Current issues essay
Personality and Social Psychology Bulletin 2017, Vol. 43(1) 60 –70 © 2016 by the Society for Personality and Social Psychology, Inc Reprints and permissions: sagepub.com/journalsPermissions.nav DOI: 10.1177/0146167216675332 pspb.sagepub.com
Article
Self-expression values have increased substantially in the last four decades, especially among younger cohorts (Inglehart, 2008; Inglehart & Oyserman, 2004). Self- expression is defined as “assertion of one’s individual traits” (Self-Expression, n.d.) and individuals have dual motivations to have their traits be seen both positively and accurately (Goffman, 1959; McKenna & Bargh, 1999; Swann, 1983). Importantly, as self-expression values are increasing, the mediums for self-expression are changing, with a rise in the use of social media platforms. This raises the critical ques- tion of whether people are able to effectively express them- selves in such contexts. The present research investigates whether the information provided on the widely used social media site Twitter enables individuals to elicit positive and accurate perceptions from unacquainted others regarding important characteristics, including self-views, relationship orientations, and behavioral tendencies.
Over 288 million active users are on Twitter, who collec- tively tweet an average of 500 million messages per day. In the United States, 23% of adult Internet users and 19% of the adult population use Twitter. In recent years, the percentage of users has increased, particularly among individuals aged 18 to 29, with 37% of this cohort using the platform (Duggan, Ellison, Lampe, Lenhart, & Madden, 2014). Although researchers
have identified personality correlates of behavior on Twitter (Golbeck, Robles, Edmondson, & Turner, 2011; Quercia, Kosinski, Stillwell, & Crowcroft, 2011), research has not examined whether perceivers are able to detect Twitter users’ personality. Given the self-expression attempts of such a large swatch of users, we investigated whether Twitter users are able to elicit positive and accurate perceptions from those who read their tweets.
We define accuracy in the current study as distinctive self- other agreement—the extent to which perceivers’ impressions of an individual’s (target’s) personality (specifically, impulsiv- ity, self-esteem, and attachment style) correspond to the tar- get’s own self-reported traits, controlling for the normative profile of responses on these traits. This form of tracking accu- racy (Fletcher & Kerr, 2010) assesses the extent to which per- ceivers are able to accurately discern a target’s unique patterning of traits (e.g., is the target higher in self-esteem than
675332PSPXXX10.1177/0146167216675332Personality and Social Psychology BulletinOrehek and Human research-article2016
1University of Pittsburgh, PA, USA 2McGill University, Montreal, Québec, USA
Corresponding Author: Edward Orehek, University of Pittsburgh, 210 S. Bouquet St., Pittsburgh, PA 15260, USA. Email: orehek@pitt.edu
Self-Expression on Social Media: Do Tweets Present Accurate and Positive Portraits of Impulsivity, Self-Esteem, and Attachment Style?
Edward Orehek1 and Lauren J. Human2
Abstract Self-expression values are at an all-time high, and people are increasingly relying upon social media platforms to express themselves positively and accurately. We examined whether self-expression on the social media platform Twitter elicits positive and accurate social perceptions. Eleven perceivers rated 128 individuals (targets; total dyadic impressions = 1,408) on their impulsivity, self-esteem, and attachment style, based solely on the information provided in targets’ 10 most recent tweets. Targets were on average perceived normatively and with distinctive self-other agreement, indicating both positive and accurate social perceptions. There were also individual differences in how positively and accurately targets were perceived, which exploratory analyses indicated may be partially driven by differential word usage, such as the use of positive emotion words and self- versus other-focus. This study demonstrates that self-expression on social media can elicit both positive and accurate perceptions and begins to shed light on how to curate such perceptions.
Keywords self-expression, first impressions, accuracy, positivity, social media, Twitter
Received February 8, 2016; revision accepted September 7, 2016
Orehek and Human 61
impulsivity?), relative to the average person or, equivalently, to determine a target’s level on a given item relative to other targets’ self-reported levels (e.g., is the target higher in self- esteem and impulsivity than other targets? Biesanz, 2010; Biesanz & Human, 2010; Kenny & Winquist, 2001).
Of note, by examining the extent to which perceivers’ impressions of the target correspond to the normative profile (in addition to their distinctive self-reported profile), we also obtain an indicator of the positivity of impressions. This is because of the normative-desirability confound (Wood & Furr, 2016): the tendency for the normative profile to be highly positive and socially desirable (Borkenau & Zaltauskas, 2009). Thus, viewing a target more normatively implies viewing that target more positively (e.g., as higher in self-esteem than impul- sivity). Note that accuracy and positivity in interpersonal per- ceptions can be independent of one another (e.g., Funder & Colvin, 1997; Gagné & Lydon, 2004), including when posi- tivity is indexed as normativity (Human & Biesanz, 2012). Thus, this approach enables us to test whether tweets can simultaneously elicit both accurate impressions (distinctive self-other agreement) and positive impressions (normative agreement).
Is it likely that people are able to accurately distill the person behind the messages on Twitter? On one hand, this seems difficult: Tweets are limited to 140 characters and remove many forms of non-verbal communication present in more dynamic impression formation contexts. On the other hand, research on thin slices has found that people are able to accurately judge personality based on minimal information, such as very brief video clips (Ambady & Rosenthal, 1992). When examining perceptions based on social media, a simi- lar picture has emerged. For example, perceivers are able to accurately perceive others’ personality traits after viewing their Facebook profiles (Back et al., 2010) and personal web- pages (Vazire & Gosling, 2004). Furthermore, personality traits can also be accurately detected based on textual infor- mation alone, including creative writing (Küfner, Back, Nestler, & Egloff, 2010) and autobiographical essays (Borkenau, Mosch, Tandler, & Wolf, 2016).
In addition, tweets may in some ways facilitate accurate self-expression. People self-disclose on the internet without strong privacy concerns (Tufekci, 2008), and self-expression on the Internet may provide an opportunity for individuals to share aspects of themselves with less fear of rejection or disapproval than face-to-face interactions with their close social network (Bargh, McKenna, & Fitzsimons, 2002). Furthermore, communicating to a large and diverse set of viewers limits a person’s ability to tailor their message to specific audience members (Marwick & Boyd, 2011), which may increase the expression of stable personality traits rather than aspects of the self that vary across interaction partners. On social media platforms, users generate content them- selves, which allows for self-expression both explicitly through self-disclosure and implicitly through word usage (VanLear, Sheehan, Withers, & Walker, 2005). Overall, then,
it is likely that individuals will be perceived with some accu- racy on Twitter.
