Ethical approval was obtained from the University of Southern Queensland Human Research Ethics Committee (H19REA248). In line with international guidelines [43], this study adhered to strict procedures for using wearable cameras in human research. Data were collected between February and March 2020 (before any effects of COVID-19 restrictions) and analyzed in August 2020. Written informed parental and adolescent consents were obtained before data collection.

Ten participants (aged 13-17 years) were recruited from a secondary school in Queensland, Australia. All students in grades 8 to 11 (age 13-17 years; N=100) attended a face-to-face information session in which they were invited to take part in a research study. At the end of the session, the principal investigator (GT) answered questions and provided research packs to adolescents interested in participating (n=17). Written parental and adolescent consents were obtained for 10 adolescents (response rate 59%).

Examples of images and coding are presented in Figure 1. Image coding was completed between April and July 2020. Images for each participant were manually coded by the first author in a spreadsheet, based on a pre-established codebook for wearable camera research on children’s screen use. The coding protocol was structured into different annotation groups (Multimedia Appendix 1). For each annotation group, the coder identified all the images and categorized only the screen components specific to that group. Images with multiple screens were coded for multiscreening, detailing the primary, secondary, or background activity. Images with inactive or absent screens were also coded, in addition to blurry or blocked images because of the position of the camera. Obscured images (such as those where the camera was facing the ceiling) in the middle of image sequences containing screens were coded based on the nonobscured preceding and subsequent images. A subset of images (10%) was repeat coded by a second researcher, and interrater reliability was tested using the Krippendorf (α) statistic to determine consistency between coders for all coding categories. Interrater reliability was interpreted using the guidelines of Krippendorf [44].

IBM SPSS Statistics version 21.0 was used for descriptive analyses. Daily camera wear time was calculated as the total number of minutes the camera was turned on. Captured time (minutes) was the number of images divided by 6 (assuming each image represented 10 seconds). The frequency and percentage of images were calculated for each screen-based device for each annotation group (eg, location and social interaction). Descriptive data are provided to describe the frequency and types of devices being used, the content being viewed, and the social and environmental context in which such behaviors occur. To analyze the temporal patterning and to compare screen-based behaviors between the different evening segments, equal time segments of 3 hours were utilized. For each time period, the frequency of each screen-based behavior was computed, and the percentage of behaviors occurring at that time period has been reported. Temporal data were analyzed and reported separately for weekdays and weekend days because they have different structures and are likely to lead to different behavioral choices. Differences between weekday and weekend screen use were analyzed using the chi-squared test, and expressed as a percentage of images (standardized by weekday and weekend wear times). Based on previously established definitions [45], weekday after-school time segments were defined as follows: “after school to 18:00,” “18:00 to 21:00,” and “21:00 to sleep” (eg, when the camera was removed prior to bedtime).

The average reliability between the 2 coders across all categories was acceptable (α=.81) [44]. With regard to agreement, the α values were .89 for device attention, .86 for device type, .77 for content type, .88 for content classification, .60 for physical setting, .85 for social setting, .84 for social interaction, .72 for co-existing behaviors, and .88 for the uncodable category.

The characteristics of the participants are presented in Table 1. Five girls and five boys participated, with an average age of 15.4 years (SD 1.6 years). The main language spoken at home was English (9/10, 90%), with an average of 4 persons living in the household. Participants’ parents who responded were mainly mothers (8/10, 80%), married (9/10, 90%), and earning a total annual household income >AUD 78,000 (9/10, 90%; 1 AUD = 0.73 USD), and had completed a university or tertiary qualification (8/10, 80%).

A total of 71,396 images, derived from 30 school weekday evenings and 10 weekend days, were coded and included in the analysis. This represented just under 200 hours of total camera wear time. Multimedia Appendix 2 shows the mean and median (IQR) numbers of images collected, camera wear time, captured time, and screen time per day for weekdays and weekend days. The camera wear time averaged 230.5 minutes on a weekday evening and 508.1 minutes on a weekend day. The camera captured, on average, 1365 images per weekday evening and 3045 images per weekend day, equating to 227.5 minutes and 504.2 minutes of captured time, respectively. Of this, 167.7 minutes were spent, on average, using screens on a weekday evening, and 371.3 minutes were spent on a weekend day. The results showed that there was no significant difference between weekday and weekend screen use (72.7% vs 73.1%, P=.23).

Table 2 shows the frequency and percentage of different screens and activities in the entire image set (N=71,396). In total, 52,842 (74.0%) images contained screens. The most common screens were televisions (25,950/71,396, 36.3%), smartphones (20,851/71,396, 29.2%), and laptop computers (15,309/71,396, 21.4%), while fewer images contained tablets (2720/71,396, 3.8%), desktop computers (20/71,396, <1%), and wearable smartwatches (1/71,396, <1%).

