- Email: [email protected]
#Proana: Pro-Eating Disorder Socialization on Twitter
Journal of Adolescent Health 58 (2016) 659e664
www.jahonline.org Original article
#Proana: Pro-Eating Disorder Socialization on Twitter Alina Arseniev-Koehler a, b, *, Hedwig Lee, Ph.D. a, Tyler McCormick, Ph.D. a, c, and Megan A. Moreno, M.D., M.S.Ed, M.P.H. b, d a
Department of Sociology, University of Washington, Seattle, Washington Center for Child Health, Behavior and Development, Seattle Children’s Research Institute, Seattle, Washington Department of Statistics, University of Washington, Seattle, Washington d Department of Pediatrics, University of Washington, Seattle, Washington b c
Article history: Received August 5, 2015; Accepted February 26, 2016 Keywords: Eating disorders; Pro-eating disorder; Social media; Social networks; Social support; Twitter; Adolescent health
A B S T R A C T
Purpose: Proeeating disorder (ED) online movements support engagement with ED lifestyles and are associated with negative health consequences for adolescents with EDs. Twitter is a popular social media site among adolescents that provides a unique setting for Pro-ED content to be publicly exchanged. The purpose of this study was to investigate Pro-ED Twitter profiles’ references to EDs and how their social connections (followers) reference EDs. Methods: A purposeful sample of 45 Pro-ED profiles was selected from Twitter. Profile information, all tweets, and a random sample of 100 of their followers’ profile information were collected for content analysis using the Twitter Application Programming Interface. A codebook based on ED screening guidelines was applied to evaluate ED references. For each Pro-ED profile, proportion of tweets with ED references and proportion of followers with ED references in their own profile were evaluated. Results: In total, our 45 Pro-ED profiles generated 4,245 tweets for analysis. A median of 36.4% of profiles’ tweets contained ED references. Pro-ED profiles had a median of 173 followers, and a median of 44.5% of followers had ED references. Pro-ED profiles with more tweets with ED references also tended to have more followers with ED references (b ¼ .37, p < .01). Conclusions: Findings suggest that profiles which self-identify as Pro-ED express disordered eating patterns through tweets and have an audience of followers, many of whom also reference ED in their own profiles. ED socialization on Twitter might provide social support, but in the ProED context this activity might also reinforce an ED identity. Ó 2016 The Society for Adolescent Health and Medicine. All rights reserved.
Eating disorders (EDs) are complex psychophysiological illnesses that can lead to serious health consequences including the highest mortality rate of any psychiatric disorder [1]. Approximately, 2.7% of US 13e17 year olds experience from an ED [2] and an additional 3.3% of adolescents struggle with subthreshold ED
Conflicts of Interest: The authors have no conflicts of interest or financial disclosures to report. * Address correspondence to: Alina Arseniev-Koehler, 16526 Densmore Ave. N, Shoreline, WA 98133. E-mail address: [email protected] (A. Arseniev-Koehler).
IMPLICATIONS AND CONTRIBUTION
This study investigated Twitter profiles that selfidentified as supportive of eating disorder (ED) lifestyles (Pro-ED). Analysis revealed that Pro-ED profiles often displayed ED references and accumulated networks of Twitter followers who also displayed ED references. These online connections could reinforce an ED identity for those with Pro-ED profiles and their followers.
symptoms such as purging or laxative use [3]. Despite the prevalence and seriousness of these disorders, it can be difficult to reach those who would benefit from treatment. Individuals often try to conceal ED symptoms [4], and many never discuss their struggles with weight and body image with professionals [3]. Given the often secretive nature of EDs, one venue in which those with ED may seek social support or resources is online. ProED online communities emerged in the early 2000s to support ED lifestyles [5]. Previous studies have illustrated that Pro-ED cyberspaces may exacerbate ED risk factors [5e7] such as poor body image [8,9], reinforcement of an individuals’ ED identity
1054-139X/Ó 2016 The Society for Adolescent Health and Medicine. All rights reserved. http://dx.doi.org/10.1016/j.jadohealth.2016.02.012
660
A. Arseniev-Koehler et al. / Journal of Adolescent Health 58 (2016) 659e664
[10,11], or learning and subsequently using unhealthy weight loss methods [5e7,12]. Among youth with EDs, engagement with online Pro-ED communities is common [6] and has recently been suggested as a screening factor for EDs [13]. Such findings highlight the importance of understanding teens’ engagement with Pro-ED content on social media. Twitter, which is used by 33% of US teens, is a unique online social media platform through which to study Pro-ED [14]. As a hub of online public social activity and information exchange, Twitter users create profiles called handles from which they can microblog, or “tweet” (microexpressions of 140 characters or less), as well as follow the feeds of other handles’ tweets, creating an interconnected social network. Users may contribute to a larger tweet conversation on Twitter by adding a hashtag to a key term within the tweet, which connects communities who also use that hashtag around shared topics of interest. With this social structure, Twitter can enable those who ascribe to Pro-ED to express their ED identity and gain a social and information network for this ED identity. For example, the use of the hashtag #proana (proanorexia) allows Twitter users to post and view content that is supportive of ED lifestyles. Although steps to counteract this content have been taken by many sites [15e19] such as Instagram [15,16] and Facebook [18], Twitter has made no attempts to limit Pro-ED content [20]. Little is known regarding the impact of Twitter as a medium to exchange information or create social connections in the Pro-ED community. This information is necessary to understand how providers can best counsel adolescent patients about these online communities, and how researchers can consider interventions to address them. The purpose of this study was to describe the content and social networks of a sample of Pro-ED Twitter profiles. To describe the social networks of Pro-ED profiles, we also examined their followers to determine the proportion which display ED references in their own profiles (hereafter referred to as ED followers). We also investigated if Pro-ED profiles with more ED content had more ED followers. We hypothesized that Pro-ED profiles with more ED expression in their own profiles would have higher proportion of ED followers compared to profiles with less ED expression. Methods Setting This descriptive study used public Twitter data from Pro-ED profiles and their followers, collected between February and April 2014. This study was determined to qualify as not human subjects’ research by the University of Washington’s institutional review board because public Twitter handles are avatars and are not identifiable, living individuals according to local and national regulations. Pro-ED profile and follower sampling strategy Our first goal was to identify a sample of Pro-ED profiles for evaluation. We used purposeful sampling to identify profiles with displayed content supporting a positive attitude toward EDs. Purposeful sampling is a strategy used for investigation into social processes of particular social groups by selecting representative members of that group [21].We applied similar
sampling and methodological framework previously used to explore social circles of Twitter profiles which displayed drug abuse [22]. To identify Pro-ED profiles, we searched for profiles using the hashtag #proana, as “proana” is an established term to describe the Pro-ED movement [5,8e12]. Using the Twitter analytic tool TwitteR [23], data were collected on days with varying intervals between them between the dates of February 18, 2014 and April 3, 2014. We collected samples of 25 tweets and retweets (shares of another’s tweet) which included #proana and were created within the week before the collection day. Sample tweets, along with handle, were collected on different days to minimize sampling of profiles engaged in direct conversations and to obtain a sample which better represented the larger Pro-ED activity on Twitter over time. Inclusion criteria for our sample of Pro-ED profiles included that the profile had least one display of a positive Pro-ED attitude in their profile information or in their most recent 15 tweets. A positive Pro-ED attitude was defined (1) selfidentifying as Pro-ED and/or as having an ED and antirecovery; or (2) expressing a desire for emaciation; or (3) ascribing to a ProED event. Examples of Pro-ED events at the time of data collection included “ABC (Ana Boot Camp) Diet” and a month-long Twitter-based Pro-ED competition to restrict calories, “#CalorieApril,” which were documented in pilot observation of Pro-ED on Twitter. For example, a profile with the tweet “I’m excited for my first day of #calorieapril! Let’s do it! #proana” would have been considered a Pro-ED profile. A total of 211 handles were viewed to achieve our predetermined sample size of 45 Pro-ED profiles. Among the 166 profiles excluded, 63.9% did not meet these criteria for explicit Pro-ED attitude, 28.9% were nonEnglish, 4.8% were inactive or had privacy set to “protected,” and 6.0% indicated a male identity. Given evidence that Pro-ED is a gendered collective identity [24] and that users tend to be female [7,25e27], we excluded explicitly male identities to minimize noise in findings due to gender. Excluded profiles may have met more than one exclusion criteria. Our second goal was to identify a sample of these Pro-ED profiles’ followers for evaluation. For each Pro-ED profile, we took a simple random sample of 100 followers for further examination. If a Pro-ED profile had fewer than 100 followers, all were collected. Codebook development and variables We developed a codebook to examine Pro-ED profiles’ tweets and profile information (profile name, reported name, location, and description) for ED references and to examine their followers’ profile information for ED references. ED variables. The codebook was designed to be applied to 3 types of ED variables: ED references in Pro-ED profiles’ profile information, ED references in Pro-ED profiles’ tweets, and ED references in followers’ profile information. Our strategy to code for ED references was designed following previous research on codebook development [22,28]. First, an extensive list of ED keywords was generated using conceptual categories and keywords from the Eating Attitudes Test-26 [29]. Similar to previous coding strategies for depression on social media [28], coding was grounded using the Eating Attitudes Test26 because this is a validated clinical questionnaire to screen for ED symptoms [29]. Next, we expanded this list by incorporating terms identified in a literature review on EDs and Pro-ED, pilot
A. Arseniev-Koehler et al. / Journal of Adolescent Health 58 (2016) 659e664
661
Table 1 ED references in Pro-ED profiles’ tweets Domain
Sample keywords
Explanations
Explicit ED terms
Anorexic, ana, bulimic, proana, Pro-ED, promia, ednos, eating disorder, wannarexic
Body and body image
Overweight, obese, fatty, skinnier, skeleton, emaciated, hipbones, backbone, bones, collarbone, thighgap, bikinibridge, hipbones, thighs, hips, thinspo, bonespo, perfection
Body weight
Weight, scale, GW, CW, LW, UGW, BMI, pound, calorie
Wannarexic is a term used in Pro-ED culture to describe individuals engaged in Pro-ED online but do not engage in ED behavior offline [11]. Thighgap refers to a gap between thighs. Bikinibridge refers to space between pelvic bones. Thinspo (Thinspiration) is a term which refers to any media or text idolizing thinness [5]. Bonespo (Bonespiration) was an observed term which referred to media or text idolizing emaciation that made bones visible. Perfection is a common theme and term observed and previously documented in Pro-ED [5,12]. GW/CW/LW/UGW were commonly observed lingo for goal weight/current weight/lowest weight/ultimate goal weight, respectively. BMI is an abbreviation for body mass index.
