Patterns of Social Media Use and Their Relationship to Health Risks Among Young Adults

Abstract

Purpose:

Social media use is pervasive among young adults, and different sites have different purposes, features, and audiences. This study identified classes of young adults based on what combination of sites they use and how frequently, and compared their health risk factors and behaviors.

Methods:

Latent profile models were developed based on frequency of using 10 sites from a national sample of young adults aged 18–24 years (n = 1,062). Bivariate analyses and multivariable regressions examined the relationship between class membership and alcohol, tobacco, and other drug (ATOD) use, and symptoms of depression and anxiety.

Results:

The optimal model identified five classes: Low Users (7.9%), High Users (63.1%), Professional Users – high use of LinkedIn (10.1%), Creative Users – high use of Vine and Tumblr (11.5%), and Mainstream Users – high use of Facebook and YouTube (7.4%). Classes differed significantly on ATOD use and depressive symptoms. Compared to High Users, Creative Users had higher odds of using most substances and lower odds of depressive symptoms, Mainstream Users had higher odds of substances used socially (alcohol and hookah), Professional Users had higher odds of using alcohol, cigarettes, and cigars, and Low Users had higher odds of using other drugs (e.g., cocaine and heroin).

Conclusions:

A young adult’s social media site use profile is associated with ATOD use and depressive symptoms. Use and co-use of certain sites may influence the volume and nature of ATOD-related content and norms young adults experience in social media. Targeting interventions to sites selected based on use patterns associated with each health risk may be effective.

Young adults are heavy users of social media (i.e., social networking sites, including Facebook, Twitter, YouTube, Twitter, Instagram, Snapchat, and Tumblr). In total, 88% of 18–29 year olds report using social media (compared to 78%–37% of older age groups) [1], and young adults spend more time (averaging over 3 hours daily) on social media than older adults [2].

Research indicates use of social media is related to health-related risk factors and behaviors among young adults, including anxiety and depression [3], sexual behaviors [4], and use of alcohol and other substances [5,6]. While many studies have not directly explored mechanisms by which social media use might increase health risks, it is possible that peer interactions on social media – including displays or positive reinforcement of risk behavior – play a key role in this process, and there is some evidence to support this [5–7].

Peer interactivity is heightened in social media compared to other media channels due to unique characteristics like the ability to create, share, critique, and endorse (e.g., likes, retweets, and comments) content [8].This heightened peer interactivity may be a powerful source of influence for young adults, who are sensitive to peer feedback as a source of understanding their identity during a time of life transition in which new identities and peer group affiliations are established [9,10]. In particular, the role of peer influence in young adults’ substance use risk behaviors has been well established [11,12].

Notably, different social media platforms have different purposes, features, content, audiences, and norms. While Facebook is typically used to share life events with family and friends, Twitter facilitates “water cooler” conversations about common interests on a global scale. Many social media sites have carved out unique identities: LinkedIn is for professional networking; Pinterest is for sharing craft, cooking, and other ideas; Tumblr is a social blogging platform where users share and discuss multimedia content; and Snapchat is a messaging application to send photos and videos that disappear once viewed. The demographics of these sites’ users also vary – prior research suggests Snapchat, Vine, and Tumblr user bases are younger; Pinterest users are predominantly women; more Blacks and Hispanics use Instagram than Whites; and LinkedIn users have higher education and income [13,14]. Given this variation, the social media platforms one uses and how often one engages with them likely influences both content exposure and nature of peer interactivity, which in turn can influence health-related norms and behaviors.

Much of the extant research examining social media use has focused on the frequency and duration of overall use, and of specific sites [1]. Recent studies using data from the Pew Research Center and the Truth Initiative Young Adult Cohort have found high use (88%–98%) of social media among young adults [1,15], with 18–24 year olds substantially more likely than older adults to use platforms like Snapchat and Instagram [1]. These studies have also found significant overlap in use of multiple sites, with young adults using 4–8 on average [1,15].

However, to our knowledge, no studies to date have created a nuanced profile of young adults’ social media use that captures use and co-use of different sites along with frequency of use. These gaps in available data on young adults’ social media use, including use of specific sites, may impede the ability of public health efforts to effectively counter health risk-promoting information, both from peers and industry interests [16].

The primary aim of this study is to aid in addressing this gap in the literature by identifying classes of young adults based on their social media use patterns (i.e., what combination of sites they use and how frequently) and assessing health risk differences between classes.

