PRIVACY PRACTICE DONE RIGHT!
INCREASING THE WILLINGNESS TO DISCLOSE PERSONAL INFORMATION FROM A SOCIAL MEDIA PERSPECTIVE
BY
PRIVACY PRACTICE DONE RIGHT!
INCREASING THE WILLINGNESS TO DISCLOSE PERSONAL INFORMATION FROM A SOCIAL MEDIA PERSPECTIVE
University of Groningen
Faculty of Economics and Business
ABSTRACT
Consumers are becoming more and more reluctant to disclose personal information on the Internet. Awareness about data usage is increasing and consumers are less inclined to sign-up for online services like social media sites. Multiple studies show growing concerns regarding online privacy, causing consumers to become more protective of their personal information. As a consequence, social media sites are forced to take action if they want their consumers to willingly disclose personal information when they sign up.
In this thesis a conjoint analysis was applied to examine how privacy policies (i.e. privacy practices) influence the decision of consumers to disclose personal information to a social media network site. The results indicate that consumers generally prefer to disclose their personal information to social media site when; as little personal information as possible is requested, data is stored semi-aggregated, personal information is not used by the social media site for any other purpose (e.g. advertising and service), the user is provided with full transparency, and users have full control over the before mentioned four practices.
Subsequently, a moderator analysis indicates that when consumers score high on privacy concerns they have an increased preference for; a lower level of personal information requested by the social media site, fully aggregated data storage, and no usage of the personal information disclosed. On the other hand previous experience with social media negatively affects the effect of control and the effect of information usage on the willingness to disclose information. No conclusion could be drawn on the moderating effect of control.
PREFACE
TABLE OF CONTENT
1. INTRODUCTION 7
2. THEORETICAL FRAMEWORK 10
2.1 Consumers (Informational) Privacy 10
2.2 Privacy Policy 10
2.3 Conceptualization of Companies Privacy Practises 10
2.4 Privacy Practices 12 2.4.1 Information Collection 12 2.4.2 Information Storage 13 2.4.3 Information Usage 14 2.4.4 Transparency 15 2.4.5 Control 16 2.5 Privacy Concerns 17
2.6 Social Media Experience 18
2.7 Control as Moderator 19 2.8 Sub-groups 20 3. METHODOLOGY 21 3.1 Method 21 3.2 Procedure 21 3.2.1 Conjoint Design 21
3.2.2 Measurement of Social Media Experience 22
3.2.3 Measurement of Privacy Concerns 23
3.2.4 Measurement of Control (Moderation) 23
3.3 Moderation 23
4. RESULTS 24
4.1 Characteristics of the sample 24
4.2 Descriptive Statistics 24
4.3 Dimension Reduction 25
4.4 Main Effects Model 26
4.4.1 Model Comparison 26
4.4.2 Conjoint Analyses Main Effect Model 26
4.4.3 Utility Levels Main Effect Model 28
4.4.4 Attribute Importance Main Effect Model 29
4.5 Moderating Effects 29
4.5.1 Conjoint Analyses Moderating Effects 29
4.5.2 Model 1: Main Effects + Privacy Concerns 31
4.5.3 Model 2: Main Effects + Social Media Experience 32
4.5.4 Model 3: Main Effects + Control 32
4.5.5 Model Comparison 33
4.6 Segmentation - Latent Class 33
5. CONCLUSION AND RECOMMENDATIONS 37
5.1 Conclusion 37
5.2 Managerial Implications 39
5.3 Limitations 39
6. REFERENCES 41
7. APPENDIX 45
Appendix 1: Exemplary Choice Set 45
Appendix 2: Spearman Correlation (demographics) 46
Appendix 3: Factor Analyses Results 46
1. INTRODUCTION
The Internet has made global and real-time communication easier than ever. Social media uses this platform and changes the way the world interacts. Companies and people all over the world benefit from this innovation, but there are serious threats to watch out for. The digitization of our society makes personal information easier accessible for companies. Social media network sites make millions by trading in personal data they obtain from their users. The fact is that the business models of social network sites and mobile social media applications like Facebook and Facebook Messenger rely on this principle. The interests of social media networks in the mining of personal data will make it increasingly harder for users to retain their privacy.
Because consumers have certain rights (i.e. the protection of privacy of personal data) companies have privacy policies. A privacy policy can be seen as a composite of several different privacy practices. These different privacy practises intent to give consumers insight in the data that is being collected, why the collection of data takes place, and wherefore the data is being used. Nevertheless, the rights of social media network sites go much further than consumers might initially think. A social media platform like Facebook provides users with free services, but in exchange has the right to use personal information gained from their users for their own benefit. It could be argued that to get certain services for free, consumers inevitably have to give up other things that they value. Thus, using services requires trading off one privilege for another. However, in the Netherlands there is an apparent change in sentiment among users and consumers.
Investigative journalism by NU.nl indicated that nearly four in ten (38 percent) Dutch- users considers to quite Facebook because they are worried about the protection of their privacy on the social network (Kraan, 2016).
& Kilcoyne, 1993; Goodwin, 1991). However, not much is known on how these privacy practices relate to each other when presented to the consumer in the context of a new social media platform. Companies like Facebook and Snapchat need a better understanding on whether and what combination of privacy practices can put off consumers to the point that it affects their decision making and willingness to disclose personal information (i.e. the willingness to acceptance a companies privacy policy) (Rust & Huang, 2014). Moreover, the willingness to disclose personal information is influenced by a number of internal and external factors. A vast literature research showed that there is a lack of research on how previous experience with social media affects the consumer’s perception of privacy practises and their willingness to disclose information to social media sites. Research in general indicates that when experience increases, the consumers’ individual willingness to disclose information also increases. The effects that the different privacy practises have on the willingness to disclose information is also affected by privacy concerns. Privacy concerns lead to individuals being less willing to share personal information with a company (Bélanger et al., 2002). Furthermore, it is important to investigate if all consumers respond in a similar manner to violations of their privacy. This paper will investigate post hoc if and which parameters of interest differ across subgroups. Ultimately the findings must provide marketers with the knowledge on how to design and target their privacy policy to different consumers in the most effective way, thereby maximizing consumer willingness to disclose information. Thus, the main goal is to gain insight regarding the influence of a social media company’s privacy practice on consumer willingness to disclose information and how the effect of the different privacy practices differs across subgroups based on their previous experience with social media and the amount of privacy concerns they experience. To find answers to this problem the following research questions (RQ) are formulated.
