Is your privacy private on mobile social media platforms?
A Quantitative Study on Privacy Concerns
20VT - 2FE21E
Independent degree project of the Marketing Program with integrated method Tutor: Michaela Sandell
Abstract
Purpose:
The purpose of this paper is to explain the effect of trust, knowledge, and control on privacy concerns on mobile social media platforms.
Methodology:
This paper used a quantitative research approach with a Cross Sectional Research Design, in form of a survey, to collect a number of 76 responses. The sample consisted primarily of swedish respondents in the ages of 18-25 with high school education living in a household earning below 19 999 SEK.
Findings:
Our study found significant negative relationships between trust and privacy concerns and knowledge and privacy concerns. This furthers the research field for trust that Milne and Boza (1999), Proudfoot, et al. (2018) and Wenjing and Kavita (2019) laid the foundation on. This also applies to knowledge, by confirming the results of Smit, Van Noor and Voorveld (2014) and Aguirre, et al. (2016). We provide a model where trust and knowledge is described to negatively affect privacy concerns on mobile social media. We also document a so-called privacy paradox from the results.
Research Implications:
Our results suggest that in order for managers to reduce privacy concerns on mobile social media platforms, increasing the levels of trust or knowledge can moderately alleviate such concerns. Knowledge to a slightly larger degree than trust. However, for such companies to customize visible cues only to appear reliable, as per Aguirre, et al (2015), can thereby be argued of little use as this would have little impact on the level of privacy concern displayed in mobile social media users.
Originality/Value:
This paper tests findings from Nowak and Phelps (1995), Milne and Boza (1999), Taylor, Davis and Jillapalli (2009), Smit, Van Noor and Voorveld (2014), Gu, et al. (2017), Proudfoot, et al. (2018), Nam (2018) and Wenjing and Kavita (2019) within a previously yet to be tested context, mobile social media platforms.
Keywords: Privacy, Concerns, Violations, Social Media, Mobile, Platforms, Facebook, Trust,
Acknowledgements
We would like to thank our tutor Michaela Sandell for her continuous support throughout this project. Thank you for your support and feedback that helped us in writing this thesis.
We would also like to thank our examiner, Åsa Devine for holding the seminars and providing crucial feedback on the project. Furthermore, all of our classmates who have been part of our seminars and provided feedback, thank you.
Also, we would like to thank Magnus Willesson for helping us with the statistics and the quantitative part of our thesis.
Lastly thank you to all of the respondents of our survey. Without your help we would not have been able to do it.
________________ ________________ _______________
Table of Contents
1. Introduction 1 1.1 Background 1 1.2 Problem Discussion 2 1.3 Purpose 4 2. Theoretical Framework 5 2.1 Privacy concerns 5 2.2 Trust 6 2.3 Knowledge 9 2.4 Control 11 3. Conceptual framework 13 3.1 Trust 13 3.2 Knowledge 14 3.3 Control 14 3.4 Conceptual Model 16 4. Method 17 4.1 Research Approach 174.2 Data Collection Method 17
4.3 Operationalization 19
4.4 Pretest 22
4.5 Sample Frame and Selection 23
6.2 Knowledge 44
6.3 Control 46
7. Conclusion 47
7.1 Limitations 48
8. Research Implications 49
8.1 Managerial and Theoretical Implications 49
8.2 Societal Implications 50
8.3 Future Research 51
Reference List 52
1. Introduction
1.1 Background
In recent years, the use of mobile applications and social media has become increasingly popular; apps were downloaded 204 billion times in 2019, each month 30 apps are used on average per smartphone owner and an app is opened by 49% of people more than 11 times per day (Blair, 2019; Clement, 2020). The enormous volume of app usage is only expected to grow in the future, as by 2023 the projected revenue from paid apps and in-app advertising is over 935 billion in U.S dollars (Clement, 2019). Moreover, since the average time spent on apps in the U.S is almost 3 hours every day (Iqbal, 2020) and the screen time on average in the UK adds up to a total of 50 days per year (Feeley, 2019), the apps and mobile devices will likely continue to be part of our everyday lives in the near future. While the apps constitute 57% of all digital media, it is worth noting that half of the time spent on apps is specifically used on social apps (Blair, 2019; Iqbal, 2020). This is not surprising however, considering that more than 3.8 billion people are active users of some form of social media (Chaffey, 2020). It has therefore, certainly been one of the biggest phenomena of the modern world since its beginning in the early 2000s (Ortiz-Ospina, 2019).
introduction of regulations such as GDPR in the European Union and the CCPA in California to name a few (Ohlhorst, 2019). Still, these regulations have not been able to downturn the concern over privacy as the issues in this regard are increasingly prevalent (Choi and Land, 2016) and online privacy in terms of data breaches remains among the highest concerns in the 2019 Global Risk Report (Security Magazine, 2019). As consumers continue to use such platforms with concerns remaining high, this has brought forward the discussion of a privacy paradox, which suggest that even though people have high concerns, they still continue to utilize the medium they are concerned with in the first place (Barnes, 2006; Taddicken, 2014)
1.2 Problem Discussion
With the continuation of increased number of users signing up to social media, the trend observed has shifted its usage from the websites to the mobile platforms instead (Chaffey, 2020). Yet, research on privacy concerns within the context of mobile apps is still scarce (Kokolakis, 2017; Pentina, et al., 2016 cited in Jozani, Ayaburi, Ko and Choo, 2020), and is encouraged to focus on applications in which users actively engage in for a longer period of time (Barth, et al, 2019). Making research on mobile platforms more relevant now than ever. Therefore, an excellent motive behind testing the previous effects of trust, knowledge and control towards privacy concerns on mobile social media exists and would most likely benefit both consumers and parties operating within such a field.
While Facebook got fined with $5b by the FTC in July 2019 over violations of consumers’ privacy (BBC, 2019) and the maximum fine allowed of £500,000 by The Information Commissioner’s office in October 2018 (BBC, 2018), its active users went up by 8% whilst revenues arose by 28% in the second quarter of 2019 (BBC, 2019). Regardless of these fines imposed on Facebook and numerous antitrust investigations being opened into Facebook (Palmer, 2019), effectively broadcasting their guilty sentence, usage rates are not dropping. Yet data breaches in online privacy are still one of the highest concerns of 2019 (Security Magazine, 2019) while at the same time not reflected in research (Kokolakis, 2017). According to Nowak and Phelps (1995) and Milne and Boza (1999) this privacy concern should appear high as a result of low levels of trust, knowledge or control. If this were to be the case, social media platforms can work to lower the high concern by alleviating the most significant variable of the three concepts; trust, knowledge and control. Which is why it is important to investigate the effects these have on the level of privacy concerns displayed on mobile social media. Only then can actions be made to lower the concern level of those consumers.