Methods

Data source

This study used data from the Truth Initiative Young Adult Cohort Study, designed to understand tobacco use among U.S. young adults, 18–34 years old at study entry, using a longitudinal cohort. The baseline survey was fielded in 2011, with subsequent surveys fielded at approximately 6-month intervals. This study used data from only Wave 9 of the survey, collected in February and March 2016, as it was the most recent wave with social media use questions available at the time of analysis. The sample was restricted to 18–24 year olds, given that these younger adults are more likely to be attending college and differ in income, employment status, smoking behavior, and social media use compared to those 25 and older [1,17]. The final sample size was 1,062 after 71 cases were dropped due to missing data on sociodemographics or variables included in the Latent Profile Analysis.

The Truth Initiative Young Adult Cohort is composed of a national sample of young adults, weighted to be nationally representative. The sample is drawn from GfK’s online KnowledgePanel, which is recruited via address-based sampling to provide statistically valid representation of the U.S. population, including cell-phone-only households. Detailed recruitment methodology is described elsewhere [18].

The household recruitment rate for the Wave 9 survey, administered online, was 13.2%. In 63.9% of these households, one member completed a survey providing key demographic information. While multiple household members could be a part of the panel, only one panel member per household was selected at random to be part of this particular study, and no members outside the panel were recruited. The Wave 9 response rate was 60.7%, for a cumulative response rate of 5.1%. The Wave 9 Truth Initiative Young Adult Cohort Study was approved by Chesapeake Investigational Review Board, Inc., and consent collected from participants before survey self-administration.

Measures

Social media use.

Social media use was assessed by asking respondents about frequency of use of ten popular social media sites (Twitter, Facebook, Instagram, Snapchat, Vine, Pinterest, Tumblr, Google+, LinkedIn, and YouTube), with response options being: multiple times per day, daily, weekly, monthly, less than one time per month, or never. Although Vine was shut down in January 2017, it was used widely at the time of this survey.

Sociodemographics.

Sociodemographic variables included in the analysis were gender, age, education, race, and a validated measure of subjective financial status (see Online Appendix A for details).

Health risk factors and behaviors.

Past 30-day alcohol, marijuana, and other drug (cocaine, heroin, ecstasy, meth, etc.) use were each measured by two items. For marijuana and other drug use, the first item asked about the frequency of use currently (every day, some days, or not at all), while for alcohol use, the first item asked about the frequency of use in the past year (from never to four or more times a week). Those who reported any use were then asked to enter on how many of the past 30 days they used the product, with those responding ≥1 day defined as past 30-day users.

In this wave of the survey, respondents were randomized to receive different tobacco use questions. For ever use of each product and separately for past 30-day use of each product, respondents received one or both of two items. The first item asked respondents to “select all that apply” with respect to their use of tobacco products (cigarettes, cigars, little cigars/cigarillos/bidis, and hookah/shisha/waterpipe; the item also included other noncombustible products not used in this analysis), with a follow up question asking them to enter on how many of the past 30 days they used the products selected. The second item asked if they had ever tried each tobacco product (yes/no) and then asked on how many days of the past 30 they smoked the product (response options ranged from 0 days to 25–30 days). A positive response to either of the two items was coded as ever or past 30-day use for cigarettes, cigars (coded as having used either cigars or little cigars/cigarillos/bidis), and hookah. Anyone who used any of these products, ever or in the past 30 days, was coded as having ever used combustible tobacco or used combustible tobacco in the past 30 days, respectively.

Variables capturing symptoms of depression and anxiety were created based on a score at or above the cut-off on the two-item Patient Health Questionnaire, which asks about the frequency of depressed mood and loss of interest in activities, and the two-item Generalized Anxiety Disorder scale, which asks about the frequency of uncontrollable worry and feelings of anxiety, respectively [19,20].

Analysis

Latent profile analysis (LPA) was conducted using Mplus 7.0 to identify mutually exclusive classes of young adults based on social media use patterns [21]. For this LPA, 10 continuous variables were entered into the model, representing the frequency of use for each of 10 social media sites from 1 (never) to 6 (multiple times a day).

