The influence of different privacy practices on the choice for an e-mail client in professional setting and private setting

The influence of different privacy practices on the choice

for an e-mail client in professional setting and private

setting

BY

Kevin van der Veer

The influence of different privacy practices on the choice

for an e-mail client in professional setting and private

setting

BY

Kevin van der Veer

University of Groningen

Faculty of Economics and Business

Msc Marketing Intelligence

Master Thesis

June 2017

Abstract

Customers are becoming increasingly concerned about their privacy. Previous research has indicated that customers may have privacy-protective responses when concerned about privacy, such as switching to a more privacy-protective competitor. In the e-mail privacy market, some privacy-protective alternatives for common e-mail services, such as Gmail, have emerged. These alternatives use different privacy practices as unique selling points for their services. E-mail clients can vary in the way they store, use and collect data. In this paper, a conjoint study is performed to examine the effect of different privacy practices on the choice for an e-mail client. Specifically, it examines the effect on choice when e-mail clients collect more or less data, store data in a country with more or less privacy legislation and varying in the way they use data. The study takes place in two settings – professional and private – to assess the effects of different privacy practices when choosing an e-mail provider for business e-mail communication or personal e-mail communication. This study finds that customers have greater preference for more privacy protective-privacy practices and are thus also willing to pay for privacy protection. However, it finds no significant effect for the two different settings. The moderating analysis for privacy concerns yields only partial proof for the interactional effect between privacy concerns and country of data storage, and no significant effect is found for data collection and data usage. The segmentation reveals that the majority of customers in the e-mail market are not strongly interested in a paid protective e-mail client; however, a clear niche segment can be formed for a more privacy-protective e-mail client.

Keywords:

Preface

6

1. Introduction

‘George is writing a critical email about Donald Trump through his Gmail account to send to one of his friends. The next day, Donald Trump targets George with a specific advertisement,

‘10 reasons not to vote for Hillary Clinton’. George feels exposed. Months later, Donald Trump is elected as the president of the United States, and weeks after that, The New York Times reports that the National Security Agency (NSA) filtered and scanned all incoming and

outgoing e-mails of Gmail users. George feels exposed again.‘

These days, Internet companies collect a vast amount of data from their customers in order to adapt their marketing mix to specific customers(Chung, Wedel, and Rust 2016). However, the increase in data collection by such companies has caused a surge in privacy concerns regarding the collection and usage of data, which has in turn resulted in customers who are increasingly worried about their privacy (Rose, Rehse, and Röber 2012). Google, for example, scans all e-mail data of Gmail users in order to serve targeted advertising and improve some functions, such as smart labelling of e-mails to allow for prioritisation and categorisation (Google 2017).

‘Our automated systems analyse your content (including emails) to provide you personally relevant product features, such as customised search results, tailored advertising, and spam and malware detection. This analysis occurs as the content is sent, received, and when it is

stored’(Gibbs 2014) - Terms of Service Gmail,

7 market. One example is ProtonMail. The most important promises that this company makes regarding its service are the following(ProtonMail 2017):

1) Swiss servers. ‘ProtonMail is incorporated in Switzerland and all our servers are located in Switzerland. This means all user data is protected by strict Swiss privacy laws’.

2) Anonymous e-mail. ‘No personal information is required to create your secure email account. By default, we do not keep any [Internet Protocol] IP logs which can be linked to your anonymous email account. Your privacy comes first’.

8 business e-mail communication serves business purposes, personal e-mail communication is designed for engaging with friends, family and acquaintances. This explains why Google has a specific option for employers to control employees’ e-mail, an option which is not available in regular Gmail: ‘The Google Apps Email Audit API allows Google Apps administrators to audit a user's email, email drafts, and archived chats’(Google 2017). There are also other privacy boundaries to be expected when using e-mail for professional matters compared to private (personal) use because, in business communications, information tends to be more shared than in private communication, as there are usually more persons and companies involved (Petronio 2012). Therefore, it is expected that consumers in personal settings compared to business settings are keener to protect their privacy concerning e-mail communication.

These developments in the market and the rise of customer privacy concerns could be a treat for companies such as Google since customers may switch to competitors to relieve privacy concerns. Previous research has indicated that customers can react negatively to the way and extent that companies store, collect and use data (Smith, Milberg and Burke 1996), which are practices that are triggers for customer privacy concerns. It is therefore important for marketers to understand if and how different privacy practices may trigger the switch to more privacy-protective alternatives (Rust and Huang 2014).

The unique selling points (USPs) of ProtonMail and the extra advertisement-free USP of Posteo, as mentioned before, can be linked to some drivers of customer privacy concerns that are covered in the literature: 1) Data storage, since storing data in Switzerland could be a way to store data safely, 2) Data collection, since they do not collect personal information during registration and 3) Data usage, since they do not sell the personal data of their customers to third-party advertising companies. So, the decision to adopt a privacy-protective e-mail client – and thus pay for privacy – depends on the extent to which customers are concerned about methods of data storage, collection and usage.

9 relative importance of the types of privacy practices. Moreover, a post-hoc segmentation is conducted for further investigation of different sub-segments since one can expect customer heterogeneity in demographics and preference utilities from the choice design. The following research questions need to be answered:

Research question 1: To what extent do the privacy practices of data storage, data collection

and data usage (relative to each other, i.e. relative importance) affect the choice for an e-mail client?

Research question 2: To what extent are customers willing to pay for more

privacy-protective levels of theprivacy practices mentioned in question 1?

Research question 3: Do customer privacy concerns moderate the effect mentioned in

question 1?

Research question 4: Is there a difference in answers to the first two questions in a

business/professional setting compared to a private/personal setting?

Research question 5: Which interesting customer segments can be formed based on the

demographics and choice utilities?

In order to address the five research questions, a choice-based conjoint analysis is performed in combination with a post-hoc segmentation. Respondents were obtained through Amazon MTurk to avoid similar respondents in a student environment (convenience sample).

2. Theoretical framework

This section provides an overview of previous literature on customer privacy concerns and privacy practices. Furthermore, it describes the difference between private and professional settings.

2.1 Customer privacy concerns

10 institutions to determine for themselves when, how, and to what extent information about them is communicated to others’. Customers are increasingly concerned about their privacy (BCG 2012), and since customer privacy concerns are the main topic of this paper, it is important to further define the concept of privacy concerns. Smith et al. (1996) have defined privacy concerns as ‘consumers’ worries or uneasiness regarding the collection, storage, and usage of their personal information’. In addition to collection, storage and usage of personal information, customers’ concerns about transparency and control regarding personal data have recently been added to the definition (Malhotra, Kim, and Agarwal 2004).

