Computers in Human Behavior 63 (2016) 823e833
Contents lists available at ScienceDirect
Computers in Human Behavior journal homepage: www.elsevier.com/locate/comphumbeh
Full length article
Psychometric validation of the Generalized Problematic Internet Use Scale 2 in a Portuguese sample Halley M. Pontes a, b, *, Scott E. Caplan c, Mark D. Grifﬁths a, b a
Psychology Division, Nottingham Trent University, Nottingham, United Kingdom International Gaming Research Unit, Nottingham Trent University, Nottingham, United Kingdom c Department of Communication, University of Delaware, USA b
a r t i c l e i n f o
a b s t r a c t
Article history: Received 28 August 2015 Received in revised form 17 May 2016 Accepted 12 June 2016
The Generalized Problematic Internet Use Scale 2 (GPIUS2) assesses individuals’ generalized problematic Internet use (PIU) cognitions, behaviors, and negative outcomes. To date, the GPIUS2 has only been validated in English, Spanish, German, and Italian language. Therefore, the aim of this study was to validate a Portuguese version of the GPIUS2 and provide a taxonomy of the potential risk of PIU among participants. A sample of 641 Portuguese-speaking Internet users was recruited online after a process of translation and back-translation of the original GPIUS2. In-depth validity and reliability analyses were conducted alongside latent proﬁle analysis (LPA) to identify the potential risk of PIU of participants. The validity and reliability analyses revealed adequate results concerning the psychometric properties of the Portuguese GPIUS2. According to the LPA results, participants were classed as “low risk” (n ¼ 289, 46.7%), “medium risk” (n ¼ 256, 40.7%), and “high risk” (n ¼ 77, 12.6%) of PIU with key differences emerging among the three classes. The present ﬁndings support the overall validity and usefulness of the Portuguese GPIUS2 and the results from the LPA may be potentially useful in informing practitioners currently working with clients struggling with PIU. © 2016 Elsevier Ltd. All rights reserved.
Keywords: Problematic Internet use Internet addiction Compulsive Internet use GPIUS2 Portuguese samples
1. Introduction 1.1. Background Ever since the ﬁrst reports of excessive and unhealthy Internet use were published almost two decades ago (e.g., Grifﬁths, 1996; Young, 1996), research in this area has grown rapidly, particularly over the last decade (Pontes, Kuss, & Grifﬁths, 2015). Some scholars conceptualize problematic Internet use (PIU) as a disease (i.e., the pathology paradigm) while others view PIU as a problem with habits and self-regulation (i.e., the cognitive-behavioral paradigm). According to Tokunaga (2015), the term problematic Internet use (PIU) has been adopted by a majority of researchers who employ the cognitive-behavioral model (i.e., Caplan, 2002, 2010; Davis, 2001) and the social-cognitive model of unregulated Internet use developed by LaRose and colleagues (i.e., Kim, LaRose, & Peng, 2009; LaRose, Eastin, & Gregg, 2001a; LaRose, Lin, & Eastin, 2003;
* Corresponding author. Nottingham Trent University, Psychology Division, Burton Street, NG14BU, Nottingham, United Kingdom. E-mail address: [email protected]
(H.M. Pontes). http://dx.doi.org/10.1016/j.chb.2016.06.015 0747-5632/© 2016 Elsevier Ltd. All rights reserved.
LaRose, Mastro, & Eastin, 2001). From these perspectives, PIU is not conceptualized as a disease, pathology, or clinical disorder. Rather, the term captures a more common and relatively less severe problem than is suggested by the Internet addiction (IA) paradigm. Accordingly, PIU is usually situated by cognitive-behavioral researchers in the middle range of the continuum [of problem severity] and emphasizes the mild, benign nature of related negative outcomes (e.g., truancy, foregoing a social event). Conversely, addiction researchers place IA at the upper end of the continuum, requiring the experience of serious negative life consequences (e.g., divorce, dropping out of school, dismissal from a job, etc.) (Tokunaga, 2015). The theoretical framework for understanding PIU initially drew upon the cognitive-behavioral theory of pathological Internet use (Davis, 2001; Davis, Flett, & Besser, 2002), which attempted to model the etiology, development, and outcomes associated with PIU. Moreover, the cognitive-behavioral model (i.e., Caplan, 2002, 2003; Davis, 2001) does not conceptualize PIU as an addiction but rather as “a distinct pattern of Internet-related cognitions and behaviors that result in negative life outcomes.” (Caplan, 2002, p. 556). Generalized PIU, which is the focus of the current study, refers
H.M. Pontes et al. / Computers in Human Behavior 63 (2016) 823e833
to “maladaptive cognitions and behaviors related to Internet use that are not linked to any speciﬁc content as individuals may develop problems due to the unique communicative context of the Internet.” (Caplan, 2002, p. 557). Conversely, speciﬁc PIU refers to the condition in which an individual uses the Internet problematically for a particular purpose (e.g., online sex, online gambling, online gaming, etc.) (Davis, 2001). From the cognitive-behavioral perspective, PIU involves a speciﬁc cycle of innate dysfunction leading to Internet use that in turn exacerbates the dysfunction (Caplan, 2003). When specifying the nature of maladaptive Internet behaviors, it is imperative to distinguish between PIU and excessive use. According to Caplan (2003, 2006), excessive Internet use involves a quantity or degree of online activity that exceeds what a person thinks of as normal, usual, or planned, whereas PIU involves difﬁculty with impulse control