Food and meals Eat or ate Caloric restriction
Food, breakfast, dinner, meals Eating, eat, ate, appetite Starve, hunger, diet, skip, fasted, CalorieApril, ProjectThin, RG, ABCdiet
Binge Compensatory behavior Exercise
Binge, bloated Laxies, laxatives, vomited, throwup, puke, purge Workout, abs, jog, elliptical, exercise, miles, gym, treadmill
CalorieApril was an observed Twitter-based competition for caloric intake in the month of April. ABCdiet and RG refer to “Ana Boot Camp” and Russian Gymnast Diet, which were commonly observed Pro-ED diets on Twitter. Laxies was a commonly observed abbreviation for laxatives
Pro-ED profiles’ tweets were categorized as ED tweets if they included at least one keyword in this codebook. ED ¼ eating disorder.
observation of Pro-ED on Twitter and other cyberspaces, as well as consultation with a clinical pediatric dietitian with expertise in adolescent EDs. Our final codebook for ED references in Pro-ED profiles’ tweets and profile information included a broad list of 143 keywords; given that coding was in context of Pro-ED we wanted to capture all references that may be related to disordered eating behaviors. Our keywords were grouped into nine domains, such as references to bingeing, body and body image, and compensatory behavior including laxative use and purging. Example keywords in these domains included “binged,” “thighgap,” and “puked.” We also included the domain explicit ED terms, which comprised references to types of EDs, such as “bulimia.” We further included domains to capture terms related to eating and exercise, given that these behaviors are clinically important in the context of disordered eating and consistent with our context of evaluating these behaviors on a Pro-ED profile. For a full list of domains and sample keywords, see Table 1. Pro-ED profiles’ tweets were categorized as ED tweets if they included at least one keyword in this codebook. Followers of the Pro-ED profiles were coded using a restricted codebook compared to that used for Pro-ED profiles and their tweets because keyword coding was no longer in context of a profile purposefully selected for a Pro-ED attitude. Thus, keywords in the domains of eating and exercise and in other domains, such as “perfection” (common lingo in Pro-ED culture [5,12]), were not coded as ED references among followers given that these references were less specific to EDs. Twenty-eight keywords were excluded in this restricted codebook. Followers were categorized as ED followers if their profile information included least one keyword or synonym to a keyword. Importantly, the inclusion criteria for Pro-ED profiles were intended to ensure these profiles displayed current, explicit, positive attitudes toward EDs, which provided assumption that content with ED keywords reflect an underlying unhealthy
perspective. Furthermore, followers were defined as those who had voluntarily subscribed to Pro-ED profiles’ tweets and our restrictive coding approach to identify ED followers meant that we identified profiles that explicitly displayed ED references in their public profile information. Therefore, a liberal keyword coding scheme for ED thoughts and behaviors was warranted in this context. Keyword coding also enabled reliable and transparent analysis. Descriptive variables. Descriptive variables for Pro-ED profiles included: number of followers, tweets, retweets, number of favorites (endorsements) of others’ tweets, days since account creation (defined as days between account creation date and analysis date), as well as profile information. We evaluated profile information for three additional variables, including: displayed age, references to the account being anonymous or secret, and displayed body mass index. Pilot testing. This codebook was pilot tested on five Pro-ED profiles and their followers and revised based on consensus before finalization. One revision following pilot testing included manual coding for followers, so that coding would include unique spellings and synonyms for keywords. These were commonly noted in profile information while pilot testing. Data collection and coding procedure The statistical program R [30] with the package TwitteR [23] was used to collect all data from Twitter. TwitteR allows users to collect public data from Twitter using the Twitter Applied Programming Interface. Pro-ED profiles’ profile information was manually coded to evaluate for descriptive variables and ED references. Tweets were coded automatically with R using text recognition for ED keywords in the codebook. Pro-ED profiles’ followers were coded manually.
662
A. Arseniev-Koehler et al. / Journal of Adolescent Health 58 (2016) 659e664
Analysis Descriptive statistics were used to describe the proportion of ED tweets on Pro-ED profiles and proportion of ED followers. Multivariate linear regression [31] was used to predict Pro-ED profiles’ proportion of ED followers from Pro-ED profiles’ proportion of ED tweets. We added the control variable days since account creation given its importance to predicting longitudinal follower accumulation [32]. As the use of proportion data is traditionally inappropriate for linear regression, after viewing the data we performed diagnostics to ensure that the effect of this violation was minimal in this data set [31,33]. We also tested an alternate measure of correlation between Pro-ED profiles’ proportion of ED tweets and their proportion of ED followers, that is, Spearman Rank Correlation [34]. Results Pro-ED profile sample characteristics Forty-five Pro-ED profiles were selected from Twitter, 86.7% (N ¼ 39) of Pro-ED profiles displayed ED references in their profile name, screen name, or description. Furthermore, 64.4% (N ¼ 29) displayed an ED reference in two or three of these locations, and 33.3% (N ¼ 15) displayed an ED reference in all three of these locations. No Pro-ED profiles displayed references in their location. Demographic data were not universally available, 13 profiles (28.9%) displayed an age; the mean was 17.7 (standard deviation [SD] ¼ 2.75) years. Table 2 includes further displayed characteristics. Finally, these accounts tended to be recently created; profiles were created a median of 4.7 months before the analysis date (May 1, 2014). Pro-ED profiles’ tweets Pro-ED profiles had a median of 49 tweets and together these profiles generated a total of 4,245 tweets for analysis. A median of 36.4% and mean of 35.7% of Pro-ED profiles’ tweets had at least one ED reference (SD ¼ 19.5%). ED tweets often included multiple ED references. Among the nine domains in our codebook for ED references, nearly all (91%) of Pro-ED profiles referenced the domain body and body image at least once in their tweets. Some (42.2%) referenced the domain compensatory behavior. The most frequently referenced domain was explicit ED terms, such as “anorexia”; on average 13.2% (SD ¼ 16.4%) of Pro-ED profiles’ tweets referenced this domain. For the prevalence of domains for ED references see Table 3. Sample observed tweets included: “Do not reward yourself with food, you are not a dog #proana,” and “crying after eating 300 cals then binging on another 5000 because you ‘blew it’ #EDlife.” For further results on Pro-ED profiles’ tweet quantity and frequency see Table 4. Pro-ED profiles’ followers Pro-ED profiles had a median of 173 followers (range 5e7,935), and the sample of 45 profiles together yielded a total audience of 25,294 followers. In total 3,719 followers were sampled for evaluation. A median of 44.5% and a mean of 39.3% of sampled followers had ED references in their profile information (SD ¼ 19.4%, range 0%e79.2%). Sample observed references in profile names
Table 2 Pro-ED profiles’ displayed characteristics N (%), Pro-ED Median Mean Standard Range deviation profiles with display Displayed age (years) 13 (28.9%) Displayed body mass 5 (11.1%) index (kg/m2) 4 (8.9%) Anonymitya
17 18.5
17.7 20
2.75 3.1
15e25 17.8e26
d
d
d
d
ED ¼ eating disorder. a Anonymity included proportion of profiles which self-describe the account as “anonymous” or “secret.”
