Are Aspects of Twitter Use Associated with Reduced Depressive Symptoms? The Moderating Role of In-Person Social Support

Abstract

In a two-wave, 4-month longitudinal study of 308 adults, two hypotheses were tested regarding the relation of Twitter-based measures of online social media use and in-person social support with depressive thoughts and symptoms. For four of five measures, Twitter use by in-person social support interactions predicted residualized change in depression-related outcomes over time; these results supported a corollary of the social compensation hypothesis that social media use is associated with greater benefits for people with lower in-person social support. In particular, having a larger Twitter social network (i.e., following and being followed by more people) and being more active in that network (i.e., sending and receiving more tweets) are especially helpful to people who have lower levels of in-person social support. For the fifth measure (the sentiment of Tweets), no interaction emerged; however, a beneficial main effect offset the adverse main effect of low in-person social support.

Social connectedness helps people cope with almost all forms of mental health problems, especially depression.^1–5 More recent evidence suggests that these benefits can derive from online as well as in-person social relations, especially those characterized by positive interactions.^6–8 Less clear, however, are the combined benefits of online and in-person social connectedness. Two theories make opposite predictions. The rich-get-richer hypothesis suggests that people who have more extensive in-person social support systems should derive greater benefits from social media, as they use it to maintain, improve, and enhance their already well-established social networks.^9–11 For example, a popular person may use social media to keep up with recent events in the lives of their large number of friends.

Alternatively, the social compensation (or poor-get-richer) hypothesis suggests that people with weaker in-person social supports should derive greater benefits from social media use, as they use it to find online niches where they are more accepted.¹⁰ For example, a shy or socially awkward person who is challenged by in-person relationships may be more comfortable in online spaces where they have more control over the pace of social interactions and can take more time to curate their responses.

The primary goal of this study was to test these two competing theories. Toward this end, we conducted a longitudinal investigation of the prospective relation of social media use (specifically, Twitter) with depressive symptoms for people with different levels of in-person social support. On the one hand, support for the rich-get-richer hypothesis would consist of an interaction between Twitter use (TU) and in-person social support, in which TU predicts greater decreases in depressive symptoms for people with higher levels of in-person social support. On the other hand, support for the social compensation hypothesis would consist of an interaction in which TU predicts greater decreases in depressive symptoms for people with lower levels of in-person social support.

Previous support for the rich-get-richer hypothesis has taken several forms. Reich's review reveals several lines of research.¹² One set of studies suggests that personality characteristics such as extraversion are correlated with time spent on social media and number of online friends.¹³ The assumption here is that the correlation between in-person and online social networks is due to a “third” variable, extraversion, which promotes both. A second set of studies reveals that online and in-person social support network parameters are positively correlated, suggesting that they complement each other in ways often seen as consistent with the rich-get-richer hypothesis.^11,14

Other research supports the social compensation hypothesis. For example, Reich's review suggests that people with social anxiety, introversion, and low self-esteem spend more time online, where they experience better social connection with others.^12,15–19 Complementing this work, Rains and Tsetsi's cross-sectional analyses of Pew survey data suggest that social inequities in the in-person world are less evident online, suggesting that the online world has the potential to provide kinds of support that may not always be available in person.²⁰

Two aspects of these studies limit their relevance. First, these results are largely based on cross-sectional analyses, whereas the underlying theories anticipate prospective relations. Second, these results largely focus on the main effects of online and in-person social variables. The current hypotheses are more nuanced, focusing on potential moderators of these relations. In this context, a critical corollary to the rich-get-richer hypothesis is that social media use will be especially helpful to people who already have strong in-person social relations. Conversely, an important corollary to the social compensation hypothesis is that social media use is especially helpful for people who have weak in-person social relations because social media platforms provide access to social supports that they cannot obtain in person.

This study examined the longitudinal effects of TU, in-person social support, and their interaction on depressive thoughts and symptoms in a general population of TUs. The rich-get-richer and social compensation hypotheses should pertain to most social media platforms; however, we elected to focus on Twitter because by 2016 it was the most popular social media platform²¹ that had public access to its content. (Facebook was the most popular; however, since Facebook upgraded its privacy settings in 2014, the proportion of its data that is publicly available for mining greatly diminished.²²)

Methods

Participants

We used Amazon.com Mechanical Turk to recruit master workers (workers with 98% approval ratings), who used Twitter regularly.²³ We offered $4.00 USD for each worker's participation. Participants were English-speaking Americans, 18–50 years old, who provided valid Twitter handles and had used Twitter in the past 3 months.^a Of 333 participants, we eliminated 27 who completed the survey too quickly (spi <1.0),²⁴ scored more than two SDs above the mean on a lie scale²⁵ or a social desirability scale,^26,27 or failed either of two validity-check items.²⁸ Between waves 1 and 2, we lost 36 participants to attrition. Attrition was unrelated to all demographic and survey variables (p's > 0.20). We used full information maximum likelihood estimation methods to include participants with partial data. Sample demographics included gender (52.2% female), age (M = 30.92 years, SD = 5.60), and ethnicity (10.54% Black, 75.20% White, 6.36% Hispanic or Mexican American, 10.02% Asian or Asian American, 0.81% Native American, 1.34% other).

