Introduction

Screen-based technologies (i.e., television, tablets) have become ubiquitous in the home and are now fully woven into the fabric of modern family life (Livingstone & Blum-Ross, 2020). The relative ease of access and low cost of such technologies has led to screen-based media activities being among the most common features of recreational time in childhood (Loprinzi & Davis, 2016). The purported risk of too much screen time has been studied extensively in recent years. And, while the psychobiological risks of childhood screen exposure remain a nascent field of research (Paulus et al., 2019), one of the more widely studied and contentious questions is whether screen time has an effect on developmental outcomes in childhood (Browne, Thompson, & Madigan, 2020). Research evidence has more consistently supported the hypothesis that higher levels of screen time in preschoolers and school-aged children are associated with psychosocial problems – specifically externalizing and internalizing behaviors (Suchert, Hanewinkel, & Isensee, 2015). However, it has also been suggested that children with behavioral dysregulation receive more access to screens to manage problematic behavior (Radesky, Peacock-Chambers, Zuckerman, & Silverstein, 2016) Accordingly, calls have been issued to increase the methodological rigor in studies of screen time and psychosocial development, as the direction of association is not well established (Kostyrka-Allchorne, Cooper, & Simpson, 2017). The primary aim of this study therefore was to estimate the longitudinal bidirectional associations between screen time and externalizing and internalizing behaviors in a nationally representative prospective cohort of children at ages 3, 5, 7, and 9 years.

Although the mechanisms that link screen time and psychopathology remain unclear (Paulus et al., 2019), consistent reporting of their covariance in both cross-sectional and longitudinal models (Suchert et al., 2015) has warranted lengthy consideration of the place and role of screen exposure in developmental psychopathology. For example, early cross-sectional research showed a significant association between exceeding pediatric screen time guidelines and externalizing and internalizing difficulties among children (e.g., Hamer, Stamatakis, & Mishra, 2009; Tomopoulos et al., 2007). Longitudinal studies have also been widely conducted and report prospective associations between screen time and later externalizing and internalizing behaviors. Such associations have been shown in cohorts across the developmental periods of early and middle childhood (e.g., McArthur, Browne, Tough, & Madigan, 2020). Though, it is also noteworthy that some studies report nonsignificant trends (Verlinden et al., 2012). Other indications of complexity in the longitudinal data can be found in studies that show differences in prospective associations for externalizing and internalizing behaviors or in studies that report differences between males and females (Carson et al., 2016).

While systematic reviews have concluded that there is strong evidence to claim that higher levels of screen time are associated with externalizing and internalizing problems during childhood (Carson et al., 2016; Suchert et al., 2015), close inspection of the evidence base reveals at least three outstanding issues. First, in addition to the possibility of nonlinear effects, further conjectures have been made about the effect sizes representing the associations between screen time and psychosocial outcomes (see, e.g., Przybylski, Orben, & Weinstein, 2020). Second, and directly related to such conjectures, counterarguments have been made about the failure of researchers to more fundamentally address the question of children’s differential susceptibility to screen exposure. As Browne et al., (2020) claim, the overemphasis on sample-level mean effects obscures the clinical reality that some children will be more affected by screen-based media exposure than others. Finally, despite the fact that some research is based on large samples of children, there is overwhelming consensus in the field that the evidence base is still far too reliant on cross-sectional designs, which considerably limits understanding of directionality of associations (Carson et al., 2016).

This latter point is critical because, while it is tacitly assumed that screen time is directionally associated with externalizing and internalizing behaviors, research also shows that children exhibiting higher levels of psychosocial difficulties receive longer durations of screen time (Radesky et al., 2016). The precise context and manner in which children ‘receive’ longer durations of screen time is still a relatively open question; however, there is a growing body of qualitative evidence that reports an increase in the instrumental use of screen time by parents, and that managing access to screens is perceived by parents to be a valuable tool and coping device for modulating children’s challenging behavior (e.g., Jago et al., 2016).