2.2 Privacy practices

The collection, storage and usage of personal data by companies are triggers for consumer privacy concerns (Smith, Milberg and Burk 1996). Therefore, companies can have different strategies around these concepts. Companies can vary in their methods of collecting, storing and using consumer data. Some companies can use privacy as a strategy if they act differently than their market competitors (Martin and Murphy 2017). It appears that previous research has not assessed the relative importance between collection, storage and usage of personal data. More insights about the influence and relative importance of these privacy practices on willingness to pay for a privacy-protective email service could be useful for marketers. Moreover, post-hoc segmentation can reveal which consumers to target with these certain privacy practices. This may assist in decision making around privacy practices. This section defines privacy practices around data collection, usage and storage.

2.2.1 Data collection

11 e-mail: ‘No personal information is required to create your secure email account. By default, we do not keep any IP logs which can be linked to your anonymous email account’ (ProtonMail 2017). When companies collect greater quantities of personal data, customers feel more vulnerable, which can encourage a switch between firms, for instance from Gmail to ProtonMail, since they are more concerned (Martin, Borah, and Palmatier 2016). Accordingly, one can argue that when a company collects less personal information, customers have a higher utility for this e-mail client. This leads to the following hypothesis:

H1: A lower level of sensitive personal data collection positively influences utility/choice for the e-mail client.

2.2.2. Data Storage

If companies collect data from their customers, they have to store this data in order to access it. So, companies must decide how and where to store the data they have collected. The way in which companies store their data is significant mainly for security reasons, as consumers worry that their information could be accessible by undesirable parties (Harris, Hoye and Lievens 2003; Martin, Borah and Palmatier 2016; Smith, Milberg, and Burke 1996). Previous literature has indicated that in order to prevent data from unauthorised use, companies can either collect less data (Section 2.2.1) or aggregate the data to make it anonymous in such a way that there is no identifiable link between the customer and the data (Verhoef, Kooge, and Walk 2016). ProtonMail prominently state that their information is stored in Swiss servers (ProtonMail 2017): ‘ProtonMail is incorporated in Switzerland and all our servers are located in Switzerland. This means all user data is protected by strict Swiss privacy laws’. Switzerland is ranked as one of the safest countries in the world to store data (Wadlow 2017). However, previous literature has not evaluated if there are benefits to data storage in a certain country next to less data collection and data aggregation.

Companies such as Google and Yahoo have been negatively featured in the news for providing data to government organisations such as the NSA. The NSA has forced Yahoo to

12 European customers may not want their cloud computing data stored on U.S. servers (Zhou et al., 2010). However, contradicting research by Ion et al. (2007) has indicated that only 7 out of 36 interview participants considered the country of storage to be important.

Governments around the world can have unique regulations around privacy. The lack of government regulation has a strong impact on customer privacy concerns (Wirtz, Lwin, and Williams 2007). The focus is more on self-regulation in the US, whereas Europe has more legislation around privacy protection (European Union’s Data Privacy Directive) (Bellman et al., 2004). In countries where a legal structure around privacy protection is visibly present, privacy features for a service or website are perceived as less important attributes (Steenkamp and Geyskens 2006). Wirtz et al. (2007) have explained that customers are less inclined to actively protect their privacy if there is legislation protection, and that one can therefore argue that storing data in a country with more privacy legislation can positively influence the utility for e-mail clients. This informs the following hypothesis:

H2: Data storage in a country with a higher level of privacy legislation positively influences utility/choice for the e-mail client.

2.2.3 Data usage

After collecting and storing consumer data, companies can take one of several approaches to using this data. Companies can employ consumer data internally in order to improve their services. Examples of internal use of data are remembering preferences of consumers or tailoring products to customer preferences (Ackerman, Cranor, and Reagle 1999) (Wedel and Kannan 2016). Google, for example, uses customer data to improve its spam filters and organise the mailbox of users to be more customised to the consumer. Furthermore, services such as Gmail generate profit through secondary usage of targeted advertisements (Marinescu 2013). They may provide external advertisers with consumer data to support targeted advertisements.

13 information, which is likely in e-mail (Mothersbaugh et al. 2012; White 2004). A step further is the use of data for external purposes, such as selling data to external advertisers. According to Alreck and Settle (2007), customers largely have a negative feeling about such advertising. Literature has called this way of using data ‘unauthorized secondary usage’ (Harris et al. 2003). Nevertheless, generating money through targeted advertisements is one reason why services such as as Gmail are able to deliver free services, and although customers may not like it, they may still use such services (Pavlou 2011). However, this does not mean that customers cannot switch to alternatives.

Therefore, one can expect that if companies use the data they have collected and stored less extensively compared to more (for internal, or even more external) usage, this would have a positive influence on the utility for e-mail clients. Research by Groopman and Etlinger (2015) has demonstrated that customers’ top privacy concern is whether companies share data with external companies. So, using data by external parties is presumed to be a higher level of data usage than internal usage, and using no data is the lowest level of data usage. This informs the following hypothesis:

H3: A lower level of data usage has a positive effect on utility/choice for the e-mail client.

2.2.4 Relative importance

14 obtain a clearer insight into the relative importance of privacy practices. This leads to the following hypotheses:

H4a: Data storage is considered relatively more important than data usage. H4b: Data usage is considered relatively more important than data collection.

2.3 Customer Privacy Concerns Moderation

It is striking that products such as Gmail are constantly growing in popularity despite the expanding privacy concerns of consumers. In 2012, Gmail had 412 million active users, and in 2016, it reached 1 billion active users (Lardinois 2016). Literature has termed this phenomenon ‘the privacy paradox’ (Berendt, Günther, and Spiekermann 2005; Norberg et al. 2007). In other words, consumers calculate the cost-to-benefit ratio of their concerns about the collection, storage and usage of their personal data (Culnan and Armstrong 1999; Dinev and Hart 2006). So, one may know about Google’s privacy policy, yet still use its service since they have weighed the costs and benefits of doing so. Other explanations of the privacy paradox could be that consumers are not aware of the privacy practices of a certain company, have no good alternatives (lock-in) or have bounded rationality (Acquisti and Grossklags 2005: Ariely 2009).

Additionally, literature has noted various consequences of high privacy concerns of consumers. If consumers have high privacy concerns, they may have privacy-protective responses. For example, consumers may refuse to disclose information, may remove information or may complain to third-party organisations (Son and Kim 2008). Privacy-protective responses may result in decreased sales for e-commerce companies (Pettey 2006) and increased costs for privacy protection (Turner 2001). These findings could also threaten products like Gmail. Companies must take these expected increased costs or decreased sales into account when making decisions about data storage, collection and usage if customers are concerned with privacy. It can therefore be argued that if customers are more concerned about their privacy, there will be higher utility of an e-mail client with higher privacy protection with respect to data collection, usage and storage. This leads to the following hypotheses:

H5a: A higher level of privacy concerns will positively influence the preference for a lower amount of data collection.

H5b: A higher level of privacy concerns will positively influence the preference for data storage in a country with a higher level of privacy legislation.

