The techno-stressor effect of intrusive technological characteristics on strain in the public context and the moderating role of age and self-

The techno-stressor effect of intrusive technological characteristics on strain in the public context and the moderating role of age and self-

efficacy.

Bart van de Vliert S2340526

MSc Change Management

Abstract:

In this research we expand on the techno-stress literature in two ways. First, we test the effect of intrusive ICT characteristics in a public context. We empirically test and adapted stress

model from the work-context to investigate which stressors are relevant within the public context. Second, we examine the moderating effects of age and self-efficacy on the relationship between strain and stressors within the working population of our sample.

Structural equation modeling was used on a sample of 152 people. Findings indicate that within the public context presenteeism influences invasion of privacy and invasion of privacy

influences strain. Within the working population it was found that presenteeism influences invasion of privacy and role ambiguity, which both influence strain. Self-efficacy was found to have a moderating role between role ambiguity and strain. The results provide insight in the

moderating role of both age and self-efficacy in the stress literature and raise questions on the applicability of certain measures used to examine the public context.

Supervisor: M. L. Hage Co-assessor: I. Maris-de Bresser

Date: 20-6-2017

Word count: 14379 (excl. references)

University of Groningen

2

Introduction

Communication and information technologies (ICT) play a large role in our modern society. They not only influence our work, but also have a huge effect on our personal life.

Technologies and their associated responsibilities and primary tasks are rapidly becoming more modern (Benbasat & Zmud, 1999). ICT are in a process of evolving and have evolved from terminals and desktop computers to for example smartphones. (Somogyi & Galliers, 1987). ICT is no longer only used for work but is now rapidly introduced for study, social and other purposes (Middleton & Cukiers, 2006).

Half of the world’s population is already connected through the internet (Internet world stats, 2017) and more than half of the world’s population currently uses a mobile phone (Statista, 2017). These are just two examples of how widespread the use of ICT has become.

In contrast to the positive aspects of ICT usage, the negative consequences of using ICT are often overlooked and more research is needed in this area (D’arcy, Gupta, Tarafda & Turel;

2014). It is very important to learn of the effects that these negative consequences have and how to deal with them. This will make it significantly easier to mitigate the future negative consequences of newly introduced ICTs. One example of a possible negative consequences was shown by Middleton & Cukiers (2006) who found that the introduction of mobile devices caused techno-stress among the users of these devices.

When individuals experience stress due to ICT use, it is also called techno-stress (Ragu- Nathan et al, 2008). Techno-stress can contribute to a wide range of health and other issues that can have far reaching consequences for the people experiencing this form of stress (Cooper et al. 1996). Because ICT is becoming more accessible for a larger audience, techno- stress is a form of stress that is likely to occur more frequently and therefore particularly interesting to look at. The World Health Organization (WHO) claims that actions aimed to reduce the negative consequences of ICT use were primarily targeted at physical risks of ICT usage and largely ignored the psychosocial risks and the effect on mental health (WHO, 2015). This study will focus specifically on the psychosocial effects of ICT usage.

Within the research focusing on the psychological effects of ICT usage, most research has

focused on stress resulting from the use of ICT (techno-stress) (Brod, 1984). Focus within the

techno-stress literature has shifted from a “disease” caused by someone’s inability to cope

with ICT in a healthy way (Brod, 1984) to stress caused by the inability to cope with demands

of organizational computer usage (Tarafdar, Tu & Ragu-Nathan; 2010). The literature shifted

3

focus to cover technostress in a work context and has mainly ignored the effects of techno- stress on users outside of the work context as well as the individual differences that could explain differences in experienced techno-stress. The focus on the workplace techno-stress is understandable since ICTs were mainly introduced for work in the first place. However, nowadays ICT are not only used in a work context but are an integral component of our lives and are often used in a public context (Czaja et al., 2006). Therefore, more research is needed on techno-stress within the public context.

Despite the usage of ICT leading to stress (Ragu-Nathan et al., 2008) it is still unclear which technology characteristics cause this stress. Ayyagari et al. (2011) came up with a framework to explain the effects of technology characteristics on technostress. They identified three technology characteristics: usability, intrusiveness and dynamism. Of these

characteristics, there is are repeated calls for a better understanding of the effect that intrusiveness has on stress (Speier et al., 2003; Ayyagari et al., 2011). Recognizing the importance of intrusive characteristics and the intrusive nature of ICT (McFarlane &

Lotorella; Weil & Rosen, 1997), we want to adhere to these calls and add to the literature by examining the effects of intrusiveness of an ICT on stress.

Because of the increasing private usage of ICTs in non-workplace environments, I will look at techno-stress not from a workplace perspective but from a public perspective. In this perspective techno-stress is considered in all instances of ICT use, not only when experienced during or at work. We build on the model of Ayyagari et al. (2011), focusing specifically on intrusiveness and examine whether this model is valid in the public context. Therefore, the first research question will be:

What are the effects of ICT characteristics on techno-stress in a public context?

Besides taking into consideration the technological characteristics previously

mentioned, examining individual characteristics could provide a more complete overview of differences in experienced techno-stress by individuals. Literature from the “digital divide”

stream provides variables that explain individual differences in experience and coping with ICT (Morrell, Mayhorn & Bennett, 2000; Rogers, Meyer, Walker & Fisk, 1998) and provides insights in the moderating role of demographic and usage-related variables. Within this digital divide literature age (Sharit, Czaja, Nair & Lee, 2003), and self-efficacy beliefs (Ellis &

Allaire, 1999) have shown to influence the stress experienced by individuals (Folkman,

Lazarus, Pimley & Novacek, 1987; Jerusalem & Schwarzer, 1992). However, recent

4

technostress research has controlled for the effect of ICT usage and demographic variables (Ayyagari et al., 2011, Tarafdar, Tu & Ragu-Nathan; 2010), but have not considered the possibility of these variables playing a moderating role. I want to add to the technostress literature by using age and self-efficacy to show the individual differences in experienced stress more clearly. Therefore, the second research question will be:

What are the effects of age and self-efficacy beliefs on the relationship between techno- stress and technology characteristics?

In the remainder of this paper I will first elaborate on the different variables of this study, their background and provide the subsequent hypotheses. After the literature section I will present the research methods, followed by the data analysis per research question and lastly the results. Lastly, I will go in depth and discuss the meaning of my findings and whether they are in line with previous research.

