Augmenting the technology acceptance model with trust model for the initial adoption of a blockchain-based system

Background In the collaborative business environment, blockchain coupled with smart contract removes the reliance on a central system and offers data integrity which is crucial when the transacting parties rely on the shared data. The acceptance of such blockchain-based systems is necessary for the continued use of the services. Despite many extensive studies evaluating the performance of blockchain-based systems, few have focused on users’ acceptance of real-life applications. Objective The main objective of this research is to evaluate the user acceptance of a real-life blockchain-based system (BBS) by observing various latent variables affecting the development of users’ attitudes and intention to use the system. It also aims to uncover the dimensions and role of trust, security and privacy alongside the primary Technology Acceptance Model (TAM)-based predictors and their causal relationship with the users’ behavior to adopt such BBS. Methods We tested the augmented TAM with Trust Model on a BBS that comprises two subsystems: a Shopping Cart System (SCS), a system oriented towards end-users and a Data Sharing System (DSS), a system oriented towards system administrators. We set research questions and hypotheses, and conducted online surveys by requesting each participant to respond to the questionnaire after using the respective system. The main study comprises two separate sub-studies: the first study was performed on SCS and the second on DSS. Furthermore, each study data comprises initial pre-test and post-test data scores. We analyzed the research model with partial least square structural equation modelling. Results The empirical study validates our research model and supports most of the research hypotheses. Based on our findings, we deduce that TAM-based predictors and trust constructs cannot be applied uniformly to BBS. Depending on the specifics of the BBS, the relationships between perceived trust antecedents and attitudes towards the system might change. For SCS, trust is the strongest determinant of attitudes towards system, while DSS has perceived privacy as the strongest determinant of attitudes towards system. Quality of system shows the strongest total effect on intention to use SCS, while perceived usefulness has the strongest total effect on intention to use DSS. Trust has a positive significant effect on users’ attitudes towards both BSS, while security does not have any significant effect on users’ attitudes toward BBS. In SCS, privacy positively affects trust, but security has no significant effect on trust, whereas, in DSS, both privacy and security have significant effects on trust. In both BBS, trust has a moderating effect on privacy that correlates with attitudes towards BBS, whereas security does not have any mediating role between privacy and attitudes towards BBS. Hence, we recommend that while developing BBS, particular attention should be paid to increasing user trust and perceived privacy.

 To enable companies to increase trust in their products and supply chains.   185 MultiChain as a solution to both on-chain and off-chain data storage, encryption, hashing and 186 tracking of data, together with Ethereum. Ethereum is used for access control and enabling 187 transactions with ethers that allow users to shop online with all the transactions stored in the 188 blockchain and get incentives for permitting to share their data as they specify in the smart 189 contracts. Figure 1 presents the interaction among the customer (data provider) and other e-190 commerce companies/apps (data consumers) of the BBS. The system comprises two subsystems: 191 Shopping Cart System (SCS) and Data Sharing System (DSS). SCS is used in the online 192 shopping cart enterprise. It has a payment mechanism supporting cryptocurrency, ether and 193 manages the mutual agreement between customers and enterprise through smart contracts. SCS 194 automatically registers the immutable timestamped information about the transactions that acts 195 as proof of existence and can be useful to settle any disputes between the stakeholders in the 196 future. Moreover, SCS deploys smart contracts that allow customers to provide their data sharing 197 preferences on a template form without needing them to write the code for the smart contracts. 198 The smart contracts support users in the following ways (Shrestha & Vassileva, 2019b): 238 In classical TAM, the main design constructs such as perceived ease of use and perceived 239 usefulness have shown significant influence on the behavioral intention of the user to adopt the 240 information systems (Davis, 1989), and the latest study by (Shin, 2019) shows the necessity of 241 considering the Trust-Security-Privacy factors in the decision model of the blockchain-based-242 solution adoption. So, we adopted the partial least square structural equation modeling (PLS-243 SEM) analyses on the augmented TAM as it is a useful technique to estimate complex cause-244 effect relationship models with latent variables and we aimed to model the latent constructs 245 under conditions of non-normality and small sample sizes (Kwong & Wong, 2013). 246 247 Many researchers often extend TAM by adding external constructs because classical TAM often 248 does not capture many key factors specific to the context of the technology (Melas et al., 2011). 