In addition to being perceived accurately, is it likely that individuals will be seen positively on Twitter? Given that first impressions generally tend to involve quite high levels of normative agreement (e.g., Human & Biesanz, 2011), it seems plausible that impressions based on social media would similarly elicit normative and positive impressions. Indeed, Twitter may be an ideal platform for curating posi- tive perceptions, as it enables individuals to have high con- trol in managing how they present themselves and such control of self-presentation has been found to foster more positive impressions (Human, Biesanz, Parisotto, & Dunn, 2012). We therefore predict that individuals will also be seen positively on the basis of their tweets, although it is possible that positivity may not be quite as high as impressions based on more dynamic information (Schroeder & Epley, 2015).
In addition to predicting that individuals on average will be able to express themselves both positively and accurately using social media, we also expect that there will be individ- ual variability in the tendency to do so. Indeed, in face-to-face interactions, individuals substantially vary in how accurately and normatively they are perceived (Human & Biesanz, 2013). Thus, extending prior research on personality percep- tions on social media, we examine not only whether impres- sions are accurate and positive but also how much people vary in how accurately and positively they are viewed. This study will also take a preliminary, exploratory look at how it is that individuals may come to elicit positive and accurate percep- tions from others through social media by examining the con- tent of tweets, particularly word usage. The types of words people use can reveal individuals’ personality traits and other internal experiences (e.g., Küfner et al., 2010; Mehl, Gosling, & Pennebaker, 2006; Rude, Gortner, & Pennebaker, 2004), including in tweets (De Choudhury, Gamon, Counts, & Horvitz, 2013), suggesting that word usage may be one way personality is detected and evaluated on Twitter.
Addressing these questions could have important implica- tions for understanding self-expression values, communica- tion via social media, and for psychological and social well-being. Indeed, being perceived accurately and posi- tively can have positive social outcomes in first impressions contexts, promoting relationship development (Human, Sandstrom, Biesanz, & Dunn, 2013), and in close relation- ships, promoting relationship quality and longevity (Murray, Holmes, & Griffin, 1996; Neff & Karney, 2005). Being accu- rately perceived in first impressions is also associated with better psychological well-being (Human & Biesanz, 2011, 2013; Human, Biesanz, Finseth, Pierce, & Le, 2014). Knowing whether and how Twitter can be used as an effec- tive medium for positive and accurate self-expression would therefore be informative with respect to whether the invest- ment in self-expressing on this platform is beneficial for tar- gets, and whether basing personality perceptions on such information is warranted for perceivers.
62 Personality and Social Psychology Bulletin 43(1)
Method
Overview
In the present study, 128 undergraduate Twitter users (tar- gets) completed a 105-item battery of personality items and provided their last 10 tweets. We chose trait measures designed to tap targets’ behavioral tendencies, views of the self, and perspectives on interpersonal relationships. Specifically, we included items from scales assessing impul- sivity, self-esteem, and attachment styles. Preliminary evi- dence suggests that tweets are likely to contain content related to self-esteem and belonging motives (Selim, Long, & Vignoles, 2014), suggesting that these are relevant domains in which to test personality perception accuracy. Eleven per- ceivers read the provided set of 10 tweets from a target and rated that target on the same 105 items. No other identify- ing information, such as gender or Twitter handle, was provided.
Targets
A total of 128 undergraduate students (targets) completed measures of their personality traits.1 This set of targets (37 male, 91 female) ranged in age from 18 to 22 (M = 18.5; SD = .73). On average, this sample had been using Twitter for 2 years (SD = 1.21 years) and had an average of 218 Twitter followers (SD = 167).
Materials and Procedure
Targets first completed scales assessing their impulsivity, self-esteem, and attachment style.2 They then entered their 10 most recent tweets, with instructions to remove the name of specific people (i.e., a friend they were describing) by replacing the person’s name with “[Name].” Targets were instructed to report all their tweets and not to skip any. They were instructed to only include tweets they composed them- selves, not retweets. No targets were excluded from data analysis.
Impulsivity. Targets completed the UPPS Impulsive Behavior Scale (Whiteside & Lynam, 2001), which is a 45-item scale with subscales assessing negative urgency (e.g., “ When I am upset I often act without thinking”), premeditation (e.g., “My thinking is usually careful and purposeful”), sensation seeking (e.g., “I generally seek new and exciting experi- ences and sensations”), and perseverance (“I generally like to see things through to the end”). In addition, targets com- pleted the positive urgency measure (Cyders et al., 2007), which contains 14 items (e.g., “I tend to lose control when I am in a great mood”). All impulsivity items included response options ranging from 1 (strongly disagree) to 5 (strongly disagree). On average, across domains, targets endorsed low-moderate levels of impulsivity (M = 2.75; SD = .39).
Self-esteem. Targets completed the Rosenberg Self-Esteem Scale (Rosenberg, 1965). This scale contains 10 items with response options ranging from 1 (strongly disagree) to 4 (strongly agree). Targets on average reported high levels of self-esteem (M = 3.04; SD = .51).
Attachment style. Targets completed the Experiences in Close Relationships–Revised (ECR-R; Fraley, Waller, & Brennan, 2000) scale. This scale contains 36 items, 18 of which tap a dimension of anxious attachment (e.g., “I’m afraid that I will lose my partner’s love”) and 18 of which tap a dimension of avoidant attachment (e.g., “I prefer not to show a partner how I feel deep down”). Targets were instructed: “We are interested in how you generally experience relationships, not just in what is happening in a current relationship.” Response options ranged from 1 (strongly disagree) to 7 (strongly agree). Tar- gets on average reported low levels of anxious (M = 3.43; SD = 1.10) and avoidant (M = 3.11; SD = 1.04) attachment.
Perceiver ratings. A set of 11 perceivers, all Twitter users who were undergraduates at the same university as the targets, were presented with the set of 10 tweets from each target, one at a time. They then rated the personality of the target using the same set of items the targets self-reported. The tweets remained on the computer screen while the perceivers made their judgments; perceivers completed all items for a single target before moving on to a new target. Raters were instructed not to judge participants if they thought they recognized the targets. No rater reported recognition of any target.
