The Privacy Paradox on Instagram: The Discrepancy Between Privacy Attitudes, Behaviors and Intentions on Instagram
Josefina Ramírez González #11804637 Master’s Thesis
University of Amsterdam Graduate School of Communications Research Master’s in Communication Science
Supervisor: Rinaldo Kühne July 19, 2019
Abstract
Social network sites, such as Instagram, are designed to motivate the constant disclosure of users’ information. While information privacy is considered as an important concern for most online users, it seems that there is a discrepancy between social media users’ concerns about keeping their personal information as private and their behavior online in terms of disclosure of this information.
Using the Ajzen’s (1991) theory of planned behavior in order to explain this dichotomy between privacy attitudes and self-disclosure behavior online, the so-called the privacy paradox (Barnes, 2006), the current study investigated the relationship between perceived privacy concerns, perceived benefits, perceived information control and subjective norm and actual self-disclosure behavior and intention on Instagram.
A different measures approaches of content analyses of 175 Instagram profiles and a survey among 175 emerging Chilean adults (age 18-25), whom are Instagram users, showed that different measurement approaches could lead to different results on the privacy paradox
hypothesis, in which privacy concerns are not associated with actual self-disclosure behavior but were found to be negatively associated with self-disclosure intention.
Other significant factors were included in the models. In the self-disclosure behavior model, no association between perceived benefits, subjective norm or perceived information control was found. In the self-disclosure intention model, Perceived benefits was found to be positively associated with self-disclosure intention, and perceived information control was found to be negative associated with self-disclosure intention, but no association with subjective norm was found.
Introduction
Social network sites (SNSs), such as Instagram, have become very popular among teenagers and young adults in recent years. These social media platforms are designed to elicit the constant update of information in the user’s profile in order to yield the full functionality of the platform, but they don’t provide adequate privacy controls (Blank, Bolsover & Dubois, 2014).
Previous research has shown that internet users noted that they were concerned about their privacy online, specifically on SNSs, but their concerns don’t match their behavior online in terms of disclosure of information (e.g. Taddicken, 2014; Tufekti, 2008; Acquisti & Gross, 2006; Debatin at al., 2009), so despite their concerns regarding possible threats while sharing personal information online they keep disclosing it, this phenomenon is also known as the ‘privacy paradox’ (Barnes, 2006). The discrepancy between attitudes toward privacy and Self-disclosure online is the focus of the current study, that is, the privacy paradox hypothesis (Barnes, 2006).
So far, however, there is a lack of consensus in previous research regarding this
dichotomy. Different levels of measurements of key variable Self-disclosure, either measured as intention, self-reported behavior or actual behavior, as well as the different research areas in which the privacy paradox has been investigated, including e-commerce and SNSs were proposed to be the major reason (Kokolakis, 2017).
Previous research has focused on different forms of Self-disclosure online, either applied to e-commerce (e.g. Dinev & Hart, 2006) or SNSs with special attention on Facebook (e.g. Dienlin & Trepte, 2015), but research on the discrepancy between privacy attitudes and behavior online has never been investigated on Instagram, the fast-growing social media site (Blank and Lutz, 2017) in which young adults and teenagers have become the biggest group of users
(Clarke, 2019) and because of its image-based and privacy settings functionality different than Facebook, it can have effects on user’s privacy perceptions (Shane-Simpson et al, 2018) .
Due to this lack of consensus, this different method approach study will consider not only Self-disclosure intention on Instagram but also actual Self-disclosure behavior through the content analysis of Instagram profiles of the same respondents that participated in the survey to see if this different measurement could have affected the results regarding the privacy paradox hypothesis.
This study aims to examine the discrepancy between privacy concerns and self-disclosure intention and actual behavior on Instagram. Consequently, some factor that might affect self-disclosure were considered as perceived privacy concerns, perceived benefits, perceived information control, and subjective norm. Also, the role of age, gender and Instagram use are examined.
It is argued that it is necessary to take into consideration the level of measurement of the key variable Self-disclosure, because the privacy paradox still exist on Instagram when actual Self-disclosure behavior is measured.
The Particularity of Instagram as a Social Media Platform
As far as is known, there are a few studies that have focused on information disclosure topics on Instagram such as the motivation for information Self-disclosure on Instagram (Al-Kandari et al., 2016) but none of this research has been done regarding the privacy paradox, in which the focus was Facebook (e.g. Krasnova et al., 2010).
Instagram is now positioned as the fastest growing social media site (Blank & Lutz, 2017), in which 15% of the users are women and 17% men between 18-24 years old, having this
cohort the largest penetration (Clarke, 2019). In the U.S the Snapchat and Instagram are
especially popular among young adults, in which 71% of users are between18 and 24 years old, however, despite of the broad popularity, only 3% of Americans trust in this social media sites (Smith & Anderson, 2018).
Instagram is characterized for being easy for mobile use, restricting its full functionality in desktop format and its image-based and filter-manipulated nature of most data (Blank & Lutz, 2017; Hu, Manikonda & Kambhampati, 2014; Khillar, 2019). The image in the form of picture or video has centrality over the text, differentiating it from other similar SNSs such as Twitter or Facebook.
Regarding functionality, the Instagram is composed by followers and people that user follows, and they don’t have to be the same, different than Facebook that has reciprocated connections, which are characterized by being close communities, but Instagram incentives the user to join other communities that share common interest (Khillar, 2019).
Also, users can set their privacy preferences by making their profiles public or private in which the user must give permission accepting a follow request from another user, but not setting the privacy within each publication like on Facebook. Thus, Shane-Simpson et al. (2018) study suggested that people’s SNSs preference is guided by privacy concerns, for example, users with high privacy concerns might prefer Facebook over Instagram due to its customizable settings.
Thus, the non-flexible settings to internally control the information within Instagram compared to other similar SNSs such as Facebook (Shane-Simpson et al., 2018), non-matching connections and the incentive to meet share communities outside the close circle can have an impact in the way users perceive that their information is in risk or not (Lee et al., 2013).
Given all this functionalities and popularity of Instagram’s image-based platform among young adults, it is relevant to investigate the phenomenon of the discrepancies between user’s privacy concerns and self-disclosure behavior in this SNS.
Privacy on SNSs
Information privacy. The functionality of new media facilitates the sharing of content uploaded on the internet. In this sense, SNSs have special characteristics that can represent a threat to privacy for users: persistence of the content once uploaded online, searchability of the information, exact copyability of the content and invisible audiences (Boyd, 2007).
This means that if the user decides to upload content it can’t be permanently erased from the platform. Also, search websites or the SNS itself allows other users to trace this content. Moreover, this content can be duplicated as many times as the user wants and this content can be reached – unintentionally – by other audiences that the user is not aware of.
This has led scholars to pursue a narrow definition of the concept of privacy in an online context in contrast to offline contexts, in which the user has less control over the information after is disclosed (Walther, 2011).
In order to narrow down the definition of privacy, this study will refer to the concept of information privacy, which is one of the spheres of privacy (Ginosar & Ariel, 2017). In online or offline contexts, information privacy refers to “the claim of individuals, groups, or institutions to determine themselves when, how, and to what extent information about them is communicated to others” (Westin, 1967, p.7, cited by Loosen, 2011). This definition highlights the right of the individual to select the personal information that is going to be available to others.
