A problematic use of the Internet has been identified in a number of studies and shows that persistent negative consequences such as job loss, academic failure, and divorce resulted from excessive Internet use (for reviews see Griffiths, 2000a,b; Chou et al., 2005; Widyanto and Griffiths, 2006; Byun et al., 2009; Weinstein and Lejoyeux, 2010; Lortie and Guitton, 2013). The clinical relevance of this phenomenon gains in importance against the background of high estimated prevalence rates ranging from 1.5 to 8.2% (Weinstein and Lejoyeux, 2010) or even up to 26.7%, depending on the scales used and criteria applied (Kuss et al., 2014).

Although the first description of this clinical issue is almost 20 years ago (Young, 1996), the classification is still discussed controversially and consequently several terms are used in the scientific literature, ranging from “compulsive Internet use” (Meerkerk et al., 2006, 2009, 2010), “Internet related problems” (Widyanto et al., 2008), “problematic Internet use” (Caplan, 2002), “pathological Internet use” (Davis, 2001) to “Internet related addictive behavior” (Brenner, 1997), to mention just a few. In the last 10 years, however, most researchers in this field have used the term “Internet addiction” or “Internet addiction disorder” (e.g., Johansson and Götestam, 2004; Block, 2008; Byun et al., 2009; Dong et al., 2010, 2011, 2013; Kim et al., 2011; Purty et al., 2011; Young, 2011b, 2013; Young et al., 2011; Zhou et al., 2011; Cash et al., 2012; Hou et al., 2012; Hong et al., 2013a,b; Kardefelt-Winther, 2014; Pontes et al., 2014; Tonioni et al., 2014). We also prefer the term “Internet addiction (IA),” because recent articles (see discussion in Brand et al., 2014) highlight the parallels between an overuse of the Internet and other addictive behaviors (e.g., Grant et al., 2013) and also substance dependency (see also Young, 2004; Griffiths, 2005; Meerkerk et al., 2009). It has been argued that mechanisms related to the development and maintenance of substance dependency are transferrable to an addictive use of Internet applications (and also other behavioral addictions), for example the incentive sensitization theory of addiction and related concepts (e.g., Robinson and Berridge, 2000, 2001, 2008; Berridge et al., 2009). This fits also nicely with the component model on addictive behaviors (Griffiths, 2005).

Many studies have been conducted on psychological correlates of IA, but this has been done – at least in most cases – without differentiating between a generalized Internet addiction (GIA) and a specific Internet addiction (SIA; Morahan-Martin and Schumacher, 2000; Leung, 2004; Ebeling-Witte et al., 2007; Lu, 2008; Kim and Davis, 2009; Billieux and Van der Linden, 2012), although psychological mechanisms might be different, also for distinct age groups or applications used (Lopez-Fernandez et al., 2014). Our study examines the mediating effects of coping styles and cognitive expectations for Internet use in the development and maintenance of GIA in order to contribute to a better understanding of underlying mechanisms and potential implications for diagnostic and treatment.

On a theoretical level, it was already postulated that IA has to be differentiated regarding the generalized Internet use (Griffiths and Wood, 2000) versus specific types of IA such as cybersex, online relations, net compulsions (e.g., gambling, shopping), information search, and online gaming for developing an addiction to the Internet (e.g., Young et al., 1999; Meerkerk et al., 2006; Block, 2008; Brand et al., 2011). However, only one subtype, Internet Gaming Disorder, has been included in the appendix of the DSM-5 (APA, 2013). Most studies either assessed IA as a unified construct or only assessed one specific subtype (in most cases Internet gaming). In his cognitive-behavioral model, Davis (2001) also differentiated between a generalized pathological Internet use (GIA) and a specific pathological Internet use (SIA). GIA was described as a multidimensional overuse of the Internet, frequently accompanied by time waste and non-directed use of the Internet. Social aspects of the Internet (e.g., social communication via social networking sites) are particularly used (see also discussion in Lortie and Guitton, 2013), which is supposed to be linked to a lack of social support offline and social deficits experienced by an individual in non-virtual situations. In addition, it has been argued that subjects may use several different Internet applications excessively without having one certain favorite, for example playing games, watching pornography, surfing on information and/or shopping sites, posting selfies, watching videos on video platforms, reading blogs of others, and so on. In this case, one may argue that the individual is addicted to the Internet and not addicted on the Internet (but see also discussion in Starcevic, 2013). Davis argues that one main difference between GIA and SIA is that individuals who suffer from GIA would not have developed a similar problematic behavior without the Internet, whereas individuals suffering from SIA would have developed similar problematic behavior within another setting. In both forms of addictive use of the Internet, GIA and SIA, dysfunctional cognitions about the self and about the world are suggested to play a fundamental role (Caplan, 2002, 2005).