The optimal number of classes was determined by running models with a successive number of classes from 1 to 8 and comparing model fit indices, entropy, and interpretability. The following model fit indices were assessed: the log likelihood (−2LL), the Akaike Information Criterion, the Bayesian Information Criterion (BIC), the sample size adjusted BIC (SSABIC), and the Lo–Mendell–Rubin and Vuong–Lo–Mendell–Rubin tests (for which significant p-values indicate that the current model is a significantly better fit for the data than a model with one fewer class) [22,23]. According to guidelines for recommended fit indices, the optimal class solution should have lowest BIC and SSABIC values [22], significant Lo–Mendell–Rubin and Vuong–Lo–Mendell–Rubin tests, and entropy values close to 1.0, demonstrating high classification precision [24]. With continuously declining BIC/SSABIC values, the recommendation is to select the class solution with minimal BIC/SSABIC differences compared to the subsequent class solution [25].

After the best-fitting LPA model was selected, class membership data were exported from Mplus for each participant and merged with the full data set in Stata IC 13. To increase interpretability, each class was given a name based on response patterns to the social media site use questions. Poststratification weights were used to account for complex sample design and nonresponse bias. Weighted bivariate analyses were conducted to estimate average frequency of use of each site, sociodemographic characteristics, and health risks among classes. Weighted chi-square tests were used to determine differences in characteristics across latent class assignment, and weighted multivariable regressions were conducted to determine the association between latent class assignment and each of the health risks, adjusting for sociodemographics.

Additionally, sensitivity analyses were conducted to see if and how the number and composition of classes differed when (1) variables capturing access to digital devices with internet access (see Online Appendix A for details) were included in the LPA model, and separately, when (2) Vine was excluded from the model (since Vine does not exist currently).

Results

Selection of the latent profile model

Model fit indices for all latent profile models are presented in Table 1. All models exhibited entropy values close to 1.0, demonstrating high classification precision [24]. Moving to models with more classes, BIC/SSABIC values declined continuously. Thus, the optimal model – the five-class solution – was selected based on the class solution with minimal BIC/SSABIC differences from the subsequent class solution, with a significant Vuong–Lo–Mendell–Rubin test, and optimal interpretability (the extent to which an additional class provided unique information).

Table 1.

Model fit indices for 1–8 class solutions of social media use patterns

Model	LL	AIC	BIC	SSABIC	LMR	VLMR	Entropy
1-class solution	−19776.677	39593.355	39692.713	39629.19	–	–	–
2-class solution	−18648.115	37358.231	37512.236	37413.775	p < .0001	p < .0001	.987
3-class solutiona	−18071.659	36227.318	36435.971	36302.571	p < .0001	p < .0001	1.000
4-class solution	−17771.92	35649.840	35913.139	35744.802	.0021	.002	.988
5-class solution b	−17507.287	35142.574	35460.521	35257.246	.0005	.0005	.996
6-class solutiona,c	−17234.247	34618.493	34991.087	34752.874	.225	.2248	.998
7-class solutiona	−16900.169	33972.337	34399.578	34126.427	.394	.3909	.937
8-class solutiona,c	−16772.473	33738.945	34220.832	33912.744	.0001	.0001	.959

AIC = Akaike Information Criterion; BIC = Bayesian Information Criterion; LL = Log likelihood; LMR = Lo–Mendell–Rubin; SSABIC = Sample size adjusted Bayesian Information Criterion; VLMR = Vuong–Lo–Mendell–Rubin.

^aModel had error indicating model nonidentification.

^bThe best fit solution is indicated in bold. The 5-class solution was deemed to have the best fit based on minimal BIC/SSABIC differences from the subsequent class solution; a significant Vuong-Lo-Mendell-Rubin test; and optimal interpretability (the extent to which an additional class provided unique information).

^cModel had error indicating the best loglikelihood value was not replicated.

Identification of social media use classes

Table 2 provides the weighted average – on a scale from 1 (never) to 6 (multiple times per day) – of social media site use for each of the 10 sites included in the model by social media use class as well as unweighted class size [15,26]. The latent profile model revealed five distinct, mutually exclusive classes of social media site use. Names were selected for each class to capture their pattern of site use (although the terms low and high are used, these are not ordinal groups): (1) Low Users –lower use of all social media sites compared to the full sample (Class 1, 7.91%); (2) High Users –higher use of all sites compared to the full sample (Class 2, 63.09%); (3) Professional Users –very high use of LinkedIn, a site primarily used for professional reasons, and extremely low use of Vine (Class 3, 10.08%); (4) Creative Users –very high use of Vine and comparatively high use of Tumblr, sites that typically involve user generation of multimedia content, and extremely low use of LinkedIn (Class 4, 11.49%); and (5) Mainstream Users –high use of Facebook and comparatively high use of YouTube, the most widely used sites among 18–24 year olds [1], and average use of other sites (Class 5, 7.44%).