Literature review

Background

Two theoretical paradigms are often used in stress phenomenon literature (Fox et al., 1993). First there is the epidemiological perspective (Karasek, 1979. This view takes

occupational conditions or characteristics and relates them to actual manifestations of a

disease, for example the condition workload can cause heart disease. Objective measures are

used to measure stressors and their subsequent outcomes (Fox et al., 1993). The second

paradigm is called the cognitive perspective (Miller, 1979). This paradigm stresses the

importance of people cognitively interpreting and appraising environmental demands and

creates stress. This paradigm focusses on how people interpret their environmental demands

and subjective measures are used. Nowadays there is a growing consensus for viewing stress

as originating neither from the individual or from the environment, but from an interaction

between the two (Edwards, 1991; Lazarus, 1990). Stress then arises from a situation in which

a person perceives that the demands arising from the environment exceed his own resources

and eventually threatening his well-being (Lazarus, 1990, Cooper et al., 2001). This whole

process is also referred to as stress. The stimuli that an individual encounter are referred to as

stressors and an individual’s psychological response to these stressors is also referred to as

strain (Cooper et al., 2001). This third synthesizing perspective is also called the person-

5

environment (P-E) model and is currently one of the most used models in stress literature (Edwards, 1991; Ayyagari et al., 2011).

Person-Environment fit

The essence of this person-environment (P-E) model is that there is a balance between an individual and their environment, which can be disturbed. The disturbed balance between an individual and his/her environment are a consequence of ignored individual needs or job demands that are not met, which in turn lead to strain (Cooper et al., 2001). This research focusses on technology instead of environment fit and therefore a refined model of the P-E fit model is used called the person-technology (P-T) model. This model is a refinement of the P- E model designed to capture the perceived fit or misfit between an individual and an ICT they are using instead of the general environment (Ayyagari et al., 2011).

There are two ways in which a misfit could occur between the person and an ICT. The first way is that there is a difference between values of a certain person and the characteristics of an ICT. Values can be defined as a conscious desire of an individual and includes interests and preferences (Edwards and Cooper, 1990; French et al., 1982). When an individual has a conscious desire about the functioning of an ICT and the ICT does not conform to this desire, a difference will be created between the values of an individual and the ICT. This difference in ICT characteristics and values of an individual is called a misfit. The second way in which a misfit could occur is when the ICT places demands on an individual that exceed the

individual’s abilities. Abilities can be defined as energy, knowledge, time and skills that an individual has (Ayyagari et al., 2011). When the intrusive characteristics of an ICT demand an individual to invest time and energy into the ICT and the individual cannot cope with these demands, the ICT exceeds the individual’s skillset (Chishold et al., 1983) and a misfit is formed. In both ways that misfit occurs, the individual experiencing this misfit will have a certain psychological response to this misfit called strain. The fit between the

values/characteristics and the demands/abilities are two complimentary approaches (Kristof, 1996). These perceived misfits will manifest in stressors and can eventually lead to strain.

Stressors are factors that are sources of strain (Cartwright & Cooper, 1997). An example is

ICT overload. If there is a misfit between an individual’s capabilities and the capabilities that

an ICT requires me to have, this leads to a stressor called ICT overload. ICT overload is the

factor in which the misfit is manifested and is called a stressor. The psychological response to

this stressor, how an individual deals with this case of ICT overload is called strain. Nelson

(1990) and Ayyagari et al (2011) state the need for deeper investigation of the ICT artifact by

6

focusing on ICT characteristics instead of treating ICT in an undifferentiated manner.

Therefore, we will first elaborate on these ICT characteristics, after which we will expand on the concept of stressors.

Intrusive characteristics

Because of this need for deeper investigation of the ICT artifact by focusing on ICT characteristics, this study will build on the important work of Ayyagari et al. (2011) and examine how ICT characteristics lead to strain in a public context and examine the moderating role of age and self-efficacy.

Within recent IS literature there are three dominant ICT characteristics identified: The usability features, the dynamic features and the intrusive features (Ayyagari et al., 2011). We focused on intrusive characteristics because of the importance of intrusive characteristics and the intrusive nature of ICTs (McFarlane & Lotorella; Weil & Rosen, 1997). There is also a call for a better understanding on this intrusive nature of ICTs and the effects on stress (Speier et al., 2003; Ayyagari et al., 2011).

The intrusiveness construct is one of the most discussed factors in technostress literature (Ragu-Nathan et al., 2008), it is important in many articles within the techno-stress literature and within this research there are two recurring variables explaining intrusiveness:

Presenteeism and anonymity (Ayyagari et al., 2011; Ragu-Nathan et al, 2007; Kakabadse et al., 2000; Tu et al., 2005; Well and Rosen, 1997). Presenteeism refers to the degree to which ICT enables individuals to be reachable. There are differences in ICT regarding reachability, some enable users to be reachable all the time but others don’t. ICTs and their communication possibilities often lead to increased communication flow and interrupt people’s attention and work (Straub & Karahanna, 1998) and these interruptions lead to stress and reduced

efficiency (McFarlane & Lotorella, 2002). The other variable often mentioned to explain intrusiveness is anonymity. Anonymity is the degree that an individual perceives that his or her use of an ICT is not traceable and cannot be referred back to them. In general, people do not like being tracked and monitored (Boyd, 1997)

Presenteeism and anonymity influence stressors by changing the earlier discussed

perceived person-technology (P-T) fit. ICT’s presenteeism and anonymity characteristics can

be different from an individual’s preferences, needs, expectations or values at that moment

(Ayyagari, Grover & Purvis, 2011). This perceived difference can in turn lead to an increase

in perceived stressors which can then in turn lead to an increase in the strain an individual

7

perceives. When someone is constantly available by using a smartphone, this person can experience that he/she can’t handle using the smartphone all the time. A misfit between the individual and the smartphone is established and in this case it is because of too much ICT usage, meaning that the stressor that is experienced is ICT overload. That person can have a psychological reaction to this stressor called strain.

Presenteeism and anonymity influence several stressors, which in turn influence perceived strain (figure 1). In the next section we develop the hypotheses to argue why the proposed relationships are likely to occur.

Figure 1: The conceptual model

Presenteeism

Anonimity

Invasion of privacy

Role Ambiguity Work Overload Work-Home

conflict

Characteristics Stressors

Strain

Strain

Age & Self efficacy beliefs

H1

H2a

H2b H3

H4

H6a-d

H7&8

8 Stressors

In the P-T fit model, four stressors are identified to be influenced by presenteeism and anonymity: Work-Home conflict, invasion of privacy, work overload and role ambiguity (Ayyagari et al., 2011; Cartwright & Cooper, 1997; Malhotra et al., 2004). Based on the literature, work overload was slightly adjusted to ICT overload to better suit the public context of this study. We also introduce social overload because this seems to be a variable that is particularly interesting in a public context (Maier, Laumer, Eckhardt & Weitzel, 2014).

First, work-home conflict can be defined as the conflict an individual perceives between demands of the work and family (Cooper et al., 2001). Individuals hold a number of roles and to fulfill these roles they consume resources like energy and time (Van Steenbergen &

Ellemers, 2009). The amount of resources is fixed and being involved in multiple roles leads to an uneven distribution of resources between these roles (Greenhaus & Powell, 2003).

