THE IMPACT OF CULTURAL DIMENSIONS ON THE TECHNOLOGY ...

Running head: IMPACT OF CULTURAL DIMENSIONS ON TECHNOLOGY ACCEPTANCE OF

VAs

THE IMPACT OF CULTURAL DIMENSIONS ON THE TECHNOLOGY

ACCEPTANCE OF VIRTUAL ASSISTANTS:

A correlational study on user acceptance towards virtual assistants

In Sri Lanka and The Netherlands.

By: Navodinee Niveditha Wickramanayake

ANR: 2014030

MASTER THESIS

THESIS SUBMITTED IN PARTIAL FULFILMENT

OF THE REQUIREMENTS FOR THE DEGREE OF

MASTER OF SCIENCE IN COMMUNICATION AND INFORMATION SCIENCES

TRACK: COGNITIVE SCIENCE AND ARTIFICIAL INTELLIGENCE

AT THE SCHOOL OF HUMANITIES AND DIGITAL SCIENCES,

TILBURG UNIVERSITY

Thesis Supervisor: Eriko Fukuda

Second Reader: Paul Vogt

Tilburg, Netherlands

January 2019

IMPACT OF CULTURAL DIMENSIONS ON TECHNOLOGY ACCEPTANCE OF VAs 2

Abstract

Prior research has identified the moderating impact of cultural dimensions on a user’s technology

acceptance. Contrastingly, this research attempts to address the question of how specific cultural

dimensions, collectivism and power distance correlate with the degree to which people accept Virtual

Assistants (VAs). This question is approached by developing three hypotheses based on constructs

derived from Hofstede’s cultural dimensions (Hofstede, 1980) and an extension of the Technology

Acceptance Model (Venkatesh & Davis, 1996; Venkatesh & Davis, 2003). To test these hypotheses, a

quantitative survey was administered to 159 Sri Lankan participants and 154 Dutch participants. The

main findings indicate that as predicted, collectivism and perceived usefulness of a VA are significantly

negatively correlated, however, this was only true of the Dutch sample. Furthermore, the hypothesis that

power distance has a negative relationship with behavioral intention to use a VA was not supported in

both samples. Finally, the hypothesis that the relationship between an individual's collectivist cultural

dimension traits and their intention to use a VA is moderated by normative beliefs was also not supported

in both samples. The research also identified several limitations with the survey instrument used and

propose adaptations that will contribute to improving the reliability of the subscales for future studies.


Table of Contents

1.1 Defining the Research Goal 5

1.2 Scientific Relevance 6

1.3 Approaching the Research Question 7

Theoretical Framework 7

2.1 Quantifying Technology Acceptance Attitudes 8

2.1.1 TAM 8

2.1.2 Other Technology Acceptance Models 9

2.2 Impact of Culture on Technology Acceptance Models 10

2.2.1 Hofstede's Cultural Dimensions 10

2.3 Research Model and Hypothesis Formulation 11

Methodology 12

3.1 Survey Design and Measures 13

3. 2 Survey Distribution 14

3.3 Survey Sample 15

3.4 Approach to Statistical Analysis 16

3.4.1 Preparation of the Data 16

3.4.2 Approach to Hypothesis Testing 17

Results 18

4.1 Descriptive Statistics of the Constructs 18

4.2 Hypothesis Testing 19

4.2.1 Hypothesis 1 19

4.2.2. Hypothesis 2 21

4.2.3 Hypothesis 3 21

4.3 Summary of Results 24

Discussion and Conclusions 24

5.1.1 Conclusions for the Sri Lankan Sample 25

5.1.2 Conclusions for the Dutch sample 27

5.2 Limitations 28

5.3 Scope for future research 29

References 30

Appendix A 33

Appendix B 34


List of Tables

Table 1: Means and Standard Deviations for the Six Constructs for Sri Lankan Respondents

Table 2: Means and Standard Deviations for the Six Constructs for Dutch Respondents

Table 3: Conditional and Interaction Effects Derived from the Multiple Regression for the Sri Lankan

Sample.

Table 4: Reduced Multiple Regression Model for the Sri Lankan Sample

Table 5: Conditional and Interaction Effects Derived from the Multiple Regression for the Dutch sample.

Table 6: Reduced Multiple Regression Model for the Dutch Sample

Table 7: Summary of Hypothesis Test Results

List of Figures

Figure 1: Amazon Alexa ("Echo Dot Smart speaker with Alexa", 2018)

Figure 2: Hofstede Insights culture comparison tool (Hofstede Insights, 2018)

Figure 3. Work Status of Sri Lankan and Dutch Participants.

Figure 4. Living Status of Sri Lankan and Dutch Participants.

Figure 5: The Best Fit Line between Collectivism and Perceived Usefulness of a VA within the Dutch

Sample.


Virtual Assistants (VAs) are an accessible artificially intelligent innovation that mimic human

interaction through consistent training of algorithms and Natural Language Processing (Claessen,

Schmidt, & Heck, 2017). This technology is distributed to consumers as a pre-installed software on

speakers and mobile devices (Figure 1). VAs, like Amazon’s Alexa, comprise of a range of features,

including making appointments, responding to emails, having simple conversations with one or more

users, and managing home utilities. A VA’s ability to perform these functions and user engagement tasks

like a human assistant have resulted in a large increase in global consumer demand and controversy

around the product (Hoy, 2018; Koetsier, 2018; Smith, 2018).

1.1 Defining the Research Goal

Despite growing consumer apprehension towards Artificial Intelligence (AI) tools like VAs, an

article published in Forbes states that Amazon’s Alexa will be in 100 million homes globally by the end

of 2018 (Cheng, 2018; Jenkins, 2018). This poses the question of what cultural elements may contribute

to how users accept, perceive, and interact with this technology across countries. Therefore, the central

goal of this research is to address the following question: How do cultural dimensions correlate to the

degree to which people accept a VA?

Figure 1. Amazon Alexa ("Echo Dot Smart speaker with Alexa", 2018)

In order to narrow the scope of the research question, two countries, Sri Lanka and Netherlands,

are selected. This choice is based on the significant differences in cultural dimensions of the two countries

(Hofstede, 2001; Hofstede, Hofstede, & Minkov, 2010; Hofstede Insights, 2018) and the access to these

subject pools. Furthermore, of the six cultural dimensions introduced by Geert Hofstede shown in Figure

2, emphasis will be placed on the two dimensions that capture the greatest deviation between Sri Lanka


and Netherlands: Individualism and power distance (Hofstede, 2001; Hofstede, et al, 2010; Hofstede

Insights, 2018).

Figure 2. Hofstede Insights Culture Comparison Tool (Hofstede Insights, 2018).

Here, individualism refers to the extent of how closely knit the social structure of a country is,

and power distance refers to the notion that society predominantly works on a strictly hierarchical

structure. To elaborate further, in a societal context, a person displaying collectivist cultural traits often

thinks of himself/herself as belonging to a group beyond their immediate family and is often loyal to this

group. Individualistic cultural traits represent the opposite end of this spectrum where individuals are

autonomous and focus only on themselves and their immediate family. Power distance represents the

extent to which society and its individuals value authority, a chain of command, and hierarchical structure

(Hofstede, 1980). In practice, this often refers to the extent to which you value and respect social or

organizational hierarchy. Thereby the research question is further specified as follows: How do cultural

dimensions (collectivism and power distance) correlate to the degree to which people accept VAs?

1.2 Scientific Relevance

The proposed research question is developed in response to two primary gaps in technology

acceptance literature: 1) The lack of quantitative analysis on technology acceptance of VAs and 2) the

lack of correlational studies between technology acceptance of AI tools and cultural dimensions.


Most waves of consumer technology innovations such as social networks, e-learning, and mobile

internet banking have had extensive research conducted on user acceptance towards these technologies

through theoretical models such as the Technology Acceptance Model (TAM) (Alalwan, Dwivedi &

Rana, 2017; Rauniar, Rawski, Yang & Johnson, 2014; Yuan, Kanthawala & Peng, 2015). However,

research about the acceptance of AI technologies are limited to a few studies on recommender systems

(e.g., software responsible for recommending music on Spotify) (Oechslein, Fleischmann & Hess, 2014).