Research that can discern the effects of such temporally stable contextual factors from estimates about the longitudinal bidirectional associations between screen time and young children’s achievement of developmental milestones has recently been conducted (Madigan, Browne, Racine, Mori, & Tough, 2019). However, such longitudinal bidirectional associations between screen time and externalizing and internalizing behaviors have yet to be tested across early and middle childhood. Thus, this study derives longitudinal bidirectional associations using random-intercepts cross-lagged panel models (RI-CLPMs) (Hamaker, Kuiper, & Grasman, 2015), which is a structural equation-based approach to modeling multivariate multilevel panel data that (a) controls for temporally stable (i.e., ‘time-invariant’ or ‘trait-like’) differences between children and thereby; (b) more precisely estimates individual child (i.e., ‘time-varying’) trajectories representing how screen time and externalizing and internalizing behaviors are linked over time. As they may have distinct mechanisms (Marceau et al., 2015), two separate RI-CLPMs were used for estimating the longitudinal bidirectional associations between screen time and externalizing and internalizing behaviors (Figures 1 and 2, respectively). In the random-intercepts components of the models, we hypothesized that there would be significant variances for, and significant covariances between, screen time and externalizing and internalizing behaviors – thereby signaling trait-like differences between children. Since these longitudinal bidirectional associations have yet to be studied prospectively across childhood with statistical models that properly account for temporally stable, or ‘unobserved’, heterogeneity (Mulder & Hamaker, 2020), our expectations for the cross-lagged components of the models were exploratory in nature. In addition to our primary analyses, we also estimate the extent to which the bidirectional associations were moderated by contextual factors (child sex, nationality, and child-care arrangements, as well as family size, caregiver marital status, stress, and socioeconomic status [SES]). Our expectations about the moderating effect of such confounders were also exploratory in nature.

Random-intercepts cross-lagged panel model illustrating the bidirectional longitudinal associations between daily screen time and externalizing behaviors between the ages of 3 and 9 years, controlling for stable between-child differences. Covariances, autoregressive, and cross-lagged paths represent standardized parameters. Solid black lines represent estimates for which the 95% CI does not overlap zero (p < .05). aPathways constrained to a factor loading of 1 to isolate the time-invariant, between child, differences in externalizing behaviors and daily screen time

Random-intercepts cross-lagged panel model illustrating the bidirectional longitudinal associations between daily screen time and internalizing behaviors between the ages of 3 and 9 years, controlling for stable between-child differences. Covariances, autoregressive, and cross-lagged paths represent standardized parameters. Solid black lines represent estimates for which the 95% CI does not overlap zero (p < .05). aPathways constrained to a factor loading of 1 to isolate the time-invariant, between child, differences in internalizing behaviors and daily screen time

Despite the exploratory focus, cross-lagged parameters estimated from RI-CLPM represent the unique amount of change in an outcome variable that is explained by an independent predictor variable at a previous time point (Schuurman, Ferrer, de Boer-Sonnenschein, & Hamaker, 2016). Therefore, in this study, we can establish with greater precision how screen time and externalizing and internalizing behaviors are potentially causally linked over time. The findings of this study will therefore be very timely for clinicians and practitioners engaged in family media planning who seek to identify critical junctures across the developmental trajectory at which to prioritize preventive interventions.

Methods Study design and population

Data are from the Growing Up in Ireland Infant Cohort, a nationally representative sample of 11194 children born in Ireland between December 1, 2007, and June 30, 2008. The Growing Up in Ireland project was administered by trained Officers of Statistics from the Irish Central Statistics Office, and recruitment and data collection were carried out using face-to-face household interviews in a manner akin to the approach used for the Census of the Population. Families were recruited when the child was 9 months old and followed up thereafter at two-year intervals. The constructs of interest for this study are from waves two to five, collected between December 2010 and February 2018, when children were ages 3, 5, 7, and 9. Socioeconomic-, household-, parent-, and child-level characteristics are summarized in Tables S1 and S2. Ethical approval for the project was overseen by the Irish Department of Children and Youth Affairs. Caregivers provided informed consent and received no financial incentive for study participation.

Psychosocial development

Caregivers reported on their child’s psychosocial development at each wave by completing the Strengths and Difficulties Questionnaire (SDQ) (Goodman & Scott, 1999). The SDQ has been shown to be as effective as semistructured interviewing for screening externalizing and internalizing behaviors and has been validated across national contexts (Goodman & Scott, 1999). The SDQ screens externalizing and internalizing symptoms using 20 items subdivided into four behavioral scales: (a) emotional symptoms (e.g., ‘nervous or clingy in new situations’); (b) conduct problems (e.g., ‘fights with other children’); (c) hyperactivity (e.g., ‘easily distracted, concentration wanders’); and (d) peer relationship problems (e.g., reverse-coded item, ‘generally liked by other children’). Scores for each item (0= ‘Not True’, 1= ‘Somewhat True’, and 2= ‘Certainly True’) are added to derive a total, ranging from 0 to 10. Following recommendations for use in population samples (Goodman, Lamping, & Ploubidis, 2010), the scales were combined to produce internalizing (emotional symptoms/peer relationship subscales, total ranging from 0 to 20) and externalizing (conduct/hyperactivity subscales, total ranging from 0 to 20) factors. Reliability correlations for internalizing and externalizing subscales at each wave were >.65 and.70, respectively.