that plays a key role in the development of negative outcomes from Internet use. As a result, the quantity or amount of time online is not necessarily indicative of a problem because many functional Internet behaviors require excessive time online (Caplan & High, 2006; Caplan, 2003, 2006). Many studies have found PIU to be associated with a variety of psychosocial problems. For example, researchers have identiﬁed associations between PIU and increased social anxiety (Weinstein, Dorani, Elhadif, Bukovza, & Yarmulnik, 2015), higher levels of depression (Pontes, Patr~ ao, & Grifﬁths, 2014), a higher incidence of attention deﬁcit and hyperactivity disorder (Sariyska, Reuter, Lachmann, & Montag, 2015), lower levels of family functioning and life satisfaction (Wartberg, Kriston, Kammerl, Petersen, & Thomasius, 2015), increased loneliness in the educational context (Pontes, Grifﬁths, & Patr~ ao, 2014), poorer emotional well-being (Piguet, Berchtold, Akre, & Suris, 2015), and increased substance use behaviors (Kuss, van Rooij, Shorter, Grifﬁths, & van de Mheen, 2013; Rücker, Akre, Berchtold, & Suris, 2015). Although there are ongoing debates regarding the best way to conceptualize PIU (see Pies, 2009; Shaffer, Hall, & Vander Bilt, 2000; Tokunaga, 2015; Van Rooij & Prause, 2014), longitudinal evidence may help shed light on the issue. A recent longitudinal study conducted by Tokunaga (2014) of 139 undergraduate students in the US found that PIU predicted later difﬁculties in familial relationships, friendships, and academic or occupational responsibilities, even when underlying psychological problems (i.e., social anxiety, loneliness, and depression) were controlled for. Despite the study’s limitations (e.g., small sample size, use of selfreport data, etc.), the longitudinal design provided preliminary evidence that PIU may be a unique construct that can be distinguished from underlying psychiatric disorders and that may be associated with psychosocial problems. As an attempt to move forward the conceptualization and measurement of PIU, Caplan’s work (Caplan, 2002, 2003, 2005, 2010; Caplan & High, 2006, 2010) sought to clarify and expand the main constructs of the cognitive-behavioral theory of PIU (Davis, 2001) by developing two theory-driven instruments to assess generalized PIU and also empirically test the underlying relationships between the main PIU constructs (Caplan, 2002, 2003, 2010). The ﬁrst of these two instruments, the Generalized Problematic Internet Use Scale (GPIUS), assesses the prevalence of cognitions, behaviors, and the negative outcomes associated with generalized PIU (Caplan, 2002). The GPIUS originally had seven factors, six of which were related to generalized PIU cognitions (i.e., mood alteration, social beneﬁts, compulsivity, excessive time, withdrawal, and interpersonal control) and one representing the negative consequences or outcomes resulting from unhealthy Internet use (i.e., negative outcomes) (Caplan, 2002). After showing that generalized PIU was a multidimensional phenomenon, Caplan (2002) noted that two of the seven original
factors of the GPIUS (i.e., social beneﬁts and perceived social control) were of particular relevance in that they could help operationally distinguish generalized PIU from speciﬁc PIU. In addition to developing the GPIUS, Caplan also found in his initial GPIUS study that negative outcomes occurring due to PIU were associated with psychosocial health (i.e., loneliness) and PIU-related cognitions and behaviors (i.e., social beneﬁts, compulsive use, excessive time, withdrawal, and social control) (Caplan, 2002). More recently, Caplan developed the Generalized Problematic Internet Use Scale 2 (GPIUS2) (Caplan, 2010), a revised version of the GPIUS (Caplan, 2002) grounded upon Davis’ cognitivebehavioral theory of pathological Internet use (Davis, 2001), Caplan’s own works with preference for online social interaction (Caplan, 2003, 2005), and the socio-cognitive model of unregulated Internet use (i.e., Kim et al., 2009; LaRose, Eastin, et al., 2001a; LaRose et al., 2003; LaRose, Mastro, et al., 2001b). The GPIUS2 operationalizes four main constructs: preference for online social interaction, mood regulation, deﬁcient self-regulation, and negative outcomes. Preference for online social interaction is an important cognitive symptom of generalized PIU characterized by beliefs that one is safer, more efﬁcacious, more conﬁdent, and more comfortable with online interpersonal interactions and relationships than with traditional face-to-face social activities (Caplan, 2003, 2010). Additionally, mood regulation is as a cognitive symptom of generalized PIU reﬂecting individuals’ motivation to use the Internet in order to enhance their mood states. Deﬁcient selfregulation refers to the failure experienced by individuals when they attempt to adequately monitor and judge their Internet usage when trying to adjust their pattern of Internet use (Caplan, 2010). Deﬁcient self-regulation is operationalized as a higher-order construct in the GPIUS2 model reﬂecting the interplay between cognitive preoccupation (i.e., obsessive thought patterns involving the use of the Internet) and compulsive Internet use symptoms (i.e., behavioral and compulsive nature of poorly regulated Internet use) of generalized PIU (Caplan, 2010). Table 1 depicts the conceptual evolution of the GPIUS to the current GPIUS2 model. 