Work-home conflict is a conflict in which pressures from public context and work context are incompatible (Greenhaus & Beutell, 1985). The constant connectivity provided by the

intrusive ICT characteristics comes at a cost of blurring the boundaries between work and home scenario’s (Mann and Holdsworth, 2003) and it has also been shown to be a source of strain (Duxbury and Higgins, 1991; Ayyagari at el., 2011). Because of the boundaries between work and home being blurred, a norm was created in which people expect other people to work from home (Middleton & Cukier, 2006). People are expected to be present and reachable every time and this presenteeism norm that is established increases demands on individual. When their abilities to meet these demands regarding presenteeism are lacking, misfits occur and the P-T gap increases. No theoretical relationship has been found that anonymity influences the conflict between work and public life. Therefore, I hypothesize:

H1: Among the working population in the sample, perceived ICT presenteeism will be positively related to perceived work-home conflict.

Second, invasion of privacy is the perception of an individual that his or her privacy is

no longer guaranteed (Eddy et al., 1999). People are increasingly concerned about their

privacy and the fact that their privacy could be compromised by computer technologies (Best

et al., 2006). While people are increasingly worried about computer surveillance and invasion

of privacy at work (Alge, 2001), the concept of invasion of privacy is much larger than only a

work setting. Malhotra, Kim & Agarwal (2004) found that all internet users experienced

privacy threats on the internet. ICTs apply pressure on people to be available all the time, not

only in a work context, but also in a public context. People value their privacy and these

9

intrusive ICT characteristics might not be able to match the values of people, a P-T misfit might be established. Therefore, I hypothesize:

H2a: Perceived ICT presenteeism will be positively related to invasion of privacy Presenteeism is not the only characteristic of ICTs that influence the feelings of individual’s regarding invasion of privacy. The anonymity characteristic is also experienced through this stressor. Anonymity is the degree that an individual perceives that his or her use of an ICT is not traceable and cannot be referred back to them. Of all the stressors, this is the only stressor that anonymity influences because it is the only stressor regarding privacy. In general, people do not like being tracked and monitored (Boyd, 1997) and research has shown that being tracked and monitored can be stressful for individuals (Parenti, 2001; Smith et al., 1992).

When an individual perceives these monitoring activities taking place, they might not match the individual’s values and the P-T gap will increase because of an anonymity misfit.

Therefore, I hypothesize:

H2b: Perceived ICT anonymity will be negatively related to perceived invasion of privacy.

The third stressor is ICT overload. work overload happens when an individual perceives that work exceeds their capabilities or level of skill (Moore, 2000). At first glance this measure seems to be inappropriate in a public context, however with minor adaptations it measures overload generated from using ICT. Within the overload literature there are two measurement instruments to measure overload (Moore, 2000). The first one is the emotional exhaustion measure by Maslach & Jackson (1981) and the second one is a measure by Pines, Aronson & Kafry (1981) called the tedium, which not only measures emotional but also physical, emotional and mental exhaustion. In this study ICT overload refers to when individuals perceive that they are emotionally exhausted by using ICT (Maslach & Jackson, 1981). The emotional exhaustion measure of overload has been found to cause strain in the public context and can thus be used in this study to assess ICT overload (Thompson, Kirk &

Brown, 2005).

Presenteeism can also manifest through the stressor ICT overload. ICT requiring individuals to be present all the time lead to continual task disruption for that individual (Tarafdar et al. 2010). Interruptions reduce the ability to sustain mental attention and reduce task accuracy (Cellier & Eyrolle, 1992). The inability to cope with the demands of the ICT lead to the user being continually distracted and frustrated and to strain (Heinssen, Glass &

Knight, 1987). These presenteeism pressures of ICT increase the demands that are placed on

10

people. The demands placed on individuals and the individual’s needs might not match. This mismatch results in a perceived misfit and the P-T gap increasing. No theoretical relationship has been found that anonymity increases or decreases overload. Therefore, I hypothesize:

H3: Perceived ICT presenteeism will be positively related to perceived ICT overload.

Fourth, role ambiguity refers to the lack of clarity on duties, objectives and

responsibilities that are required to fulfil an individual’s role (Bakker & Demerouti, 2014).

Role ambiguity has increased due to the multitasking options that ICTs enable (Kakabadse et al., 2000). These multitasking options and the interruptions of ICTs lead to uncertainty on which task or job to perform and thus constraining individual abilities (Ayyagari et al., 2011).

The measurement of role ambiguity as introduced by Rizzo, House & Lirtzman (1970) originates in the work context because of extensive research done on conflict in roles of the family and the work context. Schaubroeck et al. (1993) suggest that role clarification is needed for an individual to explicitly know their roles. Bauer & Simmon (2000) state that role clarification is most present in organizations and that because of clearer roles people

experience more role ambiguity. If people are not clear what their specific roles are they feel less ambiguity. Therefore, to have a clear measure of role ambiguity we focus on the work context for this stressor. Literature about role ambiguity during work states that using ICTs during work causes an individual to lose focus because of a need for attention of the ICT (Davis, 2002). The attention that is given to the ICT cannot be given to the work at the same time resulting in a role ambiguity on what to do, focus on work or give attention to the ICT.

This constrains an individual’s abilities. While certain ICTs do not necessarily have to get attention, individuals within an organization or system can create norms (Davis, 2002) which puts pressure on individuals to for example respond quickly. Presenteeism creates uncertainty for people working on which tasks to perform. The constant connectivity that is supplied by intrusive characteristics might not match with the individual’s expectations. This increases the P-T gap and leads to a presenteeism misfit. No theoretical relationship has found that

anonymity increases or decreases role ambiguity for an individual. Therefore, I hypothesize:

H4: Among the working population in the sample, people’s perception of ICT presenteeism will be positively related to perceived role ambiguity.

Fifth, the stressor social overload is adapted from Maier et al. (2014). The concept is

adapted from the social networking sites context to fit the ICT context because social

11

networking sites are often accessed through ICT. The relationship between ICT and social networks has been acknowledged in books and journal publications (Ellison & Boyd, 2013;

Boden and Molotoch, 1994). Being accepted and being connected by peers is very important for people (Desjarlais & Willoughby, 2010). Sometimes staying in touch with others can be a burden for people providing social support (Koroleva et al, 2011). Since ICTs are used to connect people socially and to access social networks the usage of such ICTs can lead to social overload (Maier et al., 2014). This constant connectivity of the intrusive characteristics of ICTs might not match the social values or needs of individuals. This increases the P-T gap and leads to a presenteeism misfit. No theoretical relationship has found that anonymity increases or decreases social overload for an individual. Therefore, I hypothesize:

H5: People’s perception of ICT presenteeism will be positively related to perceived social overload.

