1
Identifying Moderator Relationships in Adoption Theory for IT
Assessing the Effect of Gender and Age.
Master thesis, MscBA, specialization Business Development University of Groningen, Faculty of Business and Economics
February 15, 2013 PIETER KOOMEN Studentnumber: 1460072 Achtergracht 45-3 Amsterdam Tel.: 0624252872 E-mail: pieterkoomen@gmail.com Supervisor J.D. van der Bij
Abstract
The purpose of this thesis is to empirically test the moderating relationship of age and gender on constructs from TAM, PCI and the Marketing Mix, complemented with trust and risk. By testing these moderators on such a wide range of (potential) predictors, a large range of moderating
relationships in adoption theory are tested. Only two moderator relationships were discovered. First, age was found to negatively influence the relationship between perceived usefulness and intention to adopt. Secondly, gender was found to moderate the relationship between visibility and intention to adopt in such a way that for women the relationship was negative and for men the relationship was positive. In addition price and product were found to be significant main effect predictors of intention to adopt. These findings add to current adoption literature, were little empirical research has been done concerning moderators relationships.
Keywords: Adoption Theory, Gender, Age, Intention to Adopt, Technology Acceptance Model, Perceived Characteristics of innovation, Marketing Mix, Trust, Risk.
3
Introduction
Compared to men, women have been found to be more risk averse, more price sensitive and less instrumental (O’Neill, 1998). When looking at age, young people opposed to older people have been found to have lower resistance to change (Oreg, 2003) and higher levels of computer skills (Westerman & Davies, 2000). In addition, when people age their cognitive and learning abilities have been found to decrease (Roedder-John & Cole, 1986). The differences in gender and age have been attributed to differences in behaviour and decision making (Lizarraga, Baquedano & Cardelleelawar, 2007). These differences in decision making behaviour can easily be argued to be important influencers for a possible adopter when forming an opinion about an innovation. However when looking at current adoption theory, surprisingly little attention has been given to these (moderator) variables. This study aims to fill this gap. In order to empirically test these moderators on a large range of possible predictor relationships, a wide selection of theories and constructs will be included.
In current adoption literature there are a number of models that aim to predict the intention to adopt a technology. The most cited (Korpelainen, 2011) of these theories is the Technology Acceptance Model (Davis, 1989). This model gained widespread acceptance due to its robustness and the fact that it is a parsimonious model. The two constructs that influence
intention to adopt are; perceived ease of use and perceived usefulness. Its strong point of being a simple model with just two independent constructs is also its weakness. Because of this the model does not provide al lot of details as to what the underlying perceptions and motivations are that influence intention to adopt. In order to measure these underlying perceptions, constructs from other models will be integrated. A model that has been found to explain variance better than TAM does and provides more detailed information regarding the antecedents driving the adoption of technology innovation (Plouffe, Hulland & Van den Bosch, 2001) is the Perceived Characteristics of Innovation model (PCI) by Moore and Benbasat (1991). Therefore the constructs from this model will be included in the model.
PCI and TAM cover most of the traditional adoption factors, however when looking at adoption theory a lot of studies add one or more additional factors to the traditional models. This is often done to better mirror the setting in which the study is done. For example; in studies into the adoption of financial services, risk plays an important role and is often included as an adoption
4
factor (Featherman & Fuller, 2003). There are a number of factors like this that are often
included as an additional factor. For example trust, risk, costs and promotion have been found to be important determinants of the adoption process (Circu &Davis, 2000), (Awamleh &
Fernandes, 2006). However none of these examples are found in the original TAM or PCI models, two of the most cited models (Korpelainen, 2011). In order to get a more complete picture of the perceptions that influence intention to adopt, the Marketing Mix and the constructs trust and risk will be added in order to create a model that captures a wide pallet of perceptions which influence intention to adopt.
The main effect relationships of PCI, TAM, risk and trust have been found significant many times before and will therefore not be the focus of this study. However the Marketing Mix main effects relationships have not yet been empirically proven and will receive attention. As to the main focus of this study, the moderating effects of gender and age, little research has yet been done in this field. Moderation in TAM has received some attention in al limited office-software setting by two authors; Venkatesh and Moris (2000). However moderation in PCI was only studied for gender and only on partial or modified models. Concerning the Marketing Mix, trust and risk, no studies where found that covered gender and age as moderators on the relationship with intention to adopt.
The main purpose of this study is to fill a gap in current literature by hypothesizing and
empirically testing the moderating influence of gender and age on a wide range of perception – intention to adopt relationships. A secondary goal is to test whether the newly added factors to the model significantly influence intention to adopt.
5
Theoretical background
In this section (current) adoption theory and proposed additions will first be discussed. Due to expected overlap between the different models, a selection will be made concerning which constructs are to be included in this study and which are to be excluded. In addition the moderators and their role in current technology adoption literature will be discussed. This will result in the creation of a conceptual model.
TAM
The Technology Acceptance Model (Davis, 1989) builds on the Theory of Reasoned Action (TRA) by Ajzen and Fisbein (1975) and the Theory of Planned Behaviour (TPB) by Ajzen (1985). Where TRA and TPB contain respectively four and six independent measures in order to predict intention to adopt, TAM replaces these with only two technology acceptance measures. The two constructs that influence intention to adopt in the TAM model are; perceived ease of use and perceived usefulness. This model will be used in this study as the basis to build on because although it is highly simplified, it is still routinely found to explain around 30 to 40% of IT adoption (Legris, Ingham & Collerette, 2003). Since its introduction by Davis in 1989 TAM has found significant support. In 2003 around 10% of the space allocated to information systems publications was claimed by TAM research (Lee, Kozar & Larsen, 2003). However there has been criticism as well; the parsimonious nature of the model leads to low descriptive richness (Plouffe et al, 2001), leaving researchers and managers with little to draw conclusions upon. The model is depicted in figure 1.
6
PCI
The Perceived Characteristics of Innovation model (PCI) was developed by Moore & Benbasat (1991) by expanding the theory of Innovation Diffusion (Rogers, 1983). Rogers originally proposed 5 factors to predict innovation adoption decisions. They were; relative advantage, compatibility, complexity, trialability and observability. PCI was built on this foundation. Relative advantage, compatibility, trialability remained the same. Complexity however was renamed to ease of use, image was segregated from relative advantage and observability was replaced by two new attributes; visibility and result demonstrability. Finally, voluntariness of use was added to the new PCI model. In addition Moore and Benbasat changed what was measured by these constructs. Where in the theory of Innovation Diffusion primary attributes were being measured, the constructs in the PCI model measure secondary attributes (perceived attributes). This change was made because the behaviour of individuals is predicted by how they perceive primary attributes, not by the primary attributes directly. The model is depicted in figure 2. When comparing PCI with TAM, PCI has a higher descriptive richness than TAM because of the inclusion of more detailed constructs (Plouffe et al. 2001). Another point where PCI differs from TAM is that PCI yields a more consistent variance across cultures, making it more robust and reliable across countries (Poong & Eze, 2008). Because of these reasons the constructs from PCI will be included in this study. However because ease of use (PCI) and perceived ease of use (TAM) measure the same concept, the ease of use construct from PCI will be omitted.