249 Quality of system (QOS) (Koh, Prybutok, & Ryan, 2010), trust (T) (Wu & Chen, 2005), 250 behavioral control (Bhattacherjee, 2000) are some of the constructs that have been added as 251 influential variables to user acceptance of the information technology and are therefore inevitable 252 for evaluating a novel system, BBS as in this current study. Although in the software engineering 253 domain, security and privacy are regarded as part of QOS, in this study, we have presented 254 perceived security and perceived privacy as separate constructs. (DeLone & McLean, 1992) 255 refers to QOS as the technical details of the system interface and system's quality that produces 256 output response such that the technology attributes singularly or jointly influence user 257 satisfaction. Hence, it is assumed that the QOS affects user satisfaction and that directly or 258 indirectly through PU, affects users' intention to  267 Privacy is defined as the right to be let alone (Warren & Brandeis, 1890). Furthermore, privacy 268 has been considered as the right to prevent the disclosure of personal information to others 269 (Westin, 1968). Later, privacy has been known to be not just unidimensional (Burgoon et al.,270 1989; DeCew, 1997)) as it includes informational privacy along with accessibility privacy, 271 physical privacy and expressive privacy.

272
 Informational privacy -"how, when, and to what extent information about the self will 273 be released to another person" (Burgoon et al., 1989;DeCew, 1997), e.g. the user is 274 asked for too much personal information while using online services.

275
 Accessibility privacy-"acquisition or attempted acquisition of information that involves 276 gaining access to an individual" (DeCew, 1997), e.g. the user's contact (address, phone or 277 email) information might be left in the old system.  (Burgoon et al., 1989) e.g. viewing user screen in an unauthorized way.

280
 Expressive privacy-"protects a realm for expressing one's self-identity or personhood 281 through speech or activity"(DeCew, 1997). It restricts extrinsic social control over 282 choices and improves intrinsic control over self-expression, e.g., user data may be 283 inappropriately forwarded to others. 284 285 (Introna & Pouloudi, 1999) developed a framework of principles for the first time to study 286 privacy concerns while exploring the interrelations of interests and values for various 287 stakeholders. The study has identified that different users have distinct levels of concern about 288 their privacy. (Smith, Milberg, & Burke, 1996) developed a scale for the concern for privacy that 289 measured unidimensional aspects of privacy such as collection, errors, secondary use, and 290 unauthorized access to information factors. (Malhotra, Kim, & Agarwal, 2004) also presented a 291 model to consider multiple aspects of privacy such as identifying attitudes towards the collection 292 of personally identifiable information, control over personal information and awareness of 293 privacy practices of companies gathering personal information. However, all these studies just 294 focused on the informational privacy, so the scales to measure privacy were also based on a 295 unidimensional approach and were not even validated. Furthermore, the issue regarding the 296 benefit to giving up privacy such as offering personalization, enhanced security etc. was not 297 addressed by those studies. 298 299 Hence, to address the multidimensionality of privacy, it is particularly important to consider 300 privacy-related behaviors while studying privacy concerns and user attitudes towards privacy in 301 BBS. The constructs presented in a study by (Buchanan, Paine, Joinson, & Reips, 2007) are 302 validated and considered both privacy concerns and user behavior models. The behavioral items 303 include general caution and technical protection of privacy. Attitudinal item includes privacy 304 concern. The authors found that privacy concern correlates significantly with a general caution, 305 but not significantly with the technical protection factor. Furthermore, perceived privacy, which 306 is the attitudinal privacy or privacy concern undoubtedly plays a critical role in user accepting 307 technologies (Hoffman, Novak, & Peralta, 1999;Poon, 2008)). It sheds light on the possibility of 308 unauthorized use and access to the personal and financial information of the users by the 309 companies that they are intending to use the service of (Dwyer, Hiltz, & Passerini, 2007). 310 311 Perceived Security 312 Perceived security is the degree to which a user believes that the online service has no 313 predisposition to risk (Yenisey, Ozok, & Salvendy, 2005). The protected financial and personal 314 information may get compromised by theft and fraudulent activities leading to vulnerability on 315 the internet. Because of this, a sense of security becomes a major concern for the customers to 316 handout their details on the network (Gefen, 2000 323 Trust is an important contributing factor for users to do a certain task that can make them 324 vulnerable and yet hope the service provider on the other end to fully comply with the set of 325 protocols to complete a transaction (Dwyer et al., 2007) and eventually develop a new 326 relationship (Coppola, Hiltz, & Rotter, 2004;Jarvenpaa & Leidner, 1999;Piccoli & Ives, 2003). 327 In a virtual environment, as the users do not have any control over the outcome of their actions, 328 trust becomes one of the prime factors for them to ground some firm belief in the reliability to 329 engage with the other party (Hoffman et al., 1999). In e-commerce, when information is 330 disclosed, users tend to trust more the service provider (Metzger, 2004) resulting in users being 331 free of doubts and are more likely to engage with the other party (Hoffman et al., 1999). 332 Research has shown that trust has a positive significant impact on attitude and intentions to use 333 systems (Papadopoulou, 2007). With greater trust, users question less the authenticity of online 334 services.