In SNSs, information privacy refers to “what information will be made available in which way, to whom, when, and to what extent” (Ziegele & Quiring, 2011, p.180). The difference here is that in SNSs autonomously determine, control and restrict (Ziegele & Quiring, 2011) this private information to others becomes more difficult in SNS interactions.
For example, in an offline context, is easier to see which people access to the information that is being shared. In contrast, the user can restrict the private information to others in their network, but still this information can be reached by other audiences and be stored by the SNS itself.
Besides, it should be considered that the information could not be about the user itself, but also about another individual (Boyd, Golder, & Lotan, 2010; Naaman, Boase, & Lai, 2010 cited by Humphreys, 2014), making it even more difficult to handle the information privacy if the content can be shared by other audiences.
Private information. Scholars have agreed that the digital traces of people’s ideas and beliefs could be considered more private than basic personal information such as name, location or phone number (Papacharissi, 2010 cited by Humphreys et al., 2014), including this nuances by differentiating between different forms of self-disclosure in terms of basic, factual and sensitive information (Taddicken, 2014).
However, it is difficult to define what is private because what is considered as private, is individually appraised (Kokolakis, 2017). For example, Humphreys et al. (2014) explained that several studies found that users often uploaded posts in SNSs that they don’t consider
particularly private, while the content itself is about personal information about the user (e.g. location).
A good solution to the problem of the definition of private information was stated by Mothersbaugh (2012), that argued that previous studies failed to account for sensitivity of information and included in his study about the privacy paradox the perceived information sensitivity (low vs. high) as moderator in the relationship between online self-disclosure and information privacy behavior.
In a study about disclosure management strategies in SNSs, Masur and Scharkow (2016) defined what is private information from the user perspective, presenting different pieces of information (e.g. political beliefs, hobbies) that respondents assessed in a scale of privacy. Thus, they found that the higher users perceived a piece of information as private, the less frequently they intended to share it on SNSs.
Privacy Paradox. Scholars found a specific phenomenon called the privacy paradox (Barnes, 2006) described as “the relationship between individuals’ intentions to disclose personal information and their actual personal information disclosure behaviors” (Norberg, Horne & Horne, 2007, p.1). Some studies have shown that their information disclosing behavior does not line up with their attitudes toward privacy (Kokolakis, 2017; Gerber et al., 2018), specifically, online Self-disclosure (Kokolakis, 2017).
Thus, Self-disclosure is the process of communication about the self to other persons (Wheeless & Grotz, 1976 cited by Tadei & Contena, 2013), and it closely related with the private information that we decide to make available to others.
When it comes to research of the privacy paradox in SNSs, it is important to distinguish between the source from where privacy concerns are raised (Young and Quan-Haase, 2013; Masur & Scharkow, 2016). There are different types of privacy concerns, such as social threats (Young & Quan-Haase, 2013) that is a risk coming from other users (e.g. unwanted audiences,
bullying, stalking) or institutional (Young & Quan-Haase, 2013) or organizational threats (Krasnova et al., 2009a; Ginosar & Ariel, 2017), that is the unauthorized use, collection and sharing of private data by organizations and third-parties.
On the other hand, this discrepancy between privacy attitudes and behaviors online has not been fully explained. While there are plenty of studies regarding the privacy paradox on SNSs, contradicting results are not providing a complete explanation about the phenomenon (Kokolakis, 2017; Ziegele & Quiring, 2011) in which there is evidence supporting and challenging the dichotomy on SNSs (Kokolakis, 2017).
Regarding evidence supporting the existence of a discrepancy between privacy attitudes and behavior, Lee et al.’s (2013), in an experimental study found that users disclose personal information on SNSs despite their privacy concerns because the process of sharing information online is a risk-benefit assessment between expected benefits and risks online.
Likewise, Acquisti & Gross (2006) early study on college students found that privacy concerns of SNSs user can’t predict user’s actual self-disclosure behavior and membership on Facebook.
Also, Tufekci (2008) survey study on college students and Self-disclosure on Facebook and MySpace, found no relationship between online privacy concerns and information
disclosure. Similarity, Taddicken’s (2014) study of German users in the Social web found that privacy concerns did not have an impact on Self-disclosure self-reported behavior, but this relationship is moderated by the perceived social relevance and the number of applications used.
Likewise, Cheung et al. (2015) did not find an association between privacy risk and Self-reported self-disclosure attitudes, even controlled by other factors. Taddei & Contena (2013) did not find a direct effect of perceived concerns over self-disclosure.
On the other hand, there is evidence also challenging the privacy paradox hypothesis. Young and Quan-Haase (2013) argued that this paradox does not exist, because Facebook users self-manage their privacy employing protection strategies against possible institutional or social threats.
Zlatolas et al. (2015) and found that privacy concerns have a negative impact on self-disclosure on Facebook and Joinson et al., (2010) found that there is a negative association between privacy concerns and online self-disclosure intention (willingness), similarly to Dienlin and Trepte (2015) that found an indirect relationship between information, social privacy
concerns and self-disclosure, mediated for self-disclosure intentions.
Also, Tsay-Vogel et al. (2018) in their 5-years longitudinal study regarding privacy attitudes on Facebook users concluded that there is a relationship between online privacy concerns and online self-disclosure on SNSs, but this association has weakened over the time.
A plausible cause of these inconsistent findings is due to the differences in the operationalization and measurement of privacy behavior, specifically online Self-disclosure (Kokolakis, 2017).
Privacy paradox studies on SNSs have measured Self-disclosure intention (Dienlin & Trepte, 2015; Joinson et al., 2010; Hallam and Zanella, 2017) and Self-disclosure self-reported behavior (Taddicken, 2014; Tsay-Vogel et al., 2018; Becker & Pousttchi, 2012; Tufekci, 2008; Cheung et al., 2015; Zlatolas et al., 2015; Taddei & Contena, 2013).
However, few studies have measured actual behavior, in which Young and Quan-Haase (2013) analyzed student’s Facebook profiles together with in-depth interviews and Lee et al. (2013) in which participants shared information in a realistic experimental setting on SNS,
together with Acquisti and Gross (2006) that compared the survey data with the information retrieved from Facebook and MySpace profiles from the same participants.
In light of this, while behavior and intention are closely related in the sense that in order to perform certain behavior the individual had to have the intention to do it (Ajzen, 1991), they are not the same, and people can have the intention to act upon something but never perform the behavior.
As we can see, most of the studies that relied on self-reported or actual Self-disclosure behavior on SNSs confirmed the privacy paradox hypothesis, while studies measuring self-disclosure intention, did not find this discrepancy between privacy attitudes and behaviors. In this sense, Hallam and Zanella (2017) found privacy concerns affect distant-future self-disclosure intentions but don’t affect sensitive information self-self-disclosure behaviors.
In the same line, relying results on self-reporting behavior is not the same as measuring the actual behavior (Baumeister, Vohs & Funder, 2007) and can affect the results while
measuring self-disclosure (Nguyen, Bin & Campbell, 2012). Giving the fact that online behavior can be traceable (Boyd, 2007), it seems possible to think that we can observe the actual behavior instead of relying results on self-reported behavior.
For this reason, this study is going to measure self-disclosure intention and actual behavior, in an attempt to evaluate the impact of the different measures approaches of the dependent variable in the results of this study. Given this, the following research questions and hypothesis are proposed:
RQ1: What is the relationship between privacy concerns and Self-disclosure on Instagram among emerging adults?