Lastly, according to Jones & Bright (2001) it is important to differentiate between the environmental demands (stressors) and the psychological reaction to these demands (strain).

all of these previously mentioned stressors lead to a psychological reaction to these stressors called strain (Cooper et al., 2001). Kinman & Jones (2005) also showed the relationship between stressors and strain as every stressor results in a psychological reaction. Therefore, I hypothesize:

H6a: ICT overload is positively related to people’s perceived strain.

H6b: role ambiguity, is positively related to people’s perceived strain.

H6c: work-home conflict is positively related to people’s perceived strain.

H6d: invasion of privacy is positively related to people’s perceived strain.

Besides trying to confirm the findings Ayyagari et al. (2011) and where possible trying to expand the concepts outside of the work context, another goal of this research was to examine the role of self-efficacy and age as moderating variables. These variables originate from the digital divide literature. The interpretations of what the digital divide is ranges from the simple distinction between being able to access ICT or not (Rice & Katz, 2003) to

multilevel definitions going beyond access and use. In this research the expanded view of the

digital divide is interesting because it also includes the capability to exploit ICT and the

outcomes of exploiting ICT (Wei, Teo, Chan & Tan, 2011). Because strain can also be seen as

an outcome of using ICT the digital divide defined in this broader sense could also help to

12

explain differences in experienced strain by individuals Even though not much is written about the differences individuals experience in stress and strain, there are some indications that there might be individual differences because of certain individual characteristics. We selected age and self-efficacy because they have shown been shown to have a relationship with strain (Folkman, Lazarus, Pimley & Novacek, 1987) and because within techno-stress literature age and self-efficacy are often used as control variables making it interesting to examine possible moderating effects (Ayyagari et al., 2011, Tarafdar, Tu & Ragu-Nathan;

2010).

In the literature it is clear that a divide has been established between younger and older people (Lenhart, Rainie, Fox, Horrigan, & Spooner, 2000). There are differences in adoption between younger and older people (Sharit & Czaja, 1994) and research has also shown that there are differences in processing complex stimuli resulting from technology (Plude &

Hoyer, 1985). Age might change experienced stress as research has found that older adults expressed less comfort in using technology and had less confidence in their ability to use systems successfully (Czaja & Sharit, 2006; Ellis & Allaire, 1999; Tacken, Marcellini,

Mollenkopf, Ruoppila & Szeman, 2005). Also, research has found that older people in general experience more stress than younger people (Rudolph & Hammen, 1999). Individuals might experience the same stressors, but with the differences in comfort of use, adaptation and processing I argue that age could explain differences in experienced strain by individuals.

Therefore, I hypothesize:

H7: Age moderates the relationship between the perceived stressors and strain, in such a way that experienced strain from stressors is higher among older people.

The second variable that explain individual variations in the effect stressors have on experienced strain is self-efficacy beliefs. Self-efficacy seems to be an important variable which explains differences in experienced stress between individuals. Czaja & Sharit (2006) showed that the higher the self-efficacy of an individual was the lower their anxiety of using ICT. This would imply that self-efficacy influences the effect stressors have on strain, and can thus explain differences between individuals. Therefore, I argue self-efficacy moderates this relationship and therefore I hypothesize:

H8: Self-Efficacy moderates the relationship between the perceived stressors and strain.

13

Methods

Sample and procedures

Since the literature field around techno-stress is already quite elaborate, not scattered and there exists a specific gap in measuring a specific target group I will use a theory testing approach (Van Aken, Berends and Van der Bij, 2012). To test the previously mentioned hypotheses, online, cross-sectional survey data was collected from people ranging age 18 to 99, however the oldest respondent was 71. The questionnaire, constructs and translation can be found in appendix A. This research aims to investigate if the techno-stress model of Ayyagari et al is generalizable to the public context. To achieve this aim, we divided

participants of the online questionnaire into two groups based on if they had work or not. This was also important because while validating the model proposed by Ayyagari et al. (2011), some constructs like work-home conflict were difficult to adapt to a private setting and could only be relevant for people working. Based on if the respondent has work or not, we let them complete the full survey including the work related construct questions. We asked them whether they experienced stressors that could lead to such strain and also about technological characteristics of ICTs and how they experienced these characteristics. The difference with people that indicated to have no work was only that questions about certain stressors were left out for the people having no work because of the difficulty of changing these stressor-related questions to the public context.

The initial sample consisted of 152 respondents that filled in the questionnaire, of which 127 had work and 25 did not. One respondent was deleted due to a double entry. In the final sample, these 151 respondents 91 were female and 60 were male. The average age of the respondents was 39.32 (SD = 14.97) and ranged from 19 until 71. Upon further investigation the individual entry in which the age exceeded the earlier mentioned range of 18-67 was deleted. The average education levels were high compared to the Dutch average (CBS, 2016).

76 (50.3%) finished their university of professional education, 16 (10.6%) finished their

senior vocational education, 16 (10.6) people finished high school, 24 (15.9%) finished their

bachelor’s degree at university and 19 (12.6%) finished their master’s degree at university.

14 Measures

The online questionnaire that was conducted was largely based on the constructs Ayyagari et al. (2011) introduced. Some of Ayyagari et al’s (2011) stressors are less relevant in a public context as work is not necessarily part of an individual’s everyday life. Examples include work-home conflict and role ambiguity. These stressors are used to test the

moderating effect of age and self-efficacy, but are not included when we test stressors in the public context. Other constructs from Ayyagari et al (2011) were adapted to the work context, while the original measures were general. An example is overload, Ayyagari’s construct was:

“I feel emotionally drained from my work”. The general overload measure would be: “I feel emotionally drained” and to adapt it to an ICT-based measure the question would become: “I feel emotionally drained from using ICT). Also, considering that the previously mentioned stressors originate from a study conducted in the work context, another more socially based stressor was added to better suit the public context: Social Overload. The questionnaire can be found in appendix A.

The measurement instruments were translated from English to Dutch. The translation was compared with the translation of an English native and changes were made accordingly.

Additionally, a back-translation procedure (Brislin, 1970) was used to ensure the validity. All questions except for Age, Education, gender and work could be answered on a 5-point Likert scale that ranged from strongly disagree to strongly agree.

Presenteeism: Presenteeism is measured by asking respondents for their perception of their reachability. The items are adapted from Ayyagari et al. (2011) and an example question is: “The use of ICT’s enables others to have access to me”.

Anonymity: To measure anonymity the items from Ayyagari et al (2011) were used, which is based on the measure of Pinsonnealt & Heppel (1997) An example question is: “It is difficult for others to identify my use of ICTs”.

ICT overload: ICT overload is measured by using the items introduced by Moore (2000) and adapted in the way previously mentioned. The ICT overload variable refers to when the individual perceives the demands to be too high meaning that they exceed a person’s capabilities. An example could be “I feel busy or rushed due to ICT”.

