TPB vs. DTPB

The two models share the point of similarity that both are derived from the TRA and aims to predict an individual’s planned and deliberated behaviour. On the contrary, the two models are different in the way that TPB is a direct successor of TRA with the assertion that BU is a direct function of BI and PBC, and that BI is formed by one’s A, SN and PBC;

whereas, with DTPB, the direct successor to TPB decomposes the beliefs presented in TPB from the literature of DOI theory that are generalisable across situations and not specialised to a specific context. In addition to the context generality, another purpose of decomposing the beliefs within TPB was its lack of explanatory power in BI during the applications’ design and implementation process (Taylor & Todd, 1995a). The decomposition of beliefs was also supported by Shimp & Kavas (1984) who suggested that cognitive components of beliefs could not be organised into a single conceptual or cognitive unit, therefore needs to be fragmented according to the context of the study.

According to the published literature, within a specific context TPB is more advantageous over DTPB in explanatory power (Taylor & Todd, 1995a). In addition, due to its parsimonious structure TPB is easier to implement compared with DTPB; however, the reverse is not true (ibid). Usually a parsimonious model becomes desirable when the explanatory power of one’s intention is required (Venkatesh et al., 2003). Therefore, within a general context and situations that explore in-depth knowledge, DPTB is more advantageous over TPB.

For taking decisions to select an appropriate model, there is a scarce amount of empirical studies, out of which a few are discussed to clarify differences. The first pioneering study was Taylor & Todd (1995a), which compared three models, the TAM, TPB and DTPB, within a sample of business students. From the explanatory power perspective, the authors found that DTPB was slightly better than TPB, in that DTPB explained 60% variance in BI

68

and 76% variance in A; whereas TPB explained 57% variance in BI and 58% within A. In addition, the author concluded that DTPB provided greater diagnostic values, greater insights into the factors that influence IT usage, and suggested beliefs that can be targeted by designers or managers interested in implementation of information system (ibid).

Later on, similar results have been found by a few more studies recently. For instance, Shish & Fang (2004), in the context of Internet banking in Taiwan, found that DTPB explained 66% variance in BI and TPB explained 54%; Lin (2007), examining online shopping acceptance within the context of Taiwanese customers, found that DTPB explained 57% variance in BI and TPB explained 46%; finally, Huh et al., (2009), examining information system acceptance within South Korean hotels, found that DTPB explained 63% variance in BI and TPB explained 59%. Despite the higher explanatory power within BI in all exemplified studies, it was noticed that the path between SN and BI in all results pertaining to the DTPB was insignificant, which suggests that inclusion of SN only increased model complexity rather than explanatory power. After observing the examples above, it is inferred overall that, with the exception of the parsimonious structure condition, DTPB, compared to the TPB, is more favoured and advantageous to explain and identify the salient beliefs of the individual establishing acceptance behaviour towards new technologies.

2.11.3. TAM vs. TPB vs. DTPB

In the previous two sections (2.13.1 & 2.13.2) three models theoretically rooted in social psychological theory SCT, and derivatives of TRA were compared i.e., (i.e., TAM vs. TPB and TPB vs. DTPB). From the parsimonious perspective it was noticed that the TAM compared to TPB, and TPB compared to DTPB were better; however, from the perspective of explaining BI, the TAM compared with TPB was still better, but TPB was less effective than DTPB. Therefore, in line with the discussion in previous sections it can be inferred that at some extent both the TAM and DTPB were more advantageous than the TPB. In addition to the previous discussion, this section further examines the various issues which are thought to be making a distinction among these models and helps to favour one over another.

As discussed, the three models share points of similarity in that they all are derived from TRA and postulate that individual differences are based on influence of A, BI and BU only via the mediating construct of beliefs. However, later on, A was excluded from TAM but is still part of TPB and DTPB. On the contrary, three models differ from each other in that

69

TAM does not include SN and PBC as determinants of BI. Exclusion of the SN within the TAM at some extent is rational due to weak psychometric properties (e.g., Davis et al., 1989) and mixed results in literature i.e., significant and insignificant (e.g., Taylor & Todd, 1995a; Lin, 2007). However, omission of PBC is considered to be a major limitation (e.g., Mathieson, 1991; Venkatesh et al., 2003). Despite the fact that on one side omission of control beliefs within the TAM constrains its applicability to examine BI in situation when individuals have low level of volitional control (e.g., mandatory situations), on the other hand this is considered to be the favoured point because of its context-free generalisibility and parsimony in structure. On the other side, inclusion of the SN and PBC within TPB and DTPB makes them more favoured compared with the TAM to some extent due to applicability within mandatory situations and explaining greater insights into factors of individuals’ behaviour acceptance. At the same time this inclusion leads to them becoming more context-specific and less in parsimonious structure (see discussion of TAM vs. TPB).