7
UTAUT
The latest model that has received considerable support is the Unified Theory of Use and Adoption of Technology (UTAUT) (Venkatesh, Morris, Davis & Davis, 2003). The aim of this model was to unify the constructs of the influential adoption theories up to that moment. When empirically tested, this resulted in a model that explained 70% of variance. The three
independent constructs that influence behavioural intention are; performance expectancy, effort expectancy and social influence. Use behaviour is also included as a construct in the model and opposed to PCI and TAM it is not only influenced by intentions but also by the construct facilitating conditions which directly influences it. In addition the model adds four moderators; age, gender, experience, voluntariness. The model is displayed in Figure 3
Figure 3. Universal Theory of Use and Adoption of Technology
Although the model has been found to explain a considerable amount of variance, it has also been accused of adding chaos (Bagozzi, 2007). The argument made in the study by Bagozzi was that it uses 41 sub-constructs to predict behavioural intention and 8 sub-constructs to predict use behaviour, while still leaving out important constructs. This large number of constructs might also be the reason that in a systematic review of 450 citations of UTAUT by Williams, Rana, Dwivedi & Lal (2011) it appeared that only 16 studies had complete statistical data for each of the constructs presented in the UTAUT. In addition it was found that in several of these studies moderators were treated as external constructs instead of moderators. Because of the substantial
8
number of (sub) constructs in UTAUT and the fact that it misses (sub) constructs covering the perceptions of risk, trust and costs, the UTAUT model does not seem appropriate for this study. Therefore, although it explains a high variance and is the most recent theory, the UTAUT will not be included in this study.
Marketing Mix
The Marketing Mix can be described as a conceptual framework that identifies the principal decisions, marketing managers need to make in configuring their offerings to suit customer’s needs (Palmer, 2004). In the traditional Marketing Mix (Figure 4), this is done by configuring the 4P’s (price, promotion, place, product) in such a way that the customer is optimally targeted increasing their intention to purchase/adopt. These concepts have been found to be important in buying behaviour, but they are not directly captured by TAM and PCI. Therefore, by adding the Marketing Mix constructs, the explanatory power of the model is expected to increase. However because the original model and its constructs originate from the 1960’s, criticism on the
traditional Marketing Mix as an effective tool in a modern on and offline setting has been voiced by multiple authors (Möller, 2006). However most authors conclude that the model is still very effective when it is updated to the e-marketing setting.
Figure 4. Traditional Marketing Mix Model
Some authors add new constructs, like Lawrence, Corbitt, Fisher, Lawrence & Tidwell (2000) who add people and packaging as additional constructs. Others come up with a complete set of new factors, like the 4s model by Constantides (2002) or the internet marketing mix model by
9
Chen (2006). However a number of researchers also suggest that the current 4P’s can stay but need to be explained differently (Allen and Fjermestad, 2001), (Bhat and Emdat 2001), (Lee 2003). In this study this approach will be followed, using the definitions of Yudelson (1999). .
Risk and Trust
Trust will be defined in this study as “the willingness of a party to be vulnerable to the actions of another party based on the expectations that the other will perform a particular action important to the trustor, irrespective of the ability to monitor or control that other party” as proposed by Mayer, Davis & Shoorman (1995). When adopting a new technology, not only is it important that the supplying party can be trusted to actually deliver a product that lives up to the
expectations, trust in adequate after-sales service has become almost just as important. Especially with new technologies, it is important that a supplier can be trusted that when a product does not function as expected, adequate support/service is delivered. In addition many customers trust the suppliers to keep their products up to date through (software) updates. Customers can have a hard time controlling whether a company is actually making an effort regarding this subject. This is where trust comes into play. For perceived risk the definition of Dowlin & Stealin (1994) will be used. They formulated it as:”perceived risk can be regarded as a user´s subjective function of the magnitude of adverse consequences and the probabilities that these consequences may occur if the product is acquired”. When evaluating products or services for purchase and/or adoption, individuals both consciously and unconsciously perceive levels of risks (Peter & Ryan,1976). Also the lack of physical contact and absence of the human side, which are apparent in many new technologies and services, can be a major source of perceived risk (Lieberman &
Stashevsky, 2002). This makes risk a relevant predictor.
Gender and Age
First the role of the moderators in current adoption literature will be discussed, then the moderators age and gender will be discussed and put into context.
The use of moderators is not very common in empirical studies in adoption theory. However one of the most cited studies does uses moderators, this is the paper of Venkatesh et.al. (2003). In this paper the United Theory of Acceptance and Use of Technology is presented which reviews
10
previous adoption theories and makes an effort to describes what they have in common in a single model. In addition four moderators are introduced, being; gender, age, experience, and voluntariness of use. When testing their model they found age and gender to have the biggest moderating effect. Even though this study uses different constructs, the constructs used in the UTAUT model do draw upon the original TAM and PCI constructs. Therefore it can be expected that age and gender will have a moderating effect on these models as well.
When looking at the current literature concerning TAM, moderating relationships for age and gender have been discussed and found significant. A single study by Shumaila & Yani-de-Soriano (2012) found technology readiness, age and gender to moderate the TAM constructs. However the research included existing internet-banking customers. Therefore, in the future research suggestions of this study, it was proposed that additional research including non-users of a service should be done. This thesis studies non users, therefore including these constructs is still of academic value. After an extensive literature review, no research has been found dealing with the effects of the moderators gender and age on the PCI model. The same is true for the Marketing Mix in the adoption setting. The last two constructs, trust and risk are often used as moderators themselves in current literature (Randeree, Kishore, & Rao, 2005) (Featheman & Fuller 2003). However in the studies in which they are used as main effect constructs, no moderators are included (Kesharwani & Bisht 2012). This research aims to fill these gaps.
Age
When it comes to age, it has often been found that younger people are more tech savvy and have less problems with changes. This leads to a higher rate of adoption of computers and tech related items among young people (Venkatesh et al. 2000). However, there is reason to believe this might be changing. Over the last decades an individual’s perception of its own age has changed significantly. A customer segment comprising of individuals over 65 may be called old in terms of age, but they do no longer perceive themselves as old (Mattilla, Karjaluoto & Pento, 2003). This is supported by Marthur, Sherman & Schiffman (1998) who found new-age elderly to perceive themselves as more in control of their lives, more self-confident and younger in age and outlook. This might indicate that the new age mature and elderly have a somewhat lower degree of the characteristics that have traditionally been attributed to coming of age. However several recent studies still report significant differences between the old and the young. For example;
11
Sim et al. (2011) found that when a new technology is being deployed in an organization, young employees displayed a higher interest in the technology compared to older employees. In
addition, something that is not susceptible to a changing self-perception is the decreasing cognitive and learning abilities which accompany the coming of age (Roedder-John & Cole, 1986). Therefore it can still be expected that age will have an impact on the relationships that drive adoption.