RQ 1: To what extent do the different privacy practises; 1) data collection, 2) data storage, 3) data usage, 4) transparency, and 5) control, (relative to each other) affect the willingness of the consumer to accept a company’s privacy policy?
RQ 3: Is the effect the privacy practices; 1) data collection, 2) data storage, 3) data usage, and 4) transparency have on the willingness to disclose information affected by consumer privacy concerns?
RQ 4: Is the effect the privacy practises; 1) data collection, 2) data storage, and 3) data usage have on consumers willingness to disclose information affected when control is provided (i.e. consumer can indicate which data is collected/stored/used) over these privacy practises?
2. THEORETICAL FRAMEWORK
2.1 Consumers (Informational) Privacy
Research shows that individuals value privacy and have always been concerned about what, and how much, others know about their personal information (Schwartz, 1968). Personal information can be described as all the information that relates to an individual consumer.
Privacy, however, is more difficult to describe since it can be conceptualized and defined in many different ways. This research will focus on consumer (informational) privacy, which in this research is based on research by Westin (1967) and Goodwin (1991) and defined as; ‘’the claim of individuals to determine for themselves, when, how, and to what extent personal information about them is communicated to others.’’ Companies can give consumers the idea they are empowered about concerns regarding their privacy by creating awareness of their activities (henceforth referred to as transparency) and providing consumers with an actual choice pertaining matters of their privacy (henceforth referred to as control) (Caudill & Murphy, 2000; Foxman & Kilcoyne, 1993; Goodwin, 1991).
2.2 Privacy Policy
The privacy practices of companies that are of interest in this research are based on these findings and thus consist of; the collection, the storage, and usage of personal information, transparency, and control (Smith, Dinev & Xu, 2011). Consumers are exposed to privacy policies concerning these practices for all kinds of services and products. Consumers use companies’ privacy practices as a signal that these companies value consumer privacy. The mere presence of a privacy statement increases trust (Aljukhadar, Senecal & Ouellette, 2010), the willingness to disclose information, and even the willingness to purchase (Aljukhadar, Senecal & Ouellette, 2010). However, privacy policies of online social networks only decrease consumer willingness to disclose information (e.g. location data) (Knijnenburg, Kobsa & Jin, 2013). This indicates that more research is needed to understand how consumers respond to different privacy practises of a new social media site when confronted (to agree) with privacy policies.
2.3 Conceptualization of Companies Privacy Practises
practices have an effect on the attitude of consumers and the acceptance of innovation (dependent variable), moderated by consumers’ prior experience with social media (social media experience) and their attitudes and perceptions with regard to privacy (in this research conceptualized as privacy concerns). There may be other variables that affect the relation between the constructs of the conceptual framework. However, other variables that may affect the proposed relationships are not of primary interest.
FIGURE 1 Conceptual model
The willingness to disclose information is determined by consumers’ perception of the different privacy practices. As mentioned previously, some of the privacy practices are expected to have a negative impact on the willingness to share personal information. However, companies can influence the negative impact of these different privacy practises on the willingness to disclose information by changing the content of the privacy practice. Methods for influencing the effect of privacy practises will be discussed in Chapter 2.3. Moreover, the influence of firms’ privacy practices on consumer attitudes or behaviour can differ among consumers. We thus argue, as mentioned earlier, that consumer characteristics like social media usage and privacy concerns can moderate the privacy practice effects on the willingness to disclose personal information. Additionally, based on previous research we try to determine if control positively moderates the effect of the other privacy practices
Privacy Concerns
Social Media Experience
(collection, storage, usage of information, and transparency) on the willingness to disclose personal information.
2.4 Privacy Practices
In this section, the different privacy practices are defined. There are multiple methods that can be implemented by companies to manipulate the effect of privacy practises on consumers’ willingness to disclose information. Because previous research is conducted based on a conjoint analysis, a single method containing different levels is selected. This method is argued to manipulate the effect of the construct on the willingness to disclose information by the consumer. Because the privacy practises have different levels it is hard to determine the relative importance of the privacy practises compared to each other. Even more because the construct control is not compiled out of multiple levels. Control is argued to moderate the effect that the other privacy practices have on the willingness to disclose information. Compiling control out of different levels will make it harder to test for this moderation effect. Investigating control as moderator has the preference because control as a moderator is scientifically more compelling and newer than control as an independent variable (main effect) , which has been researched before. Other interesting insights may be uncovered when combining the findings of main effects with the effect of the moderators and the post-hoc segmentation.
2.4.1 Information Collection
The collection of personal information by companies can be described as the process of gathering and measuring information on targeted variables in an established systematic fashion. Research by Dinev et al. (2013) showed that consumers are indeed affected by which type of information a company collects. Moreover, consumers are less inclined to disclose information when firms demand more information (Hui, Teo & Lee, 2007). Facebook recently experienced this when they received negative media attention for changing its privacy policy and linking users’ Whatsapp account (a free cross-platform messaging application owned by Facebook) to their account on Facebook.
Sensitivity is defined as; ‘’the perceived intimacy level of the information (Premazzi et al., 2010).’’ Research indicates that the perceived risk of disclosing personal information increases as the requested information becomes more sensitive (Schrammel, Köffel & Tscheligi, 2009). Thus, consumers perceive the disclosure/collection of more personal information as entailing greater personal cost and are therefore less willing to disclose this kind of sensitive information (Andrade et al., 2002). This can lead to consumers who refuse to disclose personal information (i.e. won’t agree with a social media sites privacy policy) (Sheehan & Hoy, 1999).
Consumers perceive an increased level of risk as the information collected/requested becomes more sensitive (Mothersbaugh et al., 2012). It can thus be argued that information collection, in general, has a negative effect on consumer willingness to disclose information. By altering the nature of the personal information that is being collected the willingness to disclose can be influenced positively or negatively.
H1a: The effect of personal information collection on the willingness to disclose personal
information is weaker for consumers when a medium level of sensitive personal information is requested compared to a high level of sensitive personal information.
H1b: The effect of personal information collection on the willingness to disclose personal
information is lower for consumers when a low level of sensitive personal information is requested compared to medium level of sensitive personal information.
