Getting Personal: Effects of Twitter Personalization on Candidate Evaluations

Personalization has cultivated a bad reputation in politics. Initially, scholarship on the personalization of politics focused on what was often called “candidate-centered” voting: the idea that citizens would vote based on a candidate's personality. Many scholars viewed this evaluative approach as irrational and heralded the value of issue stances over charisma (see Fenno 1978; Popkin 1991). Focusing on the personal, it seems, was problematic. Another iteration of the personalization in politics was also problematic and focused on the use of the “personal frame” in news coverage of women candidates. Such news coverage focused more on women's personalities and personal lives as compared to men's (e.g., Bystrom 1999; Devitt 1999). On its surface, such coverage does not appear detrimental. However, this framing would often emphasize women's roles as mothers and wives and use that framing to question women's experience, fitness for office, and whether they could juggle domestic and political responsibilities (Braden 1996). Personalization in both iterations elicited a sense of triviality: voters’ focus on persona was deemed as a trivial way to form an opinion, and women candidates were trivialized via a focus on their personal, not political, lives.

This study reexamines the role of personalization in politics by exploring how a candidate's personalized self-disclosure may affect voter evaluations. Via an experiment, this study examines the effects of personalized versus depersonalized candidate communication on the public's evaluations of vote intention, political issue competency, and character trait portrayal. In addition to analyzing the general effects of personalization, this study examines these effects for men and women candidates. As previously mentioned, women are more likely to receive personalizing news coverage, which can be detrimental, but it is possible that when women are the ones offering the personalization, outcomes could change. For example, some women have reappropriated this framing and used their personal experiences to buttress their candidacy. A politician once told Senator Patty Murray (D-WA) that she could not accomplish anything because she was “just a mom in tennis shoes.” Murray used this phrase in her campaign and believes it was effective because “many people identified with me being an average wife and mother, they wanted to see if I could bring some down-to-earth common sense to the United States Senate” (Braden 1996, 126). This study explores whether Murray's strategy is an effective campaign style and whether women can advantageously reappropriate personalization.

Additionally, this study examines the effects of personalization in campaign Twitter feeds. Research has found that candidates discuss their personal lives in 12% to 29% of their campaign tweets (Evans, Cordova, and Sipole 2014; Kruikemeier 2014; Meeks 2016)—suggesting Twitter personalization is common in campaigns. Further, in today's Internet-driven, 24/7 communication milieu, many candidates communicate online first and most often. In the words of scholar Kathleen Hall Jamieson, “[I]nstead of working from spin rooms, [campaigns] try to push their views through Twitter” to control discourse in the fast-paced environment (Ostrow 2012). This control is possible, in part, because social media like Twitter enable candidates to communicate directly with the public and bypass the news media and their potentially negative framing. For women candidates looking to control the narrative and reappropriate personalization, Twitter can be a valuable campaign platform.

PERSONALIZATION OF POLITICS

The personalization of politics has been a growing focus in political communication research, and scholars have examined the rise of candidate-centered politics in news coverage, candidate communication, and voter behavior (Jamieson 1988; Van Aelst, Sheafer, and Stanyer 2012). The concern that voters would vote based on personality over platforms formed the initial concerns with personalization. For example, in his review of the area, Hayes (2009) noted, “[C]andidate personality is taken as a substandard criterion for a vote choice …. These considerations are often seen as inferior to partisan loyalty, with parties serving as proxies for a constellation of policy priorities and issue positions” (232). Early theorists viewed politicians as people who enact policy and who thus should be evaluated based on policy, not extrapolicy aspects (Fenno 1978). These early conceptions of personalization mostly focused on an issues-versus-image dichotomy. Contemporary notions of personalization paint a broader perspective and characterize personalization as a multifaceted concept, with subcomponents such as individualization, privatization, and emotionalization (Van Aelst, Sheafer, and Stanyer 2012; Van Santen and Van Zoonen 2010). Individualization encompasses a “shift in media visibility from parities to individual politicians,” and privatization is a “shift in media focus from the politicians as occupier of a public role to the politicians as a private individual” (Van Aelst, Sheafer, and Stanyer 2012, 205). Emotionalization includes a focus on candidate's emotions and feelings (Van Santen and Van Zoonen 2010). This shift toward the individual is not necessarily seen as a move against issue affinity as a key voter criterion; rather, it “represents an important individuating influence on political impression formation” (McGregor, Lawrence, and Cardona 2016, 3). Further, Fenno (1978) argued that a focus on policy and extrapolicy aspects provides “a more complete view of what representation is all about” (242). This study builds on these conceptions and defines personalization as occurring when candidates self-disclose information about their personal lives or use personal experiences to create a connection with a campaign topic, such as a candidate's stance on education funding.