Furthermore, as “[...] privacy concerns can result in different coping behaviors” (Jung and
such service providers to offer a decent service. With that identified, companies can thereby adapt both functional or aesthetic design to its platform to relieve such concerns for its users. Aguirre, et al. (2015) states that by adapting such cues, one could appear more reliable. Therefore, previous findings will be tested in this setting to contribute with a possible explanation as for how trust, knowledge and control affect privacy concerns when it comes to social media on mobile platforms. Such a contribution would not only enable past research to be of relevant consideration today, but also help companies that own and utilize social media platforms as their main point of business in case such users were to signal a concern for privacy.
1.3 Purpose
2. Theoretical Framework
2.1 Privacy concerns
Privacy defined by Warren and Brandeis in 1890 as “the right to be left alone” (Warren and Brandeis, 1890, cited in Peltier, Milne, Phelps, 2009, p.192). In recent years it has been described as “a context-dependent, multidimensional and dynamic concept that evolves with technological advancements” (Westin, 2003; Smith, Dinev, and Xu, 2011; Hong and Thong, 2013; Acquisti, Brandimarte, and Loewenstein, 2015, cited in Jozani, et al., 2020, pp. 1).
Research has found that online privacy concerns stem from personal encounters through hearing about privacy risks from friends and family or monetary loss due to theft of financial account information (Brandimarte, Acquisti and Loewenstein, 2012; Bryce and Fraser, 2014). Moreover it is stated that hearing about online privacy risks from media reports can result in privacy concerns (Chen, Beaudoin, and Hong, 2017). Privacy concerns have been linked to trust (Fletcher and Peters, 1997; Milne and Boza, 1999; Nam, et al., 2006), knowledge (Wirtz, Lwin and Williams, 2007; Aguirre, et al., 2015) and control (Patterson, O’Malley and Evans, 1997; Brandimarte, Acquisti and Lowenstein, 2012;Degirmenci, 2020). Research has found that routine internet activities, such as online information disclosure, are positively correlated with being an internet scam victim, which in turn increases privacy concerns (Chen, Beaudoin, and Hong, 2017). Being an internet scam victim is defined in terms of monetary loss to scammers online (Chen, Beaudoin, and Hong, 2017). Being a victim of privacy invasion increases privacy concerns (Dolnicar and Jordaan, 2006) and often privacy risks are ignored until invasions occur (Chen, Beaudoin and Hong, 2017).
Identity Damage Concerns deals with identity and financial fraud. The latter is concerned with how individuals perceive threats to their own identity and reputation, safety and how represented one's identity is online as well as the threat of financial theft. Handling of personal data entails what companies know and use of ones’ private information but also how behavior can be monitored, data shared with third parties and identity reconstructed from various sources because of information collection (Miltgen and Smith, 2015).
2.2 Trust
Fletcher (2003) proposed two dimensions which affects consumers’ privacy concerns, knowledge and attitude. Attitude means the level of trust exerted from the concerned party. This could be towards an ad or a company involving the disclosure of information (Fletcher, 2003). Within the practice of information management, concern and trust have been explained through a negative relationship where if trust is low, privacy concerns would be high, which would appear as a fundamental core relationship between the two variables (Milne and Boza, 1999). This relationship can be observed through numerous contexts. Cullen and Reilly (2007) observed this when it came to aggressive use of direct marketing practices in New Zealand, which sparked privacy concerns. These exist because responsible direct marketers have not yet, but need to, build the necessary level of trust (Cullen and Reilly, 2007). While “[...] trust in both the social network provider and social network peers influences privacy concerns, social benefits, and perceived IM affordances” (Proudfoot, et al., 2018, p. 16), breaches of privacy already have a strong link with a negative impact on trust (Cullen and Reilly, 2007). On Facebook specifically, it has been observed that users are more willing to share certain types of information than others. Where social identity information was prone to sharing to a greater extent, personal contact information was meanwhile found to be withheld. Thereby it was determined that “ Online trust and Facebook
intensity also interacted and jointly predicted privacy concerns” (Wenjing and Kavita, 2019,
p.187). This relationship of trust and privacy concerns proposed by Milne and Boza (1999) was further strengthened by Miltgen and Smith (2015) who accepted the hypothesis: “ Higher
How that trust is built seems to change as one goes through the time periods and thereby with the emergence of the internet. Milne and Boza (1999) reported that it is through tangible experiences with a company that determines the level of trust a consumer poses (Milne and Boza, 1999). However, Nam, et al. (2006) instead argue that it is the perceived safety and security that builds trust, which conforms to similar conclusions as Horn, Feinberg and Salvendy (2005) where the level of trust is determined by the structure and consistency of websites (Horn, Feinberg and Salvendy, 2005). The level of which can be traced through past experiences with a brand or company that has ultimately proved themselves trustworthy or not, determining the concern for privacy (Milne and Boza, 1999). These past experiences are still subjected to interpretation as they stem from the consumers’ perception of an invasion of privacy (Belanger, Hiller and Smith, 2002; Horn, Feinberg, Salvendy, 2005; Nam, et al., 2006). Thereby consumers’ perception of privacy invasion is determined by his or hers cognitive evaluation (Patterson, O’Malley and Evans 1997).
While trust and commitment are said to have a significant and direct relationship with willingness to disclose personal information (Fletcher and Peters, 1997), consumers are also more likely to share private data when trust towards a certain platform is exhibited (Nam, et al., 2006). Retailers, platforms, websites or other entities can leverage the characteristic ‘trustworthy’ as a unique selling point. Although these must in that case persist with efforts to actively present the consumer the benefits of using their personal data to retain a trusted entity. Otherwise, consumers may feel exploited and vulnerable having to give out personal information that will not reap any reward (Fletcher and Peters, 1997). Belanger, Hiller and Smith (2002) investigated three elements connected to websites; security, privacy and convenience features, where convenience features trumped security and privacy. Meaning that convenience features, such as accessibility, ease of use and design, could add up to higher trust if security and privacy measures would not hold up (Belanger, Hiller and Smith, 2002). The same elements from Belanger, Hiller and Smith (2002) were found in a 2012 study which concluded both user interface and perceived security being the cause of higher privacy concerns among app users compared to computer program users (Chin, et al., 2012).
perceived security. Which is how the consumer perceives the private data provided being handled (Flavian and Guinaliu, 2006). The very same connection was made in 2019 by Mardjo who stated “Perceived security protection has significant effect on trust” (Mardjo, 2019, p. 167). Looking back at the example of lower trust levels in mobile applications, one of the very reasons behind these trust issues was the perceived security of users’ phones (Chin, et al., 2012).