2.4.2 Information Storage
Moreover, aggregation of requested information is also argued to positively influence consumer willingness to disclose personal information. Data aggregation can be described as;
data collected at individual-level that have been combined for statistical or analytical purposes and that are maintained in a form that does not permit the identification of individuals, thus
providing privacy to the consumer. The levels of aggregation consist of: the personal
information provided by the consumer is stored fully aggregated, the personal information provided by the consumer is stored semi-aggregated, and the personal information provided by the consumer is stored fully disaggregated, with the first level referring to the highest level of privacy protection.
H2a: The effect of data storage on the willingness to disclose personal information is
weaker when data is stored fully aggregated compared to semi-aggregated stored data.
H2b: The effect of data storage on the willingness to disclose personal information is
weaker when data is stored semi-aggregated compared to when data is stored disaggregated.
2.4.3 Information Usage
Consumers are affected by the way their personal information is used. This personal info can be used in a number of different ways. In general, companies use information in order to better understand the needs and preferences of consumers (Wendel & Kannan, 2016). It is important to note that consumers are only aware and affected by the usage of information when companies inform the consumer, or when the usage of information is noticeable to the consumer.
(Re)Targeted banner advertisement (external use of the gained information) is another marketing technique that uses personal information. This type of advertisement is done in a high frequency on social media by targeting consumers with ads, showcasing for example products they Googled before. This type of advertising can be less effective because consumers become aware personal information is collected, stored and used (Bleier & Eisenbess, 2015; Lambrecht & Tucker, 2013). Research does, however, indicate that companies could decrease the intrusive effects of ads by making them more relevant and useful in time and location (Lambrecht & Tucker, 2013; Luo et al., 2014). Personal information is also used for selling (external use of gained personal information). In general, consumers are opposed to the sharing and selling of personal information to third parties, (Alreck & Settle, 2007) because they believe to be more at risk (Jai, Burns & King, 2013). In conclusion, it can be argued that consumers consider internal usage more acceptable than external usage (Schwaig et al., 2013). Information usage levels will be divided in: personal information is not used, personal information is collected for internal usage (personalization), and personal information is collected for external usage (third party), where the first level refers to the highest level of privacy protection.
H3a: The effect of information usage on the willingness to disclose personal information is
weaker when personal information is not used compared to internal usage.
H3b: The effect of information usage on the willingness to disclose personal information is
weaker when data is used for internal use compared to when data is used for external usage.
2.4.4 Transparency
Contradicting research by Sheehan & Hoy (2000) indicated that consumers are more worried about their privacy when they are aware of the collection and usage of personal information. However, they also argue that transparency increases trust which if beneficial in the long term and prevents future discontent about company privacy practices.
Consumers generally feel less vulnerable when companies explain and justify the use of personal information (Aguirre et al., 2015). Thus, increasing transparency will positively affect the consumers’ willingness to disclose personal information because it creates trust and strengthens the relationship between companies and consumers. Transparency will be divided into: Full-transparency, semi-transparency, and no transparency, where the first refers to the highest level of privacy protection.
H4a: The effect of transparency on the willingness to disclose personal information is
weaker when a social media site provides the consumer with semi-transparency compared to full-transparency.
H4b: The effect of transparency on the willingness to disclose personal information is
weaker when a social media site provides the consumer with semi-transparency compared to no transparency.
2.4.5 Control
Control is used to indicate that consumers have the possibility to make their own choices. Control provides the consumer with a sense of autonomy and management over the collection, storage, and usage of personal information. This gives consumers the feeling they’re less at risk, making their choices less consequential. When consumers are in control they are less concerned about their privacy and are more willing to disclose information as a result (Xu et al., 2012.) Control is therefore not about disrupting the collection, storage, and usage of information, but more about having the possibility to change which data is collected, how it is stored, and in what way it is used.
H5: Control will have a positive impact on the willingness to disclose personal info to
a social networking site.
2.5 Privacy Concerns
The Internet offers a number of possibilities to collect and trade private information, especially for social media networks. Social media networks are able to collect and combine personal information. As mentioned earlier the usage of personal information offers conveniences and privileges to users/consumers. In return users of the social media site grant the company personal information; i.e. users give up part of their privacy. Research by Van Aaken et al. (2014) indicates that a lot of consumers are unaware of this trade.
Privacy concerns can be defined as the individual’s subjective views of fairness within the context of information privacy (Malhotra et al., 2004). However, based on research by Smit, Milberg & Burke (1996) we assume that privacy concerns can be defined as ‘’the concerns individuals have with the usage of the information disclosed to organizations’’.
Privacy concerns cause the consumer to worry about the collection, storage, and usage of their personal information. Consumers that are highly concerned about their privacy are more unwilling to disclose sensitive information than consumers that consider their privacy to be less important (Mothersbaugh et al., 2012). Consequently, it can be argued that privacy concerns have a negative impact on the information collection, thereby decreasing consumer willingness to disclose information. Furthermore, privacy violations increase consumer privacy concerns. Consumers can, with respect to the way data is stored, experience privacy violations. Therefore it can be argued that privacy concerns negatively affect the relationship between data storage and the willingness to disclose personal information.
Moreover, consumers that are highly concerned and/or involved with regard to privacy pay more attention to all decisions that affect their privacy. This indicates that transparency has a negative affect on the willingness to disclose information for highly concerned consumers.
H6a: Consumers who are high in privacyconcerns will have an increased preference for a
lower level of personal information requested.
H6b: Consumers with strong privacy concerns will have an increased preference for a
storage method that not permits the identification of their personal data.
H6c: When disclosing personal information, consumers who are high in privacy concerns
will have an increased preference for no information usage over internal, and internal over external, usage.
H6d: Consumers with higher privacy concerns will have an increased preference for
transparency when disclosing personal information.
2.6 Social Media Experience
Whether or not the consumer had previous experience with Facebook is expected to influence the willingness to disclose personal information. There is no other research that investigated if previous experience with social media affects consumers’ perception with regard to their personal privacy. Moreover, no research indicated if experience affects the perception of privacy practises and subsequently influences the willingness to disclose information.