included: “AnasHostage,” “StarvingMeSmall,” and “Anasfatsister22.” Sample observed descriptions with ED references included: “Holding on to hope. Ana/mia/cat/deb. thanks for listening to my story. trigger warning” and “Just a fat girl trying to get skinny. Height: 50 6. Current weight: whale. Goal weight: lighter than air. #thinspireme.” A final sample observed description with references was “just an average girl, dying for perfection. Follow me w i dont bite (excuse the pun).” Sample references have been modified slightly to ensure confidentiality of profiles. Association between ED references in Pro-ED profiles’ tweets and followers This study further investigated Pro-ED profiles’ social circles by testing whether Pro-ED profiles with a higher proportion of ED tweets tended to have a higher proportion of ED followers. In the adjusted model, the proportion of ED tweets was positively associated with the proportion of ED followers (b ¼ .37, p < .01), controlling for the days since the profile was created. Controlling for proportion of ED tweets, more recently created profiles tended to have a higher proportion of ED followers (b ¼ .0002, p < .01). The model explained 29.4% of the variation in proportion of ED followers (R2 ¼ 29.4, p < .001). An added variable plot [31,33] in Figure 1 shows the adjusted linear relationship between proportion of ED tweets and proportion of ED followers. Pro-ED profiles’ proportion of ED tweets were positively, moderately correlated with their proportion of ED followers (S ¼ 8,193.01, Rho ¼ .46, p ¼ .001). Table 3 Prevalence of ED reference domains among 45 Pro-ED profiles and their tweets ED reference domain
N (%), Pro-ED profiles referencing domain Among 45 Pro-ED profiles
Explicit ED termsa Body and body image Body weight Food and meals Eat and ate Caloric restriction Bingeing Compensatory behavior Exercise
33 41 32 29 39 36 17 19 21
(73.3%) (91.1%) (71.1%) (64.4%) (86.7%) (80%) (37.8%) (42.2%) (46.7%)
Mean (SD) proportion of tweets referencing domain
N (%), tweets referencing domain
13.2 11.6 6.2 3.9 7.1 5.7 1.3 1.6 2.2
377 362 221 179 269 164 43 54 83
(16.4%) (13.1%) (8.9%) (4.9%) (6.5%) (6.3%) (3.1%) (2.8%) (4.6%)
Among 4,245 tweets (8.8%) (8.5%) (5.2%) (4.2%) (6.3%) (3.9%) (1.3%) (1.6%) (2.2%)
ED ¼ eating disorder; SD ¼ standard deviation. a Explicit ED terms include references to types of EDs, such as “bulimia” and “anorexia.” For a full explanation of domains see Table 1.
A. Arseniev-Koehler et al. / Journal of Adolescent Health 58 (2016) 659e664 Table 4 Quantity and frequency of Pro-ED profiles’ tweets
Total tweets Percent ED tweets Tweet frequency ED tweet frequency Total tweets and retweets Tweet and retweet frequency Total “favorites” of other tweets
Median
Mean
Standard deviation
Range
49 36.4% 2.2/week .74/week 405
94.3 35.7% 4.6/week 1.5/week 1,969.3
169.4 19.5% 5.9/week 2.2/week 3,748.5
6e1,105 0%e77.6% .2e32/week 0e13/week 8e19,381
17.9/week
57.7/week
81.8/week
.35e366.8/week
72
471.5
1,079.5
0e4,789
ED ¼ eating disorder; SD ¼ standard deviation.
Discussion This study applied keywords from a validated clinical screen to Twitter content, examining a purposeful sample of Pro-ED profiles and their followers. Findings suggest that Pro-ED profiles are indeed created and expressed on the social media site of Twitter, and these online identities tend to attract an audience of followers who are like-minded individuals. Our first finding is that more than a third (36.4%) of Pro-ED profiles’ tweets were ED related. The presence of Twitter profiles dedicated to Pro-ED identities may not be surprising. EDs are known to overwhelm individuals’ identity but offline expression of this identity may be suppressed due to factors such as stigma against EDs [4,35] and ambivalence toward recovery [35,36] which leads individuals to conceal symptoms from peers and caretakers. In contrast, explicit references to EDs were common among our Twitter profiles; 73.3% of Pro-ED profiles referenced a specific ED in their tweets and 42% of our profiles referenced compensatory behavior, including purging or laxatives. The prevalence (86.7%) of ED and of anonymity references in Pro-ED profiles’ profile information further suggests that such accounts are dedicated to the ED identity.
Figure 1. This added variable plot demonstrates the positive correlation between percent ED followers and percent ED tweets among Pro-ED Twitter profiles, after controlling both variables for any variation due to days since ProED profile account creation (R2 ¼ .37, p < .01).
663
The proportion (44.5%) of ED followers of Pro-ED profiles provides evidence that Pro-ED profiles often form social connections on Twitter with other profiles which display an interest in weight and body image. Twitter may provide Pro-ED profiles and their followers a sense of community and an audience, as is provided for general Twitter users [37]. This might provide peer reinforcement for a Pro-ED identity for both Pro-ED profiles and their followers as well as perceived integration into a Pro-ED Twitter network. Similar to other Pro-ED sites, ED socialization on Twitter might be a source of social support for Pro-ED Twitter profiles and their followers [5,7,25]. This support may be particularly important for individuals with ED, who are often stigmatized and lack in-person support [4,38]. However, studies of other Pro-ED sites have also demonstrated that socialization around EDs in the context of an underlying Pro-ED attitude can also yield outcomes such as poor body image [8,9] and learning unhealthy weight loss strategies [6], and such risks likely translate to Twitter. Our third main finding is the positive, moderate correlation between proportion of ED tweets and the proportion of ED followers (b ¼ .37, p < .01). Pro-ED profiles with more expressions of an ED identity also tended to engage a network more saturated with followers sharing this mutual interest. It is outside the scope of this study to disentangle whether an increase in ED followers or an increase in ED tweets comes first. However, it is conceivable that Pro-ED display, or quantity of display, provides a defining characteristic of profiles’ identities on which to make and break social connections. This Pro-ED saturated network may inhibit connections with outside ED communities and simultaneously attract like-minded individuals. Engagement in this network might promote forming further ED connections on Twitter and may also serve to normalize an ED identity or even foster a Pro-ED collective identity as occurs on other Pro-ED sites [10,11]. This study has several limitations, first, a small, purposeful sample was used to explore rich social network data on the ProED Twitter experience rather than produce generalizable results to all Pro-ED activity on Twitter or other cyberspaces. Notably, although just 45 Pro-ED profiles were studied, a total of 25,294 followers were represented in this study, and analysis encompassed a total of 4,245 original tweets. Thus, even this small sample of 45 Pro-ED profiles provided access to large online circles for socialization and information exchange. Second, it is not certain that all tweets with ED references reflect an unhealthy interest in weight and body image. However, as previously discussed, the strict inclusion criteria were intended to ensure that coding was in context of Pro-ED, which provided the assumption that content with ED keywords reflect an underlying unhealthy perspective. Steps such as pilot testing further intended to ensure the validity of this approach. Finally, demographic information for profiles was limited. English language inclusion criteria narrowed the geographic and cultural scope of this study to English-tweeting Twitter users. The lack of demographic information inhibits comparison of our findings to a control group of profiles not engaged in Pro-ED. Despite these limitations, findings have clinical implications. ED social expression and connection on Twitter might be a source of social support for Pro-ED Twitter profiles and their followers [5,7,25]. However, this might also normalize an ED identity or even foster a Pro-ED collective identity as has been demonstrated on other Pro-ED sites [10,11]. From a clinical perspective, reinforcement of the ED identity is particularly
