Peeling back the multiple layers of Twitter’s private disclosure onion: The roles of virtual identity discrepancy and personality traits in communication privacy management on Twitter

Abstract

This study examined multiple layers of private disclosure on the microblogging site Twitter. Survey data (N = 375) were collected from current Twitter users (N = 198), nonusers (N = 116), and dropouts (N = 61). Data from current Twitter users revealed the existence of multiple strata of private disclosure boundaries on Twitter. There were significant differences at the descriptive and inferential levels among the multiple dimensions of private information, including daily lives, social identity, competence, socio-economic status, and health. Private information regarding daily lives and entertainment was disclosed easily and located at the outermost layer of the disclosure onion. In contrast, health-related private information was concealed and located within the innermost layer of the disclosure onion. ANOVAs (N = 375) also indicated that there were significant differences among current Twitter users, nonusers, and dropouts with regard to personality traits and privacy concerns about Twitter. Theoretical implications were discussed.

Keywords

Microblogging Neo Big 5 privacy management private disclosure social media social networking tweets Twitter user-generated content (UGC)virtual identity discrepancy

Introduction

The social networking and microblogging platform Twitter allows people to post short (140 characters in length) text updates called “tweets” to a network of others known as “followers” (Marwick and boyd, 2010). In light of the growth of this innovative social media and embedded user-generated content (UGC), a burgeoning body of research has provided preliminary empirical data that reports Twitter usage across various domains and theoretical discussions about Twitter from multifarious perspectives. For example, drawing from social construction theories of technology, Arceneaux and Weiss (2010) analyzed press coverage of Twitter and found that newspapers, magazines, and blogs have actively promoted and encouraged diffusion of this innovative microblogging technology. Driven by uses and gratification theory, Chen (2011) presented empirical data supporting the notion that Twitter use gratifies people’s interpersonal need to connect with others. In the sports media domain, Hutchins (2011) examined how high-profile athletes use Twitter to communicate instantaneously with fans and observers, bypassing journalists’ gate-keeping role. In the computer-assisted interactive learning domain, Junco et al. (2011) empirically supported the positive impact of Twitter as an educational tool in increasing students’ engagement. Also in the technology-mediated learning domain, Johnson (2011) uncovered the positive effects of social tweets (versus academic tweets) on the instructor’s perceived credibility.

Despite the increasing use of Twitter as a venue for disclosing private information, there is a dearth of empirical findings about private disclosure on the site. This research attempted to fill this gap by examining (a) multiple layers and strata of private information withheld or disclosed on Twitter; (b) the social psychological mechanisms underlying the dynamic processes of privacy management and virtual identity expression; (c) the roles of extraversion and openness dimensions of Neo BIG5 personality traits in managing the dialectical tension between concealing and revealing private information on Twitter; and (d) the differences among current Twitter users, nonusers, and dropouts with regard to personality traits and privacy concerns. The overarching goals of the current article were to peel back the multiple layers of the private disclosure onion on Twitter and offer novel models of privacy management and virtual identity construction on the microblogging site. To accomplish these goals, this research drew upon communication privacy management theory (Petronio, 2000) and virtual identity discrepancy theory (Jin, 2012a).

Communication privacy management (CPM) theory

“Private information” refers to information about things that matter deeply to a person, and “private disclosure” refers to the process of communicating private information in relationships with others (Petronio, 2000). Compared to self-disclosure, which refers to revealing personal information about one’s self, “private disclosure does not restrict the process to only the self but extends it to embrace multiple levels of disclosure including self and group” (Petronio, 2002: 3). To address the breadth and depth of disclosure, CPM theory puts more emphasis on the substance (personal content) of private disclosure in a dialectical tension between withholding and disclosing private information.

When people do not disclose private information, they establish a personal boundary. When people share private information, they establish a collective boundary; the information then belongs to the relationship (Petronio, 2002). These concepts can be applied to privacy management in the dynamic processes of text posting and social networking on Twitter. If a Twitter user conceals private information and does not disclose it on Twitter, the information stays within the personal boundary and resides at the innermost layer of the private disclosure onion. Thus, withholding private information maintains the personal boundary. In contrast, if Twitter users reveal private information and share it with their followers and other people the user is following, the information belongs to the collective boundary and resides at the outer layer of the private disclosure onion. If Twitter users do not protect their Twitter account, thus allowing the general public and online lurkers to access its contents, tweets can leak into the public realm and the collective boundary further expands (Meeder et al., 2010). Thus, multiple layers of private information disclosure form and gradually expand the collective boundary due to the unique features (tweets, retweets, following, followers, listing, tweet protection, etc.) of Twitter technology. The next section provides a literature review on (a) contextual and layered approaches to privacy and (b) differences between privacy management in offline environments and online environments.

It is worth acknowledging relevant scholarship that addresses privacy issues in a larger social context. Nippert-Eng (2010) explored how people manage their secrets and provided empirical studies on social interaction and interpersonal relations in the context of privacy management. She discovered that people selectively conceal and disclose information on a daily basis and claimed that successfully managing privacy is critical for maintaining social relationships. Avoiding the private-versus-public dichotomy in confronting and resolving privacy concerns, Nissenbaum (2009) argued that “a right to privacy is neither a right to secrecy nor a right to control but a right to appropriate flow of personal information.” She further claimed that privacy must be understood in its social context, thus taking a more contextual approach to privacy. Based on the fundamental understanding of these contextual approaches to privacy at the macro/social level, the current study proposes a layered approach to privacy on social media to systematically investigate multiple strata of private disclosure at the micro/psychological level.

CPM theory was originally proposed to explain why and how people decide to reveal or conceal private information in offline environments. Prior to applying CPM theory to Twitter usage in online environments, it is crucial to discuss differences between the nature of privacy management in face-to-face (FtF) communication in offline environments and computer-mediated communication (CMC) in online environments. According to danah boyd (2010), “a conversation you might have in the hallway is private by default, public through effort,” whereas “when you engage online in equally public settings such as on someone’s Facebook wall, the conversation is public by default, private through effort.” Boyd’s observation resonates with the challenge of keeping information private and confidential in online environments. In online environments like Twitter and Facebook, lurkers’ exposure and access to others’ tweets and Facebook walls or comments might be comparable with overhearing information. Overhearing a disclosure results in a lower inclination to protect that information responsibly than is the case if being considered as an explicit confidant (Greene et al., 2003), thus leading to lower boundary linkages (connections that form boundary alliances between people).