1.2. Internet use and PIU ﬁndings in Portugal Internet use is widespread and has grown steadily over time in Portugal and yet, with the exception of a few published peerreviewed studies (e.g., Patr~ ao, Rita, & Pontes, 2013; Pontes, ~o, et al., Grifﬁths, et al., 2014; Pontes & Patr~ ao, 2014; Pontes, Patra 2014), research into Portuguese Internet use is sparse. The latest ofﬁcial report from the Portuguese government on the use of the Internet in Portugal (i.e., Instituto Nacional de Estatística, 2014) estimated that about 65% of the entire Portuguese population aged between 16 and 74 years old now use the Internet on a regular basis. Furthermore, Internet use in Portugal is more widespread amongst the younger population as the highest proportion of Internet users are those between the ages of 16 and 24 years (Instituto Nacional de Estatística, 2014). A recent study conducted by Pontes and Patr~ ao (2014) with a sample of 144 Portuguese Internet users found that 77% of the participants reported using the Internet on their cell phones while the most frequently used channel for accessing the Internet was via laptop computers (43.1%), followed by desktop computers (34%). Additionally, participants reported spending an average of 28 h a week for leisure purposes on the Internet, and had been Internet users for an average of 13 years. Additionally, a recent pilot study ~o, et al. conducted by Pontes, Grifﬁths, et al. (2014) and Pontes, Patra (2014) found that a relatively high percentage (13%) of the sample were potentially struggling with IA. However, a larger study in Portugal found a more conservative prevalence rate of IA that was
H.M. Pontes et al. / Computers in Human Behavior 63 (2016) 823e833
Table 1 The evolving model of the Generalized Problematic Internet Use Scales. GPIUS (Caplan, 2002)
Evolved to …
GPIUS2 (Caplan, 2010)
Social Beneﬁt Interpersonal Control Negative Outcomes Mood Alteration Compulsivity Withdrawal Excessive Timea
Preference for Online Social Interaction
/ / / / e
Negative Outcomes Mood Regulation Compulsive Internet Use Cognitive Preoccupation e
Note: aFactor omitted from the GPIUS2 on the basis of new empirical evidence showing that excessive amounts of time does not necessarily translated into problematic Internet use.
~o, et al., 2014). Although these ﬁgures around 1.2% (Pontes, Patra may seem incongruent due to their discrepancy, they provide some preliminary evidence that a small minority of Portuguese Internet users suffer from IA (Pontes, Grifﬁths, et al., 2014; Pontes, Patr~ ao, et al., 2014). Such ﬁndings also suggest a need for further research and better screening and assessment tools that can be used in the Portuguese cultural context. 1.3. Study aims and objectives Because most of the existing research on Portuguese samples is limited by the use of the IAT (Young, 1998), developing an alternative measure that is able to capture variables from the cognitivebehavioral model (i.e., GPIUS2) has become paramount due to the existing conceptual shortcomings present in the addiction paradigm (see Tokunaga, 2015). Consequently, several potential beneﬁts could arguably emerge from validating a measure to assess PIU in Portugal. First, Portuguese researchers will have a well validated PIU instrument that captures different aspects of PIU than the IAT measure. Second, the GPIUS2 may be a more useful tool for building and testing theory since it does not suffer from well-known shortcomings present in the IAT, such as limited utility (Kim, Park, Ryu, Yu, & Ha, 2013), and within the IA paradigm (see Starcevic & Aboujaoude, 2016; Starcevic, 2016). In light of the aforementioned rationale, the main aim of the present study was twofold: (i) to evaluate the validity and reliability of the GPIUS2 in a sample of Portuguese Internet users and (ii) to examine the GPIUS2 in relation to another similar, and yet conceptually different, instrument (i.e., the IAT) and other comorbid conditions (i.e., depression, stress, and anxiety). 2. Method 2.1. Participants and procedures A heterogeneous sample of 641 Portuguese-speaking Internet users was recruited online using convenience sampling (i.e., nonprobability sampling) over a period of 10 months (August 2014 to May 2015). Data collection was carried out online with the help of social networks websites such as Facebook, LinkedIn, and Twitter where invitations to participate in the study were sent out at different time points to recruit participants via online posts. Participants were encouraged to disseminate the survey’s link amongst their online peers. The ﬁnal sample had a relatively even gender split (52.1% male) and a mean age of 25 years (SD ¼ 9.64; range ¼ 10e74) (see Table 2). In order to develop the Portuguese version of the GPIUS2 (see Appendix A), standard back-translation protocols were adopted (i.e., Harkness, Pennell, & Schoua-Glusberg, 2004). All 15 items were subjected to a preliminary English to Portuguese translation by the ﬁrst author of this study, and then back-translated from
Table 2 Sample’s Main Socio-Demographic Characteristics and Internet use Patterns. N
Gender (male, %) Age (years) (mean, SD) Relationship status (In a relationship, %) Occupational Statusa (n, %) Working Studying Working and Studying Unemployed Daily Internet usage (hours) (mean, SD) Weekly Internet usage Age of Internet use initiation (years) (mean, SD) Preferred channel of Internet access (n, %) Cell phone Tablet Desktop computer Laptop Other Preferred location for accessing the Internet (n, %) Home Workplace School/University No speciﬁc place (e.g., mobile phone) Other
323 (52.1) 25.17 (9.6) 310 (50) 192 328 58 43 4.7 29.9 14.68
(30.9) (52.8) (9.3) (6.9) (3.4) (21.3) (7.7)
143 32 94 260 4
(26.8) (6) (17.6) (48.8) (0.8)
433 55 20 53 4
(76.6) (9.7) (3.5) (9.4) (0.7)
Note: aOccupational status at the time of survey administration.