A major contributing factor to the personalizing shift in politics was the proliferation of the television, and this study argues that social media are continuing this trend. Regarding television, Hart (1999) noted that the television brought “persons of great magnitude into our own, very modest living rooms. They share themselves with us, persons whom they have never met” (27). This new level of intimacy influenced communication styles. According to Jamieson (1988), the television “invites a personal, self-disclosing style that draws public discourse out of a private self” (84). Contemporary research supports this trend, with politicians often mixing personal anecdotes with talking points on daytime talk shows and late-night comedy shows (see Vaughn and Goren 2013). The emergence and propagation of social media seem to mimic this drive toward personalization. Reflecting on Hart's quote, social media bring candidates not only into our living rooms, but into almost every nook and cranny of our daily, heavily-mediated lives, thanks to mobile devices. Reflecting on Jamieson's assertion, the norms of social media and the intimacy associated with these norms is changing communication styles. Walton and Rice (2013) argue that social media value sharing and self-disclosure over privacy and that both of these characteristics are norms in this environment. Because we expect others to self-disclose on social media, candidates can self-disclose on social media and thus follow the expected norms of the space, and in doing so, their self-disclosure could be seen as more natural versus being seen as pretense. Scholarship has found that candidates incorporate various forms of personalization on Twitter (e.g. Evans, Cordova, and Sipole 2014; Golbeck, Grimes, an Rogers 2010; Graham, Jackson, and Broersma 2014; Meeks 2016)—suggesting candidates have taken note of this norm to self-disclose. This study examines the effects of candidates’ self-disclosure via two competing streams of research: one that trumpets the benefits of personalization for women candidates, arguing that women can reappropriate personalization to their advantage, and one that argues that women cannot effectively interweave personalization into their campaign communication due to gender stereotypes and negative perceptions.

The first body of work recommends that women candidates personalize because it is associated with a number of benefits. First, personalization can foster certain beneficial psychological effects that positively contribute to relationship building. Self-disclosure is crucial to relationship development because it promotes trust, credibility, commitment, intimacy, and feelings of connection and liking (Collins and Miller 1994; Derlega et al. 1993). Research has found that personalization can positively influence these feelings of intimacy and connection because personalization elicits higher levels of parasocial interaction (PSI) and social presence. PSI is a one-way, pseudo relationship the audience forms with a mediated personality, and such relationships create a feeling of intimacy (Horton and Wohl 1956, 215). Social presence is the extent to which mediated communication simulates face-to-face interactions, which can create the feeling of connection and “being there” physically with the other person even though communication is occurring virtually (Nowak and Biocca 2003). Lee and Oh (2012) examined the effects of personalized tweets by male politicians in South Korea and found that for people who value interpersonal connections, personalized tweets heightened their perceptions of social presence and PSI. Further, Kruikemeier et al. (2013) found that personalized campaign websites prompted Dutch participants to feel as though politicians were more willing to listen to citizens and more open to ideas from citizens, and such websites invite people in for conversation. If the saying is true and politics is about relationships, then women could use personalization to help create and foster relationships with their constituents and potential voters.

Second, when candidates personalize their communication, they may be seen as more competent in their profession. Recent research has found that when organizational leaders start with warmth, they are perceived as generally more competent. Cuddy, Kohut, and Neffinger (2013) argue this effect is possible because “warmth is the conduit of influence: It facilitates trust and the communication and absorption of ideas …. Prioritizing warmth helps you connect immediately with those around you, demonstrating that you hear them, understand them, and can be trusted by them” (4). They suggest that one way to offer warmth is to share a personal story or experience—that is, to self-disclose. Further, Coffé and Theiss-Morse (2016) found that when candidates self-disclosed information about their occupational background, it positively affected voters’ perceptions of their issue competency, and this trend held when broken down by candidate gender. This research aligns with Rosh and Offerman (2013), who posit that task-relevant self-disclosure is a good way to build trust and engender rapport. Because personalization can include using personal experiences to create a connection with a campaign topic, it is possible that when a woman self-discloses information that is relevant to an issue—for example, talking about being a parent and how it gives her insight on the education system, or being a small-business owner and how it gives her insight on dealing with the economy—it can boost perceptions of women's issue competency.

This sense of connection and competence may contribute to the third factor: personalization is associated with winning. In Lee and Oh's (2012) experiment comparing personalized and depersonalized tweets, they found that people who value interpersonal connections were more likely to express vote intention for the candidate with personalized tweets. Further, McGregor, Lawrence, and Cardona (2016) examined mixed-gender gubernatorial races in 2014 and analyzed candidates’ Twitter and Facebook posts. According to their results, when personalization was present, the more personalizing candidate won in five of the seven races, and these five races included men and women winners. Given these outcomes, this study makes the following predictions:

H_1a:

Participants exposed to personalizing candidates will report higher levels of vote intention than participants exposed to depersonalizing candidates.

H_1b:

Participants exposed to personalizing candidates will report higher levels of issue competency than participants exposed to depersonalizing candidates.