Waters and Ackerman (2011) also acknowledge a discrepancy between FtF communication and CMC with regard to privacy and disclosure, based on the empirical examination of motivations and consequences of Facebook users’ voluntary disclosure. The Internet is the apex of weak boundary linkages due to its prime feature, connectivity (Jin, 2012b; Joinson and Paine, 2007). Furthermore, information disclosed and shared in an offline setting tends to have a relatively brief retention time due to the limited human memory (Blanchette and Johnson, 2002), whereas information posted online is often stored for a long time period and is easily replicated due to the nature of the Internet’s infrastructure (Coopamootoo and Ashenden, 2011). Social media users might be aware of these risks regarding weak boundary linkages and consequently behave discreetly when engaging in private disclosure online.

Privacy management and private disclosure on Twitter

Consistent with the basic suppositions of CPM theory, Twitter users can control boundary ownership, boundary permeability, and boundary linkages. People feel and believe that they have the right to control the flow of private information to others. Boundary ownership in Twitter can be operationally defined as the rights, privileges, and amount of responsibilities accruing to co-owners (e.g., the Twitter account owner, followers, people followed by the account owner, online lurkers, the general public, random visitors, etc.) of private information revealed during Twitter-related activities (e.g., Tweets, listing, following, trending, photo/video uploading, etc.). Boundary permeability in Twitter can be operationally defined as how much information is able to pass through the boundary after Twitter users’ disclosure of private information to others. Boundary linkages in Twitter can be operationally defined as the formation of collective boundaries through private information disclosure and sharing among Twitter users. For online lurkers visiting protected Twitter accounts, the linkage is low, because they are not the intended recipient of the information.

These mechanisms of boundary ownership, permeability, and linkages on Twitter operate in a dialectical tension between private information withholding and disclosure. Twitter users can manually approve each and every person who may view that account’s tweets by “protecting” their tweets. These protected tweets (thick walls; Petronio, 2002) are visible only to people the user has approved and cannot be retweeted by those who are not approved. In contrast, accounts with public tweets (thin walls; Petronio, 2002) have profile pages that are visible to everyone without the need for approval, thus inviting the public to become shareholders of private information. If other followers further disseminate the initially shared private information to a broader audience via retweeting, the information is subject to a more extended collective boundary. As access to private information increases, boundaries become more permeable. Thus, the increase in the number of co-owners of the private information expands the collective boundary in the dynamic process of communication privacy management. Depending on users’ willingness to share private information, the collective boundary may contain multiple sub-boundaries stratified by the content and the amount of private disclosure. To investigate the existence of multiple strata and boundaries of concealing and revealing private information in Twitter, the following research question was posed:

RQ1: For current Twitter users (N = 198), are there multiple layers of private disclosure?

If there exist conceptually distinct layers of private information concealed or revealed on Twitter, the next logical step is to further examine rank ordering and significant differences among the multiple layers of private information disclosure. To address this agenda, the second research question was posed:

RQ2: For current Twitter users (N = 198), are there within-subjects differences (a) at the descriptive level (rank ordering) and (b) at the inferential level (significant t values) among the multiple layers of private disclosure?

In addition to the qualitatively different multiple layers of private disclosure, this study examined the quantitative dimension of private disclosure on Twitter. The amount of tweets posted on users’ Twitter accounts can serve as a quantitative indicator of the level of private disclosure. Previous research on social networking sites identifies three sources of information: self, other, and system (Tong et al., 2008). Profile pictures and text updates posted by an SNS user (i.e., account owner) are information generated by the self (Utz, 2010). The user’s friends’ profile pictures that appear at the corner of the user’s profile are information generated by the other. The number of friends displayed on the user’s profile is information generated by the system (Utz, 2010). Accordingly, user-generated content (UGC) on Twitter entails two dimensions: self-generated content (SGC) and other-generated content (OGC). The actual text updates (i.e., tweets) which Twitter users post on their own Twitter pages are SGC. Retweets posted by other people following or followed by the Twitter user are OGC. This process is conceptually consistent with boundary coordination as proposed by CPM theory. Departing from a unidirectional communication process, disclosed private information affects both the discloser and the recipient of a disclosure in CPM (Petronio, 2002). After Twitter users reveal private information, all involved parties (followers and people the Twitter user is following) become responsible for co-owning and comanaging the information. The service provider (company) counts and calculates the number of tweets by a user, the number of followers, and the number of other Twitter users the user is following, presenting company-generated content (CGC). Thus, Twitter consists of multifarious contents generated by multiple sources: Tweets as SGC, retweets as OGC, and the quantitative indices of tweets and followers as CGC or system-aggregated indicators of user input.

Furthermore, tweets have two dimensions: user (self and other)-generated qualitative contents and system (company)-generated quantitative indicators. This study focuses on the quantity dimension of UGC rather than analyzing the actual content dimension. The number of people the user is following and the number of followers represent system-aggregated quantitative indicators of Twitter user behavior. Despite the prominence of these multiple indicators of user behavior across various social networking and blogging sites, there is a dearth of empirical data on the relationship between the intensity of Twitter use and the scale of social networks formed and maintained via the microblogging platform. To provide preliminary empirical evidence for the relationship among the amount of tweets, the amount of people the user is following, and the amount of followers, the third research question was posed:

RQ3: For current Twitter users (N = 198), is the number of tweets quantitatively correlated with (a) the number of people followed by the user and (b) the number of people following the user (i.e., followers)?

Virtual identity discrepancy

This article further examined dynamic processes of private disclosure and privacy management in light of users’ virtual identity construction and virtual identity discrepancy on Twitter. Goffman (1959) proposed a dramaturgical theory of self, seeing the self as the process of dramatic interaction that produces multiple selves for multiple performances. In Goffman’s dramaturgical studies, the self manages its interactional ventures strategically to project an image that other interactants will find acceptable and desirable (Robinson, 2007). Goffman’s dramaturgical theory of self acknowledges the multiple, malleable, and strategic nature of self-presentation. Applying Goffman’s theory of self to online and virtual environments, Jin (2012a: 2161) defines virtual identity as “the technology-mediated or mentally processed identity being on or simulated on computers, computer networks, and any other virtual and digital media environments.” Virtual identity entails two dimensions: The virtual self and the virtual other (Jin, 2012a). The virtual self is a technology-mediated self presented in digital media environments at the intrapersonal level. When the virtual self enters the realm of interpersonal communication and social interaction with other entities, it encounters the virtual other, which refers to the technology-mediated other presented in digital media environments at the interpersonal level. Applying these concepts to the context of Twitter use, the virtual self can be operationally defined as a user’s graphical and textual representation of the self on Twitter, and virtual others can be operationally defined as the other’s graphical and textual representation of their identity on Twitter.