When it comes to personalized advertisements, usually the more personalized an ad is for an individual the chances of that ad being clicked on is generally higher. However, when presented with a personalized ad that matches the preferences of the individual, it is the amount of trust that determines how vulnerable one feels (Aguirre, et al., 2015). Much like Chin, et al. (2012) stated with convenience features, one can thereby adapt cues for the viewer to appear more reliable (Aguirre, et al., 2015). In 2018 Chin, Harris and Brookshire found a strong significant relationship between security and trust. In doing so, the results confirmed the previous study where the link between the three variables were made: “consumers that perceive more security have greater trust and reduced perceived risk” (Harris, Brookshire and Chin, 2016, p.441).
As Harris, Brookshire and Chin’s (2016) result made such a connection, Gu, et al. (2017) also revealed that an increase in perceived risk results in an increase in privacy concerns. Perceived risk is thereby determined by how sensitive the information provided is, by the consumer (Gu, et al., 2017). While Groß (2016) half dismissed these claims and suggested such variables as perceived risk and security are only marginally significant, Kim and Koo (2016) and Mardjo (2019) dismissed them completely. However, the latter author tested the variable against purchase intentions. Such a test was performed years earlier by Chang and Wu (2012) who also found strong correlations between perceived risk and purchase intentions. The results by Kim and Koo (2016) and Mardjo (2019) have also been contradicted in other contexts by Liao, Liu and Chen (2011) and Wang, et al. (2019).
application. Looking back at the study by Chin, et al. (2012), app purchases and simple tasks were approached with greater caution because of the user interface and perceived security. One of the tasks that sparked greater privacy concerns was accessing health data. Because of previous noted reasons for greater caution, the level of sensitivity in the information made the perceived risk increase (Chin, et al., 2012). This would conform to the notions of Harris, Brookshire and Chin (2016) earlier.
2.3 Knowledge
Knowledge refers to the amount of knowledge one has about data collection procedures, storing of personal information and the usage of this information in the context of privacy concerns (Fletcher, 2003).
Knowledge can be divided into three different forms: factual, procedural and experiential. Factual knowledge refers to privacy-related risks, institutional practices, individual rights and legal policies (Park, 2011; Brough and Martin, 2020). Procedural knowledge is the practical side; how to utilize privacy-enhancing strategies, tools and skills to safeguard one’s personal information online (Park, 2011; Baruh, Secinti and Cemalcilar, 2017; Brough and Martin, 2020) Experiential knowledge relates to personal experience with privacy violations and general familiarity with online technology (Park, 2011; Park and Jang, 2014; Brough and Martin, 2020)
personal information while being able to configure access to the shared information (Yerby, Koohang and Paliszkiewicz, 2019).
Subsequently, lack of knowledge, or lower levels of knowledge, lead to worries of misuse of personal data and possible violations of privacy (Smit, Van Noor and Voorveld, 2014; Aguirre, et al., 2016). Research has shown that groups with the highest level of concern and lowest level of knowledge are older people, females, people with low level of education (Smit, Van Noor and Voorveld, 2014) and low family income (Park and Jang, 2014; Smit, Van Noor and Voorveld, 2014). Furthermore, research has found affluent people to be more likely to possess knowledge (Park and Jang, 2014). Research has found that time spent and frequency of changing privacy settings on social media platforms and social networking sites (SNS) is positively related to knowledge and perceived control (Park and Jang, 2014; Barthsch and Dienlin, 2016). That is, knowledge can, and should, be accrued by conversing with SNS webpages, as it might help reduce threats to privacy (Barthsch and Dienlin, 2016).
While the perceived permission sensitivity, id est information collected is considered sensitive by the consumer, increases privacy concerns, the popularity of the platform decreases these concerns (Gu, et al., 2017). Providing the consumer with knowledge about data collection can also lower the amount of concern (Gu, et al., 2017). However, this has only been found to be the case in consumers with not a significant amount of personal experience with privacy violations (Gu, et al., 2017).
2.4 Control
Based on the relationship provided by Nowak and Phelps (1995), in the context of online social environments and digital platforms where concerns over data privacy are of high relevance, control is referred to as the individual’s ability to manage private information and to determine how this information of oneself is utilized (Lazaro and Le Métayer 2015). Control has a key role in defining what constitutes the concept of privacy: ‘ Privacy is the ability to control the acquisition and use of one’s personal information. ’(Westin 1967; cited
in Hann, et al., 2002).
Taylor, Davis and Jillapalli (2009) found that consumers’ intentions to act and privacy concerns have a negative relationship which is further strengthened by the perception of low control over personal information. The results suggest that in a context of personalized online interactions, positive intentions to engage online can be increased by providing higher levels of perceived information control since this will reduce the privacy concerns of the users (Taylor, Davis and Jillapalli, 2009). This is supported by Tucker (2014) who found that likelihood of positive behaviour online correlates with an increased ability to control private information on social networking sites. In accordance with Taylor, Davis and Jillapalli (2009), control and privacy concerns were also connected by Nam (2018), having a negative relationship and ultimately affecting people’s behavior online.
2007). Government legislation enables users to have more control over their private information due to the risk of punishment that the content providers face if private information is inappropriately collected, which in turn leads users to believe that the laws are abided thus, resulting in overall increase in perceived control over personal data (Xu, 2007). The support for government regulation was found to be stronger among consumers when the privacy concerns are high, as studied by Yang and Liu (2013) in the context of Chinese market. Additionally, mobile users’ concern levels were confirmed to be influenced by perceived control in mobile apps (Degirmenci, 2020).
The importance of perceived control and its effect on lessened privacy concerns was also highlighted in the study conducted by Hoadley, et al., (2010). They found that the sense of control being lost due to certain Facebook news-feed functions immediately resulted in concerns over the users privacy (Hoadley, et al., 2010). It is implied that in order to diminish the privacy concerns, it is vital that the perceived control and easy access to information is provided (Hoadley, et al., 2010). Similarly, a previously mentioned study by Nam (2018) suggests that one of the main predictors of low interaction in Facebook is caused by privacy invasions experienced due to low control. Furthermore, people seemed to have high willingness for control and also, experienced that they have a right to control decisions that regard their personal data (Nam, 2018).