Gender
Like age, gender is a classic demographic moderator. However it holds additional relevance in studies relating to internet and computer use because of the Gender Gap in Internet Use. From early childhood men are found to be more experienced with computers and to have more
favourable attitudes towards computers than women (Martin, 1998). The original Gender Gap in Internet Use, which was identified in the 1990’s, incorporated the fact that women were
significantly less likely than men to use the Internet at all. This gender gap in being online disappeared by 2000. Nevertheless, once online, women remain less frequent and less intense users of the internet (Ono & Zavodny 2009). Although several studies have found gender not to have a direct effect on adoption of technology (Taylor and Todd, 1995; Gefen and Straub, 1997), it has been found that there are differences in acceptance rates of specific computer technologies (Gefen and Straub, 1997). Therefore even though the gender gap seems to be closed, gender is still relevant.
12
Conceptual model
The previously discussed models, constructs and moderators form the conceptual model that is displayed in figure 5. In the following section main effect and moderator hypotheses will be developed based on current literature.
13
Main effect hypotheses
In the following section the theoretical models will be shortly discussed in order to identify whether previous empirical research has covered the relationships proposed in the conceptual model. For those where no previous empirical research is available, hypothesis will be created based on literature.
PCI & TAM
Over the years the TAM has received considerable support. After its development by Davis in 1989, the TAM has been tested in a great number of experimental studies. Adams et al. (1992) were among earlier authors to replicate and extend the Davis 1989 study. They found validity and reliability of measurement for perceived usefulness and perceived ease of use across
different settings and different information systems. Since then the findings have been replicated in several fields by various authors including; PDAs (Arning & Ziefle, 2007), the World Wide Web (Moon & Kim, 2001), mobile services (Lapczynski& Calloway, 2006) and e-portfolio systems (Shroff, Deneen & Ng, 2011). Overall TAM was empirically proven to be successful in predicting about 40% of usage intentions and 30% of system usage (Meister & Compeau, 2002).
The constructs from PCI have also been found to have a significant relationship with intention to adopt in a variety of studies covering different technologies. Examples are; smart cards
(Gagliardi & Compeau, 1995), the World Wide Web (Agarwal & Prasad, 1997) e-payment systems (Plouffe et al., 2001) and E-government services (Carter & Belanger, 2004). In these and many other empirical studies it has been empirically proven that the individual constructs from PCI and TAM have a significant positive relationship with intention to adopt, therefore no hypothesis will be created for these relationships.
Marketing Mix
The Marketing Mix is a model that is designed to help the development and communication of a marketing strategy (Kalyanam and McIntyre, 2002). In current literature no empirical link has been made between the Marketing Mix as a model and the intention to adopt. The Marketing Mix theory focuses on sales, which will increase if all four P’s of the Marketing Mix are
14
optimized for the given situation. Therefore adoption may increase when the 4P’s are executed in a way that is perceived as positive by potential customers. In order to test this, hypothesis will be constructed for each of the 4P’s of the Marketing Mix.
Price
Price can be defined as: “everything given by the acquirer in terms of money, time and effort given to obtain the product” (Yudelson, 1999). A price which is perceived to be too high will act as a barrier to adoption. In such a case a customer perceives that any benefits derived from acquiring the product are too low to offset the loss in money, time or effort. This possible negative effect holds for both the initial investment as for the expected costs of usage and maintenance of the product (Alexander et al., 1992) (Mallat & Tuuainen, 2008). In addition, the opposite can also be true; a low perceived price can stimulate adoption even if the perceived benefits are small. An aspect that can influence the perception of the price is the price of
competing products in the marketplace. The sensitivity to competitor pricing has increased due to the integration of the internet in many buy decisions. The reduced search cost enables the
potential customers to easily compare prices of similar products and services (Bakos, 1997) (Ariely and Lynch 2001). Therefore changes in price perception will quickly emerge in response to higher or lower competitor prices. Because of costs being a barrier to adoption and the
increased price sensitivity, it can be expected that that perceived price negatively influences a customer’s intention to adopt
H1: Price negatively influences a potential adopter’s intention to adopt
Promotion
Can be defined as: “all of the information that is transmitted among parties” (Yudelson, 1999). The goal of this is to increase familiarity with the positive characteristics of a brand in order to increase adoption/sales. Brand familiarity has been defined by Alba and Hutchinson (1987) as: “reflecting the brand related experiences accumulated by the customer”. Brand familiarity can be increased by exposure to the brand in (online) stores, brand advertisements, both online and offline and by recognition of the brand name (Sundaram & Webster, 1999). Multiple studies have proven the positive relationship between brand familiarity and purchase intentions. Arora and Stoner (1996) found the relationship in the field of automobiles and insurance, Hoyer &
15
Brown (1990) in peanut butter purchase intentions and Sundaram & Webster (1999) in home air conditioner purchase intentions. It can therefore be expected that promotion, which increases familiarity, will positively influence Intention to Adopt.
H2: Promotion positively influences a potential adopter’s intention to adopt
Place
Can be defined as: “Everything that is done and necessary to smooth the process of exchange” (Yudelson, 1999). Traditionally this construct only included physical efforts like providing physical sales points. However for many technologies the online environment has become the place of exchange. In this virtual environment, websites are the point of contact and can be described as the counter, helpdesk and sales outlet where the actual commercial or non-commercial transaction takes place. For digital services the website will even fulfil the task of the physical distributor (Constantinides, 2002). Within this environment multiple channels are often available to the customer. Having a multichannel distribution strategy has been found to increases customer satisfaction and loyalty just like in the physical world (Geyskens, Gielens & Dekimpe, 2002). This is because different channels typically have different cost structures / support levels. However if customers do not know where (place) to acquire the service at all this will obstruct adoption. Departing from obstruction, additional knowledge/awareness increases the likelihood of finding a suitable match in terms of support, compatibility, necessary effort and costs associated with that channel. Therefore, when the perceived possibilities/places to acquire a product/service increases, it is more likely that a customer finds a good match for its existing needs. This can be expected to increase the intention to adopt.
H3: Place positively influences a potential adopter’s intention to adopt Product
The construct product should nowadays be redefined as: “all the benefits through time that the user obtains from the exchange” (Yudelson, 1999). This definition is somewhat similar to those of relative advantage and perceived usefulness. However product focuses more broadly on all benefits combined. Relative advantage focuses primarily on the comparison with an alternate (older) technology and perceived usefulness focuses more on increased productivity,
16
effectiveness or performance. The construct product is more focused around general perception of satisfaction the product offers, do the customers “like” the product, are they satisfied?
H4: Product positively influences a potential adopter’s intention to adopt
Risk and Trust
Risk and trust have been discussed in literature in combination with intention to use by multiple authors. For example in a study conducted by Walker et al (2002) it was found that the
willingness and actual usage (adoption) are highest where the perceived risks to be associated with use are low. In the E-payment field, research has also been done on this relationship.