Daily screen time

Caregivers reported the time (hours/day) children spent watching television and using electronic devices at each wave of data collection (e.g., questionnaire prompt: ‘I would like you to think about all the time that <child name> spends on an average day looking at the TV, videos, DVDs, computer, iPad, smart phones, and electronic games systems. We are talking here about the amount of time <child> spends in front of any ‘screen’. How much time would <child> spend on this type of ‘screen time’ on an average day?’). At age 3, screen time was reported by caregivers in minutes and later converted by us into hourly categories. At ages 5, 7, and 9, daily screen time was reported by caregivers based on response categories coded by us for consistency as 1= ‘Less than one hour’; 2= ‘One to less than two hours’; 3= ‘Two to less than three hours’; and 4= ‘Three or more hours’. In line with other studies, individual wave scores for screen time were standardized to hourly categories to facilitate measurement consistency over time.

Potential moderators

Research suggests that there is a broad ecology of factors that predict levels of screen exposure during childhood (e.g., SES, ethnicity, household composition) (Cillero & Jago, 2010). While the between-person components of the RI-CLPM can control for a child’s temporally stable deviation from the grand mean for screen time – and thereby exclude the possibility of within-person components of the RI-CLPM (i.e., the longitudinal autoregressive and cross-lagged parameters) being inflated by unobserved heterogeneity relating to ‘time-invariant’ or ‘trait-like’ effects – it is still conceivable that the random intercepts, their covariances, and the lagged parameters in the model could be moderated by contextual factors (Mulder & Hamaker, 2020). As such, the following variables (measured at child age 3) were explored as potential moderators: child sex, family size, regular use of daycare services, caregiver marital status and stress, ethnicity, and SES.

Statistical analysis

Statistical analyses were conducted in R Studio, and the two RI-CLPMs were estimated using the lavaan package (Rosseel, 2012). Full information maximum likelihood estimation was used to account for potential confounding associated with missing data (see Table S1). The maximum likelihood robust estimator was also used to account for skewness and kurtosis in the distributions of children’s externalizing and internalizing behaviors. Model fit for RI-CLPMs was assessed using the guidelines and cutoff criteria proposed by Hu and Bentler (1999). Model fit was deemed adequate when the comparative fit index (CFI) and Tucker–Lewis index (TLI) were >.95 and when the root mean square error of approximation (RMSEA) and the standardized root mean squared residual (SRMR) were <.06. Since the cross-lagged parameters from RI-CLPM represent the proportion of unique explained variance in the outcome that is not shared with any other predictor (i.e., squared semipartial correlations [sr2] once standardized), thresholds for evaluating the effect sizes of sr2 values were based on the square root of thresholds for correlations recently advocated by Funder and Ozer (2019). This meant that sr2 values of .01, .04, and .09 were indicative of small, moderate, and large cross-lagged effects, respectively. Assessment of the extent to which RI-CLPM estimates differed between levels of our study moderators was undertaken using lavaan’s multiple group function (Mulder & Hamaker, 2020). Differences were estimated between groups for categorical moderators (e.g., sex) and based on a median split for numeric moderators (e.g., family size). Confidence intervals for moderating effects were derived using Fisher’s Z transformation.

Results Sample characteristics

Sample characteristics are displayed in Tables S1 and S2. The final sample was evenly split between boys and girls, who were predominantly White Irish, had multiple siblings, and were living in dual caregiver homes. Half of the children were reported as regularly attending daycare services, and, on average, caregiver reports of stress were low. On average, children’s daily screen time was between two and three hours at ages 3 and 5, and approximately two hours and one and a half hours at ages 7 and 9, respectively. On average, children’s levels of internalizing behaviors were relatively low and remained constant, whereas children’s levels of externalizing behaviors were initially higher and then decreased over time.

Preliminary analyses and model fit statistics

Cross-sectional correlations between screen time and externalizing and internalizing behaviors at ages 3, 5, 7, and 9 are presented in Table S3. Higher levels of screen time were significantly associated with higher levels of externalizing behaviors at each wave (r ranging from .10 to .12). Higher levels of screen time were also significantly associated with higher levels of internalizing behaviors at ages 3, 5, 7, and 9 (r ranging from .06 to .12).