Role ambiguity: To measure role ambiguity the role ambiguity item from Moore (2000) is used. High scores on this means that the demand of a technology exceeds the

individual’s ability and that the situations this creates are not in line with the certainty that the

user prefers. (Moore, 2000). An example that can be considered as a question is: “I am unsure

15

what to prioritize: dealing with ICT problems or my work activities”.

Work-Home conflict: Work-Home conflict is measured with the items used by Netemeyer et al., 1996) When a score on this is high it means that a user has difficulty distinguishing private time and time that is allocated for work (Netemeyer et al., 1996). An example question is: “Using technologies blurs boundaries between my home life and my job”.

invasion of privacy: To measure invasion of privacy the item from Alge (2001) is used. This measure compares the individual’s value for privacy against the technological environment that is created. If the score is high, then individuals feel that they are unable to cope with the demands placed on their privacy by the technological environment. (Alge, 2001). An example question is: “I feel my privacy can be compromised because my activities using ICTs can be traced”.

Social overload: To measure social overload a measure from Maier, Laumer,

Eckhardt et al (2014) was adapted to fit the ICT context of this study. An example question is:

“Because of ICT I take too much care of my friends’ well-being”.

Strain: Strain was measured by taking the strain item used in Ayyagari et al. (2011) which was adapted from Moore’s (2002) work exhaustion construct. An example that can be considered as a question is “I feel drained from activities that require me to use technologies”.

(Moore, 2000).

Control/moderating variables. Several control and moderating variables were included. To test the effects of the stressor and strain model introduced by Ayyagari et al.

(2011) in a public context, the model will be replicated and the variables age & self-efficacy will be used as control variables. To answer the second research question investigating moderating effects, interaction terms were introduced between the stressors, and the moderating variables age and self-efficacy. Age was measured as a scale variable where respondents were asked to fill in their age in years. Gender was coded as a dummy item in which males = 1 and females = 2. Education is also coded as a dummy variable in which High school = 1, senior vocational education = 2, University of professional education = 3,

Bachelor’s degree at university = 4 and master’s degree at university = 5. Another important

variable to control for is the self-efficacy variable. Campbell (2004) and Czaja & Sharit

(1998) showed that the higher the self-efficacy of an individual was the lower their anxiety of

using ICT was. Since the anxiety is lower, the experienced strain might also be lower and

therefore it is important to control for this effect. Self-efficacy is measured on a 5 point Likert

16

scale by adapting the items from Deng, Doll & Truong (2004) to fit the ICT context. An example question of self-efficacy is: “I have mastered the skills necessary for using ICTs”.

Variable screening. A requirement for completion of the online questionnaire was that no data could be missing. Therefore, every respondent filled in the answers that they were referred to. I have to note that in this case there are missing data since if a respondent

indicated that they had work, they were referred to the questions that were difficult to change from work to public context. If people indicated that they did not have work, those questions were skipped. Upon investigation, question 1 of invasion of privacy had to be recoded because the direction of the answer was different from the other questions. Besides the missing values, skewness and Kurtosis should not be present within the data. Within the data there was no kurtosis or skewness found (<2.2) (Sposito et al., 1983).

Analysis:

Based on the research questions specified in the introduction, there will be two analyses. The first analyses will take the whole population and see if the stressors and strain can also be used for ICT characteristics outside of the work environment. The second analysis will research whether age and self-efficacy moderate the relationship between the stressors and strain. First, an exploratory factor analysis will be used to confirm construct validity, because this is a requirement for structural equation modeling. With this we test whether the

measurement items that reflect the constructs. Because the factors are theoretically supported and based a closed factor analysis should be used based on the predefined number of factors.

In this analysis any constructs that load to multiple constructs or that don’t load into a construct will be deleted. Secondly, a reliability analysis is required to ensure that the measurement items that are kept after the initial exploratory factor analysis are reliable and can be used. To do this the Cronbach alpha’s will be measured for each construct. When the constructs are deemed as theoretically sound and reliable then the analysis itself can start. The statistical analysis that will be used is Structural Equation Modeling (SEM). SEM is selected because it offers several features needed in this research based on Byrne (2013). First, SEM uses theoretical deduction to specify the relationships among variables, which allows for easier testing of the hypotheses. Second, SEM provides error variance parameters estimates.

These estimates help the research because they correct for measurement error which raises the possibility of correct conclusions and other multivariate techniques do not give these

estimates. Third, unlike other multivariate techniques, SEM enables users to analyze

unobserved variables which is needed because the almost all variables in the proposed

17

conceptual model are unobserved variables. Last and also the most important, in this study we want to see the direct and indirect effects that our technological characteristics have on our dependent variable strain. SEM enables users to observe not only the direct effect that variables have but also the indirect effects. The minimum required sample size of 100 was reached (Hair, Anderson, Tatham & Black, 2010). Because of the constructs being made up from different questions we have reflective measures which is another condition for the SEM analysis (Barclay, Higgins & Thompson, 1995). These reflective measures reflect the latent variable/construct and should be correlated and unidimensional (Anderson & Gerbing, 1988).

The SPSS AMOS extension was used to perform this analysis.

Results

The public context

I start by analyzing the first research question of the stressors and strain in the public context. Therefore, the identified constructs that could not be translated to the public context like work-home conflict and role ambiguity were left out of the analysis. To reduce the large number of observed variables an EFA is a useful statistical method (Hadi, Abdullah &

Sentosa, 2016). The Kaiser-Mayer-Olkin (KMO) measure of sampling adequacy and Bartlett’s test of sphericity were used to assess whether the data could be factored or not (Kaiser, 1970; Bartlett, 1954). The KMO measure was above the threshold of 0.5 (0.720) and Bartlett’s test of sphericity was significant (P<0.05). Based on these two measures it can be concluded that the data is suitable for a factor analysis (Tabachnick and Fidell, 2007). The EFA was conducted by using Maximum Likelihood as an extraction method and Promax as the rotation method because these are less forgiving and because SEM also uses Maximum Likelihood (Kline, 2011). Promax is used as a rotation method because in contrast to the other rotation methods it is an oblique rotation method which is less forgiving than orthogonal rotations (Osborne & Costello, 2009). After the first run several items were changed. It was found that Strain together with ICT overload and social overload loaded onto one factor.

Since strain is the dependent variable of the research, ICT overload and social overload were

removed. All four questions of anonymity did not load significantly on any factor and also

had a Cronbach’s alpha of 0.509 which is too low and was also removed. Furthermore, the

extraction of the first question of invasion of privacy was too low (<0.2). The resulting pattern

matrix can be found in table 1. All communalities were above the cutoff point of 0.30 (Hair et

al., 2010) and all loadings in the pattern matrix are above the cutoff point of 0.40.