Awamleh and Fernandes (2006) found that when the perception of risk increases, the intention to adopt internet banking decreased. When looking at the literature written about trust and intention to adopt the relationship has been found by many authors; Chiravuri & Nazareth (2001),
Dahlberg et al. (2003) and Slyke, Belanger & Comunale (2004). An example is a study by Circu and Davis (2000) was conducted in the field of payment service providers. They found a direct and significant positive relationship between trust and intention to adopt a payment service provider. Since there are multiple studies proving the significant relationship between trust/risk and intention to adopt, no separate main effect hypothesis will be constructed for these
constructs.
Moderator hypotheses
In the next section the moderator hypotheses will be created for the proposed relationships. Before these hypotheses where constructed an Exploratory Factor Analysis was performed in order to investigate which constructs were to be included in further analysis. Perceived ease of use and image were found to have high cross loadings and thus where removed. This process will be described in the first part of the analysis section of this thesis. Because existing data is used in this study, this did not impact data collection. However it did prevent the creation of hypothesis which were not testable with the data. Due to the overlap in logic for some of the hypotheses, they will be presented ordered by main effect relationship.
17
Perceived usefulness
Perceived usefulness was defined by Venkatesh & Davis (2000) as: “the extent to which a person believes that using a technology will enhance her/his productivity, effectiveness, or
performance”. Previous empirical researches concerning the moderating effect of age and gender on perceived usefulness are mainly from the same two authors; Venkatesh and Morris. These authors primarily studied the moderators in the setting of new workplace software (Venkatesh et al, 2003), (Morris & Venkatesh, 2000).
When looking at age, Shumaila & Yani-de-Soriano (2012) found that the relationship between usefulness and continued usage behaviour was stronger for young males due to high levels of optimism and innovativeness. The authors explain that the negative impact of age is caused by; limited experience, lower perception of cognitive capabilities to learn and avoidance of anxiety provoking situations caused by new technologies. In support of this relationship; younger users have been found to place more importance on extrinsic reward, which is an equivalent for perceived usefulness (Venkatesh et al. 2003). Concerning the moderator gender, perceived usefulness was found to have a greater influence on use intentions for men than for women (Venkatesh & Morris, 2000). In support of these findings, Sanchez-Franco (2006) in a study for web-acceptance also found perceived usefulness to positively influence intention to use the web more for men than women. This can be explained by the task orientation of men typically being higher (Minton& Schneider, 1980) and the focus of perceived usefulness on task
accomplishment. The findings of these studies point in the same direction but have been scarce. In addition they were focussed on continued usage instead of adoption by non-users. Therefore this study includes the following hypothesis in order to further test these moderating
relationships in a non user adoption context.
H5:Perceived usefulness has a stronger positive influence on intention to adopt for
younger potential adopters and males than for older potential adopters and females
Compatibility
Little has been written about PCI in combination with any moderators that may be applicable. This also holds for compatibility as a factor. Compatibility can be defined as: “the degree to
18
which the use of an innovation is compatible with, or requires change, in one’s job” (Moore and Benbasat, 1991). If an innovation is perceived as not completely compatible, the degree of compatibility can be measured by the perceived amount of change needed to start using the innovation. Older individuals have been found to be more opposed to changes (Rosen & Jerdee, 1976), (Vang, 2008). The underlying explanation for this is because older individuals are more prone to routine seeking, emotional reaction to imposed change, short-term focus and cognitive rigidity (Oreg et al., 2003). Therefore it can be expected that when compatibility is low and a high level of change is required, older potential adopters are less likely to adopt the technology than their younger counterparts. Women have also been found to be more opposed to changes in routines (Davis & Songer, 2009), therefore the same rationale can be applied.
H6: Compatibility has a stronger positive influence on intention to adopt for older
potential adopters and females than for younger potential adopters and males
Relative advantage
Relative advantage can be defined as: ”the degree to which an innovation is perceived to be superior to using the existing offerings” (Moore & Benbasat, 1991). This construct is closely related to perceived usefulness. It is however different in the sense that it distinguishes between comparing the new technology to an old(er) technology, instead of examining whether a
technology is just perceived to be useful. However the moderating effects of gender and age are expected to be along the same lines as perceived usefulness due to the underlying similarities. When looking at gender, men have been found to be more instrumental in their behaviours, and to be more motivated by goal achievement (O'Neill, 1982). Because of this a technology that is perceived to be superior because it achieves a goal in a somehow better way, will be more valued for this by men. Therefore it can be expected for relative advantage that men have a higher impact on intention to adopt than women do. Concerning age, younger people have traditionally been found to try new things and be more open to new solutions. In the Diffusion of Innovations theory (Rogers 1962), the first categories to adopt an innovation are the youngest individuals. So when presented with a superior product younger individuals are expected to react stronger than older individuals.
19 H7: Relative advantage has a stronger positive influence on intention to adopt for
younger potential adopters and males than for older potential adopters and females
Trialability
Trialability can be defined as: “the degree to which an innovation may be experimented with before adoption” (Moore & Benbasat, 1991). This experimenting can be actively supported by the supplier of an innovation, but could also be initiated by the interested customer. It has been found many times that the easier it is for a customer to experiment, the higher the intention to adopt. A high trialability is basically a higher ability to experiment. Engaging in an
experimenting can be seen as a form of risk taking; a person invests time and effort, but does not know the outcome of the experiment. Steinberg (2008) claims that risk-taking declines as
adulthood progresses because of changes in the brain’s cognitive control system, changes which improve individuals’ capacity for self-regulation. Because of this it can be expected that given a certain level of (perceived) trialability, younger potential adopters will experiment more than older potential adopters. Therefore they will learn more from any pre-adoption testing increasing the impact of trialability on intention to adopt. In the same line of reasoning; Pinker & Spelke (2005) found that men are more prone to risk-taking. Because of this it can be expected that men experiment more and therefore learn more from any pre adoption testing. Therefore it is expected that:
H8: Trialability has a stronger positive influence on intention to adopt for younger
potential adopters and males than for older potential adopters and females
Visibility
Visibility was defined by Moore and Benbasat (1991) as; “the extent to which potential adopters see the innovation as being visible in the adoption context”. An example of a product that currently scores high on perceived visibility in the adoption context is the Apple Ipad. Many people see it in use at home, work or in the media. It has been found by multiple authors that visibility has a significant positive relationship with intention to adopt (Moore and Benbasat, 1991), (Agarwal Prasad, 1998), (Karahanna et al 1999), (Plouffe 2001). When looking at the role of moderators in current literature, some research has been done on this topic. Slyke, Bélanger &
20
Hightower (2005) found visibility to be a significant predictor of usage behaviour for women. It was not for men. They proposed an explanation; because women are more socially and network oriented, they might be more aware of the use of new products in their network but also more inclined to copy what others are doing. In addition, due to typical lower trust and knowledge about IT, females and older customers could need more proof that the innovation functions in a desirable way. The observation that the innovation is being used by other customers can be seen as a proof of functioning. So when visibility goes up, women and older customers are expected to have a higher intention to adopt.
H9: Visibility has a stronger positive influence on intention to adopt for older potential
adopters and females than for younger potential adopters and males.