15

2.3 Professional and Private settings

This research took place in two settings: professional and private. Ion et al. (2011) have stated that privacy requirements and willingness to pay for cloud storage differ between a consumer setting (private) and company setting (professional). Normally, business e-mail is not intended for personal use since it may distract employees from business goals and could ultimately reduce productivity (Hornung, 2005). This is main reason that consumers often have a personal e-mail account and a business e-mail account and use each in a distinct way. This is also why some companies monitor e-mail to ensure employee productivity (Agarwal and Rodhain 2002; Weisband and Reinig 1995). Consumers are more likely to share private information in deeper relationships, and deeper relationships are more likely in a private e-mail setting since business e-e-mail supports business communication instead of communication for personal matters (White 2004). An experiment by De Correspondent1 in the Netherlands highlighted some interesting differences between personal and professional e-mail usage:

Graph 1: Social network of the subject used in the experiment by De Correspondent (Tokmetzis 2014)

The experiment revealed that the social network of the subject is different in a private setting compared to a professional setting. In a business setting, conversations are less frequent between individuals than among groups (carbon copy [CC] in e-mails). Only a few individuals to whom the subject talked through his professional e-mail also appear in his

16 private e-mail. This experiment strengthens the findings of Petronio (2012), which evidenced that there is more shared information in business communications, and therefore different privacy boundaries, compared to private communication. However, one can also expect that professional e-mail communication contains valuable business information which could also trigger privacy concerns (Krutz and Vines 2010), in contrast with private e-mail communication where one can have the feeling of ‘nothing to hide’ (Wijnberg and Martijn 2013). Nonetheless, according to a report by EY (2015), employees are frequently and unintentionally the cause of company data breaches in the name of convenience. These employees may, for example, e-mail sensitive information. This could indicate that employees are not particularly eager to protect company information. In view of this, it can be anticipated that private e-mail users expect more privacy than professional e-mail users. This informs the following hypothesis:

H6: E-mail users in a private setting have a higher willingness to pay/utility for privacy practices (data collection, data storage, data usage) compared to e-mail users in a professional setting for the e-mail client.

2.4 Customer characteristics / control variables

Since customers are diverse, some (demographic) customer characteristics are included in the study. Goldfarb and Tucker (2012) have noted that older people are more concerned about their privacy. Another interesting finding is that iOS users are more concerned with privacy than Android users (Benenson et al., 2013). Furthermore, females tend to be more privacy concerned than males (Bellman et al., 2004). Finally, innovative consumers tend to have fewer privacy concerns for data-driven products (Zhao, Lu, and Gupta 2012). In addition to the previous variables, other control variables are included in the survey:

Current e-mail provider Privacy misuse

Gender Age

Mobile OS E-mail use frequency

Country Education

17

2.5 Conceptual model

The previous analysis informs the following conceptual model:

18

3. Methodology

In order to test the hypotheses, a dual-response, choice-based conjoint analysis has been conducted. The online survey was created with PreferenceLab.com. In the conjoint analysis, customers have to make a trade-off between certain levels of privacy aspects (data collection, data usage and data storage) and price. In addition to the choice of attributes and levels, a non-response was also added to determine whether respondents actually would use the e-mail service and to make it more realistic (Eggers and Sattler 2011). The dependent variable in this paper is the utility, and the willingness to pay can be calculated from the utilities if price is included as an attribute. The willingness to pay is the price that customers want to pay for a given product or service (Wertenboch and Skiera 2002; Green and Srinivasan 1990). Respondents were recruited through Amazon Mechanical MTurk.

3.2 Procedure

The survey is divided into three parts. The first part consists of an explanation of the study for respondents. The second part offers respondents a series of choice sets (conjoint analysis) through which the respondents had to make trade-offs between alternatives. The final part contains the survey questions, namely consumer characteristics and manipulation and attention checks.

3.3 Conjoint design

In the second part of the survey, respondents chose between various alternatives that differed on the attributes of storage (1GB, 5GB or 20GB), data collection (e-mail or e-mail, age, gender or e-mail age, gender, birthdate, mobile number), data storage (Iceland, United States or Brazil), data usage (not used, used internally or used externally) and price (€0, €5 or €20 per month). Respondents selected the alternative that they preferred the most and indicated if they would indeed use the alternative or not (‘none’ option). To make the conjoint part as realistic as possible, the virtual brand name used for the e-mail client was ‘DaMail’. So, respondents were asked to note the preferred product combination (levels of the attributes) for ‘DaMail’ in every choice set.

19 The conjoint part was divided into two random manipulations:

Graph 3: DaMail Business

In situation one, respondents were prompted to imagine that need to decide on an e-mail client which is used for professional matters, while the second group was using it for personal matters. The fictive brand names for the e-mail client are DaMail Business and DaMail Personal.

20

3.4 Sample Size

Since respondents were recruited through Amazon MTurk, it is interesting to determine a desirable minimum sample size. Johnson and Orme (2010) have suggested the following formula for calculating the minimum sample size: n*t*a / c = 500, where n is the number of respondents, t the number of tasks, a the number of alternatives per task and c is equal to the largest product of levels of any two attributes. This establishes a minimum sample size of 125. Fortunately, the number of respondents equals 260 in this paper, which meets the minimal requirement. In order to ensure qualitative respondents, a minimum ‘HIT Approval Rate’ of 97% has been set in Amazon MTurk. This means that at least 97% of the work on Amazon MTurk has been approved in the past by other requesters for the individual respondent.

3.5 Data Cleaning

The survey had one attention check hidden between the two questions about innovation. Respondents were asked to tick ‘1’ on a seven-point Likert scale for the attention check. Only one respondent failed the attention check and was deleted from the dataset. Since the data consist of survey data (one row per respondent) and choice data (multiple rows per respondent), the data has been merged in software package R. Normality of the data is assumed based on the central limit theorem. In the choice data, the attributes of storage, data usage, collection and price are quite straightforward. For example, respondents know that 20GB is better than 5GB. However, for country of data storage, this may not be straightforward. Therefore, after the choice sets, the following question was presented in the survey:

21 Appendix A1 displays the distribution for the countries. The distribution reveals that respondents indeed perceive Iceland to be safer than the United States, and the United States to be safer than Brazil. Although some respondents rated the countries in this question unexpectedly (for example not trusting their data anywhere and rating everything as ‘strongly disagree’), the general distribution is as intended. Therefore, no problems are expected with this attribute in the conjoint study.