The most common activities, as determined by the proportion of images recorded by the wearable camera, differed according to the screen domain. For instance, our data showed that conventional television sets were popular among adolescents, although this comprised mostly playing action games (ie, including fighting, shooter, or platform games) via gaming consoles (14,032/25,950, 54.1%), rather than watching traditional action television programs (ie, programs with real people or animals), which accounted for less than half of all television occurrences (11,803/25,950, 45.5%). Given this information, the results for “television set” occurrences were described by (1) television set: television viewing and (2) television set: action gaming. For smartphones, watching television programs through online streaming sites, such as Netflix and YouTube (10,432/20,851, 50.0%), social networking (5642/20,851, 27.1%), and communicating (1618/20,851, 7.8%) constituted the main content types, compared to creative content, such as productivity software (eg, Word, Excel, and PowerPoint), which made up 39.5% (6051/15,309) of all laptop computer occurrences. Watching television programs (4763/15,309, 31.1%) and internet use (3409/15,309, 22.3%) also made up a large proportion of content engaged on the laptop computer, while the same content accounted for 51.0% (1387/2720) and 30.3% (825/2720) across all tablet occurrences, respectively. A wearable smartwatch was captured in 1 image, showing the home screen interface. For all desktop computer images (n=20), specific content could not be determined owing to inadequate resolution.

As shown in Table 3, recreational activities made up the majority (45,218/64,856, 69.7%) of all screen occurrences, compared with other content classifications, such as educational (10,603/64,856, 16.3%) and social (7450/64,856, 11.5%) activities. Concerning individual device types, all traditional television viewing was classified as recreational, over half of laptop computer activities were educational (9361/15,309, 61.1%), and more than a third of smartphone exposure was deemed social (7265/20,851, 34.9%). Tablets were commonly used for recreational (1627/2720, 59.8%) and educational (1014/2720, 37.3%) purposes, although tablets comprised only 3.8% (2720/71,936) of total screen exposure.

As shown in Multimedia Appendix 3, more than 16% (11,976/71,936, 16.7%) of images contained multiple screens, with the most prevalent combinations of screens being (1) television-smartphone (7324/11,976, 61.2%), (2) smartphone-laptop (2558/11,976, 21.4%), and laptop-television (985/11,976, 8.2%). The majority of multiscreening involved (1) televisions as the primary screen and smartphones as the background screen (5029/11,976, 42.0%) used for gaming and watching television programs, respectively; (2) smartphones as the primary screen and televisions as the background screen (2285/11,976, 19.1%) used for watching television programs and gaming, respectively; and (3) smartphones as the primary screen and laptops as the background screen (1465/11,976, 12.2%) used for social networking and internet use, respectively.

As shown in Multimedia Appendix 4, nearly all screen exposures occurred in the home setting (62,455/64,856, 96.3%), such as the living room (37,364/64,856, 57.6%) and bedroom (19,473/64,856, 30.0%). Concerning individual screen domains, all action gaming (via television set) and the majority of traditional television viewing (10,793/11,803, 91.4%) occurred in the living room, whereas laptop computers were commonly used in the bedroom (8974/15,309, 58.6%). Smartphones were used in several areas, including the living room (8719/20,851, 41.8%) and bedroom (7932/20,851, 38.0%), and when in private transport (1564/20,851, 7.5%), while the bedroom (1709/2720, 62.8%) and kitchen/dining room (848/2720, 31.2%) served as popular locations for tablet use.

The social contexts surrounding adolescents’ screen exposure are presented in Multimedia Appendix 5. Most involved no in-person social interaction (54,430/64,856, 83.9%), although this differed by screen domain. In-person social interaction was greater when watching television programs, including co-viewing with an adult (1719/11,803, 14.6%) or child (1368/11,803, 11.6%), whereas nearly all action games were played alone (12,915/14,032, 92.0%). For smartphones, 12.8% (2668/20,851) of images involved an adult in the background, while fewer in-person social interactions were experienced with laptop computers, tablets, and desktop computers (with 92%-100% of all occurrences engaged in alone).

As shown in Multimedia Appendix 6, the majority of screen use was in isolation (56,656/64,856, 87.4%). Some co-existing behaviors that occurred alongside screen-based behaviors included writing using a pen and paper (3873/64,856, 6.0%), eating a snack (1755/64,856, 2.7%), or eating a meal (1454/64,856, 2.2%), but this varied according to screen domain. Laptop (2430/15,309, 15.9%) and tablet (800/2720, 29.4%) computers were commonly used when writing with a pen and paper, while over 10% (1288/11803, 10.9%) of television viewing occurrences involved consuming food (eg, snack or meal).