Applying Higgins et al.’s self-discrepancy theory (1985) to online and virtual environments, virtual identity discrepancy refers to the discrepancy between the actual identity in the real world and the virtual identity in the online or virtual world. Virtual self-discrepancy in the context of Twitter use can be operationally defined as the extent to which a Twitter user’s virtual self presented on the online social networking and microblogging site deviates from the actual self of the user in the offline, “real” world (Jin, 2012a). Driven by the hyperpersonal perspective, Walther (2007) examined discrepancies between offline identity in the real world and virtual identity in online friendships due to limited cues and potential asynchronicity of computer-mediated communication (CMC). Drawing upon Jin’s (2012a) original virtual identity discrepancy model, the current research empirically tested the discrepancy between the actual self and the virtual self disclosed on Twitter with regard to social/physical/task attraction (McCroskey and McCain, 1972) and personality traits (Costa and McCrae, 1985). Utz (2010) identified these attraction and personality variables as important dependent measures in research on people’s virtual identity construction and virtual impression formation on social networking sites. To provide preliminary data on virtual identity construction and virtual identity expression in the dynamic process of private disclosure on Twitter, the fourth research question was posed:

RQ4: For current Twitter users (N = 198), are there virtual self-discrepancies (i.e., within-subjects differences between the actual self in the real world and the virtual self presented on Twitter) with regard to (a) social, (b) physical, and (c) task attractions, as well as (d) Neo BIG 5 personality traits?

Neo BIG 5 personality traits and private disclosure

Building upon the examination of personality traits in light of virtual identity discrepancy, this research further examined the roles of personality traits in private disclosure on Twitter. Extraversion is one of the BIG 5 personality dimensions (Costa and McCrae, 1985). Utz (2010) found that owners of extraverted profiles on social networking sites are perceived as more popular than owners of introverted profiles, thus implicating the conceptual relevance of extraversion to active self-presentation in social media.

Given the role extraverted personalities play in SNS, this research further examined the theoretical relevance of extraversion to the dialectical tension between guarded self-disclosure and active private disclosure on Twitter. Guarded self-disclosure is a multidimensional construct consisting of privacy, self-concealment, and conflict avoidance (Barry, 2003). This research focused on the correlation of extraversion scores with privacy and self-concealment dimensions of guarded self-disclosure. As extraverted people disclose more information in CMC, it can be predicted that a current Twitter user’s extraversion is negatively correlated with guarded self-disclosure (RQ5a), whereas it is positively correlated with the multiple layers of private disclosure (RQ5d). Relational privacy preference is a multidimensional construct composed of neighbor avoidance, solitude, reservation in a relationship with a partner, and possessiveness (Craddock, 1997). Neighbor avoidance is conceptually germane to maintaining a personal boundary and avoiding the formation of a collective boundary in private disclosure processes online. Therefore, this research mainly examined the correlation between extraversion scores and the neighbor avoidance dimension of relational privacy preference. It can be predicted that extraversion is negatively correlated with neighbor avoidance (RQ5b). This research also explored the novel relationship between extraversion and virtual identity discrepancy (RQ5c). Additionally, other Neo BIG 5 personality traits’ correlations with private disclosure were tested (RQ5d). To address these relationships, the following research questions were posed:

RQ5: Is extraversion correlated with (a) guarded self-disclosure, (b) relational privacy preference, and (c) virtual identity discrepancy (versus congruity) on Twitter (N = 198)?

RQ5d: Are Neo BIG 5 personality traits correlated with the multiple layers of private disclosure on Twitter (N = 198)?

Finally, this research examined between-subjects differences regarding personality traits and privacy concerns as a function of user status (current users, nonusers, and dropouts) by posing the sixth research question:

RQ6: Are there differences among current Twitter users, nonusers, and dropouts with regard to (a) personality traits and (b) privacy concerns about Twitter (N = 375)?

Method

Data collection

This study drew upon cross-sectional survey data on college students’ Twitter usage. Participants were 375 undergraduate students (M _Age = 20.14, SD _Age = 1.05, Range _Age = 23–18 = 5; 198 current Twitter users, 116 nonusers, and 61 Twitter dropouts) enrolled in a university in the US. The recruitment procedure utilized a combination of several non-probability sampling techniques, including volunteer sampling, network sampling, and convenience sampling. Students were instructed to sign up for a research session by using an online scheduling website. To minimize the limitations of an online survey in verifying the authenticity of actual participants, students were invited to a lab to take the survey. Trained researchers instructed participants to sign the informed consent form approved by the IRB. The trained researchers directed each participant to a separate room equipped with a computer. Participants completed the online survey (Qualtrics) individually.

Measures

Private disclosure (RQ1 and RQ2) was measured by asking participants (current Twitter users N = 198) to answer the following question for each item: “Using the scale below, please indicate the appropriate number for each item. [1] indicates ‘I am reluctant to disclose my private information on Twitter’ and [7] indicates ‘I am willing to disclose my private information on Twitter.’” The items for each 7-point scale were: (1) favorite foods; (2) favorite restaurants; (3) music, movies, entertainment interests; (4) travels; (5) school; (6) occupation; (7) group memberships; (8) gender; (9) intelligence; (10) motivation; (11) strengths; (12) social skills; (13) education; (14) political affiliation; (15) socio-economic status; (16) family; (17) mental health; and (18) physical health. The order of the items was randomized to minimize sequential effects.

The intensity of Twitter use (RQ3) (N = 198) was measured by three items: (1) “tweets,” (2) “people you are following,” and (3) “followers” (see Table 1 for actual questions).

Table 1.