664
A. Arseniev-Koehler et al. / Journal of Adolescent Health 58 (2016) 659e664
worrisome given that separation from the ED identity is considered critical for recovery from EDs [36]. Accumulation of a like-minded online network might also hinder Pro-ED profiles distancing themselves from the ED identity. Public exchange of Pro-ED content is especially noteworthy in the context of Twitter given that this is a widely used public site, particularly among youth [14]. Many sites other than Twitter censor Pro-ED and content on other sites are usually private. Although censorship might minimize youth’s exposure to ProED, censorship might also push these communities further into secrecy and stigmatize individuals already feeling alienated because of their ED symptoms. Censorship and privacy might additionally insulate these communities from external perspectives. However, public ED expression on Twitter also offers a view for practitioners and researchers into ED-related thoughts and practices. Such expression might provide opportunities to reach subclinical and undiagnosed individuals struggling with disordered eating, who are often difficult to reach given the covert nature of these disorders. Future studies might investigate mechanisms underlying the association between ED tweets and ED followers found in this study, to pinpoint how Pro-ED profiles might influence and be influenced by their online social connections. This study also demonstrates the potential for a few profiles to reach a wide audience, thus studies might also investigate network-based interventions to Pro-ED on Twitter as well as other social media. For example, studies might investigate providing clinically valid resources to social groups of profiles publicly struggling with ED symptoms, such as the profiles and followers identified in this study. Acknowledgments The authors thank Laura Hooper, MS, RD, CD in Adolescent Medicine at Seattle Children’s Hospital, for special consultation on this project and Bradley Kerr for editing assistance. This research was presented in a poster symposium presentation at the Society for Adolescent Health and Medicine Meeting 2015, and at the Social Media and Adolescent Health Research Team Conference 2014. Funding Sources This research was funded by the Mary Gates Research Endowment. References [1] Arcelus J, Mitchell AJ, Wales J, et al. Mortality rates in patients with anorexia nervosa and other eating disorders. A meta-analysis of 36 studies. Arch Gen Psychiatry 2011;68:724e31. [2] Merikangas KR, He JP, Burstein M, et al. Lifetime prevalence of mental disorders in US adolescents: Results from the National Comorbidity StudyAdolescent Supplement (NCS-A). J Am Acad Child Adolesc Psychiatry 2010; 49:980e9. [3] Swanson S, Crow S, Le Grange D, et al. Prevalence and correlates of eating disorders in adolescents. Arch Gen Psychiatry 2011;68:714e23. [4] Rich E. Anorexic dis(connection): Managing anorexia as an illness and an identity. Sociol Health Illn 2006;28:284e305. [5] Overbeke G. Pro-anorexia websites: Content, impact, and explanations of popularity. Mind Matters The Wesleyan J Psychol 2008;3:49e62.
[6] Wilson JL, Peebles R, Hardy KK, et al. Surfing for thinness: A pilot study of pro-eating disorder web site usage in adolescents with eating disorders. Pediatrics 2006;118:e1635e43. [7] Ransom DC, La Guardia JG, Woody EZ, et al. Interpersonal interactions on online forums addressing eating concerns. Int J Eat Disord 2010;43: 161e70. [8] Bardone-Cone AM, Cass KM. Investigating the impact of pro-anorexia websites: A pilot study. Eur Eat Disord Rev 2006;14:256e62. [9] Bardone-Cone AM, Cass KM. What does viewing a pro-anorexia website do? An experimental examination of website exposure and moderating effects. Int J Eat Disord 2007;40:537e48. [10] Maloney P. Online networks and emotional energy. Inform Commun Soc 2013;16:105e24. [11] Giles DC. Constructing identities in cyberspace: The case of eating disorders. Br J Social Psychol 2006;45:463e77. [12] Norris ML, Boydell KM, Pinhas L, et al. Ana and the Internet: A review of pro-anorexia websites. Int J Eat Disord 2006;39:443e7. [13] Campbell K, Peebles R. Eating disorders in children and adolescents: State of the art review. Pediatrics 2014;134:582e92. [14] Lenhart A. Teens, social media, and technology overview 2015. Washington, D.C.: Pew Research Center; 2015. Available at: http://www.pewinternet.org/ 2015/04/09/teens-social-media-technology-2015/. Accessed May 1, 2015. [15] Duca L. Can thinspiration really be #banned from instagram? New York, NY: Huffington Post; 2013. Available at: Huffingtonpost.com. Accessed January 1, 2015. [16] Inc. I. Community guidelines; 2014 [cited 2014]; Available at: https://help. instagram.com/477434105621119/. Accessed January 1, 2015. [17] Tumblr. A New policy against self-harm blogs; 2012 [cited 2013 December 20]; Available at: http://staff.tumblr.com/post/18132624829/self-harmblogs. Accessed January 1, 2015. [18] Facebook. Community standards; 2013 [cited 2013]; Available at: https:// www.facebook.com/communitystandards. Accessed January 1, 2015. [19] Reaves J. Anorexia goes high tech. Time Magazine; 2001. Available at: http:// content.time.com/time/health/article/0,8599,169660,00.html. Accessed January 1, 2015. [20] Huffington C. Pro-anorexia on Twitter: Site allows ‘thinspiration’ accounts. New York, NY: Huffington Post; 2013. Available at: http://www. huffingtonpost.com/2013/01/07/pro-anorexia-on-twitter-thinspiration-_n_ 2396468.html. Accessed January 1, 2015. [21] Glaser BG, Strauss AL. The discovery of grounded theory: Strategies for qualitative research. Chicago, IL: Aldine Publishing Company; 1967. [22] Hanson CL, Cannon B, Burton S, et al. An exploration of social circles and prescription drug abuse through twitter. J Med Internet Res 2013;15:189e202. [23] Gentry J. TwitteR: R based Twitter client. R package version 1.1.7. edition; 2013. [24] Whitehead K. “Hunger Hurts but Starving Works”: A case study of gendered practices in the online pro-eating disorder community. Can J Sociol 2010;35:595e626. [25] Csipke E, Horne O. Pro-eating disorder websites: Users’ opinions. Eur Eat Disord Rev 2007;15:196e206. [26] Borzekowski D, Schenk S, Wilson J, et al. E-ana and e-mia: A content analysis of pro-eating disorder web sites. Am J Public Health 2010;100:1526e34. [27] Peebles R, Wilson J, Litt I, et al. Disordered eating in a digital age: Eating behaviors, health, and quality of life in users of websites with pro-eating disorder content. J Med Internet Res 2012;14:e148. [28] Moreno MA, Kelleher E, Pumper M. Evaluating displayed depression symptoms on social media sites. Social Networking 2013;2:185e92. [29] Garner DM, Olmsted MP, Bohr Y, et al. The eating attitudes test: Psychometric features and clinical correlates. Psychol Med 1982;12:871e8. [30] Team RDC. R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing; 2008. [31] Weisberg S. Applied linear regression. New Jersey: John Wiley and sons; 2014. [32] Hutto C, Yardi S, Gilbert E. A longitudinal study of follower predictors on twitter. CHI’ 13 Proceedings of the SIGCHI Conference on Human Factors in Computing Systems; 2013; 2013:821e30. [33] Long JS. Regression models for categorical and limited dependent variables. Thousand Oaks: Sage Publications; 1997. [34] Spearman C. The proof and measurement of association between two things. Am J Psychol 1904;15:72e101. [35] Ison J, Kent S. Social identity in eating disorders. Eur Eat Disord Rev 2010; 18:475e85. [36] Nordbø RH, Espeset EM, Gulliksen KS, et al. Reluctance to recover in anorexia nervosa. Eur Eat Disord Rev 2012;20:60e7. [37] Gruzd A, Wellman B, Takhteyev Y. Imagining twitter as an imagined community. Am Behav Scientist 2011;45:1294e318. [38] Tiller JM, Sloane G, Schmidt U, et al. Social support in patients with anorexia nervosa and bulimia nervosa. Int J Eat Disord 1997;21:408e22.
