Teens, Screens and Well-Being:
An Improved Approach
Concerns about emergent technologies form a continuous cycle: they appear when a new technology surpasses a certain popularity threshold and stop once a newer technology prompts the cycle to restart. Currently, society is concerned about teenage digital technology and social media use, which is feared to negatively impact well-being. While there has been a flurry of research in the area, remarkably little consensus has been reached. I argue in this thesis that such a lack of consensus stems from a lack of methodological rigour. Using innovations put forth by psychology’s recent credibility revolution, I devise an improved methodological framework for the study of emergent technologies. I then apply it to investigate digital technology and social media use’s association with decreased adolescent well-being, using large-scale secondary datasets from the United Kingdom, Ireland and the United States. My first study implements a methodological approach called Specification Curve Analysis, which tackles analytical flexibility by illustrating how diverse analytical approaches can produce a wide range of numerical outcomes. Furthermore, I develop a comparison specification method to put the size of the associations found into perspective. My second study focuses on diversifying measurement and implements an explicit exploratory and confirmatory hypothesis testing framework. Lastly, my third study utilises high-quality longitudinal data to highlight the bidirectional and individual nature of the effects of interest. All studies indicate that previous systematic reviews overestimated the negative link between digital technology use and well-being. It is however still unclear whether the small, complex and inherently multivariate relationship found in this thesis should engender widespread policy change.
In 1941 Mary Preston published “Children’s Reactions to Movie Horrors and Radio Crime” in The Journal of Pediatrics (Preston 1941). The American paediatrician had studied hundreds of six to sixteen-year-old children and concluded that over half were severely addicted to radio and movie crime dramas, having given themselves “over to a habit-forming practice very difficult to overcome, no matter how the aftereffects are dreaded”. Most strikingly, many of them consumed these dramas “much as a chronic alcoholic does drink” (Preston 1941). Preston therefore voiced severe concerns about the children’s health and future outcomes: those who consumed more radio crime or movie dramas were more nervous and fearful, suffered from worse general health and more disturbed eating and sleep. Drawing such inferences about the radio, the emergent technology of the era, followed naturally from Preston’s opinion that the consumption of “terrifying scenes can have an inhibitory effect on the functioning of every organ in the body” (Preston 1941).
Almost 80 years later, Mary Preston’s conclusions seem like a clear exaggeration of concerns. Today many parents would welcome if their children listened to dramas on the radio instead of playing around on their phones or chatting to friends on social media. Yet this research is a striking reminder that every decade new technologies enter human lives; and that in their wake there will arrive widespread concerns about their effects on the most vulnerable in society.
Considering the context of Preston’s work more broadly, radio was experiencing an extensive growth in popularity when the study was published. In 1936 about 91% of New York homes owned a household radio, which children spent between one and three hours a day listening to (Dennis 1998). This popularity sparked concern not limited to Mary Preston’s article. A New York Times piece considered whether too much listening to the radio would harm children and lead to illnesses, since the body needed “repose” and could not “be kept up at the jazz rate forever” (Ferrari 1929; as cited in Dennis 1998). The Director of the Child Study Association of America noted how radio was worse than any media that came before because “no locks will keep this intruder out, nor can parents shift their children away from it” (Gruenberg 1935). This view was mirrored in a parenting magazine published at the time: “Here is a device, whose voice is everywhere (…) We may question the quality of its offering for our children, we may approve or deplore its entertainments and enchantments; but we are powerless to shut it out (…) it comes into our very homes and captures our children before our very eyes” (Frank 1939; as cited in Dennis 1998).
When worries about specific technologies grip a population, it is often overlooked that such concerns form part of a constant cycle. Almost identical questions are raised about any new technology which reaches the spotlight of scientific and public attention. These are then addressed by scientists, public commentators and policy makers until a newer form of technology causes the cycle to restart. Concerns about the radio, for example, only abated once the home television became popular (Dennis 1998). This cyclical nature of technology panics was already described eloquently in 1935: “Looking backward, radio appears as but the latest of cultural emergents to invade the putative privacy of the home. Each such invasion finds the parents unprepared, frightened, resentful, and helpless. Within comparatively short member, the “movie”, the automobile, the telephone, the sensational newspaper or magazine, the “funnies,” and the cheap paper-back book have had similar effects upon the apprehensions and solicitudes of parents” (Gruenberg 1935).
In 2019, one could simply replace ‘radio’ with ‘social media’ in such a statement: most parents would agree that its arrival into the home has also left them unprepared, frightened, resentful and helpless. Mirroring Preston’s concerns about the radio, academic publications and other reports now routinely liken social media to drug use (Royal Society of Public Health 2017; see commentary Przybylski and Orben 2017). These fears once again raise the spectre of vast proportions of the adolescent population becoming addicted to a new technology (Murali and George 2007) and this having diverse and far-reaching negative consequences (Greenfield 2014; see commentary Bell, Bishop, and Przybylski 2015). While we are amused by previous parents’ fears of radio addiction, current concerns about smartphones and social media are shaping and influencing policy around the world. In the United Kingdom there have been various committees, inquiries and white papers about this topic (Davies et al. 2019; House of Commons Science and Technology Select Committee 2019; Viner, Davie, and Firth 2019; Department for Digital, Culture, Media and Sport and Secretary of State for the Home Department 2019), in Asia laws have been implemented that try to curb technology use entirely (Choi et al. 2018) and in the United States there is a movement to ban smartphones before the eighth grade (Wait until 8th 2018).
Psychology and its provision of scientific evidence play a key part in the current panic about digital technologies. This means that strict measures of methodological quality control need to be in place to ensure that the research delivered will have a positive, reasoned and scientific influence on the current technology panic and those of the future. The credibility revolution in psychology has set out many processes in which the discipline can improve the quality of its scientific endeavour (Munafò et al. 2017; Vazire 2018). This thesis will examine how such innovations and improvements can be implemented to produce a methodological framework that ensures psychology does not become an accomplice in exacerbating technology panics, but instead uses its power to promote scientific and robust reasoning in the face of societal change.
1.1 Moral Panics
Rapid increases in concern are not restricted to fears regarding technology but occur regularly throughout public life. Cohen defined those worries that reach widespread popularity as moral panics (Cohen 1972). They occur when a person, group, thing, event or other entity is perceived as challenging societal values and norms. This causes introspective ‘soul searching’ in the population and moral condemnation (Garland, 2008). The central concerns engendered are most often focused on “The Other”: a group that does not constitute the main powerholders of society. The social critics, journalists and researchers are therefore “interestingly immune” when it comes to the negative effects of the panic at hand (Grimes, Anderson, and Bergen 2008, 51). They instead focus on less-powerful subgroups like children or women, for which the entity of the moral panic becomes defined as a severe threat by presenting it in a stylised and stereotypical fashion (Cohen 1972, 28). Powerful societal actors like editors, policy-makers, religious leaders and those occupying positions of respect speak out about the problem and what they think could be a solution. The moral panic then stays in the public mind until it “disappears, submerges or deteriorates” (Cohen 1972, 28). Cohen therefore noted that moral panics can be trigged by a wide variety of things or people and can last a short or long time.
1.1.1 The Last Century
Moral panics about the youngest generations in society – what affects them and how they are developing – have been present for centuries. Greek philosophers already voiced concerns about the damage writing would do to society and youths’ increasing lack of respect centuries ago (Wartella and Reeves 1985; Blakemore 2019). Panics come and go: consistently reincarnated when a new technology or development gains popularity across society. A complete historical summary of moral panics is therefore out of the scope of this thesis. I will focus instead on the moral panics about new technologies of the last century, which have become powerful and recurring phenomena (Wartella and Reeves 1985).
Radio and movie dramas are only one example of the wide range of technologies that have been implicated in such technology panics. In the decade that followed the moral panic about radio, concerns about comic books were on the rise. In 1954, the psychiatrist Fredric Wertham wrote a scathing and scaremongering book called Seduction of the Innocent, where he voiced great concern about the effects of comic books on young people: “Slowly, and at first reluctantly, I have come to the conclusion that this chronic stimulation, temptation and seduction by comic books, both their content and their alluring advertisements of knives and guns, are contributing factors to many children’s maladjustments.” (1954, 10). Society and politicians were receptive to Wertham’s hypothesis of comic books being the cause of recent rises in juvenile delinquency, even though such a trend could have also been explained by societal or cultural issues. Yet such underlying causes would have been much more difficult to address. In a New York Times article, the sociologist C. Wright Mills lauded the book as “a most commendable use of the professional mind in the service of the public” (Wright Mills 1954; as cited in Tilley 2012). Wertham’s work was instrumental to the passing of restrictive comic book legislation that has been recognised as a major contributing factor to the demise of the genre (Tilley 2012). Yet history does not look favourably on his work. It is now known that he manipulated his evidence and exaggerated his claims while refusing to share the underlying data, privileging “his interests in the cultural elements of social psychiatry and mental hygiene at the expense of systematic and verifiable science” (Tilley 2012).