Portuguese to English by an experienced bilingual psychologist for comparison purposes. After comparing the translated and backtranslated versions of the Portuguese GPIUS2, a ﬁnal Portuguese version of the instrument was achieved by matching both versions. The items were further reﬁned via feedback from ten Internet users that piloted the Portuguese GPIUS2. This further helped in strengthening the face validity of the newly translated instrument. 2.2. Measures 2.2.1. Socio-demographics and internet usage The survey included questions related to participants’ sociodemographic characteristics and Internet-related usage patterns (e.g., age, gender, preference of Internet access, frequency and intensity of Internet use, etc.). 2.2.2. Generalized Problematic Internet Use Scale - 2 (GPIUS2) The GPIUS2 (Caplan, 2010) is a multidimensional psychometric tool that comprises 15 items assessing the degree of generalized PIU cognitions, behaviors, and negative outcomes experienced by individuals. The scale’s factor structure (Caplan, 2010) operationalizes PIU by measuring ﬁve constructs: (i) preference for online social interaction; (ii) mood regulation; (iii) cognitive preoccupation; (iv) compulsive Internet use; and (v) negative outcomes. Each construct is measured with a 3-item subscale. The GPIUS2 includes
H.M. Pontes et al. / Computers in Human Behavior 63 (2016) 823e833
a sixth second-order factor (i.e., “deﬁcient self-regulation”) comprising the compulsive Internet use and cognitive preoccupation subscales. The GPIUS2 is rated on a 7-point Likert scale (i.e., 1 ¼ “Strongly disagree”; 2 ¼ “Disagree”; 3 ¼ “Slightly disagree”; 4 ¼ “Neutral”; 5 ¼ “Slightly agree”; 6 ¼ “Agree”; and 7 ¼ “Strongly agree”) and an overall GPIUS2 index score may be computed by summing the 15 items of the scale, which will result in scores ranging from 15 to 105. Alternatively, a composite score can be created by averaging participants’ scores on all 15 items, resulting in possible scores ranging from 1 to 7, with higher scores indicating greater intensity of generalized PIU cognitions, behaviors, and negative outcomes (Caplan, 2010). 2.2.3. Internet Addiction Test (IAT) The IAT (Young, 1998) includes 20 items that are rated on a 6point Likert (i.e., 0 ¼ “Does not apply”; 1 ¼ “Rarely”; 2 ¼ “Occasionally”; 3 ¼ “Frequently”; 4 ¼ “Often”; 5 ¼ “Always”). The IAT can be used to assess the extent of a person’s involvement with the Internet and classify Internet-related addictive behaviors in terms of mild, moderate, and severe impairment based on participant’s total score (Young, 2011). Although there are currently two different ad hoc cut-off points (i.e., cut-offs not based on empirical or clinical analyses) that were proposed for the IAT (see Young, 1998, 2011), participants’ total scores can be obtained by summing up the responses provided to all items of the IAT and can range from 0 to 100, with higher scores being indicative of greater levels of IA symptoms. In the present study, the Portuguese version of the IAT was administered as it has been previously shown to be ~o, et al., 2014). valid and reliable in this population (Pontes, Patra The Cronbach’s alpha of the IAT in the present study was 0.91. 2.2.4. Depression anxiety and stress scale - 21 (DASS-21) The DASS-21 (Lovibond & Lovibond, 1995) was employed to assess participants’ levels of depression, anxiety, and stress (e.g., “I found it hard to wind down.”). The DASS-21 comprises three 7-item subscales covering depression, anxiety, and stress symptoms using a response format on a 4-point Likert scale (i.e., 0 ¼ “Did not apply to me at all”; 1 ¼ “Applied to me to some degree or some of the time”, 2 ¼ “Applied to me a considerable degree, or a good part of the time; and 3 ¼ “Applied to me very much, or most of the time”). Scores for each scale are obtained by summing the items of each subscale and may be converted to the full scale scores by multiplying by 2, thus the total scores for each subscale can range from 0 to 42, with higher scores indicating greater levels of psychological distress. The version of the DASS-21 used in the present study has been previously shown to be valid and reliable in the Portuguese context (Pais-Ribeiro, Honrado, & Leal, 2004). The Cronbach’s alpha of the DASS-21 subscales in the present study were 0.89 (depression), 0.82 (anxiety), and 0.87 (stress).