RQ2: Does the different measurement of Self-disclosure have an impact on the
relationship between privacy concerns and Self-disclosure on Instagram among emerging adults?
H1a: There is no association between the user’s perceived privacy concerns and actual Self-disclosure behavior on Instagram
H1b: There is a negative association between the user’s perceived privacy concerns and Self-disclosure intention on Instagram
Theory of Planned Behavior: Explaining the Privacy Attitude and the Behavior discrepancy
The Theory of Planned Behavior (Ajzen, 1991) is an extension the theory of reasoned action (Ajzen, 2012) and postulates that intentions to perform behaviors can be determined from the attitude toward the behavior (the degree of favorable or unfavorable evaluation of the behavior), subjective norm (perceived social acceptance to perform the behavior) and perceived behavioral control (perceived self-efficacy, degree of difficulty to perform the behavior). Thus, each one of the factors in the model would be crucial to understand that intentions are the antecedent of behaviors.
In general, the more favorable the attitude toward behavior and subjective norm, in addition to greater perceived control, the intention to perform certain behavior would be
stronger, so the individual will perform the intended behavior when the opportunity is presented (Ajzen, 1991)
Theory of Planned Behavior has been applied in some studies about motivation of self-disclosure in SNSs (Shibchurn & Yan, 2015; Zhou & Li, 2014), and also has been applied in some studies in an attempt to explain the discrepancy between privacy attitudes and
Self-Disclosure behavior online (Dienlin & Trepte, 2015; Kim et al., 2016), or integrating these concepts in order to explain which factors affect self-disclosure behavior on e-commerce and SNSs (Xu et al., 2013).
In the following, I am going to explain how the different aspects of the TPB – attitude toward the behavior, subjective norm and perceived behavioral control – can be used to give a plausible explanation of the phenomenon regarding the discrepancies between privacy attitudes and behavior on SNSs.
Attitudes toward the behavior.
According to TPB, if a person has more favorable attitudes towards certain behavior the chances that they perform it increases (Ajzen, 1991). Important here is the idea that in TPB the behavior is guided by intentions, and this assumption means that there is a strong correlation between attitudes and behaviors (Ajzen, 2012). In this case, an individual’s actions are guided about the beliefs about the evaluations of the outcomes of a certain behavior, in which attitudes would be an antecedent for behaviors (Ajzen, 1991).
Perceived concerns and benefits of Self Disclosure on SNSs.
Regarding the attitudes towards online Self-disclosure behavior, as we mentioned one of the privacy attitudes toward self-disclosure is privacy concerns of the user. In an attempt to explain the discrepancy between privacy concerns and online self-disclosure behavior, some scholars (Lee et al., 2013; Xu et al., 2013; Barth & De Jong, 2017) argued that this privacy paradox can be explained if we see the attitudes as an evaluation between costs (e.g. privacy concerns) and benefits (e.g. benefits of posting on SNSs, rewards), explaining the privacy paradox with a privacy-calculus approach, in which the discrepancy between privacy concerns and
self-disclosure could be explained if considers the perceived benefits of SNSs use (Becker & Pousttchi, 2012) as a more accurate predictor of self-disclosure.
Thus, regarding the perceived benefits of self-disclosure, research on attempting to explain the privacy paradox and research attempting to explain the motivation of self-disclosure (e.g. perceived usefulness of SNSs) that relied on the Uses and gratifications theory (Al-Kandari et al., 2016; Shibchurn & Yan, 2015) concluded that the perceived possible benefits of using SNSs on of the best predictors for self-disclosure on the Internet (Gerber’s et al., 2018)
For example, some studies found that perceived benefits of Self-disclosure in SNSs have found to be a strong predictor on actual posting behavior (Lee et al., 2013), self-disclosure intention (Xu et al., 2013), and self-reported Self-disclosure on Facebook (Krasnova et al., 2009b) and Instagram (Al-Kandari, 2016).
Cheung et al., (2015) focused both motivations of use and self-disclosure on SNSs, and identified four major types of perceived benefits of use and self-disclosure in SNSs: convenience of maintaining existing relationships, new relationship building (create new ties),
self-presentation (self-impression management) and perceived enjoyment (entertainment). They found each one of these 4 dimensions that constitute perceived benefits are significant factors that determine self-disclosure in SNSs.
In this sense, taking in consideration that the perceived benefits and privacy concerns can be attitudes that were associated with Self-disclosure on SNSs, together with previous statements about attitudes and behaviors, is proposed:
H2a: User’s perceived self-disclosure benefits in Instagram will be positively related to their actual Self-disclosure behavior on Instagram
H2b: User’s perceived self-disclosure benefits in Instagram will be positively related to their Self-disclosure intention on Instagram
Subjective norm.
Subjective norm refers to the social acceptability of performing certain behavior (Ajzen, 1991). In this sense, if social media users perceive that is socially acceptable by their peers to self-disclose information online, the likelihood that is behavior would occur increases (Taddicken, 2014).
Similar concepts have been used in studies regarding privacy and self-disclosure online. Zlatolas et al. (2015) refers to privacy social norm as a concept regarding “how people and friends influence the user into keeping their information privacy” (Zlatolas et al., 2015, p.162).
Regarding the influence of subjective norm in SNSs attitudes and behavior, the concept of subjective norm conceptualized as social influence, was found to be associated with people’s disclosure intentions in another study (Shibchurn & Yan, 2015).
Other studies, found that social influence (Cheung et al., 2015) and social relevance (Taddicken, 2014) are the variables that showed a strong effect on online self-disclosure in SNSs and Van Gool et al., (2015) subjective norms of friends and parents are predictor variables for information disclosure intention on SNSs .
Moreover, Taddicken, (2014) found that subjective norm has a negative effect on privacy concerns, because people feel would feel less concern using SNSs if they see that other
significant people also do.
Important here is to highlight that for this study, social influence and subjective norm are aiming the same concept. While social influence might be the effect of the subjective norm that people internalized as socially acceptable, that is, if people were influenced by other’s opinions, I
am going to stick to the concept of subjective norm in the sense of the evaluation of behaviors that the user considers as socially acceptable (Ajzen, 1991) by others whose opinion they value (Shibchurn & Yan, 2015) regarding use and self-disclosure on Instagram. Given, is proposed that:
H3a: There is a positive association between subjective norm regarding Instagram use and actual Self-disclosure behavior on Instagram
H3b: There is a positive association between subjective norm regarding Instagram use and Self-disclosure intention on Instagram
H4. There is a negative association between subjective norm and perceived privacy concerns on Instagram
Perceived control over information privacy.
According to Ajzen (2012) perceived behavioral control (PBC) is conceptually comparable to Bandura’s (1977, cited by Ajzen, 2012) perceived self-efficacy term, which determines how people motivate themselves and behave, and defined as individuals’ self-perception about their own ability to perform certain behavior based on perception of how easy or difficult will be for them to perform a behavior.
Perceived control might be important to explain the discrepancy between the user’s privacy concerns and self-disclosure behavior online. Brandimarte et al.’s (2013) experimental study concluded that perceived control over content affects information sharing online, in the sense that users with the sense of more control over personal information will be likely to share more personal data because the user might have a false idea of being in control of the personal data.
Likewise, research found that the perceived ease of privacy control is a predictor variable for information disclosure on SNSs among teenagers (Jia et al., 2015) and young adults (Chen & Chen, 2015), and Taddei & Contena (2013) found that there is a significant effect of control over the information mediated by the trust over self-disclosure online.
