Grounded in the cognitive-behavioral theory (Davis, 2001), our hypothesis posited that the relationship between stressful life events and OGA would be stronger for individuals with higher maladaptive cognitions or coping. College students experience stress due to academic pressure, interpersonal conflicts, job-seeking, and other factors (Ross et al., 1999; Chi and Lin, 2005). A link between stressful life events and Internet addiction in this population has been established (Zhou and Li, 2009; Yan et al., 2014), suggesting that stressors may be a robust predictor of subsequent OGA. However, coping style as a mediator has been shown to account for much of the variance in stress levels and individuals’ stress-related problems. For example, after exposure to stress, individuals who reported a greater use of emotion-focused coping (i.e., managing emotional symptoms of stress) reported poorer sleep quality than those who reported a greater use of problem-solving coping (i.e., altering stress-causing circumstances) (Morin et al., 2003). Stress in the absence of positive coping style is a very powerful predictor for substance abuse (Grusser et al., 2007). Indeed, individuals who use compulsive game-playing to cope with daily hassles or social rejection have been found to be more susceptible to addiction (Kardefelt-Winther, 2014).

It is worthy to note that classifications and diagnostic criteria for problematic Internet use diverge (Douglas et al., 2008), characteristic of this disorder subtypes and their relations to stress life events and maladaptive coping may be also different. Davis (2001) proposed two subtypes: specific pathological Internet use (SPIU) and generalized pathological Internet use (GPIU). GPIU refers to an addiction to the Internet itself and does not concern any specific online activity, while SPIU has been defined as a overuse of specific online behavior, such as online game playing, gambling, sex, and social networking. In Davis’s (2001) model, SPIU and GPIU may result from two different kinds of maladaptive cognition process. Social context of addicts, such as social isolation or lack of social support from family or friends may contribute to the cause pathway of GPIU. GPIU might be an behavioral response for coping with pressures. However, SPIU might be the result of a pre-existing psychopathology (i.e., compulsive gambling and pornography abuse). A specific online activity served as an immediate behavioral reinforcement of this psychopathology (Laconi et al., 2015). Furthermore, maladaptive cognitive styles, such as rumination, low self-worthy and self-blame are automatically enacted when individuals face with stress or interact with a stimulation associated with the Internet, which result in either SPIU or GPIU. Our previous results showed that stressful life events contributed to GPIU, and that these effects were largely mediated through ACSs, including self-blame, fantasy, withdrawal, and rationalization (Li et al., 2009). Considering that online game addicts tend to use avoidant strategies to escape everyday stressors and to avoid daily responsibilities (Kardefelt-Winther, 2014; Cheng et al., 2015), it is reasonable to postulate that stressful life events contribute to OGA mainly in individuals with ACSs.

Among personality traits, aggressiveness and impulsivity have received major attention as predictors of OGA, whereas the role of neuroticism has been partially neglected (Charlton and Ian, 2010; Collins et al., 2012). For example, adolescents with online game overuse were more likely to have aggressive behaviors during the previous year, suggesting that aggressiveness was a relevant predictor of subsequent OGA (Ko et al., 2009b). Young adults who are characterized by high impulsivity, involvement in MMORPGs can become problematic internet user (Billieux et al., 2011).

Interestingly, findings regarding the relationship between neuroticism and OGA have been mixed. Most researchers reported that online game addicts scored higher on neuroticism (Cao and Su, 2007; Mehroof and Griffiths, 2010; Kuss et al., 2013a), but some found no differences in neuroticism between addicts and non-addicts (Wang and Gao, 2008). Understanding the reasons for these mixed results is complicated by the heterogeneous nature of the diagnostic criteria, instruments, and cultural backgrounds of the subjects. Given that drug dependence shares clinical features with OGA, and neuroticism is one of the most robust factors characterizing the drug-dependent population (Mccusker and Brown, 1991; Sigurdsso and Gudjonsso, 2009), it is important to explore the role of neuroticism, as well as other personality traits, in OGA.