Furthermore, multiple cross-cultural studies applying Hofstede’s cultural dimensions have been

conducted to study the moderating relationship between acceptance of new technology (e.g., e-learning)

and culture (Tarhini, Hone & Liu, 2017; Dwivedi, Shareef, Simintiras, Lal, & Weerakkody, 2016).

However, there is a scarcity of quantitative research on possible correlations between cultural dimensions

and how users accept VAs. Therefore, this study will aim to contribute to scientific research by analyzing

the nature of the relationships that lie between cultural dimensions and technology acceptance of VA. In

addition, the paper will also aim to provide insight on the societal implications of varying levels of

technology acceptance and how AI tools such as VAs can be more effectively marketed and

communicated to potential global consumers.

1.3 Approaching the Research Question

The theoretical framework will address the research question by identifying linkages between

TAM, derived from the domain of behavioral sciences (Davis, Bagozzi, and Warshaw, 1989; Venkatesh

& Davis, 1996; Venkatesh & Davis, 2000) and Hofstede's cultural dimensions, derived from the domain

of cross-cultural communication (Mohammed & Tejay, 2017). Next, a basis for the methodology of the

study will be introduced. This methodology will then be used to conduct a quantitative survey on Sri

Lankan (N = 159) and Dutch (N = 154) participants to identify the correlation between cultural

dimensions and technology acceptance of using a VA (Section 3).

Theoretical Framework

This theoretical framework consists of three key themes. First, it addresses technology acceptance

theories and discusses prominent theoretical models used to quantify consumer acceptance of new

technological innovations and their relevance to the study. Next, the framework addresses the impact of

cultural dimensions on technology acceptance. Finally, it presents the hypotheses for this research.


2.1 Quantifying Technology Acceptance Attitudes

The research domains of business, behavioral science, and psychology have extensively studied

the user acceptance of technological innovations. These studies have resulted in the development and

implementation of theoretical models such as the Technology Adoption Curve (Rogers, 1995), TAM 1

(Davis, et al., 1989), TAM 2 (Venkatesh et al., 2003) and Unified Theory of Acceptance and Technology

Use (UTATU) (Venkatesh, Thong, & Xu, 2012). These theoretical models focus on the cognitive and/or

social determinants that affect how a user accepts new technologies. Sections 2.1.1 and 2.1.2 will

highlight the most relevant technology acceptance models to this research and their theoretical

frameworks.

2.1.1 TAM

Presented in 1989 by Davis, et al, TAM is a well-established technology acceptance model

(Alalwan, Dwivedi & Rana, 2017; Cai, Wohn, Mittal & Suresh Babu, 2018; Kessler & Martin, 2017,

Srite & Karahanna, 2006). Originally, the model was used to measure perceived usefulness and perceived

ease of use of information systems which are systems that analyze business information to help

companies make informed decisions. Here, perceived usefulness refers to the degree to which a user

believes that an information system improves his/her job performance, while perceived ease of use

measures how user-friendly a system is.

The finalized version of the original TAM theorized that perceived usefulness and perceived ease

of use of an information system influenced a user’s behavioral intention to use this system, and finally,

influenced the actual use of the system (Davis and Venkatesh, 1996). Furthermore, the model also places

significance on the role of external variables in influencing perceived usefulness and perceived ease of

use. Although this model does not provide explicit examples of what these external variables can be,

subsequent adaptations and extensions to TAM have identified a few possible variables.

For instance, adaptations of TAM have identified variables such as normative beliefs (also

referred to as subjective norms) to influence attitudes towards technology, making the model more

applicable to changing dynamics within social structures. Normative beliefs refer to the degree to which

an individual believes that people important to him/her would encourage the use of a technology (e.g.,

Believing that your sibling would think that you should use a VA). Srite & Karahanna postulated that the

addition of normative beliefs directly impacts a user’s behavioral intention to use a technology (2006).

The integration of normative beliefs as a variable to the model brings in an important social component to

TAM that reflects societal influence. Other identified variables include the introduction of hedonic

motivations to the TAM model, for instance when measuring technology acceptance towards live

streaming. Here, hedonic motivation refers to the element of enjoyment or entertainment present within


consumer technologies (Cai, et al., 2018). Therefore, even though TAM was originally intended to

measure user acceptance of information systems used within the workplace, case specific adaptations of

the model have extended their applicability outside of the workplace. Some examples of its applications

in consumer technologies are mobile banking (Alalwan, et al, 2017) and internet of things, a network of

devices and appliances that are interconnected (Kessler & Martin, 2017). This has not only increased the

model's robustness but has also strengthened its applicability in today’s growing consumer technology

market.

TAM is not without critique. The original model has shown limitations in its explanatory power

and has been criticized for not being able to holistically explain technology acceptance as it only

measures perceived usefulness and perceived ease of use (Lai, 2017; Straub & Burton-Jones, 2007).

However, this has been rectified by the extensions that have been developed specific to the type of

technology that is being measured, as explained in the latter part of section 2.1.1 (Alalwan et al., 2017;

Cai, et al., 2018; Kessler & Martin, 2017). This was taken into consideration when selecting a suitable

version of TAM for this study by using a model that captured normative beliefs.

Furthermore, even though the model has been used to quantify user technology acceptance

towards the internet of things, a technology that improves the functionality of VAs by connecting to

household appliances, TAM has still not been applied specifically to VAs. Additionally, the model has

not been used to explore user attitudes of any technology in Sri Lanka and The Netherlands specifically,

making it difficult to predict how constructs of the model will perform when applied to this technology

and cultures. Despite these limitations, this research aims to successfully contribute to these identified

gaps in research by using technology acceptance constructs derived from TAM.

2.1.2 Other Technology Acceptance Models

Over time, cognitive and social aspects of a user were incorporated into technological

infrastructure. However, this was not accurately reflected in the original version of TAM presented by

Davis et al. (1989). Therefore, TAM 2 was proposed by researchers Venkatesh and Davis to address these

limitations (Venkatesh et al., 2000). This extension followed a similar reasoning to the studies that

introduced external variables to the original TAM. In addition to perceived ease of use, TAM 2

incorporated two categories of external variables to the original TAM. These two categories were external

variables influenced by society and external variables influenced by human cognition. Here, the socially

influenced variables included factors such as normative beliefs, a user’s willingness to use a technology,

and the impression formed of a person using a certain technology. The variables influenced by cognition

included relevance to job performance and perceived quality of task fulfillment.


Even though this model is accepted as a successor to the original TAM, it can be argued that

limiting the model to these identified external variables further restricts the areas in which the model can

be applied to. As such, simply extending the original TAM based on the technology and context measured

is sufficient and effective, as supported by recent studies (Alalwan et al., 2017; Cai et al., 2018; Kessler &

Martin, 2017). This further strengthens the reasoning behind the choice of using TAM1 based constructs

for this study.

2.2 Impact of Culture on Technology Acceptance Models

Although a majority of research on technology acceptance addresses the impact of variables like

age, gender, experience, and social influence on technology acceptance, research has shown that culture

influences the outcome of TAM too (Tarhini, Hone & Liu, 2015).

For example, Muk & Chung (2015) compared technology acceptance of SMS advertising

between South Korean and American university students using TAM. The results found distinct

differences between the two countries, specifically in regards to perceived usefulness, where South

Korean students demonstrated more favorable attitudes towards SMS advertising in comparison to

American students. Furthermore, social influence had no impact on South Korean students, whereas it

showed a positive relationship amongst their American counterparts (Muk & Chung, 2015). Thus, it can

be argued that technology acceptance could vary as a result of underlying cultural factors.

Chou, Kim, Ung, Yutami, Lin, & Son (2015), in a study comparing the technology acceptance of

smart grid adoption between users in Taiwan, Korea, Indonesia, and Vietnam using an extension of TAM,

arrived at a similar conclusion. In this study, Vietnamese and Taiwanese consumers were strongly

impacted by normative beliefs, whereas this was not true of Korean and Indonesian users. These findings

further strengthen the aforementioned argument of the link between technology acceptance and

underlying cultural factors.

