• No results found

What if privacy statements would be understandable? : the effect of the presentation of information on readability and understandability

N/A
N/A
Protected

Academic year: 2021

Share "What if privacy statements would be understandable? : the effect of the presentation of information on readability and understandability"

Copied!
50
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

MSc Business Administration - Digital Business track

University of Amsterdam

What if privacy statements would be understandable? The

effect of the presentation of information on readability and

understandability.

By Mara Janssen

Name author: Mara Janssen Student number: 10082417

Date of submission: 23rd of June 2017 Version: Final

(2)

2

Statement of originality

This document is written by Student Mara Janssen who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

3 Table of contents

Abstract 4

1. Introduction 5

2. Theoretical foundations & Hypotheses 8

2.1. The privacy calculus 8

2.2. How privacy statements can influence the privacy calculus 10

2.3. Proposed new format of the privacy statement 14

2.4. Expected effects of the new privacy statement format 16

3. Research Methodology 20 3.1 Field experiment 1 20 3.1.1. Procedure 21 3.1.2. Measures 22 3.2 Field experiment 2 22 3.2.1. Procedure 23 3.2.2. Measures 24 4. Results 25

4.1. Results of Field experiment 1 25

4.2. Results of Field experiment 2 26

4.2.1. Descriptives 27

4.2.2. Checking for differences between the formats 29

4.2.3. Checking for type I Errors 30

5. Discussion 33

5.1. Discussion of the results 33

5.2. Limitations 37 5.3 Academic contributions 38 5.4 Future research 39 6. Conclusion 42 7. References 43 8. Appendix 48

(4)

4 Abstract

More and more personal data is collected online these days by companies. This leads to controversy about the potential threats to consumer privacy. Privacy statements are important means in reducing privacy concerns as they fill the information gap between the consumer and the company. However, a condition for filling this gap is that privacy statements have to be read and understood. Therefore, it is important to research how companies can present their privacy statement in a way that is comprehensible and quick to understand. In this research, a new multi-layered format with icons (MLI) was proposed with the use of the privacy calculus by Dinev and Hart (2006). This new format of the statement was tested and compared with the most often used natural language format (NL) in two field experiments. In the first field experiment an A/B test was set up on three different websites. Although the findings could not be statistically analyzed, the results show that consumers are not deterred by a clear and summarized version of how a company will process their personal data and that the accept rate for the MLI format was higher than for the NL format. In the second field experiment, the A/B test was incorporated in a survey on the customer experience of an existing website. It was found that for the MLI format, the level of understanding is higher and the format is also perceived as shorter which also has an indirect positive effect on the level of understanding. These results are a first step towards better comprehensible and shorter privacy statements.

(5)

5 1.Introduction

It seems as if the debate on privacy is slowly moving towards a clash. On the one hand, an increasing amount of consumer data is captured these days due to the widespread use of digital technologies. Analyzing this data has become a critical business capability that companies gain value from and use to obtain a competitive advantage (Chen, Chiang, & Storey, 2012). But on the other hand, the explosive growth of the internet has fired up controversy about the potential threats to privacy that the collection of data provokes (Acquisti, Brandimarte, & Loewenstein, 2015; Dinev & Hart, 2006). In books such as ‘You do have something to hide’ privacy is even entitled as the most endangered human right these days (Martijn & Tokmetzis, 2016).

Policy makers are keeping track of the growing privacy concerns among consumers and have been introducing several regulations to protect consumer privacy. In April 2016, the General Data Protection Regulation (GDPR) was approved by the European Union. The GDPR was drawn up to strengthen and unify data protection for individuals within the EU. The primary objective of the GDPR is to give consumers the control over their personal data back. Furthermore, the regulation is overarching which means that it is replacing all current data protection laws of the separate member states. Any organization that collects and/or processes data of EU citizens must comply with the regulations. The GDPR will come into force on the 25th of May 2018 (The European Parliament & the Council of the European Union, 2016).

A lot of organizations have to make big adjustments to their processes of collecting, storing and analyzing customer data to comply with the new regulation (Tankard, 2016). Many firms use their privacy statement to communicate how they go about this. Therefore, privacy statements can fill the information gap between the consumer and the firm (Culnan & Milberg, 1998; McDonald & Cranor, 2008; Milne & Culnan, 2002). However, the statements

(6)

6 that are currently provided by firms are complex, written in legal language and lengthy and are therefore rarely read by website visitors (Jensen & Potts, 2004; Steinfeld, 2016; Tsai, Egelman, Cranor, & Acquisti, 2011; Vail, Earp, & Antón, 2008). Moreover, it was found that it is practically impossible to read all the privacy statements of the websites we encounter daily. It was researched that the national opportunity cost of reading all statements and updates for America would be around $781 billion a year which equals 40 minutes a day per citizen. This research was conducted in 2008 so the current numbers would probably even be substantially higher (McDonald & Cranor, 2008).

Something clearly has to change for consumers to be prepared to read the privacy statements, which is necessary to fill the information gap between the consumer and the firm. This is also acknowledged by the European Union that places great importance on privacy statements that are transparent, clear and easy to understand (The European Parliament & the Council of the European Union, 2016). So, to comply with the new regulation, many companies will have to change the way in which they present their privacy statement. Therefore, it is important to research the optimal presentation of the privacy statement for companies. An optimal presentation would be easy and quick to understand for consumers after which they feel informed and are willing to share their data with the company. Tsai et al. (2011) found that when this is done right, consumers are prepared to pay a price premium for shopping at websites with clear and accessible privacy information. So, privacy information could even be used as a unique selling point. In this research one possible new presentation is tested with two field experiments. The research question of this paper is: What is the effect of the

presentation of the privacy statement on the level of reading and understanding by consumers and thereby their willingness to share personal data?

The remainder of this research is organized into five sections. In the section ‘Theoretical foundations & Hypotheses’, previous literature that is relevant to the research question is

(7)

7 evaluated and from this hypotheses are formulated. The next section ‘Research methodology’ describes the two experiments that were conducted to collect data. In the third section ‘Results’, the data is analyzed and some descriptive statistics and the outcomes of two parametric tests are provided. In the fourth section ‘Discussion’ the findings are discussed and interpreted. In the final section ‘Conclusion’ the findings are related to the research question.