Neuroticism can moderate perception of stress and its consequences (Hills and Norvell, 1991; Yu, 2013). The first way in which neuroticism may play a role in the relationship between stressful life events and OGA is by affecting for whom such a relationship is present or more pronounced. The association between Internet preference and Internet addiction was more stronger in neurotic adolescents (Lei et al., 2006; Kuss et al., 2013b). Further, neuroticism may exert a significant moderating effect in the relationship of coping styles to OGA. The notion was supported by the findings that neuroticism can predict negative coping styles, such as fantasy, rationalization, or withdrawal, especially in stressed young populations (Connor-Smith and Flachsbart, 2007). Neurotic individuals tended to be more anxious in face to face communication than Internet communication, which result in their staying online to relieve anxiety when exposed to stress (Rice and Markey, 2008). Thus, not all online game players are equally affected by stressful life events and that neuroticism may moderate the relationship between exposure to stress, ACSs, and OGA.

In summary, there is a need to explore the interaction effect of personality and coping styles on development of OGA. However, very few studies have explored interactions among neuroticism, coping strategies, stressful life events, and OGA in college students.

The aim of the present study was to examine how personality of neurotic characteristic influences the relationship between stressful life events and OGA via ACSs. Our previous works demonstrated that psychoticism and extroversion were two significant predictors for GPIU, but not neuroticism (Li et al., 2008a). The current study may contribute to PIU studies and make advances in our previous study by highlighting the different role of neuroticism in developing of GPIU and SPIU for gaming.

We propose a mixed model in which to examine mechanisms underlying the interaction of stressful life events and personality traits in predicting OGA (Figure 1). In this mixed model, the moderating effect of neuroticism on the relationship between stressful life events and OGA would be at least partially explained by the mediator variable of ACSs. On the other hand, an interaction between neuroticism and ACSs in determining development of OGA would suggest the presence of a moderated mediating effect.

In the present study, the incidence of OGA was determined in a sample of Chinese college students and the roles of ACSs and neuroticism on OGA were explored. We hypothesized: (1) that OGA incidence in Chinese college students, especially males, would be higher than in the general population; (2) that online game addicts would have higher levels of stressful life events, ACSs, and neuroticism than non-addicts; and (3) that the relationship between stressful life events and OGA would be mediated by ACSs, and that this mediation effect would vary with neuroticism.

A random sample of three universities was selected to be surveyed by placing all universities in the Guangzhou city on a database. Random allocation was then used to select those participants in collecting the sample. Each participants was asked to complete the questionnaires.

A total of 700 college students volunteered to participate in this study for no compensation. Because 46 were excluded due to missing data or lack of Internet experience, the final sample size was 654. The cohort included 298 (45.6%) males and 356 (54.4%) females. Their average age was 20.29 years (SD = 1.39; range 18–22). They reported spending an average of 19.39 h/wk (SD = 17.12; range 1–140) using the Internet and 2.62 h/wk (SD = 6.32; range 0–76) playing online games. The participants were majoring in social science (economics, business administration, education, philology, law, politics, or philosophy) or natural science (math, physics, chemistry, biology, life sciences, or computer sciences). This study was carried out in accordance with the recommendations of ‘Ethic of guidelines, The Institutional Review Board of Sun Yat-sen University’ with written informed consent from all participants. The Institutional Review Board of Sun Yat-sen University approved the protocol for this study. All participants gave written informed consent in accordance with the Declaration of Helsinki. Prior to the beginning of the study, participants were informed that the purpose of the research was to examine psychological factors associated with Internet use, and assured that their anonymity and privacy would be fully protected.

Questionnaires were administered in group format and took 20∼30 min to complete. The demographic section of the questionnaire included gender, age, major, and year in college. Participants were asked whether they played online games. Those who responded positively were asked about the duration of their playing and the average time spent playing online games each week. The full methodology has been published previously (Li et al., 2009). All the measurements used in the present study are presented in Table 1.