Measures

Social support was measured at wave 1. In keeping with modern conceptualizations of depression as a dimensional continuum, we operationalized the construct as a composite of depressive thoughts and symptoms, measured at both waves 1 and 2.²⁹ All Twitter variables were obtained for the time interval between the two waves.

Perceived Social Support Scales

The Perceived Social Support Scale (PSSS) is a 40-item questionnaire measuring perceived in-person social support from both friends and family. Items are rated on binary (yes/no) scales. We used only the 20-item subscale about friends' support.³⁰ The instrument has excellent psychometric support.^31,32 In this study, reliability was high (KR-20 = 0.92).

Cognitive Triad Inventory

The Cognitive Triad Inventory (CTI) is a 36-item scale assessing negative views of self, world, and future.³³ Items are rated on 7-point scales: 1 = totally agree to 7 = totally disagree. Higher scores represent higher levels of depressive thinking.³³ The CTI has strong psychometric support.^34,35 In this study, coefficient alpha was 0.97 at both waves.

Beck Depression Inventory-Version II

The Beck Depression Inventory-Version II (BDI-II) is a 21-item self-report measure of cognitive, behavioral, and emotional depressive symptoms. Items offer a series of statements ranging from 0 (no presence of the symptom) to 3 (severe form of the symptom).³⁶ The BDI-II has excellent psychometric properties.^36–38 In the current sample, coefficient alpha was 0.92 at both waves.

Twitter variables

Utilizing participants' handles, the Twitter data archives, and the official public representational state transfer application programming interfaces (REST APIs) provided by Twitter,³⁹ we obtained information about Twitter activity between waves 1 and 2 to derive five indices of TU: (1) Follows, the average number of people that a person followed between waves 1 and 2 (computed as the mean of a user's da; (2) Followers, the average number of people who follow the target person^b; (3) Tweets, the number of messages posted on Twitter by the target person; (4) Retweets, the number of messages reposted on Twitter (5) Valence, the sentiment of content posted on Twitter (assessed under the premise that more positively toned communications would be associated with more supportive relationships).⁴⁰ We measured sentiment using Nielsen's AFINN-111 glossary of microblog words and phrases.⁴¹ Each glossary entry is associated with a sentiment rating ranging from very negative (−5) to very positive (+5).^42,43 We obtained each Tweet posted by every participant over the selected time frame. From each Tweet, all whitespace and special characters were removed, leaving only words. If a word or phrase appeared in the glossary, it was assigned a numeric sentiment rating. Words not in the glossary were assigned values of 0. Within participant, sentiment ratings were summed across all Tweets posted during the interwave interval. Thus, scores increase when a participant uses either a larger number of positively valenced words or uses words that have higher valence ratings (or both). The number of neutral words is irrelevant to the sum as their valence scores are 0. Correlations among Twitter variables appear in Table 1.

Table 1.

Correlations Among Study Variables

	Variable	1.	2.	3.	4.	5.
1.	Follows	1.00
2.	Followers	0.42	1.00
3.	Tweets	0.16	0.14	1.00
4.	Retweets	0.20	0.52	0.49	1.00
5.	Sentiment	0.18	0.06	0.38	0.15	1.00

Note: Correlations ≥0.10 are significant at p < 0.05.

Procedure

Institutional Review Board approval was obtained for this study. Participants completed all surveys online. Through MTurk, we retained worker ID numbers. Approximately 4 months later, we used MTurk worker IDs to recontact all wave 1 participants who met our screening criteria, asking them to complete the online survey again. We then linked both waves of data to the Twitter data-mining results.

Results

Descriptive statistics

Means and standard deviations of all waves 1 and 2 variables appear in Table 2. Descriptive statistics are similar to other studies of community adults, although mean levels of depression were slightly elevated.^44,45 At wave 1, ∼32% reported at least mild depressive symptoms, and 21% reported at least moderate depressive symptoms on the BDI.⁴⁶

Table 2.

