Symptoms of Internet Gaming Disorder in Youth: Predictors and Comorbidity

The Trondheim Early Secure Study (TESS) comprises members of the 2003 and 2004 birth cohorts in Trondheim, Norway (N = 3456) (Wichstrøm et al. 2012). A letter of invitation together with a screen for emotional and behavioral problems, the Strengths and Difficulties Questionnaire (SDQ) 4–16 version (Crone et al. 2008), was sent to the parents of all children in the two birth cohorts prior the routine health check-up at age four of the children (n = 3358). The health nurses at the well-child clinics recruited participants. The nurses missed asking parents of 166 children, and 176 parents were excluded due to lack of proficiency in Norwegian. The study was approved by the Regional Committee for Medical and Health Research Ethics Mid-Norway and written consent was obtained. To increase variability and thus statistical power we oversampled children with higher SDQ scores. To accomplish this, children were allocated to four strata according to their SDQ scores (cut-offs: 0–4, 5–8, 9–11, and 12–40). In all, 1250 of the consenting children were drawn into the study using a random number generator with increasing probability of selection with increasing SDQ scores (0.37, 0.48, 0.70, and 0.89 in the four strata, respectively). Of the 1250 children from the four strata randomly recruited into the Study, 995 were successfully enrolled at Time 1. This sample, adjusted for stratification, was not significantly different from those who consented with respect to gender or their SDQ score (Wichstrøm et al. 2012). Statistics Norway compared the sample with available register information on all parents of 4-year olds in Trondheim in the years 2007 and 2008; the educational level of the TESS sample was virtually identical to the population’s level. The population of Trondheim is similar to the national average on several key indicators, including, average gross income per inhabitant (i.e., 99.5% of the national average), employment rate (i.e., identical to the national rate), and proportion of two-parent families (i.e., 80.0% for Trondheim and national average of 81.4%) (Statistics Norway 2010).

Children were reassessed biannually, affording investigation of 740 with valid data from the third (n = 699, 8.8 years, SD = 0.24) or fourth (n = 702, 10.5 years, SD = 0.16) assessment. Note that 43 youth did participate at the 10-year assessment did not participate at the 8-year (third) assessment. Attrition from the 8-year to the 10-year assessment was selective according to several child characteristics: symptoms of conduct disorder, OR = 1.55, p = 0.018; oppositional defiant disorder, OR = 1.38, p < 0.001; ADHD, OR = 1.16, p < 0.001; male gender, OR = 1.84, p = 0.049; problems with executive functioning, OR = 1.02, p = 0.001; low scores on word comprehension, OR = 0.93, p = 0.001; and being bullied, OR = 1.14, p = 0.039. Even so, the combined effect of these predictors on attrition was small (Cox & Snell R² = 0.05). The interviewing with respect to IGD started in September 2013 and ended in June 2015. The teacher who knew the child also completed a questionnaire about the child. Table 1 presents characteristics of the TESS participants. At ages 8 and 10 children used medication for mental health problems, mostly for ADHD (n = 8). All Norwegian children are expected to have access to a computer and the Internet, as most homework and other assignments are turned in on the Internet. As a result 100% of all families with children report to have Internet access in Norway (Statistics Norway 2017).

The importance of each symptom to the diagnosis of IGD was examined for sensitivity, specificity, and predictive value. The importance of each symptom to an underlying continuous IGD construct was evaluated by means of a one-factor exploratory analysis of binary items. To what extent the symptoms tap one or more underlying dimensions was assessed using exploratory factor analysis with oblique GEOMIN rotation. Age-8 predictors of age-10 IGD-symptoms were evaluated using linear regression, first unadjusted and then adjusted for gender and gaming at age 8. Latent IGD factors by means of confirmatory factor analysis were constructed to examine whether variables predicted different IGD factors differently when factor solution with more than one IGD construct were modelled. Finally, significant predictors were entered together in a multivariate regression analysis. To arrive at correct population estimates given oversampling, results were weighted with a factor corresponding to the number of children in the population in a particular stratum divided by the number of participants in that stratum (inverse probability weighting). Analyses were performed in Mplus 7.31 (Muthén and Muthén 1998-2015), applying a robust maximum likelihood estimator which is robust to deviations from normality, and missing data were handled according to a full information maximum likelihood procedure under the assumption that data was missing at random.