Analytical Approach
Accuracy and positivity were estimated with a multilevel model utilizing R’s lme4 package (Bates, Mächler, Bolker, & Walker, 2015; R Development Core Team, 2016), following the social accuracy modeling (SAM) procedures (Biesanz, 2010; Human & Biesanz, 2011). Specifically, in the within- perceiver part of the model (Level 1) perceivers’ ratings of each target on each trait item were regressed on (a) the tar- get’s self-report on that item after subtracting the normative mean for that item (assessing distinctive accuracy) and (b) the normative mean for that item (assessing normativity/ positivity). Each perceptual tendency was allowed to vary randomly across perceivers and targets as follows:
Yijk ij ij jk ij k ijk= + + +β β β ε0 1 2TSelf Mean . (1.1)
β β β
β β β 0 00 01 0 0 0
1 10 11 1 1
ij j i j ij
ij j i j
u u u
u u
= + + + +
= + + + +
Word
Word
( )
uu
u u u
ij
ij j i j ij
1
2 20 21 2 2 2
( )
( ).β β β= + + + +Word
(1.2)
Here Yijk is perceiver i’s rating of target j on item k, and TSelfjk is target j’s self-report on item k after partialling out
Orehek and Human 63
Meank, the mean self-report on item k across all targets. The regression coefficient β1ij thus represents the level of distinc- tive accuracy for perceiver i rating target j, or the extent to which perceiver i’s ratings of target j correspond to target j’s unique self-reported personality profile. The regression coef- ficient β2ij reflects normativity/positivity, or the extent per- ceiver i’s ratings of target j correspond to the mean self-reported personality profile. The fixed effects β00, β10, and β20 (Equation 1.2) represent the average intercept, dis- tinctive accuracy, and normativity slope, respectively, across both perceivers and targets.
Items were not reverse coded prior to analysis. We esti- mated models for the full set of items to examine accuracy overall across all traits, maximizing power, as well as for each characteristic separately (impulsivity, self-esteem, and anxious and avoidant attachment style) to examine specific- ity. Note that because different constructs were assessed on different rating scales, items were standardized within-scale prior to analyses.
Importantly, SAM also allows for the assessment of the degree of individual differences in target distinctive accuracy and normativity on average across perceivers (see Biesanz, 2010). As outlined in Equation 1.2, this approach separates true latent variability into perceiver, ui, and target, uj, main effects and the residual dyadic components, u(ij), which reflect unique effects plus measurement error. Specifically, the us represent the random (latent) effects in terms of devia- tions from the grand mean (fixed effects). For example, u1j, is target j’s unique distinctive accuracy slope averaged across the 11 perceivers, u1i is perceiver i’s unique distinctive accu- racy slope averaged across all 128 targets, and u1(ij) is the dyadic plus residual (error) component for perceiver i and target j. The variance of u1j across targets is estimated to pro- vide an indicator of the degree of reliable variance in distinc- tive accuracy due to targets and tested with a nested chi-square difference test comparing the models with versus without distinctive accuracy varying randomly as a function of targets. The same was approach was taken to assess vari- ability around targets’ unique normativity slopes, u2j, to examine individual differences in how normatively or posi- tively targets were viewed.
Equation 1.2 also demonstrates how we conducted our exploratory analyses of how word usage, such as the amount and type of words (e.g., affect, social, cognitive), might come to elicit accurate and positive perceptions. This was exam- ined by including word categories (Wordj) as moderators of the distinctive accuracy and normativity slopes at Level 2 (see Equation 1.2). A positive relationship between a mod- erator variable (e.g., word count) and the distinctive accu- racy (β11) or normativity (β21) slopes would indicate that using more words is associated with being viewed more in line with one’s distinctive traits or as more normative, respec- tively. We report effect size estimates, ds, for all interaction effects, calculated as the change in standard deviations in the dependent variable for a two standard deviation change in
the independent variable, to make these effect size estimates comparable with effect sizes for binary independent vari- ables (e.g., Cohen’s d; see Gelman, 2008).
Results
Primary Analyses
First, after viewing 10 tweets, did perceivers view targets normatively and therefore positively on average? Perceivers viewed targets with significant levels of normativity overall, b = .30, z = 7.57, p < .001, and on impulsivity, b = .17, z = 5.72, p < .001, self-esteem, b = .62, z = 10.69, p < .001, and anxious, b = .17, z = 4.85, p < .001, and avoidant, b = .28, z = 3.77 p < .001, attachment style (see Table 1). Because on average targets’ self-reported traits were relatively positive, including high levels of self-esteem and lower levels of impulsivity, anxious attachment, and avoidant attachment (see Methods), being viewed in line with the average target self-report on these traits suggests that targets were on aver- age also viewed relatively positively on these traits. These levels of normativity were generally lower than typically seen in perceptions of the Big Five after brief face-to-face interactions (e.g., b = .93; Human & Biesanz, 2011) and based on video perceptions (e.g., b = .50; Human et al., 2014). This may be due in part to the less dynamic impres- sion formation context but is also likely in part a function of the differences in scaling, as our items were standardized prior to analyses, whereas prior research used original item metrics. Indeed, when normativity was assessed within trait
Table 1. Levels of and Variability in Accuracy and Positivity.
Mean levels Target individual
differences
Model parameters b (SE) z SD
Overall Distinctive accuracy .05** (.013) 3.66 .13**
Normativity .30** (.039) 7.57 .12**
Impulsivity Distinctive accuracy .11** (.024) 4.80 .11**
Normativity .17* (.029) 5.72 .29**
Self-esteem Distinctive accuracy .03† (.016) 1.74 .12**
Normativity .62** (.058) 10.69 .16**
Anxious attachment Distinctive accuracy .02 (.018) 0.96 .16**
Normativity .17** (.036) 4.85 .07**
Avoidant attachment Distinctive accuracy .01 (.015) 0.45 .09**
Normativity .28** (.075) 3.77 .25**
Note. Mean levels are the fixed effects slopes (unstandardized). Target individual differences were assessed as the variance (estimated SDs) around targets’ unique distinctive accuracy, u1j, and normativity, u2j, main effects. †p < .10. *p < .05. **p < .001.
64 Personality and Social Psychology Bulletin 43(1)
with items in their original metric, levels were closer to that seen in video perceptions contexts but still lower than face to face impressions.