(8)

8 2. Theoretical foundations & Hypotheses

Several concepts are important to understand how consumers decide whether to share their personal data with companies online. The privacy calculus, that describes the cost/benefit analysis consumers make when they are asked to share their personal data is the most important concept discussed in the literature (Culnan & Armstrong, 1999; Culnan & Bies, 2003; Dinev & Hart, 2006; Milne & Gordon, 1993; Milne & Culnan, 2002; Milne & Culnan, 2004; Tsai et al., 2011; Xu, Teo, Tan, & Agarwal, 2009). In this section I will therefore explain the privacy calculus further and discuss the most important literature on the concept. Furthermore, I will discuss how a privacy statement can influence the consumers’ decision on whether to share data by affecting the variables in the privacy calculus. In this research, a websites’ privacy statement is defined as a document that describes the organizations’ practices on data collection, use, and disclosure and is thereby regulating the relationship between the user and the website. The statement is designed to limit the legal liability of the company (Earp, Antón, Aiman-Smith, & Stufflebeam, 2005). When a consumer accepts the privacy statement, this means that he/she is willing to share personal information with the company under the conditions described in the statement. Personal information or personal data is by Dinev and Hart (2006, p.63) referred to as “the type of information necessary to conduct online transactions. This includes credit card numbers and identifiers and any other information that might be required to purchase goods, information, or services or to register at websites, such as home addresses and other contact information, and possibly customer or product preferences”. To understand how consumers decide on whether to share their personal information online, I reviewed the literature on consumer decision making below.

2.1. The privacy calculus

When consumers are presented with a privacy statement they have to make a choice: do they accept the conditions as explained in the statement and share their personal data with the

(9)

9 company, or don’t they accept the conditions and thereby choose not to use the service provided by the company? A lot of research has been done on consumer decision making. One of the most well-known and researched theories is the expectancy theory, which implies that individuals will always act in a way that maximizes positive outcomes and minimizes the negative ones (Mitchell, 1974; Van Eerde & Thierry, 1996; Vroom, 1964; Wanous, Keon, & Latack, 1983). This idea also reflects in the first version of the privacy calculus model, that describes a trade-off between the perceived risks and perceived benefits of disclosing personal information (Laufer & Wolfe, 1977). This cost/benefit analysis was also discussed in later influential researches (Culnan & Armstrong, 1999; Milne & Gordon, 1993; Stone & Stone, 1990). Many years later, Dinev and Hart (2006) proposed an extended privacy calculus model that focused on a consumers’ willingness to share personal information online. They have attempted to better understand the trade-off between privacy risk beliefs and the incentives to transact on the internet. Additionally, they researched the influence of this trade-off on the intention of consumers to provide personal information online. In the privacy calculus model, they have drawn up, privacy risk beliefs were found to exist of privacy concerns and risk beliefs. Trust in the company and personal interest were found to be incentives to transact with a company. The results show that trust and personal interest are important factors that can outweigh privacy risk beliefs in the decision to share personal information online with a company. To make sure that consumers share their personal data with companies, which is very important for companies as explained in the introduction, it is important that trust and personal interest (the benefits) outweigh privacy risk beliefs (the costs). This study builds on the extended privacy calculus model by Dinev and Hart (2006) to show how privacy statements can play an important role in affecting the costs and benefits in the privacy calculus.

(10)

10

2.2. How privacy statements can influence the privacy calculus

In the privacy calculus by Dinev and Hart (2006), it was found that uncertainty influences privacy risk beliefs. The constructs that together form privacy risk beliefs are perceived privacy risk and privacy concerns. Both constructs were found to be associated with feelings of uncertainty. Perceived privacy risk is described as the uncertainty for consumers that is caused by the possibility that the firm takes advantage of the personal information by sharing it with 3rd parties for example. “Privacy concerns are beliefs about who has access to the information that is disclosed when using the Internet and how it is used. The greater the uncertainty about the access and use, the greater the privacy concerns.” (Dinev & Hart, 2006, p.65). So both constructs have to do with the uncertainty of consumers on how their personal data will be used by companies and who has access to it. That perceived privacy risk and privacy concerns are interrelated and have a negative influence on the willingness to share data can be found in the research by Dinev and Hart (2006) where privacy concerns were found to have a medium direct negative effect on the willingness to share data (r = -.38, p < .01) and perceived privacy risk was found to have a small direct negative effect (r = -.15, p < .01) on the willingness to share data. Perceived privacy risk also has an indirect effect on the willingness to share data by having a medium positive effect on privacy concerns (r = .33, p < .01) and a medium negative effect on trust (r = -.23, p < .01).

Dinev and Hart (2006) were not the only ones who studied the effects of several constructs on the willingness to share data online. In another influential research, Malhotra, Kim & Agarwal (2004) have examined the effects of privacy concerns, trust and risk beliefs on the willingness to share personal data online. Their focus was on privacy concerns as they also examined the antecedents of privacy concerns (collection, control and awareness) that I will discuss later. Their results show that privacy concerns effect the willingness to share personal data indirectly by having a medium sized positive effect on risk beliefs (r = .26, p < .001) and

(11)

11 a medium negative effect on trust (r = -.34, p < .001). Risk beliefs had a direct and strong negative effect on the willingness to share data (r = -.63, p < .001) and trust was found to have a small to medium positive effect (r = .23, p < .001). From the results of these studies it can be concluded that privacy concerns and risk beliefs are significant costs of sharing personal data for consumers. Not only by having a direct negative effect on the willingness to share personal information but also by decreasing trust in the company, that is as an important construct in the calculus as trust in the company can outweigh privacy concerns and risk beliefs (Dinev & Hart, 2006). Therefore, it is important to know what causes privacy concerns and feelings of risk so companies can keep these costs in the privacy calculus as low as possible.

Feelings of uncertainty and thereby privacy concerns and risk beliefs are often found to be caused by information asymmetry or a lack of information (Culnan & Milberg, 1998; McDonald & Cranor, 2008; Milne & Culnan, 2002). The three antecedents of privacy concerns that I named earlier: collection (‘whether the exchange of personal information is fair’), control (‘whether I have control over the data’) and awareness (‘whether I am adequately informed about the use of the data’) also stress the importance of the company providing information on how personal data is used by companies (Malhotra, Kim, & Agarwal, 2004). Companies can overcome information asymmetry by being open and honest about how the company uses the personal information. Culnan and Armstrong (1999) and Culnan and Bies (2003) argue that fair information practices should be in place for consumers to be willing to share personal information. Fair information practices can be explained as procedures that inform consumers on why the information is collected, how it will be used, how it will be protected and furthermore how the consumer can control the personal data that is being collected (Culnan & Armstrong, 1999). Privacy statements have been widely adopted means to inform consumers on the fair information practices of a company. Therefore,