Prior research does indicate that experience increases general privacy concerns as consumers become more aware (of the risks) of the collection, storage, and usage of personal information (Sheehan & Hoy 2000). On the other hand research by Bart et al. (2005) and Milne & Boza (1999) indicates that previous experience with a company affects their trust in the company. Furthermore, online experience decreases consumer privacy concerns (Bellman et al., 2004), as consumers become familiar with the environment and become more aware of how to protect their privacy (Milne & Boza, 1999). Not only experience with a device and channel influences consumer privacy concerns. A prior relationship with a company also positively influences to what extent company privacy practices affect consumers.
disclosing information increases, thereby weakening the negative and strengthening the positive affects the different privacy practises have on the willingness to disclose information. H7a: Social media experience negatively affects (moderates) the effect of information
collection on the willingness to disclose information.
H7b: Social media experience negatively affects (moderates) the effect of information
storage on the willingness to disclose information.
H7c: Social media experience negatively affects (moderates) the effect of information usage
on the willingness to disclose information.
H7d: Social media experience negatively affects (moderates) the effect of transparency on
the willingness to disclose information.
H7e: Social media experience negatively affects (moderates) the effect of control on the
willingness to disclose information.
2.7 Control as Moderator
Previous literature shows that consumers can be more willing to accept the collection, storage, and usage of personal information if they have control over these elements.
Research indicates that when consumers have control with regard to the usage of information, that consumers’ perception that their privacy is violated decreases (Martin et al., 2016). This indicates that enhancing control moderates their willingness to disclose information. Moreover, the impact of privacy breaches decreases when consumers are provided with control (Martin et al., 2016). This suggests that control positively moderates the effect of information storage on consumer willingness to disclose information.
Control over the collection and the way information is used enhances the acceptance of personalized ads (Schumann, Von Wagenheim & Groene, 2014). Furthermore, consumers seem more inclined to click on personalized ads on Facebook when controlling privacy is made easier (Tucker, 2014). This demonstrates that control moderates the effect of information usage on consumer willingness to disclose information.
information. Hence, when consumers are provided with a degree of control over the other privacy practises it negatively affects the individual effect the different privacy practises have on consumer willingness to disclose personal information.
H8a: Control negatively affects (moderates) the effect of information collection on the
willingness to disclose personal information.
H8b: Control negatively affects (moderates) the effect of information storage on the
willingness to disclose personal information.
H8c: Control negatively affects (moderates) the effect of usage of information on the
willingness to disclose personal information.
2.8 Sub-groups
Consumers vary in how they value privacy; some simply worry more than others. These differences influences how consumers view the privacy policy of social media sites and affect their willingness to disclose information. Consumers can thus be segmented on a number of covariates.
Research shows that as age increases people become more concerned about their privacy (Bellman et al., 2004). Gender also impacts their level of privacy concerns. In general, females tend to be more concerned about their privacy. (Bellman et al., 2004). Other research indicates that consumers with low education tend to be more concerned about their privacy as well (Milne & Boza, 1999).
3. METHODOLOGY
3.1 Method
A dual-response choice based conjoint analysis is used to test the hypothesis formulated in the previous chapter. A conjoint analysis is a decomposition method that makes use of an overall evaluation of a set of attributes (the privacy practises) to measure the relative preferences for each attribute and its corresponding levels. Moreover, a dual response no-choice option is used to determine if respondents are truly willing to disclose personal information in the given situation (Eggers & Sattler, 2011).
3.2 Procedure
The survey consisted of three parts. Firstly, respondents were provided with a brief explanation of the experiment and received instructions on the choice experiment. In the second part, respondents were given a series of choice sets. In the third part, respondents were asked five questions about their privacy concerns, four questions about their previous experience with social media and three questions about their demographics.
3.2.1 Conjoint Design
TABLE 1
Conjoint attributes & levels used in the study
Attribute Levels
Information collection • Low level of sensitive personal information requested • Medium level of sensitive personal information requested • High level of sensitive personal information requested Information storage • Fully aggregated
• Semi-aggregated • Disaggregated Information usage • No use
• Internal use • External use Transparency • Full transparency
• Semi-transparency • No transparency
Control • No control
• Full control
The experiment was composed of 10 choice sets, each comprised of 5 attributes, of which 4 (information collection, information storage, information usage, and transparency) contained 3 options and 1 (control) only 2 options. For each 10 choice sets respondents had to choose the combination of attributes they preferred the most. Subsequently, the respondents were asked if they would disclose personal information when a company would use the privacy policy they selected as most preferred. An example choice set is visually represented in Appendix 1.
3.2.2 Measurement of Social Media Experience
This scale is based of seven statements, which attempts to assess consumers’ familiarity and experience with a brand and several other aspects related to the brand. (Martin & Steward, 2001). However, because no previous study investigated the experience of consumers with online social media, not all the statement formulated by Martin & Steward, where used. Some additional question where added to better assess the usage, and thus the amount of experience, consumers had with social media sites.
TABLE 2
Measurement of Experience
Questions based on research Martin & Steward (2001) How familiar are you with social media?
How familiar are you with the privacy practises just mentioned?
How familiar are you with the privacy policy (combination of privacy practises) of the current social media sites you use?
How much experience do you have with social media sites?
Additional questions
3.2.3 Measurement of Privacy Concerns
Privacy concerns are measured based on research by Dinev & Hart (2006). Dinev & Heart combined four items to form an overall score for privacy concerns. The respondents rate their privacy concerns on a 5-point Likert scale. Answers range from not at all concerned to very concerned. Table 2 shows the questions that were used to measure privacy concerns.
TABLE 3
Measurement of Privacy Concerns
In general, I am concerned about my privacy when using the Internet. I am concerned that information I submit on the Internet could be misused. I am concerned that a person can find private information about me on the Internet.
I am concerned about submitting information on the Internet, because it could be used in a way that I cannot foresee.
3.2.4 Measurement of Control (Moderation)
Since it is argued that control, as an attribute part from the conjoint study, moderates the other attributes in the conjoint study no new scale needed to be implemented. Thus, control already has been measured in the conjoint part of the study (see paragraph 3.6).
3.3 Moderation
In order to account for moderation in a choice based conjoint analysis, new interaction variables had to be created (Eggers, 2015). The moderating effect in a choice based conjoint analysis is represented by an interaction effect between the attribute levels and the moderators (control, social media experience and privacy concerns). To estimate the effect of the moderators new variables had to be created by multiplying the factor of the moderators (see Chapter 4.2) with each of the effect-coded attribute levels (given in the data set). A total of 24 new interaction variables were created and added to the dataset.