336
The user acceptance behavioral model, as presented by (Rios, Fernandez-Gago, & Lopez, 2017; 337 Shin, 2010) for theoretical social network services, is also useful for conceptualizing the role of 338 perceived security, perceived privacy (privacy concern from attitudinal privacy) on user trust. 339 Their findings revealed that perceived security has a moderating effect on perceived privacy that 340 correlates significantly with trust the user can have on the system. 341 342 Related Work 343 Numerous studies have been conducted to examine the factors that determine the acceptance of 344 information technology in the context of an extended TAM and Trust model. We cover a cross-345 section of those studies that are related to our work.  As online activities such as online shopping generate a plethora of real-time transactions of all 375 kinds of assets and information, they are prone to security and privacy-related risks (Roca et al., 376 2009). A privacy issue mostly occurs with unwarranted access to the users' personal data, but 377 that does not necessarily involve security breaches, which can happen with poor access control 378 mechanisms in the system allowing malicious actors to control the system. However, both 379 breaches are critical issues and they often exist together on the online services where users 380 typically feel hesitant to provide private information over the internet (Hoffman et al., 1999). 381 (Shin, 2010) previously explored the statistical significance of security and privacy in the 382 acceptance of social networking sites. Later, (Shin, 2019) presented the role and dimension of 383 digital trust in the emerging blockchain context, where (Siegel & Sarma, 2019) has argued that it 384 has not been investigated how privacy/security factors affect user's behavioral cognitive process 385 of accepting the blockchain-based systems. This study, in addition to previous TAM validated 386 constructs, explores the users' perception towards the security and privacy aspect of the BBS and 387 their influence on intention to use the BBS by using the moderating effects of trust on attitudes 388 towards system. Besides, the current research aims to answer the following research questions 389 when exploring the relationship between different indicators of the augmented TAM with the 390 trust model:   Perceived privacy positively or negatively affects users' perceived security. 426 H11: Privacy concern positively affects users' behavior on general caution. 427 H12: Privacy concern positively affects users' behavior on technical protection.  The main study comprises two separate sub-studies: the first study was performed on the SCS 434 and the second on the DSS. Furthermore, each study data comprises pre-test and post-test data 435 scores. The pre-test defines the data collected from participants before they use the system, 436 whereas post-test data is collected after participants use the system. 437 The pretest study can be considered as the study associated with the prototype model. Since, the 438 present study follows the previous research work from (Shrestha & Vassileva, 2019a), the 439 pretests for the current study do not include the constructs from classical TAM as they were 440 already evaluated in the previous study. So, the pretests of the current study do not present data 441 for hypotheses H1 -H6. The post-tests for both SCS and DSS do not have behavioral privacy-442 general caution and -technical protection constructs as they are only evaluated once, during the 443 pre-test. So, the post-test data do not test hypotheses H11 -H12.   Shin, 2010Shin, , 2017 and adapted in the context 464 of our research model. The instrument consists of 6 items for perceived ease of use, 6 items for 465 perceived usefulness, 4 items for quality of system, 3 items for perceived enjoyment, 4 items for 466 intention to use, 3 items for perceived security, 9 items for trust, 4 items for attitudinal privacy 467 (perceived privacy), 4 items for behavioral privacy-general caution, 4 items for behavioral 468 privacy-technical protection and 3 items for attitude towards BBS. For our later analysis, we did 469 not consider data related to perceived enjoyment. All the respective items (questions) in the 470 constructs are provided as supplemental files. We measured the responses to the items on a 7-471 scale Likert scale from 1 = strongly disagree to 7 = strongly agree. 