Table 2.

Average use of social media sites (scale from 1 (never) to 6 (multiple times a day)) by social media use class and in the full sample (n = 1062), weighted

	Class 1 – Low Usersa (Meanb, SE)	Class 2 – High Usersa (Meanb, SE)	Class 3 – Professional Usersa (Meanb, SE)	Class 4 – Creative Usersa (Meanb, SE)	Class 5 – Mainstream Usersa (Meanb, SE)	Full sample (Meanb, SE)
Twitter	3.36 (.175)c	5.16 (.066)d	4.14 (.201)c	4.69 (.194)	4.41 (.196)c	4.84 (.058)
Facebook	3.93 (.134)c	4.36 (.065)	4.24 (.138)	4.04 (.140)	4.64 (.094)d	4.31 (.050)
Instagram	3.41 (.150)c	5.00 (.062)	3.87 (.175)c	4.62 (.170)	4.55 (.130)	4.72 (.053)
Snapchat	3.26 (.174)c	5.16 (.056)d	3.87 (.182)c	4.52 (.171)	4.46 (.154)c	4.80 (.054)
Vine	2.03 (.125)c	5.94 (.014)d	1.80 (.090)c	5.95 (.027)d	5.68 (.097)d	5.25 (.056)
Pinterest	2.81 (.234)c	5.21 (.059)d	3.61 (.235)c	4.15 (.213)c	4.56 (.201)	4.75 (.060)
Tumblr	3.14 (.257)c	5.61 (.048)d	4.59 (.213)c	5.46 (.138)	5.04 (.174)	5.29 (.051)
Google+e	2.57 (.212)c	5.29 (.064)d	4.98 (.182)	4.64 (.225)c	4.82 (.221)	4.96 (.062)
LinkedIn	1.75 (.094)c	5.98 (.005)d	5.94 (.029)d	1.47 (.053)c	3.33 (.064)c	5.10 (.058)
YouTube	3.06 (.155)c	3.90 (.063)	3.71 (.136)	3.05 (.128)c	3.74 (.109)	3.73 (.049)
Class size, unweighted (% of full sample)	7.91%	63.09%	10.08%	11.49%	7.44%
Class size, unweighted (n)	84	670	107	122	79	1,062

^aNames were selected for each class based on response patterns to social media site use questions. Low Users were .3 below the sample average across all sites; High Users were .3 above the sample average across most sites; Professional Users were .3 above the sample average only for LinkedIn (and .3 below the sample average for most other sites); Creative Users were .3 above the sample average for Vine (and had the second highest Tumblr use, after the High Users group); and Mainstream Users were .3 above the sample average for Facebook (and had the second highest YouTube use, after the High Users group).

^bWeighted average (through use of poststratification weights to account for complex sample design and nonresponse bias); scale from 1 (never) to 6 (multiple times per day).

^c>.3 below full sample average, indicating members of the class use a particular site substantially less frequently than the full sample average.

^d>.3 above full sample average, indicating members of the class use a particular site substantially more frequently than the full sample average.

^eHigh reported use of Google+ likely reflects use of Google’s search engine given the failure of the Google+ social networking site [16,28].

Adding variables capturing access to digital devices with internet access to the LPA in sensitivity analyses did not alter classes, and excluding Vine from the model changed the size and composition, but not number, of classes only slightly. While there were still classes of Low Users, High Users (9 percentage points smaller compared to the LPA with Vine), and Creative Users (6 percentage points larger compared to the LPA with Vine), there were two classes of professional users, a Professional Creative Users class with high use of both LinkedIn and Tumblr and a Professional Mainstream Users class with high use of LinkedIn along with Facebook and YouTube, and very low use of Tumblr (see Online Appendix B for detailed results).

Correlates of social media use class

Bivariate analyses of health risks by social media use class are presented in Table 3 (see Online Appendix C for bivariate analyses of sociodemographic characteristics by social media use class).

Table 3.