Beyond the general advantages of personalization, the main focus of this study is to assess whether women can capitalize on these benefits. Some research suggests that the advantages of personalization that are often available to men can extend to women as well. First, some of the aforementioned studies have found positive results for women and men regarding the benefits of personalization for vote intention and issue competency (e.g., Coffé and Theiss-Morse 2016; McGregor, Lawrence, and Cardona 2016). Second, gender stereotypes have often created a hurdle for women seeking office, but recent research suggests contemporary voters do not apply such stereotypes. Stereotypically, the public perceives men as more competent in handling the economy, crime, and national security, whereas the public perceives women as more capable in handling health care, education, and reproductive issues (Dolan 2010; Herrnson, Lay, and Stokes 2003; Sanbonmatsu and Dolan 2009). These gendered connections present a bulwark for women because voters are more likely to support a woman who is seen as competent in masculinized issues than feminized issues—meaning women need to break stereotypes and bolster perceptions on masculine issues (Dolan 2010). Alternately, Brooks’ (2013) Leaders-Not-Ladies (LNL) theory asserts that the representational progress of women in elected office has helped to minimize gender stereotypes, prompting citizens to evaluate women as leaders, not as ladies, and thus evaluate women and men similarly. Brooks did not assess vote intention, but she did find that when news articles reported that senate candidates had 10 years of legislative experience, there were no significant differences between women and men regarding perceptions of issue competency. With the LNL theory in mind, combined with the benefits of personalization, this study presents the following hypotheses:

H_2a:

Participants exposed to personalizing men and women candidates will report higher levels of vote intention than participants exposed to depersonalizing men and women candidates.

H_2b:

Participants exposed to personalizing men and women candidates will report higher levels of issue competency than participants exposed to depersonalizing men and women candidates.

This first body of work and set of hypotheses predict that women, like men, can enjoy the advantages of personalization. The second body of work presents a counter to the LNL theory and predicts that women cannot effectively incorporate personalization into their campaigns. First, research has found that when women personalize, it can hurt perceptions of their viability. The Barbara Lee Foundation (2012) had participants read different profiles of women candidates that emphasized the candidate's (a) issue expertise, (b) previous experience in elected office, (c) previous work experience, or (d) personal background and upbringing. The profile that prompted the greatest perceptions of being qualified was the issues profile (Very Qualified = 62%; Total Qualified = 93%), followed by elected office (56%; 90%), work experience (39%; 80%), and personal (21%; 63%). The personalizing woman candidate was seen as the least qualified, which pushes back on the efficacy of Cuddy, Kohut, and Neffinger's (2013) “connect, then lead” approach. The Barbara Lee Foundation (2012) also found that 90% of participants said being qualified is important for women candidates, with 63% saying it is very important, and they found that 79% of participants said it was harder for women to appear qualified as compared to men, with 46% saying it is much harder for women. It would seem that women should avoid personalization because it creates barriers for being seen as qualified.

This research showcases the classic double bind for women. Women are historically and stereotypically perceived as engaging in “rapport talk” and “feminine communication styles” that are more personal and conversational, with a focus on establishing connections and relationships (Campbell 1989; Tannen 1990). But as the Barbara Lee Foundation research showed, personalization is detrimental to being perceived as qualified, and this sets up a double bind for women. Jamieson's (1995) femininity/competence double bind argues that women can meet societal expectations of femininity at the cost of being perceived as incompetent, or meet professional standards of competency and risk being perceived as not womanly enough. Therefore women candidates can personalize and be gender-congruent, but not office-congruent, or they can not personalize and be office-congruent, but not gender-congruent. Such a situation creates a lose-lose for women, which is not prevalent for men. Rather, men's adoption of personalization is received favorably. Åström and Karlsson (2016) examined Swedish politicians’ blogs and found that when women employed feminine communication styles, including personalization and interactivity, there was no positive influence; however, men gained influence by implementing feminine communication styles. Åström and Karlsson (2016) posit that because women politicians face negative gender stereotypes, “utilizing a feminine style of blogging is perhaps not the best strategy,” whereas for men “a feminine style of blogging might very well work as a compelling complement to the generally positive masculine stereotype,” which typically includes perceptions of strength, leadership, and confidence, as well as the aforementioned connection with masculine issues (13). Therefore, personalizing men can potentially have the best of both worlds, creating a distinct advantage for them over women.

This may explain why some research has found that men personalize and self-disclose more than women. ¹ For example, McGregor, Lawrence, and Cardona's (2016) analysis of mixed-gender gubernatorial races found that men personalized twice as much as women. Bystrom et al. (2004) examined campaign ads in mixed-gender races between 1990 and 2002, and found that men were more likely to include their family in the ads than women, and Stalsburg and Kleinberg (2015) examined campaign websites in 2008 and 2010 and found that men were more likely than women to emphasize their family, especially in photos. Further, Fridkin and Kenney (2014) examined senatorial candidates’ webpages and found that 96% of men mentioned their family, while only 77% of women emphasized their family on their biography pages. Fridkin and Kenney (2014) argue, “Men want to stress their familial connections as a way of demonstrating their communal characteristics,” such as being compassionate and empathetic, whereas for women, “emphasizing family is perceived as more of a liability” (48). Fridkin and Kenney (2014) found that women were more likely than men to emphasize their political experience, committee work in the U.S. Senate, and legislative accomplishments. This emphasis on issues and elected office experience is in keeping with the Barbara Lee Foundation's (2012) recommendations regarding women and being perceived as qualified. With this research in mind, this study proposes a set of hypotheses that counter H _2a/b :

H_3a:

Participants exposed to personalizing men will report higher levels of vote intention than participants exposed to personalizing women and depersonalizing men and women.