www.jahonline.org Original article
#Proana: Pro-Eating Disorder Socialization on Twitter Alina Arseniev-Koehler a, b, *, Hedwig Lee, Ph.D. a, Tyler McCormick, Ph.D. a, c, and Megan A. Moreno, M.D., M.S.Ed, M.P.H. b, d a
Department of Sociology, University of Washington, Seattle, Washington Center for Child Health, Behavior and Development, Seattle Children’s Research Institute, Seattle, Washington Department of Statistics, University of Washington, Seattle, Washington d Department of Pediatrics, University of Washington, Seattle, Washington b c
Article history: Received August 5, 2015; Accepted February 26, 2016 Keywords: Eating disorders; Pro-eating disorder; Social media; Social networks; Social support; Twitter; Adolescent health
A B S T R A C T
Purpose: Proeeating disorder (ED) online movements support engagement with ED lifestyles and are associated with negative health consequences for adolescents with EDs. Twitter is a popular social media site among adolescents that provides a unique setting for Pro-ED content to be publicly exchanged. The purpose of this study was to investigate Pro-ED Twitter profiles’ references to EDs and how their social connections (followers) reference EDs. Methods: A purposeful sample of 45 Pro-ED profiles was selected from Twitter. Profile information, all tweets, and a random sample of 100 of their followers’ profile information were collected for content analysis using the Twitter Application Programming Interface. A codebook based on ED screening guidelines was applied to evaluate ED references. For each Pro-ED profile, proportion of tweets with ED references and proportion of followers with ED references in their own profile were evaluated. Results: In total, our 45 Pro-ED profiles generated 4,245 tweets for analysis. A median of 36.4% of profiles’ tweets contained ED references. Pro-ED profiles had a median of 173 followers, and a median of 44.5% of followers had ED references. Pro-ED profiles with more tweets with ED references also tended to have more followers with ED references (b ¼ .37, p < .01). Conclusions: Findings suggest that profiles which self-identify as Pro-ED express disordered eating patterns through tweets and have an audience of followers, many of whom also reference ED in their own profiles. ED socialization on Twitter might provide social support, but in the ProED context this activity might also reinforce an ED identity. Ó 2016 The Society for Adolescent Health and Medicine. All rights reserved.
Eating disorders (EDs) are complex psychophysiological illnesses that can lead to serious health consequences including the highest mortality rate of any psychiatric disorder [1]. Approximately, 2.7% of US 13e17 year olds experience from an ED [2] and an additional 3.3% of adolescents struggle with subthreshold ED
Conflicts of Interest: The authors have no conflicts of interest or financial disclosures to report. * Address correspondence to: Alina Arseniev-Koehler, 16526 Densmore Ave. N, Shoreline, WA 98133. E-mail address: [email protected] (A. Arseniev-Koehler).
IMPLICATIONS AND CONTRIBUTION
This study investigated Twitter profiles that selfidentified as supportive of eating disorder (ED) lifestyles (Pro-ED). Analysis revealed that Pro-ED profiles often displayed ED references and accumulated networks of Twitter followers who also displayed ED references. These online connections could reinforce an ED identity for those with Pro-ED profiles and their followers.
symptoms such as purging or laxative use [3]. Despite the prevalence and seriousness of these disorders, it can be difficult to reach those who would benefit from treatment. Individuals often try to conceal ED symptoms [4], and many never discuss their struggles with weight and body image with professionals [3]. Given the often secretive nature of EDs, one venue in which those with ED may seek social support or resources is online. ProED online communities emerged in the early 2000s to support ED lifestyles [5]. Previous studies have illustrated that Pro-ED cyberspaces may exacerbate ED risk factors [5e7] such as poor body image [8,9], reinforcement of an individuals’ ED identity
1054-139X/Ó 2016 The Society for Adolescent Health and Medicine. All rights reserved. http://dx.doi.org/10.1016/j.jadohealth.2016.02.012
660
A. Arseniev-Koehler et al. / Journal of Adolescent Health 58 (2016) 659e664
[10,11], or learning and subsequently using unhealthy weight loss methods [5e7,12]. Among youth with EDs, engagement with online Pro-ED communities is common [6] and has recently been suggested as a screening factor for EDs [13]. Such findings highlight the importance of understanding teens’ engagement with Pro-ED content on social media. Twitter, which is used by 33% of US teens, is a unique online social media platform through which to study Pro-ED [14]. As a hub of online public social activity and information exchange, Twitter users create profiles called handles from which they can microblog, or “tweet” (microexpressions of 140 characters or less), as well as follow the feeds of other handles’ tweets, creating an interconnected social network. Users may contribute to a larger tweet conversation on Twitter by adding a hashtag to a key term within the tweet, which connects communities who also use that hashtag around shared topics of interest. With this social structure, Twitter can enable those who ascribe to Pro-ED to express their ED identity and gain a social and information network for this ED identity. For example, the use of the hashtag #proana (proanorexia) allows Twitter users to post and view content that is supportive of ED lifestyles. Although steps to counteract this content have been taken by many sites [15e19] such as Instagram [15,16] and Facebook [18], Twitter has made no attempts to limit Pro-ED content [20]. Little is known regarding the impact of Twitter as a medium to exchange information or create social connections in the Pro-ED community. This information is necessary to understand how providers can best counsel adolescent patients about these online communities, and how researchers can consider interventions to address them. The purpose of this study was to describe the content and social networks of a sample of Pro-ED Twitter profiles. To describe the social networks of Pro-ED profiles, we also examined their followers to determine the proportion which display ED references in their own profiles (hereafter referred to as ED followers). We also investigated if Pro-ED profiles with more ED content had more ED followers. We hypothesized that Pro-ED profiles with more ED expression in their own profiles would have higher proportion of ED followers compared to profiles with less ED expression. Methods Setting This descriptive study used public Twitter data from Pro-ED profiles and their followers, collected between February and April 2014. This study was determined to qualify as not human subjects’ research by the University of Washington’s institutional review board because public Twitter handles are avatars and are not identifiable, living individuals according to local and national regulations. Pro-ED profile and follower sampling strategy Our first goal was to identify a sample of Pro-ED profiles for evaluation. We used purposeful sampling to identify profiles with displayed content supporting a positive attitude toward EDs. Purposeful sampling is a strategy used for investigation into social processes of particular social groups by selecting representative members of that group [21].We applied similar
sampling and methodological framework previously used to explore social circles of Twitter profiles which displayed drug abuse [22]. To identify Pro-ED profiles, we searched for profiles using the hashtag #proana, as “proana” is an established term to describe the Pro-ED movement [5,8e12]. Using the Twitter analytic tool TwitteR [23], data were collected on days with varying intervals between them between the dates of February 18, 2014 and April 3, 2014. We collected samples of 25 tweets and retweets (shares of another’s tweet) which included #proana and were created within the week before the collection day. Sample tweets, along with handle, were collected on different days to minimize sampling of profiles engaged in direct conversations and to obtain a sample which better represented the larger Pro-ED activity on Twitter over time. Inclusion criteria for our sample of Pro-ED profiles included that the profile had least one display of a positive Pro-ED attitude in their profile information or in their most recent 15 tweets. A positive Pro-ED attitude was defined (1) selfidentifying as Pro-ED and/or as having an ED and antirecovery; or (2) expressing a desire for emaciation; or (3) ascribing to a ProED event. Examples of Pro-ED events at the time of data collection included “ABC (Ana Boot Camp) Diet” and a month-long Twitter-based Pro-ED competition to restrict calories, “#CalorieApril,” which were documented in pilot observation of Pro-ED on Twitter. For example, a profile with the tweet “I’m excited for my first day of #calorieapril! Let’s do it! #proana” would have been considered a Pro-ED profile. A total of 211 handles were viewed to achieve our predetermined sample size of 45 Pro-ED profiles. Among the 166 profiles excluded, 63.9% did not meet these criteria for explicit Pro-ED attitude, 28.9% were nonEnglish, 4.8% were inactive or had privacy set to “protected,” and 6.0% indicated a male identity. Given evidence that Pro-ED is a gendered collective identity [24] and that users tend to be female [7,25e27], we excluded explicitly male identities to minimize noise in findings due to gender. Excluded profiles may have met more than one exclusion criteria. Our second goal was to identify a sample of these Pro-ED profiles’ followers for evaluation. For each Pro-ED profile, we took a simple random sample of 100 followers for further examination. If a Pro-ED profile had fewer than 100 followers, all were collected. Codebook development and variables We developed a codebook to examine Pro-ED profiles’ tweets and profile information (profile name, reported name, location, and description) for ED references and to examine their followers’ profile information for ED references. ED variables. The codebook was designed to be applied to 3 types of ED variables: ED references in Pro-ED profiles’ profile information, ED references in Pro-ED profiles’ tweets, and ED references in followers’ profile information. Our strategy to code for ED references was designed following previous research on codebook development [22,28]. First, an extensive list of ED keywords was generated using conceptual categories and keywords from the Eating Attitudes Test-26 [29]. Similar to previous coding strategies for depression on social media [28], coding was grounded using the Eating Attitudes Test26 because this is a validated clinical questionnaire to screen for ED symptoms [29]. Next, we expanded this list by incorporating terms identified in a literature review on EDs and Pro-ED, pilot
A. Arseniev-Koehler et al. / Journal of Adolescent Health 58 (2016) 659e664
661
Table 1 ED references in Pro-ED profiles’ tweets Domain
Sample keywords
Explanations
Explicit ED terms
Anorexic, ana, bulimic, proana, Pro-ED, promia, ednos, eating disorder, wannarexic
Body and body image
Overweight, obese, fatty, skinnier, skeleton, emaciated, hipbones, backbone, bones, collarbone, thighgap, bikinibridge, hipbones, thighs, hips, thinspo, bonespo, perfection
Body weight
Weight, scale, GW, CW, LW, UGW, BMI, pound, calorie
Wannarexic is a term used in Pro-ED culture to describe individuals engaged in Pro-ED online but do not engage in ED behavior offline [11]. Thighgap refers to a gap between thighs. Bikinibridge refers to space between pelvic bones. Thinspo (Thinspiration) is a term which refers to any media or text idolizing thinness [5]. Bonespo (Bonespiration) was an observed term which referred to media or text idolizing emaciation that made bones visible. Perfection is a common theme and term observed and previously documented in Pro-ED [5,12]. GW/CW/LW/UGW were commonly observed lingo for goal weight/current weight/lowest weight/ultimate goal weight, respectively. BMI is an abbreviation for body mass index.