Panics about comic books, radio and innovations like new music (Sternheimer 2003), were however overshadowed by the concerns raised about television and its possible promotion of violent behaviour. A report by the US Media Task Force found that television violence was “one major contributory factor which must be considered in attempts to explain the many forms of violent behaviour that mark American society today” (Lowery and DeFleur 1988, 309). Yet – again – high quality evidence is lacking. A seminal before-and-after study on children in Norwich who had previously not had access to home television found that television did not increase aggression levels (Himmelweit, Oppenheim, and Vince 1961). The study also highlighted that most children watched television selectively without it dominating their lives. During times of panic, however, the provision of high-quality evidence doesn’t alleviate the worries of critics and the pressure to implement policy change. An editorial in Pediatrics, for example, subsequently noted that professionals need to “avoid the intellectual trap of minimizing the importance of television’s effect on child and adolescent behaviour simply because the literature does not contain straightforward, statistically validated research” (Strasburger 1989). With the advent of newer technologies, however, the moral panic about television also waned. While television, comic books, radio, or movies were once the centre of attention, the cycle of media panics has moved on, relegating these technologies to the background of societal concerns.
1.2 The Sisyphean Cycle of Technology Panics
The continual rise and fall of concern about new technologies, driven by the want to comprehend and explain their influence on society, is an age-old component of societal debate. The nature of these debates has, however, changed in the last century. While some scientific commentators played a part in the technology panics about radio or comic books (Preston 1941; Wertham 1954), most of the debate was held outside of scientific arenas. Since the technological panic about television, however, science has been progressively included in the debate (Dennis 1998). This development is unsurprising as scientists are operating in an increasingly industrialised scientific space where they are expected to solve practical problems in society (Ravetz 1971). In other words, it is now an expectation that science can provide answers to those issues that are most prominent in the public or political eye. There are also fewer areas of life where previously inherited common-sense wisdom is valued more than the evidence provided by so-called scientific experts, and the assumption is growing “that every problem, personal and social as well as natural and technical, should be amenable to solution by the application of the appropriate science” (Ravetz 1971, 12).
This shift alters the stakeholders central to technology panics. For better or for worse, psychology – the science most closely related to child development and parenting – now plays an integral role in what I will call the Sisyphean cycle of technology panics, referencing the Greek myth of King Sisyphus. King Sisyphus was condemned by the Gods to roll a boulder up a steep hill in the underworld for eternity: every time he reaches the top, it rolls back down to the bottom, forcing him to walk back and start the cycle all over again. Similarly, psychological research on technology effects is in an intricate cycle of addressing societal worries about technologies. With every new technology treated as completely separate from any technology that came before (Wartella and Reeves 1985), psychological researchers routinely address the same questions; they roll their boulder up the hill, investing effort, time and money to understand their technology’s implications, only for it to roll down again when a novel technology is introduced. Psychology is trapped in this cycle because the fabric of moral panics has become inherently interwoven with the needs of politics, society and the scientific discipline (Grimes, Anderson, and Bergen 2008). I will outline the nature of this involvement at different stages of the Sisyphean cycle of technology panics below:
1.2.1 I. Panic Creation
Technology panics are a recurring feature of the societal landscape because of how society reacts to technological developments. The predominant approach to technological reform is technological determinism: the idea that (1) the technologies used by a society form basic and fundamental conditions that affect all areas of existence and (2) that when such technologies are innovated, these developments are the single most important driver of changes in said society (Leonardi 2012). Technology is therefore seen as a foundation for and agent of change, while society itself is assumed to have little power to influence the technologies themselves. “Utopian and dystopian views assume that technologies [therefore] possess intrinsic powers that affect all people in all situations the same way” (Boyd 2014, 15). When the internet became increasingly popular, for example, most analyses either took a utopian or dystopian point of view (Wellman 2004; Livingstone, Mascheroni, and Staksrud 2018).
Technological developments are rapidly linked to ongoing and complicated societal changes (Grimes, Anderson, and Bergen 2008, 50), an example of which is the current concern that social media is causing observed decreases in teenage mental health (Twenge 2018). What makes arguments based on technological determinism even more powerful is that they are difficult to deny. Critics are told that society will only truly understand the impact of a certain technology when a longer time-frame is available to be examined (Leonardi 2012). Technological determinism is therefore a widespread assumption which allows panics to arise quickly, by linking technological developments to current societal changes that concern the population.
1.2.2 II. Political Outsourcing to Science
Certain political benefits arise from societal panics (Garland 2008). Politicians and policy-makers routinely embrace them as an opportunity to demonstrate their willingness to stand up to emerging technology companies and their deep concern for children and other vulnerable populations (Grimes, Anderson, and Bergen 2008). In 1954, for example, Senator Estes Kefauver headed a US House subcommittee investigation into the original Superman series, making a name for himself and ultimately running for president (Grimes, Anderson, and Bergen 2008). “Then, as now, few could resist or would deny the political dynamic fuelled by the headline potential of being opposed to violence, a champion of children, and tough on a regulated industry” (Ramey 1994). Technological panics also enable politicians and society to “deflect social reform from the much more difficult issues of racial justice, economic opportunity and educational quality” (Anderson 2008). They are therefore a welcome vehicle for steering public attention away from intractable and uncomfortable issues.
While it is in the political interest to be seen to address technology panics, it has now become common to outsource the process of finding solutions to scientific research. In the last decades, politics and parenting have increasingly turned to science as a guide for addressing difficult questions. Society now treats “scientists as experts, whose opinions are regularly sought on matters of importance and for most part accepted without question.” (Okasha 2002, 121). This gives science a place in society where its specialised knowledge is used to calm fears and concerns in the general population, providing “comfort and reassurance in the face of the crucial uncertainties of the world” (Okasha 2002; Ravetz 1971, 386). Any new societal concern is not just a political event, but also a challenge to the relevant science (Ravetz 1971). Outsourcing the technological panic to science by funding, commissioning and referencing research, therefore allows politicians and policy-makers to calm and reassure the population, putting the onus on academics to provide a sense of security through the production of tailored research.
On the one hand, political outsourcing is a positive for psychology: researching a technology implicated in a panic promises funding, prestige and other outcomes aligned with current scientific incentives. In the industrialised age of science, where researchers work in more precarious positions and need to find research funding to support their existence, policy’s decision to fund science addressing the technological panic at hand promotes work in the area (Rubenstein 1982; Ravetz 1971). Furthermore, there is an attraction to investigating something society is inherently interested in, with the hope of ultimately helping vulnerable populations.
Psychologists therefore take the position of providing the public with research into a societal concern, rooting their work in the solution of a practical issue. This is best illustrated by researchers’ introductions to their work which centrally acknowledge the public concerns (Wartella and Reeves 1985). A 1937 paper considering the radio included the following introductory sentence: “The fact that parents and teachers have persistently complained about the quality of the radio entertainment provided for children makes the need for research in this field the more urgent” (DeBoer 1937) and similar introductions will be seen at the beginning of my thesis’ chapters. There is therefore a scientific shift in the area towards addressing a societal problem, rather than furthering a universal scientific theory or widely accepted research thread. Some argue that this causes additional problems. They believe that for such a scientific area to self-sustain, it now needs to uphold an aura of crisis, and keep up the face of science being necessary to cure the problem at hand (Grimes, Anderson, and Bergen 2008). By providing money, attention and prestige, technology panics therefore cater to the needs of the psychological discipline, which has to sustain itself during the 21st century. This builds a network of dependence between the public, policy and academia in the face of technological concern.
1.2.3 III. Wheel Reinvention
Yet there are also clear negatives when science becomes the provider of evidence to technological concerns. Researchers have noted the distinct similarity of research conducted to address different technological panics (Wartella and Reeves 1985). This realisation is not new; even in 1951 researchers were complaining that for each new technological concern “we seem to go through the same stages” (Seagoe 1951). Questions about addiction to relevant technologies have been raised for radio (Preston 1941), comic books (Wertham 1954), television (Lowery and DeFleur 1988), video games (Bushman and Anderson 2002) and social media (Twenge 2018). Similarly, questions about social connection, concentration and empathy are routinely found when reading psychological commentary on a range of new technologies (Greenfield 2014). It is therefore routinely overlooked that each novel technology shares more similarities than differences with its predecessors – even though it might look completely new at first glance (Seagoe 1951). Video games, for example, were first treated as a medium distinct from television but are now commonly integrated into combined reviews of so-called screen time (Dickson et al. 2018).
This reinventing of the wheel is a symptom of an area built on the existence of a practical problem, rather than the existence of a universally accepted theoretical underpinning or research thread. While there are some common ideas about technologies like the displacement hypothesis (Przybylski and Weinstein 2017; Dienlin, Masur, and Trepte 2018), such approaches are commonly used to explain findings without having a more influential role in shaping or progressing the work. Without an underlying paradigm to guide research, each researcher is “forced to build his field anew from its foundations” (Kuhn 1962, 13). The psychological discipline examining technological panics is therefore in a Kuhnian pre-paradigmatic period of science (Kuhn 1962).