Apart from influencing self-disclosure, privacy control can influence privacy concerns. Some studies found evidence that the individual’s perceived control over their information is an important factor affecting privacy concerns in SNSs (Xu et al., 2013; Jia et al., 2015). Likewise, Zlatolas et al. (2015) found that privacy control on information has a negative impact over privacy concerns on Facebook. Therefore, this study proposes the following hypotheses:
H5a: There is a positive association between the user’s perceived information control over privacy on Instagram and actual self-disclosure behavior on Instagram
H5b: There is a positive association between the user’s perceived information control over privacy on Instagram and self-disclosure intention on Instagram
H6: User’s perceived control over privacy on Instagram will be negatively related to their perceived privacy concerns on Instagram
All hypotheses are summarized in Fig.1 and 2.
Fig.1. Hypothesized privacy paradox model with actual self-disclosure behavior as dependent variable.
Fig.1. Hypothesized privacy paradox model with self-disclosure intention as dependent variable. Methods
Sample and procedure
The selected method for this study is different methods approach of self-reports and content analysis that allows measuring the dependent variable Self-disclosure in terms of self-reported intention and actual behavior observed among Chilean emerging adults’ Instagram profiles.
Self-report.
This research unit of this study is users of Instagram in Chile from 18 to 25 years old. Social media penetration in Chile was 77%, being Instagram the second most popular SNS with 7,3 million users, in which 53% are females and 71% males, in which users from 18 to 24 years old represent the second biggest group of users (We are Social & Hootsuite, 2019).
An online survey using Qualtrics platform was conducted in May and June 2019 together with a Content analysis of participant’s Instagram profiles conducted in June 2019. A
convenience sampling was used to recruit the participants, in which participants were recruited posting a call of participation in different Facebook groups and in the special Instagram account created to recruit participants named @chillenial_pavre, with about 4000 followers.
Recruitment yielded 362 respondents, however, only 321 completed the survey.
with inclusion criteria questions, based on Nationality (Chileans), Age (18-25 years old) and Instagram Use (Use Instagram at least once in the last 3 months). 21 participants were excluded either because they did not meet inclusion criteria either for nationality, age or Instagram Use.
Participants that matched the inclusion criteria and agreed to participate were asked to respond the online survey with 11 questions: account’s access, perceived concerns of Self-disclosure on Instagram, perceived benefits of Self-Self-disclosure on Instagram, Subjective Norm on Instagram use and disclosure, Perceived information control on Instagram, Perceived sensitivity of personal identifiable information, Self-disclosure intention on Instagram, gender, frequency of Instagram use, level of activity on Instagram, frequency of posting behavior on Instagram. The full questionnaire can be found in Appendix A.
At the end of the survey, participants were asked to provide their Instagram account in order to be able to participate for the incentives (Raffle of 3 gift cards) and also to have access to their Instagram profiles through the account created especially for the study named
@proyecto_uniamsterd, which would be deleted after the end of the study, and created to perform a content analyses of their Instagram profiles. Only 246 participants provided their Instagram account at the end of the survey, and the others were excluded from the study.
Finally, participants were followed to the Instagram account of the study, to grant access to their profiles and match their responses in the survey with the data collected from their profiles. Only participants that granted access to their profile by accepting to be followed by @proyecto_uniamsterd were included. The total sample consisted of 175 participants (34.9% males; 65.1% females) age 18-25 (M = 21.2, SD = 1.9).
Content-Analysis.
Instagram profiles were selected as code-units. The selection criteria for Instagram profiles were users that responded the survey was and were regular Instagram users, as those who are 1) non-personal accounts: not organizations, brands, or spammers, and 2)follow at least 30 other
accounts, has at least 30 followers (Hu et al., 2014) , no profiles were excluded due these criteria. For purposes of this study, the user doesn’t have actively update content in their profile, that is why this definition is not limited by the number of contents posted in the last year, as the original Hu et al.’s (2014) definition.
Also, for purposes of this study non-personal accounts were selected based on the description and content posted on their profiles, in which they explicitly said they were not a personal account. For example, and a non-personal account can be an online store, a meme account, a brand account, a university account, etc.
A total of 175 profiles, were collected through the Instagram @proyecto_uniamsterd that followed the Instagram accounts. For each Instagram account, the description in the account’s profile together with the last 10 single posts or multi posts (videos and/or images) and their correspondent post’s descriptions published by the user in his/her profile were considered for the analyses.
Because images can be interpreted in many ways, it is necessary the addition of text/voice to give them a meaning (Barthes, 1978 cited by Andalibi, Ozturk & Forte, 2007), inspired by Andalibi et al. (2007) approach, I analyzed visual and textual content, instead of studying them apart.
A codebook was created to measure the following variables in this order: Coder ID, Instagram’s account name, user’s account access status, use of Hashtags, third-party access to
user’s account, publications, following, followers, content categories and personal identifiable information. Full codebook can be found in Appendix B.
Measures
Coding scheme: Content Analytical measures.
Account access status. To see if the user granted access to their content and personal information to all other users or is restricted to their network (Taddicken, 2014; Tufekci, 2008)
Use of hashtags. In the post and/or description, which was included to see if the user published content that could be hyperlinked to a hashtag, and therefore, seen by other users that are not part of his/her network.
Third-party access to user’s account. Which was included to see if the participant granted access to non-personal accounts.
Number of publications. Which was included to see how many posts the user disclosed in the profile.
Number of followers. which was included to see how much people the user granted access to their account. This variable was included as demographic in other similar studies (Zlatolas et al., 2015)
Number of following. General measurement of the profile.
Content Categories. Following Hu et al.’s (2014) content analyses of Instagram’s
pictures, I also coded for the presence or absence of friends, Food, Gadget, Captioned photo, Pet, Activity, Selfie, Fashion.
Personal identifiable information. To measure the level of self-disclosure I coded the personal identifiable information of the user, defined as “information that could be directly tied
to or associated with an individual such as email, phone number, or address” (Humphreys et al., 2014, p.849) with dichotomous responses (0=absence, 1=presence).
The most categories selected were based on the personal information categories that measured Self-disclosure on SNSs defined and used by previous research or categories defined as sensitive information, which was also included in studies of Self-disclosure (Taddicken, 2014).
For each participant, the description on their account profile and the last 10 pictures and/or videos posted in their account’s profile were analyzed to find the presence or absence or the following 23 pieces of personal information: name (Acquisti & Gross, 2006; Taddicken, 2014; Tufekci, 2008; Shibchurn & Yan, 2015; Dienlin & Trepte, 2015), last name (Taddicken, 2014; Dienlin & Trepte, 2015), age (Shibchurn & Yan, 2015) and or/birth date (Acquisti & Gross, 2006; Taddicken, 2014), school/university/workplace (Shibchurn & Yan, 2015; Mainier & O’Brien, 2010), profession/career (Taddicken, 2014; Shibchurn & Yan, 2015), relationship status (Tufekci, 2008; Shibchurn & Yan, 2015; Mainier & O’Brien, 2010), sexual orientation (Acquisti & Gross, 2006), religion that the user profess (Tufekci, 2008; Shibchurn & Yan, 2015; Dienlin & Trepte, 2015), political beliefs (Acquisti & Gross, 2006; Shibchurn & Yan, 2015), interests and activities (Taddicken, 2014; Tufekci, 2008; Shibchurn & Yan, 2015; Mainier & O’Brien, 2010), moods and personal feelings (Taddicken, 2014; Shibchurn & Yan, 2015), phone number (Acquisti & Gross, 2006; Tufekci, 2008), e-mail address (Taddicken, 2014), another social media account link, home address (Acquisti & Gross, 2006; Tufekci, 2008; Dienlin & Trepte, 2015), location (Humphreys et al., 2014) , nationality, and sensitive information (Tourangeau & Yan, 2007) such as sexy/sexual pictures/videos of the user, sexy/sexual pictures/videos of others, pictures/videos of the user consuming drugs/alcohol and
pictures/videos of others consuming drugs/alcohol. Also, the categories ‘selfie’ (images of self) and ‘friends’ (images of others) (Taddicken, 2014; Hu et al., 2014; Shibchurn & Yan, 2015) from the content categories variable were included as pieces of personal information.