3.5 Measurement of Privacy Concerns and Innovativeness

Privacy concerns were measured with four questions on a seven-point Likert scale based on Dinev and Hart (2006). Innovativeness was measured with two questions on a seven-point Likert scale based on Klink and Smith (2001). Factor analysis (principal component analyses) was performed for privacy concerns and correlation analysis for innovativeness to test if these constructs are adequate to group into one factor. The Kaiser-Meyer-Olkin measure of sampling adequacy (KMO) (0.855), the Bartlett’s test of sphericity (p = 0.000) and the Cronbach’s alpha (0.909) indicated that the four questions for privacy concerns were good to load on one factor. For innovativeness, correlation was checked between the two questions, which gave a correlation of 0.75 at 1% confidence interval. Therefore, the four questions for privacy concerns and the two questions of innovativeness were grouped into one new variable by the sum of the questions divided by four and two, respectively.

3.6 Moderation

22

4. Results

4.1 Characteristic of the sample

The final sample size consists of 259 respondents. The majority of the respondents were from the United States and India, this is as expected since Amazon MTurk has the most workers from those countries and no country restriction was set. Most respondents are Android and Gmail users and all respondents have at least a high school degree. An overview can be found in table 2. Percentage Number Gender • Male 58,70% 152 • Female 41,30% 107 Age • under 18 0% 0 • 18-25 14,30% 37 • 26-39 57,90% 150 • 40-65 26,60% 69 • Over 65 1,20% 3 Country • North America 68,70% 178 • Asia 27,40% 71 • Europe 3,90% 10 Mobile OS • Android 66,40% 172 • iOS (Apple) 28,20% 71 • Other 5,40% 10 Education • Elementary school 0% 0

• High school graduate 24,30% 63

• Bachelor degree 52,10% 135

• Master degree 21,60% 56

• Doctorate degree 1,90% 5

Current E-mail provider

• Gmail 68,30% 177

• Hotmail 8,10% 21

• Yahoo 15,80% 41

• Other 7,70% 20

23

4.2 Main effects model 4.2.1 Model comparison

To calculate the main effects of the model, all attributes except the ‘none’ option were set at a nominal (part-worth) level. The main model has been calculated using LatentGOLD Choice 5.1. The following criteria were used to assess the model fit: likelihood ratio test, pseudo-R², adjusted R², predictive validity and the hit rate.

Likelihood Ratio test

The likelihood ratio test reveals if the log likelihood of the model (LL( )) is significantly different from the NULL model (LL(0)). The NULL model is calculated using the following formula: LL(0) = n*c*ln(1/m), where n is the number of consumers, c the number of choice sets per customer and m the number of alternatives per choice set. The LL( *) follows from the main model estimation from LatentGOLD.

LL(0) = 259 * 12 * In (⅓ * ½ ) = -5568.8 LL( *) -4187.1

The estimation mentioned above indicates that the log likelihood of the model is higher than the log likelihood of the NULL model. To assess significant difference with the NULL model, a chi square test has been conducted with degrees of freedom of 248 and a critical value (a=1%) of 302.7. The chi-square is calculated below:

Chisq = -2(LL(0)-LL( )) = -2(-5568.8 - (-4187.1)) = 2763.4

Therefore, one can argue that the model is significantly different from the NULL model since Chisq (2763.4) is higher than the critical value of 302.7(p<0.01).

Pseudo-R², Adjusted R² and Hit rate

The pseudo-R2 _{and the adjusted R}2 _{(McFadden’s R}2_{) were calculated and support the findings}

of the likelihood ratio test.

24 :%;$"<#% (= _{= 1 −} -- / 2>?@A

--(1) = 1 −

23456.42 44

28895.5 = 0.25

A small number, usually between 0.2 and 0.4, is considered acceptable (MacFadden, 1974). The calculated hit rate of the model is 2935 + 703 + 654 / 6216 = 69%.

This gives the following comparison with the NULL model, which confirms that the main model outperforms the NULL model:

LL Chisq Npar Df Pseudo

R2 Adjusted R2 Hitrate AIC(LL) NULL Model -5568.8 33% Main effect model -4187.1 2763.4 11 248 0.25 0.25 69% 7878.3

Table 3: Main model and null model comparison

4.2.2 Linear model

In general, one can assume a linear negative relationship of price for products and services. In other words, price has a negative price elasticity; the higher the price, the lower the demand for products and services (Bijmolt, Heerde and Pieters 2005). In models, adding more parameters usually increases fit. In order to determine if the model does not significantly decrease in fit with price as a linear variable (fewer parameters compared to part-worth), a log likelihood test was performed:

LL Number of parameters

Price part-worth -4187.1309 11

Price linear -4187.1870 10

Table 4: Log likelihood comparison linear, part-worth

25

4.3 Interpretation main effects

When examining the results of the main model in Table 5, one can see that all attributes are highly significant. This means that all attributes have a significant impact on choice. The utilities of the attributes are interpreted relative to each other, and the relative importance compared to the other attributes is calculated by the range divided by the total range.

Attribute β

Wald

statistic p-value Range

Attribute relative importance Storage <0.001 0.4435 11.47% 20GB 0.1984 90.9255 5GB 0.0467 1GB -0.2451 Data Storage 341.4650 <0.001 0.8883 22.98%

Brazil (low privacy

legislation level) -0.5479 United States (medium privacy legislation level) 0.2076 Iceland (high privacy legislation level) 0.3404 Data Usage 241.3930 <0.001 0.6796 17.58% Not used 0.2263 Used by DaMail itself 0.2270 Used by external companies -0.4533 Data Collection 20.6884 <0.001 0.277 7.17% Email 0.0935 Email, Age, Gender 0.0207 Email, Age, Gender, Birthdate, Mobile number -0.1142 Price 1026.2418 <0.001 1.5775 40.81% €0 per month 0.7934 €5 per month -0.0093 €20 per month -0.7841 None option -0.7543 270.4605 <0.001 Total range 3.8659

26 The results clearly indicate that lower price and higher storage are preferred, as the utilities of more storage and lower price are higher compared to the utilities of a higher price and less storage. The negative utility of the ‘none’ option indicates that respondents would rather use their preferred e-mail service than not use their chosen alternative.

4.3.1 Testing Hypotheses 1-3: Data Collection, Data Usage, Data Collection

In order test hypotheses 1, 2 and 3, Table 4 can be used to address the hypotheses step by step. For data collection, the presented utilities are as expected and significant. If the e-mail client collects less information about the customer, there is higher utility. The less data collected, the higher the utilities of ‘Email’ (U=0.0935) ‘Email, Age, Gender’ (U=0.0207) or ‘Email, Age, Gender, Birthdate, Mobile number’ (-0.1142). The positive utility for ‘Email, Age, Gender’ and the slighter difference with ‘Email’ compared to with ‘Email, Age, Gender, Birthdate, Mobile number’ could indicate that respondents do not see a substantial difference between ‘Email’ and ‘Email, Age, Gender’. Nevertheless, the utility for ‘Email’ is still significantly the most positive. Therefore, hypothesis H1 is supported. Moreover, the utilities for data storage are also in line with expectations. If the e-mail client stores customer data in Brazil, customers have lower utility (U=-5479) compared to in the United States (U= 0.2076) and Iceland (U=0.3404). So, one can say that customers prefer data storage in a country with a higher level of privacy legislation, which supports H2. If data is collected and stored, companies also have to decide how to use the data. The utilities for data usage are also consistent with expectations, although ‘Not used’ (U=0.2263) and ‘Used by DaMail itself’ (U=0.2270) are rather close compared to ‘Used by external companies’ (U=-0.4533). This indicates that they prefer data is not used, but see little difference in the e-mail client employing it for internal usage.