Food and meals Eat or ate Caloric restriction
Food, breakfast, dinner, meals Eating, eat, ate, appetite Starve, hunger, diet, skip, fasted, CalorieApril, ProjectThin, RG, ABCdiet
Binge Compensatory behavior Exercise
Binge, bloated Laxies, laxatives, vomited, throwup, puke, purge Workout, abs, jog, elliptical, exercise, miles, gym, treadmill
CalorieApril was an observed Twitter-based competition for caloric intake in the month of April. ABCdiet and RG refer to “Ana Boot Camp” and Russian Gymnast Diet, which were commonly observed Pro-ED diets on Twitter. Laxies was a commonly observed abbreviation for laxatives
Pro-ED profiles’ tweets were categorized as ED tweets if they included at least one keyword in this codebook. ED ¼ eating disorder.
observation of Pro-ED on Twitter and other cyberspaces, as well as consultation with a clinical pediatric dietitian with expertise in adolescent EDs. Our final codebook for ED references in Pro-ED profiles’ tweets and profile information included a broad list of 143 keywords; given that coding was in context of Pro-ED we wanted to capture all references that may be related to disordered eating behaviors. Our keywords were grouped into nine domains, such as references to bingeing, body and body image, and compensatory behavior including laxative use and purging. Example keywords in these domains included “binged,” “thighgap,” and “puked.” We also included the domain explicit ED terms, which comprised references to types of EDs, such as “bulimia.” We further included domains to capture terms related to eating and exercise, given that these behaviors are clinically important in the context of disordered eating and consistent with our context of evaluating these behaviors on a Pro-ED profile. For a full list of domains and sample keywords, see Table 1. Pro-ED profiles’ tweets were categorized as ED tweets if they included at least one keyword in this codebook. Followers of the Pro-ED profiles were coded using a restricted codebook compared to that used for Pro-ED profiles and their tweets because keyword coding was no longer in context of a profile purposefully selected for a Pro-ED attitude. Thus, keywords in the domains of eating and exercise and in other domains, such as “perfection” (common lingo in Pro-ED culture [5,12]), were not coded as ED references among followers given that these references were less specific to EDs. Twenty-eight keywords were excluded in this restricted codebook. Followers were categorized as ED followers if their profile information included least one keyword or synonym to a keyword. Importantly, the inclusion criteria for Pro-ED profiles were intended to ensure these profiles displayed current, explicit, positive attitudes toward EDs, which provided assumption that content with ED keywords reflect an underlying unhealthy
perspective. Furthermore, followers were defined as those who had voluntarily subscribed to Pro-ED profiles’ tweets and our restrictive coding approach to identify ED followers meant that we identified profiles that explicitly displayed ED references in their public profile information. Therefore, a liberal keyword coding scheme for ED thoughts and behaviors was warranted in this context. Keyword coding also enabled reliable and transparent analysis. Descriptive variables. Descriptive variables for Pro-ED profiles included: number of followers, tweets, retweets, number of favorites (endorsements) of others’ tweets, days since account creation (defined as days between account creation date and analysis date), as well as profile information. We evaluated profile information for three additional variables, including: displayed age, references to the account being anonymous or secret, and displayed body mass index. Pilot testing. This codebook was pilot tested on five Pro-ED profiles and their followers and revised based on consensus before finalization. One revision following pilot testing included manual coding for followers, so that coding would include unique spellings and synonyms for keywords. These were commonly noted in profile information while pilot testing. Data collection and coding procedure The statistical program R [30] with the package TwitteR [23] was used to collect all data from Twitter. TwitteR allows users to collect public data from Twitter using the Twitter Applied Programming Interface. Pro-ED profiles’ profile information was manually coded to evaluate for descriptive variables and ED references. Tweets were coded automatically with R using text recognition for ED keywords in the codebook. Pro-ED profiles’ followers were coded manually.
662
A. Arseniev-Koehler et al. / Journal of Adolescent Health 58 (2016) 659e664
Analysis Descriptive statistics were used to describe the proportion of ED tweets on Pro-ED profiles and proportion of ED followers. Multivariate linear regression [31] was used to predict Pro-ED profiles’ proportion of ED followers from Pro-ED profiles’ proportion of ED tweets. We added the control variable days since account creation given its importance to predicting longitudinal follower accumulation [32]. As the use of proportion data is traditionally inappropriate for linear regression, after viewing the data we performed diagnostics to ensure that the effect of this violation was minimal in this data set [31,33]. We also tested an alternate measure of correlation between Pro-ED profiles’ proportion of ED tweets and their proportion of ED followers, that is, Spearman Rank Correlation [34]. Results Pro-ED profile sample characteristics Forty-five Pro-ED profiles were selected from Twitter, 86.7% (N ¼ 39) of Pro-ED profiles displayed ED references in their profile name, screen name, or description. Furthermore, 64.4% (N ¼ 29) displayed an ED reference in two or three of these locations, and 33.3% (N ¼ 15) displayed an ED reference in all three of these locations. No Pro-ED profiles displayed references in their location. Demographic data were not universally available, 13 profiles (28.9%) displayed an age; the mean was 17.7 (standard deviation [SD] ¼ 2.75) years. Table 2 includes further displayed characteristics. Finally, these accounts tended to be recently created; profiles were created a median of 4.7 months before the analysis date (May 1, 2014). Pro-ED profiles’ tweets Pro-ED profiles had a median of 49 tweets and together these profiles generated a total of 4,245 tweets for analysis. A median of 36.4% and mean of 35.7% of Pro-ED profiles’ tweets had at least one ED reference (SD ¼ 19.5%). ED tweets often included multiple ED references. Among the nine domains in our codebook for ED references, nearly all (91%) of Pro-ED profiles referenced the domain body and body image at least once in their tweets. Some (42.2%) referenced the domain compensatory behavior. The most frequently referenced domain was explicit ED terms, such as “anorexia”; on average 13.2% (SD ¼ 16.4%) of Pro-ED profiles’ tweets referenced this domain. For the prevalence of domains for ED references see Table 3. Sample observed tweets included: “Do not reward yourself with food, you are not a dog #proana,” and “crying after eating 300 cals then binging on another 5000 because you ‘blew it’ #EDlife.” For further results on Pro-ED profiles’ tweet quantity and frequency see Table 4. Pro-ED profiles’ followers Pro-ED profiles had a median of 173 followers (range 5e7,935), and the sample of 45 profiles together yielded a total audience of 25,294 followers. In total 3,719 followers were sampled for evaluation. A median of 44.5% and a mean of 39.3% of sampled followers had ED references in their profile information (SD ¼ 19.4%, range 0%e79.2%). Sample observed references in profile names
Table 2 Pro-ED profiles’ displayed characteristics N (%), Pro-ED Median Mean Standard Range deviation profiles with display Displayed age (years) 13 (28.9%) Displayed body mass 5 (11.1%) index (kg/m2) 4 (8.9%) Anonymitya
17 18.5
17.7 20
2.75 3.1
15e25 17.8e26
d
d
d
d
ED ¼ eating disorder. a Anonymity included proportion of profiles which self-describe the account as “anonymous” or “secret.”