The prevalence of IGD (5+ symptoms) was low (unweighted n = 14), though 6 times higher among boys than girls (Table 2). The prevalence of individual symptoms varied considerably, with some symptoms being fairly common, particularly tolerance, unsuccessful attempt to control, and use of Internet games to escape or relieve negative mood. Symptoms tapping negative consequences were much less frequent, though more evident in boys.

Almost all children with the IGD-diagnosis had symptoms of tolerance and unsuccessful attempts to control gaming (see “sensitivity” column, Table 3), but fewer than 10% with such symptoms received a diagnosis (see “positive-predictive-value” column (PPV), Table 3). Most meeting the diagnostic threshold had lost interest in previous activities and hobbies, used gaming to escape or relieve negative mood, and about half had deceived family members or others concerning their gaming. Relatedly, about half of the children who presented one of these three symptoms received a diagnosis. Withdrawal symptoms were rare among those with an IGD-diagnosis and few children with this symptom met diagnostic criteria. Moreover, about half with a diagnosis fulfilled the preoccupation criterion, though most with preoccupation did not receive a diagnosis.

A one-factor factor analysis of IGD-symptoms revealed that loss of interest and continued excessive use despite psychosocial problems loaded highly, whereas withdrawal loaded moderately (Table 4). Exploratory factor analysis indicated that a two-factor solution fitted the data better than the a-priori one-factor model, Δχ² = 20.60, df = 8, p = 0.008, with a 3-factor solution not providing a better fit than the 2-factor solution, Δχ² = 8.57, df = 7, p = 0.29. Symptoms associated with heavy involvement in Internet gaming (i.e., preoccupation, withdrawal, tolerance, escape or relieve negative mood) loaded on one factor, with a second factor defined by symptoms tapping negative consequences of gaming (i.e., loss of interest, continued excessive use despite psychosocial problems, deception, jeopardizing or losing relationships or educational opportunities). The mean communality was satisfactory (MacCallum et al. 1999), albeit somewhat wide, attributable to preoccupation, withdrawal and deception. The two factors correlated moderately (r = 0.46).

The number of IGD-symptoms correlated weakly with symptoms of ADHD (r = 0.08 [95% CI: 0.00–0.16]), ODD (r = 0.15 [0.07–0.24]), anxiety disorders (r = 0.13 [0.06–0.20]), and major depressive disorder (r = 0.12 [0.03–0.20]), but was unrelated to CD (r = 0.07 [−0.01–0.14]). The two IGD factors identified above were not differentially correlated with symptoms of these psychiatric disorders (p > 0.05).

The number of age-10 IGD-symptoms was regressed on age-8 socio-demographic, parenting/family and child factors in separate analyses for each predictor, with and without adjustment for gender and age-8 h per day of gaming (Table 5). At age 8 only two children played for more than 4 h per day and none for more than 5 h. Thus, there were likely no children who fulfilled the criteria for IGD at age 8. Among family factors, only higher parental education (adjusted for gender and age-8 gaming) predicted fewer symptoms at age 10. Among child factors, better social competence and emotion regulation predicted fewer IGD-symptoms, as did a higher involvement in organized sports, and Word comprehension score following adjustments for gender and age-8 gaming.

A multivariate analysis with significant predictors was tested next. Because emotion regulation and social skills were highly correlated (r = 0.88), it was not possible to include both variables in the same model. This analysis showed previous gaming time, B = 0.27 [0.13–0.42] p < 0.001, social skills, B = −0.01 [−0.01–0.00], p = 0.007, and gender, B = 0.32 [0.13–0.51], p = 0.001, to uniquely contribute to the prediction of future IGD-symptoms, whereas sports participation was rendered insignificant, B = −0.06 [−0.16–0.02], p = 0.13), R² = 0.08. Of note, the correlation between sports involvement and social skills was fairly high, r = 0.36, p < 0.001. If social skills was removed from the model, the seeming impact of sports participation became significant, B = −0.10 [−0.18 – −0.02], p = 0.029).