Demographics of study participants (N = 375 [top]) and descriptive statistics about current Twitter users (N = 198 [bottom])

Study participants (N = 375)	N (Frequency)	% (Ratio)
Gender
Female: Male	231: 144	61.6: 38.4
Ethnicity
Asian Pacific Islander: South Asian (Indian)	37: 2	9.9:.5
African American: Latino	24: 24	6.4: 6.4
White: Middle Eastern	271: 2	72.3: .5
Mixed: Other	10: 5	2.7: 1.3
Annual household income
Less than US$20,000: US$20,001–US$40,000	56: 20	14.9: 5.3
US$40,001–US$60,000: US$60,001–US$80,000	19: 49	5.1: 13.1
US$80,001–US$100,000: More than US$100,000	56: 175	14.9: 46.7
Twitter usage status
Current users	198	52.8
Nonusers (never used it)	116	30.9
Dropouts (former users)	61	16.3
Current Twitter users (N = 198)	N (Frequency)	% (Ratio)
How many times did you user Twitter in the past month?
Less than 10: 10–30	32: 87	16.16: 43.94
31–60: 61–90	45: 3	22.73: 1.52
More than 91	31	15.66
How many minutes did you spend on Twitter in the past month?
0–30: 31–60	51: 41	25.76: 20.71
61–90: 91–120	8: 33	4.04: 16.67
121–150: 151–180	5: 2	2.53: 1.01
181–210: 211–240	8: 2	4.04: 1.01
241–270: 271–300	3: 15	1.52: 7.58
More than 300	30	15.15
How many people are you “following”?
1–30: 31–60	31: 37	15.66: 18.69
61–90: 91–120	32: 34	32.65: 17.17
121–150: 151–180	14: 12	7.07: 6.06
More than 180	38	19.19
How many “followers” do you have?
1–30: 31–60	53: 37	26.77: 18.69
61–90: 91–120	32: 22	16.16: 11.11
121–150: 151–180	15: 11	7.58: 5.56
More than 180	28	14.14
How many “tweets” do you have?
1–50: 51–100	55: 24	27.78: 12.12
101–150: 151–200	20: 7	10.10: 3.54
201–250: 251–300	6: 7	3.03: 3.54
More than 300: Missing	71: 8	35.86: 4.04

Note: Colons (:) indicate the ratio of the first item to the second item.

Social, physical, and task attractions (RQ4a, 4b, 4c) of the actual self (“I see my actual self as someone who is…” α = .93, α = .98, α = .92) and the virtual self (“I see my self-image presented on Twitter as someone who is…” α = .91, α = .97, α = .90) (N = 198) were measured by nine items (McCroskey and McCain, 1972).

Personality traits (RQ4d, RQ5, and RQ6) were measured by the Neo BIG 5 personality inventory (Costa and McCrae, 1985), consisting of extraversion (Twitter users’ [N = 198] actual self: α = .86, virtual self: α = .87; nonusers [N = 116]: α = .84; dropouts [N = 61]: α = .81), agreeableness (actual self: α = .82; virtual self: α = .85; nonusers: α = .85; dropouts: α = .75), conscientiousness (α = .84, α = .89; α = .85; α = .83), neuroticism (α = .88; α = .83; α = .84; α = .82), and openness (α = .86; α = .83; α = .88; α = .78).

Guarded self-disclosure (RQ5a) (N = 198) was measured by the six items measuring privacy and self-concealment dimensions (α = .76) from the Guarded Self-Disclosure Inventory (GSDI; Barry, 2003).

Relational privacy preference (RQ5b) (N = 198) was measured by the three items measuring the neighbor avoidance dimension (α = .66) from the Relational Privacy Preference Scale (RPPS; Craddock, 1997).

Virtual identity discrepancy (RQ5c) (α = .85) (N = 198) was measured by three items creatively adapted from the selves questionnaire (Higgins et al., 1985) and the homophily scale (McCroskey et al., 1975) with 7-point scales.

Privacy concerns (RQ6b) (N = 375) were measured by eight items (α = .92) with 7-point Likert scales (e.g., “I am concerned about disclosing my private information on Twitter”).

Results

Factor analysis

Parallel analysis and exploratory factor analysis (EFA) were conducted to answer RQ1 (N = 198). Initially, the factorability of the 18 items designed to measure private disclosure was examined. The results from parallel analysis and principal axis factoring extraction using an oblique (Direct Oblimin) rotation procedure revealed five different dimensions of private disclosure (Table 2). The Kaiser–Meyer–Olkin measure of sampling adequacy was .88, above the recommended value of .60, and Bartlett’s test of sphericity was significant (χ² (153) = 2226.673, p < .01). Based on the results of both parallel analysis and EFA, each dimension of private disclosure was conceptualized as a distinct construct and given the following labels: (Layer 1) favorite foods, restaurants, music, movies, and travel as the “daily life and entertainment” dimension; (Layer 2) school, occupation, group memberships, and gender as the “social identity” dimension; (Layer 3) intelligence, motivation, strengths, and social skills as the “competence” dimension; (Layer 4) education, political affiliation, socio-economic status (SES), and family as the “SES and education” dimension; and (Layer 5) mental health and physical health as the “health” dimension.

Table 2.

Structure matrix from exploratory factor analysis based on a principal axis factoring extraction method with Direct Oblimin rotation of 18 private disclosure items (RQ1: N = 198)

	Daily life and entertainment	Social identity	Competence	SES and education	Health	Communalities
Foods	.921					.850
Restaurants	.916					.843
Music, movies	.876					.779
Travels	.793					.639
School		.914				.838
Occupation		.953				.913
Group memberships		.456				.500
Gender		.467				.575
Intelligence			.820			.717
Motivation			.865			.768
Strengths			.894			.806
Social skills			.692			.580
Education				.713		.526
Political affiliation				.781		.638
SES				.761		.648
Family				.508		.449
Mental health					.751	.573
Physical health					.713	.527

Descriptive statistics

Descriptive statistics (RQ2a, N = 198) indicate clear rank ordering among the five layers of private disclosure (Layer 1: M = 6.27, SD = 1.51; Layer 2: M = 5.98, SD = 1.46; Layer 3: M = 5.54, SD = 1.62; Layer 4: M = 4.07, SD = 1.43; Layer 5: M = 3.99, SD = 1.75). Current Twitter users’ daily life and entertainment-related information (Layer 1) ranked highest with regard to the level or amount of private information disclosed and therefore was located at the outermost layer of the private disclosure onion. In contrast, health-related information (Layer 5) ranked lowest and therefore was located at the innermost layer.