H_3b:

Participants exposed to personalizing men will report higher levels of issue competency than participants exposed to personalizing women and depersonalizing men and women.

Finally, it is important to examine the link between personalization and character traits. As previously mentioned, self-disclosure promotes trust, credibility, connection, and liking (Collins and Miller 1994; Derlega et al. 1993). Cuddy, Kohut, and Neffinger (2013) suggested that starting with warmth, such as self-disclosure, can prompt feelings of trust and competence. Cuddy, Kohut, and Neffinger's (2013) “connect, then lead” approach may also suggest that personalization can promote positive evaluations of agentic traits—for example, leadership and strength. Fridkin and Kenney (2014) argued that men could discuss their family in an attempt to showcase communal traits, such as compassion and empathy. Further, Han (2009) found that people who self-disclosed in political appeals were perceived as more likeable. Based on this research, this study predicts that

H₄:

Participants exposed to personalizing candidates versus depersonalizing candidates will rate the candidates more positively on likeability, competency, and communal and agentic traits.

Based on LNL theory, the positive connection between personalization and traits may extend to women. Previous research has found that agentic traits—for example, decisiveness and confidence—are stereotypically more strongly ascribed to men, whereas communal traits—for example, honesty and collaboration—are more strongly ascribed to women (Eagley and Karau 2002; Fridken and Kenney 2014). These gendered connections can be detrimental to women because voters perceive agentic traits as more important than communal traits, and news coverage often employs more of a negative tone when discussing some communal traits and a more positive tone when discussing some agentic traits (Conroy 2015a; 2015b). However, in support of LNL theory, which posits a decrease in the application of gendered stereotypes, Brooks (2013) found that when news coverage reported senate candidates had 10 years of legislative experience, there were no significant differences between women and men for perceptions of trait portrayal. Given this support for LNL, women may be on a more even playing field with men when it comes to capitalizing on personalization. Thus

H₅:

Participants exposed to personalizing men and women will rate these candidates more positively on likeability, competency, and communal and agentic traits than participants exposed to depersonalizing men and women candidates.

A competing stream of research suggests that women may not reap rewards from personalizing when it comes to traits. Being likeable is crucial for all candidates, but it is especially crucial for women. The Barbara Lee Foundation (2010) found that voters were willing to vote for a man who they thought was qualified but not likeable, but they were less willing to vote for a woman who they found qualified but not likeable. In other words, there was a positive correlation between competency and likeability for women that was not present for men. Overall, the Barbara Lee Foundation (2010; 2012) found a host of other traits associated with likeability and perceptions of being qualified for women, including agentic and communal traits such as confidence, strength, honesty, and collaboration. The gendered associations between agentic traits and men, as well as communal traits and women, create another double bind for women. The Barbara Lee Foundation (2012) argues, “Women often start their campaigns with their personal stories, which makes them likeable and in touch, but often does not do much to establish their qualifications and credibility” (5). Women may personalize and be seen as likeable but not qualified/lacking agentic qualities, whereas men may be able to personalize to show their communal qualities while still relying on their stereotypical association with agentic traits. This line of work prompts the final hypothesis:

H₆:

Participants exposed to personalizing men will rate these candidates more positively on likeability, competency, and communal and agentic traits than participants exposed to personalizing women, and depersonalizing men and women candidates.

METHODS

The experiment employed a 2 × 2 posttest design: tweet type (personalized or depersonalized) by candidate gender (man or woman). The experiment was conducted online via Survey Monkey in the fall of 2013. Participants were also obtained via Survey Monkey via their pool of opt-in adult participants. The final sample consisted of 843 participants. The demographics were as follows: 52.2% of respondents were female; the median age was “45–54 years old,” and the mode was “55–64.” For education, 7.0% lacked a college education, 31% had some college, and 62% had a four-year degree or more. For race/ethnicity, 85.1% of respondents identified as white, 4.1% black or African American, 3.6% Hispanic American, 3.1% Asian/Pacific Islander, 0.9% American Indian or Alaskan Native, and 3.3% identified as “other.” Just over half, 56.6%, made less than $75,000 a year. For party affiliation, the sample was 45.6% Democrat, 31.6% Republican, and 22.7% did not identify with either major party. Political ideology was measured on a seven-point scale, ranging from extremely conservative to extremely liberal (M = 4.17, SD = 1.671). According to U.S. Census regions, the geographical distribution of participants was: South 28.1%; West 26.7%; Midwest 23%; Northeast 22.1%. The vast majority of the sample used some form of social media (86.2%), and participants were active social media users: nearly half of participants used social media several times a day (49.8%), and an additional quarter used social media about once a day (24.6%). Finally, there were no significant demographic differences between stimuli conditions.