With each new technological panic, a new group of researchers begins investigating the practical problem at hand without an overarching theoretical framework or an underlying basis for scientific understanding. The pre-paradigmatic nature of the field is then “marked by frequent and deep debates over legitimate methods, problems, and standards of solution, though these serve rather to define schools than to produce agreement” (Kuhn 1962, 48). This conflict about methodology and scientific standards has been evident in the fraught tone of scientific debate about violence in video games (Elson and Ferguson 2014) and the current discussion about smartphone use (Ophir, Lipshits-Braziler, and Rosenberg 2019; Twenge 2019). The lack of an underlying theory lets different camps emerge that are in scientific disagreement with each other, leaving the quality of scientific output relatively uncontrolled (Kuhn 1962).
In pre-paradigmatic fields such as technological panics research, the wheel is therefore routinely reinvented and little progress is made due to the lack of theoretical anchors for scientific investigation and quality control. The research area struggles while the field’s “leaders and practitioners are exposed to the temptations of being accepted as consultants and experts for the rapid solution of urgent practical problems” (Ravetz 1971, 400). The Sisyphean cycle of technology panics therefore restricts the researchers to addressing practical problems and internal debate, rather than building a long-lasting theoretical understanding that can shape science in the long term.
1.2.4 IV. No Progress; New Panic
Psychology – the science routinely chosen to investigate the technological panic at hand – therefore has few of the tools necessary to produce evidence quickly and effectively. The funding provided by politicians and government for the scientific field helps calm the nerves of the population but will provide little evidence to inform policy interventions. When faced with the decision about how to react to the technology at hand, policy makers therefore need to choose between three options: (1) they engage in evidence-based policy, and as there is not enough evidence they do not implement substantial policy change, (2) they engage in policy-based evidence, selecting or highlighting only the evidence that adheres with the policy they aim to implement or (3) they do not consider scientific evidence at all and instead adhere to the precautionary principle, implementing restrictive technology policy due to the fear of potential harm.
Many of the vocal contributors to technology panics argue for the latter approach, urging policy makers to anticipate harm even when there is no evidence that it will occur. It reverses the onus of proof so that the actor needs to show that the activity is harmless instead of vice-versa (Kriebel et al. 2001). In the area of technology panics we are reminded of an editorial by Strasburger, which highlighted that a lack of evidence should not make paediatricians shy away from examining problematic television use (1989). Furthermore, Wertham provided a gripping analogy in his book: “Gardening consists largely in protecting plants from blight and weeds, and the same is true of attending to the growth of children. If a plant fails to grow properly because attacked by a pest, only a poor gardener would look for the cause in that plant alone. The good gardener will think immediately in terms of general precaution and spray the whole field. But with children we act like the bad gardener. We often fail to carry out elementary preventative measures, and we look for the causes in the individual child” (1954, 2).
While slow scientific progress makes policy interventions difficult, sooner or later the panic will subside due to the introduction of a new technology (Dennis 1998). From the historical perspective it is clear that once a new technology is introduced, interest and concern about older technologies decreases substantially (Wartella and Reeves 1985). Public and political attention turns to the new technology, restarting the Sisyphean cycle of technology panics, while leaving behind a debate that potentially contributed little to knowledge creation, and that will be largely forgotten so that the next generation of researchers can reinvent the wheel again.
1.3 Media Effects Research: The Common Approach
It is remarkable that research conducted during different technological panics proceeds in lockstep while lacking an overarching theoretical paradigm. This is explained by researchers’ universal adherence to the same basic structures of psychological thinking and argumentation. Like in social psychology, the guiding principle for most technology research is cognitivism (Wartella and Reeves 1985). The effects of technologies are not treated as pure stimulus-response relationships, but as influencers of abstract and measurable cognitive factors (e.g. violent tendencies, positive affect or attitudes). To measure the effects of technologies one can therefore examine abstract before and after measures of cognitive factors, an approach popular for almost 90 years (Thurstone 1929).
Cognitivism requires researchers to assume that cognitions, and the biological structures that underly them, are either malleable throughout life or during certain developmental periods (Grimes, Anderson, and Bergen 2008). The malleability of the brain has been shown to be substantial in both early childhood (DiPietro 2000) and adolescence (Fuhrmann, Knoll, and Blakemore 2015). Long-term malleability in later life has been evidenced with respect to brain injury (Fraser et al. 2002) but the extent of malleability varies with respect to age and damage (Fashad and Kolb 2010; Lenroot and Giedd 2006; Anderson et al. 2005). It is also not certain whether well-being or life satisfaction are malleable over time (Lucas and Donnellan 2007). While the general assumption of malleability is readily made, it is therefore not without criticism.
Research on the effect of an emergent technology normally has four elements:
- it focuses on a technology, content or medium,
- it defines a specific audience,
- it examines a behaviour due to the technology, content or medium, and
- it provides theory or methodology to generate evidence (Grimes, Anderson, and Bergen 2008).
When initially researching a new technology both a) and b) are often broadly defined as the technology as a whole (e.g. smartphones) and a broad audience (e.g. all children) respectively. The behaviour examined (c) is the outcome which is causing concern (e.g. violence or depression), while the theoretical basis (d) is cognitivism as discussed above. This generalisation directs the early literature towards a causationist standpoint: that all members of the audience considered are affected by the new technology in the same way and that this technology is sufficient to cause long-term change (Grimes, Anderson, and Bergen 2008). This approach is similar to the hypodermic needle model which presupposes that every medium consumed affects behaviour in a direct, identical and non-individual way (Wartella and Reeves 1985).
This causationist standpoint has had many incarnations in the last century: listening to the radio causes anxiety (Preston 1941), reading comic books causes childhood maladjustment (Wertham 1954), video games cause aggression (Bushman and Anderson 2002) and smartphones and other new media cause depression (Twenge et al. 2017; Twenge 2018). The argument is however difficult to uphold because of its assumptions (Himmelweit, Oppenheim, and Vince 1961). Firstly, if we take a social developmental approach, any slight differences in upbringing and socialisation should cause differences in cognitive structures (Grimes, Anderson, and Bergen 2008), something that has been widely recognised to be true in both psychology and neuroscience (Kolb and Gibb 2011). For a causationist standpoint to work, however, cognitive structures and development need to be identical across the population studied so that they are identically affected by the technology of interest (Grimes, Anderson, and Bergen 2008). Furthermore, the causationist standpoint assumes that the technology of interest has more influence on the outcome of interest than other aspects of a person’s life (Grimes, Anderson, and Bergen 2008). Both assumptions are difficult to support in the face of widespread scientific evidence that common individual differences in the environment can substantially change a person’s cognitive structures (Kolb and Gibb 2011; DiPietro 2000).
A natural progression for research in this area is therefore to examine more specific audience types. A report by Herzog published in 1941, for example, considered how age determines whether children are affected by radio (Herzog 1941; as cited in Wartella and Reeves 1985). The 1933, the Payne Fund studies investigated the effects of films and highlighted that children’s reactions depended on gender, age, past experience and predispositions (Wartella and Reeves 1985). In addition to individual differences, research often also begins to take into account the different contents that can be consumed using the same technology. Blumer and Hauser, for example, found that the effects of films on delinquency depended on the film shown, as well as on the observer (Blumer and Hauser 1970).
When research progresses to consider the cues and justifications for a technology’s use, it adopts a multi-process approach (Huesmann, Lagerspetz, and Eron 1984). This is especially the case once researchers start considering the bidirectionality of relationships between technology use and cognitive factors. Research has, for example, shown that predispositions predict whether someone chooses to watch violent television programmes or seek out violent video games, highlighting the potential for selective exposure effects (Atkin et al. 1979; Przybylski, Ryan, and Rigby 2009). The psychological and cognitivist approach to investigating a new technology therefore opens up a natural progression of scientific thinking. Research starts by taking a causationist standpoint but then moves on to more complex, multi-dimensional and bidirectional approaches.
1.4 Teens and Screens
In this thesis I will focus on the technology panic that is currently most prominent in the western world. Concerns about digital technology use and its potential negative effects on empathy, social interaction, attention, well-being and many other important outcomes have been growing in the last decade (Greenfield 2014). While the preceding moral panic about video games is still ongoing (see Elson and Ferguson 2014), increased use of diverse screens connected to the internet has encouraged a more generalised worry about their effects on vulnerable populations (Bell, Bishop, and Przybylski 2015). The shift from a fixation on a particular technology to a more general focus on ‘screen time’, can be partly explained by our increasing inability to differentiate between various forms of screen activities. Digital devices such as smartphones, tablets or laptops can now support increasingly diverse activities and contents ranging from radio and television, to gaming, reading and social media. As written by Grimes and colleagues: “In this first decade of the 21st century, we are witnessing the final erasure of the medium (sic) as convergence makes the distinction of motion pictures, television, and other media merely an academic exercise” (2008, 41). It is therefore increasingly difficult to determine the effect of any single technology. Using a multitude of devices, 99% of UK twelve to fifteen-year-olds now go online, at an average of 20.5 hours a week (Ofcom 2019). In America, 45% of teenagers report they are online “almost constantly” through their use of many different devices (Pew Research Centre 2018). This makes ‘screen time’ a helpful umbrella term when voicing concerns about an increasingly digital childhood. It must, however, be noted that examining time spent on digital devices is an intrinsically bad measurement of their effects (Orben, Etchells, and Przybylski 2018; Orben and Przybylski 2019a, Chapter 3).