Result demonstrability
Result demonstrability can be defined as: “the tangibility of the results of using the innovation” (Moore & Benbasat, 1991). With tangibility the authors mean that results should be
noticeable/measurable, not necessarily produce material results. In the case of an online
environment, the results may also take the form of virtual deliverables. For example; a successful transaction delivered by an e-payment service provider. The importance of these demonstrated results can be expected to differ between different ages and genders. Men have been found to be more instrumental in their behaviour and to be more motivated by goal achievement than women (O’Neill 1998). In addition Minton & Schneider (1980) state that men are more task and
efficiency oriented than women. When the demonstrability of the results of using an innovation is high, males can more easily/efficiently assess whether these results will help them in achieving their goal(s). Therefore the effect of result demonstrability is expected to be stronger for males. Only one study (Wong et al. 2012) was identified that covered the topic of age as a moderator between adoption and result demonstrability. However the age groups investigated in this study were exclusively senior. The groups were; young-old, old and old-old. The results indicated a significant difference between age groups, however not a linear relationship. The middle group impacted the effect of result demonstrability the most. Because of the exclusively senior age groups and no linear relationship, no clue as to what the moderating effect of age (18-65) on the relationship between result demonstrability and adoption could be was given. However when looking at the age group of this study; 18-65, a study by Amwamleh and Fernandes (2006) has
21
found that older users tend to need more reassurance when it comes to adopting an innovation. The availability of results can function as such a reassurance, therefore it can be expected that:
H10: Result demonstrability has a stronger positive influence on intention to adopt for
older potential adopters and males than for younger potential adopters and females
Risk
The reaction to risk differs from person to person, but has also been found to change for a single person over time. Sitkin & Weingart (1995) discovered that the tendency to take or avoid risks is a predisposition that diminishes over time. Savage (1993) is one of the few that found younger individuals to be more risk averse, explaining that they tend to overestimate risks. But this research is an exception to the large body of literature that states the opposite relationship. The majority of the researchers come to the conclusion that the older a person gets, the more risk averse they are. These findings have been replicated in many fields of study; gambling (Albert & Duffy, 2012) health care (Cykert, 2004) and internet banking. Amwamleh and Fernandes (2006) found that older non-users of internet banking were more risk averse and required more
reassurance before the negative effect of risk on intention to adopt was reduced. In addition, this propensity of older people to avoid risk has also been found in managerial decisions (Vroom and Pahl, 1971). It was found that risk aversion grew due to the higher age, which had a negative effect on innovative investments and resulted in low growth strategies (Child, 1974). These findings have been replicated more recently by Antia et al. (2010). Therefore it can be expected that older potential adopters have a stronger reaction to their perceived levels of risk, which will influence their intention adopt in a negative way. The relationship between risk and gender has also been the focus of many studies. Most of these studies conclude that men have a higher propensity to take risks than women (Lauriola & Levin, 2001). A possible explanation for this is suggested by Byrnes, Milller & Schaffer (1999), they argue among other things that men are overconfident compared to women. In addition they argue that double standards of parental monitoring, which place more restrictions on girls than on boys, cause women to be more careful. Savage (1993) also found women to be more risk averse, however opposes an
22
higher perceived personal exposure to risks is given as an explanation for the gender divide. When it comes to financial risks the differences between genders remains the same, even when controlled for personal wealth (Jianakopolos & Bernasek 1998). So it can be concluded that overall the reaction to a given level of risk is different for men and women. When it comes to the intention to adopt a technology there is no reason to expect that the risk averseness of women will be different from other studies. Therefore it can be expected that:
H11: Risk has a stronger negative influence on intention to adopt for older potential
adopters and females than for younger potential adopters and males
Trust
The relationship between trust and intention to adopt has been studied many times. However no study was found in which the relationship between trust and intention to adopt had been
proposed to be moderated by age or gender. The relationships between age/gender and trust however have been discussed in literature.
For age, the amount of trust an individual has increases linearly from early childhood up until early adulthood which was defined as 18 to 21 years old. After this, the level of trust was found to be rather stable within different adult age groups (Sutter & Kocher, 2007). Because the potential adopters in this study are only able to adopt the technology after the age of 18 the respondents are all above 18. Therefore only a small portion of the respondents, 18-21 can be expected to differ in their level of trust because of their age. The reaction to this relatively stable level of perceived trust however can be expected to change among different age groups.
Experience with IT technology is typically higher among younger persons. This is important because Dutton & Sheppard (2003) found that experience tends to instil trust on the user
regarding IT and the internet as a whole. So it can be expected that younger users have a higher level of trust in IT and the internet as a whole. Older individuals however with less experience are likely to have less trust in IT. Therefore they are expected to require a higher level of trust for a service/product before they are comfortable enough to adopt them.
When it comes to the relationship between gender and the level of trust in general, women seem to have only a slightly higher propensity to trust than men. This was the finding of a
meta-23
analysis of research results from 1940 to 1992 by Feingold (1994). However in a study in the e-commerce setting by Alesina & la Ferrara (2002), it was found that when trust is lacking, this seems to have a larger negative impact on women than men. Making women less likely to purchase items over the internet. In another study on factors influencing technology satisfaction by Rogers and Harris (2003), low trust by women was found to be a predictor of low satisfaction among women, it was not for men. Therefore it can be expected that females are more influenced by lower levels of perceived trust, leading to lower adoption.
H12: Trust has a stronger positive influence on intention to adopt for older potential
adopters and females than for younger potential adopters and males.
Promotion
Exposure to promotion can vary due to smart targeting of marketing campaigns. Many marketing campaigns for lifestyle products such as clothing and perfumes are specifically designed and targeted at a certain gender or age group. However this typically is not the case for (IT) technologies. This thesis assumes that the same campaigns and levels of exposure are being applied to old, young, male and female potential adopters. However gender differences in information processing do create differences in the effects promotional ads have on the decision tree of a potential adopter. The Meyers-Levy’s selectivity model (Meyers-Levy & Maheswaran, 1991) describes men as selective processors and women as comprehensive processors. For men this means that they do not engage in comprehensive processing of all available information to form a judgement, but instead are more selective. The goal is to use efficient heuristics that are determined by the nature of the task. This results in a judgement based on a subset of all the available cues. For women being comprehensive processors means that they try to use all available cues in order to form a judgement. Restrictions in active memory may prevent women from completely achieving this goal (Darly & Smith, 1995). The influences of these different information processing strategies might explain the findings of Kempf, Palan & Laczniak (1997). The focus of their study was the influence of gender on confidence levels to form judgements in response to information exposure (ads). One of their main findings was that for women, single ad exposure was less successful at creating confidently held attitudes than for men. So for women more repetitions are needed in order to create positive confidently held attitudes about a
24
their need for comprehensive processing before forming a (confident) judgement. In the case of the adoption of IT technology, the service/product (and therefore the explanatory ads) is often relatively complex. Because of the complexity of the ads, the need for more exposure before coming to a positive confident adoption decision is likely to be present for women. Therefore promotion will be more effective with lower levels of exposure for men, making men more susceptible to promotion than females.
When looking at age, information processing might also play a role. However where for gender the strategy is different for each sex, for age the information processing capabilities are subject to change. When individuals come of age, cognitive/information processing capabilities diminish (Roedder -John & Cole, 1986). Because of this, older individuals can be expected to need more exposure to the different forms of promotion before they understand whether the communicated (complex) proposition is advantageous to them.