Descriptive Statistics for Study Variables

Measure	Wave 1		Wave 2
Measure	Mean	SD	Mean	SD
Cognitive triad (CTI)	158.84	34.58	157.88	34.34
Depression (BDI-II)	11.18	11.43	10.09	11.20
Perceived social support (PSSS)	15.97	4.76	15.57	4.92
Workplace victimization (NAQ)	29.87	8.35	33.01	16.37
Social desirability (MCSD-C)	4.75	2.93	—	—
Lie scale (LIE)	3.43	2.79	—	—

	Interval between waves 1 and 2
	Mean	SD
Number of follows (people followed by participants on Twitter)	280.60	630.12
Number of followers (people following participants on Twitter)	364.34	1617.38
Number of Tweets sent by participants	59.34	177.24
Number of times retweeted	17.68	109.14
Sum of Tweet sentiment scores	37.66	197.06

BDI-II, Beck Depression Inventory-version II; CTI, Cognitive Triad Inventory; LIE, Lie Scale of the Revised Eysenck Personality Questionnaire-Short Form; MCSF-C, Marlowe-Crowne Social Desirability-Short Form C; NAQ, Negative Activities Questionnaire-Revised; PSSS, Perceived Social Support Scale.

Primary analyses

The correlation between BDI and CTI was 0.81 at wave 1 and 0.85 at wave 2. Given these very strong relations and given that our goal was to predict depressive thoughts and symptoms, we formed a composite depressive thoughts and symptoms variable (DEP) by standardizing and averaging these two measures at each wave. (All of the following analyses were conducted as described and then repeated with additional control variables: age, sex, and race. All significant results remained significant, and all nonsignificant results remained nonsignificant.)

In our primary analyses, we examined the main and interaction effects of TU (Follows, Followers, Tweets, Retweets, and Sentiment) and in-person social support (PSSS) to predict change in depressive thoughts and symptoms. Our dependent variable was wave 2 depressive symptoms (DEP₂). Wave 1 depressive symptoms (DEP₁) served as a control variable. These analyses were based on the following general regression model: DEP₂ = b₀ + b₁DEP₁ + b₂TU + b₃PSSS + b₄TU × PSSS. In this way, TU, PSSS, and the TU × PSSS interaction are used to predict residualized change in depressive thoughts and symptoms from waves 1 to 2 (generally regarded as superior to the prediction of raw change or difference scores).⁴⁷ When the interaction was significant, we plotted the effects of the Twitter variable on the outcome at high and low levels of the in-person predictor. [In the figures, high = mean + 1 SD, and low = mean – 1 SD (or 0 when mean – 1 SD <0 and 0 was the lowest possible value on the scale.] When the interaction was not significant, we eliminated it and plotted the main effects. In each of these five analyses (one for each TU variable), wave 1 DEP served as a covariate. With N = 308, we had 0.80 power to detect even a small effect (f² = 0.02) and 0.99 power to detect a medium effect (f² = 0.15) of our key predictor in a four-predictor regression.

As shown in Table 3, four of five interactions were significant. In the fifth analysis, the Sentiment main effect was significant (Fig. 1). The four significant interactions revealed that for people with low in-person social support, following more people on Twitter, being followed by more people on Twitter, sending more Tweets, and having more of those Tweets Retweeted predicted improvement in depressive thoughts and symptoms. The same was not true for people with high levels of in-person support. For these four variables, this pattern is consistent with the social compensation corollary. The fifth analysis revealed a strong longitudinal main effect for Sentiment. Use of positive sentiment on Twitter was associated with a longitudinal reduction in depressive thoughts and symptoms for people with either high or low in-person social support. This result does not explicitly support either the rich-get-richer or the social compensation hypotheses. Rather, it suggests that use of positive sentiment in Twitter predicts reductions in depressive thoughts and symptoms for all people (i.e., everyone gets richer).

FIG. 1.

Main effects and interactions of Twitter use and in-person SS on residualized change in depressive thoughts and symptoms (depressive change) over time. DEP, mean of standardized Cognitive Triad Inventory and Beck Depression Inventory-II; SS, social support.

Table 3.