3. Conceptual framework
3.1 Trust
Nam, et al. (2006) found in the instance of high trust that consumers are more likely to share their private data, as suggested by Fletcher and Peters in 1997. Meaning that privacy concerns would appear low, which would ultimately conform to Milne and Boza’s (1999) suggestion that trust and privacy concerns are determined via a negative relationship. The very same notion can be observed through various contexts, also in relation to how vulnerable, or concerned one is, is determined by the level trust (Aguirre, et al., 2015). The implied meaning of Cullen and Reilly’s (2007) study was that privacy concerns from direct marketing were high due to low levels of trust. Also, trust is found to influence privacy concerns, among other variables, when it comes to social network providers and social network peers (Proudfoot, et al., 2018) and specifically on Facebook, users were found to withhold certain information because of privacy concerns, meaning that trust levels appeared low (Wenjing and Kavita, 2019). In another study by Miltgen and Smith (2015), high trust for entities operating within information privacy was determined to be associated with low levels of privacy concerns. As the latter study contained an accepted hypothesis with the relationship of high trust and low privacy concern in a similar setting to this study, to further strengthen such a proposition, a hypothesis built upon the negative relationship from Milne and Boza (1999) is formed specifically for the mobile social media context. While such results have been observed in older studies and in other environments, this will be used to confirm its applicability in this setting. Where if confirmed, would indicate that Milne and Boza’s (1999) results are still applicable, and explains the effect trust has on privacy concerns on mobile social media platforms. Thus the hypothesis for trust is as follows:
3.2 Knowledge
Low knowledge coupled with low levels of trust has been found to result in concerns (Aguirre, et al., 2015). With the implied meaning of a negative relationship (Nowak and Phelps, 1995), research has also found that low levels of knowledge lead to worries and concerns (Smit, Van Noor and Voorveld, 2014; Aguirre, et al., 2016) but this has yet to be proven to be the case on mobile social media platforms. While research conducted by Park, Campbell and Kwak (2012) found that higher levels of knowledge lead to more protective actions, which are preceded by concern, the results have yet to be corroborated in this context. It has also been noted that concern in and of itself is not a meaningful predictor of privacy protection behaviors (Park, Campbell and Kwak, 2012).
Interestingly higher levels of knowledge lead to more protective actions (Park, Campbell and Kwak, 2012), while high levels of concern do not (Smit, Van Noor and Voorveld, 2014; Aguirre, et al., 2016) It seems that protective actions happen only when preceded by high levels of knowledge. Thus researching whether knowledge has a negative relationship with privacy concerns on mobile social media platforms is of relevance. Thereby the following hypothesis can be formulated:
H2: Knowledge negatively affects mobile social media users’ privacy concerns
3.3 Control
information.’(Westin 1967 cited in Hann, et al., 2002). Control was also referred to as being a perception of it by Xu (2007), Hoadley, et al. (2010), Degirmenci (2020) and the sensation of being in control over data is what ultimately results in alleviated privacy concerns which can be seen through corresponding online behaviors (Brandimarte, Acquisti and Loewenstein, 2012; Tucker, 2014). Therefore, it could be argued that the distinction between actual control and perceived control is not so relevant, since online users are likely to not make a distinction between these two and the overall effect on their behavior that stems from concern levels is the same (Brandimarte, Acquisti and Loewenstein, 2012). This is to say that as long as the user is under the impression that they have the ability to control as defined by Westin (1967 cited in Hann, et al., 2002) , concern over privacy is alleviated. Again, as mentioned by Xu (2007) the perceived control can be affected by regulation, legislation and technology in hopes to reduce the privacy concerns which would be seen as positive online behavior in most scenarios (Taylor, Davis and Jillapalli 2009; Tucker, 2014; Nam, 2018).
A relationship of high control and low concern has been provided by Taylor, Davis and Jillapalli (2009) and Brandimarte, Acquisti and Loewenstein, (2012). This would suggest a negative relationship between the two variables in mobile social media platforms and as stated earlier, imply that a correlation between privacy concern and control exists. Moreover, (Hoadley, et al., 2010; Nam, 2018) have found the same relationship, stating that a high concern level would be the result of low control/perceived control. Ultimately, this gives reason to suspect there is a negative relationship between the two variables. Since, support for both low control resulting to high concern, as well as high control leading to low concern can be found, an argument for testing the prior could be made based on the consumer’s ability to identify when they are not in control as shown by (Hoadley, et al., 2010; Nam 2018.) However, since a negative relationship between control and privacy concern would imply the same results, there is no need to differentiate between testing low knowledge against high concern or the other way around. Therefore, based on the findings supported by the literature and the focus of this paper, the following hypothesis is formed in order to measure control’s significance in relation to privacy concerns.
3.4 Conceptual Model
From the stated hypotheses the following model can be made in which all the hypotheses correspond with a negative correlation between the stated entities. While this is a brand new model, it stems from the most recurrent findings in literature.
4. Method
4.1 Research Approach
This study ensued a quantitative approach, where most commonly one utilizes measurements of a specific phenomenon to measure and notice differences objectively, and often in large quantities of data. Within this approach, deductive reasoning was employed for this research, due to its inherent qualities that allow the researchers to form hypotheses from theory. Rather than generating theory as inductive research, deductive concerns that of testing theory. Where one starts with existing theory where the main goal is to confirm, reject or possibly revise it depending on what, where and how one is testing it (Bryman and Bell, 2011).
Furthermore, this study was concerned with testing existing theory on trust, knowledge and control and their effects on privacy concerns in a new context. As it was a new context these were tested on, primary data was collected using a cross sectional research design. Which pertains to data being collected on multiple cases. Most often, data is collected on a large quantity of cases with two or more variables in mind. By stacking a large level of data from cases, it is possible to see patterns which might infer correlations among the variables. Also, large quantities of data can enable a greater level of generalizability of the findings produced from this study. Cross sectional research design however, usually entails a questionnaire or survey where the point is to collect data on multiple cases at a single period of time (Bryman and Bell, 2011). This was optimal for a research of this kind due to the sheer amount of users on mobile social media platforms and the inability to attempt to generalize our results without a larger sample. Furthermore this research stemmed from existing works of many authors and as such the possibility to confirm previous results was best achieved with a cross sectional research design.
4.2 Data Collection Method
and Facebook users would naturally feel comfortable answering a survey online. For this, Google Forms was utilized because of its accessibility and fair ease of use where editing and presentation is simplistic. The data one receives also has little reason for data errors such as by providing fake information due to disclosures about how and by whom the data would be utilized. First, the user who opens the survey is faced with a short presentation of what the survey is about, why it is required and who the researchers are. Second, three filter questions are presented where information is acquired about if the respondents use social media on mobile devices, how long they have been using that particular media (in terms of years) and how frequently it is used. Third, questions about demographics are presented, where respondents answer questions regarding, age, gender, country of residence, education and household/family income. These demographics were necessarily not used to pinpoint a certain segment, but to give extra information about the respondents which might have an impact on sources of error, or simply might skew the final results of the study. It also provides some understanding as to how generalizable the data is to the population. However, there is reason to believe that some groups of specific demographics have a connection to the variables currently studied. Smit, Van Noor and Voorveld (2014) found that older people, females and uneducated people possess the highest levels of concern and lowest levels of knowledge. It could then be hypothesized that these groups likewise perform the least protective actions due to the lack of knowledge. While uneducated people with a low family income have the lowest levels of knowledge (Park and Jang, 2014), it could be argued that based on the relationship between protective actions and knowledge (Park, Campbell and Kwak, 2012), defending one’s personal information online is a commodity only available to the opulent higher class.