To gain further insight into possible mechanisms by which word comprehension and parental education might influence gaming behavior, we tested multiple mediation within a structural equation model in which social skills and sports participation were regressed on these two measures as, and symptoms of IGD were regressed on all of these measures, while controlling for gender and previous gaming time. Neither Word comprehension, B = 0.00 [−0.001–0.01] nor parental education, B = −0.03 [−0.07–0.01] had any direct effect on IGD symptoms. Notably, their respective negative effects on IGD-symptoms were mediated via social competence, with the following indirect effects: parental education, B = −0.01, p = 0.035, Word comprehension, B = −0.01, p = 0.01. Recall that sports participation was not predictive of IGD symptoms when social skills were adjusted for. Hence, sports participation could not act as a mediator. Replacing social skills with emotion regulation yielded similar results. The symptom “using Internet games to relieve or escape negative mood” could be considered a strategy for emotion regulation. We therefore constructed a measure of IGD symptoms deleting this item. Even this abbreviated 8-item IGD measure was predicted by emotion regulation, unadjusted B = −0.09, [−0.15- -0.03] and adjusted for previous gaming and gender, B = −0.09, [−0.14- −0.04].

Because the factor analytic findings revealed a Heavy involvement and a Negative consequences factor, it is possible that different variables predicted the two factors. To test this possibility a model assuming identical predictors of these two latent constructs was compared to a model in which predictors were free to vary, using a Wald test. To avoid differential scaling of the latent variables to influence the results, the residuals of the two latent factors were both set to1. None of the age-8 variables and age-6 temperament were differentially predictive of the two constructs. However, anxiety symptoms at age 10 correlated with heavy involvement, r = 0.22, p > 0.001, but not with Negative consequences, r = −0.02, p = 0.82, Wald = 8.18, p = 0.004. Symptoms of other disorders were not differentially correlated with the two IGD factors.

Previously reported prevalence rates of IGD range from 1.2% (Rehbein et al. 2015) to 8.5% (Gentile 2009). As youth continue to increase their Internet gaming into adolescence, we should expect the rates of IGD to increase as well, though a decline might emerge when the adolescents enter adulthood (Rehbein et al. 2016). Applying the DSM-5 criteria and using a diagnostic interview, the prevalence discerned herein is at the lower end. Whether the proposed symptoms actually capture a disorder or merely heavy involvement in Internet gaming is much disputed (Griffiths et al. 2016; Kardefelt-Winther 2015; Petry et al. 2014). Two correlated dimensions underlying IGD symptoms, strong gaming involvement and negative gaming consequences, resemble previous results (Brunborg et al. 2015; Charlton and Danforth 2007). Most of the symptoms tapping negative consequences had fairly high sensitivity, specificity, and PPV. As regards one of these symptoms, loss of interest in previous activities and hobbies, some have argued that there is nothing inherently pathological in switching interest from previous interests to Internet gaming (Griffiths et al. 2016). Notably, losing interest in previous interests was not normative (2.5% prevalence). Among those who did, however, more than half received an IGD diagnosis and almost two thirds of those with IGD affirmed this symptom; moreover, it loaded perfectly on a continuous IGD factor. Thus, although it may seem harmless to give up previous interests to gaming, those with IGD typically do so-- and those who do so typically have IGD; this result closely resembles that of a questionnaire study of German adolescents (Rehbein et al. 2015). Thus, although infrequent, loss of interest may be a core feature of IGD across ages.

To receive an IGD diagnosis five symptoms are needed. There are no strong arguments for why a cut-off of five was suggested in the DSM-5, understandably so, since such information was lacking. However, before the number of symptoms is decided upon, the appropriate content of IGD-symptoms should be investigated further since symptoms have different prevalence and may be of unequal importance to the IGD construct. Because the majority of the heavy involvement symptoms are prevalent, many individuals who cross the diagnostic threshold will likely manifest all or many of the four heavy-involvement symptoms, but only one or a few relating to negative consequences. Two of the heavy involvement symptoms, tolerance and unsuccessful attempt to control gaming, had low PPV, which suggests that they were unspecific to those with the disorder. This accords with the theoretical argument that tolerance does not make sense in the context of Internet gaming (Kardefelt-Winther 2015). Empirically, however, other investigators using questionnaires have reported tolerance to be central to a continuous IGD construct as well the disorder (Lemmens et al. 2015; Rehbein et al. 2015). It should also be recognized that the TESS participants were young and likely to increase their Internet gaming as they age; increased gaming should thus be expected and not pathologized. Hence, the centrality of the tolerance criterion may vary by age.