472 473 Sample Organizations 474 We recruited participants through the website announcement on the University of 475 Saskatchewan's PAWS homepage and on the social networking site, LinkedIn. Participation 476 was entirely voluntary. The participants had to read and accept the consent form to participate in 477 the study. No real identities and email addresses were collected during the data-gathering phase 478 in the surveys. The consent for participation was obtained via an implied consent form. By 479 completing and submitting the questionnaire, participants' free and informed consent was 480 implied and indicated that they understood the conditions of participation in the study spelled out 481 in the consent form. 482 To contextualize the surveys for SCS, we provided participants at the beginning of the pre-test 483 survey questionnaire (presented as the supplemental file in Article S1) with a video about a brief 484 description of blockchain technology and BBS. The inclusion criteria for the SCS survey was 485 that any individual with knowledge about the internet could participate. After participants 486 completed the pre-test survey, we presented them with another video about using the SCS and 487 hosted a remote session allowing them to use the SCS for fifteen minutes. We did not record but 488 noted down their comments and confusion during their interaction with the system. Thereafter, 489 we presented them with a post-test survey questionnaire (presented as a supplemental file in 490 Article S2) to measure different constructs of our Augmented TAM with Trust model. 491 Similarly, we conducted the pre-test and post-test surveys for the DSS part as well. The post-test 492 survey questionnaire for DSS is presented as a supplemental file in Article S3. Each participant 493 in the DSS survey was also asked to use the DSS remotely for fifteen minutes. The inclusion 494 criteria for the DSS survey was that the participants should be from a technical (computer 495 science or engineering) background because the DSS includes technical aspects that only the 496 software developer or system administrator could understand better. Most of the participants 497 completing DSS surveys also took part in the SCS surveys. 498 499 Participants Demographics 500 A total of 66 participants took part in the SCS study and 53 participated in the DSS study. 501 However, upon cleaning, 63 valid responses for SCS and 50 for DSS were left for the analysis. 502 We used a partial least square nonparametric bootstrapping procedure to test the statistical 503 significance with 5000 subsamples (Hair, J. F., Hult, G. T. M., Ringle, C. M., & Sarstedt, 2013) 504 so that the resampling process would create subsamples with observations randomly drawn from 505 the original set of data. For the study, we based our survey by collecting data from the 506 participants who understood at least something about the blockchain and smart contract 507 technologies after watching the video that we prepared on blockchain technology and BBS. The 508 mean score suggests that for SCS, 79% of participants have basic knowledge and 19% have 509 advanced knowledge of blockchain technology; whereas for DSS, 68% of participants have basic 510 knowledge and 28% have advanced knowledge of blockchain technology.  514 We used SPSS version 26 to process the collected data with descriptive statistics. We analyzed 515 the research model with structural equation modelling using smartPLS (Partial Least Squares). 516 PLS is a well-established technique for estimating path coefficients in structural models and has 517 been widely used in research studies to model latent constructs under conditions of non-518 normality and small to medium sample sizes (Kwong & Wong, 2013). The structural equation 519 model (SEM) as suggested by (Hair, J. F., Hult, G. T. M., Ringle, C. M., & Sarstedt, 2013) 520 includes the testing of the measurement models (exploratory factor analysis, internal consistency, 521 convergent validity, divergent validity, Dillon-Goldstein's rho) and the structural models 522 (regression analysis). We started by fitting the measurement models to the data and later we 523 tested the underlying structural models.   540 We had a 7-scale Likert scale for the responses to the items, so we categorized the scale in terms 541 of percentage value to analyze the average score for each item and overall impression of the 542 constructs. We collected scores for all the items in perceived ease of use, perceived usefulness, 543 quality of system, trust, security, privacy, attitudes, and intention to use constructs of our model. 544 The scores obtained for selected constructs indicate that user perceptions on the benefits of using 545 BBS should be maintained by making improvements to achieve a higher level of score category. 546 The preliminary descriptive statistic of the obtained data is shown in Fig. 4 which informs that 547 the average results of the constructs are above 71.43%, so they qualified for the quite high 548 category (Shrestha & Vassileva, 2019a). The comparatively lower pre-test scores indicate that 549 participants developed confidence and trust towards the overall usefulness, usability, attitudes 550 and intention to use the BBS after they used the SCS and DSS. Furthermore, higher scores for 551 PEOS, PU, QOS for SCS over DSS signify that the participants feel easier to use SCS compared 552 to the participants who participated in the DSS part of the study. However, all the selected 553 constructs in our study provided a significant impression in the context of both BBS.