Simple mediation is said to occur when the causal effect of an independent variable on a dependent variable is transmitted by a mediator (Baron and Kenny, 1986). Moderated mediation occurs when mediation relations are contingent upon the level of a moderator, that is when the magnitude of a mediation effect is dependent upon a modulator (James and Brett, 1984; Muller et al., 2005). If the moderator (neuroticism in our study) differs across individuals, then moderated mediation indicates that the mediating process that intervenes between the casual factor and the outcome differs between people who differ in the moderator.

Analysis of Structural Equation Model was used to explore the interactive effects of stressful life events, ACSs and neuroticism on OGA. Stressful life events were measured using the total scores of CCSQ. OGA was measured using the total scores of OGCAS. ACSs were measured using the sum of four subscale scores: self-blame, fantasy, tolerance and rationalization. Neuroticism was measured using the subscale of Neuroticism in EPQ-RSC.

The model for this hypothesis consisted of stressful life events as an independent variable, neuroticism as a continuous moderator, ACSs as a mediator, OGA as a dependent variable, and gender and college year as control variables. According to the moderated mediation analysis steps developed by Muller et al. (2005), in equation 1, we estimated the effect of stressful life events (independent variable), neuroticism (moderator), stressful life events^∗neuroticism (interaction1) on OGA (dependent variable). In equation 2, we estimated the effect of stressful life events, neuroticism, stressful life events^∗neuroticism on ACSs (mediator). In equation 3, we estimated the effect of stressful life events, neuroticism, ACSs, ACSs^∗neuroticism (interaction2), stressful life events^∗neuroticism on OGA. As recommended by Frazier et al. (2004), all predictors were centered.

The indirect effect (moderated mediation) is calculated as (a₁ + a₃ Neuroticism) × (b₁ + b₂ Neuroticism).

Statistical analysis was performed in the Statistical Package for Social Sciences (SPSS, version 20.0 for Windows; Chicago, IL, USA) and Mplus (version 5, Muthén and Muthén, 2007). Descriptive analyses were performed on all variables. Baseline characteristics were compared between groups using the independent-sample t-test for quantitative variables and the chi-squared test for qualitative variables. Online game addict and non-addict groups were compared with independent-sample t-tests. Pearson correlation coefficients were used to assess the strengths of linear relationships between pairs of variables of interest. P-values are corrected by Bonferroni method for multiple comparisons and multiple correlations.

Moderated mediation models were estimated based on data from all 654 participants with valid measures of stressful life events, neuroticism, ACSs and OGA, controlling for relevant variables (i.e., gender and college year). Missing data are handled with full information maximum likelihood (Enders, 2001). The level of significance was set at p ≤ 0.05. For the hypothesized models, we evaluated model fit by using the comparative fit index (CFI), Tucker-Lewis fit index (TLI), root-mean-square error of approximation (RMSEA), standardized root-mean-square residual (SRMR), and Bayesian information criterion (BIC) as suggested by Chen (2007). A good model fit is indicated by a CFI larger than 0.95, a TLI larger than 0.95 and an RMSEA smaller than 0.05 (Kline, 2010).

Of the 654 participants included in this study, 31 (4.7%) were identified as online game addicts (OGCAS: 44.97, 8.14; CIAS: 5.38, 1.78), and 623 (95.3%) were non-addicts (OGCAS: 19.38, 4.78; CIAS: 1.90, 1.37). Of the 31 addicts, 21 (77.4%) were males, leading to a gender difference between addicts and non-addicts (χ² = 18.21, p < 0.001). Among non-addicts, 277 (44.5%) were males and 346 (55.5%) were females. There were no significant differences between the OGA group and the non-OGA group in terms of age, year in college, or major.