Since the start of my doctoral training, social media use and smartphones have become an additional focal point of technology concerns. Smartphones and tablets allow mobile and continuous engagement with digital technologies and the internet. In the United States, 95% of teenagers (aged thirteen to seventeen) report having access to or owning a smartphone (Pew Research Centre 2018). In 2018, 83% of British twelve to fifteen-year-olds owned a smartphone, and 50% owned a tablet (Ofcom 2019). Social media is often hosted on either of the two devices and encompasses “sites and services that emerged during the early 2000s, including social network sites, video sharing sites, blogging and microblogging platforms, and related tools that allow participants to create and share their own content” (Boyd 2014, 6). Social media is different compared to other communication methods as it is non-directed and does not need to be topically orientated. Previous internet platforms often focused on specific topics, but social media sites downplayed the necessity of having a common topic and instead made “friendship the organizing tenant of the genre” (Boyd 2014, 7).
Social media completes the erasure of the medium as it is inherently diverse and ever-changing: its content is highly individualised and can differ from person-to-person on an hour-by-hour basis. This uncontrollability poses a challenge to parents and policymakers: “Both the efficacy of regulation (of content) and governmentality (of use), the traditional comfort blankets in relation to television, are challenged by the internet because children and young people are theoretically free to roam anywhere in the places and spaces of the cyber.” (Brown 2005, 151; France 2007).
The diversity of social media, and its inherently social nature, makes it attractive to younger generations. In the UK, 69% of twelve to fifteen-year-olds have a social media profile (Ofcom 2019). Asked about its effect on their lives, 31% of American adolescents believe it is “mostly positive”, in comparison to 24% saying it is “mostly negative” (Pew Research Centre 2018). One of the most appreciated aspects of social media use is the ability to connect and communicate with friends and family (Pew Research Centre 2018), something that has been a key part of adolescent life for decades (Blakemore and Mills 2014). Social media creates so-called “networked publics”, spaces in society where adolescents can form social relationships and spend time together (Boyd 2014). “Teens engage with networked publics for the same reason they have always relished publics; they want to be a part of the broader world by connecting with other people and having the freedom of mobility” (Boyd 2014, 10). As adolescents are less able to access public places, because they are spending more time at home, the ability to interact in such networked publics is an opportunity for them to create their own social environment in the 21st century (Boyd 2014). What makes social media concerning is therefore its sheer diversity and uncontrollability. This emergent technology provides opportunities for adolescents to construct places that allow them to socialise and connect, away from the eyes and ears of their parents and even when their use of physical public spaces is restricted.
Most concerns about digital technologies and social media focus on adolescents, the earliest extensive adopters of these innovations. While adolescence used to be considered the age between puberty and marriage/parenthood, the endpoints are less clear cut today. The end of adolescence is now often linked to adult responsibilities (Patton et al. 2016) or a reduction in dependency on others (Griffiths 1996). There is however no universally accepted age range or traits of adolescence. The World Health Organisation defines adolescents as ten to nineteen-year-olds, with early adolescence being between ten and fourteen years of age and late adolescence being between fifteen and nineteen years of age (Patton et al. 2016). The lack of clear age limits stands in contrast to the distinct role teenagers take on in society. Adolescents are often viewed as having distinct attributes not found in children or adults (Seagoe 1951). But what makes an adolescent an adolescent is still relatively unclear. Historical, sociological and psychological approaches provide important insights into this stage of life.
188.8.131.52 Historical Perspective
Society’s stance towards adolescents has undergone considerable fluctuations over time. A study of child portraits concluded that in the 17th century adolescents were seen as small versions of adults (Ariès 1960). Researchers have however criticised this conclusion as portraiture was only available to the most privileged and the study can therefore not generalise to other subsets of the population (France 2007; Griffiths 1996). Furthermore, adolescents seem to have been perceived as a distinct ‘group’ for multiple centuries. Aristotle already remarked how “youth is the age when people are most devoted to their friends” and are also “lacking in sexual self-restraint, fickle in their desires, passionate and impulsive” (Blakemore 2018a).
Throughout the 18th and 19th century, state involvement and general concern about the adolescent age group increased; concerns specifically about leisure time started to appear in the 19th and 20th century (France 2007). Adolescents were then seen as an age group that needed shielding, even though the amount of safeguarding considered necessary fluctuated over time. In the 1960s and 1970s the arrival of youth countercultures made adolescents take a more prominent place in society as consumers, while there was also a push towards more youth education and culture (Livingstone 2009). The last half century then saw more products and technologies being marketed specifically at the adolescent age group – a development still ongoing today. Looking back, it becomes evident that our definition of adolescence, and adolescents’ place in our society, is still in continuous flux.
184.108.40.206 Sociological Perspective
The sociological idea of increasing individualisation in society is a concept worthy of further attention (Beck and Beck-Gernsheim 2002), especially when considering current concerns about adolescent well-being trajectories. The last decades have been marked by this societal development: people want increasing control over their own lives. This need for individualisation is also reflected in psychological approaches like Self-Determination Theory (Ryan and Deci 2000). With individualisation progressing in society, individual self-fulfilment is aspired to at the highest level and the previous social orders based around the traditional family, nation states, class and ethnicity, decline in importance (Beck and Beck-Gernsheim 2002, 22). This is said to affect the youngest in society. They are put under pressure to aspire to a ‘biographical project’, where they plan their lives and wilfully navigate career and lifestyle choices (Beck and Beck-Gernsheim 2002, 60–61). Failures to deal with the increasing pressure or expectation to build a biographical narrative are attributed to a failure of the self. Furthermore, problems in society and “social crises appear as individual and are no longer – or only very indirectly – perceived in their social dimension”, making people responsible for their own failures (Beck and Beck-Gernsheim 2002, 24). While such a sociological change is difficult to study from a psychological perspective, it is important to consider when discussing current trends and determinants of adolescent well-being.
220.127.116.11 Psychological Perspective
Some have argued that considering adolescence a separate life stage is arbitrary (Seagoe 1951). There is, however, an increasing consensus in the psychological discipline that adolescence is a “distinct period of biological, psychological and social development” (Blakemore 2019). There are three pillars of evidence: 1. There exist behaviours, not dependent on culture, that set adolescents apart from other age groups, 2. Such adolescent-type behaviours are found across species, and 3. Such adolescent-type behaviours have been documented throughout history (Blakemore and Mills 2014; Blakemore 2019). The activity most commonly associated with adolescence is social risk taking (Steinberg 2004). While risk taking might seem inherently negative, it could be evolutionarily beneficial for adolescents maturing in an unstable and changing world (Blakemore 2018b). Furthermore, adolescents are very attuned to their social peer groups (Boyd 2014) and especially affected by social rejection (Blakemore and Mills 2014; Blakemore 2019, 2018b). This again can be seen as something beneficial for their maturation, while it does exacerbate problems and behaviours seen as unique to that age group.
In addition to psychological changes in cognition and behaviour, there are biological developments that make adolescence a unique window of cognitive maturation (Fuhrmann, Knoll, and Blakemore 2015). Such periods of biological change give reason to believe that adolescence is a distinct stage of development. During adolescence, the brain is more susceptible to social signals, when compared to the brains of adults or children (Blakemore and Mills 2014). Furthermore, the brain is undergoing long-term changes in grey and white matter, axonal myelination and synaptic pruning (Blakemore 2019). Not only in the brain, but around the body, changes in stress (i.e. glucocorticoids) and sex hormones (e.g. testosterone), controlled either through gene expression, physical changes or epigenetic mechanisms, play a key role in “sculpting the adolescent brain” (Andersen and Teicher 2008). This increased malleability is important when theorising about potential technology effects (Grimes, Anderson, and Bergen 2008), and it supports the view that adolescence is a distinct developmental period worthy of psychological attention.