Gaming to relieve or escape negative moods was fairly prevalent (15%). Indeed, gaming to relieve negative moods is typical of those with heavy involvement (Griffiths et al. 2016), including those who do not necessarily have the disorder. Even so, the PPV was among the highest (50%); the specificity was high; and the item also loaded highly on the one-factor IGD dimension. Thus, empirically, the escape symptom seems to be central to IGD in this population.

As regards withdrawal, its sensitivity was very low; its PPV was low; and it loaded moderately on an underlying IGD factor. Contrary to previous factor-analytic (Lemmens et al. 2015) and inference-tree (Rehbein et al. 2015) findings, then, withdrawal does not seem to define IGD in the present population when a clinical interview was applied. In sum, if our findings are replicated, the content, number of and balance between heavy involvement and negative consequences symptoms should be considered in future revisions of the IGD diagnosis.

Finding symptoms of IGD to correlate (somewhat) positively with symptoms of most common psychiatric disorders in youth is in line with previous research (Kuss et al. 2014). Notably, though, associations were considerably weaker than between symptoms of other disorders (Costello et al. 2003; Wichstrøm et al. 2018), and no specific pattern of associations was seen. Thus, the data presented herein does suggest that IGD belongs to neither the internalizing nor externalizing spectrum. The presence of comorbidity does not speak directly to whether IGD should be considered a disorder or not. However, the weak associations with symptoms of more established psychiatric disorders indicate that the latter are not likely predictors or consequences of IGD symptoms, at least in this young age group.

Two prior prospective studies identified poor social competence and emotion-regulation skills as predictive of later symptoms of pathological video-gaming using self-completed questionnaires (Gentile et al. 2011; Liau et al. 2015). We replicated these Singaporian findings using a diagnostic interview. Possibly due to allowing more time for deliberate response and not giving away non-verbal cues of social insecurity or appearance, children who are less socially competent in face-to-face interaction may prefer the Internet for social interaction (Kowert and Oldmeadow 2013). Whether poor social competence is indeed a risk factor for IGD or for simply spending much time on the Internet awaits future research. Like others, we find that emotion-regulation deficits predict IGD symptoms, but further show that this also holds when IGD is examined by a diagnostic interview and when “using gaming to reduce negative mood” is not an included symptom. Finally, low parental education and less child word comprehension emerged as predictors of IGD symptoms, seemingly through reduced child social competencies.

The gender difference in gaming and IGD in particular has been known for a long time. What is it about boys that make them more susceptible to IGD than girls? One may speculate that it is because they simply game more than girls, or that other risk factors such as poor social competence, and high levels of aggression and impulsiveness are more prevalent in boys (Rehbein et al. 2010). However, even when such trait-like characteristics were taken into account, boys evinced more IGD-symptoms than girls in the present study. This increase risk of becoming addicted in young males is seen in a range of areas, including gambling (Dowling et al. 2017) and alcohol use (Rohde et al. 2001). Thus, common explanations for all addictions should be considered and investigated. However, one should not overlook the possibility that gaming-specific factors such as boys’ stronger preference for games that may be particularly addictive such as role-playing and first-person shooting games may explain the gender difference, as one study suggested (Rehbein et al. 2016). Relatedly, games may be constructed to be more attractive to boys.

