Adolescent media use, parent involvement and health outcomes: a latent class analysis approach

ABSTRACT

Media use among early adolescents is nearly ubiquitous and has been associated with important health outcomes such as physical activity, sleep and problematic internet use (PIU). Parent involvement has been recommended as a prevention strategy; it remains unclear how it is associated with media use and health outcomes. The purpose of this study was to develop profiles of media use, parent involvement and health outcomes among adolescents. Early adolescents were recruited to a cross-sectional online survey using the Qualtrics platform and panels. Media use measures included ownership and bedroom use of devices, social media platforms and video games. Parent media involvement assessed media rules and role-modeling. Health measures included physical activity, sleep and PIU. We used latent class analysis (LCA) to identify distinct profile groups across these three areas. The 1155 participants had a mean age of 13.6 years (SD = 1.1), of whom 49.7% were female, 73.7% were White and 61.1% had parent education with a college degree. We found that most participants owned personal media devices, including smartphones (81.4%), computers (64.6%) and video game systems (58.9%). The LCA identified three distinct profile groups: (1) Active Autonomous Media Users, (2) Young Low-Tech Sleepers and (3) Risky Regulated Media Users. Findings support that media use patterns vary across adolescents, suggesting different education and prevention approaches may be needed. Targeting educational messages to different media profiles may be an effective strategy to optimize productive media use and health.

KEYWORDS:

• Adolescent
• social media
• latent class analysis
• physical activity
• sleep
• problematic internet use

Introduction

Adolescents today are often referred to as ‘digital natives’ given they are growing up in an immersive technological society. The majority of adolescents have a personal smartphone and engage with digital media; approximately 45% of adolescents describe that they are online ‘almost constantly’ (Anderson & Jiang, 2018a, 2018b). Many studies have documented teens’ frequent and near-ubiquitous digital media use as a group (Duggan et al., 2014; Duggan & Brenner, 2013; Lenhart, 2015). Adolescents represent an important population in which to study technology and health links given their nearly universal use and access to technology, their unique developmental stage of emerging independence, and their critical stage of cognitive and brain development (Casey et al., 2008; Giedd, 2008; Spear, 2000).

Benefits and risks to media use

Adolescents’ frequent and consistent digital media use has both benefits and risks. Benefits include opportunities for content creation and social support (Ellison et al., 2007). Previous studies have shown that digital media offers opportunities to provide novel interventions for patients for asthma (Nickels & Dimov, 2012), to advance sexual health (Bull et al., 2012) and to provide new approaches for mental health support (Naslund et al., 2016). Another potential benefit is the capacity for digital media to enhance patient education and access (Atkinson & Gold, 2002; Wong et al., 2014).

Risks include negative health consequences. First, digital media use can negatively affect sleep by delaying bedtime as adolescents engage with digital media, or by repeated awakenings from phone notifications (Levenson et al., 2016). Screen-based media use can also disrupt sleep through exposure to light from screens disrupting melatonin levels (Wahnschaffe et al., 2013). Second, decreased physical activity has been associated with the sedentary nature of most media use (de Jong et al., 2013; Goris et al., 2010). Media use has also been attributed to increased obesity risk as adolescents often increase caloric intake while watching media (Blass et al., 2006). Third, problematic internet use (PIU) is defined as ‘Internet use that is risky, excessive or impulsive in nature leading to adverse life consequences, specifically physical, emotional, social or functional impairment’ (Moreno et al., 2013). Studies support that components of PIU include compulsive use and anxiety when not able to access the internet (Jelenchick et al., 2016).

One approach to prevent these adverse health outcomes is parent involvement through establishing media rules and role-modeling appropriate use. Previous studies illustrate parents’ challenges in enforcing rules, such as removing technology from children’s bedrooms (Buxton et al., 2015; Dinleyici et al., 2016).