Were targets’ distinctive traits also viewed accurately? On average, targets were viewed with significant levels of dis- tinctive accuracy overall across traits, b = .05, z = 3.66, p < .001, and on impulsivity, b = .03, z = 2.06, p < .04, and with marginally significant levels of accuracy on self-esteem, b = .03, z = 1.74, p = .08 (see Table 1). However, targets were not viewed with significant levels of accuracy on either anxious, b = .02, z = 0.96, p = .34, or avoidant, b = .01, z = 0.45, p = .65, attachment style. These levels of distinctive accuracy are lower than what is typically observed in first impressions based on face-to-face interactions and video-based percep- tions of the Big Five (e.g., b = .08; Human & Biesanz, 2011; Human et al., 2014), both when standardized and when left in their original metrics. Thus, the overall patterning of traits across all characteristics, and within impulsivity and self- esteem, were somewhat detectable based on tweets, but attachment style was harder to discern.3
Importantly, however, there was also significant variability in target distinctive accuracy and normativity, indicating that there were reliable individual differences in how accurately and positively targets were perceived based on tweets (see Table 1). Figure 1 provides the distributions of targets’ distinc- tive accuracy and normativity slopes (i.e., how accurately and positively each target was seen on average across perceivers, respectively). Panel A demonstrates that there was substantial variability in how accurately targets were viewed, with some being viewed with high levels of accuracy (Max b = .39) and others being viewed with very inaccurately (Min b = −.28). Similarly, Panel B demonstrates the substantial variability in how positively targets were viewed, with some being viewed with high levels of normativity (Max b = .61), and others being viewed with much less normatively (Min b = −.24).
Exploratory Analyses
The aim of our exploratory analyses was to test whether word usage predicted how accurately and positively targets were viewed. According the realistic accuracy model (RAM; Funder, 1995), for a target to be perceived accurately, cues to the targets’ personality must be both available and relevant. Thus, the quantity and quality of information provided in tweets may increase the cue availability and relevance stages of RAM, respectively, enabling accurate perceptions. Furthermore, both the quantity and type of information pro- vided could influence how positively targets are viewed. To code word quantity and quality/type, we utilized the Linguistic Inquiry and Word Count program (LIWC2015; Pennebaker, Booth, Boyd, & Francis, 2015), which provides indicators of the amount of information (e.g., total word count, dictionary words) as well as a variety of information types, many of which could convey both diagnostic and evaluative informa- tion, including words that involve affect, other- versus self- focus, cognitive processes, drives, and personal concerns.
We examined whether each of the 93 LIWC categories were related to the accuracy and positivity in perceptions on average across all traits and for each trait separately. Given that this resulted in a very large number of analyses and there- fore high potential for spurious findings, we focus on present- ing cues that were a significant predictor of accuracy or positivity in the analyses that included all trait categories, as these may be more reliable effects (see Table 2 for key results and Table 3 for examples). Nevertheless, we emphasize that these analyses are highly exploratory, included to provide preliminary insights and suggest avenues for future research.
Predictors of normative perceptions. What aspects of tweets predicted being viewed in a more normative, positive man- ner? First, providing more information in the form of more
Figure 1. The degree of individual differences in how accurately (Panel A) and normatively (Panel B) targets were viewed are displayed with a kernel density plot of the distribution of the random effects (n = 128). The sigma-hat is the estimated standard deviation across unique target distinctive accuracy, u1j, and normativity, u2j, slopes.
Orehek and Human 65
total words overall was associated being perceived more posi- tively (see Table 2). Furthermore, more affectively positive
versus negative tweets were quite consistently associated with being viewed more positively (see Table 3 for examples).
Table 2. Associations Between Accuracy and Positivity and Linguistic Characteristics of Tweets.
Accuracy Normativity
Word category M (SD) d z d z
Information quantity Word count 111.10 (33.88) −0.11 −0.59 0.37* 1.99 Information quality Affect Tone 52.99 (36.54) 0.13 0.67 0.67*** 3.74 Positive emotion 4.35 (2.54) −0.04 −0.22 0.47* 2.52 Negative emotion 2.77 (2.00) −0.31 −1.55 −0.56** −2.83 Anger 0.98 (1.26) −0.12 −0.59 −0.29 −1.43 Sad 0.68 (0.89) −0.23 −1.23 −0.35† −1.94 Anxiety 0.38 (0.71) −0.42* −2.37 −0.09 −0.50 Swear 0.70 (1.20) −0.14 −0.60 −0.49* −2.15 Sexual 0.26 (0.60) −0.26 −1.30 −0.46* −2.34 Negations 1.90 (1.36) 0.18 0.93 −0.42* −2.32 Certainty 1.69 (1.39) −0.03 −0.15 0.38* 2.12 Clout 52.57 (26.42) 0.42* 2.24 0.46* 2.50 All punctuation 28.52 (18.43) 0.34† 1.84 0.47** 2.59 Quotations 1.04 (2.01) −0.16 −0.88 0.43* 2.35 Periods 5.95 (4.21) 0.43* 2.31 0.27 1.46 Self/other focus Personal pronouns 10.05 (3.37) −0.39* 1.98 −0.10 −0.53 I-talk 6.84 (3.13) −0.56** −3.06 −0.17 −0.92 Male referents 0.71 (1.01) 0.55* 2.47 0.36 1.62 Experiential Feeling 0.60 (0.90) 0.38* 2.07 0.06 0.35 Time 6.78 (3.00) 0.43* 2.34 0.17 0.91
Note. Means represent mean total (word count, tone, clout, punctuation) or percentage across all 10 tweets. †p ≤ .10. *p < .05. **p < .01. ***p < .001.
Table 3. Example Tweets for Word Categories Associated With High and Low Levels of Accuracy and Positivity.
Perception Linguistic category Example
Accuracy Low Anxiety Prob failed bio . . . time to stress out about planning for the next 3 hrs I-talk Every time I cyberstalk I end up with a pit in my belly because of how I feel toward
what I view, you’d think I’d stop, right? High Male referents It all started here just last year. So many memories with these guys! Feeling/time It was freezing! Showered in the waterfall! I am back!!! Positivity Low Negative emotion/swear Sometimes I really fucking hate going to class Sad I knew this day would come but it’s sort of sad reminiscing High Positive emotion So excited for Christmas! Family fun:))))) Certainty That was hands down the best match ever, great game everyone! Accuracy and positivity High Punctuation When your little sister asks what “marijuana” is . . .