Inferential statistics

t-tests

Paired-samples t-tests were conducted to answer RQ2b (N = 198). Within-subjects comparisons revealed significant differences among the different layers of private information (Table 3). Alpha levels were adjusted using Bonferroni correction methods.

Table 3.

Paired-samples t-tests for within-subjects private disclosure comparison (RQ2b: N = 198)

	Paired differences effect sizes
	M (Mean difference)	SD	SE	95% confidence interval		t	Adjusted alpha (Bonferroni correction)	Cohen’s d	r
				Lower	Upper
Pair 1: (Layer 1 – Layer 2)	.29	1.19	.09	.11	.46	3.26*	.005	.48	.24
Pair 2: (Layer 1 – Layer 3)	.74	1.25	.09	.55	.92	7.94**	.001	1.18	.51
Pair 3: (Layer 1 – Layer 4)	2.14	1.74	.12	1.89	2.39	17.30**	.001	2.47	.78
Pair 4: (Layer 1 – Layer 5)	2.22	1.97	.14	1.94	2.50	15.81**	.001	2.25	.75
Pair 5: (Layer 2 – Layer 3)	.45	1.49	.11	.23	.67	4.05**	.001	.60	.29
Pair 6: (Layer 2 – Layer 4)	1.90	1.76	.13	1.64	2.16	14.58**	.001	2.17	.73
Pair 7: (Layer 2 – Layer 5)	1.99	2.03	.15	1.69	2.29	13.18**	.001	1.96	.70
Pair 8: (Layer 3 – Layer 4)	1.45	1.68	.12	1.21	1.70	11.68**	.001	1.74	.66
Pair 9: (Layer 3 – Layer 5)	1.54	1.95	.14	1.25	1.83	10.63**	.001	1.58	.62
Pair 10: (Layer 4 – Layer 5)	.08	1.66	.12	−.16	.31	.65	.005	.09	.05

Note: Using Bonferroni correction methods, the alpha levels were adjusted.

p * < .005, p ** < .001.

Paired-samples t-tests were conducted (RQ4, N = 198). There are significant differences between the actual self and the virtual self with regard to social attraction, physical attraction, task attraction (trustworthiness), and personality traits (Table 4).

Table 4.

Paired-samples t-tests for within-subjects actual self versus virtual self comparison (RQ4: N = 198)

	Paired differences effect sizes
	M (Mean difference)	SD	SE	95% confidence interval		t	Adjusted alpha (Bonferroni correction)	Cohen’s d	r
				Lower	Upper
Pair 1: Social attraction (Actual – Virtual)	.16	.61	.04	.08	.25	3.72**	.00125	.53	.26
Pair 2: Physical attraction (Actual – Virtual)	.04	.71	.05	−.06	.14	.87	.00625	.12	.06
Pair 3: Task attraction (Actual – Virtual)	.33	.77	.05	.23	.44	6.09**	.00125	.87	.40
Pair 4: Extraversion (Actual – Virtual)	.24	.83	.06	.12	.36	4.08**	.00125	.58	.28
Pair 5: Agreeableness (Actual – Virtual)	.43	.99	.07	.29	.57	6.03**	.00125	.86	.39
Pair 6: Conscientiousness (Actual – Virtual)	.52	1.1	.08	.37	.67	6.74**	.00125	.96	.43
Pair 7: Neuroticism (Actual – Virtual)	.43	1.1	.08	.28	.58	5.66**	.00125	.81	.37
Pair 8: Openness (Actual – Virtual)	.34	.77	.05	.24	.45	6.32**	.00125	.90	.41

Note: Using Bonferroni correction methods, the alpha levels were adjusted.

p * < .00625, p ** < .00125.

Correlation

Regarding RQ3a and RQ3b (N = 198), the amount of tweets was positively correlated with the number of people the user is following (Pearson r = .57, p < .01) and the number of people following the user (r = .31, p < .01). There was also a positive correlation between the number of followers and the number of people the user is following (r = .28, p < .01). Regarding RQ5a, RQ5b, and RQ5c, Twitter users’ extraversion scores were negatively correlated with guarded self-disclosure, relational privacy preference, and virtual identity discrepancy. Virtual identity discrepancy was positively correlated with guarded self-disclosure and neighbor avoidance, implying that people with higher self-concealment and neighbor avoidance tendencies present online selves that are different from their actual selves.

Regarding RQ5d, Twitter users’ (N = 198) extraversion scores were positively correlated with (a) layer 1, r = .26, p < .01; (b) layer 2, r = .41, p < .01; (c) layer 3, r = .32, p < .01; and (d) layer 4, r = .20, p < .01. Agreeableness was positively correlated with layer 2, r = .15, p < .05. Conscientiousness was negatively correlated with layer 5, r = -.21, p <.01. Neuroticism was not correlated with any layers of private disclosure. Openness scores were positively correlated with (a) layer 1, r = .22, p < .01; (b) layer 2, r = .25, p < .01; (c) layer 3, r = .20, p < .01; and (d) layer 4, r = .16, p < .01.

One-way ANOVAs

ANOVAs and Tukey’s HSD tests were conducted to answer RQ6 (N = 375). There were significant differences among current Twitter users, nonusers, and dropouts with regard to extraversion, agreeableness, openness, and neuroticism, as well as privacy concerns about Twitter (Table 5).

Table 5.