Media use resources

The American Academy of Pediatrics (AAP) policy statement: Media Use Among School-aged Children and Adolescents was released in 2016 (Council on Communications and Media, 2016). This policy statement provided guidance for parents to ensure that adolescents get the recommended amount of sleep and physical activity daily. The policy statement recommended that parents establish a Family Media Use Plan (FMUP) to clearly state the rules/guidelines for media use and enforce those for both parents and children. Examples of rules/guidelines from the FMUP include setting media-free times in the home, such as dinnertime, and engaging in co-viewing media as a family. The FMUP is intended to be personalized for each child in the family, emphasizing that different approaches may be needed with each individual.

Challenges in understanding media use across populations

The growing awareness that media use may vary within populations is further supported by large population studies of adolescents and their media use. In the area of risk for depression and media use, some studies have advocated for a relationship between increased media use and poorer mental health based on small positive associations (Twenge, 2020), some studies have found no significant relationship between risk of depression and media use (Przybylski & Weinstein, 2017). It may be that the variation in media use by individual adolescents leads to no major population effect, supported by a recent research review (Odgers & Jensen, 2020). This area of work supports that studies to examine adolescent media use to identify groups rather than assuming homogeneity within a population may help advance our understanding of the associations between health and media use.

One approach to understand patterns within a study population is using latent class analysis (LCA). LCA provides a person-centered analysis approach to understand how different behaviors or factors may cluster together to reflect distinct categories or groups within a larger heterogeneous population (Hensel, 2020). Previous studies have utilized LCA to understand complex relationships such as between health lifestyle and suicide (Xiao et al., 2019) and understanding patterns of violence victimization and perpetration among teens (Sessarego et al., 2021). While this approach does not overcome limitations of cross-sectional data, such as the inability to assess causal directions, it does allow for the identification of groups with shared characteristics within a single sample of participants.

A further area of consideration in studies of technology is whether types of media, such as social media versus video games, have different influences or effects. There are similarities in proposed gains from these types of media, including connecting to others. Social media allows the creation of a network, while multiplayer video games allow for playing alongside others in real time. Social media use is nearly ubiquitous among adolescents (Anderson & Jiang, 2018a, 2018b). Video games are also incredibly common among teens, with rates of approximately 60% of teens who game regularly. There are also differences in the activities involved in these types of media, as social media can promote personal sharing of experiences while gaming often centers on the game play. There are some blurred lines between these types of media, as social media can be used to access games, and gamers can stream themselves playing via social media (Guskin, 2018). Thus, this study will investigate both types of media.

Study purpose

The emerging understanding that all adolescents’ media use is not the same, and the critical role of parents’ in moderating youth media use, informed this study’s purpose. The purpose of this study was to develop profiles representing early adolescents’ media use, parent involvement, and association with health outcomes using LCA.

Materials and methods

A national Qualtrics cross-sectional online research panel was engaged during the month of June 2017. The Institutional Review Boards at the University of Washington and the University of Wisconsin approved this study.

Study design: survey

We selected an online cross-sectional survey as our data collection method for this LCA. Advantages of online surveys include access to a broad population of individuals, as the data collection is not limited to a specific geographic area (Garton et al., 1999). Another advantage is the capacity to recruit large numbers of participants expediently (Wright, 2005). This is of particular benefit for LCA analyses as this approach benefits from larger sample sizes to support nuanced class analysis. Disadvantages of online survey approaches include self-selection bias, as individuals make choices about what surveys to complete based on their interests and experiences (Thompson et al., 2003). A second consideration is that cross-sectional studies cannot assess the directionality of findings, which we illuminate below as we consider our results and limitations. A third disadvantage is concerns related to online sampling and ensuring that participants are unique and represent their identities. This concern can be mitigated by engaging with an online survey panel, as we describe below.