(12)

12 privacy statements can mediate the privacy concerns and feelings of risk that consumers can have when disclosing personal information (Milne & Culnan, 2002; Tsai et al., 2011). This was also found by Xu et al. (2009) who argued that when consumers can make a well-informed choice on whether to disclose personal information, this can lead to a lower perception of risk and privacy concerns. Furthermore, it was found that reading privacy statements is important for consumers who want to reduce the risk of disclosing personal information online (Bansal, Zahedi, & Gefen, 2016). In another research, the results showed that just the existence of a privacy statement on a website led to decreased feelings of risk and thereby more consumers disclosing their personal information online (Hui, Teo, & Lee, 2007). But privacy statements do not only have an influence on the costs of sharing personal data. If consumers feel informed by the privacy statement, this can increase the consumers’ perception that the company is trustworthy (Bansal & Gefen, 2015; Culnan & Armstrong, 1999). Trust is one of the constructs in the privacy calculus that can outweigh the previously described costs in the privacy calculus. It is defined as “a set of specific beliefs about another party that positively influence an individual's intention to conduct online transactions. These beliefs embody the expectation that another party will not engage in opportunistic behavior” (Dinev & Hart, 2006, p.66). Trust is very important in the privacy calculus as the positive relationship between trust and the willingness to share data was found to be the strongest in the entire calculus (r = .59, p < .01). In the research by Malhotra et al. (2004) a direct positive relationship was also found (r = .23, p < .001), although not as strong as in the research by Dinev and Hart (2006). Yet another study found that trust can lead to the possibility for the company to build a positive relationship with consumers whereby they feel comfortable enough to provide the company with personal information (Schoenbachler & Gordon, 2002). It is therefore important to know how trust in the company can be increased. Several studies have found that privacy statements have a positive effect on trust (Culnan & Armstrong,

(13)

13 1999; Milne & Culnan, 2004; Pan & Zinkhan, 2006; Wu, Huang, Yen, & Popova, 2012). However, it is also argued that a privacy statement can only have a positive effect on trust when consumers perceive the statement as comprehensible. Furthermore, if privacy statements are presented in a straightforward manner, consumers are more likely to trust the privacy statement and therefore the company (Milne & Culnan, 2004). This was also found in earlier research by Mile and Culnan (2002) where they found that a privacy statement must be clear and reliable for the consumer to trust the company.

So when a company wants to decrease privacy concerns and feelings of risk and increase feelings of trust, it should draw up a clear statement to inform consumers (Culnan & Milberg, 1998; Culnan & Armstrong, 1999; Culnan & Bies, 2003; Milne & Culnan, 2002; Milne & Culnan, 2004; Pan & Zinkhan, 2006; Tsai et al., 2011; Wu et al., 2012; Xu et al., 2009). However, to feel informed, consumers do have to read and understand the statement. But the form, length and legal context of the privacy statements that are currently used by most companies make it hard for consumers to understand the statement and to make a well-informed decision, hence they are rarely read by website visitors (Jensen & Potts, 2004; Tsai et al., 2011; Vail et al., 2008).

So to conclude, privacy statements can be a mean to make the benefits outweigh the costs in the privacy calculus. It is argued that when the benefits outweigh the costs, consumers are willing to share their personal data with the company. A condition would however be that consumer have to read and understand the privacy statement. Since the format of the privacy statements that are currently most often used is not comprehensible and often not read, it is important that other ways of presenting the privacy statement are researched. By analyzing the literature, suggestions of policy makers and considerations of companies I propose a new format of the privacy statement that is expected to improve the comprehensibility and readability of the statement.

(14)

14

2.3. Proposed new format of the privacy statement

For an optimal privacy statement, companies have to pay attention to both the content (whether the company has fair information practices in place) and the format (whether the statement is easy to read and understandable) of the statement (Milne & Culnan, 2002). The content of the privacy statement is usually written by lawyers since the company needs to comply with certain regulations (Earp et al., 2005). Therefore, possible changes to the content of the statement are limited. However, to make the content easier and less time consuming to understand, the format of the statement can play a role. The format of the privacy statement is therefore the focus of this research.

To make a privacy statement easier to understand and more transparent, a multi-layered format has been recommended by the European Commission in 2004, and is now recommended again in the GDPR (Data Protection Working Party of the European commission, 2004; The European Parliament & the Council of the European Union, 2016). A multi-layered privacy statement is a short summary of the content of the statement that communicates the most important points. If users want to get more information they can access the next layer that contains more extensive information (The European Parliament & the Council of the European Union, 2016). With a multi-layered format, the statement can be short and summarized, which was found to be strongly preferred by respondents in a study by Milne and Culnan (2004). When I interviewed Daniël Okma, who is the Data Protection Officer for the Persgroep (a Dutch publisher of several newspapers, magazines and websites), he stated that they are considering using the multi-layered format for the privacy statements of the websites in their portfolio. Since the multi-layered format is recommended by previous research, policy makers and considered by companies, this is the type of format that will be tested in this research.

(15)

15 One of the reasons that consumers do not read the privacy statement can also be the fact that the statement is often not actively provided on the website, meaning that consumers have to search for it. The majority of the websites only states that users agree to the privacy statement when accepting the terms of service of a website, without actually providing the statement itself. Sometimes a link to the privacy statement is provided. However, results of a lab experiment show that when a privacy statement is presented by default when opening a webpage, participants tend to read it quite carefully. On the other hand, when the participants were given the option to agree to the terms and conditions without reading the privacy statement, the majority would skip it. Furthermore, it was found that presenting the privacy statement by default had a positive effect on the understanding of the content of the statement (Steinfeld, 2016). Next to the positive effect the presentation of the statement by default can have on understanding the statement, the GDPR also contains a section on conditions for accepting the privacy statement. It requires companies to actively ask consumers for permission to use their data. Pre-ticked boxes for accepting the privacy statement without providing the content will not be permitted anymore (The European Parliament & the Council of the European Union, 2016). Because of the positive effects and the changes in regulations, the privacy statement in this research will be presented by default.