Note. These example tweets were constructed to resemble actual participant tweets that fit into these categories. The actual tweets were changed to protect participant confidentiality. Bolded words fall within the corresponding linguistic category.
66 Personality and Social Psychology Bulletin 43(1)
This can be seen in the summary affective tone variable, which is an index of the ratio of positive to negative words, as well as the individual positive and negative emotion vari- ables. Examining the negative emotion subscales indicates that this association was strongest for sadness-related words, although the association was marginally significant.
Using more swear and sexuality-related words were also associated with being perceived less positively. This may be due in part to the overlap in content between both of these word categories and negative emotion words (see Table 3 for an example). Furthermore, using more negations in speech (e.g., couldn’t, don’t, won’t) was also associated with being viewed less positively, which may also be because such word use conveys greater negativity.
In contrast, tweets that involved more certainty words (e.g., always, every) were associated with being viewed more positively (see Table 3 for example). A similar pattern emerged for tweets higher in “clout,” a summary indicator of confidence or leadership indicated by using fewer first person singular pronouns or I-talk (e.g., I, me) and more first person plural pronouns (e.g., we, us; Kacewicz, Pennebaker, Davis, Jeon, & Graesser, 2014). Finally, the use of more punctuation and quotations in particular were also associated with being viewed more positively. Although LIWC is not able to code for humor, an informal qualitative look at tweets higher in punctuation and quotations suggests they may have elicited more positive perceptions because they were often used as comic devices (see Table 3 for example). Thus, words that convey more positivity, confi- dence, and humor, and less negativity, may promote being viewed more positively.
Predictors of accurate perceptions. What aspects of tweets were associated with being viewed more accurately? Greater information quantity was not significantly associated with the accuracy of perceptions and only one affect word was significantly associated with accuracy: The use of more anx- iety-related words was associated with being viewed less accurately (see Table 2). Using more personal pronouns (e.g., I’m, we) and, in particular, I-talk, was also associated with being viewed less accurately. This indicates that more self-focused tweets that expressed insecurity may have hin- dered accurate perceptions (see Table 3 for examples).
In contrast, using more male referents (e.g., boyfriend, brother), exhibiting greater clout (i.e., less I-talk), and using more punctuation, in particular periods, was associated with being viewed more accurately. Using more feeling (e.g., warm, cold, felt) and time (e.g., today, wait, back) words, potentially indicative of a more experiential, action-oriented focus, was also associated with being viewed more accu- rately. Taken together, these findings suggest that tweets that were more other-focused, confident, potentially humorous, and experiential in nature may have facilitated more accurate perceptions (see Table 3 for examples).
Discussion
Social media outlets such as Twitter represent widely used platforms for self-expression. Self-expression is a strongly held value in modern culture (Inglehart, 2008; Inglehart & Oyserman, 2004) that involves the dual motivations to be seen both positively and accurately (Goffman, 1959; McKenna & Bargh, 1999; Swann, 1983). In turn, being per- ceived positively and accurately in first impression settings predicts positive social consequences (Human et al., 2013) and is associated with better psychological functioning (Human & Biesanz, 2011; Human et al., 2014), suggesting that being perceived in such a manner on Twitter could ben- efit individuals. Thus, it is important to understand whether attempts at self-expression on social media platforms pro- duce both veridical and positive perceptions of targets.
The present study provides evidence that individuals can be seen accurately on Twitter, or at least in line with their distinctive self-views, extending previous research examin- ing other social media platforms, including Facebook (Back et al., 2010) and personal webpages (Vazire & Gosling, 2004). These findings also extend prior research, which has primarily focused on the Big Five personality traits, by examining other important dimensions of personality: impul- sivity, self-esteem, and attachment style. In particular, this study indicates that it is possible to detect impulsivity and self-esteem on the basis of tweets, but that attachment style is more difficult to perceive. However, even the accuracy of perceptions of impulsivity and self-esteem tended to be lower than accuracy for the Big Five based on face-to-face interactions (Human & Biesanz, 2011) and video clips (Human et al., 2014), suggesting that, although possible, conveying a distinctively accurate impression on Twitter can be a difficult task and perceivers should be cautious about inferring too much from tweets alone.
We also examined how positively targets’ personalities were perceived by examining the degree of normativity in per- ceptions (Wood & Furr, 2016), finding that impressions based on Twitter did tend to be quite positive. Interestingly, on aver- age, targets were viewed with similar levels of normativity to those being rated on the Big Five after exposure to brief video clips (Human et al., 2014), but perceptions were less norma- tive than those based on face-to-face impressions (Human & Biesanz, 2011). This suggests that dynamic, interactive social contexts may foster more positive impressions than more pas- sive ones (see also Schroeder & Epley, 2015). However, research is needed that directly compares impressions based on Twitter and other social media contexts with impressions based on more dynamic and interactive social contexts.
Importantly, this study also extends previous research by demonstrating that there were significant individual differ- ences in how accurately and positively targets were perceived on Twitter. Indeed, even though on average levels of accuracy were quite low, some individuals were viewed with much
Orehek and Human 67
higher levels of distinctive accuracy, well above the average levels seen in face-to-face interactions. Similarly, even though impressions tended to be relatively normative, some individu- als were viewed with quite low levels of normativity and therefore positivity. This variability suggests that there may be ways that individuals can actively attempt to elicit more (or less) accurate and positive perceptions on Twitter. To try to gain initial insight into why some individuals were per- ceived more accurately and positively than others, we explored what characteristics of tweets, such as their informa- tion quantity and quality, might contribute to this variability.
Interestingly, greater information quantity in tweets in the form of a greater number of words predicted being perceived more positively. This may be because more information in tweets may imply greater self-disclosure (Collins & Miller, 1994) and enhance familiarity (Reis, Maniaci, Caprariello, Eastwick, & Finkel, 2011), thereby increasing liking. Furthermore, just as a lower volume of tweets can indicate depression (De Choudhury et al., 2013), shorter tweets may similarly reflect and convey a less positive internal state, eliciting less positive perceptions.