H1: A lower amount of personal data collection positively influences utility/choice for the e-mail client. - Supported

H2: Data storage in a country with a higher level of privacy legislation positively influences utility/choice for the e-mail client. – Supported

H3: A lower level of data usage has a positive effect on utility/choice for the e-mail client. –

27

4.3.2 Testing hypothesis 4: Relative importance

The relative importance was calculated in a percentage to indicate the importance of the attributes relative to each other. This yielded the following result:

Graph 6: Relative importance attributes, in % without ‘none’ option

If customers select an e-mail client, price is considered most important (41%). It is interesting to notice that data storage (23%) and data usage (18%) are considered more important than the amount of storage in gigabytes (11%). Apparently, customers care more about where their data is being stored and how their data is being used than about the amount of storage available in their account. The least important is data collection (7%), supporting hypotheses H4a and H4b.

H4a: Data storage is considered relatively more important than data usage. – Supported H4b: Data usage is considered relatively more important than data collection. - Supported

0 5 10 15 20 25 30 35 40 45

28

4.3.3 Incremental Willingness to pay

The incremental willingness was calculated to determine the extent to which customers are willing to pay extra for additional protective privacy practices. Table 6 illustrates that customers are willing to pay more for more privacy-protective privacy practices. The incremental willingness to pay clearly indicates that customers are willing to pay €1.12 extra per month to have their data stored in Iceland compared to the United States (€-0.69 + €0.43), €0.76 more for data not used compared to used external (€0.29 + €0.57) and €0.25 more for only collecting e-mail compared to e-mail, age, gender, birthdate and mobile number (€0.11 + €0.14). Strikingly, customers are equally willing to pay extra for ‘not used’ compared to ‘used internally’. This may indicate that respondents are willing to accept that companies use their data for internal purposes. Another explanation is that ‘not used’ is not realistic since e-mail companies always have to use data to process information.

Attributes beta Incremental Willingness to Pay per month

Calculation Price val. -0.7915 Data Storage Brazil €-0.69 -0.5479/ 0.7915 United States €0.26 0.2076/0.7915 Iceland €0.43 0.3404/0.7915 Data Usage Not used €0.29 0.2263/0.7915 Used internal €0.29 0.2270/0.7915 Used external €-0.57 -0.4533/0.7915 Data Collection Email €0.11 0.0935/0.7915 Email, Age, Gender €0.03 0.0207/0.7915 Email, Age, Gender, Birthdate, Mobile number €-0.14 -0.1142/0.7915

Table 6: Incremental willingness to pay

4.3.4 Absolute Consideration Willingness to Pay

29 customers want to pay compared to the ‘none’ option. It can be argued that asking more than this maximum for a product will cause customers to reject buying it, but they will buy it if asking for the maximum or lower. To find the consideration willingness to pay, a demand function was created of the most ideal attribute combination. In order to determine the absolute consideration willingness to pay for an average consumer, the most preferred e-mail service (j1) was compared to the ‘none’ option by using a demand function. To calculate the probability of the most preferred option compared to the ‘none’ option, the following estimation was used:

j1 = {20GB, Stored in Iceland, Data Not used, only ‘e-mail’ collected, β*price } j2 = {None option}

prob(i|J) = exp(j1) / exp(j1) + exp(j2)

The different utilities and probabilities can be found in Appendix A2.

Graph 7: Demand function

Graph 7 conveys that the absolute consideration willingness to pay is higher than the maximum of €20 per month that is used in this paper. This is not realistic since it entails extrapolation to a willingness to pay that is not in the conjoint design. Consequently, it is not useful to state the precise willingness to pay per month, though it does indicate that people are at least willing to pay something for privacy protection. A shortcoming of this method is that all other market conditions are assumed to be stable; for example, no competitors are in this demand function, which would not be the case in a realistic market situation.

30

4.4 Moderating effects 4.4.1 Privacy concerns

The result of the main effects model including the moderator effects of privacy concerns are shown in table 8.

Attributes Class1 Wald p-value Mean

Storage 20GB 0,1679 2,8076 0,25 0,1679 5GB -0,0151 -0,0151 1GB -0,1528 -0,1528 Data Storage Brazil -0,2251 15,9968 <0,001* -0,2251 United States 0,4297 0,4297 Iceland -0,2046 -0,2046 Data Usage Not used 0,0807 4,4789 0,11 0,0807 Used internal 0,1589 0,1589 Used external -0,2396 -0,2396 Data Collection Email, -0,0545 0,3047 0,86 -0,0545

Email, Age, Gender, 0,0486 0,0486

Email, Age, Gender, Birthdate, Mobile number

0,0059 0,0059 Price -1,9125 323,9436 <0,001* -1,9125 None_option -4,5153 451,9382 <0,001* -4,5153 BrazilxConcerns -0,0731 8,7803 0,0031* -0,0731 UnitedStatesxConcerns -0,0434 3,8095 0,051 -0,0434 IcelandxConcerns (reference) 0,1165 Not usedxConcerns 0,0332 2,2357 0,13 0,0332 Used internalxConcerns 0,0159 0,5156 0,47 0,0159 EmailxConcerns 0,0326 2,1632 0,14 0,0326

Email, age, genderxConcerns

-0,0042 0,0339 0,85 -0,0042 0euroxConcerns -0,2409 122,8335 <0,001* -0,2409 5euroxConcerns 0,0075 1,77 0,18 0,0075 20GBxConcerns 0,0088 0,1549 0,69 0,0088 5GBxConcerns 0,0127 0,3285 0,57 0,0127

31 Table 7 indicates that only data storage remains significant in the main effects for the concepts together with price, storage and the ‘none’ option. The interaction BrazilxConcerns is significant with a negative utility (a=1%) (U=-0.0731), which means that people who score highly on privacy concerns have a more negative utility for storing data in Brazil compared to the baseline Iceland. UnitedStatesxConcerns is slightly insignificant (a=5%). The utility of the reference level (Iceland) is the sum of moderating utilities of Brazil and the United States multiplied by -1, which yields (U=0.1165). Since these interactions are relative to the baseline (Iceland), one can say that someone who scores highly on privacy concerns is more likely to have a high utility to store their data in a country with a high level of privacy legislation (Iceland), but that there is no significant difference between storing their data in low (Brazil) or medium (the United States) levels of legislation. Therefore, hypotheses H5a and H5c are not supported and H5b, partly since it is not significant for UnitedStatesxConcerns.