Questionnaire scores are presented in Table 2. In terms of stress, assessed with the CSSQ, the two groups’ scores were similar on the academic stress, job stress, daily hassles, interpersonal conflicts, and major events. Briefly, the CSQ revealed that the OGA group showed a greater propensity toward ACSs than the non-OGA group (self-blame: t = -3.81, p < 0.001; fantasy: t = -3.28, p = 0.001). Additionally, the OGA group’s scores were significantly higher on the neuroticism (t = -3.90, p < 0.001) subscales of the EPQ-RSC than those of the non-OGA group.

Correlations between variables of interest are presented separately for males and females in Table 3. Significant gender differences were observed, with only males showing statistically significant correlations among stressful life events, academic stress, ACSs, psychoticism and the OGCAS scores. Specifically, in males, ACSs correlated with stressful life events (r = 0.33, p < 0.001), academic stress(r = 0.23, p < 0.001), psychoticism (r = 0.34, p < 0.001) and the OGCAS scores (r = 0.25, p < 0.001).

In females, ACSs correlated with stressful life events (r = 0.27, p < 0.001), academic stress (r = 0.24, p < 0.001), and psychoticism (r = 0.25, p < 0.001). There was no significant correlation between ACSs and the OGCAS scores (Table 3).

The measurement model was first tested for an acceptable fit to the data through a CFA, and it had good fit to the data, CFI = 0.95, RMSEA = 0.052 [90% CI = 0.042–0.061].

After an acceptable measurement model was developed, the structural model was tested. Figure 2 depict the results of Structural equation modeling (SEM) analyses. The first stage was to establish a simple mediation model which OGA was regression on stress mediated by ACSs. The simple mediation model displayed satisfactory model fit indices (χ2 = 2.48, df = 3, p < 0.001, CFI = 1.00, TLI = 1.00, RESEA < 0.001, SRMR = 0.01, BIC = 6714.64). The standardized path coefficients from stressful life events to ACSs [β = 0.277 (SE = 0.043), p < 0.001] and from ACSs to OGA [β = 0.195 (SE = 0.045), p < 0.001] were significant. The path from stressful life events to OGA was also significant [β = 0.09 (SE = 0.05), p < 0.05].

The second stage was to test the moderated mediation model. The hypothesized model obtained acceptable model fit indices (χ² = 1.78, df = 1, p < 0.001, CFI = 0.99, TLI = 0.91, RESEA = 0.036, SRMR = 0.01, BIC = 9957.67). Analysis of mediation indicated that the standardized path coefficients from stressful life events to ACSs [β = 0.275 (SE = 0.039), p < 0.001] and from ACSs to OGA [β = 0.199 (SE = 0.046), p < 0.001] were significant. The path from stressful life events to OGA was non-significant [β = 0.025 (SE = 0.044), p > 0.05]. As hypothesized, ACSs fully mediated the positive relationship between stressful life events and OGA. Moreover, the interaction effect of neuroticism on stress life evens toward OGA was at a marginally significant level [β = -0.093 (SE = 0.049), p = 0.06]. Importantly, the interaction effect of neuroticism on ACS toward OGA was positive and statistically significant [β = 0.134 (SE = 0.051), p < 0.001] (Figure 2). The regression coefficients of three models are presented in Table 4. The significant interactions observed were indicative of moderated mediation, meaning that ACS (the mediating process) that intervened between stress (the independent variable) and OGA (the dependent variable) was different for people who differ with respect to neuroticism (the moderator).

The present study had several limitations. First, the cross-sectional, correlative design does not allow conclusions to be made about the causal relationships among stress, ACSs, and neuroticism on OGA. In the future, a longitudinal study with individual interviews would be needed to better elucidate the influence of stress on OGA among college students. Second, even though there is similar diversity in students’ places of origin, major and cultural circumstances in the three universities, the possible cluster effect by university should be taken into account. Third, this sample was relatively small and not fully representative of Chinese college students. Therefore, its generalizability is limited. A larger representative sample should be studied to confirm the present results.