18 Table 1: Pattern Matrix

Strain Presenteeism Self- efficacy

Invasion of privacy

IP2 .907

IP3 .891

IP4 .767

ST1 .871 ST2 .867 ST3 .792 ST4 .915

PR1 .827

PR2 .827

PR3 .773

PR4 .637

SE1 .922

SE2 .880

SE3 .862

Extraction Method: Principal Component Analysis. Rotation Method: Promax with Kaiser Normalization. Rotation converged in 5 iterations.

The model explains 73.6% of the variance, which exceeds the cutoff point of 50% (Merenda, 1997). Because of the communalities above 0.3 and the significant loadings on factors were above .4, I conclude that the items found within these factors are correlated and have good convergent validity. Moreover, there are no strong cross-loadings and thus evidence of discriminant validity. The correlation matrix in table 2 shows that there are no correlations higher than .7 which means that there is good discriminant validity.

Table 2: Factor Correlation Matrix Componen

t

Strain Presenteeism Self- efficacy

Invasion of privacy

1 1.000 -.071 -.278 .078

2 -.071 1.000 .363 .251

3 -.278 .363 1.000 .311

4 .078 .251 .311 1.000

Extraction Method: Principal Component Analysis.

Rotation Method: Promax with Kaiser Normalization.

19 Confirmatory factor analysis

The different factors that are found through the EFA will have to be confirmed. This confirmation is done through a confirmatory factor analysis (CFA). This CFA is a

confirmatory technique that results in a measurement model (Schreiber, Nora, Stage, Barlow

& King, 2006). To continue with the CFA and the checks, first a relatively decent model fit has to be observed (Vandenberg, Lance, 2000). To assess this fit the CFA model was drawn and tested in AMOS. According to Hair, Anderson, Babin & Black (2010) in a model fit three indexes are important. The first important index is the CMIN/DF which stands for the

measure of differences between estimated and observed covariance matrixes divided by the degrees of freedom. The second important index is the CFI which measures how well the model holds up against alternative models (Hair et al., 2010). Lastly the RMSEA is an indicator for how well the model shows the observed data (Kenny & McCoach, 2003). The CFA showed that the model did not pass the bottom line requirements of these measures (Hu

& Bentler, 1999). While CMIN/DF was appropriate (<3) as well as the CFI (>0.90), the RMSEA was not acceptable (0.08), which means that the model does not show the observed data optimally and there is no optimal fit. By looking at the modification indices some relations were found between latent variables from different observed variables. This meant that there were probably still some cross loadings for questions among different factors. After deleting invasion of privacy question 3, strain question 1 and presenteeism question 4, an acceptable model was found with appropriate CFI (>0.90), CMIN/DF (<3) and RMSEA (<0.05).

Validity and reliability checks. Multiple validity and reliability checks were

performed. First, construct reliability (CR) was generated and analyzed. To have a consistent

and reliable measure, the CR of a variable should be larger than 0.7, since this indicates good

reliability and internal consistency (Hair et al., 2010). All variables had a CR of at least 0.7

and this can be observed in table 3. Another important indicator of validity and reliability is

the average variance explained (AVE). The AVE is a measure that gives a summary of the

convergent validity of items loading in a construct (Hair et al. 2010). All items scored higher

than the bottom line of 0.50 which means that the items converge adequately in the constructs

(Hair et al., 2010). It can be concluded that all of these newly established factors have good

validity and reliability.

20

Table 3: Construct Reliability and Average Variance Explained

CR AVE

Self-efficacy 0.896 0.742 Invasion of privacy 0.951 0.922 Strain 0.849 0.654 Presenteeism 0.721 0.642

Common method bias. A common method bias test was performed to see if there was variance that could be attributed to the model rather than the constructs the measures represent (Podsakoff, MacKenzie, Lee & Podsakoff, 2003). A dummy variable was added and connected to all observed items in the model. The significance of the model was then tested with and without the dummy factor. If there is no difference between the significance of the model with the dummy variable and the model without the dummy variable, then that means that there is significant shared variance and that the dummy variable has to stay in the model (Posakoff et al., 2003). In the next analyses there will be a correction for common method bias by leaving the dummy variable in the model.

Multivariate assumption. Cook’s distance test was done to see if there were any influentials in the data set (>1). There were no influential found and thus no entries had to be deleted (Cook & Weisenberg, 1988). There was no multicolinearity since all VIF’s are below the threshold of 3 and the tolerances are all above 0.1 (Neter, Wasserman, Kutner & Irwin, 1990), the table can be found in appendix B.

Structural equation modeling. After doing the EFA and the CFA, Structural equation modeling (SME) could begin. The factors that were found with the observed items during the EFA and CFA were imputed in our model and thus added to the data set. Since ICT overload could not be measured due to the problems in the EFA, the only construct that can be tested with this SEM is the invasion of privacy variable. This concerns hypothesis 2 which states: People’s perception of ICT presenteeism will be positively related to invasion of privacy and also hypothesis 6, whether invasion of privacy influences strain. In this model the control variables age, gender, education and self-efficacy are added to ensure the same

research conditions as Ayyagari et al. (2011). The model answering the first research question

can be found in figure 1.

21

Figure 2: Public context variables and effects

Presenteeism Invasion of

privacy

Characteristics Stressors

Strain

Strain

-.354 0.276

Self-Efficacy Age

-0.162 -0.154

-0.549

Hypothesis testing. The testing of the structural model lead to the following results and can be found in table 4. The results from the SEM suggest a significant positive effect between presenteeism and invasion of privacy within the public context (β = 0.35, p < 0.05).

Also a positive significant effect was found between the stressor invasion of privacy and

strain (β = 0.28, P < 0.05). Other interesting effects found are the negative significant effects

between both strain and the stressor with age, the negative significant effect of education and

the large negative significant effect of self-efficacy on strain. I will discuss these findings

about the control variables in more depth in the discussion section. Based on these findings

hypothesis 2 is confirmed.

22

Table 4 the SEM outcomes: Regression Weights: (Default model) Main

effects

S.E. C.R. P estimates

Presenteeism *---> Invasion of privacy

0.116 3.743 *** 0.354

Invasion of privacy

*---> Strain 0.096 3.971 *** 0.276

Controls

Age *---> Invasion privacy

0.003 -1.965 0.049 -0.154

Age *---> Strain 0.004 -2.286 0.022 -0.162

Education *---> Invasion privacy

0.043 1.016 0.31 0.081

Education *---> Strain 0.053 -1.429 0.153 -0.102 Gender *---> Invasion

privacy

0.093 0.155 0.877 0.012

Gender *---> Strain 0.115 0.403 0.687 0.027 Self-efficacy *---> Invasion

privacy

0.081 -0.421 0.674 -0.041 Self-efficacy *---> Strain 0.082 -7.657 *** -0.549

The moderating effect of age and self-efficacy

Now that the first research question is answered, the effect of age and self-efficacy will be examined. As mentioned before, we will now look only at the working population since we want to include all stressors and SEM cannot be performed with missing values.