H5a: A higher level of privacy concerns will positively influence the preference for a lower amount of data collection. - Not supported

H5b: A higher level of privacy concerns will positively influence the preference for data storage in a country with a higher level of privacy legislation. - Partly supported

H5c: A higher level of privacy concerns will positively influence the preference for a lower level of data usage. - Not Supported

4.4.2 The difference of professional and private settings

Since respondents were randomly allocated between the two settings (1 = professional with n=129), 2 = private with n=130) with a decent sample size, one would not expect significant differences for the control variables in the different groups. To confirm, a chi-square test and independent samples t-test were conducted to check whether the control variables significantly differ across the groups. For the categorical variables ‘current e-mail provider’, ‘privacy misuse’, ‘gender’, ‘age’, ‘mobile OS’, ‘e-mail use frequency’, ‘country’ and ‘education’, a chi-square test was performed. This test is appropriate since the setting and the control variables consist of two or more categorical, independent groups. For the interval variable ‘innovativeness’, an independent samples t-test is performed since the setting is categorical and innovativeness interval.

32 significantly from respondents in private setting (m=3.8, sd = 1.55). The chi-square tests for the categorical variables do not reveal any significant differences. In a chi-square test, the assumption is that there must be at least 10 expected frequencies in each group of the categorical variable in a 2x2 table, and for a table larger than 2x2, the assumption is that the expected count is no fewer than five in more than 20% of the cells. When the assumption is violated, one can consider the likelihood-ratio p-value. Table 9 presents the results.

Control variable Degrees of freedom Pearson Chi-square p-value Assumptions violated Likelihood-Ratio p-value Mobile OS 2 0.699 No Gender 1 0.112 No E-mail provider 3 0.273 No Country 2 0.696 No Education 3 0.974 Yes (25%) 0.974 E-mail use frequency 2 0.925 No Privacy misuse 1 0.145 No Age 3 0.177 Yes (25%) 0.174

Table 9: test results chi-square test to investigate differences in control variables between groups

In order to assess differences in preferences for data storage, data usage and data collection between a professional setting and private setting, a two set-class model is created. Two classes are created in this model: one with the professional setting and one with the private setting. The log likelihood is almost equal to the main effects model (-4187.189). The result reflects no significant effect for data storage, data usage and data collection, so hypothesis H6 cannot be supported. The output can be found in Appendix A3.

H6: Email users in a private setting have a higher willingness to pay/utility for privacy practices (data collection, data storage, data usage) compared to e-mail users in a professional setting for the e-mail client. – Not supported

33 important when choosing an e-mail client for professional matters. An alternative explanation could be that the relative importance is lower because employees themselves do not pay for the service; thus, price is likely to be less important for them. For the ‘none’ option, the relative importance is higher for private 36% 2.4757) than for professional 40% (U=-2.1929), which may indicate that chosen alternatives in the choice design are generally perceived as more interesting when using the e-mail client for private use compared to business use. Graph 8 provides an overview of the relative importance for the two settings.

Graph 8 – Relative importance professional versus private setting

4.5 Latent class analysis - segmentation

The presented choice models neglects heterogeneity and assumes that all respondents react the same way. However, it is possible that consumers differ in their preferences. It is logical to create segments to accommodate these differences. Latent class segmentation is an appropriate method to create segments of respondents (Eggers 2016). The latent class method assumes that respondents belong to discrete segments that differ in terms of preferences, and that respondents are homogeneous within segments. For the segmentation, the covariates of

gender, age, country, education, privacy concerns and innovativeness are added to the model.

To find the ideal number of segments, the model is estimated for 1-12 classes. The model with the lowest value for the Bayesian Information Criterion (BIC) and Consistent Akaike Information Criterion (CAIC) is the preferred model, which was used for defining the segments. Table 10 indicates that the BIC and CAIC score for the nine-class model are the lowest, so this model was thus used for further analysis.

0 10 20 30 40 50

Storage (GB) Data Storage Data Usage Data Collection Price None-option

Relative importance

34

Table 10: number of segments selection

At 83%, the hit rate of the nine-class model is higher than the 69% of the main effect model. The likelihood ratio test was performed to calculate if the nine-class model significantly improves in fit compared to the main effect model. The Chi-square statistic exceeded (chisq = 2370.66) the critical value of 183.200 found in the Chi-square table (df=113, α=1%), indicating that the fit of the model significantly improved:

LL Chisq Npar Df Hitrate AIC(LL)

9-class model -3001.77 2370.66 146 113 83% 5777.54 Main effect model -4187.1 11 248 69% 7878.3

Table 11: Main model versus main effect model

The nine segments are each relatively small, with the largest class comprising 16% of the sample and smallest representing 4% of the sample. Fortunately, all attributes for all segments are significantly different from one another. An overview can be found in Tables 12 and 13. When adding up the segment sizes of segment 1, segment 2 (price most important), segment 7 (storage GB most important) and segment 6 (‘none’ option not negative), the total size is 47%. This indicates that nearly half of the consumers were not really interested in the privacy protective attributes. However, segment 9 has a strong negative utility for data storage in Brazil (U=-4.9385) and a strong positive utility for Iceland (U=4.5703). For this segment, country of data storage seems particularly important, with a relative importance of 47% compared to only 4% for price. Segment 4 also has a strong negative utility for Brazil (U=-2.8962) and a high one for Iceland (=1.6771). Segments 4 and 9 together are 18% of the sample size, which indicate that there could a niche market for e-mail service with country of data storage as USP.

35 Attributes 1 2 3 4 5 6 7 8 9 p-value p-value2 Segment Size 16,25% 14,97% 14,09% 13,96% 13,35% 11,46% 6,29% 5,36% 4,27% Storage 20GB 0,202 0,1038 0,0341 0,2286 0,2736 0,289 2,1636 0,2581 -0,0136 <0,001* <0,001* 5GB 0,328 0,1225 0,2123 0,0909 0,0658 -0,0249 0,5111 -0,2635 -0,843 1GB -0,5299 -0,2263 -0,2465 -0,3195 -0,3395 -0,2641 -2,6748 0,0054 0,8566 Data Storage Brazil -0,9744 -0,7451 -0,2263 -2,8962 -0,7345 -0,5916 -0,1598 -0,1768 -4,9385 <0,001* <0,001* United States 0,4218 0,5284 0,0526 1,2191 0,1161 0,3908 0,0723 0,2227 0,3681 Iceland 0,5526 0,2167 0,1737 1,6771 0,6184 0,2008 0,0875 -0,0459 4,5703 Data Usage Not Used 0,7504 0,228 0,3594 0,1061 1,5897 0,3927 0,3044 -0,0081 0,2587 <0,001* <0,001* Used by DaMail itself 0,3058 0,0013 0,558 0,2663 1,2111 -0,0226 0,0317 1,379 0,3435 Used external -1,0562 -0,2293 -0,9174 -0,3724 -2,8008 -0,3701 -0,3361 -1,3709 -0,6023 Data Collection E-mail 0,2447 0,1466 0,1886 0,2381 0,2385 0,6073 -0,0341 -0,1773 -0,1167 <0,001* <0,001* Email, Age, Gender 0,0031 0,0991 0,2056 0,1344 0,0704 -0,0298 -0,3256 -0,0863 -0,2804 Email, Age, Gender, Birthdate, Mobile number -0,2478 -0,2457 -0,3943 -0,3725 -0,3089 -0,5775 0,3597 0,2636 0,3971 Price -2,5163 -0,2501 -3,7214 -1,5449 -1,5951 -0,3319 -1,2091 -0,1717 0,383 <0,001* <0,001* None_option -2,2852 -4,286 -8,4349 -2,6928 -2,5738 0,4146 -3,1423 -10,6822 -6,6262 <0,001* <0,001*

