while overlooking the interactions with other independent variables [58]. In ANN, sensitivity analysis enables the evaluation of input variables in terms of the significance of their impact on the output variable and the identification of factors that could be omitted without compromising network quality and the critical key factors [59]. The mean importance of ANN sensitivity analysis (Table 9) in this study dataset indicated that PT was the most important variable, followed by LC and HM. Fig 2 represents the ANN model, and Fig 3 offers the predicted values by observation.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Hidden layer.

https://doi.org/10.1371/journal.pone.0273849.g002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Predicted by observed.

https://doi.org/10.1371/journal.pone.0273849.g003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 9. Sensitivity analysis.

https://doi.org/10.1371/journal.pone.0273849.t009

Theoretical implications

The main objective of the current study is examining factor affecting WPD adoption behaviour in China. In doing so, this study developed a more sophisticated framework based on a well-known UTAUT2) of Venkatesh et al. [26] to capture contributing factors to the formation of the intention to adopt WPD. Our model capture 68% of IWPD which confirms the validity of the UTAUT2 in context of WPD, and the findings highlighted UTUAT2 factors are important in formation of IWPD. Thus, it extends the literature of contactless payment. Furthermore, as mentioned earlier, wearable payment devices are currently in the early stages of commercialization and there is lack of studies in this context, our study contributes to the model and theoretical development in WPD adoption behaviour. Moreover, unlike previous studies on contactless payment adoption, this study included perceived trust and lifestyle compatibility as influencing factors of IWPD, which were never explored in the context of contactless payment especially wearable device. The significant influences of perceived trust and lifestyle compatibility adds new evidence to the literature in contactless payment.

In addition to SEM-PLS, the current study employed a two-step analytical approach (SEM-ANN) to confirm factors that contribute to the formulation of an IWPD. This ANN validated the excellent predictive accuracy of data fitness by focusing on the impacts of the most important variables. Such empirical estimation is still limited in previous research of contactless payments in China that used UTAUT2. The current study is the first attempt in using an integrated ANN model to observe the predictors of WPD adoption variables at the individual level. In our cases, the dual analytical strategy offered consistent findings and suggested that more efforts be undertaken to extend the UTAUT model to explore technology adoption. The ANN findings suggested that the four prime factors contributing to the formation of an intention to adopt WPDs are PT, FC and PE, respectively. Hence, it offers important contribution to the literature of contactless payment. Finally, the current study explored in the context of the China market thus improving literature of the China market and offering literature to compare with other countries.

Practical implications

The primary goal of this study is to identify and validate a set of attributes that enhance the behavioural intention to use WPD systems in China. The current study offers a few practical implications for product innovators, financial institutions, and shoppers. The current study shows that PE, EE, FC, HM, LC, and PT have a substantial impact on WPD adoption. Such knowledge can assist financial institutions and device manufacturers in developing new plans and marketing methods to persuade WPD adopters. For example, relating to performance expectations, wearable payment device manufacturers and financial institutions should improve the functionalities and features of their devices, where they may focus on ensuring availability, performance, effectiveness, and efficiency, including 24*7 services and customer support. Referring to effort efficiency, WPD should be easy-to-use with an adaptable user interface and a high-quality product experience. The significance of facilitating conditions suggests that WPD manufacturers should include a practical benefit and full instructions to be effectively conveyed to the consumers via marketing and social media to spread information and the advantages of WPD among users.

Furthermore, the significance of lifestyles implies that the adoption and usage of different models of WPD is likely to change in China with changing lifestyles and fashion trends because wearable device are currently deemed as fashion accessory. In market like China, owning a high-end technology device symbolize the high social status [61]. Hence, in an attempt to hold and increase market share in the WPD industry, device manufacturers should architect and develop new products with continuous market assessment where they need to assess how their products suit customers’ lifestyles and current fashion trends. Additionally, device makers must evaluate the needs and lifestyles of customers in various segments in order to design and manufacture the products that best suit those customers. In advertising, WPD manufacturers and service providers should consider the living standards and lifestyle patterns of their target customer segments, where they may hire a fashion icon as their brand ambassador to promote products.

Moreover, trust has always been a critical key factor in financial services and technological products. The relevance of trust and security-related concerns is further highlighted by the discovery that perceived trust is a crucial determinant of the intention to use WPD. Hence, manufacturers and service providers should determine what customers perceive as trust-related risks associated with WPD usage in order to make WPD more trustworthy [15]. Therefore, in the first place, manufacturers should establish the safest possible infrastructure and environment for WPD services. Then, WPD providers should disclose information detailing data privacy rules, end-to-end encryption, biometric authentication capabilities (e.g., fingerprint readers), secured networks, and sensor-based procedures, which make customers feel trustworthy and protected. Additionally, they need to ensure that their WPD is error free in terms of software functioning. On the other hand, financial institutions, as a service provider, have to provide data security, service accessibility, platform security for wearable device payments, and user assurance that they can execute a variety of financial transactions efficiently and safely utilizing their wearable payment devices. Financial institutions may make use of a number of sophisticated encryption and authentication techniques to increase security. All these measures will ultimately increase trust of users towards device and servicer providers.

Finally, the findings would assist the government and key official policymakers in the governing bodies in improving the policies to enhance wide acceptance of WPD and accomplish the goal of developing a cashless society.

Study limitations and future research options

However, three pertinent limitations were found in this study. Firstly, the use of cross-sectional design restricted the controllability of unobserved heterogeneity, preventing a strong foundation for proving causality. Therefore, future longitudinal research was suggested to allow variable arrangements to be designed, created, and measured more efficiently using data acquired over a long period of time. Secondly, while the current findings offered a better match in the use of two more variables (LC and PT), other important aspects such as innovativeness, risk, and brand image were overlooked despite their potential to provide a more thorough explanation of the adoption of WPD. By extending the UTAUT2 in the context of Chinese consumers’ intention to use WPD, this study has offered prospects for academics and researchers to conduct future research that includes these components. Considering that UTAUT2 was used to examine behavioural intention, the study scope was broadened to include other behavioural intention theories (e.g., Diffusion of Innovation, Theory of Planned Behaviour) to study the WPD innovation adoption in China. Lastly, the study focused on young people in China who are financially stable and advanced technology users. Thus, future studies were suggested to examine the model focusing on various demographic populations from a cross-country perspective, which comprises diverse cultural, environmental, economic, and technological circumstances, to explore the behavioural intention of adopting WPD.