Specifically, inclusion of SN due to unstable results remains a topic of debate and confusion in IT acceptance and favours TAM conceptualisation. For example, specifically looking at the DTPB, which is favoured over the TAM due to its higher explanatory power (Taylor & Todd, 1995a), some researchers found a significant effect of SN over BI (e.g., Taylor & Todd, 1995a; Huh et al., 2009) while others found no effect (e.g., Chau & Hu, 2002; Shish & Fang, 2004; Lin, 2007).

Another point which favours the TAM over TPB and DTPB is exclusions of A which showed partial mediated effect at the time of creation (e.g., Davis et al., 1989). This exclusion made the TAM more parsimonious, and avoided the possibility of a mediating impact between behavioural beliefs and behaviour itself. However, on the other hand, TPB and DTPB still examine behaviour through the mediating effect of BI between A and BU;

this might invite the risk of distorted results in situations where intentions are ill-formed (i.e., partial or no mediation effect) (e.g., Bagozzi & Yi, 1989). In favour of omission of A, Chau & Hu (2001) recently excluded A from DTPB and found a significant impact of COMP on PU and PEOU. This result favours our argument that inclusion of A is a limitation rather than an advantage of TPB and DTPB models.

Apart from the above rationales, the limitations of the TAM, which favours TPB and specifically DTPB, include: its self-reported usage instrument (e.g., Davis, 1993);

construct BU not being examined with actual measures of usage behaviour (e.g., Matheision, 1991); inability to explain the explanatory variance more than 30% to 40%

(Venkatesh et al., 2003; Taylor & Todd, 1995a); and finally, lack of external variables

70

examination (e.g., Igbaria and Iivari, 1995). From the explanatory perspective, Taylor &

Todd (1995a) compared all three models and found that DTPB provided an increased explanatory power for BI i.e., 60%; this compared with 57% of TPB and 52% of the TAM.

Similar results were also found for explaining intention by Lin (2007) i.e., TAM=0.41%, TPB=0.46%, DTPB=0.57; and by Huh et al., (2009) i.e., TAM=0.61%; TPB=0.59%;

DTPB=0.63%.

In summary, the importance of all three models, specifically the TAM and DTPB, is clear and indisputable. What is not clear, however, is the extent to which models need to be parsimonious as well as capable of explaining individuals’ differences that matters in establishing acceptance behaviour. Up to this point in the discussion, it is concluded that TAM is the most favoured model over others if it is extended with situational and volitional factors similar to TPB and external factors similar to DTPB. This argument can be read in the literature that extended the TAM. For instance, Yi et al., (2006) developed an integrated model based on the TAM, TPB and IDT to assess the acceptance and design of hand-held devices in the health-care sector; it found a significant increase in explaining BI (i.e., 57%) as compared to pure TAM-based studies’ variance (i.e., 30% to 40%).

2.11.4. TPB vs. DTPB vs. TRA

Up until this point in the discussion, the models compared were rooted in the social psychological theory SCT and directly derived from TRA. Nevertheless, models and theories are derived with the aim of overcoming the limitations of preceding one, but this does not mean that they would be better in all aspects from their earlier versions. In accordance with that assumption, the next few sections compare the models derived from TRA (e.g., the TAM, TPB) and TRA itself. The comparison helps to comprehend the real needs that were required at to be complete successive models.

The first extension of TRA was TPB. The two models were similar in that dependent variable of interest was an overt and observable manifestation of the focal behaviour.

Specifically, both theories posit that such BU is influenced by an individual’s BI, which in turn is determined by the individual’s A and SN towards BI (Ajzen, 1985). However, unlike the TRA, TPB introduces an additional construct of PBC as a predictor of BI as well as BU. The inclusion of this additional construct within TPB was to overcome the limitation of TRA when predicting behaviour under conditions where individuals were having low or no volitional control (e.g., Ajzen, 1991; Taylor & Todd, 1995a). According to TPB, volitional control of individuals is unpredictable towards behaviour which needs to

71

be observed with external variable PBC (Ajzen 1985; Madden et al., 1992). Examining the effect of newly added construct PBC within TRA, Madden et al., (1992) compared two models within the student sample to examine the 10 behavioural activities. The authors found that PBC presented a significant increase in the prediction of BI, on average, from R²=48% to 59%, and within BU, R²=28% to 38% (ibid). These results suggest that inclusion of PBC significantly enhances the prediction of BI as well as target behaviour.

Although TPB provided a solution for the TRA’s volitional control assumption, but it still lacks a solution for the inherent assumption of the proximity between BI and BU, which requires specific situations to predict the actual behaviour. In other words, beliefs to measure were still context specific (Foxall, 1997). This limitation was acknowledged by Taylor & Todd (1995a) in the model comparison study. The author decomposed beliefs of TPB that were generalisable across the situations and named the model as DTPB. When comparing the three models, the TAM, TPB and DTPB, Taylor & Todd found that DTPB provide increased explanatory power compared with others, however, it had a less parsimonious structure. Recently, Shish & Fang (2004) compared TRA with its two extensions TPB and DTPB when examining the acceptance of Internet banking in Taiwan.