4. RESULTS
4.1 Characteristics of the sample
A total of 122 respondents took part in the survey. All respondents were subtracted out of the researcher’s circle of acquaintances. A conjoint choice design with five attributes; of which four have three levels, one with two levels and a total of ten choice sets per survey respondent, requires a statistical minimum of 50 respondents in order to the give the data analysis statistically valuable output (Eggers, 2016). The data was checked for missing values and outliers. A few respondents were removed from the sample to ensure data quality. The final sample contained a total of 114 respondents, an adequate number of respondents. The full factorial design based on these attributes contained 162 possible stimuli. 114 valid respondents completed the survey. Due to the nature of the choice based conjoint data, every respondent had 30 lines. This means the sample contained a total of 3420 observations.
4.2 Descriptive Statistics
As can be seen in Table 4 the ratio between males and females in the sample is nicely distributed. However, there is a difference among the respondents in age. The sample has an average age of 28 (median = 24). Moreover, almost all the respondents have a HBO Bachelor degree or higher. This means the sample consists mainly of young, high-educated people (students). Table 4 shows an overview of the sample characteristics.
TABLE 4 Sample characteristics
Percentage Number of respondents Gender • Man 63.20% 72 • Female 36.80% 42 Age • 19-21 6.20% 7 • 22-24 45.70% 52 • 25-27 24.50% 28 • 28-30 7.00% 8 • >30 16.60% 19 Education • Secondary school 2.60% 3 • MBO 3.50% 4 • HBO 27.20% 31 • Bachelor’s degree 26.30% 30 • Master’s degree 40.4-% 46
these socio-demographics. The Spearman correlation was used. Only age and education showed a low positive correlation (Correlation Coefficient = -0.300, p<0.001) (Appendix 2.) This correlation indicates that when age increases education level drops, which can easily be explained because the bigger part of the sample contains highly educated students. To see if this multicollinearity caused any problem(s) when estimating the model, a simple linear regression was performed to find the Variance Inflation Factors (VIF) for all independent variables, including the conjoint analysis. Based on the VIF scores (all below the threshold of 5) no multicollinearity problems were expected.
4.3 Dimension Reduction
In order to measure the moderating effect between social media experience and privacy concerns on the different privacy practices a new variable had to be created. A Principal Component Analyses (PCA) was conducted to find out if the dimension reduction needed to create these new variables was appropriate. PCA creates a principal component, which is a linear combination of the original variables. Moreover a Cronbach’s alpha test was conducted to assess the internal consistency of the scales. Since the measurement of privacy concerns was adopted from prior literature, the multi-item scales had been tested before on the appropriateness of dimension reduction. “Social media experience” had never been tested before on the appropriateness of dimension reduction, because the multi-item scale was not only based on questions that were modeled after an existing scale but also partially consisted of self-fabricated questions. Thus, a PCA was conducted for “privacy concerns” and “social media experience” to make sure the data gained in this study about the moderators was appropriate for data reduction.
The Kaiser-Meyer-Olkin Measure of Sampling Adequacy (KMO)(0.779), the Bartlett’s Test of Sphericity (p<0.001), 878), and the communalities all showed values over the threshold of 0.4, indicating that the questions about privacy concerns factor well. The KMO of 0.682, the Bartlett’s Test of Sphericity (p<0.001) and communalities were all over 0.4 indicating that social media experience was also appropriate for factoring.
4.4 Main Effects Model 4.4.1 Model Comparison
The main model was calculated with all the attributes (privacy practices) at a nominal level. All attributes were assumed to be part worth by default. To determine the goodness of fit, the main model was compared to the NULL model using the Chi-square. The Chi-square compared the log-likelihood of the NULL model (LL(0) = -1252.41) to that of the main model (LL(β*) = -958.2644):
Chi-square = −2(𝐿𝐿(0)−𝐿𝐿(𝛽∗))
This resulted in a Chi-square of 588.29, which exceeded the critical value of 137,208 found in the Chi-square table (df=97, α=1%). It can thus be said that the main model parameters are significantly different from zero and the main model clearly outperforms the NULL model. Furthermore, McFadden’s R² confirmed the finding that the main model outperforms the NULL model with a R² of 0.2446 and an adjusted R² of 0.2441. An overview of the model comparison can be found in Table 5.
TABLE 5
Main effect model compared to null model
LL X2 Npar Degrees of Freedom
R2 Adjusted AIC(LL) Hit rate Null model -1252.41
Main effect model -958.2644 588.29 10 114 0.2441 1934.5287 60,88%
4.4.2 Conjoint Analyses Main Effect Model
Table 6 shows the utility estimates, the Wald values and the p-values of the attribute levels of the main effect model that are given based on estimations with dummy coded variables. The p-values in Table 6 reveal that; information collection, information storage, information usage, transparency, and control have a highly significant effect on the utilities and that the attributes significantly differ from each other.
The results in Table 6 reveal that consumers prefer a low level of requested sensitive personal information the most (β=0.3483), followed by a medium level of sensitive personal information requested (β=-0.0590) and high level of sensitive personal information requested (β=-0,3414). H1a states that consumers would prefer a medium level of sensitive personal
personal information. Based on the utility level and significant p-value it can be assumed that H1a is supported. Moreover, H1b, which states that consumers prefer a low level of sensitive
personal information compared to a medium level of sensitive information, is also supported based on the before mentioned utility levels and p-values. Thus, H1a and H1b are supported.
In order to test H2a and H2b, the difference between the different storage methods had to be
investigated. As can be seen in Table 6, consumers prefer that their personal information is stored fully aggregated (β=0.6970) the most, followed by semi-aggregated (β=-0.0797) and disaggregated (β=-0.6173). The results indicate (p<0.001) that there is a significant difference between the attribute levels. Both H2a and H2b are supported, because the hypothesis
formulated proposed that consumer prefer semi-aggregated over disaggregated (H2a) and
prefer fully aggregated over semi-aggregated (H2b).
H3a and H3b infer about the difference in information usage. The results shown in Table 6
confirm that consumers prefer no usage of personal information (β=0.3968) to internal use (β=0.2306) and internal usage to external use (β=-0.6274). A p<0.001 again indicates that the attributes differ significantly. Hence, H3a and H3b are supported.