ANOVA and post-hoc analysis with Tukey HSD tests for between-subjects multiple comparisons among users, nonusers, and dropouts (RQ6: N = 375)

Dependent variables	Usage status comparison	F (partial η²)	Mean difference	SE	95% confidence interval
					Lower	Upper
Twitter privacy concerns	Users – Nonusers	8.56**	−.99**	.24	−1.57	−.42
	Users – Dropouts	(.06)	−.07	.33	−.86	.71
	Nonusers – Dropouts	7.48**	.92*	.37	.06	1.79
Extravert	Users – Nonusers		.45**	.12	.18	.72
	Users – Dropouts	(.04)	.16	.15	−.18	.50
	Nonusers – Dropouts		−.29	.16	−.66	.08
Agreeable	Users – Nonusers	3.39*	.30*	.12	.02	.58
	Users – Dropouts	(.02)	.04	.15	−.31	.39
	Nonusers – Dropouts		−.26	.16	−.64	.11
Conscientious	Users – Nonusers	1.62	.21	.13	−.08	.50
	Users – Dropouts	(.009)	−.02	.16	−.39	.35
	Nonusers – Dropouts		−.23	.17	−.63	.17
Neuroticism	Users – Nonusers	3.14*	−.34*	.13	−.65	−.02
	Users – Dropouts	(.02)	−.12	.17	−.51	.28
	Nonusers – Dropouts		.22	.18	−.21	.65
Open	Users – Nonusers	4.05*	.26†	.13	−.04	.56
	Users – Dropouts	(.02)	−.20	.16	−.57	.17
	Nonusers – Dropouts		−.46*	.17	−.86	−.06

Note: p†< .10, p*< .05, p** < .01.

Discussion

Summary of findings and interpretations

The data indicate the existence of stratified relationships among multiple layers of private disclosure on Twitter. First, factor analyses addressing RQ1 (Table 2) indicate that there are five different components of private disclosure: (1) daily lives and entertainment; (2) social identity; (3) competence; (4) SES; and (5) health. Therefore, this research provides empirical data supporting the thesis that there are multiple layers of private disclosure on Twitter. Figure 1 illustrates these multiple boundaries of the private disclosure onion and privacy management.

Figure 1.

Twitter’s private disclosure onion.

Descriptive statistics and inferential statistics addressing RQ2 demonstrate the rankings among the five distinct components of private disclosure (RQ2a) and statistically significant within-subjects differences among the five layers (RQ2b). Daily lives-related information and entertainment-related private information are located in the outermost layer of the private disclosure onion, whereas health-related private information is located in the innermost layer of the private disclosure onion (Figure 1). The within-subjects mean score differences between each layer are statistically significant, except for the mean difference between the most closely adjacent layers of SES/education and health (non-significant mean difference between layer 4 and layer 5 [pair 10]) (Table 3). This finding resonates with the onion analogy from Altman and Taylor’s (1973) social penetration theory (SPT). Self-disclosure is at the core of social penetration, the process of bonding that moves a relationship from superficial to more intimate. The outer, peripheral layers of the social penetration onion represent an individual’s public image, whereas the inner, central layers of the social penetration onion represent an individual’s private self. Similarly, the outer, peripheral layers of Twitter’s private disclosure onion represent information that users reveal and exchange more frequently (the breadth dimension of private disclosure), whereas the inner, central layers of Twitter’s private disclosure onion represent information that users are reluctant to reveal to the public (depth dimension). Thus, the current empirical findings in the novel domain of social media-based CMC can be integrated into existing interpersonal communication theories (SPT and CPM).

The amount of tweets sent was significantly and positively correlated with the number of people the user is following and the number of people following the user (RQ3). This finding serves as one of the first empirical supports for the relationship between active Twitter use and the scale of social networks built and maintained through the Twitter platform. This study also identified multiple sources of information disclosed on Twitter (UGC [SGC and OGC as user input] and CGC as system-aggregated indicators of user input) and positive correlations among them, thus enriching and refining the extant literature on self-generated content and other-generated content in CMC (Utz, 2010; Walther et al., 2009).

In line with the virtual identity discrepancy model (RQ4), the data revealed significant within-subjects differences between the actual self and the virtual self with regard to social and task attractions and personality traits (Table 4). However, the data demonstrate an empirical rebuttal of the theoretical propositions of the hyperpersonal perspective (Walther, 2007). The hyperpersonal model states that the asynchronicity and anonymity of many online environments allow people to construct idealized self-presentations. According to the hyperpersonal model, the virtual self presented on Twitter should be more idealized than the actual self. Contrary to this theoretical proposition of the hyperpersonal perspective, findings from the current data empirically demonstrate that Twitter users’ perceptions of their actual selves were more positive than their perceptions of their virtual selves. Participants perceived their actual selves in the real, offline world to be more socially attractive, trustworthy, extraverted, agreeable, conscientious, and open than their virtual selves presented in the online environment. This finding, which is seemingly inconsistent with the basic premise of the hyperpersonal perspective, can be interpreted from two theoretical angles: (a) the dialectical tension between concealing and revealing the true self in the dynamic private disclosure process outlined by CPM theory; and (b) the discrepancy between the actual self and the virtual self proposed in the virtual identity discrepancy model. First, with regard to the private disclosure process, Twitter users may conceal some aspects of their true personality on Twitter, which may explain more positive perceptions of actual personality in the “real,” offline world than of virtual personality partially presented in the online world. Second, with regard to virtual identity discrepancy perceptions, Twitter users may perceive their actual selves as more attractive than their virtual selves because the SNS is limited in its presentation of other multifarious aspects of their actual identities. Consequently, the discrepancy between the actual self and the virtual self may be attributed to the limited verbal and nonverbal cues in CMC-based impression formation and self-presentation. Furthermore, a Twitter user’s perceptions of the actual self and the virtual self from their own standpoint may be different from followers’ perceptions of the user from the other’s standpoint. The theoretical advantage of the original virtual identity discrepancy model over the hyperpersonal perspective is its ability to explain both overestimated (idealized) self-presentation and underestimated (modest) self-presentation, as well as to account for both the self and the other dimensions of virtual identity construction and virtual social interaction. To clarify this finding about more positive perceptions of the actual self over the virtual self, the length of Twitter use needs to be controlled to rule out the alternative explanation regarding effects of the amount of self-disclosure on hyperpersonal self-presentation and virtual identity discrepancy. Additionally, Back et al. (2010) claim that Facebook profiles reflect actual personality instead of self-idealization, thus suggesting no significant discrepancy between the actual self and the online self. This theoretical claim provides an alternative lens that can explain candid and transparent online self-presentation and complement the hyperpersonal perspective and the virtual identity discrepancy model.