Descriptive statistics and description of all coding variables can be found in Appendix C. Krippendorff’s alphas were calculated for intercoder reliability for each variable, in which another trained Chilean coder coded 10% of the sample (see Appendix C).
Self-report
Account access status. To assess the privacy settings in the participant’s Instagram accounts following the Shane-Simpson et al. (2018) and Taddicken’s (2014) approach measuring the level of privacy of the user’s account, participants were asked if their account was public (Any other user can have access to their profile) or private (only their followers can see their profile). 61.1% have a public account and 37.7% has a private account.
Perceived concerns of Self-disclosure. To measure the perceived information privacy (Ziegele & Quiring, 2011) institutional and social concerns regarding information disclosure on Instagram, Shibchurn & Yan’s (2015) scale of privacy concerns scale on social media was adapted for Instagram statements.
This measurement was presented in two different questions. The first scale was regarding institutional/organizational privacy concerns (Young & Quan-Haase, 2013; Ginosar & Ariel, 2017) and incorporated statements about Chilean government, companies and third parties (“I am worried that Instagram may sell my personal information to third parties”) as sources of these concerns. The second scale was related to social privacy concerns (Young & Quan-Haase, 2013) and incorporated statements regarding privacy threats from other users (“I am worried that my personal information could be stolen from Instagram by other users”).
Participants rated the degree to which they agreed with statements on a scale ranging from 1 (totally disagree) to 5 (totally agree). The 9 total items were summed to create a total score with higher scores reflecting a higher degree of privacy concern. Scores ranged between 1 and 5 (M = 4.00, SD = .80). The scale has good reliability, in which Cronbach's alpha was α =.89. A Principal component analysis (PCA) revealed that only one component (“I’m worried that Instagram can sell my personal information to third-parties”) has an eigenvalue above 1 (eigenvalue 5.09), and items loaded on this one factor that explained 56.5% of the variance. All items correlate positively with the first component.
Perceived benefits of Self-disclosure. To assess the perceived benefits on online self-disclosure on Instagram, I adapted the previous measurements of perceived benefits of social media use from Cheung et al., (2015) to statements measuring specifically Instagram use.
The scale is composed mainly of 5 different benefits (Cheung et al., 2015): (1) Convenience of maintaining existing relationships (“I find Instagram efficient in sharing
information with my friends”) (2) New relationship building (“I get to know new people through Instagram”) (3) Self-presentation (“Instagram helps me to present my best sides to others”) (4) Enjoyment (“I find Instagram entertaining”). Each topic was also composed of 3 items each, given a total of 12 items. Participants were asked to what extent they agree with the statements. Answer options ranged from 1 (totally disagree) to 5 (totally agree). The 12 items were summed to create a total score regarding perceive benefits of Instagram’s disclosure. Scores ranged between 1 and 5 (M = 3.68, SD = .53). Cronbach's alpha was α =.78.
A PCA revealed that only 3 components have an eigenvalue above 1 (eigenvalue 3.7; eigenvalue 2.39; eigenvalue 1.58) and items loaded on this one factor that explained 31%, 19% and 13% of the variance. All items correlate positively with the first component (“Instagram is
useful to inform about my activities”) but not all items correlated positively with the second (“Instagram allows me to save time when I want to share something new with my friends”) and third component (“Instagram is efficient when I want to share information with my friends”). Subjective norm. Based on the original definition of subjective norm proposed by Ajzen (1991) the concept of subjective norm in Instagram and Subjective norm scale used by Shibchurn & Yan (2015) , it refers to the user’s assessment of social acceptability (Ajzen, 1991) of
information disclosure on Instagram by people who are important to them.
The scale was composed of items regarding participant’s beliefs regarding people whose opinion they value and their Instagram use (“People whose opinion I value use
Instagram”). Participants were asked to evaluate from 5-point Likert agreement scale (1=strongly disagree to 5=strongly agree). I followed the Shibchurn & Yan (2015) adapted subjective norm scale from Venkatesh and Davis (2002, cited by Shibchurn & Yan, 2015). The 4 items were summed to create a total score for subjective norm. Scores ranged between 1 and 5 (M = 3.4, SD = .66). Cronbach's alpha was α =.71.
A PCA revealed that 2 components (“people whose opinion I value would be willing to share their personal information in Instagram in exchange for a benefit”; “people whose opinion I value would advise me to share my personal information in Instagram for a benefit”) have an eigenvalue above 1 (eigenvalue 2.17; eigenvalue 1.24) and items loaded on these factors that explained 54.3%, 31% of the variance. All items correlate positively with both components.
Perceived information control. Perceived information control on Instagram was assessed using an adapted version to Instagram from Krasnova et al., (2010) and reflects the perceived control (Ajzen, 1991) over the privacy of the information that users shared on
Instagram (“I feel in control over the information I provide on Instagram”) to which participants assessed on a scale from 1 (totally disagree) to 5 (totally agree). Scores ranged from 1 to 5 (M = 3.6, SD = .82). Cronbach's alpha was α = .69.
A PCA revealed that only 1 component (“I feel in control over the information I provide on Instagram”) have an eigenvalue above 1 (eigenvalue 1.86) and items loaded on this one factor that explained 62.1% of the variance. All items correlate positively with both components.
Perceived sensitivity of personal identifiable information. To measure in what extent the individual considers certain information as sensitive in terms of personal information as Mothersbaugh et al. (2012) suggested and inspired by Masur and Scharkow (2016), participants were asked to evaluate from a 5-point scale (1= not personal at all to 5=very personal) how personal they evaluate the pieces of information. The same items used with the content analyses for personal identifiable information (name, last name, etc.) were presented to the participants.
Self-reported Self-disclosure intention on Instagram. To assess self-disclosure intention in Instagram based on Dienlin & Trepte’s (2015) and Hallam and Zanella
(2017) approach regarding general willingness of online self-disclosure, participants were asked to indicate which of the 24 pieces of personal identifiable information (used in the content analyses) they would be willing to share in their profile’s descriptions and Instagram’s account. The participant could select more than one response.
Dependent Variables
Actual Self-disclosure Behavior. Self-disclosure behavior was constructed with the personal identifiable information items retrieved from the participant’s Instagram profiles together with the variable account access status, as in Taddicken’s (2014) construct of self-disclosure and perceived sensibility of personal identifiable information (Mothersbaugh et al.,
2012). This approach takes into consideration that the information disclosed in restricted profiles is still available for followers included in the user’s network and that certain pieces of
information are perceived as more or less personal than others. A weighted sum index was constructed.