As they may have distinct mechanisms, separate RI-CLPMs were used to estimate the longitudinal bidirectional associations between screen time and externalizing and internalizing behaviors, respectively. The chi-squared difference test confirmed that the RI-CLPM (as opposed to traditional CLPM) and freely estimated (as opposed to constrained) autoregressive and cross-lagged parameters better fit the data for both externalizing and internalizing behaviors (Table S4). Both absolute and relative fit indices also confirmed that both RI-CLPMs were a good fit to the observed data (Table S4). The outcomes from the RI-CLPMs are reported in detail in Tables S4 to S7. The primary estimates representing the longitudinal bidirectional associations between screen time and externalizing and internalizing behaviors are shown in Figures 1 and 2, respectively.

Associations between random intercepts

Random intercepts had a significant variance in the between-person component of the RI-CLPMs (Table S5). This implies the presence of temporally stable (i.e., ‘time-invariant’ or ‘trait-like’) differences between children for daily screen time and externalizing and internalizing behaviors. The between-person component of the model also resulted in significant correlations between random intercepts. This means that, in general, children with higher levels of daily screen time also exhibited higher levels of externalizing and internalizing behaviors on average.

Autoregressive parameter estimates

In the within-person, or time-varying, component of the RI-CLPM (Table S6; Figure 1), positive autoregressive parameter estimates suggest that children’s longitudinal trajectories for externalizing behaviors were relatively stable across adjacent time points. The increase in these autoregressive parameter estimates over time indicates greater stabilization of externalizing behaviors in later development periods. For internalizing behaviors, on average, children’s internalizing behavior scores at age 3 were not significantly associated with scores at age 5 (Table S6; Figure 2). This signals a lack of carryover effect for internalizing pathology in the preschool period. Longitudinal trajectories for internalizing behaviors did not begin to stabilize until between the ages of 5 and 9, with very high levels of stabilization during later childhood. Longitudinal trajectories for screen time were initially stable between the ages of 3 and 5. However, a subsequent nonsignificant autoregressive parameter indicates a lack of carryover of children’s preschool screen viewing habits between the ages of 5 and 7 (likely coinciding with school entry). Children’s trajectories for daily screen time began to stabilize again between the ages of 7 and 9.

Cross-lagged parameter estimates Screen time ←→ externalizing behaviors

Higher levels of externalizing behaviors at age 3 were associated with a moderate increase in screen time at age 5 (Table S7; Figure 1). The obverse association of earlier screen time predicting later externalizing behaviors was not observed. Furthermore, there were no significant cross-lagged associations between screen time and externalizing behaviors beyond age 5. Taken together, these findings suggest that children’s preschool levels of externalizing behaviors were prospectively associated with later increases in screen time, but not vice versa. Moderation analyses revealed a large difference in the cross-lagged association between externalizing behaviors at age 3 and screen time at age 5 between children who were regularly placed in daycare and those who were not. The association was larger for children who were regularly placed in daycare (Table S8). Cross-lagged parameters linking screen time and externalizing behaviors did not differ significantly between levels of other moderators (Tables S9–S14).

Screen time ←→ internalizing behaviors

Higher levels of internalizing behaviors at age 3 were associated with a moderate increase in screen time at age 5 (Table S7; Figure 2). The obverse association of screen time at age 3 predicting increased internalizing behaviors at age 5 was also observed and also a moderate effect size, thereby signaling a substantial bidirectional relationship. A moderate-sized directional association between higher levels of screen time at age 5 and increased internalizing behaviors at age 7 was also observed. Finally, higher levels of screen time at age 7 were associated with a small decrease in internalizing behaviors at age 9, indicating a tapering off and possible change in the direction of the association between screen time and internalizing behaviors as children enter preadolescence. No further significant associations were observed. Moderation analyses revealed that the cross-lagged parameter linking internalizing behaviors at age 5 and screen time at age 7 differed significantly between the sexes, with the magnitude of the association larger for boys (Table S15). The magnitude of this specific association was also greater for larger family sizes (Table S16). Cross-lagged parameters did not differ significantly between levels of other moderators (Tables S17–S21).

Discussion

To date, studies of screen time and psychosocial development have been limited by the extent to which they can control for temporally stable – that is, ‘time-invariant’ or ‘trait-like’ – differences between children that persist over time. The current study used a multivariate multilevel modeling procedure (RI-CLPM) that can account for such ‘unobserved heterogeneity’, and we applied it to a nationally representative cohort of children followed at age 3, 5, 7, and 9. As such, the significance of this study rests on the establishment of greater precision in the estimates of whether and how screen time and externalizing and internalizing behaviors are causally linked over time.