Health risk characteristics of the five social media use classes and in the full sample (n = 1062), weighted

	Class 1 – Low Users	Class 2 – High Users	Class 3 – Professional Users	Class 4 – Creative Users	Class 5 – Mainstream Users	Full sample	Chi-squared test p valuea
Drug and alcohol use
Past 30-day marijuana use	14.10%	11.33%	13.25%	15.58%	14.56%	12.31%	.7797
Past 30-day other drug use (cocaine, heroin, ecstasy, meth, etc.)	5.25%	.69%	.72%	4.04%	2.83%	1.46%	.0458
Past 30-day alcohol use	55.10%	42.21%	56.48%	65.85%	61.01%	47.85%	< .0001
Tobacco use
Ever combustible tobacco use	44.80%	42.35%	55.57%	52.64%	50.47%	45.26%	.0872
Past 30-day combustible tobacco use	22.97%	32.69%	46.16%	45.86%	39.05%	34.93%	.0034
Ever cigarette use	36.65%	35.34%	42.47%	38.61%	32.79%	36.26%	.7209
Past 30-day cigarette use	19.65%	25.49%	30.25%	30.41%	25.65%	26.00%	.5437
Ever cigar use	24.14%	23.18%	32.23%	33.57%	24.49%	25.16%	.1768
Past 30-day cigar use	7.97%	16.17%	26.97%	28.65%	18.58%	17.94%	.0011
Ever hookah use	28.16%	19.34%	27.14%	35.68%	31.80%	23.00%	.0033
Past 30-day hookah use	11.17%	12.90%	19.12%	28.68%	24.79%	15.58%	.0001
Depression/anxiety symptoms
Depression (MDD)	6.30%	6.98%	5.63%	1.48%	2.18%	6.00%	.143
Anxiety(GAD)	12.70%	7.34%	10.32%	3.81%	2.18%	7.35%	.0669

GAD = Generalized Anxiety Disorder; MDD = Major Depressive Disorder.

Statistically significant (p < .05) p values are given in bold.

^aWeighted chi-squared tests were used to determine differences in each health risk characteristic across social media use class assignment. Poststratification weights were used to account for complex sample design and nonresponse bias.

There were significant differences in ever and past 30-day substance use between social media use classes. A greater proportion of Creative Users used other drugs (4.04% in contrast to .69% of High Users), alcohol (65.85%), hookah (35.68% ever use and 28.68% in the past 30 days), cigars (28.65% in the past 30 days – over three times that of Low Users), and combustible tobacco (45.86% in the past 30 days). A greater proportion of Mainstream Users used alcohol (61.01%) and hookah (31.80% ever use and 24.79% in the past 30 days – over twice that of Low users); a greater proportion of Professional Users used cigars (26.97% in the past 30 days – over three times that of Low Users) and combustible tobacco (46.16% in the past 30 days); and a greater proportion of Low Users used other drugs (5.25% in contrast to .69% of High Users).

Multivariable analyses of health risks by social media use class

Thirteen multivariable regression analyses, adjusting for socio-demographics, were conducted to assess the association between a categorical variable capturing each social media use class and dichotomous variables capturing each health risk. Results are shown in Table 4, with High Users (the largest class) set as the reference group.

Table 4.

Multivariable logistic regressions of health risks by social media use class, weighted

	Past 30-day marijuana use OR [CI]a	Past 30-day other drug use OR [CI]a	Past 30-day alcohol use OR [CI]a	Ever use combustible tobacco OR [CI]a	Past 30-day use combustible tobacco OR [CI]a	Symptoms of depression OR [CI]a	Symptoms of anxiety OR [CI]a
Low versus High	1.51 [.63, 3.63]	9.54 [1.58, 57.81]b	1.61 [.87, 2.99]	1.21 [.69, 2.14]	.61 [.36, 1.06]	.88 [.31, 2.47]	1.84 [.73, 4.64]
Professional versus High	1.31 [.66, 2.60)	1.61 [.18, 14.31]	1.82 [1.16, 2.87]b	1.99 [1.26, 3.16]c	1.90 [1.17, 3.08]c	.91 [.31, 2.72]	1.52 [.70, 3.29]
Creative versus High	1.70 [.83, 3.48]	10.05 [1.82, 55.47]c	2.00 [1.21, 330]c	1.78 [1.11, 2.87]b	1.68 [1.04, 2.70]b	.24 [.07, .84]b	.56 [.22,1.43]
Mainstream versus High	1.53 [.69, 3.37]	7.13 [.67, 75.57]	1.86 [1.09, 3.18]b	1.62 [.93, 2.83]	1.32 [.75, 2.31]	.36 [.08, 1.56]	.31 [.07, 1.37]
	Ever use cigarettes OR [CI]a	Past 30-day use cigarettes OR [CI]a	Ever use cigars OR [CI]a	Past 30-day use cigars OR [CI]a	Ever use hookah OR [CI]a	Past 30-day use hookah OR [CI]a
Low versus High	1.23 [.66, 2.29]	.77 [.42, 1.39]	1.16 [.56, 2.40]	.45 [.21, .95]	1.55 [.79, 3.02]	.71 [.37, 1.37]
Professional versus High	1.62 [1.01, 2.61]b	1.43 [.83, 2.48]	1.75 [1.02, 3.01]b	2.03 [1.13, 3.65]b	1.63 [.94, 2.83]	1.60 [.92, 2.78]
Creative versus High	1.49 [.91, 2.42]	1.43 [.85, 2.40]	1.83 [1.09, 3.06]b	1.92 [1.12, 3.29]b	2.10 [1.28, 3.44]c	2.09 [1.23, 3.54]c
Mainstream versus High	1.11 [.60, 2.06]	1.14 [.59, 2.21]	1.14 [.62, 2.10]	1.11 [.60, 2.09]	1.87 [1.07, 3.27]b	1.88 [1.02, 3.44]b