Several other types of linguistic information were also associated with the positivity of perceptions. In particular, positive emotion words and indicators of greater confidence, such as more certainty words and lower self-focus, elicited more desirable perceptions. In contrast, expressing greater negative emotion, particularly sadness, using more nega- tions, and swearing more elicited less desirable impressions. Greater use of negative emotion words have also been linked to greater depression, both in tweets (De Choudhury et al., 2013) and other written communications such as essays (Rude et al., 2004); thus, it is not surprising that such linguis- tic patterns would elicit less positive perceptions from per- ceivers. More surprisingly, greater use of punctuation, especially quotations, was also associated with more positive perceptions. This may have been due to their use as comical devices, but this speculation requires further examination. Thus, to be seen positively on traits such as impulsivity, self- esteem, and attachment style, individuals may wish to aim for positive, confident, and humorous tweets.
In contrast, even though greater information quantity has been linked to more accurate impressions (e.g., Blackman & Funder, 1998), information quantity as indexed here was not significantly associated with greater accuracy. This may be because the upper limit of 140 characters may have limited how much greater information could actually increase accu- racy. Indeed, the benefits of greater information have been demonstrated when comparing larger differences in quantity, such as 5- to 10-min versus 25- to 30-min video clips (Blackman & Funder, 1998) or months versus years of acquaintanceship (Biesanz, West, & Millevoi, 2007). Thus, a difference of several words may not be sufficient enough to influence cue availability and in turn accuracy.
Although the amount of information in tweets did not sig- nificantly increase accuracy, other types of linguistic cues in
tweets were linked to accuracy. In particular, more other- focused tweets, indicated by greater use of male referents and less I-talk, were associated with being perceived with greater accuracy, as were tweets that may have involved more experiential information. It is possible that tweets that focus more on others and actions, relative to the self, convey more relevant, diagnostic information about an individual than tweets that are more self-focused. Alternatively, self- focused tweets may not necessarily decrease cue relevance themselves, but may instead be a marker of worse psycho- logical adjustment, which has been linked to the provision of less personality-relevant cues and lower accuracy in other social contexts (e.g., Human et al., 2014). Indeed, more self- focused language has been linked to negative psychological states and outcomes, including depression (De Choudhury et al., 2013; Rude et al., 2004) and suicidal tendencies (Stirman & Pennebaker, 2001). In line with this, the use of more anxiety-related words in tweets, which could be a mar- ket of maladjustment, was also associated with being viewed less accurately. Whether such linguistic patterns directly contribute to less accurate perceptions, or are an indicator of broader psychological functioning that influences accuracy, is a critical question for future research.
Of course, given the exploratory nature of these analyses, these effects must be interpreted cautiously and will ideally be replicated in future research with larger samples and in different social contexts. For example, although strangers may not respond very positively to more negative tweets, those within an individual’s online social networks may view such self-disclosures more sympathetically (e.g., Forest & Wood, 2012). In addition, the sample involved college stu- dents who may use Twitter differently from other samples and age groups. Indeed, more generally, our data are limited because we examined undergraduate Twitter users who were rated by other undergraduates. Future research should inves- tigate whether similar patterns are found for other types of users and raters, including older individuals, celebrities, and politicians. Furthermore, the sample was predominantly female, and although some gender differences emerged (see Footnote 3), more gender-balanced samples are needed to explore these effects.
More generally, the present results call for additional research and suggest avenues for future research on how per- sonality is perceived and expressed on social media. One major question concerns the potential generalizability versus specificity of the findings. Twitter is now populated by more than 288 million active users who tweet about 500 million messages per day, making it a particularly important domain to investigate self-expression. If the present results are lim- ited to Twitter and do not extend to other platforms, then such a finding would itself be interesting, and would call for research to investigate the critical differences across plat- forms. However, there is reason to believe the results will generalize to other social media platforms, as research inves- tigating self-expression on Facebook (Back et al., 2010) and
68 Personality and Social Psychology Bulletin 43(1)
personal webpages (Vazire & Gosling, 2004) similarly dem- onstrate that accuracy can be achieved. However, different traits were examined in those contexts and different approaches were used; thus more research is needed on (a) which personality traits are accurately perceived on social media, (b) which social media platforms facilitate more ver- sus less accuracy, and (c) whether specific platforms are more effective than others at communicating specific traits. It would also be interesting to examine whether these results generalize to other contexts in which brief text serves as the communication mode, such as text messages.
One potential limitation of these studies is the use of self- reports as the accuracy criterion. Although self-reports are a common and valid accuracy criterion (Funder & Colvin, 1997), using additional realistic accuracy criteria (Funder, 1995) to validate perceptions, such as close-other reports or behavioral indicators, would further strengthen conclusions about whether personality can be perceived accurately on Twitter. Indeed, it is possible that the apparently less accurate perceptions of those who use more I-talk are due to a lack of self-knowledge on the part of targets rather than inaccuracy on the part of the perceiver. This is especially possible given the evaluative nature of these traits, which the self may not always have accurate insight into (Vazire, 2010). On the con- trary, the less observable nature of many of the traits exam- ined in the present study does suggest that reliance on self-reports is valid, as the self tends to have unique insight into such traits, relative to outside observers (Vazire, 2010). Moreover, given that being viewed in line with one’s self- views, even when negative, is in itself an important and often positive experience, socially, psychologically, and physio- logically (e.g., Ayduk, Gyurak, Akinola, & Mendes, 2013; Swann, 1983), self-reports remain an important comparison with outside observers’ perceptions.
Conclusion
Overall, these results contribute to research demonstrating that social media platforms can be an effective outlet for accurate and positive self-expression of personality traits, extending such findings to the widely used platform of Twitter and to previously unexplored personality traits, such as impulsivity, self-esteem, and attachment style. Indeed, just 10 tweets from a person provided enough information for perceivers to accurately distill the person behind the mes- sage, particularly for impulsivity and self-esteem, while also eliciting generally normative and therefore positive percep- tions. There was also, however, significant variation in how accurately and positively individuals were perceived on Twitter, with some individuals being relatively open and enjoyable books, and others being more difficult and less enjoyable to read. Exploratory analyses suggest that word usage in tweets contributed to these individual differences, with affective content and degree of other- versus self-focus appearing to play particularly important roles. Ideally, future
work will continue to examine what characteristics are accu- rately and positively perceived on social media, through what cues and processes, and whether such perceptions are in turn consequential for both targets and perceivers.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, author- ship, and/or publication of this article.
Supplemental Material
The online supplemental material is available at http://pspb.sagepub .com/supplemental.
Notes
1. We recruited 150 targets. Twenty-two potential participants did not have Twitter accounts and therefore completed an unrelated study.