Results from the temporally stable between-child components of the RI-CLPMs (i.e., the random intercepts and their covariances) confirmed the predictions that children would differ significantly in their daily levels of screen time and externalizing and internalizing behaviors, and that, in general, differences in children’s levels of screen time would be associated with differences in their levels externalizing and internalizing behaviors. These results provide further empirical support that broader contextual factors predict differences in children’s levels of daily screen time (Cillero & Jago, 2010) and that heritable differences in children’s levels of externalizing and internalizing behaviors remain largely stable during childhood (Hatoum, Rhee, Corley, Hewitt, & Friedman, 2018). Further, the significant associations between the estimated random intercepts signal the presence of common causes for these temporally stable effects. For example, this temporally stable co-occurrence of screen time and externalizing behaviors could simply be an epiphenomenon of family dysfunction – manifested in a lack of reinforcement strategies to reduce problematic behaviors and guide media usage (Radesky et al., 2016).

As for our exploratory analyses, this study supports previous findings that have shown stability in children’s screen time prior to age 5 (McArthur et al., 2020). The findings of this study are particularly novel in showing that the longitudinal trajectories for children’s screen time were unstable between the ages of 5 and 7, however. In fact, once the confounding effect of temporally stable common causes (e.g., family dysfunction) was addressed by estimation of RI-CLPMs, the nonsignificant, near zero, autoregressive parameter suggests that children’s screen exposure was highly susceptible to change between the ages of 5 and 7, likely due to the transition into middle school and the added demands and school hours that accompany this transition. In contrast, children’s externalizing and internalizing behaviors were consistent across time and stabilized at an increasing rate between the ages of 7 and 9. Taken together, the results of this study suggest that early childhood could be the critical juncture at which to prioritize preventive interventions focusing on problematic screen time and developmental psychopathology more broadly.

Results from the cross-lagged parameters of the RI-CLPM examining externalizing behavior revealed that, after controlling for temporally stable differences between children, higher levels of externalizing behaviors in the preschool period (age 3) were associated with moderate increases in screen time by school entry (age 5). The obverse association was not found, and no further significant cross-lagged associations were observed. Although we can only speculate, these findings may suggest that screens could be used as a tool to modulate challenging behaviors or to placate dysregulated children, in the early years, but do not support the notion that increased screen time leads to later externalizing problems. In fact, despite systematic review evidence for a directional association between screen time and later behavioral problems (Suchert et al., 2015), in this study, differences in children’s daily levels of screen time were not associated with subsequent increases in externalizing behaviors at any point across childhood.

Results from our moderation analyses suggests that the early years cross-lag from externalizing behaviors to later screen time could be exacerbated in families where the child was regularly placed in daycare. It is possible, for example, that busier schedules associated with dual-parent employment and the need to manage drop-offs to and pickups from daycare means that parental stress is elevated and that resources are limited for managing children’s challenging behaviors (Walton, Simpson, Darlington, & Haines, 2014). Thus, in the context of what has become known as ‘household chaos’ (Matheny, Wachs, Ludwig, & Phillips, 1995), parents might resort to instrumental screen use as an accessible and reliable strategy to address challenging behavior, especially in the context of low levels of perceived parenting efficacy (Chen, Chen, Wen, & Snow, 2020).

Cross-lagged parameters from the RI-CLPM examining internalizing behavior indicate that screen time and internalizing behaviors were significantly bidirectionally associated within preschool children. Later in childhood, however, this bidirectional association was not observed. Rather, higher levels of screen time at age 5 predicted moderate increases in internalizing behaviors at age 7, and higher levels of screen time at age 7 predicted a small decrease in internalizing behaviors at age 9. These results shed new light on the prospective associations between screen time and internalizing behaviors across childhood. The initial bidirectional associations suggest that, at a young age, screen time and internalizing behaviors are mutually reinforcing. For example, for children exhibiting higher levels of internalizing symptomatology, screen time might be a welcome escape from unwanted social interaction (Acevedo-Polakovich, Lorch, & Milich, 2007). However, reward mechanisms could, in turn, lead to greater levels of screen use, which could limit some children’s ability to ultimately overcome anxiety-provoking experiences.