Participants that have restricted access profiles were given a score of 1 and open access one’s a 2 score. These scores were multiplied by the sum of all the personal identifiable
information (0 = absent; 1= present) which were scored together with the perceived sensitivity (1= not personal at all to 5=very personal) that the user gave to each item. Scores ranged from 2 to 80 (M = 28.5, SD = 15.8).
Self-disclosure intention index. Self-disclosure intention was constructed with 3
different indicators: self-reported Self-disclosure intention on Instagram, account’s access status as in Taddicken’s (2014) construct of self-disclosure, and perceived sensibility of personal identifiable information, as Mothersbaugh et al. (2012) suggested to take into account. A weighted sum index was constructed.
Participants that have restricted access profiles were given a score of 1 and open access one’s a 2 score. These scores were multiplied by the sum of all the personal identifiable
information (0 = absent; 1= present) which were scored together with the perceived sensitivity that the user gave to each item (1= not personal at all to 5=very personal). Scores ranged from 6 to 122 (M = 34.3, SD = 18.4).
Control variables.
Because self-disclosure behavior on Instagram may depend on Social media use (Tsay-Vogel et al., 2018; Jia et al., 2015; Al-Kandari et al., 2016), in this case, Instagram, gender and age (in
years) (Gerber et al., 2018; Blank et al., 2014), I controlled for these variables in all analyses. Age and gender were reported using multiple choice question.
Instagram use. It was constructed as sum index of the following variables (1) level of activity in Instagram, in which participants were asked “What is your level of activity on Instagram?). Responses ranged from 1 = plenty activity (I participate, I comment regularly, I updated content and photos regularly) to 4 = no activity/almost no activity, the question was adapted to Instagram use from AIMC (2018). (2) Frequency of Instagram use, in which
participants were asked: “Usually in a day, how much time do you spend on Instagram?” (1=less than 15 minutes to 7= more than 8 hours), adapted to Instagram use from AIMC (2018) in which items were inverted. (3) frequency of posting behavior, based on Dienlin & Trepte (2015), participants were asked “How frequently do you post content on your Instagram account?” in which responses ranged from 1= several times a day to 7 = never. Scores ranged between 1 and 5.6 (M=3.2; SD=.97).
The total 3 items were added to create a total score with higher scores reflecting a higher degree of Instagram use. Scores ranged between 1 and 5.6 (M = 3.2, SD = .97). A PCA revealed that only one component (level of activity in Instagram) has an eigenvalue above 1 (eigenvalue 1.90), and items loaded on this one factor that explained 63% of the variance. All items
correlated positively with this factor.
Results Descriptive
Some descriptive analyses were performed to test the normality of distribution in all the variables measured in the models. Perceived information concerns, perceived benefits, subjective norm
and perceived information control scaled indicated right bias in the distribution, and kurtosis indicated a relatively peaked distribution.
Positive skewness values in the dependent variables, gender and Instagram use indicate too many low scores, positive values in gender indicate right bias in the distribution. Positive values in Kurtosis for the dependent variable indicate pointy and heavy-tailed distribution, while the negative values in age, gender and Instagram use indicate flat and light-tailed distribution (Field, 2013)
In addition, Shapiro-Wilk test was performed to test normality of the distribution. Shapiro-Wilk test yielded that variables are not normally distributed (Ghasemi & Zahediasl, 2012; Laerd Statistics, 2018; Field, 2013). Table 2 summarizes the results:
Table 2 - Summary of skewness, kurtosis and normality test for variables in the models (N = 175).
Skewness Kurtosis Shaphiro-Wilk test
Privacy Concerns -1.25 1.96 0.92* Perceived Benefits -0.68 2.04 0.96* Subjective Norm -0.41 1.53 0.96* Perceived Control -0.47 0.36 0.95* S.D behavior 1.50 3.32 0.89* S.D intention 0.95 0.44 0.92* Age 0.13 -0.81 0.95*** Gender -0.64 -1.60 0.98* Instagram use 0.07 -0.49 0.60*** Note. p <.05. ** p <.01. *** p <.001. Hypothesis Testing
A data analysis was performed with the use of IBM SPSS Statistics 20.0. a multiple regression modeling was used, and hypothesis testing was done.
Actual Self-disclosure behavior model.
Before the analysis, regression analysis assumptions were tested. The actual Self-disclosure behavior model does not met the assumption of independent errors, but independent residuals assumption should not raise concern (Durbin-Watson value= 1.48) as the obtained value was close to 2, and in an acceptable range (Field, 2013), and also assumption of collinearity indicated that multicollinearity was not a concern (Privacy concern, Tolerance = .92, VIF = 1.06;
Perceived benefits, Tolerance= .77, VIF =1.29; Subjective norm, Tolerance= .86, VIF= 1.15; Perceived control, Tolerance = .96, VIF= 1.03) (Field, 2013). P-P plot for this model suggested residuals are not normally distributed, violating this assumption and the scatterplot of
standardized residuals showed that the model met the assumptions of homoscedasticity (Field, 2013).
To test hypothesis 1a that stated that there is no association between privacy concerns and actual Self-disclosure behavior on Instagram (H1a) and hypotheses that stated a positive
association between Perceived benefits (H2a), Subjective norm (H3a) Perceived information control (H5a) and Self-disclosure behavior on Instagram, a multiple linear regression analysis test was conducted.
The regression model with actual Self-disclosure behavior as dependent variable and perceived privacy concerns, perceived benefits, perceived subjective norm and perceived
information control on Instagram, including Instagram use, age and gender as control variables is not significant, F (7, 167) = 1.22, p =.293.
The analysis yielded that there is no significant association between perceived privacy concerns, b* = .03, t= 0.48, p =.628, perceived benefits, b* = -0.04, t=-0.52, p =.600, subjective norm b* =0.11 , t=1.35, p =.179, perceived information control b* =-0.04 , t=-0.56, p =.573 and the dependent variable actual Self-disclosure behavior. For all these effects Instagram use, gender and age are assumed to be held constant.
Self-disclosure intention model.
Likewise, regression analysis assumptions were tested for this model. The Self-disclosure intention model does not meet the assumption of independent errors (Durbin-Watson value= 1.42) but this value does not cause concern as the obtained value was close to 2, and in an acceptable range (Field, 2013). In addition, result for previous results for assumptions regarding multicollinearity, normally distributed errors and homoscedasticity could be applied for this model too (Field, 2013).
To test hypothesis 1b that stated that there is a negative association between privacy concerns and self-disclosure intention on Instagram and hypotheses that stated a positive association between Perceived benefits (H2b), Subjective norm (H3b) Perceived information control (H5b) Self-disclosure intention on Instagram, a multiple linear regression analysis test was conducted.
The regression model with Self-disclosure intention as dependent variable and perceived privacy concerns, perceived benefits, perceived subjective norm and perceived information control on Instagram, including Instagram use, age and gender as control variables is significant, F (7,167) = 2.80, p =.009. The regression model can therefore be used to predict self-disclosure intention on Instagram, but the strength of the prediction is low: 10% of the variation in
self-disclosure intention can be predicted based on privacy concerns, perceived benefits and perceived control (R2 = .10)
Results yielded a significant negative association between perceived privacy concerns and Self-disclosure intention on Instagram, b* =-0.20, t=-2.70, p <.05, CI [-7.01, -1.10]. On the other hand, perceived benefits, b* =0.17 , t=2.12 , p <.05, CI [0.36, 10.1], is significatively associated with Self-disclosure intention on Instagram, and contrary to H5b, there is a negative significant association between perceived information control, b* =-0.14 , t=-1.92 , p <.05, CI[-5.59, 0.06], and Self-disclosure intention on Instagram. No significant association was found between Subjective norm, b* =0.02, t=0.30, p =.764, CI [-3.14, 4.27], and Self-disclosure intention on Instagram.