The procedure for the experiment was as follows. After being randomly assigned, participants were instructed that they would read a random selection of tweets from a candidate running for the U.S. Senate. The gender of the candidate was conveyed via first names that are socially perceived as male or female in the United States. The male candidate was named Steve Adams, and the female candidate was named Sarah Adams. Participants then read a total of 14 mocked-up tweets from the candidate's Twitter feed. The tweets were from a hypothetical candidate with the username @AdamsForCongress. Prior to the experiment this username and all usernames featured in the stimuli were checked to ensure they were not currently in use by any Twitter users. The tweets included 10 treatment and four “filler” tweets (see the supplementary material for the stimuli; note the Twitter formatting of the stimuli has been removed to ease readability). ² Treatment tweets featured personalized or depersonalized content, with personalized tweets including some form of self-disclosure or connecting some aspect of the candidate's private life to tweet content. Prior to this experiment, a content analysis of thousands of tweets by U.S. Senate candidates was conducted, and personalized tweets were pulled from this examination and modified slightly to fit the parameters of this experiment, thus adding more realism to the stimuli. An example of a depersonalized tweet included, “Most farms are family owned. Farming is a labor of love. Support is needed for this vital part of America.” The personalized version of this tweet was, “Most farms are family owned—like my grandpa's. I saw firsthand that farming is labor of love. Support this vital part of America.” The main topic of the treatment tweets was held constant. For example, the tweets above both focus on agriculture. There were also four “filler” tweets to make the Twitter feeds more realistic. Filler tweets were held constant, and included information regarding campaign events that did not include any personalization. An example of a filler tweet was, “Be one of the first to watch our new campaign ad, and RT to pass it along! AdamsForCongress.com/WorkingTogether.” The average length of the tweets in the conditions was similar (personalized: M = 114.50, SD = 31.147; depersonalized: M = 110.29, SD = 30.555).

Once participants finished reading the tweets, they filled out a questionnaire. Descriptives for the evaluative criteria are in Tables 1 and 2. The following sections featured a matrix question format, with a Likert scale running across the columns. The rows were randomized within each matrix to avoid order bias. To measure vote intention, participants were asked to what extent they agreed with two statements, adapted from Lee and Oh (2012), and then provided seven-point, Likert scales ranging from “strongly disagree” (1) to “strongly agree” (7). The statements included, “I would like this candidate to run in the next election” and “I would vote for this candidate if they ran in the next election.” To examine the likeability/competency double bind, participants were asked to what extent they agreed with two statements using the same scale: “I think Adams is likeable” and “I think Adams is competent.”

Table 1. Descriptive statistics and correlations for vote intent and traits

Note: Criteria 1–3 were based on a seven-point scale, and 4 and 5 on a five-point scale. All correlations were significant at p < .01.

Table 2. Descriptive statistics and correlations for issue competency criteria

Note: Issue evaluations based on a five-point scale. All correlations were significant at p < .01.

To measure perceptions of issue competency and trait portrayal, participants were asked to assess the candidate's handling of eight issues and eight traits. The issues are common, topical issues in American politics and included the economy, taxes, crime, national security, health care, education, women's issues, and environment. The eight traits aligned with Eagley and Karau's (2002) classification of agentic and communal traits. Agentic traits included strength, leadership, decisiveness, and confidence. Communal traits included compassion, collaboration, honesty, and friendliness. In both cases participants were provided five-point, Likert-type scales ranging from “not at all” (1) to “extremely” (5). The Cronbach's alphas for all composite variables in Table 1 met standard conventions for internal consistency (Nunnally 1978).

Prior to analysis, three questions, adapted from Lee and Oh (2012), were examined as a manipulation check to ensure that tweets were perceived as intended and that participants were attentive to the stimuli. Participants were asked how they would evaluate Adams's tweets based on three seven-point, semantic differential scales that contained public/private (M = 3.09, SD = 1.648), nonintimate/intimate (M = 3.86, SD = 1.657), and impersonal/personal (M = 4.16, SD = 1.700) as the semantic anchors. The order of the semantic anchor points was flipped for each participant to avoid within-item bias. According to literature, we would expect participants receiving personalized tweets to be more likely than those receiving depersonalized tweets to rate Adams's tweets as more private, intimate, and personal, and thus produce higher means. The results of independent sample t-tests in Table 3 show this expectation was met for all three questions. The manipulation of the tweet content appeared to function as expected.

Table 3. Manipulation checks by treatment

Note: Measures were based on seven-point, semantic differential scales. Higher means indicate that the tweets were considered more private, intimate, or personal.