2.2.1 Hofstede's Cultural Dimensions

In recent years Hofstede’s cultural dimensions have been examined as a moderating variable to

existing TAM models (Srite and Karahanna, 2006; Tarhini, et al., 2015; Tarhini et al, 2017). His research

scores countries on the following cultural dimensions: masculinity/femininity, uncertainty avoidance,

long-term orientation, indulgence, power distance, and individualism/collectivism (Hofstede, 1980).

Research published by Srite and Karhanna (2006) identified that these dimensions have

significant moderating effects on TAM with regards to personal computers. These findings have been

observed across participants of 30 nationalities. Similar results were observed in a more recent study


conducted in Lebanon which examined the moderating role of cultural dimensions on e-learning tools

(Tarhini et al., 2017).

However, research has not explored the possible linear correlation cultural dimensions may have

with technology acceptance of VAs. It is possible that the results detailed in Muk & Chung (2015) as well

as Chou et al. (2015) are due to the direct influence of cultural dimensions. Interestingly, Sri Lanka and

Netherlands, show largely contrasting scores on individualism (Sri Lanka: 35 vs. Netherlands: 80) and

power distance (Sri Lanka: 80 vs. Netherlands: 38) (Hofstede, 2001; Hofstede, et al., 2010; Hofstede

Insights, 2018). This may have an impact on how technology is accepted in the two countries.

2.3 Research Model and Hypothesis Formulation

Although a majority of research surrounding cultural dimensions and TAM have explored the

moderating effect of cultural dimensions on the model, Yeniyurt and Townsend (2003) discover

interesting insights into the direct relationships that lie between cultural dimensions and the acceptance of

new products. Even though Yeniyurt and Townsend did not use a TAM based model or focus on VAs, the

results derived draw clear connections between Hofstede’s cultural dimensions and the acceptance of new

products. The study specifically finds that individualistic cultural dimensions have a positive impact on

the acceptance of new products (2003). When keeping in mind that individualism is on the opposite end

of the same spectrum as collectivism, it could give the assumption that collectivist cultural dimensions

will have a negative impact on the acceptance of new products. Here, the acceptance of new products

closely relates to perceived usefulness as it refers to a product positively contributing to a user’s daily life.

Supported by this, a similar conclusion can be expected from the findings of the current research, where

collectivism will have a negative relationship on the TAM construct of perceived usefulness. Therefore,

the following hypothesis is presented: H1: Collectivist cultural dimension traits have a negative

relationship with perceived usefulness of a VA.

Next, a study conducted by Baptista and Oliveira (2015) on the technology acceptance of a

mobile banking app in Mozambique found power distance to have a significant influence on intention to

use the app. Using UTATU presented by Venkatesh et al. (2003), the study also found that social

influence played a large role in the country due to Mozambique’s high power distance which in turn

impacted the acceptance of the app (Baptista & Oliveira, 2015). UTATU is an elaborate extension of

TAM, created using a combination of eight individual theoretical models surrounding technology

acceptance from the academic domains of psychology and sociology. Due to the similarities of the

UTATU model and TAM, similar results could be expected when assessing the nature between power

distance and TAM construct of behavioral intention to use. Therefore, it could be postulated that power

distance has a strong relationship to the TAM construct of behavioral intention to use a VA.


A VA’s ability to perform tasks like a human and eliminate the need for a human assistant or to

ask for help from another individual counteracts with inherent high power distance attitudes like

delegating tasks to another individual or group. This argument is supported in Yeniyurt and Townsend

(2003) where power distance was found to have a negative relationship with the acceptance of new

products in general. On this basis it could be argued that: H2: Power distance has a negative

relationship with the behavioral intention to use a VA. This hypothesis is based on the notion that

individuals who value hierarchy would rather transfer a task over to another individual than do it

themselves via an assistive technology (i.e., Recruiting a travel agency to book flight tickets on your

behalf).

Finally, attention is drawn to the moderating effect of the TAM construct, normative beliefs. A

few studies have been conducted on the relationship between collectivism and normative beliefs. Ra &

Cho (2018) found that normative beliefs are a significant moderator to smoking intentions among risk

taking women in Korea, predominantly described as a collectivist nation (Kim & Park, 2010). This begs

the question if the moderating effect of normative beliefs can be generalized to collectivism with regard to

the technology acceptance of VAs. Furthermore, as discussed before, recent TAM based models have

identified normative beliefs as a direct influencer to behavioral intention to use a technology when

moderated by collectivism (Srite and Karahanna, 2006; Tarhini, et al., 2015; Tarhini et al, 2017).

However, it is unclear if the same effect will be observed if the roles were reversed and the relationship

between collectivism and behavioral intention to use a VA was moderated by normative beliefs instead.

This effect would most likely impact those who are more collectivist than those who are individualistic

and do not place too much importance on what society thinks of them. Based on that reasoning, the

following hypothesis is proposed: H3: The relationship between an individual's collectivist cultural

dimension traits and their intention to use a VA is moderated by normative beliefs. Therefore, it can

be predicted that there is a significantly positive relationship between a person’s degree of collectivism

and behavioral intention to use a VA when moderated by normative beliefs.

Methodology

This chapter will lay out the methodology used to test the hypotheses presented in the preceding

section. It begins with laying out the process in which the survey instrument was developed and

administered to the sample; this also includes the characteristics of the sample. This chapter concludes

with information on the preparation of the data, as well as arguments for the selected approach to test the

hypotheses presented in the previous chapter. The measurement instrument can be found in Appendix B.


3.1 Survey Design and Measures

The survey was created in English and comprised of demographic questions, a scale, and a video

that was displayed to respondents based on their nationality. All questions in exception to the question

about disposable income was configured to elicit a response in order to move forward in the survey. The

structure of the survey was divided into the following three stages:

1) The respondents were presented with six demographic questions. These questions explored

age, gender, work/study status, living situation (e.g., with parents, living alone), monthly disposable

income and nationality. At this stage, respondents older or younger than the predetermined age range of

20-25 were automatically exited from the survey.

2) Next, respondents were shown two subscales in random order that measured two of the

preselected Hofstede’s cultural dimensions using a 7-point Likert scale (1 = strongly disagree up to 7 =

strongly agree). One subscale measured collectivism using six items (e.g., group success is more

important than individual success.) The other subscale measured power distance using four items (e.g.,

managers should make most decisions without consulting subordinates).

3) Respondents were directed to one of two 2-3-minute video advertisements published by

Amazon highlighting the features of a VA, specifically, Amazon’s Alexa (Amazon, 2015; Amazon,

2018). The Sri Lankan respondents were shown a video featuring an Indian cast, while the Dutch

respondents were shown a similar video featuring an American cast to create a sense of familiarity for the

respondents, please see Appendix A.

4) After respondents viewed the video, they were shown four more subscales measuring several

constructs of TAM in random order. Here, the perceived usefulness subscale (e.g., using a VA will

enhance my productivity), perceived ease of use subscale (e.g., I would find VAs easy to use) and

normative beliefs subscale (e.g., my friends would believe I should use a VA) had four items each, while

the behavioral intention to use subscale (e.g., I intend to use a VA in my daily life) had two items. These

four subscales were also scored on a 7-point Likert scale (1 = strongly disagree up to 7 = strongly agree).

All subscales mentioned in stages 2 and 4 were adapted from an original scale used to measure

technology acceptance of university students in a study conducted in America by Srite and Karahanna

(2006). This scale was more recently adapted and used by Tarhini et al. (2017) in a study conducted in

Lebanon, introduced in section 2.2.1. With regards to the reliability of the scale used in this correlational

study, Tarhini et al. (2017) found that all cultural dimension and TAM subscales displayed a Cronbach's

alpha ranging between .85 and .90, showing high reliability within a Lebanese sample. In addition, the

TAM subscales were measured in a cross-cultural study between Lebanese and British students which

displayed a Cronbach's alpha of .74 - .90 within the Lebanese sample and .92 - .83 within the British

sample. These high reliability scores suggest that this scale can be used in varying countries and cultures.