For each time period, the frequency of each screen-based behavior was computed. These data represent the percentage of behaviors occurring at that time period; thus, the results reported are necessarily descriptive. Figures 2 (weekdays) and 3 (weekends) compare the temporal patterns of screen-based behaviors.

This study aimed to use automated wearable cameras to describe adolescents’ screen exposure. Consistent with a recent systematic scoping review [46], our data indicated that adolescents are exposed to high amounts of screen use. In particular, our data showed a greater percentage of time on smartphones, with engagement in various activities, such as watching programs, social networking, and communicating online, compared to time using television sets and engaging in conventional television viewing. This finding supports the perception that use of newer digital media is increasing, with some displacement of traditional forms of media for adolescents [16]. Indeed, reports show a decline in watching programs on conventional television sets, despite an increase in consuming television content on the internet [47]. This is likely caused by the multiple functions that smartphones offer, including the social and recreational tasks performed online [48]. Moreover, the portability of smartphones allows adolescents to use these devices ubiquitously [49]; almost anywhere in free-living conditions as reported here. Future studies need to determine effective strategies for the responsible use of contemporary screen engagement, paying attention to the use of smartphones [13]. Moreover, these findings have implications for the assessment of television viewing, a common category for screen time measurement. If television programs are watched on television sets, as well as other devices, we may be estimating behaviors incorrectly, depending on the nature of the question asked. Hence, it is possible that there may be conflating of the assessment of behaviors and devices. Better measurement that captures the types of devices used for watching television programs, in addition to the social environmental contexts of such viewing, is warranted.

This study also showed that multiscreening is evident in screen use among adolescents, supporting conclusions from a previous study [50]. Here, we revealed that multiple screens were identified in approximately 17% of images across the entire image set. This rate was more than 3 times the rate reported by Smith et al among adolescents in New Zealand [39]. An important part of understanding multiscreening is examining the combinations of tasks undertaken. Contrary to previous findings based on self-report data [19,51], our data showed that gaming via television together with watching programs on a smartphone was the most common combination of screen exposure. Previously, it has been argued that gaming is harder to combine with another screen because it demands many cognitive capacities and behavioral responses [52]. One explanation for our findings is that although gaming was used concurrently with smartphones, the latter was predominantly used in the background and therefore was less likely to interfere with adolescents’ cognitive demands of gaming. Other possible explanations are that smartphones were used to temper impatience or boredom whilst waiting for a game to load [17,48], or offered an opportunity to socialize with friends whilst watching television shows [48]. Further investigations on why adolescents engage in certain multiscreening behaviors (eg, social functions) are needed to help researchers deliver effective interventions to change screen-based behaviors, if deemed necessary. Models and theories, specific to multiscreening behaviors among adolescents, might also warrant further enquiry.

The home environment may serve as an important setting for interventions that aim to influence adolescents’ screen exposure. Participants in this study spent the majority of their time at home and, as such, engaged in most of their screen time at home. Consistent with previous findings [21], the living room was strongly linked to television viewing. While expected, this finding suggests that if reductions in television viewing are sought, this location may be a target for interventions, such as through environmental restructuring involving reconfiguring seating arrangements or family rules for behaviors while in the room. In addition to the living room, the bedroom may also be an important context for screen use, particularly involving smartphones, tablets, and laptop computers, as shown in the current study. It is possible that adolescents feel they have greater privacy and have fewer interruptions from family members in this setting. This might encourage prolonged recreational use of screens in this location, often whilst sitting or lying down [48]. Indeed, reducing access to screen-based devices in the bedroom has been identified as a facilitator to reduce screen use [53], although this will not guarantee a reduction in sedentary behavior as other sedentary pursuits may be adopted as a substitute.

Corroborating recent qualitative data [41], the present study found that adolescents had very little in-person social interactions with others whilst using screens. Such findings support the hypothesis of time displacement for social interaction, that is, more adolescents spending time on digital media and less time on face-to-face social interaction [54]. This has led to concern over the detrimental impacts of screen use on adolescents’ psychological well-being, with some finding links to depression, loneliness, and lower social connectedness [55-57], although the associations might be small and complex [8,56,58]. Others argue that digital media may instead compliment in-person social interaction, particularly media involving opportunities to interact online [59]. For instance, despite a lack of in-person social interactions observed in this study, it is highly likely that adolescents engaged in numerous online interactions whilst using screens, for example, through online communication or playing interactive video games. The mechanisms of the putative effects of screen use on psychological well-being should be explored further to better understand the impact of lower in-person social interactions that might be characteristic of traditional solitary screen use, in addition to higher digital social interactions that might be typical of newer digital media. Such research may need to account for the types of devices as well as physical and social settings.