Humans are motivated to live a happy life, and therefore decreases in the well-being of certain populations attract societal concern. Well-being is not only integral for mental health, but also a key predictor of productivity (Bryson, Forth, and Stokes 2015) and general health outcomes (Steptoe, Deaton, and Stone 2015). Yet it is much more difficult to quantify than GDP or mortality rates. Adolescent well-being levels, or rather the current decrease of adolescent well-being levels, are however at the forefront of a current panic about digital technologies and social media (The Children’s Society 2018; Twenge 2018). This thesis will therefore focus on investigating adolescents’ subjective well-being, defined as an “overall evaluation of the quality of a person’s life from her or his own perspective” (Diener, Lucas, and Oishi 2018). In this section I will discuss the definition of subjective well-being and how it can be measured. I will examine its stability over the life course and whether we need to consider some of its distinct qualities when studying adolescents.
While there exists an inherent difficulty in defining what exactly well-being is (Kesebir and Diener 2008), subjective well-being is most often separated into four components (Kesebir and Diener 2008; Diener 2000):
- Life satisfaction: global judgements about one’s life
- Satisfaction with certain domains: e.g. satisfaction with friends, family or school (Huebner 2004)
- Positive affect: pleasant moods and experiences
- Negative affect: negative moods and experiences
The measurement’s subjective nature limits the scope of the construct: it probes quality of life from an individual’s perspective, instead of trying to obtain a more general overview (Huebner 2004). The measure should therefore not be assumed to quantify overall objective well-being. Yet subjective well-being might be one of the best proxies available for measuring well-being, and its subjectivity could be a strength. “Different people likely weight different objective circumstances differently depending on their goals, their values, and even their culture. Presumably, subjective evaluations of quality of life reflect these idiosyncratic reactions to objective life circumstances in ways that alternative approaches (such as the objective-list approach) cannot” (Diener, Lucas, and Oishi 2018).
Subjective well-being is separable from other conceptualisations of well-being prominent in the literature. Eudaimonic well-being, for example, assumes well-being is determined by having a sense of purpose and regular positive contact with others (Ryan and Deci 2001). This form of well-being might contribute to subjective well-being measurements. For example, meaningful relationships could benefit subjective well-being (Rohrer and Lucas 2018). However, eudaimonic well-being also falls outside of the subjective well-being conglomerate: good relationships might not be necessary for every person’s happiness.
For people’s subjective judgements about well-being to be valid, the following needs to be true: 1. All experiences people have should additively sum to make some real global well-being entity, 2. These feelings of global well-being should be relatively stable and 3. People need to be able to summarise and report these feelings in an accurate manner (Campbell 1981). Whether the third point is true is still relatively unclear. There are both top-down and bottom-up theories trying to describe how people arrive at subjective well-being judgements. Top-down theories argue that people view their life in a certain way and this judgement determines their responses to subjective well-being questionnaires. The bottom-up view argues that people aggregate their evaluations and experiences of many diverse aspects of their lives to derive their general subjective well-being judgement (Diener, Lucas, and Oishi 2018).
In parallel to this distinction, it is unclear whether respondents fill out questionnaires ‘experientially’ (i.e. considering how they feel now) or ‘evaluatively’ (i.e. considering how they think they have been feeling retrospectively). At times, judgements are made too rapidly to be the result of a full memory search, hinting that responses might be heavily dependent on experiential rather than evaluative thinking (Diener, Lucas, and Oishi 2018). Yet if people fill out questionnaires experientially, we would expect short-term changes in the environment that affect mood to also change their well-being ratings. Yet studies have shown that mood’s effect on subjective well-being judgements is minor and inconsistent (Eid and Diener 2004). Furthermore, a study of one million US residents’ life satisfaction found that the weather on the day of questionnaire completion did not substantially affect well-being judgements (Lucas and Lawless 2013). As mood and day-to-day circumstances fail to pose major threats to the reliability and validity of well-being and life-satisfaction measurements (Diener, Lucas, and Oishi 2018; Huebner 2004), participants are probably using a mixture of experiential and evaluative techniques to answer the questions.
Whatever way participants approach their completion, there is sufficient evidence showing that subjective well-being questionnaires are valid by traditional psychometric baselines (Diener, Lucas, and Oishi 2018; Diener et al. 1985). While participants’ scores can be slightly influenced by socially desirable responding and emotional states, these influences can be discounted as non-detrimental to the measurement procedure (Diener 2000). It is however still unclear how the many different measures of subjective well-being relate to each other. The measures range from quantifying life satisfaction in various domains (Knies 2017; University of Essex, Institute for Social and Economic Research 2018) to asking about mood either on the day or over a specific past timeframe (University of Michigan. Survey Research Center 2018). As there is still little consolidation of measures, researchers expert in the area recommend that “when possible, researchers should include a broad array of measures, including both judgment-focused measures like life satisfaction and more affective measures.” (Diener, Lucas, and Oishi 2018). I therefore include diverse and partially overlapping scales of subjective well-being in the work completed for this thesis.
18.104.22.168 Measurement Stability
A question central to this thesis is whether subjective well-being can be effectively changed or whether it is stable throughout life. Only if well-being is malleable can media significantly affect it (Grimes, Anderson, and Bergen 2008). While there has been much disagreement about the stability of well-being, the conflicting evidence hints of the existence of a small proportion of well-being that can be changed over time.
22.214.171.124.1 Happiness pie hypothesis.
The disagreement about stability is at its clearest when comparing the positive psychology movement to the hedonic treadmill hypothesis. Positive psychology, a popular area of psychology for over two decades, stresses that small changes to how people live their lives can increase long-term happiness (Seligman 2002). Researchers have floated ideas like the Happiness Pie Hypothesis: that approximately 50% of individual differences in happiness are due to genetics, 10% are dependent on life circumstances and the remaining 40% can be influenced by the activities people decide to do in their lives (Lyubomirsky, King, and Diener 2005). This idea of 10% of happiness being controlled by your circumstances and 40% by your volition became very popular, as it fit “into a pervasive contemporary ‘feel-good’ narrative in which self-improvement is claimed to be not only desirable, but also eminently achievable” (Brown and Rohrer 2019). But this is an exaggeration. Researchers have criticised the Happiness Pie Hypothesis for confusing within- and between-subject variance, conflating demographic factors and life circumstances, failing to include a necessary error term and not accounting for important interactions and the uncertainty around how much genes determine happiness (Brown and Rohrer 2019). The Happiness Pie Hypothesis therefore overestimated the percentage of well-being that is susceptible to change through lifestyle choices by making multiple errors. While the hypothesis might be an exaggeration, however, one should note that some researchers have simulated agent-based models showing that even a small percentage of happiness malleability can cause significant differences in the happiness levels of monozygotic twins (Nes and Røysamb 2017).
The area of positive psychology, and its message that any person is the maker of their own happiness, has been increasingly criticised as well. Some believe that the narrative about real-world pursuits increasing happiness are money-making ideals lacking scientific evidence, which cause great opportunity costs for individuals, science and society (Brown and Rohrer 2019; Moreau, Macnamara, and Hambrick 2019). Moreau and colleagues stress that “overemphasizing the role of environmental factors in success may lead to failure being stigmatized, despite the fact that individual differences in many real-world endeavours may in part reflect factors that are not under people’s control” (2019).
126.96.36.199.2 Hedonic treadmill hypothesis.
In contrast to the Happiness Pie Hypothesis, other groups of researchers have long argued that happiness might not be as malleable as often proclaimed. The Hedonic Treadmill Hypothesis suggests that people return to their baseline level of happiness after significant life events (Diener 2000; Brickman and Campbell 1971). The original authors of this theory found that while new paraplegics and lottery winners exhibited changes in happiness, they returned back to baseline levels in the long term (Brickman and Campbell 1971). There has been much support for this theory (Lykken and Tellegen 1996), but it has been supplemented to better address more recent research findings. The Dynamic Equilibrium Model, for example, now includes personality as a predictor (Diener 2000). Furthermore, people are now known not to fully return back to their original baseline of subjective well-being after an extreme life event like the loss of a spouse (Diener and Oishi 2005). Yet it is still agreed that, while life events or daily activities perturb well-being, they oftentimes have little long-term impact.
This conclusion is supported by research using a variety of methods to quantify the stability of happiness levels. It has been suggested that up to 50% of variance in life satisfaction is stable (Lykken and Tellegen 1996). A more recent study, however, has adjusted this estimate using the STARTS method, finding that the stable trait component of subject well-being is about 34-38%, while the autoregressive component is about 29-34% (Lucas and Donnellan 2007). The relatively high stability of well-being and life satisfaction could be driven by genetics (Nes and Røysamb 2017). Three meta-analyses found that the heritability of overall happiness measures is about 32-41% (Nes and Røysamb 2017). Taking what we now know about well-being’s stability into account, one could conclude that some proportions of happiness can be changed by life events and recent actions, while over a third is purely stable and outside technology effects’ reach. This supports the theoretical validity of hypotheses about activities like technology use affecting small proportions of well-being.