Furthermore, Srite and Karahanna’s subscales measuring cultural dimensions were originally

derived from Hofstede (1980) and Dorfman and Howell (1988); the subscales measuring perceived ease

of use, perceived usefulness, and behavioral intention to use were originally derived from Davis (1989),

and lastly, the subscale measuring normative beliefs was developed by the authors themselves (Srite &

Karahanna, 2006). The adaptations to the original scale for the purpose of the current research ensured

that the survey addressed respondents who were students and/or employees, and also referred to their

perceptions towards VAs. This was done in order to make the scale relevant to VAs and the Sri Lankan

and Dutch demographic groups surveyed.

3. 2 Survey Distribution

The survey was created and distributed via Qualtrics, an online survey builder. As a control for

bias caused due to differing access to technology between respondents of the two countries, the sample

was restricted to those from a similar age group, educational background, and access to technology.

Furthermore, the survey was administered to respondents who do not presently own a VA, this decision

was made in order to capture the perceptions toward a VA instead of capturing retrospective user

experience.

All steps were taken to ensure full anonymity by configuring Qualtrics to refrain from collecting

IP addresses and any identifiable personal data. The survey was distributed using purposive and snowball

sampling methods, where the survey was only distributed to respondents who fit the profile described

above and were encouraged to share the survey link with friends fitting the same profile. In practice, this

was done using two primary methods: 1) distribution of survey links to respondents via Facebook and

WhatsApp with a request to share among friends; 2) in-person distribution of the surveys with a similar

request, where the participant completes the survey on the spot. Due to difficulties experienced when

distributing the surveys to Dutch respondents, two additional distribution methods were adopted for this

demographic, they are: 3) collaborating with two lecturers at Tilburg University to present a link to the

survey before four lecture sessions for consenting respondents to complete on the spot, and 4) collecting

email addresses from consenting respondents gathered at a class and sending the survey link via email for

them to complete and requesting them to share the link with their friends.

The distribution of surveys to the Sri Lankan respondents was delegated to associates who

distributed the survey according to methods 1 and 2. The survey distribution to Dutch participants was

conducted solely by the primary researcher by approaching students at Tilburg University, Talent Square

- a student housing complex, posting on Facebook groups, WhatsApp messaging, and collaborating with

lecturers as detailed above.


Participation in the survey was completely voluntary and the respondents were notified of their

right to withdraw from the study at any time. Furthermore, no financial incentives or rewards were

provided to respondents for completing the survey.

3.3 Survey Sample

A total of 159 Sri Lankan respondents and 154 Dutch respondents responded to the survey. Of the

Sri Lankan respondents, 57.9% were female and 42.1% were male, while 59.7% of the Dutch respondents

were female and 40.3% were male. The mean age of the Sri Lankan sample was 22.31 years of age (SD =

1.68) and the Dutch sample had a similar mean age of 22.77 years (SD = 1.51).

Figure 3. Work Status of Sri Lankan and Dutch Participants.

Furthermore, as seen in Figure 3, the Sri Lankan sample included 30 employees who were recent

graduates, 80 students, and 49 students who were also employees, while the Dutch sample included 3

employees who were recent graduates, 93 students, and 58 students who were also employees.


Figure 4. Living Status of Sri Lankan and Dutch Participants.

As seen in Figure 4, 129 Sri Lankan respondents lived with their parents, while 15 lived by

themselves, 13 lived with housemates, and 2 lived with their partners. In contrast, most of the Dutch

respondents lived with housemates (N = 74) while 36 lived by themselves, 30 lived with their parents, and

only 14 lived with their partners. In addition, on average, Sri Lankan respondents had a disposable

income of EUR 116 (SD = 186) while the Dutch respondents had a higher disposable income of EUR 372

(SD = 402).

3.4 Approach to Statistical Analysis

3.4.1 Preparation of the Data

The data preparation and analysis for this research was done using the statistical software, SPSS.

In order to test the hypotheses presented in section 2.3, the data derived from the sample was cleaned and

prepared in the following manner:

First, data from respondents who did not belong to Sri Lanka or The Netherlands were discarded.

With regards to the treatment of missing data belonging to the Sri Lankan and Dutch datasets: 33 out of

346 respondents partially completed the survey. Of these 33 partial responses only 14 respondents

proceeded beyond all the demographic questions and none of them responded to any of the technology

acceptance subscales. This is most likely because the respondents did not want to watch the video placed

before the technology acceptance subscales, indicating that the missing data can be classified as missing

at random (MAR). Following the example of the methodology used in TAM related studies conducted by


Srite & Karahanna (2006), Tarhini, et al (2017), and Chou et al. (2015), a decision was made to remove

these missing responses in its entirety using the list-wise deletion technique. This method is also

supported by Bennett (2001) who states that list-wise deletion is acceptable in the instance that the

missing data accounts for less than 10% of the sample as is not considered large enough to cause a bias to

subsequent statistical analysis.

Next, all disposable income of Sri Lankan respondents was converted into Euros for

standardization. Then, mean scores were computed for the cultural dimension and technology acceptance

subscales thereby converting this data from ordinal to continuous data. Finally, the reliability of the scores

were measured using a Cronbach's alpha and descriptive statistics of the constructs measured through the

subscales were recorded.

3.4.2 Approach to Hypothesis Testing

This research adopts a simple approach to identifying the impact of cultural dimensions,

collectivism and power distance on the technology acceptance of VAs in Sri Lanka and The Netherlands.

To refresh, (H1) explores the potential existence of a direct negative relationship between collectivism

and perceived usefulness. As such, a one-tailed correlation analysis was selected as the appropriate

approach to test this directional hypothesis. Furthermore, due to the nature of the mean score variables

used to measure collectivism and perceived usefulness of a VA, the distribution of the data, as well as the

large sample size (Sri Lanka: N=159 and Dutch: N=154), a Pearson’s correlation test was selected as the

best fit for the data and the desired analysis. Here, the Dutch sample was bootstrapped based on 1,000

bootstrap samples due to the non-normal distribution of data within the sample.

The same parameters were taken into consideration when deciding the best fit for the analysis of

H2: Power distance has a negative relationship with behavioral intention to use a VA. Due to the

similarities in the nature of the analysis, a one-tailed Pearson’s correlation test was selected as the

appropriate analysis in this instance as well.

H3: The relationship between individuals scoring high on collectivism and their intention to use a

VA is moderated by normative beliefs, attempts to explore the moderating effect of normative beliefs on

the relationship between collectivism and behavioral intention. As such, a moderated multiple regression

using PROCESS by Hayes (2013) - a computational tool that integrates many of the functions of existing

statistical tools, was selected as the appropriate test given the desired exploration of the variables.


Results

4.1 Descriptive Statistics of the Constructs

As mentioned in section 3.4, a reliability analysis was done to assess the internal consistency of

the subscales measuring the cultural dimension constructs and TAM constructs for the Sri Lankan and

Dutch samples. As seen in Tables 1 and 2, all subscales showed high reliability scores for both countries

except for the power distance construct. This construct had a low reliability score of Cronbach’s α = .63

within the Sri Lankan sample and a Cronbach’s α =.50 within the Dutch sample. Even though power

distance showed low reliability, it was decided that this scale would nonetheless be included in the

analysis. This decision was supported by the fact that most methodologists deem a Cronbach's alpha

lower than 0.5 to be unacceptable, but both values fall above this threshold (Taber, 2017).

Next, the mean scores shown in Tables 1 and 2 were computed for each construct, the maximum

score obtainable was 7.0. Both Table 1 and 2 show that none of the mean scores appear to be

exceptionally high or low for both samples:

Table 1

Means and Standard Deviations for the Six Constructs for Sri Lankan Respondents

M SD Cronbach’s a

Collectivism 4.17 1.05 .72

Power Distance 2.36 0.91 .63

Perceived Usefulness 4.94 1.24 .92

Perceived Ease of Use 5.33 0.98 .82

Behavioral intention to use 4.30 1.48 .89

Normative Beliefs 3.91 1.24 .89

Note: Maximum score obtainable = 7.0

A notable discovery within Table 1 is that Sri Lankan respondents displayed a moderately low

power distance which is contrary to the assumption made in the theoretical framework which suggested


that a high power distance could be expected from this sample (Hofstede, 2001; Hofstede, et al., 2010;

Hofstede Insights, 2018).