By investigating the temporal patterns of screen use, we were able to understand the typical schedule of adolescents and identify timeframes when specific screen-based behaviors compete against each other. Similar to previous self-report data [21,23], screen-based behaviors were shown to be prominent throughout the evening, starting immediately after school until the end of the day. As shown in a systematic review [42], the after-school period is linked to high screen use in adolescents. This was particularly apparent for laptop computers, which were likely used to complete educational tasks, such as homework. Despite the sedentary nature of this behavior, such tasks are generally considered to be important and valuable; thus, whether it should be reduced or replaced with other behaviors, such as physical activity, is debatable [23]. This supports the argument that sedentary behaviors should not be viewed in isolation [21].

As for television viewing, this behavior most likely occurred in the middle evening segment, corroborating previous findings using self-report [21]. However, the same behavior reduced rapidly in the prebedtime hours, comprising just 2.4% of all screen occurrences. This may be due to the changes in patterns of media consumption among young people. Here, we found that smartphones peaked in the hours before bedtime, supporting evidence that portable devices are increasingly part of the adolescent sleeping environment [60]. For instance, using data from a large population-based survey of adolescents in Norway (n=9846), Hysing et al showed that approximately 80% of boys and 90% of girls used a cell phone in the hour before going to sleep [61]. Together, these findings may cause a rise in public health concern, especially given the evidence for associations between prebedtime screen use and a number of poor sleep markers (eg, inadequate sleep quantity, poor sleep quality, and excessive daytime sleepiness) [62]. As such, smartphones may be an important target for interventions that aim to mitigate the risks associated with prebedtime screen use [63] and sleep interventions in general.

A strength of this study was the measurement of adolescents’ screen exposure, which was significantly enhanced through the use of wearable cameras. Such devices offer an improvement over existing self-report measures of lifestyle behaviors and the contexts in which they occur. A high agreement between coders was reported, similar to a previous study using wearable cameras [39]. However, these devices also have limitations. First, the 10-second epoch between image capture may have missed possible screen exposure, particularly quick and sporadic smartphone checking [49]. Future studies that compare a continuous video or 1-second epoch with longer intervals between image capture are warranted. Second, if we wish to ascertain the function (eg, relaxation and entertainment) that different screens serve for adolescents, we are unlikely to do this through camera images. While we were able to infer the content being viewed, further qualitative work will enable a more in-depth understanding of what functions are being served by engaging in different devices and platforms. Third, there is the possibility of the Hawthorne effect, whereby participants modified their behavior in response to wearing an automated wearable camera. This may have implications for the validity of this study. Fourth, the annotation of wearable camera images was based on decisions made by the coders. This limitation may have been offset by conducting interrater reliability tests, showing an almost perfect agreement between coders [44]. The recent development of an annotation protocol for sedentary behavior in children using wearable cameras [64] shows promise, and once applied to larger samples of children, this protocol can help better understand adolescents’ contemporary screen engagement. In addition to coding issues, the data processing and coding times are limiting factors and may be unsuitable for use in large-scale studies, unless an automatic recognition algorithm is developed to classify different aspects of human behavior [26].

Other limitations include a small sample size and the relatively homogenous demographic characteristics of the sample. Therefore, the results are unlikely to be generalizable to the wider adolescent population. Future research needs to consider other sociodemographic groups to confirm the key findings observed in this study. A further limitation was that due to ethical concerns raised by school principals regarding camera wearing on school grounds, we only examined screen use during the after-school period and weekends. Since screen use before and during school may yield different results and patterns, future studies are needed to examine the exposure to screen use across the day. Finally, as with other wearable camera studies, the sample size was small, and thus, the study was insufficiently powered to use temporal patterns as a means for testing differences. This should be considered in future studies with larger sample sizes.

Among a small sample of adolescents, we showed high amounts of screen use, most of which occurred in the home, with little social interaction. This information might be used when designing interventions to inform new policy to influence adolescents’ screen use. For example, Australian guidelines for physical activity and sedentary behavior recommend no more than two hours of recreational screen use daily for this age group [63]. Moreover, we showed that wearable cameras may provide a new approach to collect more accurate data on screen-based behaviors in free-living conditions, and with some volume. As such, we were able to both enhance traditional self-report and provide context and temporal specificity surrounding screen-based behaviors in free-living settings. Our findings may be used to inform guidelines and protocols for visual research on screen-based behaviors, and form a basis for larger-scale studies for comparisons.