4.3 Operationalization
4.4 Pretest
Pretests or pilot studies are utilized with self completion questionnaires in order to locate inherent problems with the questions or the design of the questionnaire. Due to the nature of self completion questionnaires, administrators cannot be present at the time of responding, thus it is important that the questions are not misunderstood and have the ability to produce desirable answers while being certain that the instrument as a whole functions as intended (Bryman and Bell, 2011).
A pretest was conducted on people comparable to the actual sample, whilst not being a part of it, to not employ the possibility of contaminating the actual sample pool. This was also considered due to its ability to combat tampering with the representativeness of the sample (Bryman and Bell, 2011). The procedure consisted of the subject, or responder, going through the survey on their own in front of a researcher either online or in person, whilst going through each question and asked to ‘think out loud’ to demonstrate that they understand it and make a decision on the correct intended information. Without interference from the researchers, notes were taken down to indicate if the subject understood the questions as intended or how they perceived them instead if that was the case. Moreover, if certain questions/statements were unclear, the subjects informed of this and provided feedback on their understanding at the end of the survey.
Afterwards the survey was published and sent out. However, after consideration it was determined that the questions related to privacy concerns were not reliable enough and did not measure the intended concept. Therefore, new measurements were adopted. The new questions related to privacy concerns have been used from Miltgen and Smith’s (2015) study where they measured the level of privacy concerns using two dimensions: Data Tracking Concerns and Identity Damage Concerns. In this article, 11 questions could be found of which eight questions have been adopted for this paper. The main reason for this was that the questions were already proven to be successful indicators of privacy concerns and they had been tested by an expert workshop. However, the reduction of questions from eleven to eight was due to some already being connected to the independent variables for this paper, which needed to be measured separately (Miltgen and Smith, 2015). Furthermore, from the initial data gathered, it was possible to observe errors in the measuring items. Subsequently, one question related to knowledge was removed as deemed inaccurate and four questions related to trust were revised to better measure the intended items.
4.5 Sample Frame and Selection
4.5.1 Population
younger generations are inherently more active on social media compared to older generations (Chaffey, 2020), a compromise was made since the population has no exclusion on specific age demographics. Thereby, those who use mobile social media weekly or more, are considered as active users.
4.5.2 Sample
To represent social media the chosen platform was Facebook as this is the most popular social media platform (Clement, 2020), and would result in a higher probability of inclusion to represent the entire population. The sample is thereby Facebook users who use the platform on their mobile devices weekly or more and for a period of minimum five years. To reach this sample a survey was simply published on the platform from the authors individual accounts. The sample size for a representative sample was calculated based on the independent variables with the formula of “n > 50 + 8 x M”, where n = sample size and M = number of independent variables. N> 50 + 8 x 3 gave us a representative sample size of 74. While this formula is a good “rule of thumb” for multiple regression analysis it has its limitations. It is not always applicable and researchers should also pay attention to the minimum effect size to determine the proper sample size (Green, 1991). However, since the population includes millions of users it would be hard to determine effect size therefore, the stated formula has been used as a bassline. Bryman and Bell (2011) advocate that a higher number of people within the sample creates a better representation of the population, the sample size of 74 is only the bare minimum needed for this study.
The sample for this study consisted of 76 respondents in which a clear majority of Swedish people (58.7%) were between the ages of 18-25 (61.3%) followed by 25.3% of 26-34-year-olds. Including fairly level distribution of males and females where 49.3% had at minimum a high school education as well as 34.7% were undergraduates and a majority was living in a household earning between 10 000 - 19 999 SEK / 1,000 - 1,999$ per month.
4.5.3 Sources of Error
Sampling error to a degree is highly likely due to the unlikeliness of achieving a truly representative sample (Bryman and Bell, 2011). However, the degree of error can be mitigated to subnormal levels. Data collection errors are occurrences related to the wording of questionnaires and other flaws in the administration of the research instrument (Bryman and Bell, 2011). The effects of these can be mitigated by pretesting the instrument in order to find and correct any and all problems associated with the wording and/or administration of the instrument. For the aforementioned reason, we have conducted a careful and thorough pretest. After the pretest, a survey was published which meant to serve as the main collection of data. However, this survey was revised after the dependent variable was observed as not reliable enough of a measure. Instead, measurement scales and questions that had already been tested and proven to work by Miltgen and Smith (2015) were adopted and revision of other faulty questions was conducted. This all added up to a mitigated degree of error that was catered to by publishing a new survey. This all helps with reliability and validity of the study which is discussed later in this chapter. However, because of time constraints, the new survey could not collect a large quantity of respondents within the given time frame. As previously mentioned, this may result in a sample size that is not generalizable enough to apply to the population and may contain hidden sample bias. Nevertheless, the new survey managed to collect data from just more than the stated minimum of 74 respondents, with 77 in which one answer was discarded as it was not qualified for the sample criteria.
other local communities. This would serve the sample as an immediate bias towards people with a certain age, education, country of residence as well as household income. Thereby, the likelihood of respondents in the ages of 18-25 with a household income below 19 999 SEK / 1,999$ situated in Sweden has been part of a higher probability of representation than other demographics. This should be made aware as a limitation to the study.
4.6 Ethical Considerations
Bryman and Bell (2011) discuss the ethical considerations in relation to the harm of participants, respondent consent issues, privacy invasions, deception, trust, conflict of interest, copyright and data management issues. These have all been taken into consideration in this study and it was made sure that these were not disregarded in hopes to achieve results. When it comes to the survey itself, respondents were made aware in the initial introduction that their data will be used for this research for overall conclusions and that it will not be used to look into individual data, thereby allowing the respondents to fill the survey anonymously. However, respondents may not trust this statement and answer accordingly as if their identity is not completely anonymous. For this, the researchers can only hope that respondents trusted the intentions of the researchers. Such trust and privacy issues may also arise due to the survey platform used, which in this case was Google Forms. Any potential issues the respondents might have had with Google can affect the ethical discussion in general. Further, on the same point of trust and privacy issues is the fact that these are the very issues being studied in this paper. This means that if there were any prior ethical issues in this subject matter for the respondents answering the survey, it could have heavily influenced the answers they chose to use in the survey. There is also the chance that when people did the survey, they became aware of the issues of privacy violations, and thereby, just by participating could have influenced their choices and opinions. Reading about privacy violations could in itself create more privacy concerns.