Table 2

Means and Standard Deviations for the Six Constructs for Dutch Respondents

M SD Cronbach’s a

Collectivism 3.69 0.97 .81

Power Distance 2.53 0.77 .50

Perceived Usefulness 4.06 1.35 .93

Perceived Ease of Use 5.43 1.22 .91

Behavioral intention to use 3.45 1.70 .95

Normative Beliefs 2.82 1.15 .87

Note: Maximum score obtainable = 7.0

Table 2 is consistent with the expected outcomes of this sample with low levels of collectivism,

power distance and normative beliefs.

4.2 Hypothesis Testing

The following sections will present the results for the hypothesis testing for each country. Results

will be split and reported for both countries separately. Note that all tests are interpreted based on a 5%

level of significance.

4.2.1 Hypothesis 1

To test H1: Collectivist cultural dimension traits have a negative relationship with perceived

usefulness of a VA, a one-tailed Pearson’s correlation analysis was performed taking into account the

large sample size and data type (Sri Lanka: N=159 and Dutch: N=154).

Sri Lanka: The average collectivism score was 4.17 (SD = 1.05), the average perceived

usefulness score given to a VA was 4.94 (SD = 1.24). The assumptions of normality for this sample were


met: The z-scores for collectivism were: skewness z-score= -1.04 and kurtosis z-score = 0.74, the z-scores

for perceived usefulness were: skewness z-score = -0.26 and kurtosis z-score = 1.22. The results of the

Pearson’s correlation analysis revealed a non-significant positive correlation between the two variables,

Pearson’s r (157) =.12, p = .062. Therefore, H1 is not supported within the Sri Lankan sample.

Netherlands: The average collectivism score for Dutch respondents was 3.69 (SD = 0.97),

showing more individualistic cultural dimension traits. The average perceived usefulness score given to a

VA by this sample was 4.06 (SD = 1.35). As the assumption of normality was not met within this sample

(kurtosis z-score for collectivism = -3.74), a bootstrapped Pearson’s correlation analysis was done. The

results suggest that a significant negative correlation between collectivism and perceived usefulness was

found, Pearson’s r (152) = -.16, 95% CI [-.32,-.01], p = .022. The 95% CI based on the percentile method

confirms the significance of the correlation. Therefore, H1 is supported within the Dutch sample.

Figure 5: The Best-Fit Line between Collectivism and Perceived Usefulness of a VA within the

Dutch Sample.

The best-fit line shown in Figure 5 suggests that as collectivism decreases (displaying higher

individualistic cultural dimension traits), the perceived usefulness of a VA increases. That is, people


displaying individualistic cultural dimension traits find VAs more useful than those displaying collectivist

cultural dimension traits, as predicted.

4.2.2. Hypothesis 2

A one-tailed Pearson’s correlation was performed to test H2: Power distance has a negative

relationship with behavioral intention to use a VA. This test was once again selected for the analysis after

taking the data type and the large size of the sample into consideration (Sri Lanka: N = 159 and Dutch: N

= 154).

Sri Lanka: An average power distance of 2.36 (SD = 0.91) was observed within the sample. The

average behavioral intention score was 4.30 (SD = 1.48). The assumptions of normality were met. The z-

scores for power distance were: skewness z-score= 0.26 and kurtosis z-score = 1.15, the z-scores for

behavioral intention to use a VA were: skewness z-score = -0.45 and kurtosis z-score = -0.85. Next, the

Pearson’s correlation test performed showed a positive correlation between the two variables. However,

this was not statistically significant, Pearson’s r (157) = .01, p = .458. Therefore, H2 is not supported

within the Sri Lankan sample.

Netherlands: The average power distance score was 2.53 (SD = 0.77). The average behavioral

intention to use a VA score was 3.45 (SD = 1.70). The assumptions of normality were met: The z-scores

for power distance were: skewness z-score = 1.35 and kurtosis z-score = -0.74, the z-scores for behavioral

intention to use a VA were: skewness z-score = 1.17 and kurtosis z-score = -0.32. Subsequently, the

Pearson’s correlation test showed a positive relationship between the two variables. This too was not

statistically significant, Pearson’s r (152) = .13, p = .056, therefore, H2 is also not supported for the Dutch

sample.

4.2.3 Hypothesis 3

In order to test H3: The relationship between an individual's collectivist cultural dimension traits

and their intention to use a VA is moderated by normative beliefs, a moderated multiple regression

analysis was performed using the tool PROCESS Model 1 (Hayes, 2012).

Sri Lanka: The average normative beliefs score observed for the sample was 3.91 (SD=1.24).

The model summary for the multiple regression analysis was as follows: F(3,155) = 11.92, p <.001, R2

= .19, suggesting that as a set, collectivism, normative beliefs, and the interaction between the two


account for 19% of the variance in behavioral intention to use a VA. Table 3 illustrates the conditional

and interaction effects individually.

Table 3

Conditional and Interaction Effects Derived from the Multiple Regression for the Sri Lankan Sample.

β SE t p

Collectivism .07 .10 .67 .501

Normative Beliefs .48 .09 5.53 <.001

Interaction .09 .07 1.20 .231

Interpreting Table 3 suggests that normative beliefs do not have a significant moderating effect on

the relationship between collectivism and behavioral intention to use a VA, β= .09, t(155)= 1.20, p

=.231. As such the results do not support H3. Furthermore, the addition of the interaction did not make a

significant change to the model as a whole, F(1,155) = 1.45, p =.231, with R2 change = .01, representing a

small sized effect. As previously mentioned, the moderating effect of normative beliefs on collectivism

and behavioral intention to use a VA is not statistically significant, thereby not supporting H3 for the Sri

Lankan sample.

As the interaction effect of the model was deemed non-significant a follow up analysis was

conducted by removing the interaction and conducting a multiple regression on the reduced model. The

diagnostics suggested that the residuals deviated from normal (normative beliefs skewness z-score = -

2.84, behavioral intention to use a VA skewness z-score=-2.21). As such, bootstrapping based on 1000

samples was done.

The model summary for the reduced model was as follows: F (2, 156) = 17. 11, p<.001, R2 = .18,

suggesting that this model accounts for 18% of the variance in behavioral intention to use a VA.

Table 4

Reduced Multiple Regression Model for the Sri Lankan Sample.

b β SE t p LLCI ULCI

Collectivism .10 .06 .11 .85 .435 -0.15 0.30

Normative Beliefs .50 .41 .12 5.66 .001 0.27 0.72

Note. LLCI= Lower level confidence interval, ULCI= Upper level confidence interval.


As seen in Table 4, the bootstrapped coefficients showed consistent results. Collectivism did not

significantly predict behavioral intention to use a VA: b = .10, p = .396, 95% CI [-0.15, 0.30]. However,

normative beliefs were able to significantly predict behavioral intention to use a VA: b = .50, p = .001,

95% CI [0.27, 0.72]. Furthermore, the bootstrapped 95% confidence interval does not cross zero,

indicating that the model generalizes to the population.

Netherlands: The average normative beliefs score observed for the sample was 2.82 (SD = 1.15).

Similar to the Sri Lankan sample, a moderated multiple regression analysis was performed using the tool

PROCESS Model 1 (Hayes, 2013) to test if H3 is supported within the Dutch sample. The model

summary was as follows: F(3,150) = 9.62, p <.001, R2 = .16, suggesting that as a set, collectivism,

normative beliefs, and the interaction between the two account for about 16% of the variance in

behavioral intention to use a VA. Table 5 illustrates the conditional and interaction effects individually.

Table 5

Conditional and Interaction Effects Derived From the Multiple Regression for the Dutch Sample.