deemed univariate outliers if they scored ±3.29 standard deviations from the GPIUS2 z-scores. This threshold was chosen because it includes around 99.9% of the normally distributed GPIUS2 z-scores (Field, 2013). As a result, one additional case was excluded. Finally, the data were also screened for multivariate outliers using Mahalanobis distances and the critical value for each case based on the chi-square distribution values, which resulted in no further exclusion of participants. Thus, the ﬁnal sample size for all subsequent analyses was N ¼ 622. All the analyses were performed using n & Muthe n, 2012) and SPSS Statistics v.20 (IBM MPLUS 7.2 (Muthe Corp, 2011). The statistical analysis comprised (i) descriptive statistics of the main sample’s characteristics (i.e., frequencies and percentages), (ii) assessment of the dimensionality and factorial structure of the GPIUS2 with conﬁrmatory factor analysis (CFA), (iii) validity analysis by investigating convergent and discriminant validity of the GPIUS2 based on the average variance extracted [AVE] coefﬁcients of each latent variable and further examination of the bootstrapped Pearson’s correlation coefﬁcients with 95% Bias-corrected and accelerated (BCa) conﬁdence intervals between the GPIUS2 and other relevant measures (i.e., IAT, depression, anxiety, stress, and daily and weekly hours spent on the Internet), (iv) analysis of the reliability of the GPIUS2 using different coefﬁcients and indicators of internal consistency (i.e., Cronbach’s alpha, composite reliability, and factor determinacies), and (v) a latent proﬁle analysis (LPA) to identify and describe the latent proﬁles of the sample in terms of their potential risk of PIU. The LPA was based on participants’ responses to the four subscales of the GPIUS2, with the resulting ﬁnal classes being subjected to Wald’s chi-square test using several socio-demographic and Internet-related variables alongside the psychological distress subscales to further explore the validity of the GPIUS2. This provided additional information on the characteristics and speciﬁcities of each identiﬁed latent class. To test the dimensionality of the GPIUS2, two main models were examined via CFA: a six-factor solution model (i.e., Model A) as originally suggested by Caplan (2010), which included preference for online social interaction, mood regulation, and negative outcomes as ﬁrst-order factors and deﬁcient self-regulation as a second-order factor comprised by the cognitive preoccupation and compulsive Internet use subscales, and a model previously validated by Fioravanti, Primi, and Casale (2013) with preference for online social interaction, mood regulation, negative outcomes, and deﬁcient self-regulation as ﬁrst-order factors (i.e., Model B), in which deﬁcient self-regulation was operationally deﬁned and modeled as a ﬁrst-order factor as opposed to a second-order factor, as in Model A.
2.3. Data analytic strategy and statistical analysis
3.1. Descriptive statistics
Prior to the statistical analyses, the data were cleaned across two stages. In the ﬁrst stage, the data were cleaned via a thorough analysis of each case presenting with missing values above the threshold of 10% in all relevant instruments of the study. As a result, 18 cases were excluded due to severe incompleteness. The second stage of the data management process involved the analysis of the (i) univariate normality of all 15 items of the GPIUS2, (ii) univariate outliers, and (iii) multivariate outliers among the dataset. As for the univariate normality, no item of the GPIUS2 had absolute values of Skewness >3.0 and Kurtosis >8.0 (Kline, 2011), thus warranting univariate normality of the study’s main measure. In order to screen for univariate outliers, a standardized composite sum score of the GPIUS2 using all 15 items was created and participants were
Participants’ main socio-demographic characteristics and Internet use patterns are presented in Table 2. In terms of relationship status, half of the participants were in a relationship (n ¼ 310, 50%). Additionally, most participants were students (n ¼ 328, 52.8%) and spent an average of 4 h (SD ¼ 3.4 h) per day and 29 h (SD ¼ 21.3 h) per week on the Internet for leisure purposes. Participants reported initiating the use of the Internet around at the age of 14 years (SD ¼ 7.7 years) while the most preferred channel for accessing the Internet was via laptop computer (n ¼ 260, 48.8%) followed by cell phone (n ¼ 143, 26.8%). Finally, participants’ most preferred location for accessing the Internet was at home (n ¼ 433, 76.6%) and the least preferred place was at the school/university (n ¼ 20, 3.5%).
H.M. Pontes et al. / Computers in Human Behavior 63 (2016) 823e833
3.2. Dimensionality and factorial validity A CFA was performed on the 15 items of the GPIUS2 using maximum likelihood estimation method with robust standard errors (MLR) on the whole sample (N ¼ 622) in order to test the two main competing models (i.e., Model A vs. Model B) of the GPIUS2 suggested by previous studies (i.e., Caplan, 2010; Fioravanti & Casale, 2015). The following ﬁt indices and thresholds were adopted to examine the goodness of ﬁt of the models being tested: c2/df [1; 4], Root Mean Square Error of Approximation (RMSEA) [0.05; 0.08], RMSEA 90% conﬁdence interval with its lower limit close to 0 and the upper limit below 0.08, p-close > 0.05, Standardized Root Mean Square Residual (SRMR) [0.05; 0.08], Comparative Fit Index (CFI) and Tucker-Lewis Fit Index (TLI) [0.90;0.95] (see Bentler & Bonnet, 1980; Bentler, 1990; Hooper, Coughlan, & Mullen, 2008; Hu & Bentler, 1999). The results obtained when testing the original six-factor solution (i.e., Model A) showed a poor overall ﬁt (c2  ¼ 374.2, c2/ df ¼ 4.6; RMSEA ¼ 0.076 [90% CI: 0.068e0.084], p-close ¼ < 0.001; SRMR ¼ 0.051, CFI ¼ 0.89; TLI ¼ 0.86). Additionally, the results of the CFA performed on Model B also yielded poor ﬁt (c2  ¼ 377.7, c2/df ¼ 4.5; RMSEA ¼ 0.075 [90% CI: 0.067e0.083], p-close < 0.001; SRMR ¼ 0.051, CFI ¼ 0.89; TLI ¼ 0.86). However, an inspection of the modiﬁcation indices, suggested covarying three error terms of items pertaining to the same factor (i.e., 13 and 14, 4 and 9, and 11 and 1). This may be a result of the relatively similar wording and underlying latent construct being assessed by these indicators. After adding these constraints, a third nested model of the