In terms of the directional association between screen time at age 5 and internalizing symptoms at age 7, some candidate explanations deriving from the displacement hypothesis have been recently tested and tentatively supported. For example, trends toward greater internalizing problems for children who failed to meet pediatric guidelines for screen time and sleep duration suggest that screen time may be displacing sleep (Carson et al., 2016). Additionally, recent research has shown evidence of an indirect effect, where children’s sleep mediates the association between screen time and internalizing pathology, further supporting the displacement hypothesis (Guerrero, Barnes, Chaput, & Tremblay, 2019).

Compositional analysis of 24-hr movement continues to be a fruitful avenue for explaining screen time’s co-occurrence with internalizing psychopathology. However, it has also been argued that excessive screen time could displace other childhood experiences activities (including parent–child and peer interactions) that directly foster socioemotional development. Kostyrka-Allchorne, Cooper, Simpson, and Sonuga-Barke (2020) have tested this displacement hypothesis in part in a recent study and reported no such effect, however: Frequency of engagement in nondigital recreation did not mediate the relationship between screen time and emotional symptoms. However, perhaps the opportunity cost of screen time is less about displaced time than it is about missed opportunities for ‘teachable moments’ (Madigan et al., 2019). For example, self-regulation has been shown to moderate the relationship between contextual risk factors and growth in internalizing adjustment problems during childhood (Lengua, Bush, Long, Kovacs, & Trancik, 2008). Therefore, with the inflection point of the growth curve for self-regulation occurring between the ages of 3 and 9 years (Weintraub et al., 2013), it is possible that instrumental use of screens by parents could displace opportunities for their child to practice precursor skills for developing self-regulation during this time period (Baker, Morawska, & Mitchell, 2019).

A final point needs to be made about the shift in direction that was observed: greater screen time at age 7 predicting fewer internalizing behaviors at age 9. To our knowledge, this is the first study to show such a relationship during childhood. These findings are consistent with postulations that, in older children, the context of screen time (e.g., Internet-mediated video games) can provide social connection and thereby offset the relative risk of excessive screen time on psychosocial difficulties (Casiano, Kinley, Katz, Chart ier, & Sareen, 2012). Therefore, although not measured in the current study, the positive association between screen use and internalizing behaviors in later childhood could reflect children taking more ownership over the content and context of their screen time at this age and subsequently using screen-based devices to socialize with peers. Remote access to peers could in principle increase the frequency of social interactions, which has been linked to reward system activation and positive affect (Kawamichi et al., 2016). Such hypotheses require empirical validation.

Limitations

Despite the novelty and methodological strengths of this study, some limitations should be noted. First, although caregiver reports were consistent across waves of data collection in the Growing Up in Ireland project, the use of single informant reporting is a potential limitation. Lack of objective measurement of screen time represents another potential limitation; though, objective measurement of screen remains a methodological challenge for the field as a whole rather than a unique issue for this study. Although our measurement of screen time was constant across waves of data collection, it is also worth noting that the pace of technological innovation has changed rapidly over the last decade. Thus, measurement invariance is also a challenge for the field to work to overcome. Finally, in this study, we measured the quantity, rather than the content, of screen time. Whether our findings are moderated by the content of screen time – as well as by the context within which screen time is consumed (e.g., social coviewing, parent–child active mediation) – merits further investigation to better understand how the quantity and quality of screen use interact and (potentially) differentially affect psychosocial outcomes.

Conclusions

The findings of this study provide further evidence that higher levels of externalizing and internalizing behaviors in the preschool period predict greater screen time exposure by school entry. The findings also show that screen exposure has a complex and nonlinear association with internalizing symptomatology across childhood. With screen-based technologies becoming more fully woven into the fabric of modern family life, it is important that we move beyond restrictive or prohibitive mediation and begin to direct families to both the risk and benefits of screen use.

Acknowledgements

The authors acknowledge the Irish Social Science Data Archive for granting permissions to access the ‘Growing Up in Ireland’ dataset. Data were accessed via the Irish Social Science Data Archive at www.ucd.ie/issda. The authors have declared that they have no competing or potential conflicts of interest. Open access funding provided by IReL.

Key points The long-term effects of excessive screen time on psychosocial development across childhood remain poorly understood. This study explores the directionality and temporal sequencing of the associations between daily screen time and externalizing and internalizing behaviors in a large prospective cohort. Higher levels of daily screen time co-occur, but are not prospectively associated, with increases in externalizing behaviors. Higher levels of daily screen time were associated with increased internalizing behaviors between ages 3, 5, and 7 and with a small decrease in internalizing behaviors between ages 7 and 9. This is the first study to signal such a change in the direction of the association between screen time and internalizing symptomatology as children enter preadolescence.