included: “AnasHostage,” “StarvingMeSmall,” and “Anasfatsister22.” Sample observed descriptions with ED references included: “Holding on to hope. Ana/mia/cat/deb. thanks for listening to my story. trigger warning” and “Just a fat girl trying to get skinny. Height: 50 6. Current weight: whale. Goal weight: lighter than air. #thinspireme.” A final sample observed description with references was “just an average girl, dying for perfection. Follow me w i dont bite (excuse the pun).” Sample references have been modified slightly to ensure confidentiality of profiles. Association between ED references in Pro-ED profiles’ tweets and followers This study further investigated Pro-ED profiles’ social circles by testing whether Pro-ED profiles with a higher proportion of ED tweets tended to have a higher proportion of ED followers. In the adjusted model, the proportion of ED tweets was positively associated with the proportion of ED followers (b ¼ .37, p < .01), controlling for the days since the profile was created. Controlling for proportion of ED tweets, more recently created profiles tended to have a higher proportion of ED followers (b ¼ .0002, p < .01). The model explained 29.4% of the variation in proportion of ED followers (R2 ¼ 29.4, p < .001). An added variable plot [31,33] in Figure 1 shows the adjusted linear relationship between proportion of ED tweets and proportion of ED followers. Pro-ED profiles’ proportion of ED tweets were positively, moderately correlated with their proportion of ED followers (S ¼ 8,193.01, Rho ¼ .46, p ¼ .001). Table 3 Prevalence of ED reference domains among 45 Pro-ED profiles and their tweets ED reference domain
N (%), Pro-ED profiles referencing domain Among 45 Pro-ED profiles
Explicit ED termsa Body and body image Body weight Food and meals Eat and ate Caloric restriction Bingeing Compensatory behavior Exercise
33 41 32 29 39 36 17 19 21
(73.3%) (91.1%) (71.1%) (64.4%) (86.7%) (80%) (37.8%) (42.2%) (46.7%)
Mean (SD) proportion of tweets referencing domain
N (%), tweets referencing domain
13.2 11.6 6.2 3.9 7.1 5.7 1.3 1.6 2.2
377 362 221 179 269 164 43 54 83
(16.4%) (13.1%) (8.9%) (4.9%) (6.5%) (6.3%) (3.1%) (2.8%) (4.6%)
Among 4,245 tweets (8.8%) (8.5%) (5.2%) (4.2%) (6.3%) (3.9%) (1.3%) (1.6%) (2.2%)
ED ¼ eating disorder; SD ¼ standard deviation. a Explicit ED terms include references to types of EDs, such as “bulimia” and “anorexia.” For a full explanation of domains see Table 1.
A. Arseniev-Koehler et al. / Journal of Adolescent Health 58 (2016) 659e664 Table 4 Quantity and frequency of Pro-ED profiles’ tweets
Total tweets Percent ED tweets Tweet frequency ED tweet frequency Total tweets and retweets Tweet and retweet frequency Total “favorites” of other tweets
Median
Mean
Standard deviation
Range
49 36.4% 2.2/week .74/week 405
94.3 35.7% 4.6/week 1.5/week 1,969.3
169.4 19.5% 5.9/week 2.2/week 3,748.5
6e1,105 0%e77.6% .2e32/week 0e13/week 8e19,381
17.9/week
57.7/week
81.8/week
.35e366.8/week
72
471.5
1,079.5
0e4,789
ED ¼ eating disorder; SD ¼ standard deviation.
Discussion This study applied keywords from a validated clinical screen to Twitter content, examining a purposeful sample of Pro-ED profiles and their followers. Findings suggest that Pro-ED profiles are indeed created and expressed on the social media site of Twitter, and these online identities tend to attract an audience of followers who are like-minded individuals. Our first finding is that more than a third (36.4%) of Pro-ED profiles’ tweets were ED related. The presence of Twitter profiles dedicated to Pro-ED identities may not be surprising. EDs are known to overwhelm individuals’ identity but offline expression of this identity may be suppressed due to factors such as stigma against EDs [4,35] and ambivalence toward recovery [35,36] which leads individuals to conceal symptoms from peers and caretakers. In contrast, explicit references to EDs were common among our Twitter profiles; 73.3% of Pro-ED profiles referenced a specific ED in their tweets and 42% of our profiles referenced compensatory behavior, including purging or laxatives. The prevalence (86.7%) of ED and of anonymity references in Pro-ED profiles’ profile information further suggests that such accounts are dedicated to the ED identity.
Figure 1. This added variable plot demonstrates the positive correlation between percent ED followers and percent ED tweets among Pro-ED Twitter profiles, after controlling both variables for any variation due to days since ProED profile account creation (R2 ¼ .37, p < .01).
663
The proportion (44.5%) of ED followers of Pro-ED profiles provides evidence that Pro-ED profiles often form social connections on Twitter with other profiles which display an interest in weight and body image. Twitter may provide Pro-ED profiles and their followers a sense of community and an audience, as is provided for general Twitter users [37]. This might provide peer reinforcement for a Pro-ED identity for both Pro-ED profiles and their followers as well as perceived integration into a Pro-ED Twitter network. Similar to other Pro-ED sites, ED socialization on Twitter might be a source of social support for Pro-ED Twitter profiles and their followers [5,7,25]. This support may be particularly important for individuals with ED, who are often stigmatized and lack in-person support [4,38]. However, studies of other Pro-ED sites have also demonstrated that socialization around EDs in the context of an underlying Pro-ED attitude can also yield outcomes such as poor body image [8,9] and learning unhealthy weight loss strategies [6], and such risks likely translate to Twitter. Our third main finding is the positive, moderate correlation between proportion of ED tweets and the proportion of ED followers (b ¼ .37, p < .01). Pro-ED profiles with more expressions of an ED identity also tended to engage a network more saturated with followers sharing this mutual interest. It is outside the scope of this study to disentangle whether an increase in ED followers or an increase in ED tweets comes first. However, it is conceivable that Pro-ED display, or quantity of display, provides a defining characteristic of profiles’ identities on which to make and break social connections. This Pro-ED saturated network may inhibit connections with outside ED communities and simultaneously attract like-minded individuals. Engagement in this network might promote forming further ED connections on Twitter and may also serve to normalize an ED identity or even foster a Pro-ED collective identity as occurs on other Pro-ED sites [10,11]. This study has several limitations, first, a small, purposeful sample was used to explore rich social network data on the ProED Twitter experience rather than produce generalizable results to all Pro-ED activity on Twitter or other cyberspaces. Notably, although just 45 Pro-ED profiles were studied, a total of 25,294 followers were represented in this study, and analysis encompassed a total of 4,245 original tweets. Thus, even this small sample of 45 Pro-ED profiles provided access to large online circles for socialization and information exchange. Second, it is not certain that all tweets with ED references reflect an unhealthy interest in weight and body image. However, as previously discussed, the strict inclusion criteria were intended to ensure that coding was in context of Pro-ED, which provided the assumption that content with ED keywords reflect an underlying unhealthy perspective. Steps such as pilot testing further intended to ensure the validity of this approach. Finally, demographic information for profiles was limited. English language inclusion criteria narrowed the geographic and cultural scope of this study to English-tweeting Twitter users. The lack of demographic information inhibits comparison of our findings to a control group of profiles not engaged in Pro-ED. Despite these limitations, findings have clinical implications. ED social expression and connection on Twitter might be a source of social support for Pro-ED Twitter profiles and their followers [5,7,25]. However, this might also normalize an ED identity or even foster a Pro-ED collective identity as has been demonstrated on other Pro-ED sites [10,11]. From a clinical perspective, reinforcement of the ED identity is particularly