Research addressing GIA demonstrated that subjective complaints in everyday life resulting from Internet use are correlated with diverse personality characteristics. Indeed, it was shown that GIA is linked to psychopathological comorbidities, such as affective or anxiety disorders (Whang et al., 2003; Yang et al., 2005; Weinstein and Lejoyeux, 2010) as well as to the personality traits shyness, neuroticism, stress vulnerability, tendencies to procrastinate, and low self-esteem (Niemz et al., 2005; Ebeling-Witte et al., 2007; Hardie and Tee, 2007; Thatcher et al., 2008; Kim and Davis, 2009). Also, factors of social context, e.g., lack of social support or social isolation (Morahan-Martin and Schumacher, 2003; Caplan, 2007) and even loneliness in the educational setting in adolescents (Pontes et al., 2014), seem to be related to GIA. Moreover, it has been argued that using the Internet as a tool for coping with problematic or stressful life events contributes to the development of GIA (Whang et al., 2003; Tang et al., 2014). Persons with IA show also high tendency toward impulsive coping strategy (Tonioni et al., 2014). Some authors even conceptualize IA as a type of coping with everyday life or daily hassles (Kardefelt-Winther, 2014). There are still only some first studies, which explicitly compared predictors of different types of SIA. Pawlikowski et al. (2014) reported that shyness and life satisfaction are related to an addictive use of Internet games, but not to a pathological use of cybersex or the use of both games and cybersex.

Based on previous research, in particular on the arguments by Davis (2001), and also considering current literature on neuropsychological and neuroimaging findings in subjects who are addicted on the Internet, we have recently published a theoretical model on the development and maintenance of GIA and SIA (Brand et al., 2014). Some aspects included in the model have already been mentioned in the context of the use of social networking sites, for example the expectancy of positive outcomes (Turel and Serenko, 2012). It has also been shown that an excessive or addictive use of online auctions is correlated with changes in individuals’ beliefs about the technique and this determinates future use and use intentions (Turel et al., 2011). This is in line with our theoretical model on GIA, in which we assume that beliefs or expectancies about what the Internet can do for a person influence the behavior, i.e., the Internet use, which in turn also influences future expectancies. However, in our model we have focused on the mediating role of expectancies and coping strategies in developing and maintaining a GIA and specific types of SIA.