Because we are using a different sample, both an exploratory and confirmatory factor analysis is performed again to ensure validity. When the model for working people of our sample is established, another analysis using SEM will be conducted.

Exploratory factor analysis

To reduce the large number of observed variables an EFA is a useful statistical method

(Hadi, Abdullah & Sentosa, 2016). The Kaiser-Mayer-Olkin (KMO) measure of sampling

adequacy and Bartlett’s test of sphericity were used to assess whether the data could be

factored or not (Kaiser, 1970; Bartlett, 1954). The KMO measure was above the threshold of

0.5 (0.720) and Bartlett’s test of sphericity was significant (P<0.05). Thus, the data is again

suitable for a factor analysis. Just like the previous analysis the EFA was conducted by using

23

Maximum Likelihood as an extraction method and Promax as the rotation method because these are less forgiving and because SEM also uses Maximum Likelihood (Kline, 2011). After the first run several items were changed. It was found that Strain in this case also loaded highly onto one factor together with ICT overload and social overload. Since strain is the dependent variable of the research ICT overload and social overload were removed. The first question of Invasion of privacy did not load on any factor and was also removed. All four questions of anonymity did not load significantly on any factor and also had a Cronbach’s alpha of 0.514 which is too low and was also removed. After a second run it was found that question four of role ambiguity had a significant loading on a different factor and was also removed. Moreover, the questions regarding work-home conflict had a Cronbach’s alpha that was too low (0.622) and was thus also removed. The resulting pattern matrix can be found in table 5. All communalities were above the cutoff point of 0.30 (Hair et al., 2010) and all loadings in the pattern matrix are above the cutoff point of 0.40. The model explains 73.5% of the variance, which exceeds the cutoff point of 50% (Merenda, 1997). Because of the

communalities above 0.3 and the significant loadings on factors were above .4, I conclude that the items found within these factors are correlated and have good convergent validity.

Moreover, there are no strong cross-loadings and thus evidence of discriminant validity. The

correlation matrix in table 6 shows that there are no correlations higher than .7 which means

that there is good discriminant validity.

24

Table 5: Pattern Matrix Strain Presenteeism Self-

efficacy

Invasion of privacy

Role ambiguity

RA1 ..805

RA2 .899

RA3 .759

IP2 .908

IP3 .892

IP4 .766

ST1 .842

ST2 .862

ST3 .847

ST4 .891

PR1 .806

PR2 .833

PR3 .762

PR4 .654

SE1 .903

SE2 .871

SE3 .848

Extraction Method: Maximum Likelihood.

Rotation Method: Promax with Kaiser Normalization.

Rotation converged in 5 iterations.

Table 6: Factor Correlation Matrix

Factor Strain Presenteeism Self-

efficacy

Invasion of privacy

Role ambiguity

1 1.000 -.056 -.283 .083 .347

2 -.056 1.000 .338 .263 -.136

3 -.283 .338 1.000 .296 -.307

4 .083 .263 .296 1.000 -.057

5 .347 -.136 -.307 -.057 1.000

Extraction Method: Maximum Likelihood. Rotation Method: Promax with Kaiser

Normalization.

25 Confirmatory factor analysis

Just like with the previous CFA, the model was drawn in AMOS and the CMIN/DF, CFI and RMSEA were looked at to determine model fit. However, during an initial test before the validity and reliability checks, it was found that the model did not pass the bottom line requirements of these measures (Hu & Bentler, 1999). While CMIN/DF was appropriate (<3) as well as the CFI (>0.90), the RMSEA was not acceptable (0.08), which means that the model does not show the observed data optimally. By looking at the modification indices some relations were found between latent variables from different observed variables. This meant that there were probably still some cross loadings for questions among different factors. After deleting invasion of privacy question 3, strain question 1 and presenteeism question 4, an acceptable model was found with appropriate CFI (>0.90), CMIN/DF (<3) and RMSEA (<0.05).

Validity and reliability checks. Another validity check was performed since the sample used this time is different. Construct reliability and average variance explained were generated again and can be found in table 7. Again, CR was good because they were all larger than 0.7 and AVE was also good because all items scored higher than the bottom line of 0.50.

All newly established factors have good validity and reliability.

Table 7: Construct Reliability and Average Variance Explained

CR AVE

Presenteeism 0.780 0.545 Privacy 0.790 0.661 Strain 0.856 0.668 Selfefficacy 0.893 0.737 Role 0.798 0.574

Common method bias. A common method bias test was performed to see if there was variance that could be attributed to the model rather than the constructs the measures

represent (Podsakoff, MacKenzie, Lee & Podsakoff, 2003). A dummy variable was added and

connected to all observed items in the model. The significance of the model was then tested

with and without the dummy factor. If there is no difference between the significance of the

model with the dummy variable and the model without the dummy variable, then that means

that there is significant shared variance and that the dummy variable has to stay in the model

(Posakoff et al., 2003). In the next analyses there will be a correction for common method

bias by leaving the dummy variable in the model.

26

Multivariate assumption. Cook’s distance test was done to see if there were any influentials in the data set (>1). In contrast to the previous analysis, now there were 2

influentials which had a value higher than 1. Because these two cases influence the strength of the regression, deletion of these cases is justified (Cook & Weisenberg, 1988). Also according to appendix C, there is no multicolinearity since all VIF’s are below the threshold of 3 and the tolerances are all above 0.1 (Neter et al., 1990).

Structural equation modeling. After doing the EFA and the CFA, another SME model was build. This time the model has two stressors, role ambiguity and invasion of privacy. This time age and self-efficacy were not added as control variables but as moderating variables. Because this new model is different from the model fit examined earlier, model fit has to be assessed again. In this new model the CMIN/DF is sufficient (<3), the CFI is also great (>0.9) and the RMSEA also meets the required standard(<0.05) (Hu & Bentler, 1999).

Therefore, it can be concluded that the overall structural model has a good fit. Due to the complexity of the model, the significant effects from the model can be found in figure 2 and the both significant and non-significant effects can be found in table 8.