H13: Promotion has a stronger positive influence on intention to adopt for younger
potential adopters and males than for older potential adopters and females
Product
A significant positive impact of the factor product on the overall Marketing Mix was found for men (Goi, 2011). However no such effect was found compared to the women sample in the same study. From this it can be derived that men place more importance on the product characteristics than women. Although no explanation for this effect was given by the author, this effect may stem from the more goal achievement oriented approach to new products by men. The
relationship that was found implies that when men rate the product construct as high/positively, this would have a positive effect on the other constructs of the Marketing Mix whereas for women no such relationship was found. Therefore it can be expected that the perceived product will have a stronger positive effect on intention to adopt for men than for women. For age no such relations have been found in current literature. However when we look at the younger age groups, a tendency towards higher product involvement has been found which is absent in the older groups (Strizhakova, Coulter & Price, 2008). Product involvement can be defined as "an internal state variable that indicates the amount of arousal, interest or drive evoked by a particular stimulus or situation" (Mitchell, 2002b). So for a young(er) person the perceived product can be expected to result in a higher level of arousal, interest and drive than those with
25
low product involvement (older age groups) These higher response levels of arousal, interest and drive for a given product are likely to increase adoption intentions. A possible explanation why younger people have a higher level of product involvement is because they are more accustomed to search for additional information about the product. For example through the internet or social networks. This is relevant because a higher level of subjective product knowledge has been found to result in higher product involvement (Lutz, Mackenzie & Belch, 1983). A positive relationship is expected between product and intention to adopt (H4). For younger people this effect is expected to be more pronounced due to the higher level of product involvement.
H14: Product has a stronger positive influence on intention to adopt for younger potential
adopters and males than for older potential adopters and females.
Price
As a main effect predictor, price is expected to be negatively correlated to intention to adopt. Age and gender are expected to have an effect on how individuals respond to perceived
costs/price. Regarding age, Ericsson & Starc (2012) found that there is substantial heterogeneity in price sensitivity, with younger individuals more than twice as price sensitive as older
individuals. Similar results were found by Cruz et al. (2010). An investigation by the Dutch Bureau for Economic Policy Analysis (Carman 2006) found similar results for Dutch citizens. An explanation could be that when the average individual comes of age, their personal wealth has increased during their working life (Shorrocks, 1975). This increased personal wealth has been found to be accompanied by a decline in price sensitivity (Chaston, 2009). The higher price sensitivity by younger individuals is expected to impact the relationship between price and intention to adopt. When looking at gender, women generally are found to be more price sensitive than men (Ericsson & Starc, 2012), (Cruz et al., 2010). In the light of technology adoption, Venkatesh et al. (2012) give a possible explanation. In their study they explain that compared to women, men are found to assign higher value to technologies due to the inclination of men to play with technology. Because of this higher assigned value, the price for them is less important than it is to women who assign lower value to the technology. Therefore the higher price sensitivity of women is expected to impact the relationship between price and intention to adopt.
26 H15: Price has a stronger negative influence on intention to adopt for younger potential
adopters and females than for older potential adopters and males
Place
As discussed in the creation of the main effect relationship between place and intention to adopt, the physical locations/marketing channels intended in the original place construct become virtual for many IT related technologies and services. Therefore the level of ease of finding these
websites and level of user friendliness when navigating these websites are new important elements in the updated place construct (Kim &Yoo 2010). These new underlying elements are also important when discussing the relationship of place, intention to adopt and the moderators age and gender.
When individuals perceive that they know where to go in order to adopt a technology, this is expected to be of a positive influence on intention to adopt. The lack of this knowledge is expected to have a negative effect. However this negative effect may be stronger for those who expect to have a hard time gaining this knowledge on the internet. When an individual is not fully aware where to go in order to adopt a technology, it becomes necessary to conduct a search effort before they can adopt. The self-assessed skills an individual perceives to possess will determine whether this necessary search effort will discourage the intention to adopt the technology. Age has been found to be a critical factor influencing success in web search due to the sophisticated information processing demands involving perceptual and cognitive functions (Sutcliff and Ennis, 1998). This age relationship has also been found for relatively simple search efforts, for well defined, specific information that could easily be found in a single source. This is important because it means that even if an individual knows well what they are looking for and the availability of the information is good, the age relationship has been found to maintain
significance (Chin, Fu & Kannampallil, 2009). Therefore it can be expected that age will influence the relationship between place and intention to adopt. When tested, the actual
performance of men and women on searching for information do not differ much. However the perceptions of both sexes on their skills do differ in such a way that women typically
underestimate their search skills (Hargittai & Shafer, 2006). This perception by women that makes them think they will have a harder time finding information will make them react stronger
27
to any increase in the place construct. This is expected to positively influence the relationship between place and intention to adopt.
H16: Place has a stronger positive influence on intention to adopt for older potential
28
Method
The data-source used in this thesis is an existing data set created for the study of Oosterkamp (2012). This study covered the adoption of a payment service provider. The data from this study is used in order to measure each of the constructs which were discussed in the previous section. In order to further clarify the empirical origins of the items, they will shortly be discussed in the following section. All constructs except the control and moderator variables were measured on a 5 point Likert scale ranging from “strongly agree” to “strongly disagree”. For the TAM
constructs six items were adopted from Davis (1989). Since these items did not accurately capture the perceptions of the intended subjects, slight adjustments were made. For the PCI model, items were adopted from both Plouffe et al. (2002) and Moore & Benbasat (1991). Although the original model and items originate from the Moore and Benbasat study, Plouffe et al. developed and validated items for the PCI constructs that were in certain cases more suitable for the empirical setting of the study. The items for the Marketing Mix constructs have been adopted from Yoo, Donthu & Lee (2000) and Kim and Hyun (2010). These authors both applied parts of the Marketing Mix in the IT sector. For the constructs trust and risk, scale items were developed specifically for the collection of data in the setting of an e-commerce provider. In order to measure the intention to adopt, four items from Plouffe et al (2001), Davis (1989) and Venkatesh & Davis (2000) were adopted. In addition to the items that measured the constructs, a number of control variables and the items that measured the moderators were added. Gender was recorded and for age five intervals where created for respondents to select.
The questionnaire was targeted at Dutch merchants only. Therefore after development the items were translated to Dutch to improve the understand ability for the recipients. In addition a pre-test was undertaken with 5 respondents to see whether the items proved clear and
understandable. Only slight translation modifications were necessary the items and the modifications that have been made can be found in Appendix 1.
Data collection
Data was collected through an online questionnaire for which a link was send to merchants which enlisted for the payment provider its newsletter. A dedicated direct mail was send to the respondent containing only a short explanation about the purpose of the questionnaire and the
29
link to the questionnaire. In order to make sure that the person filling in the questionnaire was the person making the adoption decision, a control question was asked in the beginning of the
questionnaire.