Regression of Wave 2 DEP onto Twitter Variables, Perceived Social Support Scale, and Their Interaction, Controlling for Wave 1 DEP

Model	Predictor	B	SE (B)	β	t-Test	p
1	Intercept	0.165	0.081		1.98	0.048
	DEP wave 1	0.928	0.052	0.94	17.76	0.001
	PSSS	−0.315	0.095	−0.17	−3.32	0.001
	Number of Twitter follows	−0.00021	0.00016	−0.01	−1.29	0.199
	PSSS × Twitter follows	0.00075	0.00015	0.35	5.22	0.001
2	Intercept	0.251	0.082		3.32	0.001
	DEP wave 1	0.967	0.052	1.04	18.69	0.001
	PSSS	−0.217	0.088	−0.12	−2.47	0.013
	Number of Twitter followers	−0.00020	0.00006	−0.33	−3.48	0.001
	PSSS × Twitter followers	0.00030	0.00007	0.55	4.18	∼0.000
3	Intercept	0.135	0.080		1.75	0.079
	DEP wave 1	0.838	0.046	0.81	18.08	0.001
	PSSS	−0.250	0.098	−0.14	−2.55	0.011
	Number of Tweets	0.00001	0.00051	0.01	0.00	0.971
	PSSS × Tweets	0.00130	0.00052	0.10	2.50	0.013
4	Intercept	0.134	0.071		1.82	0.069
	DEP wave 1	0.835	0.047	0.80	17.61	0.001
	PSSS	−0.232	0.095	−0.12	−2.44	0.015
	Number of Retweets	−0.00069	0.00070	−0.04	−0.99	0.323
	PSSS × Retweets	0.00240	0.00103	0.07	2.33	0.021
5	Intercept	0.166	0.070		2.22	0.026
	DEP wave 1	0.850	0.048	0.83	17.75	0.001
	PSSS	−0.180	0.098	−0.10	−1.84	0.066
	Tweet sentiment	−0.00125	0.00040	−0.13	−3.10	0.002
	PSSS × Tweet sentiment	−0.00014	0.00075	−0.01	−0.19	0.849

DEP, mean of standardized Cognitive Triad Inventory and Beck Depression Inventory-II.

Discussion

This study examined the longitudinal effects of TU as a means for offsetting the adverse effects of low social support on depressive thoughts and symptoms. Two key results emerged: first, support emerged for our corollary to the social compensation (or poor-get-richer) hypothesis but not for the rich-get-richer hypothesis. Four aspects of TU were related to reductions in depressive thoughts and symptoms, but only for people with low initial levels of in-person social support. Second, conveying positive sentiment through Twitter predicted a reduction in depressive thoughts and feelings, irrespective of people's level of in-person social support. Below, we elaborate on these findings and their implications.

Our first set of findings was consistent with our corollary to the social compensation hypothesis. People with low social support showed improvements in depressive thoughts and feelings over time if they reported four markers of TU: following more people on Twitter, having more people follow them on Twitter, posting more Tweets, and having more of their posts retweeted by others. These markers were unrelated to depressive thoughts and feelings of people who already had high levels of in-person social support. These results support Baker and Algorta's observation that the effect of social media on depression-related outcomes is complicated by social, psychological, and behavioral moderators.⁸ The current research provides longitudinal evidence that in-person social support may be one such moderator. These results also suggest that social media might be a way to combat the adverse effects of low social support on mental health. This possibility is commensurate with the conventional wisdom that having one or two good friends in one social niche can offset social adversity in other social niches.^48,49 Perhaps social media platforms represent a modern version of such niches. We urge caution along these lines as previous research has also shown that the relation of Twitter with depression-related outcomes varies as a function of how (and when) Twitter is used^6,7,50–54 (also refer literature reviews by Guntuku et al.⁵⁵ and Hur and Gupta⁵⁶).

Three reasons for this finding are possible.^9,57 One is that meeting people with similar interests or characteristics may be easier online than in person, especially when such people are not available within one's in-person social networks.^57–61 For people who lack these affiliations, connecting with others online may have an especially strong impact. A second explanation is that the online channels of communication are simpler, such that people who find it challenging to develop supportive in-person social networks may be more effective in the more restricted online world of social media. One's ability to interpret nonverbal cues, one's physical characteristics, one's proper use of vocal tone, and one's timing of social responses may be less important online than in person. Nesi et al. referred to this as cue absence in their transformation theory.^62–65 In a related vein, online interactions tend to be asynchronous. Delays between online communications might allow people the time to compose more effective responses.⁶² Understanding the mechanisms that underlie these results represents an important avenue for future research. A third explanation for these results is that the value added by having online followers may not be as beneficial to people who already have strong in-person social support, at least insofar as reducing depressive thoughts and symptoms are concerned. Some evidence even suggests that having a very large number of online friends may actually be associated with negative outcomes.^66,67 The current interaction plots in Figure 1 somewhat reflect this possibility, in that some of our TU variables appeared to have adverse effects for people who had strong in-person social supports. We caution against overinterpreting this result, however, as the slopes for participants with low social support were not statistically significant.