To help transfer information much faster and in a small amount of space, icons can be added to text (Hemenway, 1982). This was also acknowledged by some researchers who studied using icons in addition to written statements to increase consumers’ awareness on the use of their personal information. In despite of the positive effects of using icons, such as better and faster information finding, none of the icons that have been developed over the past few years have gained practical relevance (Hansen, 2009; Holtz, Nocun, & Hansen, 2010; Kelley, Bresee, Cranor, & Reeder, 2009). In The Netherlands, there has been a recent attempt to design icons for privacy in the De Nationale Denktank, a Dutch foundation that selects

(16)

16 twenty multidisciplinary, young and talented academics to brainstorm on practical solutions for social problems (Stichting de Nationale Denktank, 2017). Eva de Leede, who participated in De Nationale Denktank in 2014, designed icons for the Datawijzer that provides consumers a quick overview of how a website is using their personal information. Consumers can then use this information the make an informed decision on whether they want to share their personal information under the conditions of the website (De Nationale DenkTank, 2014). That this idea was developed in De National Denktank shows that there is an interest from society in using icons for privacy information purposes. However, specific icons for this purpose were never developed and tested. Since previous literature shows that the use of icons can be beneficial for transferring privacy information, icons are added to the new format.

The new format of the privacy statement that will be tested in this research will be referred to as ‘Multi-layered format with icons’. The format that is currently most often used by companies will be referred to as ‘Natural language format’.

2.4. Expected effects of the new privacy statement format

Privacy statements have the potential to influence the privacy calculus by decreasing privacy concerns and feelings of risk and by increasing feelings of trust (Culnan & Milberg, 1998; Culnan & Armstrong, 1999; Culnan & Bies, 2003; Milne & Culnan, 2002; Milne & Culnan, 2004; Pan & Zinkhan, 2006; Tsai et al., 2011; Wu et al., 2012; Xu et al., 2009). One condition would be that consumers have to read and subsequently understand the statement. I proposed a new format of the privacy statement to fulfill this condition. Whether this multi-layered format with icons (MLI) can really improve understandability and readability when compared with the commonly used natural language format (NL), is tested in this research with the use of hypotheses.

(17)

17 Milne and Culnan (2004) have found that consumers did not read privacy statements because they perceived the statements as being too long. This was also argued by McDonald and Cranor (2008) who stated that the opportunity cost of reading privacy statements is too high and therefore consumers do not read the statements. In a longitudinal study of online privacy statement readability, it was argued that long statements can be formatted differently in order to promote navigation and thereby improve readability. Layered statements were proposed as an alternative format that has the potential to improve the usefulness of the statement for consumers since it can be perceived as shorter (Milne, Culnan, & Greene, 2006). In the same research, the results of a fairly old but relevant survey were shared in which it was found that the respondents had a strong preference for short privacy statements (77%) and also for companies to provide a consistent summary or checklist of the privacy statement (70%) (Harris Interactive Inc, 2002). Furthermore, studies have found that if consumers feel it will take them a long time to read the statement they will not read it (Jensen & Potts, 2004; Milne & Culnan, 2004). This was also found by Pan & Zinkhan (2006) who conducted two experiments with privacy statements of which the results showed that when consumers are exposed to too much information at once, they will not read it to save time and energy. Since the MLI format is a summarized, layered version of the natural language format I expect the following:

H1. Consumers take more time to read the Multi-layered format with icons than the Natural

language format.

H2. Consumers perceive the Multi-layered format with icons to be less long than the Natural

language format.

However, readability is not the same as comprehension. If consumers read the privacy statement they need to understand the content to feel informed (Milne et al., 2006). As I have

(18)

18 argued earlier in this research, consumers feeling uncertain on how their personal data is used by companies and who has access to them is one of the biggest costs in the privacy calculus. Privacy statements can fill the information gap since they contain information that can empower consumers by explaining what their rights are, what information about them is collected and stored and how they can control this information (Steinfeld, 2016). Privacy statements cannot only empower consumers but if they understand the statement, this can also decrease privacy concerns and risk beliefs (Milne et al., 2006; Xu et al., 2009). Furthermore, several studies have found that when consumers perceive a privacy statement as understandable, they are more likely to read and trust it (Milne & Culnan, 2004; Milne et al., 2006; Schoenbachler & Gordon, 2002). It is therefore important to test the level of understanding for both privacy statement formats. Pan & Zinkhan (2006) found that consumers perceived short and straightforward privacy statements as easier to understand. Due to the simplified and short representation of the privacy statement for the MLI format I expect the following:

H3. The level of understanding for the Multi-layered format with icons is higher than for the

Natural language format.

It becomes clear from the literature that I discussed up until now that privacy concerns, risk beliefs and trust are the constructs of the privacy calculus that have been researched most often. However, Dinev and Hart (2006) have also discussed the construct personal interest in the privacy calculus. This is the intrinsic motivation to transact on the internet which gives consumers access to a huge collection of information, goods, and services. In the survey used to collect data, they compared the importance of personal interest with the importance of for example privacy concerns: “The greater my interest to obtain a certain information or service from the internet, the more I tend to suppress my privacy concerns.”. This variable is a benefit of sharing personal data since it provides consumers access to online information and

(19)

19 services. Dinev and Hart (2006) found that Personal interest is an important construct in the privacy calculus as the relationship that was found between personal interest and the willingness to share personal information was the second largest effect found in the study (r = .48, p < .01) after trust. I would expect that when consumers are interested in the information or service of a specific website, this increases their willingness to share data and therefore will spend less time reading the statement. Although a large effect was found between personal interest and the willingness to share data, I could not find any research that has also tested this relationship. It is therefore important to test for personal interest in this study:

H4. The website where personal interest is higher leads to a lower reading time of the

privacy statement when compared with the other two websites.

If the hypotheses formulated above are accepted in this research, this would mean that trust and personal interest outweigh privacy concerns and risk beliefs in the privacy calculus for the MLI format and therefore more consumers should be willing to share personal data for the MLI format when compared with the NL format. However, I argue that this is not the case due to the Privacy Paradox. Consumers state that they have growing concerns regarding their online privacy and the way in which their personal data is used by companies. However, actions do not reflect in these concerns as consumers still freely provide firms with their personal data (Taddicken, 2014). This conflicting situation is called the ‘Privacy Paradox’ which is referred to as “the relationship between individuals’ intentions to disclose personal information and their actual personal information disclosure behavior” (Norberg, Horne, & Horne, 2007, p.100). Because of this paradox, I expect the following:

H5. The Natural language format is equally often accepted as the Multi-layered format with

icons.

(20)

20 3. Research Methodology

As explained above, the online behavior of consumers does not mirror their expressed privacy concerns which is called the privacy paradox (Norberg et al., 2007; Taddicken, 2014). The implication of the privacy paradox for this research is that when you tell participants your research is about privacy they will most likely change their behavior and express privacy concerns way more than their actual behavior would show. This could mean that they do not accept the privacy statement where they normally would for example. Since for this study it is important to know how consumers would respond to the new format of the privacy statement in real life, I decided to conduct field experiments to study consumers’ real online behavior.