Setting and participants

Our goal was to achieve a national sample of youth to complete an online survey. Compared to traditional survey approaches, such as in person, phone or mail recruitment, online survey panels offer broader reach and lower costs in data collection (Dillman et al., 2008). We selected the online survey platform Qualtrics for several reasons. First, while online survey platforms do not use weighting, previous studies have shown that online survey approaches using tools such as Qualtrics can achieve demographic attributes that are typically within a 10% range of their corresponding values in the US population (Heen et al., 2014). Second, unlike other platforms such as Mechanical Turk, Qualtrics allows recruitment of youth via approaching parents for consent as a first step. Third, there is a strong and growing literature around the use of Qualtrics to recruit youth samples in the US, including studies of media (Bushman et al., 2012; Len-Rios et al., 2016).

A Qualtrics survey manager recruited adult panel participants between June and August 2017 who indicated they had adolescent children who spoke English. Parents who met these criteria were provided information about the survey and an opportunity to complete informed consent for their child’s participation. Once parent consent was given, the adolescent was provided study information and an opportunity to provide assent. Adolescents who provided assent were allowed to begin the survey.

The target population for this study was 12–15-year-olds who were US residents and English-speaking. While most social media companies require that a youth be at least 13 years old to establish a profile on the platform, many reports suggest that parents often allow their children to bypass these guidelines and create accounts earlier than that age (Beilinson, 2014; Gordon, 2020). Thus, we wanted to include 12-year-olds in this study to capture a full range of early adolescents. Qualtrics recruits via adult panel members who then provide informed consent for their children to participate. Thus, we requested Qualtrics to focus on adult panel members with children in the adolescent age range. We established parameters for Qualtrics to recruit a sample consistent with race/ethnicity representative of the US census population (Heen et al., 2014). In specifying this approach, Qualtrics allows recruitment across all race/ethnicity groups until the proportion for that race/ethnicity matches the US census population. Sample size estimates were calculated using estimates for LCA analysis (Dziak et al., 2014; Park & Yu, 2017), which supports approximately 1:3–1:4 ratios of number of items to number of participants in order to achieve a full range of a potential number of latent classes with a minimum of 0.8 power. As we estimated that approximately 30 items would be included in the final LCA model, our sample size goal was to achieve a sample of 1100 participants.

Survey measures

Media use

Media use measures included device ownership and bedtime access, social media platform ownership, video game use, and parent involvement.

Device ownership and bedroom access: hardware

To assess media device ownership, we modeled questions after those in previous Pew Internet and American life surveys (Anderson & Jiang, 2018a, 2018b; Duggan, 2015; Duggan et al., 2014; Duggan & Brenner, 2013). Given the AAP recommends limiting media device use in bedrooms, we included questions about which devices were allowed to be used in the adolescent’s bedroom. Participants were asked, ‘Which of the following devices do you own? Select all that apply’ and ‘Which of the following do you have access to in your bedroom? Select all that apply’ and given the options to select television (TV), computer, tablet, video games, smartphones with internet access, cell phones without internet access, other and/or none for both questions.

Social media platform ownership and video game use: software

We again modeled questions from Pew Internet and American life surveys (Anderson & Jiang, 2018a, 2018b; Duggan, 2015; Duggan et al., 2014; Duggan & Brenner, 2013). We asked about the use of specific media platforms, including social media and video game engagement. Social media profile ownership was assessed using the question, ‘Which social media profiles do you have? Select all that apply.’ Response options included Instagram, Snapchat, Facebook, Twitter, other, and/or none.

Video game playing included console-based games as well as online single and multiplayer games. Video game playing was assessed with the question, ‘Which games do you play on a regular basis?’ Answer options were chosen among the top 10 games at the time the survey launched, which included answer options of Minecraft, Grand Theft Auto, Call of Duty, Lego, Harry Potter, Jurassic Park, other and/or none.

Parent media involvement: behavior modification

Participants were asked how strongly they agree/disagreed with nine statements related to parent media involvement at home. The statements were modeled after the AAP FMUP’s suggested parent rules and role-modeling. These statements were tested in a previous intervention (Moreno et al., 2016). Example statement included: ‘My parents and I talk about my media use’ and ‘My parents follow screen time rules before bedtime.’ Participants were asked to select from a Likert scale of Strongly Agree, Agree, Not Sure, Disagree, and Strongly Disagree for each of the statements.