GPIUS2 was tested (i.e., Model Bmodiﬁed) and the goodness of ﬁt improved considerably in comparison to the previous two models (c2  ¼ 242.9, c2/df ¼ 2.9; RMSEA ¼ 0.057 [90% CI: 0.049e0.065], pclose ¼ 0.09; SRMR ¼ 0.045, CFI ¼ 0.94; TLI ¼ 0.93). This latter model resulted in overall better ﬁt indices and acceptable standardized item loadings (i.e., lij 0.50) (see Fig. 1) and thus, the model ﬁt the data well. In order to assess which model had a statistically signiﬁcant better ﬁt, information theory ﬁt indices such as the Akaike Information Criteria [AIC], Bayesian Information Criteria [BIC], and Sample-size Adjusted Bayesian Information Criteria [SSABIC] were examined to compare the two non-nested models of the GPIUS2 (i.e., Model A vs. Model B). Additionally, a Satorra-Bentler scaled chi-square difference test (i.e., S-BDc2) was computed to ascertain the degree of ﬁt of the nested models being tested against each other (i.e., Model B vs. Model Bmodiﬁed). As shown in Table 3, Model Bmodiﬁed ﬁt the data signiﬁcantly better than Model B. 3.3. Validity and reliability analysis Convergent and discriminant validity were analyzed using the approach suggested by Fornell and Larcker (1981). Accordingly, convergent validity is achieved if a given latent construct presents an AVE 0.50 whereas discriminant validity can be demonstrated when the square root of the AVE for each latent construct is higher than the correlations between it and the rest of the latent constructs (Fornell & Larcker, 1981; Hair, Black, Babin, & Anderson, 2010). As shown in Table 4, the values obtained for the AVE of each latent construct fell between 0.50 (Deﬁcient Self-Regulation) and 0.68 (Mood Regulation). With regard to the discriminant validity of the GPIUS2, the square root of the AVE for each latent construct is also presented in Table 4 and is located in bold on the diagonal of the table. Results demonstrated that the value for each latent construct was higher than the correlations between it and the other constructs, further warranting the discriminant validity of each latent construct of the GPIUS2. In order to further examine the validity of the GPIUS2 as an
overall construct reﬂecting the degree of generalized PIU cognitions, behaviors, and negative outcomes experienced by participants, a correlational analysis using bootstrapped Pearson’s correlation coefﬁcients with 95% BCa conﬁdence intervals tested its associations to other relevant measures (i.e., IAT, depression, anxiety, stress, daily and weekly Internet use), that have been previously shown to be associated with PIU (see Barke, Nyenhuis, & €ner-Herwig, 2014; Fioravanti & Casale, 2015; Ga mez-Guadix, Kro Villa-George, & Calvete, 2012; Laconi, Tricard, & Chabrol, 2015; Meerkerk, Van den Eijnden, Franken, & Garretsen, 2010; Pontes & € Grifﬁths, 2015; Tokunaga, 2014; Ozdemir, Kuzucu, & Ak, 2014). As shown in Table 5, the associations between the overall GPIUS2 score and the main variables of interest ranged from medium to large effects. Most notably, GPIUS2 was positively associated with the IAT (r ¼ 0.75, p < 0.01) to a higher degree and positively associated with daily Internet use (r ¼ 0.30, p < 0.01) to a lower degree. These ﬁndings lend further empirical support to the overall convergent, concurrent, and criterion-related validity of the GPIUS2. Finally, the reliability of the GPIUS2, as assessed by Cronbach’s alpha, ranged from 0.78 (Negative Outcomes) to 0.86 (Deﬁcient Self-Regulation) (see Table 4). Additionally, factor determinacies n were all well above the conventional threshold of 0.80 (Muthe n, 2012; Mo nok et al., 2012; Schembre & Geller, 2011) & Muthe while all composite reliability coefﬁcients also exceeded the desired threshold of 0.70 (Fornell & Larcker, 1981; Hair et al., 2010). Overall, these results illustrate that the GPIUS2 demonstrated adequate validity and reliability at several levels. 3.4. Latent Proﬁle Analysis (LPA) An LPA was carried out in order to characterize participants’ potential risk of PIU based on their class membership. The LPA procedure is a mixture modeling statistical procedure used to identify groups of individuals (categorical latent variables) that give similar responses to speciﬁc continuous variables (Collins & Lanza, 2010), which in the context of the present study were participants’ responses to the four latent constructs comprising the GPIUS2 (i.e., preference for online social interaction, mood regulation, negative outcomes, and deﬁcient self-regulation). The LPA was conducted with two to four classes and multiple ﬁt indices were used to help establish the optimal number of latent classes, such as (i) the AIC, BIC, and SSABIC with lower values indicating more parsimonious models, (ii) the entropy criterion which is used in order ascertain the accuracy of classifying participants into their respective proﬁles (i.e., higher values suggesting better ﬁt), and (iii) the Lo-MendellRubin Adjusted Likelihood Ration Test (L-M-R Test) to help determining the ﬁnal number of classes, where a signiﬁcant p value (i.e., <0.05) indicates that the tested model ﬁts better than the previous n & Muthe n, 2012). one (Muthe Table 6 summarizes the results obtained from the LPA, with two to four classes, that was performed on the four dimensions of the GPIUS2. Using the aforementioned criteria, the three-class solution was chosen as it was deemed the optimal solution for the data collected. Although the values of the information theory ﬁt indices (i.e., AIC, BIC, and SSABIC) did not decrease convincingly and the entropy values obtained suggested that the two-class solution had the best value for this statistic while the four-class solution the worst, a further inspection of the L-M-R test values and their levels of signiﬁcance clearly indicated that the four-class solution should be rejected in favor of the preceding three-class solution, even though the decrease in the information criteria ﬁt indices were not too consistent across the three class solutions as one would have expected. The ﬁnal three latent classes and their characteristics are
H.M. Pontes et al. / Computers in Human Behavior 63 (2016) 823e833
Fig. 1. Conﬁrmatory factor analysis of the generalized Problematic Internet Use Scale 2 (Model Bmodiﬁed). Abbreviations: POSI: Preference for online social Interaction; MR: Mood regulation; DSR: Deﬁcient self-regulation; NO: Negative outcomes.