We failed to identify several correlates and predictors of IGD symptoms and related Internet addiction, as well as of pathological video gaming, that have emerged in prior cross-sectional (e.g., Ko et al. 2012; Kuss et al. 2014; Li et al. 2014) and prospective (e.g., Henchoz et al. 2016; Liau et al. 2015) studies, including family climate, parenting, child temperament, symptoms of psychiatric disorders, physical activity/sport participation, victimization, and self-esteem. Conceivably, this is due to the fact that most prior work used self-completed questionnaires to assess all construct, thus raising the possibility that reported associations are inflated due to common methods. Such inflation seems less likely in the work reported herein given that we relied on diverse methods and information sources. Moreover, the lack of adjustment for likely confounders may have inflated some previous estimates. To illustrate, as in a Swiss study (Henchoz et al. 2016) we found sports participation to predict future IGD, over and above previous gaming, gender, and objectively measured level of moderate and vigorous physical activity. However, unlike the prior investigation, once we adjusted for the higher social competence of those involved in sports, the effect of sports vanished.

We must acknowledge detection of only a few IGD cases, resulting in somewhat large confidence intervals for the prevalence estimate, 0.7% to 2.7%. This uncertainty should also be considered in the context of moderate reliability of the IGD disorder, which might be due to the low prevalence of IGD in the population and thus uncertainty in the reliability sub-sample. Second, we investigated 10-year olds in one Norwegian community. Prevalence may increase as children age--because gaming is highest in 12–14 year olds (Gentile 2009; Medietilsynet 2015). Although it has been difficult to discern country differences in prevalence because of the wide variety of assessment instruments and definitions of IGD employed (Petry et al. 2015), the prevalence documented herein may be different in other countries, and in other populations such as those seeking help for mental health problems. The fact that all Norwegian children have access to the Internet should also be considered when generalizing the present findings. Nonetheless, the rate of Internet gaming was similar to that found in Norwegian (Medietilsynet 2015), German (Rehbein and Baier 2013), Spanish (Pujol et al. 2016), and British (Przybylski 2014) preadolescent samples, indicating generalizability of results at least to European youth. That said, the rates of other mental health problems are generally lower among Norwegian children than in many other Western countries (Heiervang et al. 2007; Wichstrøm et al. 2012; Wichstrøm et al. 2018), suggesting that generalization to other populations should be tentative. We relied on child-reports to determine IGD because many of the IGD symptoms are likely difficult to reliably and validly assess for others (e.g., anticipation, gaming to escape negative mood) or youth may actively conceal their gaming from parents. Moreover, it is possible that youth lose track of for how long they game once they are absorbed in gaming, or that they consciously underreport gaming time or other symptoms when queried for social desirability reasons. Even though parents may not be in an expert position to rate all symptoms, they could certainly report on some of them, and this lack of parental information is a limitation, possibly deflating the number of symptoms and the rate of disorder. Even though the over-determination was good in our 2-factor solution, the communalities evinced some width, implying that it should be interpreted with caution as we cannot rule out the possibility that an even higher number of participants might have revealed different patterns. We were only able to adjust for time spent on gaming at age 8 and not IGD-symptoms when examining the predictors of age 10 IGD-symptoms, though very few children played for an extended time at age 8 and thus few were expected to IGD-symptoms. Moreover, some predictors had only modest reliability (e.g., parental monitoring and inconsistent discipline) and this may have deflated associations with symptoms of IGD. Finally, we interviewed only the youth on IGD. As previously mentioned we had strong reasons to believe that parents might underestimate the amount of gaming and the prevalence of IGD symptoms due to a lack of information. However, we cannot discount the possibility that the youth might also underreport gaming and IGD symptoms, in part for social desirability reasons, and that at least some parents might have provided information that is more correct.

DSM-5-defined IGD is already present in some 10-year olds, with a strong male preponderance. The nine proposed symptoms tap two underlying factors, heavy involvement in gaming which proved quite common, and negative gaming consequences which was rather rare. Two of the common heavy-involvement symptoms, tolerance and unsuccessful attempts to control gaming, were not very specific to IGD. The withdrawal symptom was weakly related to IGD, both as a diagnosis and as continuous construct. It is important to investigate the content and relevance of these three criteria further. Symptoms of IGD correlate modestly with symptoms of other psychiatric disorders. Although a wide range of potential child, family and demographic predictors were considered, only low social competence and poor emotion regulation skills predict more IGD-symptoms.