To understand how technology might be affecting well-being outcomes, it is important to understand what activities predict increases or decreases in well-being. Social variables seem to play a large part in predicting subjective-well-being, illustrating that happiness is not an individual pursuit (Diener and Oishi 2005; Bradburn 1969). This has been reflected in seminal theories of social psychology that highlight the ‘need to belong’ as fundamental to human happiness (Ryan and Deci 2000; Baumeister and Leary 1995). Recent pre-registered and large-scale research found that those who aim to socialise to increase their well-being are most successful in reaching their goal (Rohrer and Lucas 2018). Furthermore, other research has highlighted how social connection is a core aspect of human life, both in modernity and throughout our evolutionary past (Dunbar 2018). The substantial influence of social variables on subjective well-being is found across cultures: research using the World Values Survey has highlighted that social capital is a key predictor of happiness around the world (Helliwell and Putnam 2004). When evaluating this research, we however need to keep in mind that social variables are easier to change than, for example, health variables. This could partially explain their high predictive power (Rohrer and Lucas 2018).
In addition to social circumstance, a variety of other factors predict subjective well-being. High levels of income, strong relationships and religiosity are successful predictors (Diener, Lucas, and Oishi 2018). Surprisingly, education and gender are less predictive – at least in adult populations (Diener, Lucas, and Oishi 2018). The high stability of subjective well-being and its moderate heritability also show that internal factors like personality can be predictors (Diener, Lucas, and Oishi 2018). While this is of no importance to this thesis, which focuses only on UK, US and Irish populations, well-being judgements’ variability between cultures should also be considered in cross-cultural work (Diener, Lucas, and Oishi 2018).
Lastly, it is important to note that well-being should not only be considered as an outcome – “happiness may also be functional” (Oishi, Diener, and Lucas 2007, Chapter 3). Not being completely happy can be a motivating factor for people to engage in activities that stimulate future achievement (Oishi, Diener, and Lucas 2007). Happiness can therefore also be a stimulus for action in our lives, rather than just being an outcome (Nes and Røysamb 2017). We will see the importance of this notion in Chapter 4 of this thesis.
188.8.131.52 Adolescent Well-Being
While adolescents are routinely perceived as being at their healthiest, a deterioration in well-being during this time of development can have long-lasting consequences on later quality of life (Patton et al. 2016; Blakemore 2019; Andersen and Teicher 2008). Previous studies have found that the stable portion of subjective well-being is much smaller in young people (Lucas and Donnellan 2007). Furthermore, while gender is not predictive of subjective well-being in adulthood, adolescent females are twice as likely to suffer from depression than adolescent males (Andersen and Teicher 2008). There is also a higher potential for predisposed adolescents to start exhibiting the symptoms of a psychiatric disorder that will extend into their adult life (Andersen and Teicher 2008). Adolescence is therefore also a distinct stage of life when considering well-being and mental health.
Social and biological factors could underly the qualitative differences in adolescent well-being. Socially, adolescents are known to be more receptive to social rejection (Blakemore and Mills 2014), especially as self-reported quality of friends is a key predictor of future adolescent mental health (Harmelen et al. 2017). Biologically, adolescent brains are undergoing key developmental changes and show a qualitatively different response to stress (Andersen and Teicher 2008; Eiland and Romeo 2013). Stress might also have a more negative effect on well-being because the adolescent hypothalamic-pituitary-adrenal axis is still maturing, and glucocorticoid development is not yet finalised (Romeo 2013; Eiland and Romeo 2013). Adolescence is therefore a very sensitive period for stress affecting future mental health (Fuhrmann, Knoll, and Blakemore 2015). These factors combine to make adolescent well-being a distinct and influential aspect of life that can determine important long-term outcomes.
1.4.4 Current Evidence
“There is, as yet, no scientific consensus on the impact of screen-based lifestyles on the mental health of young people” (Frith 2017). Yet there have been well over 80 systematic reviews and meta-analysis published that examine this link in a range of populations (Dickson et al. 2018). This number is bound to increase further, as the production of evidence in the area is still advancing at accelerating speeds. A review of these systematic reviews and meta-analyses found that they are of varying quality standards (Dickson et al. 2018). In this literature review of the link between digital technology use and well-being, I will therefore not consider those reviews ranked as having a medium or high risk of being biased (Dickson et al. 2018). If the reviews were not ranked by Dickson and colleagues, for example if they were published in 2019, I will also include them in this review. I will however exclude those reviews specifically focused on sexting, gaming, aggressive behaviour, internet addiction or those that only examined a specific sub-population (e.g. Rice et al. 2016; Wang et al. 2017; Mitrofan, Paul, and Spencer 2009). It is important to note that many reviews were published using the keyword ‘sedentary behaviour’ and have therefore been routinely overlooked by the psychological or communication sciences literature.
The many competing meta-analyses and reviews regarding social media and screen use rely predominantly on cross-sectional evidence, as this makes up the vast majority of evidence in the field (Dickson et al. 2018). The quality of evidence reviewed is therefore very low (Carson et al. 2016). My literature review will begin with a review of reviews about digital technology use and psychological outcomes. I will then proceed to examining reviews specifically about social media use, before summarising particularly high-quality studies in the area. I will also discuss potential improvements and current limitations of research addressing digital technologies and their link to well-being.
184.108.40.206 Systematic Reviews and Meta-Analyses: Digital Technologies
Systematic reviews in the field have routinely been confronted with a mixture of conflicting results. If averaged, these results provide evidence for a positive association between time spent using digital technologies (i.e. screen time) and depressive symptoms (Hoare et al. 2016). Reviews of studies on very young children found low to moderate quality evidence that TV use is linked to unfavourable outcomes (Poitras et al. 2017; LeBlanc et al. 2012). Systematic reviews examining older populations highlight that 1 in 8-12 studies find a null result, while the rest find a positive association between screen time and unfavourable psychological outcomes (Tremblay et al. 2011; Dennison, Sisson, and Morris 2016). The relation is however not exceedingly clear. Some systematic reviews noted that a link between screen time and depressive symptoms only exists in cross-sectional and not in longitudinal studies (Liu, Wu, and Yao 2016). In contrast, others find that it is the longitudinal studies that report a negative or null relation (Carson et al. 2016). To make sense of such conflicting reviews, the “very low” quality of research in the area must be taken into account (Carson et al. 2016; World Health Organisation 2019). The conflicting results highlight that the evidence is still too weak to promote a uniform interpretation of the effect of interest.
The evidence base for the link between screen time and self-esteem is even weaker (Hoare et al. 2016). Just like for depression, there are many mixed results and slightly more studies find negative results (Carson et al. 2016). There has however been a randomised control trial showing that limiting television use increased self-esteem, which has been used by many systematic reviews to argue for a link (Tremblay et al. 2011). But one high-quality study on a specific intervention cannot make up for the many low-quality studies in the area that find mixed evidence.
220.127.116.11 Different Uses
Different uses and utilisations of social media might therefore be important to consider in order to obtain a better understanding of social media effects (Burke, Marlow, and Lento 2010). One major distinction is that between active and passive use, with active use representing activities like chatting, messaging and liking while passive use including activities like browsing newsfeeds, profiles or scrolling through photos and news items (Ellison, Steinfield, and Lampe 2007). Researchers have hypothesised that active use increases social capital and connectedness, therefore positively affecting well-being, while passive use increases upward social comparisons and envy, in turn decreasing well-being (Verduyn et al. 2017). Studies have found that active use increases bonding social capital and decrease loneliness, while passive use doesn’t have such positive outcomes (Burke, Marlow, and Lento 2010). Experimental and experience sampling studies support this idea by highlighting that passive use decreases well-being, potentially by increasing envy (Verduyn et al. 2015). It is therefore important to differentiate between active and passive uses of social media. Yet it is important to note that the results are still not clear cut. A study of 10,557 Facebook users whose Facebook data were examined for three months prior to them filling out a questionnaire, found that active Facebook use did not influence well-being. Only direct communication with close friends and family was linked to positive results (Burke and Kraut 2016).
When considering different uses of social media, one also needs to examine the style of a user’s online self-presentation. A qualitative synthesis of 21 observational studies examining Facebook self-presentation and mental health outcomes found that inauthentic self-presentation was related to low self-esteem and high social anxiety. More authentic or positive self-presentation was associated with increased levels of self-esteem and social support (Twomey and O’Reilly 2017). A two-wave longitudinal study found that people who were more authentic on their profile reported higher positive affect and life satisfaction, and lower negative affect six months later (Reinecke and Trepte 2014). In addition to active and passive use, a person’s self-presentation might therefore be an important factor to consider in order to understand the link between social media use and well-being.
18.104.22.168 Finding Common Ground
While the research area is filled with conflicting findings based on cross-sectional evidence, there is however some common ground. Many studies and meta-analyses find a small negative association between social media use and well-being of about r = -0.15 to r = -0.10. Correlations and observed effects in this ballpark have been shown in meta-analytic studies considering anxiety and depressive outcomes (e.g. McCrae, Gettings, and Purssell 2017; Yoon et al. 2019; Vahedi and Zannella 2019; Hancock et al. 2019), but have also been found in longitudinal research (Kross et al. 2013; Frison and Eggermont 2017; Reinecke et al. 2018; Bayer et al. 2018) and experimental work (Allcott et al. 2019). As mentioned above, it is still unclear what such a range of effects can tell us about well-being and how it is affected by social media use. This is because there are a range of third factors that can influence both variables, and there have been sources of bias not addressed properly in a literature that is largely cross-sectional and exploratory.