This research examined the influence of personality characteristics and various social psychological processes underpinning private disclosure (RQ5) and found that Twitter users’ extraversion scores were negative predictors of guarded self-disclosure (RQ5a), relational privacy preference (RQ5b), and virtual identity discrepancy (RQ5c). In addition, extraversion and openness were positively correlated with the multiple layers of disclosure (RQ5d). This is one of the first empirical findings about the roles that personality factors play in the dynamic process of private disclosure and privacy management in the microblogging-based CMC context. This research demonstrates the roles of extraversion, openness, agreeableness, and conscientiousness in privacy management and private disclosure on Twitter.

Finally, this research provides empirical evidence for significant differences among current Twitter users, nonusers, and dropouts with regard to personality characteristics and privacy concerns about Twitter (RQ6) (Table 5). Although the finding about the difference between nonusers and dropouts with regard to privacy concerns cannot prove a cause-and-effect relationship between Twitter usage and privacy concerns, it is probable that nonusers have concerns about privacy and consequently are reluctant to sign up for a Twitter account. Further empirical tests are needed to illuminate this causal relationship.

This research elucidates various social psychological mechanisms underpinning multiple layers of private disclosure on Twitter, drawing upon the solid theoretical foundations germane to privacy management on social networking and microblogging platforms. It provides rich empirical evidence for the existence of stratified layers of private information that is either revealed, forming a collective boundary, or concealed, staying within the personal boundary. Private information that is revealed by the self and shared with others in Twitter-based social relationships is located at the outer layers of the private disclosure onion. In contrast, private information that is concealed by the self and not shared with others is located at the inner layers of the private disclosure onion.

Limitations and suggestions for future research

First, the foci of this research were on the virtual self, virtual self-discrepancy, and virtual identity expression at the intrapersonal level. This research examined users’ private disclosure on their own Twitter account and did not test users’ perceptions of others’ private disclosure. Other people on Twitter include both “followers” of the user and “people the user is following.” Follow-up studies need to investigate Twitter users’ perceptions of virtual others presented on Twitter and to measure the amount of others’ private disclosure and the subsequent formation and expansion of collective boundaries via social networking and microblogging technology. These follow-up studies will add to the tapestry of research on virtual identity by examining CMC at the interpersonal level, as well as the theoretical counterparts of the virtual self (i.e., the virtual other) and self-discrepancy (i.e., other-discrepancy).

Second, one challenge in measuring Twitter users’ virtual other-discrepancy perceptions is that many users follow or are followed by online interlocutors that they never encounter in FtF communication. Twitter users may follow or be followed by public figures and celebrities with whom they do not have FtF encounters. To address this challenge, follow-up studies may compare Twitter-based interactions with online-only followers and Twitter-based interactions coupled with FtF contact. Empirical research on significant differences between these two different ways of social networking (purely online social networking versus hybrid forms of online social networking combined with existing FtF connections) would serve as valuable discourse on bridging social capital (weak ties) and bonding social capital (strong ties) (Putnam, 2000). Social networking and microblogging-based social media can function as an innovative venue for both building and expanding the scale or range of new social networks (bridging social capital) and maintaining and strengthening the depth or intimacy of existing relationships (bonding social capital) (Phua and Jin, 2011).

Third, this research was based on cross-sectional survey data from participants’ retrospective self-reports and analysis of the “quantity” dimension of user behavior. Utilization of multiple methodologies, such as focus group interviews and content analysis of the “quality” dimension of actual tweets, may provide deeper insights into Twitter users’ motivations for social networking and microblogging in relation to private disclosure. Different motivations for Twitter use may contribute to varying amounts of private information disclosure. Focus group interviews among Twitter nonusers and dropouts may provide qualitative data on their concerns about private disclosure and privacy management patterns different from those of active users. More specifically, the investigation of Twitter dropouts who stopped using Twitter due to privacy concerns will help elucidate (a) the process of privacy boundary turbulence, which refers to the violation of privacy boundary or conflicts about boundary expectations and regulation (Petronio, 2002); and (b) the phenomenon of leaked tweets, meaning the act of retweeting that “enables protected tweets to leak to the public sphere” (Meeder et al., 2010: 1) in social media. Future research can also use a longitudinal panel design to systematically examine (a) the relationship between the increase in tweets and followers and (b) the relationship between the amount of private disclosure and the expansion of collective boundaries and social capital via Twitter in dynamic privacy management processes.

Fourth, participants in the current study were not representative of the population. The median age of Twitter users was 31 years old in 2009 and 39 years old in 2010 (WSI Internet Marketing Services in Survey, 2011). Although Twitter use among 18–24-year-olds increased dramatically between May 2011 and February 2012 (Pew Internet & American Life Project, 2012), participants in the current study exclusively consist of undergraduate students with an average age of 20.14 years and with a range of five years, thus raising the issue of sample representativeness and generalizability. Furthermore, the homogeneity of the sample results in a restricted range of the measures and attenuates correlations among variables, thus leading to low estimates of factor loadings and correlations between factors (Fabrigar et al., 1999).

Last, this study did not empirically test online privacy mechanisms. The option to protect one’s tweets has not been integrated into the data collection procedure and data analyses. This study did not observe actual online behavior, but rather asked participants to think about whether they would engage in online private information disclosure. This lack of objective behavioral measures raises issues regarding ecological validity. Also, it would be interesting to examine different patterns and levels of private disclosure in multiple sub-components of the Twitter interface (profile, tweets, recent images, favorites, lists, etc.), but the current dataset contains no information about private disclosure in various sub-components of the Twitter interface. These agendas merit future attention.

This research underscores the theoretical importance of addressing private disclosure and virtual identity construction in social networking and microblogging-based social media. The current study’s illustration of Twitter’s private disclosure onion may serve as a vignette providing preliminary data for future pursuit in this strain of research.

Footnotes

Acknowledgements

The author thanks Drs Nicholas Jankowski and Steve Jones and the three anonymous reviewers for their valuable insights and constructive comments.

Funding

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Author biography

Seung-A Annie Jin (PhD, University of Southern California) is an assistant professor of Marketing Communication at Emerson College, USA. Her research interests include new media, communication technologies, consumer behavior, and computer-mediated communication (CMC).

References

Altman

Taylor

(1973) Social Penetration: The Development of Interpersonal Relationships. New York: Holt, Rinehart and Winston.

Arceneaux

Weiss

(2010) Seems stupid until you try it: press coverage of Twitter, 2006-9. New Media & Society 12(8): 1262–1279.