The most interesting finding arose when hypothesis H4b was tested. H4b hypothesizes that
‘’the effect of transparency on willingness to disclose personal information is weaker when a social media site provides the consumer with full transparency compared to transparency’’. However, when looking at the utilities it appears that consumers prefer semi-transparency (β=0.0825) to full semi-transparency (β=0.0395). This finding is surprisingly significant p=0.041, although slightly less significant as the other attributes in the main model. Therefore, H4b is not supported. H4a is supported; consumers do prefer
semi-transparency (β=0.0825) to no semi-transparency (β=-0.1220).
Finally, H5 was investigated. H5 states that ‘’control will have a positive impact on the
-0,4 -0,2 0 0,2 0,4 0,6 0,8 Utility
In the survey respondents were asked if, when provided by the fictional social media with the most preferred privacy policy of the three, they would indeed disclose their personal information. In all the estimations the so-called ‘no choice option” variable caused an iteration error. Moreover, the variable was in all the estimations highly insignificant. To make sure the iteration error did not affect the other variables in the estimation, in this and further estimations the variable was disregarded.
TABLE 6
Utility Estimations of the Main Effects only Model
Attributes Utility Wald P-value
Information collection
- Low level of sensitive personal information requested 0.3483 52.9211 <0.001 - Medium level of sensitive personal information requested -0.0590 52.9211
- High level of sensitive personal information requested -0.2893 52.9211 Information storage - Fully aggregated 0.6970 232.3062 <0.001 - Semi-aggregated -0.0797 232.3062 - Disaggregated -0.6173 232.3062 Information usage - No use 0.3968 128.4071 <0.001 - Internal use 0.2306 128.4071 - External use -0.6274 128.4071 Transparency - Full transparency 0.0395 6.3840 0.041 - Semi-transparency 0.0825 6.3840 - No transparency -0.1220 6.3840 Control - Full control 0.4628 148.6952 <0.001 - No control -0.4628
4.4.3 Utility Levels Main Effect Model
Figure 2 displays the utility levels for each category and clearly show that consumers prefer the attribute levels; low level of sensitive personal information requested, fully aggregated data storage, no use of personal information, and full control the most. Moreover, the figure shows how semi-transparency is slightly preferred over full transparency.
0,00% 5,00% 10,00% 15,00% 20,00% 25,00% 30,00% 35,00% Information
Collection Information Storage Information Usage Transparency Control
Rel at ive Im p or ta nce
4.4.4 Attribute Importance Main Effect Model
To find out to what extent each attribute influences the decision of the consumer to disclose personal information, the range within each attribute had to be calculated. The results can be found in Figure 3. The results reveal that the most important attributes in the respondents’ decision to disclose personal information was the amount of information storage (29,83%), followed by information usage (21,93%), control (16,87%), information collection (15,51%) and transparency (15,24%). An important detail; the control attribute has one level less than the other attributes. In a conjoint study, attributes with more levels have a higher importance. Thus, since not all the attributes are equally balanced is it hard to compare their relative importance.
FIGURE 3
Relative Attribute Importance
4.5 Moderating Effects
4.5.1 Conjoint Analyses Moderating Effects
Table 7 shows the estimated utilities and the significance levels of the parameters. To assess the stability of the results, three models were produced. In each model one of the three moderators was added to the main model. It is interesting to notice that the utility levels of the attributes are similar between the three models, also compared to the utilities in the main effects model.
TABLE 7 Overview Utilities
Attributes Model 1 Model 2 Model 3
Information collection
- Low level of sensitive personal information requested 0.3482* 0.3476* 0.3583* - Medium level of sensitive personal information requested -0.0555* -0.0579* -0.0465* - High level of sensitive personal information requested -0.2927* -0.2897* -0.3118*
Information storage - Disaggregated 0.6993* 0.7092* 0.7011* - Semi-aggregated -0,0751* -0.073* -0.0861* - Fully aggregated -0.6242* -0,6362* -0.6151* Information usage - No use 0.3894* 0.411* 0,3989* - Internal use 0.2374* 0.2269* 0,2385* - External use -0.6268* -0,6379* -0.6375* Transparency - Full transparency 0.0463** 0.0414*** 0.0387** - Semi-transparency 0.0748** 0.0737*** 0.0832** - No transparency -0.1212** -0.115*** -0.1219** Control - Full control 0.4614* 0.4771* 0.4818* - No control -0.4614* -0.4771* -0.4818*
Privacy Concerns*Lower level of sensitive info requested 0.1298** Privacy Concerns*Medium level of sensitive info requested -0.1299** Privacy Concerns*High level of sensitive info requested 0.0001
Privacy Concerns*Fully aggregated 0.1057**
Privacy Concerns*Semi-aggregated -0.0491
Privacy Concerns*Disaggregated -0.0566
Privacy Concerns*No use 0.1326**
Privacy Concerns*Internal use -0.0329
Privacy Concerns*External use -0.0997
Privacy Concerns*Full transparency -0.0033
Privacy Concerns*Semi-transparency 0.0801***
Privacy Concerns*No transparency -0.0768
SM Experience*Lower level of sensitive info requested 0.0307
SM Experience*Medium level of sensitive info requested -0.0338
SM Experience*High level of sensitive info requested 0.0003
SM Experience*Fully aggregated -0.0274 SM Experience*Semi-aggregated -0.0083 SM Experience*Disaggregated 0.0357 SM Experience*No use -0.1369** SM Experience*Internal use 0.0633 SM Experience*External use 0.0736 SM Experience*Full transparency 0.0393 SM Experience*Semi-transparency 0.0237 SM Experience*No transparency -0.063 SM Experience*Full control -0.1176** SM Experience*No control 0.1176
Control*Lower level of sensitive info requested -0.0607
Control* Medium level of sensitive info requested -0.0622
Control*High level of sensitive info requested 0.1229
Control*Fully aggregated -0.0639 Control*Semi-aggregated 0.0597 Control*Disaggregated 0.0042 Control*No use -0.0097 Control*Internal use -0.0437 Control*External use 0.0534
*Significant at a 1% level, ** Significant at a 5% level, *** Significant at a 10% level Model 1: Main effects
Model 2: Main effects + Privacy Concerns
4.5.2 Model 1: Main Effects + Privacy Concerns
The main effect of information collection is significant (p<0.001), meaning that asking for less information increases consumers’ utility. Moreover, consumer privacy concerns significantly moderate the effect of information collection on utility. Specifically, the more a consumer is concerned about their privacy, the more a consumer prefers giving a low level of personal information when disclosing information to a social media site. The effect of a lower level of sensitive personal information requested increases by β=0.1298 for every unit of privacy concerns (p=0.0075). There is also a significant moderating effect of privacy concerns on a medium level of sensitive info requested which becomes more negative (β=-0.1299) than the main effect the more a consumer is susceptible to privacy concerns (p=0.0072). H6a, which
infers that when consumers experience privacy concerns they prefer a lower level of personal information requested, is supported by these results.