4.7 Societal Considerations
forward the idea of a privacy paradox, or what is often referred to as the personalization-privacy paradox. This, in essence, entails that people are concerned about their privacy only to then forgo these concerns to experience the benefits that stem from compromised privacy, i.e. personalization (Barnes, 2006; Taddicken, 2014). As data management and privacy violations are matters of the online world and therefore not of tangible entities where the negative effects could more easily be spotted, one can argue it is easy to forgo these privacy protections for immediate benefits. This leads to the ethical concerns that studies such as this one can have upon society. While such studies might be available for everyone to take part in, who will actually exploit such revelations? Even though consumers appear to demand more privacy protection, the laws and regulations fail to sufficiently provide this due to the technological advancements as well as the navigation of the political field when it comes to passing laws and regulations. It is clear that companies are not ready to give up their personalized offerings and thereby valued products. This brings upon an ethical discussion among researchers themselves. On the one hand, studies of this nature may be used by companies to alleviate privacy concerns before they become privacy violations for the overall ‘greater good’ of society. On the other hand, these results can instead be used to ‘hide’ privacy concerns from users without actually solving them. Meaning, one could create a false sense of security by customizing the cues that indicate concern among consumers, similarly to what Aguirre, et al. (2015) proposed as an effort to appear more reliable. Knowledge generation is then dependent on how researchers argue for its use. If it is the case that such knowledge should be distributed to the world, researchers can only hope that it will be used for the greater good. Or, in the other case that it is used for commercial gains, one can only hope that consumers become aware of how such contributions are used by different entities, leaving the responsibility to the individual to make their own choice: give up your privacy or give up the benefit that could be gained by submitting to the lack of privacy.
4.8 Data analysis
median, mode, standard deviation, variance, skewness, and kurtosis. Along with basic demographics, these are meant to provide an initial picture of how the data looks. An overall idea regarding distribution could thereby be made. Skewness shows how symmetrically distributed the answers are for each item (Freeman, et al., 2015) enabling us to see towards what opinion a majority leaned in regards to a certain question. The closer to value 0 the skewness is, the more equally the responses are distributed (Freeman, et al., 2015). Kurtosis on the other hand, lets us observe how uniform the distribution is where small or light tails having small amounts of outliers means a low kurtosis (platykurtic) and heavy tails would indicate a high kurtosis (leptokurtic) (McAlevey and Stent, 2018). A very high or low kurtosis would entail possible issues with the data since largely differing outliers in relation to the normal distribution could affect the mean value drastically and therefore, misrepresent the actual responses of the survey.
Next, Cronbach’s Alpha has been used for a reliability analysis followed by a Spearman correlation analysis used to test for validity. These will be explained further in chapters 4.9.1 and 4.9.2. Lastly, a regression analysis was performed to accept or reject the proposed hypotheses.
4.8.1. Regression analysis
the explanation behind the relationship should be included and it should be based on relevant theories (Freeman, et al., 2015).
For the regression analysis to be appropriate, some measures needed to be taken into account. The F-test was used to determine whether the results of the regression could be deemed significant when there are more than one independent variable whereas with only one independent variable, a t-test should be considered (Freeman, et al., 2015). Since in this paper, there was an interest to look into multiple independent variables and their relation to the dependent variable, the use of F-statistics for significance of the model was appropriate. However, since an independent-samples t-test will compare the values of two different means and reveal any large variances they may have (Freeman, et al., 2015), it is beneficial in examination of variables that are not ranked or do not follow a numeric scale. Consequently, this t-test was used to determine if gender has a significant impact on the phenomenon being investigated.
Testing the hypotheses with the regression analysis is conducted by looking at the standardized coefficient beta (β) and the significance level p.The purpose of the p-value is to use a predetermined confidence level and see whether the results are significant at this level or not in regards to the p-value. Commonly a 95% confidence level interval is used to verify statistical significance in results (Freeman, et al., 2015) and while higher confidence intervals could arguably be beneficial for very large sample sizes (Moore, McCabe and Craig, 2014), the use of the 95% level was deemed appropriate based on the size of the sample and thus, was utilized in this paper. This implies that the results that show a p-value of less than .05 can be deemed as statistically significant.
Further, the β-value should be in line with the hypotheses for them to be accepted. Even if the relationship would be significant in terms of the p-value, a β-value that denotes a linear relationship in the opposite direction in the regression, would lead to the rejection of the hypothesis. The reason for why the standardized coefficient beta is used instead of the unstandardized beta, is that since the β-values are standardized to fit the model it allows us to compare the relative effect of each predictor variable in the model and helps in determining which are the most meaningful ones.
Finally, the measure of R squared and adjusted R square is used to interpret how much of the variance in the dependent variable can be explained by the model. This is called the coefficient of determination (Freeman, et al., 2015). The higher this number is, the more precisely the model predicts the change in the dependent variable. The adjusted R square was used for this analysis, since it takes into consideration the amount of variables used and adjusts the values to fit the number of predictors. The adjusted R will be within the range of 0-1, where 1 would mean that the model explains 100% of the total variance in the dependent variable (Moore, McCabe and Craig, 2014).
4.9 Research quality
4.9.1 Reliability
Cronbach’s alpha refers to the internal reliability of the model being tested and reveals how well the variables are measuring the same concept (Tavakol and Dennick, 2011). The higher the alpha value is, the less error in the test results can be expected and thus, a higher value is preferred. The values of Cronbach’s alpha are between 0 and 1 and a reliable range for the test is above 0.7 but no higher than 0.9 (Tavakol and Dennick, 2011). If the value is less than the low end of the range, it implies the variable in question might not measure what it is supposed to. When there are multiple concepts being tested, the values should be accounted for each concept individually or otherwise, the alpha may not accurately represent the internal reliability of the data due to falsely inflated figures (Tavakol and Dennick, 2011).
4.9.2 Validity
Validity refers to the conclusions of the research, whether or not they can be considered as valid. Measurement validity attempts to explain the measure’s reflection of the actual concept. In the case that the measures do not accurately measure the concept, the validity of the research can be called into question. It is often related to reliability; an unstable or inconsistent measure produces an invalid measure. Thus it is of utmost importance to have robust and consistent measures to secure the validity of the research (Bryman and Bell, 2011).
External validity is an important issue concerning quantitative research. This concerns whether or not the results of the research can be used to generalize a larger population than the sample itself (Bryman and Bell, 2011). In this research we have carefully selected and outlined the sample for this research in an effort to produce a representative sample. We formulated our sample based on the works of Park, Campbell and Kwak, 2012, Park and Jang, 2014, Smit, Van Noor and Voorveld, 2014 and Barth, et al., 2019 and chose the platform based on popularity (Statista, 2020). This has given this research the ability to produce as valid conclusions as possible within this framework. Additionally from the adopted scales for privacy concerns (independent variable) by Miltgen and Smith (2015) who stated that in order “ to verify their content validity, all scales were first pre-tested and
members of the academic community, was utilized. During this worksop, all the items in the survey were discussed, and a number were revised. We then pre-tested the survey with 117 young UK subjects” (Miltgen and Smith, 2015, p. 726).