Back

Stopfer

Vazire

. (2010) Facebook profiles reflect actual personality not self-idealization. Psychological Science 21(3): 372–374.

Barry

(2003) Cultural and demographic correlates of self-reported guardedness among East Asian immigrants in the US. International Journal of Psychology 38(3): 150–159.

Blanchette

Johnson

(2002) Data retention and the panoptic society: the social benefits of forgetfulness. Information Society 18(1): 33–45.

boyd

(2010) Making sense of privacy and publicity. In: SXSW, Austin, TX, 13 March. Available at: http://www.danah.org/papers/talks/2010/SXSW2010.html (accessed 26 November 2012).

Chen

(2011) Tweet this: a uses and gratifications perspective on how active twitter use gratifies a need to connect with others. Computers in Human Behavior 27(2): 755–762.

Coopamootoo

Ashenden

(2011) A systematic evaluation of the communicability of online privacy mechanisms with respect to communication privacy management. In: Marcus

(ed.) Design, User Experience, and Usability, Pt II HCII 2011. Orlando, FL, USA: Springer, pp. 384–393.

Costa

McCrae

(1985) The NEO Personality Inventory Manual. Odessa, FL: Psychological Assessment Resources.

10.

Craddock

(1997) The measurement of privacy preferences within marital relationships: the relationship privacy preference scale. American Journal of Family Therapy 25(1): 48–54.

11.

Fabrigar

Wegener

MacCallum

. (1999) Evaluating the use of exploratory factor analysis in psychological research. Psychological Methods 4: 272–299.

12.

Goffman

(1959) The Presentation of Self in Everyday Life. New York: Doubleday Anchor.

13.

Greene

Derlega

Yep

. (2003) Privacy and Disclosure of HIV/AIDS in Interpersonal Relationships: A Handbook for Researchers and Practitioners. Mahwah, NJ: LEA Publishers.

14.

Higgins

Klein

Strauman

(1985) Self-concept discrepancy theory: a psychological model for distinguishing among different aspects of depression and anxiety. Social Cognition 3: 51–76.

15.

Hutchins

(2011) The acceleration of media sport culture: Twitter, telepresence and online messaging. Information, Communication & Society 14(2): 237–257.

16.

Jin

(2012a) The virtual malleable self and the virtual identity discrepancy model: investigative frameworks for virtual possible selves and others in avatar-based identity construction and social interaction. Computers in Human Behavior 28: 2160–2168.

17.

Jin

(2012b) “To disclose or not to disclose, that is the question:” A structural equation modeling approach to communication privacy management in e-health. Computers in Human Behavior 28(1): 69–77.

18.

Johnson

(2011) The effect of Twitter posts on students’ perceptions of instructor credibility. Learning, Media and Technology 36(1): 21–38.

19.

Joinson

Paine

(2007) Self-disclosure, privacy, and the Internet. In: Joinson

McKenna

KYA

Postmes

. (eds) The Oxford Handbook of Internet Psychology. Great Britain: Oxford University Press, pp. 237–252.

20.

Junco

Heiberger

Loken

(2011) The effect of Twitter on college student engagement and grades. Journal of Computer Assisted Learning 27(2): 119–132.

21.

McCroskey

McCain

(1972) The measurement of interpersonal attraction. Speech Monographs 41: 261–266.

22.

McCroskey

Richmond

Daly

(1975) The development of a measure of perceived homophily in interpersonal communication. Human Communication Research 1: 323–332.

23.

Marwick

boyd

(2010) I tweet honestly, I tweet passionately: Twitter users, context collapse, and the imagined audience. New Media & Society 13(1): 114–133.

24.

Meeder

Tam

Kelley

. (2010) RT@IWantPrivacy: widespread violation of privacy settings in the Twitter social network. In: Web 2.0 privacy and security workshop, IEEE symposium on security and privacy 20 May 2010, Oakland, California.

25.

Nippert-Eng

(2010) Islands of Privacy. Chicago, IL: University of Chicago Press.

26.

Nissenbaum

(2009) Privacy in Context: Technology, Policy, and the Integrity of Social Life. Palo Alto, CA: Stanford University Press.

27.

Petronio

(2000) The boundaries of privacy: praxis of everyday life. In: Petronio

(ed.) Balancing the Secrets of Private Disclosure. Mahwah, NJ: Erlbaum, pp. 37–49.

28.

Petronio

(2002) Boundaries of Privacy: Dialectics of Disclosure. Albany, NY: SUNY Press.

29.

Pew Internet & American Life Project (2012) Twitter use 2012. Available at: http://pewinternet.org/Reports/2012/Twitter-Use-2012/Findings.aspx (accessed 26 November 2012).

30.

Phua

Jin

(2011) “Finding a home away from home:” The use of SNSs by Asia-Pacific students in the United States for bridging and bonding social capital. Asian Journal of Communication 21(5): 504–519.

31.

Putnam

(2000) Bowling Alone: The Collapse and Revival of American Community. New York: Simon & Schuster.

32.

Robinson

(2007) The cyberself: the self-ing project goes online, symbolic interaction in the digital age. New Media & Society 9(1): 93–110.

33.

Tong

Van Der Heide

Langwell

. (2008) Too much of a good thing? The relationship between number of friends and interpersonal impressions on Facebook. Journal of Computer-Mediated Communication 13(3): 531–549.

34.

Utz

(2010) Show me your friends and I will tell you what type of person you are: how one’s profile, number of friends, and type of friends influence impression formation on social network sites. Journal of Computer-Mediated Communication 15(2): 314–335.

35.

Walther

(2007) Selective self-presentation in computer-mediated communication: hyperpersonal dimensions of technology, language, and cognition. Computers in Human Behavior 23(5): 2538–2557.

36.

Walther

Van Der Heide

Hamel

. (2009) Self-generated versus other-generated statements and impressions in computer-mediated communication: a test of warranting theory using Facebook. Communication Research 36: 229–253.

37.

Waters

Ackerman

(2011) Exploring privacy management on Facebook: motivations and perceived consequences of voluntary disclosure. Journal of Computer-Mediated Communication 17: 101–115.

38.

WSI Internet Marketing Services in Survey (2011) The average age of social media users is going up fast! Available at: http://www.wsiims.co.uk/blog/?p=530 (accessed 26 November 2012).