The main effect of information storage is significant (p<0.001). Providing a more secure way of data storage increase consumers’ utility. Moreover, consumer privacy concerns significantly moderate the effect of data storage when data is stored fully aggregated. The effect of storing sensitive personal information fully aggregated increases the utility by β=0.1057 for every unit of privacy concerns (p=0.025). The moderating effect of privacy concerns on the other levels of the attribute data storage is not significant. H6b, which argues
that consumers with strong privacy concerns will have an increased preference for a storage method that prohibits identification of their personal records, is only partially supported since for the other attributes no significance is found.
The same holds true for information usage. The main effect of information usage is significant (p<0.001). Moreover, consumer privacy concerns significantly moderate the effect of information usage when no personal information is being used. The effect of using no personal informal increases the utility by β=0.1326 for every unit of privacy concerns (p=0.006). The moderating affects of privacy concerns on the other levels of the information usage were not significant. H6c states that when disclosing personal information, consumers
who are high in privacy concerns will have an increased preference for no usage over internal or external usage of their personal data. H6c is thus only partially supported. People who
The main effect of transparency is significant (p=0.049). Providing more transparency about the privacy policies will increase consumers’ utility. Moreover, consumer privacy concerns significantly moderate the effect of transparency when data is stored semi-transparent. The effect of providing semi-transparency increases the utility by β=0.0801 for every unit of privacy concerns (p=0.1). The moderating effects of privacy concerns on the other levels of the attribute were not significant. H6d, which argues that consumers with higher privacy
concerns will have an increased preference for transparency when disclosing personal information, cannot be supported. The more preferred level of transparency, full transparency, is not significant. The preferred option is semi-transparency over transparency.
4.5.3 Model 2: Main Effects + Social Media Experience
As mentioned before, the main effect for information usage is significant (p<0.001). Furthermore, social media experience also significantly moderates the effect of information usage. The effect of not using the information decreases the utility by β=-0.1369 for every unit of social media experience (p=0.0061). The moderating effects of social media experience on the other levels of the attribute are not significant. H7c, which argues that social
media experience negatively moderates the effect of information usage on the willingness to disclose information, can be partially supported. Only one level of the attribute is significant. The main effect of control is significant (p<0.001). Providing control increases consumers’ utility. Moreover, a consumer’s experience with social media significantly moderates the effect of control on utility. Specifically, when a consumer is more experienced with social media the effect of control decreases by β=-0.1176 for every unit of social media experience (p=0.0044). H7e, which states that social media experience negatively affects (moderates) the
effect of control on the willingness to disclose information, is supported.
No significant moderation between social media experience and information collection, social media experience and information storage, and social media experience and transparency were found. Therefore H7a, H7b, andH7d cannot be supported.
4.5.4 Model 3: Main Effects + Control
4.5.5 Model Comparison
Findings show that when comparing the moderation/interaction models based on their utility levels for the different attributes and the attribute importance (shown in Appendix 4 and Appendix 5) they appear to be very similar. In other words the moderators do not cause consumers to act different to the privacy practices of a social media site. The most appropriate model to compare to the main affects the model is the main effects model including only the significant interaction variables from the moderators; privacy concerns, social media experience, and control. This model, as indicated in Table 8, has the best R2, the lowest log likelihood, the lowest AIC, and the lowest BIC. Thus, all findings indicate that main effects + all significant interaction/ moderation model has the best fit and explains the most of the variance.
The goodness of fit of the model, with only the significant moderation, was compared to the main model using the Chi-square test to see if the model was also significantly better as the main effects model. A Chi-square of 33.20 not exceeded the critical value of 137,208 found in the Chi-square table (df=98, α=1%). Therefore, the main model with only the significant interaction (moderations) does not significantly differ from the main model. An overview of the model comparisons can be found in Table 8.
TABLE 8
Models with interaction (moderation) compared to the Main effect model
LL Npar R2 Adjusted AIC(LL) BIC(LL)
Main Effects + Privacy Concerns -947.8607 17 0.2524 1929.7215 1976.2369 Main Effects + Social Media Experience -949.2027 18 0.252 1934.4055 1983.6570 Main Effects + Control -956.0169 15 0.2466 1942.0338 1983.0768 Main effects + all
significant interaction/ moderation
-941.6642 16 0.2571 1915.3285 1959.1076
Main effect model -958.2644 10 0.2441 1934.5287 1959.1545
4.6 Segmentation - Latent Class
The main effects model, which is determined to be the best model, is estimated for one to four classes. To determine the most appropriate amount of classes, the model with the lowest Bayesian Information Criteria (BIC) was selected as the preferred model and was used for further analyses. Table 9 shows us that the two-class model had the lowest BIC value.
TABLE 9
Comparison Latent class analysis
LL BIC(LL) AIC(LL) Npar L² df p-value Class.Err. 1-Class Choice -958.2644 1959.1545 1934.5288 9 1916.5288 105 <0.001 0
2-Class Choice -873.0065 1959.142 1836.0131 45 1746.0131 69 <0.001 0.0218 3-Class Choice -794.6808 1972.9936 1751.3615 81 1589.3615 33 <0.001 0.0226 4-Class Choice -740.1778 2034.4908 1714.3555 117 1480.3555 -3 . 0.0395
Looking at the measurement concerning the models’ goodness of fit, in the one-class model the R² and R²(0) improved from respectively 0.2441 and 0.2446 for the one-class model to 0.3486 and 0.3492 for the two-class model. This indicates that the two-class model explains more of the variance. This is supported by the Chi-square statistic (chisq = 170.52.8332). The Chi-square statistic exceeded the critical value of 112,317 found in the Chi-square table (df=69, α=1%) indicating that the fit of the model improved significantly. The two-class model parameters are significantly different from zero; the two-class model clearly outperforms the main model.