3.1. Field experiment 1

First, I set up a field experiment that consisted of A/B testing on three different websites where consumers are likely to experience a different level of ‘personal interest’. The strength of this method is that it observes the behavior of real website visitors that do not know they are participating in an experiment. Therefore, I should be able to measure real behavior as the privacy paradox should not effect this method of data collection.

The first website was a recruitment website (www.worksprout.com), where a high level of personal interest was expected due to the possibility of finding a job. The second website was a web shop (www.biano.nl) where a lower level of personal interest was expected. The third website was a price comparison website (www.vindproduct.nl) where a lower level of personal interest was expected as well. The two different treatments consisted of the Natural language format (NL) versus the Multi-layered format with icons (MLI). Because it is important that consumers can interpret the icons, the design of the icons was based on those already in use by companies such as Facebook. The experiment design was 2x3 where

(21)

21 website visitors were randomly assigned to one of the two treatments (between subjects). This leads to the following graphical representation of the design:

Recruitment Web shop Find product

Natural Language (NL) NL - Recruitment NL - Web shop NL – Find product Multi-layered with Icons

(MLI) MLI - Recruitment MLI - Web shop MLI – Find product

3.1.1. Procedure

When a website visitor entered the website, the privacy statement was presented by default as explained earlier. The visitor was presented with one of the two treatments and was asked to disagree with or accept the privacy statement. When the option ‘disagree’ was chosen, the visitor could not use the services provided by the website. After the visitor had chosen to decline or accept the privacy statement, a new screen would pop-up in which the visitor was asked to participate in a short survey. This survey consisted of four short statements about the comprehensibility, length and overall perception of the privacy statement. The four statements were drawn up following a research on metrics for analyzing the form of security policies by Goel & Chengalur-Smith (2010). In this research, three dimensions were found to be important for effective security policies: clarity, breadth and brevity. For this research, only the items for clarity and brevity were used since breadth focuses on legal and security content which is not considered in this research. Since I wanted to keep the survey as short as possible so more visitors would participate, only the items with the highest reliability were used.: 1. “The statement is easy to understand” 2. “The statement is easy to read” 3. “The statement presented the information clearly” (clarity) 4. “The statement is too long” (brevity) (Goel & Chengalur-Smith, 2010). All statements were rated on a 5-point Likert scale where 1 = strongly disagree 2 = disagree 3 = neutral 4 = agree 5 = strongly agree. Screenshots of the different treatments on the different websites can be found in the Appendix.

(22)

22

3.1.2. Measures

When website visitors were presented with the statement, their response (disagree, accept) was recorded. Since in a field experiment it is not possible to know if a participant has actually read the privacy statement, the time a participant spends on the webpage when the statement is displayed was analyzed. This operationalization of reading the privacy statement was also used in a study by Steinfeld (2016) on how consumers read privacy statements. The A/B tests were programmed via Google Analytics. After two days, all three A/B tests had more than enough participants so the tests were stopped. When the results were exported from Google Analytics it turned out that only aggregated data was available. This data could not be analyzed in SPSS and was therefore not suitable to make any inferences on the hypotheses. Furthermore, none of the participants in the A/B test filled out the survey. Although the results from the A/B tests have been very positive for the research (as will be explained in the results section), more data had to be collected to be able to make inferences on the hypotheses in this study. Therefore, another field experiment was set up.

3.2. Field experiment 2

In the second field experiment, the A/B test was processed in a survey on the customer experience of one of the three websites that were used in the first experiment. Since participants now know their responses will be observed and analyzed, this could make the method less reliable for making inferences on real behavior. However, since the experiment was carried out in the beginning of the survey and it is not communicated that it is an experiment that deals with privacy, I still consider this method reliable enough to test the hypotheses.

Biano.nl was the chosen website because it is a new website that is not very specialized such as worksprout.com for example, which is a recruitment platform for developers. It addresses

(23)

23 quite a general audience and therefore is considered fit for a survey that is distributed to a broad population. Furthermore, the results of the survey are of interest to Biano.nl to improve their customer experience. For this experiment, the two treatments were the same as for the first experiment: Natural language format versus Multi-layered format with icons. The icons that were used were the same as the ones used in the first experiment. The experiment design was 2x1 where website visitors were randomly assigned to one of two treatments (between subjects). This leads to the following graphical representation of the design:

Biano.nl customer experience survey Natural Language

(NL) NL – Biano.nl customer experience survey Multi-layered with

Icons (MLI) MLI – Biano.nl customer experience survey

3.2.1. Procedure

It was communicated to participants that the survey was about the customer experience of the website Biano.nl. Because the survey involved a website visit, participants had to agree with the privacy statement first. Just like in the first experiment, participants were randomly assigned to one of the two treatments. When the participant declined the statement he/she was directed to the end of the survey. After accepting the privacy statement, the participants were asked to answer the same four statements as for the first experiment that were rated on a 5-point Likert scale again. When the participants had answered these statements, they were asked to visit biano.nl and then answer some questions on the customer experience of the website. At the end of the survey the participants were asked to provide some demographic information (gender, age and level of education). Because this experiment design only involved one website, the results could not be compared with other websites to check for the influence of personal interest. To be able to make some inferences on the influence of

(24)

24 personal interest, the following statement was added to the customer experience survey: “After this survey I am interested to use Biano.nl.”. Participants had to rate this statement on the same 5-point Likert scale as described above. Because this experiment was set-up using Qualtrics, it was not possible to redirect customers to the specific topic of the privacy statement using the ‘read more’ button. However, this is not considered a problem in the research since not even 3% of website visitors clicked on one or more of the ‘read more’ buttons in field experiment 1. Furthermore, a link to the full privacy statement could be provided below the layered version in Qualtrics like in field experiment 1.

3.2.2. Measures

When participants were presented with one of the two different privacy statement formats their response was recorded (Accept Yes/No) and at the same time how much time they spend looking at the statement. If a participant decided not to accept the privacy statement they were redirected to the end of the survey.

(25)

25 4. Results

In this section I will elaborate on the results that have been found in the two different field experiments that were carried out.