A summary of all results is presented in Table 3:
Table 3 -Summary of multiple regression models predicting Self-disclosure on Instagram (N = 175).
Actual behavior model Intention model
Variable b SE b b* b SE b b* Constant 38.3 22.4 47.1 18.7 Privacy concerns 0.86 1.79 0.03 -4.05 1.49 -0.20* Perceived benefits -1.55 2.96 -0.04 5.52 2.47 0.17* Perceived control -0.96 1.71 -0.04 -2.76 1.43 -0.14* Subjective norm 3.03 2.24 0.11 0.56 1.87 0.02 Age -0.18 0.74 -0.01 -0.29 0.62 -0.03 Instagram Use -3.00 1.58 -1.15* -0.74 1.32 -0.04 Gender 2.94 3.05 0.07 -3.02 2.55 -0.09
R2 0.04 1.22 0.10 2.80* F Note. * p <.05. ** p <.01. *** p <.001. Hypothesis 4 and 6.
Regression analysis assumptions were tested. The model met the assumption of
independent errors (Durbin-Watson value= 2.0) as the obtained value is 2 (Field, 2013), and also assumption of collinearity indicated that multicollinearity was not a concern (Subjective norm, Tolerance= .92, VIF= 1.07; Perceived control, Tolerance = .98, VIF= 1.01) (Field, 2013). P-P plot for this model suggested residuals are normally distributed and the scatterplot of
standardized residuals showed that the model met the assumptions of homoscedasticity (Field, 2013)
To test hypothesis 4 that stated a negative association between subjective norms and perceived privacy concerns on Instagram and Hypothesis 6, that stated a negative association between perceived information control and perceived information privacy concerns on Instagram, a multiple regression analysis was conducted.
The regression model with privacy concerns as dependent variable and perceived privacy perceived subjective norm and perceived information control on Instagram, including Instagram use, age and gender as control variables is not significant, F (5,169) = 1.69, p =.138.
Results yielded no significant association between subjective norm, b* =-0.01 t=-0.19, p =.849, CI -0.20,0.16], perceived information control b* =-0.12, t=-1.66, p =.099, CI [-0.26, 0.02] and perceived privacy concerns on Instagram. For all these effects Instagram use, gender and age are assumed to be held constant.
Discussion and Conclusion
The present study was conducted to examine in the privacy paradox hypothesis (Barnes, 2006) on Instagram that states that there is a discrepancy between privacy attitudes and behaviors by analyzing the relationship between perceived privacy concerns and Self-disclosure actual behavior and intention on Instagram. Using Ajzen (1991; 2012) theory of planned behavior, other significant factors that might be associated with Self-disclosure behavior and intention including perceived benefits, subjective norm, and perceived information control were included, and factors that might be associated with privacy concerns, as perceived information control and subjective norm.
Besides, this study postulated that different measurements of the outcome variables self-disclosure would lead to different results regarding the relationship between privacy concerns and self-disclosure.
The actual self-disclosure behavior model showed that there is no association between privacy concerns and actual Self-disclosure behavior, even controlled by other factors. These results are in line with previous studies that find evidence supporting the privacy paradox hypothesis on SNSs (Tadicken, 2014; Tufekci, 2008; Cheung et al., 2015; Taddei & Contena, 2013). Particularity Lee et al.’s (2013) experimental controlled setting, and Acquisti & Gross (2006) survey and data retrieved from the same participant’s profiles that measured actual self-disclosure behavior, that found that user’s privacy concerns do not predict self-self-disclosure behavior, as the logic might suggested.
The Self-disclosure intention model showed that there is a negative association between privacy concerns and Self-disclosure intention on Instagram. These results that challenge the privacy paradox hypothesis on SNSs are in line with similar studies (Young & Quan-Haase,
2013; Tsay-Vogel et al. 2018; Zlatolas et al., 2015), particularly with studies that measured the outcome variable as Self-disclosure intention (Dienlin & Trepte, 2015; Joinson et al., 2010; Hallam and Zanella, 2017), that found that user’s that are more concerned about their privacy will significantly be less willing to disclose information on their SNSs.
Contrary to the expectations (Lee et al., 2013) no association between perceived benefits and actual Self-disclosure behavior was found. A possible explanation for these results is the way perceived benefits were measured. Previous studies that found significant results measured the perceived benefits in terms of need and motives of using Instagram (Al-Kandari et al., 2016), including motives like self-expression that weren’t included in this study’s scale and which was directly related with information sharing on Instagram, perceived usefulness of SNS (Shibchurn & Yan, 2015) or just measuring the construct in an unidimensional manner as is perceived enjoyment (Krasnova et al., 2009b), which is not the same that measuring perceived benefits of SNSs sharing.
However, in line with it was initially proposed, this study found a positive association between perceived benefits and self-disclosure intention, in line with previous studies that found an association (Lee et al., 2013; Xu et al., 2013; Krasnova et al., 2010; Al-Kandari et al, 2016; Cheung et al., 2015; Gerber et al., 2018).
Contrary to the expectations (Brandimarte et al., 2013, Jia et al., 2015; Chen & Chen, 2015) it was found that there is a negative association between the Instagram user’s perceived information control of the information shared, and self-disclosure intention and no association between perceived control and actual self-disclosure behavior. On the same line, contrary to previous findings and what it was postulated (Cheung et al., 2015; Xu et al., 2013), no
association between subjective norm and Self-disclosure intention neither actual Self-disclosure behavior was found.
Results of this study revealed that the different measurement of Self-disclosure leads to different results in the models in agreement with similar argumentations (Kokolakis, 2017). At the measurement level, as was explained, having the intention to perform a behavior is not enough to translate this intention in actual behavior (Ajzen, 1991; Ajzen, 2012). In this sense behavior is intention are not interchangeable concepts and do not measure the same.
Moreover, according to Kokolakis (2017) studies that measure intention instead of behavior overlook the essence of the privacy paradox, that is, privacy intention does not always match with privacy behaviors, that is why the self-disclosure intention model refutes the privacy paradox while the actual self-disclosure behavior model confirms it.
Dienlin and Trepte (2015) argued that some studies did not find an association between privacy behaviors and attitudes due to the operationalization of the outcome variable (e.g. Self-disclosure). Thus, the behavior is often measured as a dichotomous answer (e.g. Taddicken, 2014) and attitudes are measured on metric scales (e.g. Krasnova et al., 2009; Zlatolas et al., 2015). Dichotomy measurement might imply limitation of variance, reducing statistical power (Schmidt et al. 1976 cited by Dienlin and Trepte, 2015). However, this study measured the personal identifiable information (part of the self-disclosure variable) as a dichotomous answer, so we can discard that the discrepancy in results is due this.