TABLE 10 Two-Class Model
Attributes Class 1 Wald P-value Class 2 Wald P-value Information collection
-Low level of sensitive personal information requested 0.4989 88.2491 <0.001 0.0816 24.3228 <0.001 - Medium level of sensitive personal information requested -0.0352 -0.2003
By looking at the relative importance of the attributes we can determine what typifies the classes. Class 1 had a higher importance for information collection (22.53%), information storage (36.05%) and information usage (28.47%), thus looking for the main indicators of a privacy policy. However, for class 2, control was by far the most important attribute (52.55%). The importance of the attributes for each of the two classes can be found in Table 11.
TABLE 11
Importance of attributes (two-class model)
Attributes Class 1 Class 2
Information Collection 22.53% 7.13%
Information Storage 36.05% 20.96%
Information Usage 28.47% 14.01%
Transparency 3.39% 5.34%
Control 9.56% 52.55%
The descriptive statistics; gender, age and education allows to further define the two-class model. However when looking at Table 12, no big differences were found. Both class 1 and class 2 consisted of more men than women. A larger difference became visible when looking at the age distribution between the two models. Class 1 consisted mainly of 22 to 24 year olds, while class 2 consisted for a large part of people aged 25 and older. Average age difference between the two classes was 5 years. When looking at education level it again became clear that the sample consisted mainly out of higher educated respondents. However class 2 seems a little more representative since 13.48% of the sample had a MBO degree while in class 1, no respondents held this degree.
TABLE 12
Descriptive Statistics (two-class model)
TABLE 13 Summary Results
Hypothesis Conclusion
H1a The effect of personal information collection on the willingness to disclose personal information is weaker for consumers when a medium level of sensitive personal information is requested compared to a high level of sensitive personal information.
Supported
H1b The effect of personal information collection on the willingness to disclose personal information is lower for consumers when a low level of sensitive personal information is requested compared to medium level of sensitive personal information.
Supported
H2a The effect of data storage on the willingness to disclose personal information is weaker when data is stored fully aggregated compared to semi-aggregated stored data
Supported
H2b The effect of data storage on the willingness to disclose personal information is weaker when data is stored semi-aggregated compared to when data is stored disaggregated.
Supported
H3a The effect of information usage on the willingness to disclose personal information is weaker when personal information is not used compared to internal usage.
Supported
H3b The effect of information usage on the willingness to disclose personal information is weaker when data is used for internal use compared to when data is used for external usage.
Supported
H4a The effect of transparency on the willingness to disclose personal information
is weaker when a social media site provides the consumer with semi-transparency compared to full-semi-transparency.
Supported
H4b The effect of transparency on the willingness to disclose personal information
is weaker when a social media site provides the consumer with semi-transparency compared to no semi-transparency.
Not Supported
H5 Control will have a positive impact on the willingness to disclose personal
info to a social networking site.
Supported
H6a Consumers who are high in privacy concerns will have an increased preference for a lower level of personal information requested.
Supported
H6b Consumers with strong privacy concerns will have an increased preference for a storage method that not permits the identification of their personal data.
Partially supported
H6c When disclosing personal information, consumers who are high in privacy concerns will have an increased preference for no information usage over internal, and internal over external, usage.
Partially supported
H6d Consumers who higher privacy concerns will have an increased preference for transparency when disclosing personal information.
Not Supported
H7a Social media experience negatively affects (moderates) the effect of
information collection on the willingness to disclose information.
Not Supported
H7b Social media experience negatively affects (moderates) the effect of
information storage on the willingness to disclose information.
Not Supported
H7c Social media experience negatively affects (moderates) the effect information
usage on the willingness to disclose information.
Partially supported
H7d Social media experience negatively affects (moderates) the effect of
transparency on the willingness to disclose information.
Not Supported
H7e Social media experience negatively affects (moderates) the effect of control
on the willingness to disclose information.
Supported
H8a Control negatively affects (moderates) the effect of information collection on
the willingness to disclose personal information.
Not Supported
H8b Control negatively affects (moderates) the effect of information storage on the
willingness to disclose personal information.
Not Supported
H8c Control negatively affects (moderates) the effect of usage of information on
the willingness to disclose personal information.
5. CONCLUSION AND RECOMMENDATIONS
5.1 Conclusion
The increasingly rapid digitization of our society in the last few decades has made it harder to retain our privacy. This study tried to shed some light on both this phenomenon and the reaction of consumers to this problem. To make the study more specific and measurable the example of a privacy policy of a social network site was used. The research goal was to find out how consumers react to different fictive privacy policies of social network sites when asked to provide their personal information. The survey results give an overview of how different consumers react to specific privacy policies and if privacy concerns, previous experience with social media, and provided control affect this response.
Based on these finding the most obvious conclusion that can be drawn from the different hypotheses testing the main effects (H1a, H1b, H2a, H2b, H3a, H3b, H4a and H5) is that by altering
privacy practises it is possible to increase consumer willingness to disclose personal information to a social media site. The most interesting finding presented itself when testing for hypothesis H4b, since it appears that consumers prefer semi-transparency to full
transparency. It may be that too much transparency is perceived as to confronting. Consumers in a fully transparent environment are fully aware and confronted with the knowledge of how and what is done with their personal information. It can be hypothesized that consumers prefer to be kept in the dark about these confronting, possibly intimidating, personal matters.
usage, and internal to external usage. We found prove that consumers high on privacy concerns significantly prefer that the social media site does not use their personal information. However, no prove was found that consumers high on privacy concerns prefer internal use to external usage. This adds to existing research by Culhan (1993) and Belanger et al. (2002). The assumption that consumers who are higher in privacy concerns will have an increased preference for transparency when disclosing personal information could not be proven.
There can also be made some interference about the moderator social media experience. Results indicate that the hypothesis that social media experience negatively affects the effect of control on the willingness to disclose information can be supported. Other research by Schumann, Von Wagenheim and Groene (2014), which indicates that social media experience negatively moderates the effect information usage has on the willingness to disclose information, can only be partially be supported. This because only the effect for the attribute no usage could be supported. Thus, social media experience negatively moderates, to some extent, the effect information usage has on the willingness to disclose information. The other formulated hypothesis on the effect of social media experience on privacy practises could not be supported.
In the current study no conclusion could be drawn on the moderating effect of control. Therefore it could not be proven that, when consumers are provided with a degree of control over the other privacy practises, it negatively affects the individual effect that the different privacy practises have on consumers’ willingness to disclose personal information.