For the development and maintenance of GIA, we argue that the user has certain needs and goals which can be achieved by using certain Internet applications. Based on prior research, we incorporated several of those findings to develop a comprehensive model to tie these elements together. Initially, a person’s core characteristics are associated with IA and include psychopathological aspects, personality aspects, and social cognitions. In the first section, we included psychopathological symptoms, in particular depression and social anxiety (e.g., Whang et al., 2003; Yang et al., 2005), dysfunctional personality facets, such as low self-efficacy, shyness, stress vulnerability, and procrastination tendencies (Whang et al., 2003; Chak and Leung, 2004; Caplan, 2007; Ebeling-Witte et al., 2007; Hardie and Tee, 2007; Thatcher et al., 2008; Kim and Davis, 2009; Pontes et al., 2014), and social isolation/lack of social support (Morahan-Martin and Schumacher, 2003; Caplan, 2005) in the development of GIA. However, we suggested that the influence of those person’s primary characteristics and cognitions on the development of an addictive use of the Internet should be mediated by certain Internet-related cognitions, in particular Internet use expectancies (Turel et al., 2011; Xu et al., 2012; Lee et al., 2014), and certain strategies to cope with everyday life requirements or daily hassles (Tang et al., 2014; Tonioni et al., 2014). In the third section of the model, as a consequent behavior, if the user goes online and receives reinforcement in terms of dysfunctional coping with problems or negative mood and the person expects that Internet use will distract them from problems or negative feelings, then the more likely they will turn to the Internet to escape those feelings evidenced by a loss of control, poor time management, cravings, and increased social problems. The role of reinforcement and conditioning processes has been described well in the literature on the development and maintenance of substance related disorders (e.g., Robinson and Berridge, 2001, 2008; Kalivas and Volkow, 2005; Everitt and Robbins, 2006). We have also argued that the positive and negative reinforcement of coping style and Internet use expectancies successively result in a loss of cognitive control over the Internet use, which is mediated by prefrontal (executive) functioning (Brand et al., 2014).

Although this model fits well with previous literature on key findings with regards to psychological mechanisms behind IA (see overviews by Kuss and Griffiths, 2011a,b; Griffiths, 2012) and also with very recent neuropsychological and neuroimaging correlates of GIA and distinct types of SIA (Kuss and Griffiths, 2012; Brand et al., 2014), this model still needs empirical evidence in terms of incremental validity. In this study, we aimed at translating the hypotheses summarized in the theoretical model on GIA outlined above into a statistical model on latent variables level and tested the predictor and mediator effects on the severity of GIA symptoms using a large scale Internet population. Using validated psychological and personality measures, we first assessed a persons’ core characteristics in predicting an excessive and addictive use of the Internet in a generalized way. Using a validated measure of coping and a newly developed measure of Internet use expectancies, we tested if poor coping skills and Internet use expectancies (such as using the Internet to escape negative feelings or unpleasant situations) mediate the link between person’s core characteristics and symptoms of GIA.

We have introduced a new theoretical model on the development and maintenance of an addictive use of the Internet (Brand et al., 2014), which is based on the main arguments by Davis (2001) who first suggested a differentiation between a generalized overuse of the Internet (GIA) and a specific addiction to certain Internet applications (SIA). In the current study, we translated the theoretical model on GIA into an operationalized model on latent level and statistically tested the SEM using an online survey on an Internet population of 1019 respondents. We found an overall good model fit with the data and the hypothesized SEM, which represents the main facets of the theoretical model and explained 63.5% of the variance of GIA symptoms as measured by the s-IAT (Pawlikowski et al., 2013).

The model is the first to tie together elements associated with IA such as depression, social anxiety, low self-esteem, low self-efficacy, and higher stress vulnerability. Based on the emphasis of cognitions related to developing IA and to addictive behavior in general (Lewis and O’Neill, 2000; Dunne et al., 2013; Newton et al., 2014), the model investigates if two mediator variables (coping styles and Internet use expectancies) impact the direct effects of the predictor variables (psychopathology, personality, and social cognitions) on the development of GIA. Results show that both coping styles and Internet use expectancies play a significant role.