Manuscript to be reviewed
Computer Science 556 We checked the measurement model with the exploratory factor analysis by testing the 557 convergent validity, reliability of measures and discriminant validity. 558 559 For Exploratory Factor Analysis, we first checked the factor loadings of individual items, as 560 shown in Table 2, to see whether the items in each variable loaded highly on its own construct 561 over the other respective constructs. According to (Chin, Peterson, & Brown, 2008), factor 562 loadings exceeding 0.60 can be considered as significant. In our study, all the indicators in the 563 measurement models had a factor loading of value greater than 0.60 except for Item 4 in the 564 construct Behavioral Privacy-Technical Protection (BP-TP4). Since the square of factor loading 565 is directly translated as item's reliability, the item BP-TP4, "I regularly clear my browser's 566 history" with a very low loading value of 0.39 indicated that its communality value would be 567 only 0.15, and thus should be avoided in the model. Although we used the validated constructs, 568 our exploratory analysis detected that the item BP-TP4 had a weak influence on the Behavioral 569 Privacy construct. 570 571 For the Convergent Validity of each construct measure, we calculated the Average Variance 572 Extracted (AVE) and Composite Reliability (CR) from the factor loading. AVE for each 573 construct should exceed the recommended level of 0.50 so that over 50% of the variances 574 observed in the items were accounted for by the hypothesized constructs, and CR should also be 575 above 0.75 to publish results (Hair, Black, Babin, & Anderson, 2014). In our study, the AVE 576 reported in Table 3 exceeds 0.50 for all the constructs except for Beh Privacy-Technical 577 Protection (BP-TP). However, CR for each construct was above 0.75 (acceptable), confirming 578 that it measures the construct validity of the model. Since the BP-TP had the item BP-TP4 of 579 very low factor loading along with an AVE value of 0.469, it suggests that the factor BP-TP did 580 not bring significant variance for the variables (items/questions) to converge into a single 581 construct which means BP-TP items are a less-than-effective measure of the latent construct. We 582 also justify this with the exceptionally low rho_A value for the construct BP-TP.  Table 3 shows the calculated rho_A value (Dillon-Goldstein's rho) for checking the internal 585 consistency to justify the reliability of each measure. The rho_A evaluates the within-scale 586 consistency of the responses to the items of the measures of constructs and is a better reliability 587 measure than Cronbach's alpha in SEM (Demo, Neiva, Nunes, & Rozzett, 2012). In our study, as 588 recommended, rho_A for each construct was greater than 0.70 except for BP-TP which had a 589 0.28 rho value. Therefore, this also supports our decision of removing the behavioral privacy 590 constructs from the post-tests for both SCS and DSS. We assumed that using the BBS simply 591 does not influence the user's behavioral perception of privacy. So, we were interested to see if 592 there is any significant effect on the attitudinal aspect of privacy.  Table 4. To lean towards discriminant Figure 1 Blockchain-Based System of the current study (BBS)  Manuscript to be reviewed Manuscript to be reviewed Computer Science