Our second set of findings was the significant main effect of Twitter sentiment, which offset the adverse effect of low in-person social support. Two aspects of this result deserve emphasis. First, this finding cannot be explained as consequences of depression, as it derives from longitudinal analyses in which prior levels of depression were statistically controlled. Second, these results were not moderated by level of social support. The effects of positive sentiment applied to people at all levels of social support. Indeed, positive Twitter sentiment offset much of the depressive effects of low in-person social support. People with problematic social networks but highly positive Twitter sentiment had similar levels of depressive symptoms as did people with strong social networks but more negative Twitter sentiment, reminiscent of Granovetter's early work on the strength of weak ties.⁶⁸

Shapiro and Margolin's extensive literature review describes at least four reasons why effective use of online social media platforms could offset the adverse effects of problematic face-to-face relationships, especially with respect to cognitive and emotional outcomes.⁵⁷ First, people can engage in selective self-presentation more easily online than in person. By crafting carefully their online communications and constructing their online persona, some people can accrue more positive feedback online than they can in person, which may in turn result in improvement on psychological outcomes.^64,65 Second, connecting with similar people or with people who share similar interests may be easier for some people online than in person, especially when such affiliations are not available within in-person social networks.^58–61 Third, through the Internet, communicating with others from more diverse intellectual, political, and social backgrounds can expand one's self-identity while enhancing feelings of belongingness and affiliation.⁶⁹ Trepte et al. hypothesized that large, diverse groups may feel more connected with each other online and are thus more likely to support each other.⁷⁰ Fourth, self-disclosure may be easier online than in person, potentially facilitating online social relationships or enabling people to practice for in-person relationships.^71,72

Taken together, these results begin to suggest interesting supplemental strategies in the prevention of depression in people who are at risk because of low social support. The current findings, derived from one of very few longitudinal studies in this area, increase our understanding about prospective (not just correlational) relations and could have implications for the use of social media in prevention research.¹² A powerful next step will be true experimental research designs in which positive use of social media is actively manipulated, so that its causal effect on mental health outcomes can be assessed. If successful, online social skills training could become a valuable component of comprehensive depression prevention efforts.

Several shortcomings of this study suggest important avenues for future research. The first focuses on our sentiment analysis. Although examining the actual sentiment conveyed by Twitter communications is a powerful step, in-depth content analysis of people's Tweets could reveal more about more specific aspects of people's communications that might be responsible for the relation of sentiment with depression-related outcomes. Furthermore, in short textual passages (such as Tweets), it is extremely difficult to reliably measure issues such as sarcasm and irony. Also, some kinds of negatively toned messages (e.g., expressing distress) could serve as triggers for positive responses (e.g., emotional support). Second, depressive thoughts and symptoms are extremely important mental health outcomes, emblematic of one of the most common and debilitating classes of mental illnesses; however, many other important clinical outcomes should be explored, including Internet addiction, social anxiety, and obsessive-compulsive disorder.^73–75 Third, our study focused only on Twitter. Other social media platforms exist generate very different kinds of risks and benefits, which should be explored. Fourth, we used an observational/correlation research design, which leaves various “third variables” uncontrolled. Random assignment to high versus low Twitter conditions could control for self-selection factors such as extraversion or level of depression. Fifth, although use of MTurk for participant recruitment has certain strengths, weaknesses have also been documented. These include crosstalk among participants, misrepresentation of personal characteristics to qualify for studies, and provision of unreliable results.^76–78 Although these issues do seem to be characteristic of some MTurk participants, research shows that these problems actually occur at similar rates in samples obtained from more conventional methods.⁷⁹ Future studies should examine the generalizability of the current results across a wider variety of populations.

Footnotes

Notes

Author Disclosure Statement

No competing financial interests exist.

Funding Information

This research was supported in part by Patricia and Rodes Hart and a Peabody Small Grant awarded to the first author by Peabody College, Vanderbilt University. Contributions by Elizabeth Nick and Rachel Zelkowitz were supported by NIMH T32MH018921-26 and F31MH108241-01A1, respectively.

References

Reijntjes

, Kamphuis

, Prinzie

, et al. Peer victimization and internalizing problems in children: a meta-analysis of longitudinal studies. Child Abuse and Neglect, 2010; 34:244–252.

Gariépy

, Honkaniemi

, Quesnel-Vallée

. Social support and protection from depression: systematic review of current findings in Western countries. British Journal of Psychiatry, 2016; 209:284–293.

Blai

. Health consequences of loneliness: a review of the literature. Journal of American College Health, 1989; 37:162–167.