All variables (predictors and mediators) included in the model were significantly correlated with the s-IAT score on a bivariate level. This is basically consistent with previous research on bivariate relationships between symptoms of IA and personality aspects, psychopathological symptoms and other person variables, as mentioned in the Introduction. However, in the SEM analysis, all direct effects of the three main predictors (on latent dimension) were no longer significant when including the hypothesized mediators into the model. This means that psychopathological aspects (depression, social anxiety), personality aspects (self-esteem, self-efficacy, and stress vulnerability) as well as social cognitions (emotional loneliness, perceived social support) do not impact symptoms of GIA directly, but that their influence is mediated by either a dysfunctional coping style, or Internet use expectancies, or both. However, psychopathological aspects and personality aspects significantly predict both dysfunctional coping style and Internet use expectancies. Social cognitions, however, are not significantly related to coping and expectancies, when their relative impact is controlled for the effects of psychopathological and personality aspects (but note that the three predictor latent dimensions were correlated significantly and that the effect from social cognitions to Internet use expectancies slightly failed to reach significance). The direct effects of both coping style and expectancies on symptoms of GIA were significant. In summary, the current study, although with a non-clinical population, not only confirms the previous findings on the relevance of coping style and dealing with stressful life events (Kardefelt-Winther, 2014; Tang et al., 2014; Tonioni et al., 2014) as well as Internet use expectancies (Turel and Serenko, 2012; Xu et al., 2012; Lee et al., 2014) for developing or maintaining symptoms of GIA, but explicitly highlights the role of coping and expectancies as mediators in the process underlying GIA.

The model was tested with a large online population. Model must be tested with clearly defined clinical samples, such as treatment-seeking individuals. The meaning of the model would be more robust with a clinical population to draw more accurate clinical implications. Although 11.3% of the sample reported a problematic Internet use and 3.7% described themselves as having an addictive Internet use, this study is considered only an initial look to see if the model works and draws statistical inferences that could potentially have clinical relevance. However, as a new model with statistical significance using a variety of psychological and personality tests on online users, a few clinical implications, which may inspire future research, can be made with caution.

First, individuals with dysfunctional coping to deal with problems in their life and who have expectancies that the Internet can be used to increase positive or reduce negative mood may be more likely to develop GIA. Moreover, the effects of psychopathological aspects on both dysfunctional coping and Internet use expectancies were positive indicating that higher symptoms of depression and social anxiety can increase the risk for dysfunctional coping strategies and also for the expectancies that the Internet provides help for dealing with stress or negative mood. Only when these processes act in concert, meaning the combination of psychopathological symptoms and coping/expectancies, the probability of using the Internet addictively seems to increase.

Secondly, although the number of studies addressing treatment of GIA is limited, the meta-analysis published by Winkler et al. (2013) argues that cognitive-behavioral therapy is the method of choice. This is particularly based upon the analysis of treatment effects on time spent online, depression, and anxiety symptoms. In fact, cognitive-behavioral therapy for IA (CBT-IA; Young, 2011a) has been identified as the most prevalent form of treating IA (Cash et al., 2012). Within cognitive-behavioral treatment of GIA proposed by Young (2011a), individual characteristics as well as coping and Internet use expectations have already been hypothesized to be relevant within the treatment of GIA, but the empirical evidence was very sparse (e.g., Young, 2013).

The findings presented in this study provide one further source of evidence to show that cognitive-behavioral therapy and CBT-IA can work to treat IA. The person’s specific cognitions (coping style and Internet use expectancies) mediate the impact of psychopathological symptoms (depression, social anxiety), personality traits, and social cognition (loneliness, social support) on GIA symptoms. Using cognitive therapy, an emphasis in assessment should include identifying dysfunctional cognitions to be addressed. That is, upon examination, clinicians should examine Internet use expectancies to understand the needs of the client and what ways the client believes the Internet may help to satisfy.