Health outcomes

Physical activity

Physical activity was assessed using the validated International Physical Activity Questionnaire short-form version (Craig et al., 2003). Participants were asked to consider physical activities within the last seven days then answer yes or no if they had engaged in any of the four following activities: vigorous physical activity, moderate activity, walking and sitting. For each item, participants responded yes to they were then asked to indicate how many days of the week they had engaged in that activity and average minutes per day. For this study, we focused on reported vigorous physical activity.

Sleep

Sleep was assessed using the validated Cleveland Adolescent Sleepiness Questionnaire (CASQ) (Spilsbury et al., 2007). The Chronbach alpha for this scale was 0.89. Participants were prompted to answer 16 questions related to sleep using the answers options of Never (0 times per month), Rarely (less than 3x/month), Sometimes (1–2 times/week), Often (3–4x/week), Almost every day (5 or more times/week). CASQ responses are assigned a numerical value (1 = never, 5 = almost every day) and then summed to produce an overall score, higher scores indicate more sleep issues. Sample statements included ‘I feel wide awake the whole day,’ ‘I feel drowsy if I ride in a car longer than 5 minutes,” and ‘I feel alert during my classes.’

Problematic internet use

PIU was measured using the validated Problematic and Risky Internet Use Screening Scale (PRIUSS) survey (Jelenchick et al., 2013, 2014). This instrument was created based on a PIU conceptual model (Moreno et al., 2013), and then further evaluated for psychometrics and convergent/divergent validity (Jelenchick et al., 2013, 2015). The Chronbach alpha for this scale was 0.96. Participants responded to 18 questions related to internet using a Likert scale, with Never = 1, Rarely = 2, Sometimes = 3, Often = 4, Very Often = 5. Example statements included ‘do you feel anxious when you are away from the internet’ and ‘do you choose to socialize online instead of in person.’ Overall scores were compiled for each participant for all 18 questions with a range of 0–72. Scores of 26 and higher were considered high risk, consistent with previous studies (Jelenchick et al., 2014).

Analysis

Latent class analysis is a statistical method that characterizes otherwise unobservable groups based on individuals’ response patterns of multiple observable variables (Bray et al., 2010). The analysis results in mutually exclusive latent classes of individuals based on their responses to observed measurements. Thus, LCA can be used to identify distinct subjects’ profile groups of multifaceted constructs, in this study these constructs included media use, parent involvement and health outcomes. In the initial model we included measures to represent these constructs, as well as demographic information as observable items in the LCA. Because LCA requires complete data for participants to be included in the models, any participants with missing data were excluded from the LCA process.

Measures included in LCA

For media device ownership, bedroom media use and media platform variables, responses were included as whether the participant reported owning the device (yes/no), whether the participant reported access to the device in their bedroom (yes/no) and whether the participant reported using the media platform (yes/no), respectively. For parent media involvement outcomes, responses were included as continuous variables, measured on a 5-point Likert scale with a higher score corresponding to more parent media involvement.

For health outcomes measures, physical activity was included as a categorical variable of high vs. low physical activity, separated at the median value of minutes of activity per week. Sleep outcomes were included as the summary score on the instrument, with higher scores indicating more sleep concerns. PIU was included as a binary outcome of meeting criteria as at-risk or not at risk (≥ 26 vs.  < 26).

The demographic variable of age was included as a binary variable (12–13 years vs. 14–15 years). Our rationale for combining 12–13 years is that these age groups are typically newer to social media in particular, while 14–15-year-olds typically have at least a year of experience on these platforms (Anderson & Jiang, 2018a, 2018b). Furthermore, gender was reported as a binary (male including biological male and trans male vs. female including biological female and trans female). Finally, parent education was included as a binary variable (college degree or higher vs. no college degree).