The intercorrelation matrix for the evaluative criteria is provided in Tables 1 and 2. Because the conceptually-linked dependent variables in this study had strong correlations, ANCOVAs were conducted instead of MANCOVAs (see Rovai, Baker, and Ponton 2014). Given the conceptual focus on gender and research that shows that a participant's party affiliation can alter candidate evaluations regarding likeability, competency, issues, and traits (e.g., Hayes 2011; Meeks and Domke 2016), the participants’ gender and party were controlled in the analysis.

RESULTS

The first set of hypotheses predicted that participants exposed to personalizing candidates would report higher levels of vote intention ( H _1a ) and issue competency ( H _1b ) than participants exposed to depersonalizing candidates. ANCOVA results for vote intention were not significant, F(1, 812) = 1.131, n.s., ηp² = .001 (personalized: M = 3.76, SD = 1.550; depersonalized: M = 3.65, SD = 1.523). Therefore H _1a was not supported. The personalizing candidate had the highest mean in every issue comparison, but the ANCOVA results for four issues were not significant, including taxes, national security, crime, and environment. Results were significant for the following issues. First, the personalizing candidate was seen as more competent in handling the economy (M = 2.89, SD = 1.076) than the depersonalizing candidate (M = 2.73, SD = 1.026), F(1, 810) = 4.937, p = .027, ηp² = .006. Second, the personalizing candidate was seen as more competent in handling health care (M = 2.91, SD = 1.017) than the depersonalizing candidate (M = 2.76, SD = 1.046), F(1, 810) = 4.036, p = .045, ηp² = .005. Third, the personalizing candidate was seen as more competent in handling education (M = 2.98, SD = 1.006) than the depersonalizing candidate (M = 2.75, SD = 1.079), F(1, 805) = 10.230, p = .001, ηp² = .013. Finally, the personalizing candidate was seen as more competent in handling women's issues (M = 3.13, SD = 1.060) than the depersonalizing candidate (M = 2.82, SD = 1.061), F(1, 806) = 18.172, p = .000, ηp² = .022. Thus, there was partial support for H _1b : The personalizing candidate was seen as more competent than the depersonalizing candidate in handling economy, health care, education, and women's issues.

The next set of analyses examined competing hypotheses: H ₂ predicted that personalizing women and men candidates would receive higher evaluations regarding vote intention ( H _2a ) and issue competency ( H _2b ) than depersonalizing men and women. Alternately, H ₃ predicted that personalizing men would receive higher evaluations than all of other candidate types for vote intention ( H _3a ) and issue competency ( H _3b ). Personalizing men were rated highest for vote intention, closely followed by personalizing women, but ANCOVA results were not significant, F(1, 812) = .632, n.s., ηp² = .002. Therefore neither H _2a nor H _3a were supported.

Turning to issue competency, Table 4 provides the means and standard deviations for each comparison. Perceptions of taxes and environment were not significant. Perceptions regarding the other six issues provided mixed support for H _2b and H _3b . Results for education supported H _2b : personalizing men and women were viewed as more competent than depersonalizing men and women, F(3, 805) = 3.473, p = .016, ηp² = .014. For education, personalizing men were seen as more competent than depersonalizing men (p = .033) and women (p = .053), and personalizing women were seen as more competent than depersonalizing men (p = .010) and women (p = .019).

Table 4. Means for men and women by personalization for issue competency

Note. Criteria based on a five-point scale.

Evaluations for economy and health care supported H _3b : Personalizing men were seen as more competent than any other candidate type. For the economy, personalizing men were seen as more competent than personalizing women (p = .028) and depersonalizing men (p = .030) and women (p = .001), F(3, 810) = 3.775, p = .010, ηp² = .014. For health care, personalizing men were seen as more competent than personalizing women (p = .040) and depersonalizing men (p = .035) and women (p = .004), F(3, 810) = 2.994, p = .030, ηp² = .011. Results for national security and crime also supported H _3b to some extent. For national security, personalizing men were seen as more competent than personalizing women (p = .028) and depersonalizing women (p = .031), F(3, 806) = 2.206, p = .086, ηp² = .008. However, there was no significant difference between personalizing and depersonalizing men. For crime, personalizing men were seen as more competent than personalizing women (p = .018) and depersonalizing women (p = .025), F(3, 806) = 2.754, p = .042, ηp² = .010. There was no significant difference between personalizing and depersonalizing men. ³ These four issues provide partial support for H _3b .

Finally, results for women's issues did not support either hypothesis, F(3, 806) = 11.046, p = .000, ηp² = .040. For women's issues, personalizing women were seen as more competent than personalizing men (p = .003) and depersonalizing men (p = .000) and women (p = .001). Further, personalizing men were seen as more competent than depersonalizing men (p = .007), and depersonalizing women were seen as more competent than depersonalizing men (p = .018). Thus, all of the other candidate types were seen as more competent on this issue than depersonalizing men, and personalizing women were seen as the most competent on women's issues.