4.1. Results of Field Experiment 1

As stated before it was not possible to draw statistically significant inferences from the A/B tests on the different websites that were run with Google Analytics. However, some interesting results were found in the tests that were summarized in table 1 below. It was found that on all websites, the accept rate for the MLI format was higher than for the NL format. The biggest difference in accept rate between the formats was found on Vindproduct.nl where the difference was 21%. Decline rates were lower for the MLI format for Vindproduct.nl and Worksprout.com, the rate was the same for Biano.nl. Another noticeable difference between the two formats was found in the No Choice rate that was substantially higher for the NL format. This confirms the findings of Jensen & Potts (2004) and Milne & Culnan (2004) that if consumers think reading the statement will take them too much time they will not do it and decide not to accept the statement. The average time on page was also measured in this experiment. On Vindproduct.nl the average time on page was 2.5 times higher for MLI compared with NL (5 seconds vs. 2 seconds). For Biano.nl the average time on page was the same for the two formats (5 seconds) and for Worksprout.com the average time on page was 40% higher for the MLI format (7 seconds vs. 5 seconds). As I am assuming that no one can read any of the formats within the average time on page of 5 seconds, consumers are still not reading the statement. However, this experiment shows that the MLI format of the statement can have positive results for companies as more consumers accept the privacy statement in this case. So, consumers are not deterred by an explicit summary of the privacy statement.

(26)

26 As described in Hypothesis 4, I expect that the website that has a higher level of personal interest (which in this case would be the recruitment platform Worksprout.com), would lead to a lower reading time of the privacy statement in comparison with the other two websites. However, from the results it becomes clear that the average time on page for the MLI format is a little bit higher for Worksprout.com when compared with the other websites, but for the NL format the average time on page for Biano.nl was higher. Since these data are not very convincing, I argue that the effect stated in the hypothesis will not hold. Another noticeable result is that the accept rate for Worksprout.com is lower when compared with the other two websites. Since this website collects data that could be perceived as more sensitive (job-related data), consumers could be less willing to accept the statement and share this data.

Table 1. Results Field Experiment 1

4.2. Results of Field Experiment 2

In the second field experiment was tested whether the results from field experiment 1 can be replicated with statistical significance. Furthermore, there was the possibility to look into the level of understanding and perceived length of the statement as well.

The customer experience survey was distributed online via Qualtrics which resulted in 159 respondents in total, of which 15 respondents did not accept the privacy statement. For the

Website Format # Users # Accept % Accept # De-cline % De-cline #No Choice % No Choice Average Time on Page* MLI 144 133 92% 6 4% 5 3% 00:00:05 NL 140 100 71% 10 7% 30 21% 00:00:02 MLI 401 366 91% 11 3% 24 6% 00:00:05 NL 394 295 75% 11 3% 88 22% 00:00:05 MLI 126 111 88% 3 2% 12 10% 00:00:07 NL 112 93 83% 5 4% 14 13% 00:00:05 MLI 671 610 91% 20 3% 41 6% 00:00:05 NL 646 488 76% 26 4% 132 20% 00:00:04 * Displayed as hh:mm:ss 1. Vindproduct.nl 2. Biano.nl 3. Worksprout.com Total

(27)

27 144 participants that did accept the privacy statement there was an even distribution across the two experimental conditions. Of the participants that also filled out some demographic information (65%), 34% was male and 66% female, the average age was 31 years old and the majority of the sample was holding a University degree (70%).

4.2.1. Descriptives

As a first step in the analysis of the second field experiment, a frequencies check was done. The cases that had missing data for the 5 statements about the privacy statement were deleted listwise, except for the participants that rejected the privacy statement. In the NL format, it took participants 3 seconds on average to get to the bottom of the page where they could find the question ‘Do you accept the privacy statement?’. For the MLI format this question was on the same page. Therefore, to not inflate the results, 3 seconds were subtracted from the time

on page cases of the NL format. To detect outliers, the z-scores of all variables were

standardized. For time on page there were z-scores that were bigger than |3|. These outliers were removed from the dataset.

As explained in the research design section, items for clarity and brevity were used to express the effectiveness of the statement (Goel & Chengalur-Smith, 2010). Since the level of understanding is used as a variable in this research, the three items for clarity (statements 1,2 and 3) were combined into one variable: level of understanding. A reliability analysis was conducted for the variable, Cronbach’s alpha for level of understanding α=.872, furthermore all 3 items have a corrected item-total correlation well above 0.3 and the Cronbach’s Alpha if item deleted is Δ<.10 for all 5 items. Therefore, the scale is considered reliable.

To check for normality, I computed the z scores for skewness and kurtosis of the variables. For the variables of level of understanding, length (earlier mentioned as brevity, statement 4 of the research) and personal interest the z-scores for skewness and kurtosis were all

(28)

28 significant at p<.001. The variable time on page had non-significant z-scores for skewness and kurtosis since the variable is positively skewed and leptokurtic. To test for homogeneity of variance, Levene’s test was carried out. For the level of understanding, the variances were equal for the NL and MLI format, F(1,102) = 3.35, ns, for personal interest the variances were equal as well, F(1,102) = 1.78, ns. For the variables length, F(1,102) = 4.36, p < .05 and

time on page, F(1,102) = 8.19, p < .01 Levene’s test was significant and therefore equal

variances could not be assumed. For these variables, the variances ratio was calculated that were 0.94 and 2.87 respectively. The ratio of length was smaller than the critical value 1.67 with p < .01, therefore homogeneity of variance can be assumed for this variable. For time on

page homogeneity could not be assumed.

Since the variable time on page was not normally distributed and homogeneity of variances could not be assumed, I could decide to transform the data to correct this. However, since the tests that will be carried out are considered robust and due to the possible drawbacks of transformation such as changing the hypothesis to be tested and negatives consequences of the wrong transformation, I decided to not transform the data (Field, 2013; Games, 1984).

After computing the means and standard deviations for the variables, I calculated the correlation coefficient for the different variables to indicate the relationships between them. For this analysis, the non-parametric Spearman’s rho was chosen because the variable time on

page was not normally distributed. There was a significant relationship between the length of

the privacy statement and the level of understanding, r = -.39, p <.01. The negative relationship shows that the longer the privacy statement gets, the lower the level of understanding.

(29)

29 The fourth hypothesis of this research: H4. The website where personal interest is higher

leads to a lower reading time of the privacy statement when compared with the other websites should be rejected because from the correlation matrix it can be concluded that personal interest and time on page are not correlated.

4.2.2. Checking for differences between the formats

Since the focus of this research is on the difference in the level of understanding, time on

page and perceived length between the two formats, the means for the different variables

were compared with the independent samples t-test.