β SE t p

Collectivism -.23 .13 -1.74 .085

Normative Beliefs .60 .11 5.29 <.001

Interaction -.10 .11 -.80 .396

When interpreting Table 5, normative beliefs do not have a significant moderating effect on the

relationship between collectivism and behavioral intention to use a VA, β= -.10, t(150) = -.80, p =.396,

thereby not supporting H3. Furthermore, the addition of the interaction did not bring by a significant

change to the model as a whole, nor did it meaningfully contribute to its explanatory power,

F(1,150)= .62, p =.431, R2 change = .00.

As the interaction effect was non-significant, the interaction term was removed and a multiple

regression analysis was done on the reduced model as a follow up analysis. The diagnostics suggested

that the residuals were not normally distributed (behavioral intention to use a VA kurtosis z score = -

3.15), as such bootstrapping based on 1000 samples was done prior to continuing with the analysis. The

reduced model summary was as follows: F (2, 151) = 14.15, p<.001, R2= .16, suggesting that the reduced

model accounts for 16% of the variance in behavioral intention to use a VA within the Dutch sample.


Table 6

Reduced Multiple Regression Model for the Dutch sample.

b β SE t p LLCI ULCI

Collectivism -.22 -.12 .14 -1.70 .096 -.51 .10

Normative Beliefs .60 .40 .11 5.23 <.001 .35 .80

Note. LLCI= Lower level confidence interval, ULCI= Upper level confidence interval.

Furthermore, as seen in Table 6, collectivism did not significantly predict behavioral intention to

use a VA; b = -.22, β= -.12, p = .096, 95% CI [-.51, .10]. However, normative beliefs was able to

significantly predict behavioral intention to use a VA; b =.60, β= .40, p <.001, 95% CI [.35, .80].

Furthermore, the bootstrapped 95% confidence interval does not cross zero, indicating that the model

generalizes to the population.

4.3 Summary of Results

Table 7

Summary of Hypothesis Test Results

Hypothesis Sri Lanka Netherlands

H1: Collectivist cultural dimension traits have a negative

relationship with perceived usefulness of a VA

Not supported Supported

H2: Power distance has a negative relationship with

behavioral intention to use a VA

Not supported Not Supported

H3: The relationship between an individual's collectivist

cultural dimension traits and their intention to use a VA

Not supported Not supported

Discussion and Conclusions

This research set out to answer the question of how cultural dimensions, specifically collectivism

and power distance correlate to the degree to which people accept a VA. To answer this question, three


hypotheses were developed and tested. The following section will discuss the conclusions derived from

the hypothesis tests summarized in section 4.3.

5.1.1 Conclusions for the Sri Lankan Sample

With regard to the Sri Lankan sample, the results derived from H1 suggest that within this

sample, collectivist cultural dimensions of an individual (individualism) have a non-significant positive

correlation with perceived usefulness of a VA. This does not support H1 and contradicts its proposed

direction. Furthermore, it is not consistent with the theoretical findings proposed by Yeniyurt and

Townsend which suggest that collectivism has a negative relationship on the acceptance of new products

or innovations (2003).

One reason for the contradicting direction of the observed result could be attributed to the

measurement instrument used for this analysis. Specifically, the subscale measuring collectivism may

have hidden qualitative biases that could have potentially contributed to a measurement error. Item

phrases such as “individual gain” imply selfishness which has a negative connotation (Van Essen,

Thomas, Van Berkum, Chorus, 2016). This may evoke cognitive dissonance in participants when

responding, despite the survey being anonymous. This effect may be especially pronounced in the Sri

Lankan sample where selflessness is embedded deeply into the culture through majority Buddhist

practices such as “Dhaana” or almsgiving (Deepananda, Amarasinghe & Jayasinghe-Mudalige, 2016),

thus rendering the item unreliable. More objective and neutral phrasing would help increase the reliability

of the subscale and obtain a more accurate measurement of collectivism.

Alternatively, the rejection of the hypothesis may be due to an effect of a third variable not

measured in this analysis, or perhaps because of the existence of a nonlinear relationship such as a

quadratic relationship. Therefore, further statistical analysis on an improved measurement instrument

which takes the possible effect of confounding or control variables such as living situation and disposable

income into account, may help identify and confirm the nature of these relationships within the sample.

In reference to H2 within the Sri Lankan sample, the results obtained from the one-tailed

correlation analysis of power distance and behavioral intention to use a VA did not support the

hypothesis. It showed a negative but non-significant relationship between the two constructs. Even though

a negative correlation was observed consistent with literature (Baptista and Oliveira, 2015; Venkatesh et

al. 2003), the low effect size suggests that it is too weak to partially support the hypothesis, thereby

concluding that power distance is not directly correlated to behavioral intention to use a VA, within this

sample. This outcome may also have been influenced by possible errors in the measurement instrument

used. As highlighted in section 4.1, the power distance subscale faced reliability issues with regards to its

low Cronbach's alpha. A similar problem was found in the original subscale adopted by Srite and


Karahanna (2006) who in turn omitted four items to correct this before finalizing the instrument for

distribution. This may represent an indication that the subscale is weak in general and requires

reformation in its entirety.

Furthermore, the low mean scores for power distance contradict the findings of extensive research

conducted by Hofstede (Hofstede, 2001; Hofstede, et al., 2010; Hofstede Insights, 2018) and local

research (Dissanayake, 2015) which suggest that the country has a very high power distance. This

discrepancy is most likely due to the sample tested. The sample only consisted of students or recent

graduates who are not in a relative position of power when considering professors or managers. They

often feel disengaged and desire to be listened to by their superiors and this plight is reflected in the

politically charged protests occurring across the country (Haviland, 2018). These underlying emotions

may have played a role on how they scored this subscale; as a reflection of their desired reality as

opposed to their reality. Perhaps basing the subscale on a different context as opposed to their reality

could derive a more reliable response or statistically significant result (e.g., Pilots should make most

decisions without consulting their co-pilots). Alternatively, the rejection of the hypothesis may have also

been due to confounding or control variables that have not been accounted for in the research, the nature

of the relationship itself, or the sample size. Further analysis of the data using an improved measurement

instrument is essential to arrive as a more definite conclusion.

H3 proposing normative beliefs as a moderator for the relationship between collectivism and

behavioral intention to use a VA was rejected through statistical analysis. Thus, suggesting that

collectivism may not have a direct impact on behavioral intention to use a VA when moderated by

normative beliefs. The follow-up analysis done on the reduced model also suggested that collectivism was

not a predictor of behavioral intention to use a VA, but normative beliefs was. This suggests that the

intention to use non-essential technologies such as VAs vastly depends on the recommendations of

trusted individuals, regardless of their level of collectivism. This is interesting as conventional wisdom

would argue that normative beliefs are endogenous to collectivism, however, this is not reflected in either

one of the tested models for this sample. Nonetheless, the change in the model specifications did not have

a significant impact on exploratory power which suggests that further analysis including other variables is

necessary to gain a better understanding of the model and its performance.

It could also be argued that this outcome may also have been influenced by an error in the

measurement instrument. As pointed out in Srite and Karahanna (2006), the subscale for normative

beliefs included both one’s inner circle (i.e., family, friends, and classmates) and outer circle (i.e.,

professors and managers). However, it is unclear if Sri Lankans are equally influenced by their outer

circle as they are by their inner circle. It could be postulated that students may have an average high


normative belief score for their inner circle, but a lower score for their outer circle, thereby rendering the

measurement inaccurate.

Furthermore, given that the technology being tested is a non-essential tool (VA), it could be that

participants are unable to accurately decide if their inner or outer circle would want them to use this

technology. Realistically, it is likely that non-essential technologies such as VAs would not be an

encouraged purchase, especially in a lesser developed country like Sri Lanka (GDP per capita: USD

4350.00) (World Bank, 2019). This would depend on the disposable income of the respondents, which

have not been accounted for in this analysis.