2. Targets also completed measures of their self-regulatory modes and perceived supportiveness of their Twitter followers. These measures are not considered because raters did not evaluate tar- gets on these dimensions.
3. There were significant gender differences in how targets were perceived, such that women were viewed less positively on self- esteem and anxious attachment, and more positively on avoid- ant attachment, as well as less accurately on both anxious and avoidant attachment. Given that the male target sample was small (N = 37; 29%), gender differences are not a focus of this study but suggest a future research direction.
References
Ambady, N., & Rosenthal, R. (1992). Thin slices of expressive behavior as predictors of interpersonal consequences: A meta- analysis. Psychological Bulletin, 111, 256-274.
Ayduk, O., Gyurak, A., Akinola, M., & Mendes, W. B. (2013). Consistency over flattery: Self-verification processes revealed in implicit and behavioral responses to feedback. Social Psychological and Personality Science, 4, 538-545. doi:10.1177/ 1948550612471827
Back, M. D., Stopfer, J. M., Vazire, S., Gaddis, S., Schmukle, S. C., Egloff, B., & Gosling, S. D. (2010). Facebook profiles reflect actual personality, not self-idealization. Psychological Science, 21, 372-374. doi:10.1177/0956797609360756
Bargh, J. A., McKenna, K. Y., & Fitzsimons, G. M. (2002). Can you see the real me? Activation and expression of the “true self” on the internet. Journal of Social Issues, 58, 33-48.
Bates, D., Mächler, M., Bolker, B., & Walker, S. (2015). Fitting linear mixed-effects models using lme4. Journal of Statistical Software, 67(1), 1-48. doi:10.18637/jss.v067.i01
Biesanz, J. C. (2010). The social accuracy model of interpersonal perception: Assessing individual differences in perceptive and expressive accuracy. Multivariate Behavioral Research, 45, 853-885.
Biesanz, J. C., & Human, L. J. (2010). The cost of forming more accurate impressionsaccuracy-motivated perceivers see the
Orehek and Human 69
personality of others more distinctively but less normatively than perceivers without an explicit goal. Psychological Science, 21(4), 589-594.
Biesanz, J. C., West, S. G., & Millevoi, A. (2007). What do you learn about someone over time? The relationship between length of acquaintance and consensus and self-other agreement in judgments of personality. Journal of Personality and Social Psychology, 92, 119-135. doi:10.1037/0022-3514.92.1.119
Blackman, M. C., & Funder, D. C. (1998). The effect of information on consensus and accuracy in personality judgment. Journal of Experimental Social Psychology, 34, 164-181. doi:10.1006/ jesp.1997.1347
Borkenau, P., Mosch, A., Tandler, N., & Wolf, A. (2016). Accuracy of judgments of personality based on textual information on major life domains. Journal of Personality, 84, 214-224.
Borkenau, P., & Zaltauskas, K. (2009). Effects of self-enhancement on agreement on personality profiles. European Journal of Personality, 23, 107-123.
Collins, N. L., & Miller, L. C. (1994). Self-disclosure and liking: A meta-analytic review. Psychological Bulletin, 116, 457-475. doi:10.1037//0033-2909.116.3.457
Cyders, M. A., Smith, G. T., Spillane, N. S., Fischer, S., Annus, A. M., & Peterson, C. (2007). Integration of impulsivity and positive mood to predict risky behavior: Development and validation of a measure of positive urgency. Psychological Assessment, 19, 107-118.
De Choudhury, M., Gamon, M., Counts, S., & Horvitz, E. (2013, July 8-11). Predicting depression via social media. Proceedings of the ICWSM 2013: International AAAI Conference on Weblogs and Social Media, Cambridge, MA.
Duggan, M., Ellison, N. B., Lampe, C., Lenhart, A., & Madden, M. (2014). Social media update 2014. Washington, DC: Pew Internet & American Life Project.
Fletcher, G. J. O., & Kerr, P. S. G. (2010). Through the eyes of love: Reality and illusion in intimate relationships. Psychological Bulletin, 136, 627-658.
Forest, A. L., & Wood, J. V. (2012). When social networking is not working: Individuals with low self-esteem recognize but do not reap the benefits of self-disclosure on Facebook. Psychological Science, 23, 295-302. doi:10.1177/0956797611429709
Fraley, R. C., Waller, N. G., & Brennan, K. A. (2000). An item response theory analysis of self-report measures of adult attach- ment. Journal of Personality and Social Psychology, 78, 350-365.
Funder, D. C. (1995). On the accuracy of personality judgment: A realistic approach. Psychological Review, 102, 652-670. doi:10 .1037/0033295X.102.4.652
Funder, D. C., & Colvin, C. R. (1997). Congruence of others’ and self-judgments of personality. In R. Hogan, J. Johnson, & S. Briggs (Eds.), Handbook of personality psychology (pp. 617- 647). San Diego, CA: Academic Press. doi:10.1016/B978- 012134645-4/50025-1
Gagné, F. M., & Lydon, J. E. (2004). Bias and accuracy in close relationships: An integrative review. Personality and Social Psychology Review, 8, 322-338.
Gelman, A. (2008). Scaling regression inputs by dividing by two standard deviations. Statistics in Medicine, 27, 2865-2873.
Goffman, E. (1959). The presentation of self in everyday life. Garden City, NY: Doubleday.
Golbeck, J., Robles, C., Edmondson, M., & Turner, K. (2011, October). Predicting personality from twitter. In 2011 IEEE
International Conference on Privacy, Security, Risk and Trust (PASSAT) and 2011 IEEE Third International Conference on Social Computing (pp. 149-156). Los Alamitos, CA: IEEE Computer Society.
Human, L. J., & Biesanz, J. C. (2011). Target adjustment and self- other agreement: Utilizing trait observability to disentangle judgeability and self-knowledge. Journal of Personality and Social Psychology, 101, 202-216.
Human, L. J., & Biesanz, J. C. (2012). Accuracy and assumed similarity in first impressions of personality: Differing asso- ciations at different levels of analysis. Journal of Research in Personality, 46, 106-110. doi:10.1016/j.jrp.2011.10.002
Human, L. J., & Biesanz, J. C. (2013). Targeting the good target: An integrative review of the characteristics and consequences of being accurately perceived. Personality and Social Psychology Review, 17, 248-272.