Alternatively, findings also suggest that therapy should address maladaptive cognitions associated with dysfunctional use of the Internet. These findings confirm earlier studies that showed maladaptive cognitions such as overgeneralization, avoidance, suppression, magnification, maladaptive problem solving, or negative self-concepts are associated with addictive Internet use (Young, 2007). A clinical implication of these findings is that therapy should apply cognitive restructuring and reframing to combat thoughts that lead to addictive use of the Internet. For instance, a patient suffering from GIA may have signs of social anxiety and shyness and therefore a few friends and also trouble with others at school. She may then think that communicating with other people via social networking sites gratifies her social need without having the scary situational aspects of a “real” social interaction. In addition, she may have the expectancy that also playing an online game may distract her from the problems at school and that buying online or searching information on the Internet may reduce the feelings of loneliness. Therapy would focus her on seeing alternative places at school or in private life where she can build up esteem and gratify social needs. If she stops justifying that the social networking sites, games and shopping sites are the only places she feels good about her life and she finds other healthier outlets, the less reliant she will be on the different Internet applications. Knowing the role that cognitions play in the development of GIA, cognitive therapy can help clients restructure the assumptions and interpretations that keep them online. Again, these potential clinical implications of the study’s results must be treated with caution, since they must be replicated in a treatment-seeking, clinical sample.

From a broader perspective, however, these findings gain insights into how therapists can specifically apply CBT-IA to Internet-addicted patients. Behavior modification can help clients develop and adapt new and more functional coping strategies in order to deal with daily hassle. Therapy needs to focus on helping clients find healthier ways of coping than turning to the Internet. A major component of CBT-IA is behavior therapy to help clients cope with underlying issues contributing to IA, specific or generalized (Young, 2011a, 2013). The findings suggest that improving coping skills would reduce the need to go online for clients. Although studied in a sample of the general population, we believe that the finding that coping and expectancies are mediators in the development and maintenance of GIA contribute to a better understanding of the mechanisms of GIA and that they likely have some treatment implications, as mentioned above. Another aspect that was not focused in the current study is the role of prefrontal cortex integrity. Efficacy of CBT-IA may also depend upon the patient’s prefrontal functioning, because strengthening cognitive control of the Internet use in the course of the therapy is most likely related to executive functions and other higher-order cognitive processes. This is important to address in future studies, because most recently there have been a couple of articles published showing that prefrontal cortex functions are likely reduced in patients with IA (see overview in Brand et al., 2014).

In our sample, age was inversely correlated with symptoms of GIA, but with a very low effect size (explaining 1.96% of the variance, only). Considering recent articles on Internet use in older individuals (e.g., Eastman and Iyer, 2004; Vuori and Holmlund-Rytkönen, 2005; Campbell, 2008; Nimrod, 2011), one may certainly except age effects on several aspects of using the Internet, such as using motives and the way elderly experience fun and satisfaction on the Internet. Given that elderly people also have a higher chance to develop executive dysfunctions due to prefrontal cortex changes with increasing age (Alvarez and Emory, 2006), which are also linked to decision-making reductions (Brand and Markowitsch, 2010), one may speculate that those older individuals with executive reductions, who experience a large amount of pleasure on the Internet may develop GIA. However, this is not represented by our data, since our sample did not include older subjects. Future studies may investigate the specific vulnerability factors linked to the risk of GIA in older adults.

Gender did not affect the overall structure of the model. In previous articles, gender effects have been found for specific types of IA, such as online gaming (e.g., Ko et al., 2005) and particularly cybersex (Meerkerk et al., 2006; Griffiths, 2012; Laier et al., 2013, 2014), but it has also been argued that both genders are generally at risk for developing an addictive use of the Internet (Young et al., 1999, 2011). In our study, the effects of gender on GIA, as measured by the s-IAT, was very low (d = 0.19, see results), indicating that at least in a general population both gender are equally at risk for developing GIA. Although gender did not affect the general data structure in the SEM, there were some differences between men and women with respect to three direct effects from predictor variables to the mediators. As summarized in the results section, psychopathological aspects had an effect on expectancies in women, not in men, in the negative effect of personality aspects on coping and expectancies is more present in men than in women. These effects fit with the literature on gender differences with respect to depression and social anxiety (Sprock and Yoder, 1997; Moscovitch et al., 2005), as well as self-esteem and self-efficacy (Huang, 2012). However, the facets which are the focus of the study, namely the mediation effects of coping and expectancies and their importance for GIA were not affected by gender (see results of the mean structure analysis). So independently of how gender may influence social anxiety, depression or some personality aspects, coping and expectancies should be considered in CBT-IA in both genders.