Table 12: Utilities for the segments. *Significant at a 1% level

Segment 1 2 3 4 5 6 7 8 9 Storage (GB) 7% 5% 2% 5% 5% 12% 40% 3% 8% Data Storage 13% 18% 2% 38% 11% 22% 2% 3% 47% Data Usage 15% 6% 8% 5% 35% 17% 5% 18% 5% Data Collection 4% 5% 3% 5% 4% 26% 6% 3% 3% Price 42% 7% 40% 25% 25% 15% 20% 2% 4% None_option 19% 59% 45% 22% 20% 9% 26% 71% 33%

36 Only the covariates of innovativeness, privacy concerns and gender had a significant effect. Accordingly, only these variables are taken into account in further defining the segments. Appendix A4 depicts the p-values of the covariates.

Segment 3 has the lowest score on privacy concerns (3.16), a high relative importance for price (40%) and a very low importance for the difference privacy practices attributes. Although hypotheses H5a and H5c were rejected based on the moderator analysis, this finding still indicates that customers who score highly on privacy concerns will have a higher utility for an e-mail client that is more protective of their privacy, although this is not significant based on the moderator analysis. Segment 9, with a very high importance on country of data storages, is clearly male-dominate. This segment has high score for innovativeness and privacy concerns (5.23 and 5.61), which gives more weight to the partly supported hypothesis H5b. Table 14 presents an overview.

37

4.5 Summary of the results

The following table gives an overview of hypothesis:

Hypothesis Supported?

H1: A lower amount of personal data collection positively influences utility/choice for the e-mail client

Yes

H2: Data storage in a country with a higher level of privacy legislation positively influence utility/choice for the e-mail client

Yes

H3: A lower level of data usage has a positive effect on utility/choice for the e-mail client

Yes

H4a: Data storage is considered relatively more important than data usage Yes

H4b: Data usage is considered relatively more important than data collection Yes

H5a: A higher level of privacy concerns will positively influence the preference for lower amount of data collection

No

H5b: A higher level of privacy concerns will positively influence the preference for data storage in a country with a higher level of privacy legislation

Partly

H5c: A higher level of privacy concerns will positively influence the preference for a lower level of data usage

No

H6: Email users in private setting have a higher willingness to pay/utility for the privacy practices (data collection, data storage, data usage) compared to e-mail users in professional setting for the e-mail client

No

38

5. Conclusion

This section discusses the main results of this paper and presents the answers to the research questions and hypotheses. The answer to the first research question covers H1, H2, H3, H4a and H4b.

RQ 1: To what extent do the privacy practices of data storage, data collection and

data usage (relative to each other, i.e. relative importance) affect the choice for an email client?

In line with the presented literature and expectations (Hui, Teo and Lee 2007), if a larger amount of data is collected, the utility for this e-mail provider is lower than when less data is collected. However, the relative importance for data collection is only 7%, which is rather low and is consistent with the literature (TRUSTE 2016). Hypothesis H1 is therefore supported. The most important privacy practice is data storage, with a relative importance of 23%. In agreement with the literature, customers indeed have a higher utility for an e-mail client when data is stored in a country with a higher level of privacy legislation. Accordingly, hypothesis H2 can be supported. The difference in utilities between data not used and data used internally is not substantial, but there is a clear negative utility for data used externally. This means that customers are negative towards an e-mail client who, for example, sells advertisements to external companies. This supports the findings of previous research that customers in general have a negative feeling about external usage of data, such as selling data to external advertisers (Alreck and Settle 2007). With a relative importance of 18%, data usage is found to be the second most important privacy practice; therefore, hypotheses H3, H4a and H4b are supported.

The second research question was not linked to a specific hypothesis, but follows from the results of the utilities of the main effects, and is therefore connected with the first research question.

RQ2: To what extent are customers willing to pay for more privacy-protective levels

of privacy practices mentioned in question 1?

privacy-39 protective levels, which corroborates previous findings regarding the cloud storage market (Martin and Murphy 2017; Tsai et al. 2011). Notably, respondents are equally willing to pay extra for ‘using data internally’ and ‘not using,’ which could imply that customers are willing to accept when companies use their data for internal matters, and maybe even see some associated benefits. However, since utilities in conjoint are relative to each other and the level ‘not used’ is in also in this attribute, this may convey the wrong image since ‘not used’ is not realistic; e-mail companies may always use data to process requests.

The answer to the third research question also addresses hypotheses H5a and H5b.

RQ 3: Do customer privacy concerns moderate the effect mentioned in question 1?

If a person scores higher on privacy concerns, there is a significantly high utility to store their data in a country with a high level of privacy legislation, but there is no significant difference between storing their data in a country with low or medium levels of legislation. Privacy concerns did not significantly moderate the other attributes, so customers with higher scores on privacy concerns did not significantly differ in preference for data collection and data usage compared to customers with lower scores on privacy concerns. This is contrary to the findings of previous research, which have indicated that privacy-concerned customers are more likely to exhibit privacy-protective responses (Son and Kim 2008). This paper has only found a clear link between privacy concerns and country of data storage. One possible explanation for the insignificance could be that the majority of respondents rated themselves as ‘privacy concerned’, even those who do not actually have behaviour (choices in the conjoint study) that matched their own privacy concerns rating in the survey questions, which may also explain the privacy paradox (Berendt, Günther, and Spiekermann 2005; Norberg et al. 2007). In view of this, hypothesis H5b is partly supported, and hypotheses H5a and H5c are not supported.

The fourth research question is linked to the last hypothesis, H6.

RQ4: Is there a difference in answers to the first two research questions in a

business/professional setting compared to a private/personal setting?

40 However, it is possible that there are differences in preferences for privacy protection which the conjoint design does not cover.

The last research question regards the segmentation analysis, which did not have specific hypothesis.

RQ5: Which interesting customer segments can be formed based on the

demographics and choice utilities?