Heinrich

, Gullone

. The clinical significance of loneliness: a literature review. Clinical Psychology Review, 2006; 26:695–718.

West

, Kellner

, Moore-West

. The effects of loneliness: a review of the literature. Comprehensive Psychiatry, 1986; 27:351–363.

Choi

, Toma

. Social sharing through interpersonal media: patterns and effects on emotional well-being. Computers in Human Behavior, 2014; 36:530–541.

Robinson

, Bonnette

, Howard

, et al. Social comparisons, social media addiction, and social interaction: an examination of specific social media behaviors related to major depressive disorder in a millennial population. Journal of Applied Biobehavioral Research, 2019; 24:e12158.

Baker

, Algorta

. The relationship between online social networking and depression: a systematic review of quantitative studies. Cyberpsychology Behavior and Social Networking, 2016; 19:638–648.

McKenna

KYA

, Green

, Gleason

MEJ

. Relationship formation on the Internet: what's the big attraction?. Journal of Social Issues, 2002; 58:9–31.

10.

Kraut

, Kiesler

, Boneva

, et al. Internet paradox revisited. Journal of Social Issues, 2002; 58:49–74.

11.

Vergeer

, Pelzer

. Consequences of media and Internet use for offline and online network capital and well-being. A causal model approach. Journal of Computer-Mediated Communication, 2009; 15:189–210.

12.

Reich

. Connecting offline social competence to online peer interactions. Psychology of Popular Media Culture, 2017; 6:291.

13.

Gosling

, Augustine

, Vazire

, et al. Manifestations of personality in online social networks: self-reported Facebook-related behaviors and observable profile information. Cyberpsychology Behavior and Social Networking, 2011; 14:483–488.

14.

Saunders

, Chester

. Shyness and the internet: social problem or panacea?. Computers in Human Behavior, 2008; 24:2649–2658.

15.

Bonetti

, Campbell

, Gilmore

. The relationship of loneliness and social anxiety with children's and adolescents' online communication. Cyberpsychology Behavior and Social Networking, 2010; 13:279–285.

16.

Ebeling-Witte

, Frank

, Lester

. Shyness, Internet use, and personality. CyberPsychology and Behavior, 2007; 10:713–716.

17.

Ellison

, Steinfield

, Lampe

. The benefits of Facebook “friends”: social capital and college students' use of online social network sites. Journal of Computer-Mediated Communication, 2007; 12:1143–1168.

18.

Fernandez

, Levinson

, Rodebaugh

. Profiling: predicting social anxiety from Facebook profiles. Social Psychological and Personality Science, 2012; 3:706–713.

19.

Shen

, Williams

. Unpacking time online: connecting internet and massively multiplayer online game use with psychosocial well-being. Communication Research, 2011; 38:123–149.

20.

Rains

, Tsetsi

. Social support and digital inequality: does Internet use magnify or mitigate traditional inequities in support availability?. Communication Monographs, 2017; 84:54–74.

21.

EBiz. Top 15 Most Popular Social Networking Sites. 2016. www.ebizmba.com/articles/social-networking-websites (accessed Mar. 31, 2016)

22.

Magid

. Facebook changes new user default privacy setting to friends only—Adds privacy checkup. Forbes. https://www.forbes.com/sites/larrymagid/2014/05/22/facebook-changes-default-privacy-setting-for-new-users/%7B#%7D33937a8759ac (accessed May 22, 2014).

23.

Amazon. Amazon Mechanical Turk: General questions. 2017. https://www.mturk.com/mturk/help?helpPage=overview%7B#%7Dwhat%7B_%7Dis (accessed Jul. 27, 2017).

24.

Wood

, Harms

, Lowman

, et al. Response speed and response consistency as mutually validating indicators of data quality in online samples. Social Psychological and Personality Science, 2017; 8:454–464.

25.

Eysenck

SBG

, Eysenck

, Barrett

. A revised version of the psychoticism scale. Personality and Individual Differences, 1985; 6:21–29.

26.

Reynolds

. Development of reliable and valid short forms of the Marlowe-Crowne Social Desirability Scale. Journal of Clinical Psychology, 1982; 38:119–125.

27.

Crowne

, Marlowe

. A new scale of social desirability independent of psychopathology. Journal of Consulting Psychology, 1960; 24:349–354.

28.

DeSimone

, Harms

. Dirty data: the effects of screening respondents who provide low-quality data in survey research. Journal of Business and Psychology, 2018; 33:559–577.

29.