5.1.2 Conclusions for the Dutch sample

H1: Collectivist cultural dimension traits have a negative relationship with perceived usefulness

was supported within the Dutch sample, suggesting that collectivism and perceived usefulness of a VA

are significantly negatively correlated. This means that those exhibiting individualistic cultural traits (low

collectivist scores) will find VAs more useful. As theorized in prior research (Yeniyurt & Townsend,

2003), this finding supports the argument that people with individualistic cultural traits value new

technologies and find them useful in daily life. This information could be advantageous to manufacturers

and public relations companies when marketing VAs to new demographics. The finding also contributes

to the research question by establishing that there is a direct correlation between the cultural dimension of

collectivism and the TAM construct the perceived usefulness. However, it is important to note that this

correlation is weak, this may be improved by amending the measurement instrument based on the

suggestions made for the Sri Lankan sample. Further research needs to be conducted keeping in mind the

identified drawback with regard to the subscale to solidify this significant finding.

The results obtained from the testing of H2 within the Dutch sample showed a positive non-

significant relationship between power distance and behavioral intention to use a VA, thus contradicting

the hypothesized direction and rejecting the hypothesis proposed. However, the result is near to being

statistically significant at 5%. As the low mean power distance score observed in this sample aligns with

findings in existing research (Hofstede Insights, 2018; Hennekam, 2015), the subscale does not give

reason to suggest errors in its ability to measure power distance. Based on the direction of the correlation,

it could be speculated that Dutch people perceive a VA as another entity and thus, giving it voice

commands is considered a delegation of tasks to ‘another’, similar to enlisting a personal secretary to run

an errand. However, a larger sample is required to test if a significance can be observed then. Regardless

this will need to be interpreted with caution as the effect size remains small.

H3 was also rejected for the Dutch sample as no significant moderating effect by normative

beliefs was found between collectivism and behavioral intention to use a VA. In addition, the reduced


model did not show a difference in exploratory power. However, the reduced model did confirm the

predictive power of normative beliefs on behavioral intention to use a VA within the sample. A similar

discussion to that of the Sri Lankan sample could be made in this instance. Perhaps young people place

greater trust on the opinions of those closer to them when purchasing non-essential technology tools such

as VAs. Again, the non-significance of the predictive power of collectivism is an interesting outcome, as

inherently, the value placed on the opinions of others is intertwined with one’s level of collectivism.

However, it is difficult to interpret why this may be without ensuring the measurement instrument is

operating optimally. As such, it is important to reiterate that the same drawbacks applied to the reliability

of the normative beliefs subscale for the Sri Lankan sample may also be applicable here. Therefore,

further research with a more robust measurement instrument is necessary to explore if these changes will

yield a significant moderating effect for the original model of this sample, or a significant predictive

power for collectivism in the reduced model.

5.2 Limitations

The objective of this study was to address the research question of how cultural dimensions such

as collectivism and power distance correlate to the degree to which people accept a VA. Based on the

conclusions made in sections 5.1.1 and 5.1.2, this question has been sufficiently addressed. However, this

research has numerous limitations. In addition to the drawback of the measurement instrument

extensively discussed in sections 5.1.1 and 5.1.2, a larger sample size could have strengthened the

reliability of the significance testing.

Furthermore, the survey was distributed at a time where respondents are burdened with exams

and filling out multiple surveys for ongoing research during the end of the semester. It may be that this

environment made respondents answer hastily to questions, thereby compromising quality and reliability

of the survey responses. It is advised that future researches provide incentive for respondents to accurately

fill in the survey distributed.

Additionally, another element to take into consideration is the video that the respondents based

their responses for the TAM constructs on. The two videos used were promotional content produced by

Amazon, as such it is created to attract customers. This may have impacted the responses on perceived

usefulness and perceived ease of use of VAs. It is beneficial for future studies using this methodological

approach to create their own video content that highlights both the features and drawbacks of VAs to

probe a more unbiased response from the participants.

Finally, in regards to generalizability, given the limitations of the measurement instrument it is

advisable to duplicate this research on an improved measurement instrument in order to more accurately

make these assumptions.


5.3 Scope for future research

This research is not a comparative analysis between Sri Lanka and The Netherlands. Instead, it is

a correlational study exploring the possible relationships between cultural dimensions and TAM

constructs within these two countries. A possible future study, after conducting a measurement invariance

analysis, could be a comparative analysis measuring technology acceptance towards VAs between Sri

Lanka and The Netherlands. This approach could possibly derive interesting insights into how

comparable cultural differences between the countries influence technology acceptance towards VAs.

Additionally, the study can also be extended to include other cultural dimensions outlined by

Hofstede. For instance, is uncertainty avoidance correlated to technology acceptance of VAs? Uncertainty

avoidance may have interesting correlations to the technology acceptance of VAs as the tool relies

heavily on its ability to connect to household appliances, security devices, and your schedule to operate

optimally. Giving a VA this power may be difficult for individuals that are risk averse and skeptical of

delegating ‘responsibility’ to machines. Another possible research question would be exploring the

influence of inner and outer circle normative beliefs on behavioral intention to use a VA. Since the

reduced models testing H3 suggests that normative beliefs predict behavioral intention to use a VA, there

could be significant societal contributions, especially in the areas of development, marketing, and

communications to be made by exploring which circle (inner or outer) influences a user more.

In conclusion, the scientific and societal contributions of this research have provided interesting

insights into technology acceptance research and potential marketing methods for communicating VAs to

two demographics. Furthermore, the commentary made on the measurement instrument used to

operationalize the widely accepted cultural dimension theory and TAM act as a solid base upon which to

conduct further studies on the intersection between culture and technology acceptance. In addition, the

proposed scope for future research act as an indicator of the future direction of technology acceptance

research within AI. Thereby, this research will undoubtedly contribute to yielding interesting and

impactful insights into the future of VAs in a growing global consumer market.


References

Abbasi, M. S., Tarhini, A., Elyas, T., & Shah, F. (2015). Impact of individualism and collectivism over

the individual’s technology acceptance behaviour: A multi-group analysis between Pakistan and

Turkey. Journal of Enterprise Information Management, 28(6), 747-768.

Amazon (2017, October 29) Amazon Echo Plus #AskAlexa [Video File]. Retrieved from

https://www.youtube.com/watch?v=9qkW75JsY3U&ab_channel=AmazonEchoIndia

Amazon (2015, June 23) Amazon Echo - Now Available [Video File]. Retrieved from

https://www.youtube.com/watch?v=FQn6aFQwBQU&ab_channel=amazon

Bennett, D. A. (2001). How can I deal with missing data in my study? Australian & New Zealand Journal

of Public Health, 25(5), 464–469. https://doi.org/10.1111/j.1467-842X.2001.tb00294.x

Besikci, E., Agnew, C. R., & Yildirim, A. (2016). It's my partner, deal with it: Rejection

sensitivity, normative beliefs, and commitment. Personal Relationships, 23(3), 384-395.

Davis, F. D., Bagozzi, R. P., & Warshaw, P. R. (1989). User acceptance of computer technology: a

comparison of two theoretical models. Management science, 35(8), 982-1003.

Deepananda, K. A., Amarasinghe, U. S., & Jayasinghe-Mudalige, U. K. (2016). Neither bust nor boom:

Institutional robustness in the beach seine fishery of southern Sri Lanka. Ocean & Coastal

Management, 128, 61-73.

Dissanayake, D. M. S. (2015). A comparative study of positions of collectivism in Japanese and

Sri Lankan societies.

Dorfman, P. W., & Howell, J. P. (1988). Dimensions of national culture and effective leadership patterns:

Hofstede revisited. Advances in international comparative management, 3(1), 127-150.

Echo Dot Smart speaker with Alexa. (2018). Retrieved from https://www.amazon.com/Amazon-Echo-

Dot-Portable-Bluetooth-Speaker-with-Alexa-Black/dp/B01DFKC2SO

Haviland, C. (2018). Sri Lanka protest students evicted. Retrieved from

https://www.bbc.com/news/world-asia-16469642

Hayes, A. F. (2013). Introduction to mediation, moderation, and conditional process analysis.

Hennekam, S., & Tahssain-Gay, L. (2015). Changing attitudes towards diversity: The Netherlands and

Morocco compared. Management Decision, 53(9), 2135-2155.