Table 3 Summary of the goodness of ﬁt statistics for the various nested and non-nested model comparisons of the Generalized Problematic Internet Use Scale 2 (GPIUS2). Model
Model A Model B Model Bmodiﬁed
30,677 30,689 30,514
30,912 30,915 30,753
30,744 30,753 30,582
e e 767.23
e e 1
e e <0.001
0.89 0.89 0.94
0.86 0.86 0.93
0.076 0.075 0.057
<0.001 <0.001 0.09
Note: Model A: initially proposed model for the GPIUS2 scale as in Caplan (2010); Model B: the four ﬁrst-order factors model with deﬁcient self-regulation as a ﬁrst-order factor as suggested by Fioravanti et al. (2013); Model Bmodiﬁed: Model B with the inclusion of covariances between the three residuals of the items 13 and 14, 4 and 9, and 11 and 1. Abbreviations: AIC: Akaike Information Criteria; BIC: Bayesian Information Criteria; SSABIC: Sample-size Adjusted Bayesian Information Criteria; S-BDc2: Satorra-Bentler scaled Chi-square difference test; Ddf: difference in degrees of freedom between nested models; p: p-value of the S-BDc2; CFI: Comparative Fit Index; TLI: Tucker-Lewis Index; RMSEA: Root Mean Square Error of Approximation; p-close: Close ﬁtting model probability.
presented in Fig. 2. The ﬁrst identiﬁed class (i.e., “low risk”) features Internet users presenting a low risk of PIU (n ¼ 289, 46.7%) as they scored generally low on all four dimensions of the GPIUS2, and thus exhibited lower levels of PIU. Additionally, participants belonging to this class had an average score of 24.68 on the GPIUS2 (SD ¼ 7.28, 95% CI [23.84e25.52]). Moreover, the second identiﬁed class (i.e., “medium risk”) included Internet users exhibiting a relatively medium risk of PIU (n ¼ 256, 40.7%). The main feature of Internet users in this class is that they scored markedly high on the “Mood
Regulation” dimension and low on the remaining dimensions of the GPIUS2. Furthermore, the mean score for these participants on the GPIUS2 was 43.06 (SD ¼ 9.90, 95% CI [41.84e44.28]). Finally, the third class (i.e., “high risk”) included a smaller proportion of users presenting the highest risk for developing PIU (n ¼ 77, 12.6%) as their scores on all four dimensions of the GPIUS2 were generally higher than the other two classes. Although participants in this class are not considered to be problematic Internet users per se, they scored particularly high on the “Mood Regulation”, “Deﬁcient
H.M. Pontes et al. / Computers in Human Behavior 63 (2016) 823e833
Table 4 Reliability, convergent, and discriminant validity of the Generalized Problematic Internet Use Scale 2. Construct
POSI MR NO DSR
0.80 0.84 0.78 0.86
0.91 0.96 0.91 0.93
0.77 0.86 0.78 0.85
0.54 0.68 0.54 0.50
Correlation matrix POSI
0.73 0.42 0.55 0.51
0.82 0.39 0.47
Abbreviations: AVE: Average Variance Extracted; POSI: Preference for Online Social Interaction; MR: Mood Regulation; NO: Negative Outcomes; DSR: Deﬁcient SelfRegulation. Note: The Cronbach’s alpha obtained for all 15 items was 0.90. The coefﬁcients in bold on the diagonal of the table represent the square root of the Average Variance Extracted for each latent construct.
Table 5 Bootstrapped correlation matrix with 95% Bias-corrected and accelerated conﬁdence interval between the GPIUS2 and relevant study variables. Measure
BCa 95% CI
IAT Depression Anxiety Stress Daily Internet use Weekly Internet use
0.75* 0.42* 0.32* 0.37* 0.30* 0.32*
0.71;0.77 0.34;0.50 0.23;0.40 0.30;0.47 0.20;0.39 0.23;0.40
0.56 0.18 0.10 0.14 0.10 0.10
Abbreviations: GPIUS2: Generalized Problematic Internet Use Scale 2; IAT: Young’s Internet Addiction Test. Note: Bootstrap results are based on 100.000 bootstrap samples; * Correlation is signiﬁcant at 0.01.