LCA analysis procedure

The LCA created a series of candidate models, each of which divided study subjects into a discrete number of mutually exclusive latent classes for characterizing media use, parent involvement and health outcomes. Candidate LCA models with between one and six different latent classes were evaluated and compared. The Lo–Mendell–Rubin (Lo et al., 2001) likelihood ratio test was used to identify the number of classes. In this analysis, the likelihood function of LCA model with k classes was compared with the likelihood function of a LCA model with k–1 classes. A p-value < .05 indicated that the model with k classes provided a better fit than the model with k–1 classes. When the difference in fit between two models was no longer significant, the more parsimonious model with k–1 classes was selected as a final model.

Initially, the LCA analysis was conducted using all available outcome measures (34 items). After identifying a final model based on available outcome measures, analysis of variance (ANOVA) and Fisher Exact test were used to compare individual items between classes for continuous and categorical items, respectively. Outcome measures with no significant differences between classes were then removed and a refined LCA was conducted on the remaining items using the same iterative process described above. Only one variable was identified as not different between classes, which was race. We assessed item-response patterns for each class to summarize and identify a final model with the best combination of homogeneity and separation (Bray et al., 2010). Several rounds of iterative discussion involving all investigators were then held to achieve consensus on latent class names and characterizations. The statistical analyses were conducted using SAS software (SAS Institute Inc., Cary NC), version 9.4 and M-Plus software (Muthen & Muthen, 1998–2017), version 8.

Results

The 1155 participants had mean age 13.6 years (SD = 1.1), 49.7% were female, 73.7% were White and 61.1% had parent education with a college degree (Table 1). A total of 1124 participants completed all assessments and were included in the LCA.

Table 1. Demographic characteristics of 12–15 year old US participants recruited via Qualtrics.

Variable	Statistics
Age (mean/SD)	13.6 (1.1)
Female (n/%)	571 (49.7)
Caucasian (n/%)	841 (73.7)
Parent Education College and above (n/%)	698 (61.1)
N = 1155

Media use

Most adolescent participants owned their own smartphone (82%), and 67% owned their own tablet. For social media use, approximately half of participants used common social media platforms including Instagram (52%), Snapchat (52%) and Facebook (43%). Approximately three-fourths of adolescent participants reported using video games on a regular basis, including Minecraft (37%) and Call of Duty (28%). Table 2 describes these data.

Table 2. Media ownership, access and use among 12–15 year old adolescent participants.

Device use	Media device: Own n (%)	Media device: Bedroom use n (%)
TV	753 (67)	607 (54)
Computer	725 (65)	455 (41)
Smartphone	918 (82)	795 (71)
None	27 (2)	93 (8)
Cell Phone	102 (9)	70 (6)
Other	21 (2)	0 (0)
iPod	12 (1)	11 (1)
Video Game	669 (60)	356 (32)
Tablet	748 (67)	620 (55)
Platform use	n (%) report use
Social media use
Facebook	486 (43)
Instagram	588 (52)
Snapchat	587 (52)
Twitter	211 (19)
Other	35 (3)
None	247 (22)
Video games
Minecraft	417 (37)
Grand Theft Auto	228 (20)
Call of Duty	314 (28)
Lego	139 (12)
Other	212 (19)
N = 1124

LCA findings

The LCA revealed three distinct profile groups to describe media use, parent involvement and health outcomes in this study population. The results of the LCA by class are shown in Table 3.

Table 3. LCA results across 12–15 year old adolescent participants.