The same kind of effect size has, however, also been found bidirectionally: for social media use decreasing well-being and well-being decreasing social media use (Wang et al. 2018). The importance of bidirectional effects is clearly evident, but the results remain unclear. An early group of experimental and correlational studies found that while disconnection drives the use of Facebook, connection results from Facebook use (Sheldon, Abad, and Hinsch 2011). This does not fall in line with those studies finding negative relations in both directions (Wang et al. 2018; Frison and Eggermont 2017; Aalbers et al. 2018), only in the direction of social media use decreasing well-being (Kross et al. 2013) or only in the direction of loneliness leading to Facebook use (Song et al. 2014). It is therefore clear that more work considering bidirectional effects needs to be completed before true effects become evident. To start finding common ground, research therefore needs to increase transparency, while doing more to interpret the size and importance of effects and highlight their bidirectionality.
22.214.171.124 Limitations and Future Improvements
The low quality and conflicting state of the literature highlights many areas of the research field that could be improved further. On the one hand there needs to be an increased focus on individual differences. This would be helped by the study of more diverse and rigorously recruited samples (Erfani and Abedin 2018). More studies should also account for factors like gender or age. While age is not a routine focus of studies (Hancock et al. 2019), gender has been shown to be a predictive factor in recent work (Frison and Eggermont 2016; Twenge et al. 2017). While the predominant line of reasoning is that girls are more negatively affected by social media than boys (Twenge, Martin, and Campbell 2018), one meta-analysis found that screen time is linked to higher depressive symptoms and lower self-esteem in boys, but not in girls (Tremblay et al. 2011), and other studies found screen time effects to be independent of gender (Hoare et al. 2016; Frison and Eggermont 2017). It is however important to note that social media and general screen time are two different activities and can therefore have different gender effects.
“Ultimately, our findings demonstrate the lack of a uniform overall ‘Facebook effect’ on individuals, and illustrate the need to build temporal and spatial components into future research on Facebook and the wider social media ecosystem.” (Bayer et al. 2018). It is therefore important to conduct more longitudinal work (Dickson et al. 2018; Carson et al. 2016, 2016; Frost and Rickwood 2017) with more diverse time frames (Bayer et al. 2018) ranging from short-term experience sampling (Aalbers et al. 2018) to long-term annual studies (Wang et al. 2018).
It also needs to be noted that there has been increasing discontent about the measurement practices used in the area. Researchers argue that there are now the psychometric tools available to move away from measuring self-reported screen time (Andrews et al. 2015; Wilcockson, Ellis, and Shaw 2018; Ellis 2019; Ellis et al. 2019), which is known to be a flawed measure of media effects (Scharkow 2016). Better measurement could lead to more exact and consistent results in the literature. More exact tracking would also allow different types of social media and technology use to be examined in more nuanced and diverse ways, distinguishing different activities and timings of use. This would support important processes of triangulation (Munafò and Davey Smith 2018). Furthermore, it would enable researchers to home in on possible non-linear dose-response relationships between technology use and psychological outcomes, which have been shown in previous work (Hoare et al. 2016, 2016; Przybylski and Weinstein 2017). This review therefore suggests multiple areas of potential improvements ranging from better measurement to more longitudinal work and a keener eye for individual differences.
126.96.36.199 Cultural Critiques
A variety of other criticisms have been voiced about research into digital technologies and their effects. Commentators have noted that blaming social media and digital technologies for societal concerns is easy to do, but can lead public conversation astray: “All too often, it is easier to focus on the technology than on the broader systemic issues that are at play because technical changes are easier to see” (Boyd 2014, 16). The broad popularity of smartphones makes them a clearly visible scapegoat, while increasing inequality, austerity and deficiencies in children mental health services are less easy to spot. Current fears about digital technologies can therefore be “dangerously misguided when we consider all the real, though not as sexy, issues that get pushed out of the headlines in favour of media fears. Poverty, family violence, child abuse and neglect, and the lack of quality education and health care are problems that merit public attention way before media culture” (Sternheimer 2003, 3). Yet it should be noted that while there having been rapid changes in societal life (e.g. austerity measures in the UK), such developments are much more long-term and gradual than technological developments. Furthermore, important aspects of a child’s life, like their risk of maltreatment, have been improving over the last decades (Degli Esposti, Humphreys, and Bowes 2018). While it is easier to pinpoint the blame for negative developments on technologies, however, technology panics are used as a political tool to 1. take attention away from other societal issues and 2. to allow political figures to profile themselves. The media also profits from them: news consumers feel more obliged to read or watch scaremongering coverage, because not doing so might be detrimental to their, or their children’s, safety (Altheide 2002). The resulting widespread media coverage then makes the concerns spread more widely throughout the public conversation.
Furthermore, technology panics might be overemphasising the effect of technologies, in the face of other life factors: “It often seems as though media researchers presume that people have none but a life in the media. People hang out, hook up, and live in face-to-face relationships every day” (Grimes, Anderson, and Bergen 2008, 69). Grimes and colleagues add that the resulting “findings are banal (is that so … people can learn and be influenced by the media?), a scientizing of the obvious. Its primary effect has been to advance disciplinary interests and individual careers. Its unintended consequence has been to deflect attention from the hard issues of social reform by promising a quick regulatory fix” (2008, 91)
1.5 An Improved Approach
Is the psychological study of digital technologies and social media predestined to be subsumed into the ever-continuing Sisyphean cycle of technology panics? Will the research outputs be forgotten and ignored once a new technology gains widespread societal traction? Parts of this introduction paint a bleak picture of the research area and its interplay with policy and the public. Some of the following chapters will also uncover further quality control issues stifling progress in the field. Yet there are multiple factors which make the study of digital technologies and social media an important and influential research area.
Firstly, we are only on the cusp of a digital technology revolution that will change society for decades to come. The potential for digital technology and social media to alter the most basic features of society easily surpasses that of previous innovations like comic books, radio and television. It is still unclear how such changes will affect us and future generations. This makes it especially important for scientists to determine how such technologies are already affecting society today: not just to increase our knowledge of what is to come, but also to help shape future policy approaches and technology design. Psychology therefore has the potential to play a crucial and important role in ensuring that the digital revolution will have a net positive effect on society.
Secondly, the last decade has seen psychology in a time of crisis and reform (Vazire 2018; Munafò et al. 2017). Discontent about past methods and scientific approaches (Simmons, Nelson, and Simonsohn 2011; Open Science Collaboration 2015) has led to the development of many ideas aimed at improving the discipline’s scientific processes. We are therefore at the beginning of a so-called credibility revolution (Vazire 2018). As the field of technology panics research is affected by quality control issues and lacks a methodological framework, current initiatives to improve the quality and efficiency of psychological research could provide an impetus for improvement. Better quality of evidence in the field would allow science to effectively address the technology questions of the future, while it would also limit the opportunity costs and financial burdens created by current technology panics.
Thirdly, science is now facing an ever-accelerating technological revolution that is making research on novel technologies a race against time (Valkenburg and Piotrowski 2017). Innovations to improve research efficiency could therefore provide an important boost to psychology’s ability to address those technological innovations introduced to society at accelerating speeds. As Lewis Carroll once wrote: “My dear, here we must run as fast as we can, just to stay in place. And if you want to get somewhere else you must run at least twice as fast as that” (Carroll 1871; as cited in Valkenburg and Piotrowski 2017).
Taking these three factors to heart, my thesis will develop and implement a new methodological framework to improve, accelerate and innovate the provision of scientific evidence during technology panics. Such work has the potential to better policy approaches and inform changes that will ultimately benefit society. While the methodological framework I introduce and develop during this thesis cannot halt the Sisyphean cycle of technology panics, it can improve the research output considerably. This would ensure that the public and policy are better informed about the effects technologies are having on the population. In total, my thesis will consider five potential innovations: speeding up research through openness, improving transparency to decrease biases, splitting exploratory and confirmatory research to guide policy and interpreting effect sizes to limit hyperbole.