^aOdds ratio [confidence interval] determined via multivariable logistic regressions assessing relationship between each health risk and a categorical variable capturing social media use class, adjusting for sociodemographics (gender, age, education, race, and a validated measure of subjective financial status). Thirteen regression models were run (one for each health risk) with the independent variable being the categorical variable capturing social media use class, thus providing comparisons of health risks for each social media use class relative to the reference group of High Users (the largest group). All models were weighted through use of poststratification weights to account for complex sample design and nonresponse bias.

^bp < .05.

^cp < 01.

^dp < .001.

Compared to High Users, Creative Users had 10 times the odds of using other drugs; twice the odds of using alcohol, cigars (ever and past 30 days), and hookah (ever and past 30 days); 70%–80% higher odds of using combustible tobacco (ever and past 30 days); and 75% lower odds of depressive symptoms. Compared to High Users, Mainstream Users had twice the odds of using alcohol and hookah (ever and past 30 days). Compared to High Users, Professional Users had twice the odds of using alcohol, cigars, or combustible tobacco in the past 30 days, and 60%–75% higher odds of ever using cigars. Compared to High Users, Low Users had 10 times the odds of other drug use. There were no significant differences between High Users and other social media use classes for marijuana use, anxiety symptoms, or past 30-day cigarette use.

Discussion

Key findings

Overall, this analysis finds that young adults fall into distinct classes of social media use patterns, and these patterns are associated with differences in health-related risk factors and behaviors. Formation of two of the social media use classes was determined by overall frequency of use across sites. Formation of the other three classes was driven by high and/or low frequency use of particular sites, providing insight into the primary reasons these groups use social media (creation and sharing of multimedia content vs. professional networking vs. use of general sites like Facebook and YouTube). These classes remained relatively stable even in sensitivity analyses adding digital device access variables to or excluding Vine from the model, suggesting these classes may be useful even as specific devices and social media sites come and go, assuming the primary reasons users turn to sites remains the same.

Adjusted regressions assessing the relationship between social media use classes and health risks indicated that compared to High Users, Creative Users were more likely to use most substances (but less likely to have depressive symptoms), Mainstream Users were more likely to consume alcohol and hookah, substances young adults use socially in commercial establishments like bars and lounges [27,28], Professional Users were more likely to use alcohol, cigarettes, and cigars, and Low Users were more likely to use hard drugs (cocaine, heroin, etc.). While it is not possible to determine the causal ordering of these associations, it is possible that the kinds of young adults who exhibit certain social media use profiles are those that use substances more than their peers due to uncaptured confounders (e.g., a stressful home environment may influence both low use of social media and higher use of hard drugs). Alternatively, it is possible that higher exposure to pro-tobacco/alcohol/drug content or norms on sites frequented by a particular social media use class increases their use of that substance. For example, anonymity on Tumblr may contribute to the site’s high degree of disclosure and sharing around sensitive topics [15,29], so perhaps Creative Users are exposed to more content depicting substance use; Facebook is one of the social media sites often used to share pictures of social events, which may contribute to social use of products like alcohol and hookah being more accepted.