Additionally, the use of spearman correlation analysis has been utilized in this research. The spearman correlation is used to measure the correlation and its direction between the variables of interest. The spearman is commonly used instead of pearson’s correlation whenever the data consists of rank values or ordinal scale data (Freeman, et al., 2015) Moreover, the spearman correlation reveals important insights of the statistics by demonstrating whether the correlations are significant enough to be relevant and thus, providing validity for further and more complex data-analysis. Significance in correlations ensures the independent variables and the dependent variable are not independent but rather, have an effect on each other. If no significant correlations exist, the variables in the data do not have sufficient relationship to explain the outcome of the dependent variable. Since the conducted survey has ranked the answers to represent low or high levels of each predictor as well as the dependent variable, spearman correlation was deemed to be appropriate.
5. Results
5.1 Descriptive Statistics
Age <18 18-25 26-35 36-45 46-55 >55 6.7% 61.3% 25.3% 4% 1.3% 1.3% Gender Female Male
Prefer Not To Say Other
42.7% 53.3% 2.7% 1.3%
Country of Residence Sweden Norway Denmark Germany Finland China USA Switzerland Portugal Netherlands Italy Great Britain France 58.7% 10.7% 1.3% 4% 2.7% 4% 5.3% 2.7% 1.3% 1.3% 1.3% 2.7% 4% Level of Completed Education
Middle school/comprehensive school High school
Undergraduate (BSc or equivalent) Graduate school (MSc or equivalent) PhD 4% 49.3% 34.7% 9.3% 2.7% Family/Household Income < 1000 $ 1001$ - 1999$ 2000$ - 2999$ 3000$ - 3 999$ 4000$ - 4 999$ > 5000$ 14.7% 45.3% 14.7% 1.3% 10.7% 13.3% Table 2: Demographics
privacy concern was very similar between the genders. The same applied for all of the predictor variables where the responses were similar to one another with a small exception to control where females reported slightly lower levels of control against males. Other genders and those who preferred not to say consisted of only 3 answers in total, which means that the representation of these groups is too insignificant to determine if their gender was the predicting factor of their levels of privacy concern, trust, knowledge and control.
T-test Gender N Mean
Privacy Concern Female Male
Prefer not to say Other 33 40 2 1 3.8030 3.5063 4.0050 4.2500 Trust Female Male
Prefer not to say Other 33 40 2 1 2.5871 2.5906 2.7500 3.0000 Knowledge Female Male
Prefer not to say Other 33 40 2 1 2.8232 2.7833 2.5850 2.6700 Control Female Male
Prefer not to say Other 33 40 2 1 2.8586 3.3417 3.1700 3.6700
Table 3: T-test: a comparison of means N=76
them. Furthermore, a rather high variance within separate items can be observed across all variables. More so than others, C1 has the highest variance which would indicate that there is a mixed perception of the amount of control one has over what information one discloses on mobile social media.
Item Minimum Maximum Mean Median Mode
Standard
Deviation Variance Skewness Kurtosis
T1 1 5 3.21 3.00 3 1.111 1.235 -0.012 -0.735 T2 2 5 3.93 4.00 4 1.011 1.022 -0.660 -0.610 T3 1 5 3.26 3.00 4 1.248 1.556 -0.138 -1.063 T4 1 5 2.41 2.00 2 1.122 1.258 0.470 -0.571 T5 1 5 2.63 2.00 2 1.274 1.622 0.453 -0.884 T6 1 4 1.78 1.50 1 0.947 0.896 1.049 0.128 T7 1 5 2.04 2.00 1 1.089 1.185 0.875 -0.006 T8 1 5 1.53 1.00 1 0.901 0.813 1.882 3.239 K1 1 5 4.05 4.00 5 1.031 1.064 -1.081 0.716 K2 1 5 2.45 2.00 2 1.310 1.717 0.574 -0.844 K3 1 5 1.75 1.00 1 1.133 1.283 1.529 1.393 K4 1 5 2.30 2.00 1 1.255 1.574 0.484 -0.890 K5 1 5 2.74 3.00 3 1.182 1.396 0.183 -0.773 K6 1 5 3.47 4.00 4 1.341 1.799 -0.453 -1.099 C1 1 5 3.41 4.00 5 1.462 2.138 -0.457 -1.224 C2 1 5 2.63 2.50 2 1.253 1.569 0.278 -0.980 C3 1 5 3.36 4.00 4 1.334 1.779 -0.477 -0.978 PC1 1 5 4.12 5.00 5 1.166 1.359 -1.066 -0.052 PC2 1 5 3.86 4.00 5 1.262 1.592 -0.906 -0.292 PC3 2 5 4.38 5.00 5 0.966 0.932 -1.388 0.691 PC4 1 5 4.00 4.00 5 1.155 1.333 -1.121 0.516 PC5 1 5 3.54 4.00 4 1.183 1.398 -0.519 -0.549 PC6 1 5 2.87 3.00 1 1.379 1.902 0.055 -1.245 PC7 1 5 2.53 2.50 1 1.390 1.933 0.296 -1.264 PC8 1 5 3.97 4.00 5 1.177 1.386 -1.056 0.262
5.2 Reliability Test
The Cronbach’s alpha values for trust, knowledge and privacy concerns were .781, .772 and .864, respectively. However, for control the value is .539. Deleting items within the control variable would not increase the statistic over the threshold, hence control must be regarded as a non-reliable measure. Effectively, the results for control will be discussed but they must be regarded with the impact of reliability measures in mind. A further discussion regarding control will be found under discussion and limitations.
Variable Cronbach’s Alpha Cronbach’s Alpha if Item Deleted Privacy Concern PC1 PC2 PC3 PC4 PC5 PC6 PC7 PC8 0.864 0.851 0.847 0.842 0.833 0.839 0.855 0.879 0.836 Trust T1 T2 T3 T4 T5 T6 T7 T8 0.781 0.764 0.775 0.752 0.746 0.775 0.741 0.746 0.757 Knowledge K1 K2 K3 K4 K5 K6 .772 0.792 0.718 0.733 0.723 0.702 0.752 Control C1 C2 C3 .539 0.364 0.664 0.188
5.3 Validity Test
The spearman correlation shows that the negative correlations of trust, knowledge and control in relation to privacy concerns were apparent with respective values of -.331, -.398 and -.350. These correlations also showed statistical significance at the 95% confidence level. Positive correlations amongst the other variables was also found, yet only the correlation between control and knowledge was statistically significant with a positive correlation of .493. All of the values were within the range of -1 to 1 which implies that there were no calculation errors that have occured in the process of computing the validity test. Moreover, multicollinearity was not deemed to be an issue since none of the correlations displayed a value that would be extremely high or close to 1. For improved validity however, the VIF statistics was computed to see if the somewhat high correlation between control and knowledge would be problematic. The VIF test proved that there are no multicollinearity issues between knowledge and control as the VIF was 1.314. While most of the correlations between the independent variables trust, control and knowledge were not significant, the correlations between the independent and dependent variables were found to be significant. Since the latter correlations represent the purpose of this paper and as multicollinearity among the independent variables was non-existent, the spearman correlation was deemed to be effective thus, validating the measures for further analysis.