Moreover, unlikely other studies that measured Self-reported behavior (Taddicken, 2014; Tsay-Vogel et al., 2018; Becker & Pousttchi, 2012; Tufekci, 2008; Cheung et al., 2015; Zlatolas et al., 2015; Taddei & Contena, 2013) this study tried to increase the accuracy (Nguyen, Bin &
Campbell, 2012) to measure Self- Disclosure, in terms of actual behavior, in which content analyses were performed on same users that responded the survey.
In addition to those models, it was proposed that perceived control and subjective norm could have been associated with privacy concerns, with the argument that users with more perception of control over the information online on Instagram and users that consider that people whose opinion they value use Instagram would be more likely to be less concern about their privacy on Instagram, but no significant results were found in these relationships either, contrary to previous finding over subjective norm (Taddicken, 2014) and perceived information control (Jia et al., 2015; Zlatolas et al., 2015)
Besides, to capture the complexity of Self-disclosure, important pieces of information that the user could have disclosed 23 pieces of personal identifiable information were included. Also, because the functionality of Instagram allows having open or restricted access profiles only, this variable was included as previous studies suggested (Taddicken, 2014). The perceived sensibility of the information (Mothersbaugh et al., 2012) was also considered. While
Mothersbaugh (2012) argued that the privacy paradox may be a result of not considered information sensitivity, but the actual self-disclosure behavior model still confirms the discrepancy.
At the beginning, this study argued that was necessary to investigate the privacy paradox in different SNSs, the image-centric Instagram, that had become popular in the recent years (Blank & Lutz, 2017) among young adults (Clarke, 2019) and might have different
functionalities that similar SNSs such as Facebook (Shane-Simpson et al., 2018), in which the privacy paradox has been largely studied (Kokolakis, 2017).
This study attempted to considers the different functionalities of Instagram at the moment to investigate the discrepancy between privacy attitudes and behaviors online, limited account access privacy settings (Shane-Simpson et al., 2018) and the centrality of the image (Hu et al., 2014), taking in consideration information from both text and images (Andalibi et al., 2007) when the data was retrieved to measure personal information.
Thus, this study suggested that the discrepancies between privacy concerns and actual self-disclosure behavior still exist on Instagram contributing to the idea that concerns regarding social or organizational privacy threats on SNSs are not enough for users to stop them to share private information on Instagram.
Finally, there are some social contributions to this research. As was mentioned, the functionalities of SNSs are designed to elicit the information disclosure of the users (Blank, Bolsover & Dubois, 2014), and despite user’s privacy concerns, there is no association with their actual self-disclosure behavior on Instagram. Content analyses revealed that the number of publications per user is relatively considerable, and there is still a high number of users that keep their information as public.
Moreover, content analyses revealed that most of the users allowed third-party accounts to be part of their social network (such as Institutions, online stores, meme accounts), and therefore have access to their publications. Just as an example the account that I created to invite participants (@chillenial_pavre) follows 4.711 users and have access to their information
updated in their profiles. This means -theoretically- that with limited resources, a user can collect the personal identifiable information of 4711 individuals without their knowledge or consent.
Considering this, and as Walther (2011) suggested, education about the online footprint (Boyd, 2007) seems to be the key to conscientize individuals about privacy threats by institutions
or unwanted audiences. Particularly the group of emerging adults seems (and possibly teenagers) to be the focus for future policies regarding online privacy protection. Effective efforts are necessary to help social media users to understand the vulnerabilities and unintended consequences while using SNSs.
Education of the digital footprint and privacy online must be a concern, when we see that users are getting used to sharing personal information online over the years. In this sense, for example Tsay-Vogel et al. (2018) 5 years study on privacy attitudes showed that users are getting less concern over their privacy online over the time, particularly teenager that feel comfortable with the idea of sharing information on SNSs (Camacho, Minelli, & Grosseck, 2012), and research showed that the frequent use of SNSs increase the disposition to self-disclose online (Trepte & Reinecke, 2013), as our actual behavior model also suggest this association.
Limitations and Directions for Future Research
Despite these interesting approaches and practical and theoretical implications of this study, there as some limitations in this study. This study has limitations regarding external validity (Rossi, Wright & Anderson, 1983 cited by Cook et al., 2002) because the sample is not
representative in terms of age (18-25 years old) or cultural background (only Chilean participants were included), or SNSs (only Instagram users) and an important group of Instagram users such as the underage teenagers (Anderson & Jian, 2018) was not included in the results of this study. Moreover, assumption of normally distributed errors in both models was violated and the distribution in all variables is not normal, although, according to Field (2013) in larger sample sizes this violation does not invalidate the significance test, because of central limit theorem, but results should be read with caution.
On the other hand, results from multiple regression analyses do not allow to make causal inferences in the relationships tested (Field, 2013). Also, these results should be interpreted with cautious because some of the composed variables included in the model such us self-disclosure actual behavior, intention, and Instagram use incorporated different response scales combined into one measure, affecting factor analyses which are sensitive to different response scales, because factor analyses depend on variance (Field, 2013).
This study can also have a self-selection bias, as Acquisti & Gross (2006) raised this concern in a similar study of actuals self-disclosure behavior, because the participants that were willing to share their Instagram accounts and agreed to share their information in this study are less privacy-conscious than non-participants.
Also, a reliability limitation was pointed out in a similar study by Taddicken (2014), in the sense that social desirability bias (King & Bruner, 2000) could have affected the way privacy concern and self-disclosure intention was measured. For example, self-disclosure intention scale contained sensitive information (e.g. sexual behavior, drugs consumption) that could have led respondents to overreport socially desirable answers (Krumpal, 2013).
Furthermore, one of the questions that should be answered, in which other significant factors might explain the privacy paradox and weren’t considered in this study, such as
personality traits and individual characteristics, which might affect online self-disclosure (Gerber et al., 2018; Shane-Simpson et al., 2018; Taddicken, 2014).
Another important limitation of the study is a sample of pictures and videos selected for each code-unit (Instagram profiles). A sample of 10 posts per user might be considered as too small when the mean number of publications of the sample is 152. More developed techniques such as Automated Content analysis (Trilling & Jonkman, 2018) might be considered for other
privacy paradox studies on SNSs (Acquisti & Gross, 2006) an to retrieve bigger samples of information. If this possible, future studies might take in consideration not just the presence of absence of certain information, but also the amount of times that information was disclosed.
Future research might consider to investigate the privacy paradox in other social media, such as dating apps (e.g. Tinder, Happn) which had become popular in the recent years and in which, unlikely the other SNSs the purpose of the app is to meet new people, outside the social circle of the user, so it may raise different institutional and social concerns and motivations (Lutz & Ranzini, 2017).
While this study measured self-disclosure intention and actual behavior, self-reported self-disclosure behavior was not considered, and it could have been important to measure given the fact that most studies on the privacy paradox considered self-reported behavior and not actual behavior (Kokolakis, 2017; Gerber et al., 2018) and self-reported measurements can also differ from actual behavior (Baumeister et al., 2007). Future studies might consider measuring the three concepts, that while similar, do not reflect the same.
In conclusion, the survey and content analyses approach from this study shows that the discrepancy between privacy attitudes and actual behavior exist on Instagram among Chilean emerging adults, but no other significant factors in this study contributed to explain why users actually disclosure information on Instagram. Moreover, it was argued that the measurement of Self-disclosure as self-disclosure intention could affect the results, given that this discrepancy was not confirmed, and self-disclosure behavior was negatively associated with self-disclosure intention.
References