Figure 3: Moderation effect

27

Table 8: standardized SEM outcomes

S.E. C.R. P Estimates

Presenteeism *--> Role 0.083 4.597 *** 0.383 Presenteeism *--> Privacy 0.082 4.913 *** 0.405 Invasion of privacy *--> ZStrain 0.083 2.318 0.02 0.195 Role ambiguity *--> ZStrain 0.083 2.723 0.006 0.229 Age *--> ZStrain 0.084 -1.489 0.136 -0.127 Self-efficacy *--> ZStrain 0.094 -2.278 0.023 -0.214 age*invasion of

privacy

*--> ZStrain 0.089 0.072 0.943 0.006

Self-efficacy

*invasion of privacy

*--> ZStrain 0.076 -1.001 0.317 -0.086

Self-efficacy * role ambiguity

*--> ZStrain 0.081 1.994 0.046 0.174 Age * role

ambiguity

*--> ZStrain 0.11 1.302 0.193 0.115

Gender *--> Privacy 0.082 0.009 0.993 0.001 Gender *--> Role 0.083 0.099 0.921 0.008 Gender *--> ZStrain 0.084 1.327 0.185 0.112 Education *--> ZStrain 0.085 -1.259 0.208 -0.108

Hypotheses testing. The results from the SEM again suggest a significant positive effect between presenteeism and invasion of privacy within the public context (β = 0.40, p <

0.05). This means that H2: People’s perception of ICT presenteeism will be positively related to invasion of privacy is confirmed. Also, a positive significant effect was found between presenteeism and role ambiguity (β = 0.38, p < 0.05) and thus H5 that states: within the work context, people’s perception of ICT presenteeism will be positively related to perceived role ambiguity, is confirmed. Moreover, both the effect of privacy as well as the effect of role ambiguity on strain was confirmed (β = 0.20, p < 0.05) and (β = 0.23, p < 0.05) respectively.

Meaning that both hypothesis 6b as 6d regarding the positive effect of these stressors on strain

are confirmed. Of the interaction effects, self-efficacy seems to have a significant negative

moderating effect (β = 0.17, p < 0.05), meaning that the more self-efficacy an individual has

the less strain is experienced through the perception of role ambiguity. This means that H8:

28

Self-Efficacy moderates the relationship between the perceived stressors and strain is partly confirmed and only for the moderating effect on perceived strain through the role ambiguity stressor. An overview of the hypotheses that were confirmed or rejected can be found in table 9.

Table 9: Result of the hypotheses H1 Among the working population in the sample,

perceived ICT presenteeism will be positively related to perceived work-home conflict

NOT ABLE TO TEST

H2a Perceived ICT presenteeism will be positively related to invasion of privacy

Confirmed

H2b Perceived ICT anonymity will be negatively related to perceived invasion of privacy.

NOT ABLE TO TEST

H3 Perceived ICT presenteeism will be positively related to perceived ICT overload.

NOT ABLE TO TEST

H4 Among the working population in the sample, people’s perception of ICT presenteeism will be positively related to perceived role ambiguity.

Confirmed

H5 People’s perception of ICT presenteeism will be positively related to perceived social overload.

NOT ABLE TO TEST

H6a ICT overload is positively related to people’s perceived strain.

NOT ABLE TO TEST

H6b role ambiguity, is positively related to people’s perceived strain.

Confirmed

H6c work-home conflict is positively related to people’s perceived strain.

NOT ABLE TO TEST

H6d invasion of privacy is positively related to people’s perceived strain.

Confirmed

H7 Age moderates the relationship between the perceived stressors and strain, in such a way that experienced strain from stressors is higher among older people.

Rejected

H8 Self-Efficacy moderates the relationship between the perceived stressors and strain.

Partly confirmed

29

Discussion

The goal of this research was twofold: to expand the techno-change concept beyond the work-context into the public context and to try and examine the role that demographic and psychological variables such as age and self-efficacy play in experiencing stressors and strain.

The model heavily drew on Ayyagari et al (2011), but focused exclusively on intrusive characteristics and the misfits between an individual’s abilities and the demands from the environment resulting from stressors. Stressors in turn were modelled as predictors of strain.

In this section I will first elaborate on the findings and implications of testing applicability of Ayyagari et al’s (2011) model in another context. After that I will focus on the implications that the findings have for the digital divide and the techno-stress literature. Lastly, I will discuss managerial implications of this study as well as the limitations.

Techno-stress in the public context.

Adhering to the call for a better understanding of the interruptive nature of technology (Speier et al., 2003), this study sought to translate the effect of interruptive ICT characteristics in work context to the public context. We found that within the public context presenteeism is positively related to invasion of privacy, which in turn is positively related to strain. This suggests that the effects of presenteeism on invasion of privacy and of invasion of privacy on strain are generalizable from the work context to the public context. These findings are in line with the literature as Greenberg & Firestone (1977) found that personal space intrusion through surveillance in a non-work context lead to stress. In addition, Härenstam & Hagberg (2011) found that people in the public context experienced this personal space intrusion through a high frequency of calls or texts, which also lead to stress.

In this study we found that work-overload and social-overload could not be

constructed as expected. First regarding ICT overload, works from Price & Spence (1994) already mention the large differences between home stressors and work stressors. They conclude that exhaustion in males was predicted by severity of daily hassles, mostly originating from the public context. Since we have used an adopted exhaustion (overload) measure of Maslach & Jackson (1981), it could be that more public context variables should have been examined. Moreover, looking at the variables we used to examine strain and

overload, they both seem to be emotional measures. Not much difference is apparent in “I feel

busy or rushed due to ICT” and “I feel tired from my ICT activities”. Both measures examine

30

the emotional state, which can explain the high loadings of ICT overload on strain.

Secondly, the items of social overload also loaded on the dependent variable. Maier, Laumer, Eckhardt & Weitzel (2014) conclude that social overload is a stressor but specific to the SNS context. Using social overload in a different context might have caused the

measurement problems. Both ICT overload and social overload are forms of overload

resulting from interruption and receiving messages/interruptions. An explanation for the high cross loadings is that both measures actually measure information overload (Bawden &

Robinson, 2009). Information can come to us through ICT in the form of social messages or work related messages, but they are both information. Moreover, the measure for social overload refers to friends, however colleagues can also be friends which complicates loadings even more

The Cronbach’s alpha of one of the technology characteristics: Anonimity was too low to include it in the research (<0.5). While anonymity has been shown to influence invasion of privacy and strain within a work context (Ayyagari et al., 2011; Parenti, 2001), within this study the measure of this construct was not valid in a public context and thus could not be used. One possible reason is the measure itself. The measure of anonymity in this study was taken by adapting the anonymity measure from Ayyagari et al. (2011) and Ayyagari et al’s measure was based on Pinsonneault & Heppel (1997). Pinsonneault & Heppel (1997) explained anonymity to consist of five variables: Diffused, proximity, knowledge of other group members, confidence in the system and public self-awareness. In hindsight, it might have been better to develop questions based on these five variables, because Ayyagari et al.

(2011) based questions around just one of these variables, proximity. Proximity is especially important in the context of computer-based communications in groups (Pinonsault & Heppel (1997). Proximity is the physical distance to colleagues and how observed an individual feels.

This will be much less in a public context and therefore the proximity measure might not be

sufficient. Future research could research whether taking different anonymity measures

influences the outcome of the research.