Measurements
The e-mail with the hyperlink was send to an e-mail list containing 63354 e-mail addresses. 19% (12982) of the recipients opened the email. From these 12982 recipients 17% (2155) clicked through to the questionnaire and 7% (880) completed the questionnaire without missing data. The focus of this study is to investigate the impact moderators have on intention to adopt, not actual adoption. Therefore only responses from merchants who have not yet adopted the
transaction provider are of interest for this study. These merchants were identified by a negative response to the question: “I already integrated X in my web shop”. 335 merchants did not yet adopt the transaction provider and will be used for this study. The demographics used and relevant for the moderator analysis are displayed in Table1.
Table 1. Respondents Categories Responses (N=335) Gender Male 271 Female 64 Age <20 2 21-30 24 31-40 56 41-50 101 >51 153
Analysis
First a check was performed in order to make sure all items within a construct were coded in the same direction. This is critical in order have all the items that make up a construct measure the same concept. Table 2 shows the items that were reverse coded. In addition the moderator variable age was dummy coded were male=0 and female=1
30 Table 2 recoded items
Construct Recoded items
Relative Advantage -Compared to other transaction providers, I think that X is Riskier for our business.
Result Demonstrability -I would find it hard to explain why adopting X is beneficial or detrimental
Visibility -It is not very visible how other Merchants have adopted X
Risk -Adopting X is safe for a merchant
Price -The discount rate which I need to pay for X is low
-The adoption Costs for adopting X are low
Intention to Adopt -I do not see much use in adopting X
Exploratory factor analysis
After this an Exploratory Factor Analysis (EFA) was conducted on the entire dataset (=335) in order to uncover the underlying structure of the collected data. First the suitability of EFA was assessed. With a sample size of 335 the sample size can be labeled “good” when adhering to the standards set by Comrey & Lee (1992). In this rating system, a sample size of a 100 is labeled poor, 300 as good and 1000 as excellent. The Sample to Variable Ratio for the dataset is 6,7:1 which also falls inside the recommended range of between 5 and 10 subjects per variable (Kass& Tinsley 1979). After this a visual inspection of a correlation matrix found that there were
sufficient correlations greater than .30. In addition as can be seen in table 3, the MSA value is 0.917.Therefore it was found to be appropriate to continue with the EFA.
The goal of EFA is to reduce and consolidate the variables into factors. The design of the
questionnaire was that for each construct from the models, multiple questions would be asked. In essence this should lead to 2 TAM, 6 PCI, 4 Marketing Mix, 1 trust and 1 risk factors. Therefore the a priori criterion was set to 14 factors. The extraction method to be used is Principal
Component Analysis and for the rotation Oblimin with Kaiser Normalization is used because of the more accurate results it produces for research involving human behavior (Costello & Osborn, 2005).
With all variables and 14 factors a vast amount of cross-loading was present. Because of the sample size the cutoff point where a variable was identified to be loading significantly was set on .30, following the guidelines set out by Hair et al. (1998). One by one the highest cross loading
31
variables where removed. Only when all the variables belonging to Image (PCI) and Perceived ease of use (TAM) where removed in addition to several variables belonging to the remaining factors, a stable Pattern matrix emerged. This brought the number of factors back to 12 (table 4) with a MSA value of .862 (Table 3). The variables that were removed are shown in appendix 2. With 21 variables left for 12 factors, the MSA remained high enough; 0.862. Although it is not favorable in a factor analysis for that several factors only one variable remained, these factors however still captured the underlying theoretical concept.
Table 3 KMO and Bartlett’s test
All variables Factor solution
Kaiser-Meyer-Olkin Measure of Sampling Adequacy. ,917 ,862
Approx. Chi-Square 10502,858 3113,613
Bartlett's Test of Sphericity df Sig.
1275 210
,000 ,000
In addition to the dependent constructs from PCI, TAM, Marketing Mix, risk and trust, the dependent construct “intention to adopt” also consisted of 4 variables. A factor analysis was run for this single factor as well. The results which can be seen in Table 4 all pointed in the same direction so there was no need to eliminate any variables.
Table 4. Pattern Matrix Intention to Adopt
Component 1 IntentiontoAdopt2 ,906 IntentiontoAdopt3 ,898 IntentiontoAdopt1 ,779 IntentiontoAdopt4R ,650
Labeling was done according to the corresponding underlying constructs from TAM, PCI, Marketing Mix, risk and trust. In order to check for internal consistency, the Crohnbach Alpha’s of these new factors were determined. These values can be seen in Table 6. No gain was to be had if any variables were to be removed. Hair, Anderson, Tatham & Black (1998) states that an alpha of 0.7 or more indicates that the items are homogeneous and measure the same constant. Alpha’s between 0.6 and 0.7 are considered the lower limit of acceptability. Only Trialability
32
scored within this lower limit of acceptability, the other factors where either scoring higher or were created with only one variable dismissing the concept of internal consistency. Therefore all factors could be used for further analysis. The variables where then combined to create one new variable for each construct/factor resulting in 12 dependent and 1 independent variable.
Table 5. Pattern Matrix
Component 1 2 3 4 5 6 7 8 9 10 11 12 Compatibility 2 ,886 ,034 ,011 -,041 ,023 ,022 -,037 ,004 ,004 ,002 -,033 ,022 Compatibility 4 ,868 -,058 -,020 ,015 -,030 -,013 ,023 -,084 -,018 ,026 -,032 ,007 Price 1 ,021 ,907 ,012 -,037 -,049 ,036 ,002 -,050 ,032 ,046 ,027 ,021 Price 2 -,051 ,874 ,032 -,074 -,101 ,091 -,033 ,025 -,064 -,014 -,041 -,006 Price 3R ,003 ,826 -,053 ,096 ,150 -,102 ,024 ,019 ,084 -,072 -,011 -,083 Visibility 1 -,052 -,043 ,770 -,107 ,059 ,101 -,101 -,065 ,049 -,019 ,021 ,110 Visibility 4 ,032 ,029 ,925 ,052 -,040 -,104 ,051 ,035 -,046 ,040 -,037 -,056 Promotion 1 ,096 ,104 ,023 -,782 ,107 -,058 -,107 ,113 -,102 ,022 ,178 ,140 Promotion 2 -,009 -,035 ,042 -,867 -,035 ,039 ,048 -,154 ,063 ,022 -,166 -,108 Result Demonstrability 4R -,008 -,028 ,000 -,033 ,969 -,039 ,001 ,017 ,023 -,010 -,056 -,036 Risk 1 ,005 ,044 -,040 ,005 -,046 ,939 -,012 ,070 ,053 -,030 -,025 -,037 Place 2 -,002 ,002 ,002 ,028 -,003 ,006 -,995 -,030 ,000 ,020 -,020 -,040 Product 3 ,103 ,016 ,015 -,042 -,025 -,082 -,057 -,885 ,004 -,039 ,000 ,049 Trust 2 -,068 -,040 -,090 -,071 -,098 -,184 -,043 -,086 -,730 ,057 -,168 ,083 Trust 3 ,112 -,040 ,102 ,029 ,021 ,050 -,015 ,041 -,867 -,004 ,031 -,006 Trialability 1 ,001 ,046 ,076 ,046 ,146 ,234 ,033 -,189 -,258 ,595 ,046 ,047 Trialability 2 ,037 -,046 ,024 -,044 -,056 -,102 -,064 ,087 ,068 ,923 -,049 ,004 Perceived Usefulness 1 ,048 ,036 ,028 -,010 ,112 ,031 -,036 -,025 -,021 ,061 -,839 ,107 Perceived Usefulness 2 ,273 -,027 ,039 -,038 -,082 -,004 -,059 -,002 -,133 -,032 -,616 ,084 Relative Advantage 1 ,023 ,029 ,006 ,039 -,025 -,047 ,034 -,038 ,038 ,032 ,002 ,930 Relative Advantage 3 ,015 -,099 ,018 -,026 -,013 ,030 -,011 ,007 -,053 -,035 -,104 ,800
33 Table 6.Crohnbach Alpha values
Construct Items Alpha
Compatibility 2 0,852 Visibility 2 0,706 Trialability 2 0,655 Relative Advantage 2 0,821 Result demonstrability 1 x Perceived usefulness 2 0,850 Price 3 0,862 Promotion 2 0,718 Product 1 x Place 1 x Trust 2 0,771 Risk 1 x Intention to Adopt 4 0,828
With these new constructs a multiple regression analysis will be conducted in order to test the effect of the independent (predictor) variables on the dependent (criterion) variable. Simple regression analysis will be used as the mode of entering the variables into the model because no specific order is put forward by the theories used. The outcome of this analysis will be used to test H1-H4. In order to test the remaining hypothesis two moderated multiple regression analysis will be conducted using interaction terms (figure 6). For all regressions a check for
multicollineairity will be conducted in order to make sure that the independent variables are not strongly correlated with each other.