The independent t-test assumes a normally distributed sampling distribution, homogeneity of variance, data measured at least at the interval level and independent scores. Like indicated before, the first two assumptions are violated due to the variable time on page. Since the t-test is robust and gives a separate score for when homogeneity of variance cannot be assumed, I expect the results to be reliable. However, to make sure the results are reliable, a non-parametric test was performed to check the results of the t-test.

To test for the first hypothesis H1. Consumers take more time to read the Multi-layered

format with icons than the Natural language format a t-test was performed. On average, the time on page for the NL format was higher (M =32.62, SE = 5.54) than for the MLI format

Table 2: Means, Standard Deviations and Correlations

Variables M SD 1 2 3 4

1. Level of Understanding 3.44 0.90 (.87)

2. Length 3.44 1.19 -.39**

-3. Personal Interest 2.73 1.03 .01 -.07

-4. Time on Page 30.01 39.54 .09 -.02 .06

-**Correlation is significant at the 0.001 level.

(30)

30 (M =27.36, SE = 2.90). However, this difference was not significant t(119.26) = .841, p > .05. It also represented only a small effect r = .08. Therefore, H1 has to be rejected.

To test for H2. Consumers perceive the Multi-layered format with icons to be less long than

the Natural language format and H3. The level of understanding for the Multi-layered format with icons is higher than for the Natural language format, independent t-tests were carried

out as well. Participants perceived the NL format to be longer on average (M = 4.22, SE = .11) than the MLI format (M = 2.67, SE = .11). This difference was significant t(142) = 10.31, p<.01 and the effect was strong r = .65. For the level of understanding, participants perceived the NL format on average to be less understandable (M = 3.12, SE = .11) than the MLI format (M = 3.75, SE = .09), this difference was significant t(142) = -4.51, p<.01, the effect was medium sized r = .36. Since some of the assumptions of the t-test were violated, a non-parametric test was performed to check the outcomes of the t-test. The Mann-Whitney test verified the findings of the t-tests conducted.

The last hypothesis of the research, H5. The Natural language format is equally often

accepted as the Multi-layered format with icons was also tested using a t-test, where the value

of accept was 1 if the answer was ‘Yes’ and 2 when the answer was ‘No’. As expected, no significant difference was found between the accept rate of the NL format (M =1.10, SE = .03) when compared with the MLI format (M =1.09, SE = .03), t(157) = .24, p > .05. From this result, Hypothesis 5 can be accepted.

4.2.3. Checking for Type I errors

Since carrying out multiple t-tests can increase the probability of making a Type I error, I wanted to conduct another test as well. Since the conceptual model consists of one independent variable (format of the statement), for which I want to compare groups for several dependent variables (level of understanding, perceived length and time on page) a

(31)

31 MANOVA should be used. However, since the correlation matrix shows that time on page does not correlate with the other two dependent variables and since the t-test has already verified that there is no difference between the group means, the MANOVA is executed for

level of understanding and length only.

A MANOVA analysis assumes independence of observations, random sampling, multivariate normality and homogeneity of covariance matrices. Multivariate normality cannot be checked with SPSS, but instead it was checked that all dependent variables had a univariate normally distribution which is the case for level of understanding and length. Homogeneity of variance can be assumed for both dependent variables and a non-significant Box’s test showed that homogeneity of covariances matrices can also be assumed F(3,3629520) = 5.52, ns.

The multivariate tests show significant results for all test statistics (Pillai’s Trace, Wilks’Lambda, Hotelling’s Trace and Roy’s Largest Root). Since group samples are equal, I chose Pillai’s Trace for reporting, showing a significant effect of the privacy statement format on the level of understanding and length, V = .44, F(2, 141) = 54.84, p < .01. Separate univariate ANOVAs show that the groups significantly differ on level of understanding F(1, 142) = 20.33, p < .01 and on length as well F(1, 142) = 106.23, p < .01. The size of the differences was already found in the t-test. The SSCP Matrix showed a negative correlation between length and level of understanding of r = -.24. It was argued that no assumptions were violated in this test, however, when Levene’s test was carried out in the MANOVA, the statistic showed a significant test while the Levene’s test conduct before was not. I expected this would not impact the results since the Box’s test was not significant. However, to be sure this did not affect the results, a non-parametric version of MANOVA was conducted with the use of the R software. The computations made in Rstudio replicated the findings of the

(32)

32 parametric MANOVA so it can be concluded that the results were not influenced. From the t-test and MANOVA results hypotheses 2 and 3 can be accepted.

(33)

33 5. Discussion

In this section I will discuss the results after which I will use the results to explain the academic contributions of the research. Then I will discuss the limitations of the research and finally the possibilities for future research.

5.1. Discussion of the results

This research examined the effect of two different formats of privacy statements on the level of reading and understanding of the statement by consumers and thereby their willingness to share data. The results suggest that a new format of the privacy statement presented and tested in the research has a positive effect on the level of understanding and perceived length of the privacy statement and furthermore a positive effect on the willingness to share data. The findings from the first field experiment show that consumers accept the privacy statement more when they were presented with the new MLI format. When combined with the findings of field experiment 2, that showed that the level of understanding of the privacy statement was significantly higher and perceived length significantly lower for the MLI format, this builds a strong case for companies to change the format of their privacy statement.

However, the two field experiments did not complement each other on all studied constructs. In the theoretical foundations section, the importance of consumers reading the privacy statement was discussed. Several authors showed the importance of a format that is perceived as comprehensible and that can be read within a relatively short period of time. If these factors are met, consumers should be more likely to read the statement. This is important for them to feel informed and to decrease the costs in the privacy calculus (Jensen & Potts, 2004; Milne & Culnan, 2004). Although it could not be checked whether consumers actually read the statement, it could be checked how much time they spend on the page where the

(34)

34 statement was displayed. For field experiment 1, the time on page was on average only 5 seconds for the MLI format and only 4 seconds for the NL format. From these numbers, it can be argued that most consumers did not read the privacy statement for either one of the formats. Although I expected the average reading time to be longer for the MLI format when compared with the NL format (H1), this was not the case in field experiment 2. It was even the other way around, the mean of the MLI format was 27 seconds and for the NL format this was 33 seconds. Although this difference was not found to be significant and the results of field experiment 1 could not be statistically verified, the difference between the two experiments is remarkable. In field experiment 1 consumers clearly did not read the privacy statement whereas in field experiment 2 it was more likely that consumers did read the statement. However, in the results section it was also shown that time on page was positively skewed, meaning that most scores for time on page were low in the number of seconds spent on the page. In total, 70% of the sample had a time on page that was lower than the overall mean of 30 seconds. From this it can be argued that the majority of the participants had scores for time on page that were low and therefore it could be argued that it would not be possible that they have read the statement. Furthermore, as discussed in the results section, it took participants more time to scroll down the page for the NL format to get to the question ‘Do you accept the privacy statement?’. This could have influenced the scores for time on page for the NL format as well, although I already corrected the NL format scores for some differences.