Human, L. J., Biesanz, J. C., Finseth, S., Pierce, B., & Le, M. (2014). To thine own self be true: Psychological adjustment promotes judgeability via personality-behavior congruence. Journal of Personality and Social Psychology, 106, 286-303.
Human, L. J., Biesanz, J. C., Parisotto, K. L., & Dunn, E. W. (2012). Your best self helps reveal your true self: Trying to make a good impression leads to more accurate personality impres- sions. Social Psychological and Personality Science, 3, 23-30.
Human, L. J., Sandstrom, G. M., Biesanz, J. C., & Dunn, E. W. (2013). Accurate first impressions leave a lasting impression: The long-term effects of accuracy on relationship development. Social Psychological and Personality Science, 4, 395-402.
Inglehart, R. F. (2008). Changing values among western publics from 1970 to 2006. West European Politics, 31, 130-146.
Inglehart, R., & Oyserman, D. (2004). Individualism, autonomy, self- expression: The human development syndrome. International Studies in Sociology and Social Anthropology, 93, 74-96.
Kacewicz, W., Pennebaker, J. W., Davis, M., Jeon, M., & Graesser, A. C. (2014). Pronoun use reflects standings in social hierarchies. Journal of Language and Social Psychology, 33, 125-143. doi:10.1177/0261927X1350265
Kenny, D. A., & Winquist, L. (2001). The measurement of inter- personal sensitivity: Consideration of design, components, and unit of analysis. In J. Hall, & F. Bernieri (Eds.), Interpersonal sensitivity: Theory and measurement (pp. 265-302). Mahwah, NJ: Lawrence Erlbaum.
Küfner, A. C. P., Back, M. D., Nestler, S., & Egloff, B. (2010). Tell me a story and I will tell you who you are! Lens model analy- ses of personality and creative writing. Journal of Research in Personality, 44, 427-435. doi:10.1016/j.jrp.2010.05.003
Marwick, A. E., & Boyd, D. (2011). I tweet honestly, I tweet pas- sionately: Twitter users, context collapse, and the imagined audience. New Media & Society, 13, 114-133.
McKenna, K. Y., & Bargh, J. A. (1999). Causes and consequences of social interaction on the internet: A conceptual framework. Media Psychology, 1, 249-269.
Mehl, M. R., Gosling, S. D., & Pennebaker, J. W. (2006). Personality in its natural habitat: Manifestations and implicit folk theories of personality in daily life. Journal of Personality and Social Psychology, 90, 862-877.
Murray, S. L., Holmes, J. G., & Griffin, D. W. (1996). The ben- efits of positive illusions: Idealization and the construction of satisfaction in close relationships. Journal of Personality and Social Psychology, 70, 79-98.
70 Personality and Social Psychology Bulletin 43(1)
Neff, L. A., & Karney, B. R. (2005). To know you is to love you: The implications of global adoration and specific accuracy for marital relationships. Journal of Personality and Social Psychology, 88, 480-497.
Pennebaker, J. W., Booth, R. J., Boyd, R. L., & Francis, M. E. (2015). Linguistic inquiry and word count: LIWC2015. Austin, TX: Pennebaker Conglomerates. Available from www.LIWC.net
Quercia, D., Kosinski, M., Stillwell, D., & Crowcroft, J. (2011, October). Our Twitter profiles, our selves: Predicting person- ality with Twitter. In 2011 IEEE International Conference on Privacy, Security, Risk and Trust (PASSAT) and 2011 IEEE Third International Conference on Social Computing (pp. 180-185). Los Alamitos, CA: IEEE Computer Society.
R Development Core Team (2016). R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing. Retrieved from https://www.R-project .org
Reis, H. T., Maniaci, M. R., Caprariello, P. A., Eastwick, P. W., & Finkel, E. J. (2011). Familiarity does indeed promote attrac- tion in live interaction. Journal of Personality and Social Psychology, 101, 557-570. doi:10.1037/a0022885
Rosenberg, M. (1965). Society and the adolescent self-image. Princeton, NJ: Princeton University Press.
Rude, S. S., Gortner, E.-M., & Pennebaker, J. W. (2004). Language use of depressed and depression-vulnerable college students. Cognition and Emotion, 18, 1121-1133.
Schroeder, J., & Epley, N. (2015). The sound of intellect: Speech reveals a thoughtful mind, increasing a job candidate’s appeal. Psychological Science, 26, 877-891.
Self-Expression. (n.d.). In Merriam-Webster’s online dictionary. Retrieved from http://www.merriam-webster.com/dictionary/ self-expression
Selim, H. A., Long, K. M., & Vignoles, V. L. (2014). Exploring iden- tity motives in Twitter usage in Saudi Arabia and the UK. Annual Review of Cybertherapy and Telemedicine, 199, 128-132.
Stirman, S. W., & Pennebaker, J. W. (2001). Word use in the poetry of suicidal and nonsuicidal poets. Psychosomatic Medicine, 63, 517-522.
Swann, W. B., Jr. (1983). Self-verification: Bringing social real- ity into harmony with the self. In J. Suls & A. G. Greenwald (Eds.), Social Psychological Perspectives on the Self (Vol. 2, pp. 33-66). Hillsdale, NJ: Lawrence Erlbaum.
Tufekci, Z. (2008). Can you see me now? Audience and disclosure regulation in online social network sites. Bulletin of Science, Technology & Society, 28, 20-36.
VanLear, C. A., Sheehan, M., Withers, L. A., & Walker, R. A. (2005). AA online: The enactment of supportive computer mediated communication. Western Journal of Communication, 69, 5-26.
Vazire, S. (2010). Who knows what about a person? The self-other knowledge asymmetry (SOKA) model. Journal of Personality and Social Psychology, 98, 281-300. doi:10.1037/a0017908
Vazire, S., & Gosling, S. D. (2004). e-Perceptions: Personality impressions based on personal websites. Journal of Personality and Social Psychology, 87, 123-132. doi:10.1037/0022- 3514.87.1.123.
Whiteside, S. P., & Lynam, D. R. (2001). The five factor model and impulsivity: Using a structural model of personality to under- stand impulsivity. Personality and Individual Differences, 30, 669-689.
Wood, D., & Furr, R. M. (2016). The correlates of similarity estimates are often misleadingly positive: The nature and scope of the problem, and some solutions. Personality and Social Psychology Review, 20, 79-99. doi:10.1177/1088868315581119