Figure 6. Moderated multiple Regression equation
Y= b0 + b1x1 + b2x2 + b3x3 + b4x4 + b5x5 + b6x6 + b7x7 + b8x8 + b9x9 + b10x10 + b11x11 + b12x12 + b13z1 + b14(x1*z1) + b15(x2*z1) + b16(x3*z1) + b17(x4*z1) + b18(x5+z1) + b19(x6*z1) b20(x7*z1)b21(x8*z1)b22(x9*z1)b23(x10*z1)b24(x11*z1) b25(x12*z1)+ε Y= b0 + b1x1 + b2x2 + b3x3 + b4x4 + b5x5 + b6x6 + b7x7 + b8x8 + b9x9 + b10x10 + b11x11 + b12x12 + b13z1 + b14(x1*z2) + b15(x2*z2) + b16(x3*z2) + b17(x4*z2) + b18(x5+z2) + b19(x6*z2) b20(x7*z2)b21(x8*z2)b22(x9*z2)b23(x10*z2)b24(x11*z2) b25(x12*z2)+ε
34
In order to make sure the data is suitable for using regression analysis and to validate the eventual results found, certain assumptions should be met (Palant 2007). First the number of respondents or sample size should be large enough. Hair et al, (2000) provides a general rule for assessing this for multiple regression; the ratio of number of subjects to number of independent variables should be 5:1 at a minimum. Provided that the number of subjects is 335 and the
number of variables is 12, the subsequent ratio of 27:1 is adequate to perform multiple regression analysis. Another assumption is that there is no multicollinearity. When checking for
multicollinearity in the main effects model, all the Variance Inflation Factor (VIF) scores remained well below the common cut-off point of 10 Cohen et al (2003). The highest VIF value in this model was 2,342 for compatibility, therefore it can be concluded that for this model multicollineairity is no issue. However when checking for multicollinearity for the moderated multiple regression analysis, unacceptable VIF values as high as 70,318 were found. This is due to the interaction terms which are created with variables that are both on a positive scale. When multiplied low values produce low products and high values produce high products, this causes the product variables to be highly correlated with the original constructs. In order to solve this problem all independent variables were mean centered as suggested by Aiken & West (1991). This is done by subtracting the mean from each variable. After checking for
multicollinearityagain, the highest VIF value found was 3,017 for the product term
Compatibility_Age. Another assumption that should be met is that there should be no outliers. However because of the use of a 5-point Likert scale for all variables this should not be a
problem. Finally the data should be normally distributed; this was assessed by examining normal Q-Q plots for each variable (appendix 3). The data points were close to the diagonal line for all variables, so it can be said that the data is normally distributed. Because all assumptions are met, a multiple regression analysis will be conducted in order to indentify significant relationships. The final analysis that will be conducted is a simple slope analysis. This analysis will be
conducted for all interaction effects that have been found to significantly p>0.05 affect Intention to Adopt. Simple slope analysis for a given IV-DV relationship shows the slopes for a high and a low level of the moderator. This will provide additional insight and will help test the
36
Results
Before continuing to the moderator analysis the direct effect hypothesis are tested. This was done by feeding the factors that were created in the EFA into a regression analysis using the Enter method. The model was found to be significant (p<0.01) and to have an adjusted R-squared of 0,676 meaning the main effects model explain 68% of the variance. Since the scope of this study only included main effect hypothesis which have not been empirically proven before, only part of the constructs from table 6 will be discussed here. The full results of the simple regression analysis are presented in Table 8. Price was found to have a negative
relationship with intention to adopt (β= -0,104, p<0,01), therefore H1 was accepted. Promotion was not found to be significantly and positively related to intention to adopt. Neither was place found to be significantly and positively related to intention to adopt. Because of this H2 and H3 were not accepted. Finally the construct product showed a positively significant relationship with intention to adopt (β= -0,104, p<0,01) leading to the acceptation of H4.
Table 8 Coefficients of the main effect model
Unstandardized Coefficients Standardized Coefficients t Sig. Collinearity Statistics
B Std. Error Beta Tolerance VIF
(Constant) 1,396 ,232 6,012 ,000 Perceived usefulness ,283 ,046 ,289 6,120 ,000** ,435 2,299 Relative Advantage ,210 ,041 ,222 5,082 ,000** ,507 1,972 Compatibility ,282 ,043 ,313 6,568 ,000** ,427 2,342 Visibility ,007 ,040 ,006 ,168 ,867 ,676 1,479 Trialability -,105 ,042 -,100 -2,494 ,013* ,605 1,654 Result demonstrability -,032 ,031 -,034 -1,040 ,299 ,887 1,127 Trust -,044 ,045 -,043 -,985 ,325 ,514 1,946 Risk -,114 ,030 -,128 -3,757 ,000** ,835 1,198 Promotion -,018 ,047 -,014 -,376 ,707 ,672 1,488 Price -,104 ,038 -,098 -2,705 ,007* ,742 1,347 Place ,007 ,035 ,007 ,202 ,840 ,787 1,270 Product ,170 ,037 ,176 4,549 ,000** ,645 1,550 Significance: * p < .05; ** p < .01
For both the moderator age and the moderator gender a separate regression analysis was run in order to test the hypotheses constructed for each moderator separately. The results are shown in table 9 and table 10. It was found that age has a significantly negative impact on the relationship