	Latent Class
	1	2	3
Media use
Media device ownership
% of participants reporting owning the media device
TV	0.95	0.40	0.74
Computer	0.84	0.48	0.66
Tablet	0.79	0.59	0.61
Video game	0.83	0.38	0.62
Smartphone	0.98	0.64	0.89
Cell phone	0.08	0.11	0.07
Bedroom media device use
% of participants reporting access to the device in their bedrooms
TV	0.81	0.26	0.66
Computer	0.59	0.22	0.48
Tablet	0.67	0.45	0.56
Video game	0.54	0.09	0.39
Smartphone	0.92	0.45	0.83
Cell phone	0.05	0.08	0.04
Social media platform ownership and video gaming
% of participants reporting owning a profile on the platform
Facebook	0.59	0.26	0.51
Instagram	0.67	0.34	0.62
Snapchat	0.65	0.34	0.66
Twitter	0.31	0.07	0.23
Plays 1 or more video games regularly	0.86	0.64	0.69
Parent media involvement
Average Likert score (1–5) response to each statement (with 5 as very often)
My parents and I talk about my media use	2.05	2.27	3.32
My parents set rules or boundaries for my media use	2.01	1.97	3.55
Screen free zones	4.29	4.04	6.11
Screen free times	2.55	2.37	5.44
No screen time at meals (parents)	2.08	2.17	4.69
No screen time at meals (teens)	2.20	2.15	4.30
No screen time before bed (parents)	3.68	3.63	5.10
No screen time before bed (teens)	4.34	4.24	5.63
My parents co-view media with me	2.66	2.69	3.75
Health outcomes
Vigorous physical activity higher score = more physical activity	0.55	0.54	0.43
Sleep score sum higher score = more sleep issues	33.35	29.64	35.27
PIU % of participants high risk for PIU	0.77	0.64	0.85
Demographic information
Age groups
% of participants who were 12 & 13	0.41	0.58	0.34
% of participants who were 14 & 15	0.59	0.42	0.66
Gender
% of participants who were female	0.55	0.47	0.48
Parent education
% of participants with parent education of college degree or higher	0.42	0.32	0.46
N = 1124

The first profile group (Latent Class 1) included 399 participants, 35% of the total sample. It was characterized by a higher proportion of females compared to Latent Classes 2 and 3 (55% versus 47–48%). For media use, these participants had the highest reported ownership and use, including multiple devices and platforms. Notably, 98% of this group reported owning their own smartphone. Latent Class 1 participants were high users of both social media with rates of Facebook, Instagram and Snapchat use all above 50%, and 86% of participants in this class reporting regular use of video games. Latent Class 1 participants reported low-to-moderate levels of parent involvement in their media use; there were no parent media involvement measures in which this group scored the highest of the three profile groups. Across the three profile groups, Latent Class 1 had the highest reported physical activity, with 55% of the class having above the median for physical activity, moderate sleep and moderate PIU risk. Because of the high level of media use and ownership, as well as physical activity, and low levels of parent involvement, we characterized this group as Active Autonomous Media Users.

The second profile group (Latent Class 2) was the largest and included 480 participants, or 42% of the sample. It was characterized as a younger group of participants compared to Latent Classes 1 and 3. This profile group reported that their parents were less likely to be college graduates (32%) compared to the other two profile groups (42–46%). These participants were less likely to report device ownership compared to both other groups and had the lowest frequency of access to media devices in their bedrooms (with the exception of ownership of and access to cell phones without internet access). Latent Class 2 participants reported low-to-moderate levels of parent media involvement, including the lowest agreement with screen-free time restrictions and avoiding screen time at meals among the three groups. However, this group had a high agreement that they and their parents avoided screen time before bed. This group had moderate physical activity, the lowest reported sleep issues and lowest risk for PIU. Because of the low media use, low parent involvement and positive health outcomes particularly for sleep, we characterized this group as Young Low-Tech Sleepers.

The third profile group (Latent Class 3) was the smallest, including 276 (24%) participants characterized as an older group of participants who were more likely to be male (52% of this class). This group had moderate device ownership, bedroom device use and social media/video gaming, consistently in between Latent Class 1 and 2. This group reported the highest levels of parent involvement, including the highest rating for following screen time rules before bed, the highest levels of reporting rules around screen free zones and times, and most agreement with co-viewing media with parents in comparison to Latent Classes 1 and 2. This profile group had the lowest level of vigorous physical activity with only 43% in the above-median group for physical activity, the highest sleep issue scores and the highest proportion of participants with PIU (85% of this class). Thus, we characterized this profile group as Risky Regulated Media Users. Full results of the LCA are shown graphically in Figure 1 to illustrate overlap in key variables across classes.