As expected, the author found that DTPB was the most successful model followed by TPB and TRA respectively. Specifically, explaining BI and BU, the authors found that DTPB explained 66% and 23% variance, TPB explained 54% and 24% variance, and TRA explained 46% and 20% variance respectively. In summary, it is observed that DTPB was more favoured over others from the perspective of the context generalisability as well as explanatory power. Therefore, it can be argued that extending the model to understand the in-depth knowledge is an essential requirement rather than just desirable.

2.11.5. TAM vs. TPB vs. TRA

In line with the discussion on comparing the extensions of TRA with its original conceptualisation and empirical findings, this section aims to examine another extension i.e., the TAM with TRA and its extension, TPB. Before commencing the discussion, it is worth noting that a comparison of the TAM and TPB has already been discussed in section (2.13.3), and a comparison of TRA and TPB is presented in section (2.13.4). Here, the researcher only highlights the main differences between TRA and the TAM with some empirical evidence.

TAM is an immediate succession of TRA. Two models, TRA and the TAM, share the point of similarity that BI is the major determinant of BU. Both models share the limitation

72

of volitional control, where it is assumed that individuals are usually rational when making the decision to engage in a specific behaviour (e.g., Fishbein & Ajzen, 1975; Davis et al., 1989). The two models differ from each other due to two main reasons. First, unlike TRA, the TAM does not include SN as a predictor of BI due to its uncertain theoretical and psychometric properties (e.g., Davis et al., 1989). Second, unlike expectancy formulation of beliefs examined in TRA, the TAM posits only two beliefs: PU and PEOU to predict an individual’s A (however in the final TAM, A was removed due to partial mediaion effect) and BI. The two differences elicited makes the TAM more advantageous compared with TRA. For instance, it was noticed previously that SN remained an unstable predictor to explain BI (e.g., Chau & Hu, 2002; Shish & Fang, 2004; Lin, 2007) therefore its inclusion in a model only increases the complexity rather than explanatory power. The second difference, the addition of normative beliefs (e.g., system design characteristics, individuals’ characteristics, task characteristics, nature of development process, political factors, and organisational factors) and their expectancy formulation with A is also considered to be a limitation of TRA, because for each new context new beliefs need to be elicited that are idiosyncratic in nature and cannot be generalised for other systems (Davis et al., 1989).

Overall, the importance of the two models remains unarguable. Davis et al., (1989) in a paper entitled ‘User acceptance of computer technology: a comparison of two theoretical models’ compared two models in a longitudinal study with a sample of 107 MBA students.

Upon comparing the results of two models in voluntary settings the authors found the TAM to be better than TRA in explaining BI. Specifically, at two time intervals TRA explained 32% and 26% variance, whereas the TAM explained 47% and 51% variance.

Additionally, as theorised in TRA, Davis did not find a significant impact of SN on BI, and hence supports the TAM’s conceptualisation.

Very little research is found in published literature on comparing the three models, TRA and its two extensions, the TAM and TPB. One reason could be their dichotomous differences in conceptualisation. For example, there are studies that compare the conceptualisation of the TAM vs. TRA (e.g., Davis et al., 1989), TAM vs. TPB (e.g., Mathieson, 1991), or TRA vs. TPB (Madden et al., 1992), but studies that compare all the models together are very scarce. Gentry & Calantone (2002) compared three models to examine the buyer intention on the web and found that the TAM explained higher variance in BI i.e., 91% followed by TPB with 85% and TRA with 57%. In another study Venkatesh et al., (2003), during the development of UTAUT, compared the results of eight prominent

73

models including TRA, TAM, TPB and DTPB as well. The authors found that within voluntary settings the TAM was better than the other two models. For instance, explaining BI, the TAM explained 38% variance, whereas TRA explained 30% and TPB/DTPB explained 37% variance. Also, in mandatory settings unexpectedly the TAM was better than the other two. For example, TAM explained 39% variance, TRA explained 26%

variance, and TPB/DTPB explained 34% variance (ibid).

In conclusion, all three models have clear strengths over each other. However, the TAM precedes the other two due to its simple structure and consistent explanatory power, while in the design and implementation process, the other two models are considered to be better than the TAM. Considering the advantages, Venkatesh & Davis (2000) integrated all three models together and named it the TAM2. The authors’ integration approach was successful and the model explained a 60% variance in BI within four different organisational contexts (ibid). The lesson learned from Venkatesh & Davis’ (2000) findings suggest that selecting constructs from the multitude models is the favoured approach to overcome the limitations of earlier models and equally contributes to extending the present theoretical frameworks.