Hofstede Insights. (2018). Retrieved from https://www.hofstede-insights.com/

Hofstede, G. (2001). Culture’s consequences: Comparing values, behaviors, institutions, and

organizations across nations. Thousand Oaks, CA: Sage Publications.

Hofstede, G., Hofstede, G. J., & Minkov, M. (2010). Cultures and Organizations: Software of the Mind.

Revised and expanded third edition. New York.



https://doi.org/10.1111/j.1467-842X.2001.tb00294.x

https://www.amazon.com/Amazon-Echo-Dot-Portable-Bluetooth-Speaker-with-Alexa-Black/dp/B01DFKC2SO

https://www.amazon.com/Amazon-Echo-Dot-Portable-Bluetooth-Speaker-with-Alexa-Black/dp/B01DFKC2SO

https://www.bbc.com/news/world-asia-16469642


Koetsier, J. (2018). Aug 2, 2018, 07:44 pm Amazon Echo, Google Home Installed Base Hits 50

Million; Apple Has 6% Market Share, Report Says. Retrieved from

https://www.forbes.com/sites/johnkoetsier/2018/08/02/amazon-echo-google-home-

installed-base-hits-50-million-apple-has-6-market-share-report-says/#15e1bbea769c

Lai, P. C. (2017). The literature review of technology adoption models and theories for the novelty

technology. JISTEM-Journal of Information Systems and Technology Management, 14(1), 21-38.

Mohammed, Z. A., & Tejay, G. P. (2017). Examining privacy concerns and ecommerce adoption in

developing countries: The impact of culture in shaping individuals’ perceptions toward

technology. Computers & Security, 67, 254–265.

https://doi.org/10.1016/j.cose.2017.03.001http://search.ebscohost.com/login.aspx?direct=true&db

=bsh&AN=94947442&site=ehost-live

Muk, A., & Chung, C. (2015). Applying the technology acceptance model in a two-country study of SMS

advertising. Journal of Business Research, 68(1), 1-6.

Oechslein, O., Fleischmann, M., & Hess, T. (2014, January). An application of UTAUT2 on social

recommender systems: Incorporating social information for performance expectancy. In System

Sciences (HICSS), 2014 47th Hawaii International Conference on (pp. 3297-3306). IEEE.

Ra, J. S., & Cho, Y. H. (2018). Role of social normative beliefs as a moderating factor in smoking

intention among adolescent girls in Korea. Nursing & health sciences, 20(4), 530-536.

Rauniar, R., Rawski, G., Yang, J., & Johnson, B. (2014). Technology acceptance model (TAM) and social

media usage: an empirical study on Facebook. Journal of Enterprise Information Management,

27(1), 6-30.

Rogers, E. M. (1995). Lessons for guidelines from the diffusion of innovations. Joint Commission Journal

on Quality and Patient Safety, 21(7), 324-328.

Smith, C. (2018). Alexa and Siri Can Hear This Hidden Command. You Can’t. Retrieved from

https://www.nytimes.com/2018/05/10/technology/alexa-siri-hidden-command-audio-attacks.html

Srite, M., & Karahanna, E. (2006). The role of espoused national cultural values in technology

acceptance. MIS quarterly, 679-704.

Straub, D., & Burton-Jones, A. (2007). Veni, vidi, vici: Breaking the TAM logjam. Journal of the

Association for Information Systems, 8(4), 5.

Tarhini, A., Hone, K., & Liu, X. (2015). A cross‐ cultural examination of the impact of social,

organisational and individual factors on educational technology acceptance between British and

Lebanese university students. British Journal of Educational Technology, 46(4), 739-755.

Tarhini, A., Hone, K., Liu, X., & Tarhini, T. (2017). Examining the moderating effect of individual-level

cultural values on users’ acceptance of E-learning in developing countries: a structural equation

https://www.forbes.com/sites/johnkoetsier/2018/08/02/amazon-echo-google-home-installed-base-hits-50-million-apple-has-6-market-share-report-says/#15e1bbea769c

https://www.forbes.com/sites/johnkoetsier/2018/08/02/amazon-echo-google-home-installed-base-hits-50-million-apple-has-6-market-share-report-says/#15e1bbea769c

https://doi.org/10.1016/j.cose.2017.03.001

https://www.nytimes.com/2018/05/10/technology/alexa-siri-hidden-command-audio-attacks.html


modeling of an extended technology acceptance model. Interactive Learning Environments,

25(3), 306-328.

Taber, K. S. (2017). The use of cronbach’s alpha when developing and reporting research instruments in

science education. Research in Science Education, 1-24.

Van Essen, M., Thomas, T., van Berkum, E., & Chorus, C. (2016). From user equilibrium to system

optimum: a literature review on the role of travel information, bounded rationality and non-selfish

behaviour at the network and individual levels. Transport reviews, 36(4), 527-548.

Davis, F. D., & Venkatesh, V. (1996). A critical assessment of potential measurement biases in the

technology acceptance model: three experiments. International Journal of Human-Computer

Studies, 45(1), 19-45.

Venkatesh, V., & Davis, F. D. (2000). A theoretical extension of the technology acceptance model: Four

longitudinal field studies. Management science, 46(2), 186-204.

Venkatesh, V., Morris, M. G., Davis, G. B., & Davis, F. D. (2003). User acceptance of information

technology: Toward a unified view. MIS quarterly, 425-478.

Venkatesh, V., Thong, J. Y., & Xu, X. (2012). Consumer acceptance and use of information technology:

extending the unified theory of acceptance and use of technology. MIS quarterly, 157-178.

World Bank. (2019). World Bank national accounts data, and OECD National Accounts data files.

Retrieved from https://www.worldbank.org/en/country/srilanka

Yuan, S., Ma, W., Kanthawala, S., & Peng, W. (2015). Keep using my health apps: Discover users'

perception of health and fitness apps with the UTAUT2 model. Telemedicine and e-Health, 21(9),

735-741.


Appendix A

Videos shown during the survey:

● https://www.youtube.com/watch?v=FQn6aFQwBQU&ab_channel=amazon (For Dutch

Participants)

● https://www.youtube.com/watch?v=9qkW75JsY3U&ab_channel=AmazonEchoIndia (For Sri

Lankan Participants)




Appendix B

Scale used for the survey:

Collectivism:

● Being accepted as a member of a group is more important than having autonomy and

independence.

● Being accepted as a member of a group is more important than being independent.

● Group success is more important than individual success.

● Being loyal to a group is more important than individual gain.

● Individual rewards are not as important as group welfare.

● It is more important for a professors/manager to encourage loyalty and a sense of duty in

students/subordinates than it is to encourage individual initiatives.

Power distance:

● Professors/Managers should make most decisions without consulting students/subordinates.

● Professors/managers should not ask students/subordinates for advice because they might appear

less powerful.

● Decision-making power should stay with senior professors/senior management in the

institution/organization and not be delegated to lower level students/employees.

● Students/employees should not question their professor's/manager's decisions.

Perceived usefulness:

● Using a Virtual Assistant would enhance my productivity.

● I find Virtual Assistants would be useful in my daily activities.

● Using Virtual Assistants would enhance the effectiveness of my daily life.

● Using Virtual Assistants would improve my daily performance.

Perceived ease of use:

● It would be easy for me to become skillful at using a Virtual Assistant.

● I would find Virtual Assistants easy to use.

● I would find it easy to get a Virtual Assistant to do what I want.

● Learning to operate a Virtual Assistant would be easy for me.

Behavioural intention to use a VA:

● I intend to use a Virtual Assistant in my daily life.


● I intend to use a Virtual Assistant frequently in my daily life.

Normative beliefs:

● My relatives would think I should use a Virtual Assistant.

● My friends would think I should use a Virtual Assistant.

● My professors/managers would think I should use a Virtual Assistant.

● I believe that my classmates/colleagues would think I should use a Virtual Assistant.

THE IMPACT OF CULTURAL DIMENSIONS ON THE TECHNOLOGY ...

Documents