Variable Privacy Concerns
Trust Knowledge Control Privacy Concerns 1.000 -.331* -.398* -.350* sig. 0.004 0.000 0.002 Trust -.331* 1.000 .220 .162 sig. 0.004 0.057 0.161 Knowledge -.398* .220 1.000 .493* sig. 0.000 0.057 0.000 Control -.350* .162 .493* 1.000 sig. 0.002 0.161 0.000
Collinearity statistics Tolerance VIF
Knowledge .761 1.314
Table 7: Dependent variable: Control VIF test
5.4 Hypothesis Test
In order to answer the hypotheses H1, H2 and H3, a regression analysis was conducted. In the following table the results from computing the 5 different models are presented. The model 1 represents the effect of the control variables in relation to the dependent variable and is computed to conclude whether these factors have an impact on the outcome of the model. Models 2, 3 and 4 are simple linear regressions to determine their individual relationship and significance in predicting the outcome of the dependent variable. Model 5 is a multiple regression analysis that takes all of the factors into consideration and is used to test whether the hypotheses should be accepted or rejected.
The F-values were significant at the test level of p < .05 for all of the models except for the Model 1. Same applies for the adjusted R-square value as it lands between the acceptable range of 0 and 1 for models 2 (=.092) , 3 (=.190), 4 (=.160) and 5 (=.258) but not for the model 1 (= -.026). Therefore, the results indicate that the control variables usage, age, education and income are not significant in predicting the model and can be disregarded in the further analysis.
In order to determine whether the hypotheses should be accepted or rejected, the model 5 that takes all of the independent variables into consideration was computed against the dependent variable. For H1 the β was -.232 and significant by displaying a p-value of .035. The β-value was negative and in accordance to the hypothesized outcome and therefore, were accepted. H2 was also significant with p=.036 and β= .-271, showing that knowledge and privacy concern have a significant negative relationship and the hypothesis was also accepted. While H3 shows a negative relationship in relation to the dependent variable (β= -.228), this relationship was not significant as the p-value (= .085) was not below the required significance level of .05.
Exp. Sign Model 1 (Controls)
Model 2 Model 3 Model 4 Model 5 (All) HP Accept/Reject Intercept 2.645 4.246 4.185 4.384 5.437 Control variables: Usage .161 (.180) .086 (.452) .130 (.222) .146 (.181) .088 (.398) Age -.015 (.908) -.064 (.612) -.002 (.988) -.051 (.673) -.057 (.623) Education .053 (.686) .061 (.620) .034 (.768) -.042 (.727) -.001 (.996) Income .047 (.731) .100 (.441) .049 (.689) .015 (.905) .065 (.578) Trust H1: - -.365* (.002) -.232* (.035) Accept Knowledge H2: - -.465* (.000) -.271* (.036) Accept Control H3: - -.452* (.000) -.228 (.085) Reject R Squared .029 .152 .244 .216 .327 Adjusted R Square -.026 .092 .190 .160 .258 F-Value .528 2.513* 4.525 3.864* 4.723* Degrees of Freedom 4 5 5 5 7
Table 8: Regression Model With Independent and Dependent Variable * Significant at level = 95% (p < .05)
6. Discussion
6.1 Trust
Consistent with the findings of Milne and Boza (1999) our findings suggest a negative correlation between privacy concerns and trust. This corresponds with further results by Cullen and Reilly (2007), Miltgen and Smith (2015), Proudfoot, et al. (2018) and Wenjing and Kavita (2019). High trust has been found to result in low levels of privacy concerns in other environments (Milne and Boza, 1999; Miltgen and Smith, 2015). Our study matched the proposition of a negative relationship between trust and privacy concern by Milne and Boza (1999) and found it applicable to mobile social media platforms. While the results display a significant negative relationship, the effects between the two are only moderate (-.232). Although, from this study it is possible to observe low levels of trust and high levels of privacy concerns partly predicted by Chin, et al., (2012), privacy concerns would appear high regardless of the level of the predictors. Moreover, the control variables, demographics, do not play a role in the level of trust and privacy concern one possesses. While the effect between trust and privacy concern is negative and a relationship between the two can be established, there is only a moderate negative effect meaning that privacy concerns can be alleviated to a small degree by increasing trust on mobile social media platforms.
trustworthy as a unique selling point, by actively presenting the consumer of the benefits of the company utilizing their personal data. Meaning, that as long as mobile social media platforms continue to utilize users’ personal data in a satisfactory manner, the users accept a loss of privacy regardless of the amount of trust displayed. Which makes reports by Nam, et al. (2006) not relevant in this context as regardless of trust, users still share their private data. Moreover, by utilizing private data from users, online social media platforms can personalize content to its users, and when presented with this content it is the amount of trust that determines the level of vulnerability (Aguirre, et al., 2015). While it is not possible to confirm this relationship, it is possible to see a generally low level of trust from perceived experience and safety, and that privacy concerns are reported as very high translating to very high vulnerability.
While experience and safety perceptions displayed lower levels, convenience features raised the general trust level among mobile social media users in accordance with Belanger, Hiller and Smith’s proposition (2002). It was said that accessibility, ease of use and design could add to higher trust if security and privacy measures would not hold up, which appears to be the case on mobile social media platforms as well (Belanger, Hiller and Smith, 2002). However, our results do not portray that this would matter per Milne and Boza’s (1999) negative relationship, since privacy concerns in this case are still reported as very high. It might be argued though, that privacy concerns would be even higher if convenience features were not taken into account, but even then the effect seems marginal which in any case results in high privacy concerns. This would mean that Chin’s, et al (2012) and Aguirre’s, et al (2015) notion of being able to adapt cues to appear more reliable only affects the total amount of trust marginally and thereby the level of privacy concern as well. This needs to be studied more in detail however to see an exact effect.
disregarding convenience features of trust in this context which speaks against the dismissal of such relationship by Kim and Koo (2016) and Mardjo (2019). However, even with a reported high perceived risk on mobile social media, the question concerning it scored high (3.236) in terms of its kurtosis value. This leptokurtic kurtosis would imply that such an inference should be treated with caution due to the responses having a high abnormality from the normal distribution of the answers. The relationship itself between perceived security, perceived risk and trust by Harris, Brookshire and Chin (2016) cannot be confirmed in this study. Nevertheless, with the reported low levels of trust in mobile applications as compared to computer applications (Chin, et al., 2012), a similar observation can be made once again in the context of mobile social media platforms that trust is indeed low, in the case of disregarding convenience features of trust.
6.2 Knowledge
Previous studies had found low levels of knowledge to result in high levels of concern (Smit, Van Noor and Voorveld, 2014; Aguirre, et al., 2016) and we were able to confirm these results in mobile social media platforms with a significant negative relationship between knowledge and concern.