The next hypothesis predicted that participants exposed to personalizing candidates versus depersonalizing candidates would rate the candidates more positively regarding likeability, competency, and agentic and communal traits ( H ₄ ). Results for perceptions of being competent were not significant, but results were significant for the other perceptions and supported H ₄ . Personalizing candidates (M = 4.76, SD = 1.476) were perceived as more likeable than depersonalizing candidates (M = 4.51, SD = 1.370), F(1, 813) = 5.637, p = .018, ηp² = .007. Personalizing candidates (M = 3.14, SD = .931) were perceived as more likely to portray agentic traits than depersonalizing candidates (M = 2.99, SD = .955), F(1, 810) = 5.302, p = .022, ηp² = .007. Personalizing candidates (M = 3.22, SD = .902) were perceived as more likely to portray communal traits than depersonalizing candidates (M = 2.95, SD = .889), F(1, 810) = 17.336, p = .000, ηp² = .021.

The next set of analyses examined competing hypotheses: H ₅ predicted that personalizing women and men candidates would receive higher evaluations regarding likeability, competency, and agentic and communal traits than depersonalizing men and women. Alternately, H ₆ predicted that personalizing men would receive higher evaluations than all of the other candidate types for these traits. Table 5 provides the means and standard deviations for each comparison. Personalizing men and women had the highest means across the board, but results for three of the comparisons were not significant: likeability, F(1, 813) = 1.964, n.s., ηp² = .007; competency, F(1, 811) = .687, n.s., ηp² = .003; agentic traits, F(1, 810) = 1.843, n.s., ηp² = .007. Perceptions regarding communal traits were significant, F(1, 810) = 7.021, p = .000, ηp² = .026. Personalizing women were seen as more likely to portray communal traits than depersonalizing men (p = .004) and women (p = .000), and personalizing men were perceived as more likely to portray communal traits than depersonalizing men (p = .003). Overall, perceptions of communal traits supported H ₅ because personalizing men and women were seen as more likely to portray communal traits than depersonalizing candidates.

Table 5. Means for men and women by personalization for character traits

Note. Likeability and competency were based on a seven-point scale, and agentic and communal traits were based on a five-point scale.

DISCUSSION

Whether scholars see it as a boon or bane, politics is getting more personal. The television elicited a more personal, self-disclosing communicative style (Jamieson 1988), and social media appears to have continued this communicative trend and may have even accelerated it. This study sought to explore the effects of candidates’ self-disclosure with a central focus on the effects of women candidates’ self-disclosure. Due to men's historic and contemporary dominance in American politics, men have experienced a number of electoral advantages as compared to women. When men self-disclose, they can be perceived as more likeable and compassionate while also maintaining perceptions of competence due to positive male stereotypes (Åström and Karlsson 2016; Fridkin and Kenney 2014). Women, on the other hand, have often faced a double bind with personalization: they can self-disclose and gain likeability but lose perceptions of competence, or they can avoid self-disclosure and come off as competent but not likeable. This study sought to examine whether this double bind still stands and whether women can advantageously personalize like men. Overall, this analysis yielded three patterns, which focus on the effects of personalization in general and for women and men separately. First, in the aggregate and once divided by candidate gender, there was no significant relationship between personalization and vote intention. In other words, participants were not more likely to vote for personalizing candidates over depersonalizing candidates. Thus, even though personalization can create a sense of connection between candidates and citizens (Cuddy, Kohut, and Neffinger 2013; Kruikemeier et al. 2013; Lee and Oh 2012) and has been associated with winning (Lee and Oh 2012; McGregor, Lawrence, and Cardona 2016), this study did not find a significant positive relationship between personalization and vote intent for women or men.

Second, personalization positively contributed to perceptions of issue competency and trait portrayal. It is important to emphasize that the issues and traits discussed in all Twitter feed conditions were the same. Therefore, the presence of personalization, not additional policy or character statements, drove the more favorable impressions. Regarding issues, the personalizing candidate was seen as more competent than the depersonalizing candidate in handling economy, health care, education, and women's issues. That personalization was effective on these particular issues is important for two reasons. First, according to Gallup, the public consistently views the economy as the most important problem/issue for the United States, and health care is commonly ranked in the top four issues (Saad 2015). In this experiment, personalizing candidates were seen as more effective in handling these omnisalient issues. Second, though education and women's issues are not typically ranked as top issues in the United States, they are highly topical, and candidates address these issues at various levels of office. In 2016 we saw presidential and down-ticket candidates routinely discuss free or affordable higher education, Common Core, reproductive rights, fair pay, and more. Overall, these results suggest that personalization is a valuable addition to candidates’ discussion of salient, topical issues.