1.5.1 Enabling Research Through Openness
Making data and code openly accessible has the potential to accelerate scientific progress, while providing an additional impetus to improve research quality. Accelerating scientific progress without compromising on quality is especially important in times of increasing technological innovation rates (Valkenburg and Piotrowski 2017). Since the publication of the 6th edition of the APA manual in 2009, researchers are required to share their data on request for up to five years after publication (American Psychological Association 2010). This is however not common practice in the field (Houtkoop et al. 2018; Ceci 1988). A reluctance to share research data has been linked to weaker evidence and more errors in statistical reporting (Wicherts, Bakker, and Molenaar 2011). When asked about the barriers to sharing data, 77% of researchers highlighted the fear of alternative analyses exposing invalid conclusions. Furthermore, 73% feared discoveries of errors in data (Houtkoop et al. 2018). While posing barriers to widespread adoption, these fears highlight how openness would benefit the scientific community by facilitating better quality control (Stark 2018). Naturally not all data can, or should, be shared: a discussion of ethical and privacy issues is however out of the scope of this thesis (see Gilmore, Kennedy, and Adolph 2018). But even when data need to be kept private, the provision of analysis code can still provide many benefits – exposing errors and allowing researchers to reuse work to speed up progress in the field (as noted in Orben and Przybylski 2019b). Openness would therefore allow research to progress at greater speeds, providing evidence faster and adapting quicker to changes in technologies.
1.5.2 Improving Transparency to Decrease Biases
Flexibility in how researchers analyse and report their data is an ingrained and substantive problem throughout science. Simmons and colleagues showed that “undisclosed flexibility in data collection and analysis allows presenting anything as significant”, even the finding that undergraduates get physically younger when listening to specific music (2011). The extent of analytical flexibility is best explained by the garden of forking paths analogy (Gelman and Loken 2014). Any researcher needs to make multiple decisions when analysing their data (e.g. what outliers to exclude, what control variables to add, etc.). When making these decisions while analysing their data, they can unconsciously or consciously choose those data analysis methods that lead them towards the result that they were expecting or hoping to find. Researchers wander through this garden of forking paths of analytical choices, but only report the one analytical pathway they took to obtain their final result. Looking back, they often believe this analytical path is the only one they would have taken all along. But the garden of forking paths shows that there are oftentimes thousands of equally respectable ways of analysing the data, which were not reported and might have provided very different results. The ability to choose the preferred analytical path, so-called researcher degrees of freedom, enables studies to find physically impossible results (Simmons, Nelson, and Simonsohn 2011). They also allow researchers to report (false-) positive results in their work (Bishop 2019). The power of researcher degrees of freedom has been compellingly demonstrated in recent crowdsourced analyses where multiple statistical teams analysed the same dataset to answer the same research question, producing a wide range of possible end-results (Silberzahn et al. 2018; see also MAPS Project 2019).
Researcher degrees of freedom are hugely influential and can increase the false positive rate in a discipline, especially when there are cognitive biases and widespread pressures to publish positive results (Wagenmakers et al. 2012; Bishop 2019). This has negative consequences for science as “research findings that do not replicate are worse than fairy tales; with fairy tales the reader is at least aware that the work is fictional” (Wagenmakers et al. 2012). In the last years, the problem of analytical flexibility has been increasingly recognised as substantial: previously “everyone knew it was wrong [to do flexible data analysis], but they thought it was wrong the way it is wrong to jaywalk [… but now we recognise] it was wrong the way it is wrong to rob a bank.” (Simmons, Nelson, and Simonsohn 2018). Researchers have therefore been advocating for more transparent disclosures of analytical pathways (Simmons, Nelson, and Simonsohn 2018; Stark 2018; Bishop 2019).
Preregistration and Registered Reports have been suggested as two possible ways to combat analytical flexibility (Wagenmakers et al. 2012; Chambers 2013, 2014; van’t Veer and Giner-Sorolla 2016). Preregistration entails registering the process of data analysis before accessing the data, meaning that the researcher decides on their path through the garden of forking paths before the data can bias their choice. Registered Reports further aim to remove publication bias, by providing peer review before data collection and giving a paper ‘in principle acceptance’ at a journal before the results are known (Chambers 2014). They have already been successfully applied to the research considering new media and technologies (Elson and Przybylski 2017). Such initiatives have shown the potential for transparent research to better inform policy, the public and academia. Transparency therefore has the potential to hugely benefit the provision of evidence about new technologies.
1.5.3 Splitting Exploratory and Confirmatory Research to Guide Policy
Preregistration and Registered Reports allow researchers to clearly distinguish between exploratory and confirmatory work (Wagenmakers et al. 2012; Elson and Przybylski 2017). The difference between exploratory and confirmatory evidence is important to highlight, especially when the evidence routinely informs policy decisions (House of Commons Science and Technology Select Committee 2019). When implementing conventional statistics, a hypothesis test should only be used once on a dataset; it is however extremely common that multiple analyses are run on data until a favourable result is found (Wagenmakers et al. 2012; Simmons, Nelson, and Simonsohn 2011). Pre-registration – specifying the analysis plan before the data are available to the researcher – avoids the Texas sharp shooter fallacy where the shooter fires his bullets before painting on the targets afterwards, therefore achieving so-called perfect performance even though this was clearly not the case. While some have criticised strictly confirmative approaches as limiting creativity (Scott 2013), others disagree and believe that differentiating exploratory and confirmatory work channels scientific creativity into more fruitful pursuits (Wagenmakers , Dutilh, and Sarafoglou 2018). Some go as far as arguing that policy should only be based on confirmatory research, while exploratory research should be solely used to inform future studies and theorising. As policy change can have a substantial impact on society, it is a compelling argument that only gold-standard confirmatory evidence should be used as its foundation.
1.5.4 Interpreting Effect Sizes to Limit Hyperbole
With many psychological fields becoming increasingly big-data driven (Lang 2013), new approaches for communicating effect sizes will also become important. Using large-scale datasets is often wrongly equated to doing better quality research (for discussion see Orben and Przybylski 2019b). The problems associated with large sample sizes are best explained by Trout’s 1994 metaphor: a study with too low power (a too small sample size) is like trying to catch fish with a too large meshed net, you will not catch many fish and if you do, the fish will be uncharacteristically large. If you increase your power and sample size, your smaller meshed net will catch more of the fish you want (Trout 1994). Taking this analogy one step further, however, it is also true that if your sample size becomes even larger, the higher power will provide you with such a tightly meshed net that you will also catch tiny correlational dirt and not just the fish you were looking for. Because larger datasets can ‘catch’ increasingly smaller effects (Orben and Lakens 2019; Meehl 1967), statistical significance ceases to be a good indicator of practical significance.
This problem has been evident in recent research where statistically significant results were reported, but they were so small they should have little to no effect on our actions in the real world (Kramer, Guillory, and Hancock 2014). To ensure that minute, but statistically significant, effects are not over-reported, researchers have suggested defining a Smallest Effect Size of Interest: the smallest possible effect that will be reported as practically ‘significant’ in a study (Lakens, Scheel, and Isager 2018). Defining such a value is however very difficult (Anvari and Lakens 2019) and depends on the perspective that one takes about what populations will be affected (Rose 2008). It is however increasingly clear that effective communication of effect sizes will become crucial for both academia and policy in times of increasingly large-scale data research.
1.6 Thesis Outline
This thesis will devise a new methodological framework for the study of teenagers and their use of digital technologies. To do so it will apply and develop multiple methodological and statistical innovations that have the potential to increase the quality, efficiency and transparency of research in the field. Taking into account the many areas of reform highlighted by psychology’s credibility revolution (Vazire 2018; Crüwell et al. 2018), it will outline an improved approach for current and future work on emergent technologies.
In Chapter 2, I focus on how bias in data analysis, p-hacking and exaggeration of effect sizes can be understood, estimated and their impact minimised in the research area. The chapter introduces Specification Curve Analysis as a way to add transparency and illustrate how the selection of different analytical choices can lead to the production of skewed results. Furthermore, I detail a method I devised called comparison specifications to put the effect sizes of important relations into a novel perspective. Together, the approaches make the inherent uncertainty surrounding research results more transparent, allowing academia, policy and the public to better judge results’ relative importance.
In Chapter 3, I focus on the measurement issues afflicting the quantification of digital technology effects by focusing on diversifying measurement. I analyse three datasets that include time-use diaries and retrospective self-report measures of digital technology use. I also use an explicit exploratory and confirmatory hypothesis testing framework to introduce a more confirmatory approach to the research area. This chapter therefore focuses on additional methodological innovations, while further developing the statistical approaches pioneered in Chapter 2.
Chapter 4 concludes the thesis by detailing one of the first high-quality and transparent longitudinal analyses considering the role of social media in the lives of adolescents. The research examines social media use and its long-term effect on life satisfaction using Random Intercepts Cross-Lagged Panel Modelling executed in a Specification Curve Analysis framework, allowing me to disentangle within- and between-person effects. In this study, I also consider gender differences for the first time. Just like the previous two chapters, the work is fully accessible online with open code for all the analyses reported. This should help accelerate error detection and future scientific progress.
The thesis therefore provides an overview of an improved methodological framework that can address some of the major problems holding back research in the area of digital technology use and its effect on adolescent well-being. It takes initial steps towards ultimately addressing parts of the Sisyphean cycle of technology panics, while future-proofing a research field which is already struggling to keep up with an increasingly accelerating and influential digital technology revolution.