Solomon

ARI

, Haaga

DAF

, Arnow

. Is clinical depression distinct from subthreshold depressive symptoms? A review of the continuity issue in depression research. The Journal of Nervous and Mental Disease, 2001; 189:498–506.

30.

Procidano

, Heller

. Measures of perceived social support from friends and from family: three validation studies. American Journal of Community Psychology, 1983; 11:1–24.

31.

Cole

, Milstead

. Behavioral correlates of depression: antecedents or consequences?. Journal of Counseling Psychology, 1989; 36:408–416.

32.

Dubois

, Neville

, Parra

, et al. Testing a new model of mentoring. New Directions for Youth Development, 2002; 2002:21–57.

33.

Beckham

, Leber

, Watkins

, et al. Development of an instrument to measure Beck's cognitive triad: the Cognitive Triad Inventory. Journal of Consulting and Clinical Psychology, 1986; 54:566–567.

34.

McIntosh

, Fischer

. Beck's cognitive triad: one versus three factors. Canadian Journal of Behavioural Science, 2000; 32:153.

35.

Zauszniewski

, Panitrat

, Youngblut

. The children's Cognitive Triad Inventory: reliability, validity, and congruence with Beck's cognitive triad theory of depression. Journal of Nursing Measurement, 1999; 7:101–115.

36.

Beck

, Steer

, Ball

, et al. Comparison of Beck Depression Inventories-IA and-II in Psychiatric Outpatients. Journal of Personality Assessment, 1996; 67:588–597.

37.

Steer

, Beck

. (2000) The Beck Depression Inventory-II. In Craighead

, Nemeroff

, eds. The Corsini Encyclopedia of Psychology and Behavioral Science. 3rd ed. New York: Wiley, pp. 178–179.

38.

Steer

, Ball

, Ranieri

, et al. Dimensions of the Beck Depression Inventory-II in clinically depressed outpatients. Journal of Clinical Psychology, 1999; 55:117–128.

39.

Twitter. Twitter developer documentation: REST APIs. https://dev.twitter.com/rest/public (accessed Jul. 27, 2017).

40.

Twitter. New user FAQs. https://support.twitter.com/articles/13920 (accessed Jul. 27, 2017).

41.

Nielsen FÅ. (2011) A new ANEW: Evaluation of a word list for sentiment analysis in microblogs. In: Proceedings of the ESWC2011 Workshop on “Making Sense of Microposts”: Big Things Come in Small Packages. 718 in CEUR Workshop Proceedings, pp. 93–98.

42.

Hansen

, Arvidsson

, Nielsen

FÅ

, et al. (2011) Good friends, bad news-affect and virality in twitter. In: Park

, Yang

, Lee

, eds. Future Information Technology. Berlin Heidelberg, Germany: Springer, pp. 34–43.

43.

Nielsen FÅ. AFINN. http://www2.imm.dtu.dk/pubdb/views/publication%7B_%7Ddetails.php?id=6010 (accessed Jul. 27, 2015).

44.

Shean

, Baldwin

. Sensitivity and specificity of depression questionnaires in a college-age sample. The Journal of Genetic Psychology, 2008; 169:281–292.

45.

Whisman

, Richardson

. Normative data on the Beck Depression Inventory–second edition (BDI-II) in college students. Journal of Clinical Psychology, 2015; 71:898–907.

46.

Dozois

DJA

, Dobson

, Ahnberg

. A psychometric evaluation of the Beck Depression Inventory–II. Psychological Assessment, 1998; 10:83–89.

47.

Cronbach

, Furby

. How we should measure “change”: or should we?. Psychological Bulletin, 1970; 74:68–80.

48.

Eisenberg

, Vaughan

, Hofer

. (2009) Temperament, self-regulation, and peer social competence. In Rubin

, Bukowski

, Laursen

, eds. Social, Emotional, and Personality Development in Context. Handbook of Peer Interactions, Relationships, and Groups. New York, NY: The Guilford Press, pp. 473–489.

49.

Frankel

, Cantwell

, Myatt

. (1996) Helping ostracized children: social skills training and parent support for socially rejected children. In Hibbs

, Jensen

, eds. Psychosocial Treatments for Child and Adolescent Disorders: Empirically Based Strategies for Clinical Practice. Washington, DC: American Psychological Association, pp. 595–617.

50.

Cavazos-Rehg

, Krauss

, Sowles

, et al. A content analysis of depression-related tweets. Computers in Human Behavior, 2016; 54:351–357.

51.

Garett

, Liu

, Young

. The relationship between social media use and sleep quality among undergraduate students. Information Communication and Society, 2018; 21:163–173.

52.