Regarding traits, personalizing candidates were seen as more likeable and more likely to portray agentic and communal traits than depersonalizing candidates. These results support previous work that found that self-disclosure was linked with honesty, competence, and likeability (Collins and Miller 1994; Cuddy, Kohut, and Neffinger 2013; Derlega et al. 1993; Han 2009). Being likeable is a key attribute for candidates. Likeable candidates are praised, and unlikeable candidates struggle. The results of this study suggest that personal self-disclosure could aid struggling candidates in perceptions of likeability. Further, personalization can aid in evaluations of agentic and communal traits. Agentic traits, such as strength and decisiveness, are often associated with leadership, and the public often privileges these traits when considering elected officials (Conroy 2015a; Hayes 2011). Alternately, voters also want candidates who demonstrate integrity and empathy, which are communal traits (Conroy 2015a). Personalizing candidates were viewed as more effective at balancing and fulfilling these dual requests from voters.

Third, once gender was considered, the results for issues and traits became more nuanced, suggesting that the advantages of personalization do not evenly apply to men and women. This portion of the study was guided by competing hypotheses and associated bodies of research. The first line of work argued that personalizing women could obtain the same benefits as personalizing men, and thus would receive higher/more positive evaluations than depersonalizing men and women. This area of work received some support in this study. Personalizing men and women were viewed as more competent than depersonalizing men and women regarding education. Further, personalizing men and women received higher evaluations than depersonalizing men and women across all of the traits, though only perceptions of communal traits produced significant differences. This support suggests that women candidates can effectively personalize and have their own “mom in tennis shoes” moment like Sen. Murray. The second line of work argued that due to factors like Jamieson's (1995) femininity/competency double bind that is present for women and not men, personalizing women would not reap the same rewards as personalizing men. Thus, personalizing men would receive higher evaluations than personalizing women and depersonalizing candidates. Results also provide partial support for this area of work. Personalizing men were seen as more competent than any other candidate type in their handling of the economy and health care. Further, personalizing men were seen as more competent than personalizing and depersonalizing women regarding their ability to handle crime and national security. The economy, health care, and national security are routinely seen as the most important issues for Americans (Saad 2015). Therefore even though the content of the tweets was the same and only the candidate's gender differed, personalizing men had an advantage over personalizing women regarding these highly salient issues. Future research could examine whether men's personalizing advantage persists or is mitigated by other factors, such as the candidate's party affiliation, incumbency status, race, ethnicity, or type of office.

These findings are not without limitations. First, there may be limits to the representativeness and generalizability of the study due to the use of Survey Monkey. People who volunteer to take part in their service may share some unknown underlying characteristic that could affect the results. However, at the time of this study, 30 million people take Survey Monkey surveys each month, and using their services provided a relatively diverse sample in terms of demographics. This diversity does not completely assuage the limitations of the sample, but along with random assignment of participants, it does add validity to this study's findings. Further replication with other samples is necessary. Second, as with any experiment in which the stimulus is altered between conditions, there is the potential for an information effect. To minimize this potential, the focus of each personalized and depersonalized tweet was kept the same—for example, both discussed the same political issue. Finally, the effect size for most of the significant findings would be categorized as “small” based on partial eta squared scales (see Rovai, Baker, and Ponton 2014). This categorization does not diminish this study's findings, but it suggests that alterations to the stimuli may be needed to see larger effects. Tweets can be no more than 140 characters in length, and this study featured 14 tweets. This level of exposure may not be enough to spur larger effects or shift perceptions of vote intention. Future work could test increasing the overall number of tweets and the number of treatment tweets. Research could also examine whether more intense self-disclosure can increase effects—though it is important to keep in mind that perceptions of over-sharing may inhibit the effectiveness of personalization. Collins and Miller's (1994) meta-analysis found that the effects of self-disclosure depend on factors such as situation, timing, and volume and type of information shared—for example, high/low intimacy or negative/positive valence. More work is needed to examine how voter perceptions and evaluations change due to these factors.

Finally, it is important to note an additional gendered dynamic in these findings. As previously mentioned, agentic traits and issue competency on the economy, crime, and national security are more positively associated with men, and communal traits and issue competency on health care, education, and reproductive rights are more positively associated with women (Dolan 2010; Eagley and Karau 2002). If we look across the results, we find that personalizing men were seen as (a) more competent than any candidate type for economy and health care, (b) more competent than women on crime and national security, and (c) equally competent to personalizing women on education and communal traits. To a varying extent, personalizing men had an advantage on issues ascribed to men, issues ascribed to women, and traits ascribed to women. Personalizing men were able to transcend certain stereotypical boundaries and fulfill several evaluative criteria. Personalizing women also had positive evaluations, which showcases that women can reappropriate personalization and succeed, but that success was confided to stereotypical areas. Personalizing women were seen as (a) equally competent to personalizing men and more competent than depersonalizing candidates for education and communal traits, and (b) more competent than all candidate types for women's issues. Personalization for women appears to be a favorable approach, but not transcendent. These findings align with Åström and Karlsson's (2016) results, and their theory that men's adoption of a feminine communication style may complement a positive masculine stereotype that is not afforded to women. Further replication of this study is needed to determine whether these observations are indicative of larger phenomena, but these findings suggest that while women can effectively reappropriate personalization, men's appropriation of this communication style may result in greater rewards.