664
A. Arseniev-Koehler et al. / Journal of Adolescent Health 58 (2016) 659e664
worrisome given that separation from the ED identity is considered critical for recovery from EDs [36]. Accumulation of a like-minded online network might also hinder Pro-ED profiles distancing themselves from the ED identity. Public exchange of Pro-ED content is especially noteworthy in the context of Twitter given that this is a widely used public site, particularly among youth [14]. Many sites other than Twitter censor Pro-ED and content on other sites are usually private. Although censorship might minimize youth’s exposure to ProED, censorship might also push these communities further into secrecy and stigmatize individuals already feeling alienated because of their ED symptoms. Censorship and privacy might additionally insulate these communities from external perspectives. However, public ED expression on Twitter also offers a view for practitioners and researchers into ED-related thoughts and practices. Such expression might provide opportunities to reach subclinical and undiagnosed individuals struggling with disordered eating, who are often difficult to reach given the covert nature of these disorders. Future studies might investigate mechanisms underlying the association between ED tweets and ED followers found in this study, to pinpoint how Pro-ED profiles might influence and be influenced by their online social connections. This study also demonstrates the potential for a few profiles to reach a wide audience, thus studies might also investigate network-based interventions to Pro-ED on Twitter as well as other social media. For example, studies might investigate providing clinically valid resources to social groups of profiles publicly struggling with ED symptoms, such as the profiles and followers identified in this study. Acknowledgments The authors thank Laura Hooper, MS, RD, CD in Adolescent Medicine at Seattle Children’s Hospital, for special consultation on this project and Bradley Kerr for editing assistance. This research was presented in a poster symposium presentation at the Society for Adolescent Health and Medicine Meeting 2015, and at the Social Media and Adolescent Health Research Team Conference 2014. Funding Sources This research was funded by the Mary Gates Research Endowment. References [1] Arcelus J, Mitchell AJ, Wales J, et al. Mortality rates in patients with anorexia nervosa and other eating disorders. A meta-analysis of 36 studies. Arch Gen Psychiatry 2011;68:724e31. [2] Merikangas KR, He JP, Burstein M, et al. Lifetime prevalence of mental disorders in US adolescents: Results from the National Comorbidity StudyAdolescent Supplement (NCS-A). J Am Acad Child Adolesc Psychiatry 2010; 49:980e9. [3] Swanson S, Crow S, Le Grange D, et al. Prevalence and correlates of eating disorders in adolescents. Arch Gen Psychiatry 2011;68:714e23. [4] Rich E. Anorexic dis(connection): Managing anorexia as an illness and an identity. Sociol Health Illn 2006;28:284e305. [5] Overbeke G. Pro-anorexia websites: Content, impact, and explanations of popularity. Mind Matters The Wesleyan J Psychol 2008;3:49e62.
[6] Wilson JL, Peebles R, Hardy KK, et al. Surfing for thinness: A pilot study of pro-eating disorder web site usage in adolescents with eating disorders. Pediatrics 2006;118:e1635e43. [7] Ransom DC, La Guardia JG, Woody EZ, et al. Interpersonal interactions on online forums addressing eating concerns. Int J Eat Disord 2010;43: 161e70. [8] Bardone-Cone AM, Cass KM. Investigating the impact of pro-anorexia websites: A pilot study. Eur Eat Disord Rev 2006;14:256e62. [9] Bardone-Cone AM, Cass KM. What does viewing a pro-anorexia website do? An experimental examination of website exposure and moderating effects. Int J Eat Disord 2007;40:537e48. [10] Maloney P. Online networks and emotional energy. Inform Commun Soc 2013;16:105e24. [11] Giles DC. Constructing identities in cyberspace: The case of eating disorders. Br J Social Psychol 2006;45:463e77. [12] Norris ML, Boydell KM, Pinhas L, et al. Ana and the Internet: A review of pro-anorexia websites. Int J Eat Disord 2006;39:443e7. [13] Campbell K, Peebles R. Eating disorders in children and adolescents: State of the art review. Pediatrics 2014;134:582e92. [14] Lenhart A. Teens, social media, and technology overview 2015. Washington, D.C.: Pew Research Center; 2015. Available at: http://www.pewinternet.org/ 2015/04/09/teens-social-media-technology-2015/. Accessed May 1, 2015. [15] Duca L. Can thinspiration really be #banned from instagram? New York, NY: Huffington Post; 2013. Available at: Huffingtonpost.com. Accessed January 1, 2015. [16] Inc. I. Community guidelines; 2014 [cited 2014]; Available at: https://help. instagram.com/477434105621119/. Accessed January 1, 2015. [17] Tumblr. A New policy against self-harm blogs; 2012 [cited 2013 December 20]; Available at: http://staff.tumblr.com/post/18132624829/self-harmblogs. Accessed January 1, 2015. [18] Facebook. Community standards; 2013 [cited 2013]; Available at: https:// www.facebook.com/communitystandards. Accessed January 1, 2015. [19] Reaves J. Anorexia goes high tech. Time Magazine; 2001. Available at: http:// content.time.com/time/health/article/0,8599,169660,00.html. Accessed January 1, 2015. [20] Huffington C. Pro-anorexia on Twitter: Site allows ‘thinspiration’ accounts. New York, NY: Huffington Post; 2013. Available at: http://www. huffingtonpost.com/2013/01/07/pro-anorexia-on-twitter-thinspiration-_n_ 2396468.html. Accessed January 1, 2015. [21] Glaser BG, Strauss AL. The discovery of grounded theory: Strategies for qualitative research. Chicago, IL: Aldine Publishing Company; 1967. [22] Hanson CL, Cannon B, Burton S, et al. An exploration of social circles and prescription drug abuse through twitter. J Med Internet Res 2013;15:189e202. [23] Gentry J. TwitteR: R based Twitter client. R package version 1.1.7. edition; 2013. [24] Whitehead K. “Hunger Hurts but Starving Works”: A case study of gendered practices in the online pro-eating disorder community. Can J Sociol 2010;35:595e626. [25] Csipke E, Horne O. Pro-eating disorder websites: Users’ opinions. Eur Eat Disord Rev 2007;15:196e206. [26] Borzekowski D, Schenk S, Wilson J, et al. E-ana and e-mia: A content analysis of pro-eating disorder web sites. Am J Public Health 2010;100:1526e34. [27] Peebles R, Wilson J, Litt I, et al. Disordered eating in a digital age: Eating behaviors, health, and quality of life in users of websites with pro-eating disorder content. J Med Internet Res 2012;14:e148. [28] Moreno MA, Kelleher E, Pumper M. Evaluating displayed depression symptoms on social media sites. Social Networking 2013;2:185e92. [29] Garner DM, Olmsted MP, Bohr Y, et al. The eating attitudes test: Psychometric features and clinical correlates. Psychol Med 1982;12:871e8. [30] Team RDC. R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing; 2008. [31] Weisberg S. Applied linear regression. New Jersey: John Wiley and sons; 2014. [32] Hutto C, Yardi S, Gilbert E. A longitudinal study of follower predictors on twitter. CHI’ 13 Proceedings of the SIGCHI Conference on Human Factors in Computing Systems; 2013; 2013:821e30. [33] Long JS. Regression models for categorical and limited dependent variables. Thousand Oaks: Sage Publications; 1997. [34] Spearman C. The proof and measurement of association between two things. Am J Psychol 1904;15:72e101. [35] Ison J, Kent S. Social identity in eating disorders. Eur Eat Disord Rev 2010; 18:475e85. [36] Nordbø RH, Espeset EM, Gulliksen KS, et al. Reluctance to recover in anorexia nervosa. Eur Eat Disord Rev 2012;20:60e7. [37] Gruzd A, Wellman B, Takhteyev Y. Imagining twitter as an imagined community. Am Behav Scientist 2011;45:1294e318. [38] Tiller JM, Sloane G, Schmidt U, et al. Social support in patients with anorexia nervosa and bulimia nervosa. Int J Eat Disord 1997;21:408e22.