Figure 1. Figure illustrating three latent classes and distribution of variables within classes. Overlap in circles represents variables shared across distinct classes.

Discussion

This study used LCA to develop profiles of media use and parent involvement and their associations with health outcomes. While previous studies have illuminated links between media and individual health variables (Levenson et al., 2016; Lobelo et al., 2009), the LCA classification provides a rich understanding of patterns in which adolescents use media, as well as an opportunity to integrate these patterns with the role of parents and the association with health outcomes. In particular, the role of parents is a critical component to understanding the intersection of media and health.

Profile group 1

The first profile group of Active Autonomous Media Users brings to mind a busy adolescent who may be involved in multiple sports or activities, who may be social both offline and online, and who may need to fit media use around other activities. This group may have lower parent involvement for their media use just because these adolescents were busy with other activities. Adolescents in this group may be perceived by their parents as requiring less regulation of media use based on having less time available for media. Alternatively, these adolescents may value media less and thus seek out more extracurricular activities. A potential implication for a clinician interacting with a patient who is an Active Autonomous Media User may be to consider an emphasis on screen time limitations as less critical. Clinicians providing media guidance to adolescents from this group may wish to emphasize the importance of maintaining adequate sleep, and recommend limiting media use around bedtime to achieve that goal. Future research directions for this profile group may include examining this group further with measures such as extracurricular activities to better understand media in their daily lives.

Profile group 2

The second profile group of Young Low-Tech Sleepers reported the lowest levels of access to devices and platforms. This group was younger, and more likely to have parents without college degrees which may suggest lower socioeconomic status. However, a previous study found that children in lower socioeconomic households had more access to electronic media devices in their bedrooms (Tandon et al., 2012). Adolescents in this profile group also reported lower parent involvement and media rules. This group brings to mind a younger teen without consistent access to media devices at home. Alternatively, this group may value media less and thus not request access to these devices. For a clinician interacting with a patient who is a Young Low-Tech Sleeper, an emphasis may be on commending patients for positive health behaviors around sleep and physical activity. The clinician may want to explore ideas for how media rules will be approached either in settings outside the home, or if devices eventually are brought into the home. Future research directions for this group may include understanding whether they continue their lower risk status for health outcomes as they begin to obtain personal media devices.

Profile group 3

The third profile group was perhaps the most intriguing and represented approximately a quarter of our sample. The Risky Regulated Media Users profile group had high parent involvement yet poor health outcomes, which is challenging to explain. On the one hand, this profile group may represent high-risk media users who would have even worse health outcomes without high levels of parent involvement. Alternatively, this group may represent adolescents whose parents create rules, but are ineffective in enforcing them. This explanation brings to mind a family in which media rules are present, but not enforced, such that the adolescent may sneak the phone into the bedroom at night. A previous study found that increased parental restriction was associated with increased youth media use (Barbovschi et al., 2015). Another study found that parent monitoring, in comparison to restriction, was associated with lower screen time (Gentile et al., 2014). It is also possible that the Risky Regulated Media Users group represents teens who are engaging in risky media use, thus their parents are trying to become more involved and regulate their use. And finally, as we cannot ascertain directionality in this cross-sectional study, it is also possible that these adolescents who are already struggling with their health seek out media-based activities, under the guidance of their parents. The Risky Regulated Media Users group merits further study to understand ways to prevent media-related negative health outcomes in partnership with parents.