McIver

, Hawkins

, Chunara

, et al. Characterizing sleep issues using Twitter. Journal of Medical Internet Research, 2015; 17:e140.

53.

Seabrook

, Kern

, Fulcher

, et al. Predicting depression from language-based emotion dynamics: longitudinal analysis of Facebook and Twitter status updates. Journal of Medical Internet Research, 2018; 20:e168.

54.

Yin

, Fabbri

, Rosenbloom

, et al. A scalable framework to detect personal health mentions on Twitter. Journal of Medical Internet Research, 2015; 17:e138.

55.

Guntuku

, Yaden

, Kern

, et al. Detecting depression and mental illness on social media: an integrative review. Current Opinion in Behavioral Sciences, 2017; 18:43–49.

56.

Hur

, Gupta

. Growing up in the web of social networking: adolescent development and social media. Adolescent Psychiatry (Hilversum), 2013; 3:233–244.

57.

Shapiro

LAS

, Margolin

. Growing up wired: social networking sites and adolescent psychosocial development. Clinical Child and Family Psychology Review, 2014; 17:1–18.

58.

Grasmuck

, Martin

, Zhao

. Ethno-racial identity displays on Facebook. Journal of Computer-Mediated Communication, 2009; 15:158–188.

59.

Hillier

, Harrison

. Building realities less limited than their own: young people practising same-sex attraction on the internet. Sexualities, 2007; 10:82–100.

60.

Larson

, Wilson

, Brown

, et al. Changes in adolescents' interpersonal experiences: are they being prepared for adult relationships in the twenty-first century?. Journal of Adolescent Research, 2002; 12:31–68.

61.

Tynes

, Giang

, Thompson

. Ethnic identity, intergroup contact, and outgroup orientation among diverse groups of adolescents on the Internet. CyberPsychology and Behavior, 2008; 11:459–465.

62.

Nesi

, Choukas-Bradley

, Prinstein

. Transformation of adolescent peer relations in the social media context: part 1—A theoretical framework and application to dyadic peer relationships. Clinical Child and Family Psychology Review, 2018; 21:267–294.

63.

Walther

. Computer-mediated communication: impersonal, interpersonal, and hyperpersonal interaction. Communication Research, 1996; 23:3–43.

64.

Walther

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

65.

Walther

. Theories of computer-mediated communication and interpersonal relations. The Handbook of Interpersonal Communication, 2011; 4:443–479.

66.

Blease

. Too many “friends,” too few “likes”? Evolutionary psychology and “Facebook depression.”. Review of General Psychology, 2015; 19:1–13.

67.

Tong

, Van Der Heide

, Langwell

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

68.

Granovetter

. The strength of weak ties. American Journal of Sociology, 1973; 78:1360–1380.

69.

Markstrom

. Identity formation of American Indian adolescents: local, national, and global considerations. Journal of Adolescent Research, 2011; 21:519–535.

70.

Trepte

, Dienlin

, Reinecke

. Influence of social support received in online and offline contexts on satisfaction with social support and satisfaction with life: a longitudinal study. Media Psychology, 2015; 18:74–105.

71.

Valkenburg

, Peter

. Preadolescents' and adolescents' online communication and their closeness to friends. Developmental Psychology, 2007; 43:267.

72.

Valkenburg

, Sumter

, Peter

. Gender differences in online and offline self-disclosure in pre-adolescence and adolescence. British Journal of Developmental Psychology, 2011; 29:253–269.

73.

Carli

, Durkee

, Wasserman

, et al. The association between pathological internet use and comorbid psychopathology: a systematic review. Psychopathology, 2013; 46:1–13.

74.

Kraut

, Patterson

, Lundmark

, et al. Internet paradox: a social technology that reduces social involvement and psychological well-being?. American Psychologist, 1998; 53:1017.

75.

Widyanto

, Griffiths

. “Internet addiction”: a critical review. International Journal of Mental Health and Addiction, 2006; 4:31–51.

76.

Edlund

, Lange

, Sevene

, et al. Participant crosstalk: issues when using the mechanical Turk. Tutorials in Quantitative Methods for Psychology, 2017; 13:174–182.

77.

Hydock

. Assessing and overcoming participant dishonesty in online data collection. Behavior Research Methods, 2018; 50:1563–1567.

78.

Thomas

, Clifford

. Validity and Mechanical Turk: an assessment of exclusion methods and interactive experiments. Computers in Human Behavior, 2017; 77:184–197.

79.

Necka

, Cacioppo

, Norman

, et al. Measuring the prevalence of problematic respondent behaviors among MTurk, campus, and community participants. PLoS One, 2016; 11:e0157732.