Finally, there are several limitations of this study. It is a newly developed model that needs further testing on a clinical population to fully see its clinical efficacy in treatment. It should also be tested using the longer version of the IAT (Young, 1998; Widyanto and McMurran, 2004) as a more tested measure in the literature. We used the shorter version given the length of the assessment tool we used for the whole model but if replicating this work with a clinical sample, it would be suggested to use the IAT along with additional measures of IA, such as the Assessment of Internet and Computer game Addiction as scale (AICA-S) or clinical interview (AICA-C) developed and validated with clinical groups by (Wölfling et al., 2010, 2012). Furthermore, we developed and tested the Internet use expectancies questionnaire for the purposes of this study. While we were methodologically conservative and careful in the development of the scale, this measure should be evaluated on additional populations for validity and the questionnaire needs further empirical testing in future studies. Additional and more detailed scales and interviews should also be applied to clinical samples, since most of the facets assessed in our study were measured using short questionnaires with a restrictive number of items, due to practical reasons (time limitation in the context of online surveys). A further potential problem is that of common method variance (Podsakoff et al., 2003). Unfortunately, no clear marker variable, which should theoretically be unrelated to all other variables, has been included in the study for practical reasons (the survey took almost 25 min, which is a critical threshold for online surveys). Although we cannot exclude the effect of common method variance on the results, we argue that this effect unlikely account for the whole data structure reported. When inspecting the bivariate correlations (Table 3) one can see that some of those are very low (e.g., r = –0.08, r = –0.09, r = 0.12 etc.). We think that these low correlations give some tender hints for the assumption that common method variance does not affect the main analyses dramatically. Nonetheless, the model should be tested with a systematic multi-trait-multi-method approach (Campbell and Fiske, 1959) in future studies.

The current study focuses on GIA, which means that the model on SIA, as described by Brand et al. (2014), still needs to be tested empirically. Different forms of SIA (e.g., gaming, online porn, or Internet gambling) should be tested to see if coping skills and Internet use expectancies play a similar role in development of the problem. It is also still a debate if the concept of GIA is principally adequate for covering the problematic behavior in patients. We found evidence for the link between self-reported problems related to an unspecific use of several different Internet applications and the variables suggested in the model. The concept of GIA was operationalized by the s-IAT instruction and item formulations, but also by the fact that more than 97% of the participants reported to regularly use three or more different Internet applications, such as communication, gaming, gambling, cybersex, shopping, or information seeking. From a clinical perspective, it is nevertheless a topic of debate if GIA can be a reason for seeking treatment or if treatment-seeking patients basically suffer from a loss of control over the use of one certain application, only. We suggest to consider this point in clinical research by systematically investigate the critical behavior in the context of the Internet use and analyze how frequent the uncontrolled and addictive use of more than one Internet application is in clinical samples. In addition, not all components proposed in the theoretical model on GIA could be included in this study. For example, additional personality traits or other psychopathological disorders may be included in future studies.

The main hypotheses of the model on GIA are supported by empirical data. Person’s core characteristics are related to symptoms of GIA, but these effects are mediated by person’s specific cognitions, in particular coping style and Internet use expectancies. These cognitions should be addressed in the treatment of an addictive use of the Internet.

Matthias Brand wrote the first draft of the paper, supervised the data collection, and analyzed and interpreted the data. Christian Laier contributed particularly to the conceptualization of the empirical study and data collection, and revised the manuscript. Kimberly S. Young edited the draft, revised it critically, and contributed intellectually and practically to the manuscript. All authors finally approved the manuscript. All authors are accountable for all aspects of the work.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.