Placing findings in the context of existing literature

This study’s findings expand upon the current literature that has mainly focused on large population-based studies. Some of these studies have found small positive associations between poor mental health and technology (Twenge, 2020), while others have found no significant relationship between these constructs (Przybylski & Weinstein, 2017). A large research review recently concluded that adolescents are not a homogenous population, and that future work needs to consider approaches beyond what has informed the current evidence (Odgers & Jensen, 2020). Our study builds upon Odgers and Jensen’s recommendations. Rather than using variable-centered whole-population analyses, we used a person-centered analysis approach of LCA. This approach allows one to identify classes within populations. In our study, we identified three distinct classes with unique patterns of technology and health behavior. While this approach does not overcome limitations of cross-sectional data, such as the inability to assess causal directions, it does allow for the identification of groups with shared characteristics within a single sample of participants. These novel analysis approaches represent a step forward towards understanding the heterogeneity within the adolescent population and allow researchers to consider new models, prevention approaches and paradigms for adolescent health and technology use. Future studies should include the role of media for learning platforms, given the increases in virtual learning associated with the COVID-19 pandemic.

Limitations

The first limitation of this study is that our results may not generalize beyond a study population of early adolescents recruited via Qualtrics. Recruiting from an online panel of participants meant that we could designate study population size and criteria but limited our ability to assess the external validity of the sample. Further, we were able to match US census data for race/ethnicity of the parent. Future studies could consider approaches to represent race/ethnicity for adolescents based on census data. However, the Qualtrics platform and panels have been used in other studies of early adolescents (Len-Rios et al., 2016), and the panels have been found to have close approximations of US populations (Heen et al., 2014). Second, LCA provides a systematic approach to create profiles representing critical variables, though our interpretations of those profile groupings may have inaccuracies. Because our data is cross-sectional, we cannot assess directionality in the relationships between our measured variables. Further, other unmeasured variables, such as sports or extracurricular activities, may have impacted sleep and physical activity. This study used cross-sectional data, which is common within the LCA approach but does not allow an understanding of long-term predictors or consequences. Future studies should include measures of other offline activities to understand the range of offline and online activities that adolescents engage in each day.

Conclusion

Study findings suggest that media use is complex, and health outcomes are not fully explained by device ownership or specific platform use. We did not find specific latent classes centered on particular media platforms or games. Given that we did not identify specific differences by platforms, future studies may consider summaries of platforms. Further, future studies may want to examine nuances related to motivations for use or priorities in use rather than platform differences. The profile groups illuminated through this large sample LCA approach suggest that media use and parent involvement are intertwined and associated with specific health outcomes. Thus, we encourage further research that delves in a more nuanced way into the relationship between pediatric media use patterns and health, with parent involvement as a critical moderating variable.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

Funding for this project was provided by Seattle Children’s Research Institute; Center for Child Health, Behavior and Development.

Notes on contributors

Megan A. Moreno

Megan A. Moreno (MD, MSEd, MPH) is a Professor of Pediatrics and an Affiliate Professor of Educational Psychology at the University of Wisconsin-Madison. She is the Vice Chair of Digital Health for the Department of Pediatrics, and Division Chief for General Pediatrics and Adolescent Medicine. Dr Moreno is the Principle Investigator of the Social Media and Adolescent Health Research Team (SMAHRT) [email: [email protected]; [email protected]].

Aubrey D. Gower

Aubrey D. Gower is currently a first-year medical student at the University of Washington School of Medicine. She worked as a research specialist at the University of Wisconsin-Madison. She previously completed her bachelors at the University of Washington in Biology and Psychology. Her research has included topics related to adolescent health, screen time and problematic internet use

Daniel Pham

Danny Pham is an undergraduate student at the University of Washington. He completed a research internship at the University of Wisconsin-Madison. His area of study is biostatistics and informatics.

Qianqian Zhao

Qianqian Zhao (MS) is a Biostatistician at the University of Wisconsin Department of Biostatistics & Medical Informatics. She has experience in the design and analysis of clinical and pre-clinical studies. Her research interests are in pre-clinical modeling and clinical trials with a focus on oncology, while serving my collaborators in answering their scientific research questions across a broad range of areas.

Jens Eickhoff

Jens Eickhoff (Ph.D.) is a senior biostatistician at the University of Wisconsin Department of Biostatistics and Medical Informatics. His statistical methodology research is focused on latent variable modeling and includes LCA approaches.