Limitations and strengths

Key limitations of this study include the challenging nature of measuring social media site use. The use of self-reported survey data may paint an incomplete picture of social media use due to an incomplete list of sites or misinterpretation of sites listed, and is subject to recall bias, especially when use of sites blends together, as with watching a YouTube video on Facebook. Additionally, the dataset used in this analysis does not capture user perception of different sites or the kind of health and other information these individuals receive via these sites. It is possible there is substantial overlap in the content an individual receives across platforms due to the accounts they choose to follow and the way they engage with these sites, but the measures in this analysis do not capture such information. Site use also likely changes over time with changing life circumstances (entering college, leaving a job, becoming a parent, etc.) and preferences, so each individual’s social media site use in this cross-sectional analysis represents only a snapshot in time. Despite these limitations, prior research has found self-reports of social network site use can be accurate [30], and self-report requires fewer resources and demands on participant privacy compared to tracking use on devices [31].

Social media site use was reported on a scale from 1 (never) to 6 (multiple times per day), and it was treated as a continuous variable in latent profile analyses to facilitate interpretation. While similar scale variables have been treated as continuous in other published latent profile analyses [32], it is important to note that this approach assumes the distance between each category is the same.

Although the study sample’s completion and cumulative response rates are low, they are similar to other studies that have relied on KnowledgePanel [33,34], and there is little indication of nonresponse bias on core variables in other KnowledgePanel samples or in this sample specifically [35,36]. Other limitations include that individuals received different tobacco use measures, cross-sectional analyses preclude understanding causality, important confounders may have been missed, and type I error (finding a significant relationship when none exists) may be increased due to the use of multiple comparisons without applying a correction. A correction was not employed in this analysis due to the exploratory nature of this research and desire to minimize type II errors (deeming important relationships nonsignificant) [37,38].

The primary strength and novel contribution of this study is the use of a large, nationally representative sample of young adults to form a more detailed and nuanced picture of social media use than currently exists in the literature. By examining how social media use across sites forms distinct use patterns, this study goes beyond typical approaches of examining use individually by site [1], without attention to patterns of co-use across sites. The use of sensitivity analyses and weighted multivariable regression analyses provide a clearer picture of the association between social media use patterns and health risks.

Contribution and implications

Although some prior studies did not find a relationship between overall social media use and risk behaviors such as alcohol and tobacco use, these studies did not explore the full range of sites used by participants [4,39]. One previous study conducted with the same dataset used in this analysis demonstrated that the number of social media sites used by young adults was associated with alcohol use but not marijuana use or symptoms of depression or anxiety [15]. However, their analysis did not explore what specific combinations of sites might be associated with risk behaviors.

The analysis presented in this paper is the first to our knowledge to explore how patterns of social media site use are associated with health risks and suggests that patterns of use of specific sites may be more important than overall frequency of use of social media – or at least an additional factor to consider – in understanding the relationship between social media use and health risks. A more nuanced understanding of these social media use patterns may enable public health professionals to disseminate effective risk-mitigating and preventive information to young adults through targeted social media platforms. Social media sites vary in content produced, norms endorsed, and audiences reached, and exposure to media content depicting risky health behaviors has been shown to influence actual risk behaviors [7,39]. Thus, young adults’ pattern of social media site use may influence the volume and nature of their exposure to health risk promoting content – as well as norms surrounding such content – which in turn may influence their health risk behaviors.

Future research should explore the precise nature of health risk promoting content and norms youth are exposed to on different social media sites, and how this exposure may vary based on nuanced site use patterns. More research is also needed to understand how the negative effects of health risk promoting content and norms on social media can be mitigated or countered, and whether strategies applied offline such as warning labels or notices about harmful products or sensitive content could be effective in a social media environment. Targeting normative interventions or health campaigns through a combination of sites selected based on the use patterns associated with each of these risk behaviors may be an effective strategy to reach the young adults most likely to engage in these behaviors. Further research is needed to determine whether and what preventive health information via social media would be most influential [4,40].

Supplementary Material

Supplementary data

Supplementary data related to this article can be found at doi:10.1016/j.jadohealth.2018.06.025.

IMPLICATIONS AND CONTRIBUTION.

This study identified classes of young adults based on use of different social media sites, and found these classes differed significantly in their alcohol, tobacco, and other drug use, as well as depressive symptoms. Targeting interventions to sites selected based on use patterns associated with each health risk may be effective.