41

6. Discussion

6.1 Managerial and theoretical implications

This paper has some interesting managerial implications. By comparing the relative importance of the privacy practices, one can say that customers care less about what is collected, more about how their data is used and the most about where their data is stored. This is useful for e-mail providers and other cloud services since it may be advantageous to incorporate this in their marketing strategies. Customers are generally fine with collecting more data if this data is used for internal purposes (for example improving the service) and stored in a country with a higher level of privacy legislation. Only using data internally (no advertising) and storing it in a country with more privacy legislation could be more expensive or generate less money, but doing this increases the utility and the willingness to pay. No significant differences were found for the privacy practices between professional and private settings – only a lower price sensitivity for the professional setting, through this may be the case because the company pays for the service and not the employee themselves. For the privacy practices, no differences in marketing around these aspects for professional and private are needed, but when choosing an e-mail client for business, less privacy sensitivity is expected. However, in general, price has a relative importance of 41% and is thus the most important attribute, which explains the current business models of companies such as Google and Yahoo. Nonetheless, when evaluating the segments, there is a clear (niche) market for an e-mail client that focuses on certain privacy practices. Country of data storage is especially an aspect in which companies can make a difference by specifically targeting privacy-concerned male customers. The dominant e-mail providers at this moment totally do not mention this in their marketing. It can be argued that these findings are not only useful for e-mail providers, but also for other cloud services, such as cloud storage, or cloud collaborating services, such as Google Docs.Theoretically, this paper as aimed to academically contribute to the concepts of privacy practices. As far as is known, the approach of using as realistic of a conjoint study as possible in combination with segmentation analysis is unique to offer an understanding of the influence of data collection, data storage and data usage on the choice of an e-mail client.

6.2 Limitations

42 alternatives and that there are no switching costs, which is not an accurate reflection of reality. For the attributes of data collection and data usage, the levels ‘data not used’ and only collecting ‘e-mail’ are not very realistic, as companies always have to use data, and to have an e-mail client, IP-logs and other website meta-data are collected in order to process the information.

6.3 Future Research

43

6. Appendix

6.2 Appendix A2

44

6.3 Appendix A3

Model for Choices

Attributes Professional Private Wald p-value Wald(=) p-value Mean Std.Dev.

45

6.4 Appendix A4

Covariates Wald p-value X4_Gender 13,5322 0,095*** X37_age 7,5756 0,48 X41_country 9,1499 0,33 X49_education 18,5425 0,018 Innovative 28,5115 <0,001* PrivacyConcerns 41,8495 <0,001*

46

7. References

Ackerman, M. S., Cranor, L. F., & Reagle, J. (1999, November). Privacy in e-commerce: examining user scenarios and privacy preferences. In Proceedings of the 1st ACM

conference on Electronic commerce (pp. 1-8). ACM.

Acquisti, A., and Grossklags, J. (2005). Privacy and rationality in individual decision making. IEEE Security & Privacy, 3(1), 26-33.

Agarwal, R., and Rodhain, F. (2002). Mine or ours: email privacy expectations, employee attitudes, and perceived work environment characteristics. ystem Sciences, 2002.

HICSS. Proceedings of the 35th Annual Hawaii International Conference, 2471-2480.

Alreck, P. L., & Settle, R. B. (2007). Consumer reactions to online behavioural tracking and targeting. Journal of Database Marketing & Customer Strategy Management, 15(1), 11-23.

Amazon. (n.d.). Special Offers on Your Fire or Kindle. Retrieved April 2017, from Amazon: https://www.amazon.com/gp/help/customer/display.html?nodeId=200671290

Andrade, E.B., Kaltcheva, V., and Weitz, B. (2002), “Self-disclosure on the web: The impact of privacy policy, reward, and company reputation,” Advances in Consumer Research, 29, 350–54.

Ariely, D. (2009). The end of rational economics. Harvard Business Review, 87(7-8), 78-84. Awad, N. F., & Krishnan, M. S. (2006). The personalization privacy paradox: an empirical

evaluation of information transparency and the willingness to be profiled online for personalization. MIS quarterly, 13-28.

BCG (2012), “The value of our digital identity.”

Bellman, S., Johnson, E. J., Kobrin, S. J., & Lohse, G. L. (2004). International differences in information privacy concerns: A global survey of consumers. The Information

Society, 20(5), 313-324.

Benenson, Z., Gassmann, F., & Reinfelder, L. (2013, April). Android and iOS users' differences concerning security and privacy. In CHI'13 Extended Abstracts on Human

Factors in Computing Systems (pp. 817-822). ACM.

47

Communications of the ACM, 48 (4), 101–6.

Bijmolt, T. H., Heerde, H. J. V., & Pieters, R. G. (2005). New empirical generalizations on the determinants of price elasticity. Journal of marketing research, 42(2), 141-156. Casadesus-Masanell, R., and Hervas-Drane, A. (2015). Competing with privacy. Management

Science, 61(1), 229-246. Retrieved from & Hervas-Drane, A.

Chung, T., Wedel, M., and Rust, R. (2016). Adaptive personalization using social networks.

Journal of the Academy of Marketing Science, 44(1), 66-87.

Culnan, M. J., and Armstrong, P. K. (1999). Information privacy concerns, procedural fairness, and impersonal trust: An empirical investigation. Organization science, 10(1), 104-115.

Dinev, Tamara, Massimo Bellotto, Paul Hart, Vicenzo Russo, Ilaria Serra, and Christian Colautti (2006), “Privacy calculus model in e-commerce–a study of Italy and the United States,” European Journal of Information Systems, 15 (4), 389–402.

EEN Vandaag. (2016). Onderzoek: Privacy, belangrijk of niet? EEN Vandaag.

Eggers, F. (2016). Choice-based Conjoint Preference-based Segmentation [PowerPoint slides]. Retrieved from https://nestor.rug.nl/bbcswebdav/pid-8136160-dt-content-rid-

9029200_2/courses/EBM079A05.2016-2017.1B/ME_Conjoint_Lecture_3%283%29.pdf

Eggers, F., & Sattler, H. (2011). Preference measurement with conjoint analysis: Overview of state-of-the-art approaches and recent developments. GfK Marketing Intelligence

Review, 3(1), 36.

EY (2015), “Creating trust in the digital world. EY’s Global Information Security Survey.” Gibbs, S. (2014, April 15). Gmail does scan all emails, new Google terms clarify. Retrieved

March 2017, from The Guardian:

https://www.theguardian.com/technology/2014/apr/15/gmail-scans-all-emails-new-google-terms-clarify

Goldfarb, A., & Tucker, C. (2012). Shifts in privacy concerns. The American Economic

Review, 102(3), 349-353.

Google. (n.d.). Google Apps Email Audit API Developer's Guide . Retrieved April 2017, from Developers Google: https://developers.google.com/admin-sdk/email-audit/