Another difference that was found between the two experiments was the difference in accept rate. I expected that the accept rate of the NL format would be equal to the MLI format (H5), however in field experiment 1, the accept rate was higher for the MLI format on all three websites. Whereas in field experiment 2, the accept rates between the two formats did not show a significant difference. This could be due to the privacy paradox that has been

(35)

35 discussed before (Norberg et al., 2007; Taddicken, 2014). The privacy paradox influencing the behavior of participants could also explain the difference in time on page between the two experiments. If this is the case then the results from field experiment 1 should be more reliable since this experiment measured the real behavior of participants better than the second field experiment where participants could act differently because they knew they were filling in a survey that would be analyzed. These findings show the importance of taking into account the privacy paradox in future research as it can influence results.

It was not only tested whether participants (had enough time to) read the statement, but as well what their level of understanding of the statement was and how they perceived the length. As stated earlier, these factors are important for consumers to feel informed and thereby decreasing the costs in the privacy calculus by lowering feelings of risk and privacy concerns (Xu et al., 2009) and increasing levels of trust (Milne & Culnan, 2004; Schoenbachler & Gordon, 2002). Since there were no participants that filled in the survey in field experiment 1, the level of understanding and perceived length was only researched in the second field experiment. The results show that participants perceive the MLI format as being significantly shorter and more understandable than the NL format (H2 and H3). Furthermore, the effect for perceived length was strong (r = .65, p<.01) and the effect for level of understanding medium (r = .36, p<.01). These constructs are therefore important in explaining the difference in perception of the MLI format and the NL format. After checking for Type I errors, a medium sized negative correlation effect (r = -.24, p<.01) was found between the two variables. Meaning that the longer the participant perceives the statement, the less understandable the statement gets. This is in line with the research by Pan & Zinkhan (2006) who found that consumers perceived short and straightforward privacy statements as easier to understand. This finding is a strong argument for companies to consider the layered version with icons of the statement.

(36)

36 This research was built on the revised privacy calculus model by Dinev and Hart (2006), in this model not only the costs of sharing personal data are discussed but the benefits as well. In the model, the benefits were represented by the level of personal interest and trust. I have tried to make predictions on the effects of the level of personal interest in this research, but unfortunately no significant results were found (H4). In field experiment 1, the website that was thought of having the highest personal interest was found to have a similar time on page as the other two websites. This finding is in contradiction with hypothesis 4 in which I expected that a website with a higher level of personal interest would lead to a lower time on page, which could also be explained as participants accepting the statement without reading due to their high personal interest. It was however found that the accept rate of the website with higher personal interest was lower compared to the other two websites, which is unexpected since personal interest had a positive effect on the willingness to share data in the privacy calculus (Dinev & Hart, 2006). This could be due to the different type of personal information that is collected by Worksprout.com. Since this website collects data that could be perceived as more sensitive (job-related data), consumers could be less willing to accept the statement. In the research by Malhotra et al. (2004) it was indeed found that the request for more sensitive information lead to a negative effect on the willingness to share data which could explain the lower accept rate for Worksprout.com. In field experiment 2, no significant relationship was found between personal interest and time on page either. This could be due to the difficulty of measuring personal interest in the second experiment since only one website was used so no comparison was possible. In this experiment, personal interest was tested with how likely participants were to use the website Biano.nl in their daily life. The answer to this question also depends on other factors such as how they rate the website for example. This could have influenced the results. The internal validity of this variable could therefore be improved in future research.

(37)

37 Although it could not be confirmed that the MLI format lead to consumers reading the statement (H1), the results of field experiment 2 show that the MLI format is indeed perceived as shorter (H2) and more understandable (H3). It is argued that due to these factors, the costs in the privacy calculus are decreased (Bansal et al., 2016; Hui et al., 2007; Milne & Culnan, 2002; Tsai et al., 2011; Xu et al., 2009) and trust is increased (Culnan & Armstrong, 1999; Milne & Culnan, 2004; Pan & Zinkhan, 2006; Wu et al., 2012). The benefits outweighing the costs for the MLI format lead to a higher willingness to share data as was found in field experiment 1. This finding is in line with the privacy calculus as described by Dinev and Hart (2006). Next to these positive outcomes, the MLI format also helps companies to comply with the GDPR making it even more beneficial for companies to use the MLI format instead of the NL format.

5.2. Limitations

Although the results of the research were positive, there have been some limitations to the research. The first limitation was that the findings of the A/B tests in field experiment 1 could not be statistically analyzed with SPSS due to the basic version of Google Analytics. Replication of this experiment would be valuable since it could increase generalizability and internal validity by reporting results with statistical significance. A second limitation of the study is the sample in the second field experiment that was not representative of the population. Of the participants that filled out their demographic information, 34% was male and 66% female, the average age was 31 years old and the majority of the sample was holding a University degree (70%). This could have influenced the results. In a longitudinal assessment of privacy statement readability it was for example found that 53.8% of all the evaluated statements were written above high school reading level (Milne et al., 2006). This study was conducted in the United States but if a similar study in The Netherlands would find a similar result, this would be a problem since only 29% of the Dutch population holds a

Referenties

GERELATEERDE DOCUMENTEN

(a) Brian: All Entities (b) Brian: Wiki Entities (c) Brian: Non-Wiki Entities Figure 3: Disambiguation results over different top k frequent terms added from targeted tweets.. the

After grafting 5 wt% and 39 wt% of CNT/CNF on a carbon fabric it was found that: (1) compressibility of the fabric is seriously decreased even with 5 wt% grafting;

At least we expect a moderating effect of AC characteristics on the relationship between auditor gender and the readability of KAMs due to the possible impact AC can have

First, the relationship between the length of recessions and the change in average government expenditures per GDP during a recession (compared to a five-year average benchmark

When the user has a high perceived privacy risk, the users has the perception that everything that is shared with the fitness tracker can potentially be violated by those

In the back matter section of this dictionary there is a complex outer text, indicated by the table of contents in the front matter functioning as a primary outer text as Wortfelder

In addition, in the first part of the questionnaire, respondents were asked to provide the name of a specific retailer they had a personal omni-channel experience with (using both an

[r]