6 5.088 5 4.352
4 3.348 3
Self-Regulation”, and “Negative Outcomes” dimensions. Participants belonging to this class had an observed mean score of 65.16 on the GPIUS2 (SD ¼ 9.75, 95% CI [62.95e67.37]), which was signiﬁcantly greater than the other two classes. In order to help characterize the ﬁnal three classes obtained in the LPA, while also providing further evidence of the GPIUS2’s validity (i.e., convergent, concurrent, and criterion-related validity), Wald’s Chi-square tests were computed. The Wald’s chi-square test of mean equality is used for latent class predictors in mixture modeling as it takes into account the probabilistic nature of the LPA classes (see www.statmodel.com/download/meaantest2.pdf for further information). Therefore, the three classes were compared across a relevant set of variables related to PIU such as gender, age, relationship status, age of Internet use initiation, daily and weekly Internet usage, GPIUS2, IAT, depression, anxiety, and stress total scores. As shown in Table 7, age, relationship status, and age of Internet use initiation were not signiﬁcantly different across the three classes. However, the gender variable differed signiﬁcantly between the “medium risk” and “high risk” classes (c2 ¼ 4.5, p ¼ 0.03), with the “high risk” group containing a greater proportion of males. In terms of daily and weekly hours spent on the Internet, differences were relatively consistent and steady across the three classes with participants showing increased risk of PIU spending generally more hours on the Internet during their leisure
POSI MR Low risk (n = 289; 46.7%) High risk (n = 77; 12.6% )
DSR NO Medium risk (n = 256; 40.7%)
Fig. 2. The three-class solution obtained from the Latent Proﬁle Analysis (LPA) and the potential risk of problematic Internet use associated with each class on the basis of participants’ responses to the four subscales of the Generalized Problematic Internet Use Scale 2 (GPIUS2). Abbreviations: POSI: “Preference for Online Social Interaction”; MR: “Mood Regulation”; DSR: “Deﬁcient Self-Regulation”; NO: “Negative Outcomes”.
time. Finally, all three classes differed signiﬁcantly in terms of participants’ overall scores on the GPIUS2, IAT, depression, anxiety, and stress scales, with participants with membership to the “high risk” class exhibiting higher levels of PIU and psychological distress in general in comparison to the “low risk” and “medium risk” classes. These ﬁndings provide preliminary support for the validity of the Portuguese GPIUS2.
4. Discussion The main aims of the present study were to further investigate the psychometric properties of the GPIUS2 and validate it in the Portuguese cultural context using a heterogeneous sample of Portuguese-speaking Internet users. To achieve these goals, several statistical analyses examined and compared competing models of the GPIUS2 that had been previously established in the literature. The analyses also assessed the validity and reliability of the
Table 6 Summary of the goodness of ﬁt obtained from the latent proﬁle analysis (LPA)a. Number of Latent classes
2 3 4
7968 7702 7609
8025 7782 7711
7984 7724 7638
0.959 0.893 0.864
708 267 99
<0.001 <0.001 0.27
Abbreviations: AIC: Akaike Information Criteria; BIC: Bayesian Information Criteria; SSABIC: Sample-size Adjusted Bayesian Information Criteria; L-M-R Test: Lo-MendellRubin Adjusted Likelihood Ratio Test Value; P: p value associated with the L-M-R Test. Note: aThe best loglikelihood value has been successfully replicated across all analyses even after a twofold increasing of the random starts. The coefﬁcients in bold of the table represent the square root of the Average Variance Extracted for each latent construct.
H.M. Pontes et al. / Computers in Human Behavior 63 (2016) 823e833
Table 7 Comparison of the three latent classes: Testing equality for latent class predictors. Low risk (n ¼ 289)
Gender (Male %) Age (years), Mean (SE) Relationship status (In a relationship %) Age of Internet use initiation (years), Mean (SE) Daily Internet usage (hours), Mean (SE) Weekly Internet usage (hours), Mean (SE) GPIUS2 (total score), Mean (SE) IAT (total score), Mean (SE) Depression (total score), Mean (SE) Anxiety (total score), Mean (SE) Stress (total score), Mean (SE)
53.3 (0.03)ab 25 (0.62)a 53 (0.03)a 14.5 (0.49)a 4.1 (0.21)a 25 (1.23)a 16.8 (0.82)a 28.6 (0.71)a 2.4 (0.20)a 1.7 (0.16)a 3.2 (0.21)a
Medium risk (n ¼ 256)
47.2 25.2 49 14.4 4.6 30.1 34.1 37.1 3.9 2.5 4.6
(0.03)a (0.60)a (0.03)a (0.51)a (0.20)a (1.32)b (1.26)b (0.80)b (0.28)b (0.21)b (0.27)b
High risk (n ¼ 77)
62 25.6 40.7 14.19 6 39.1 58.8 53.9 6.5 3.8 6.5
(0.06)b (1.03)a (0.06)a (0.79)a (0.46)b (3.04)c (2.5)c (1.34)c (0.61)c (0.42)c (0.49)c
Overall test Wald’s c2
5.8 0.2 3.1 0.1 12.5 19.1 257.5 239.2 39.9 22.3 36.9
0.05 0.89 0.22 0.93 <0.01 <0.01 <0.01 <0.01 <0.01 <0.01 <0.01
Note: Means having different subscript letters are different on at least p < 0.05 level according to the pairwise Wald’s chi-square test of mean equality for latent class predictors in mixture modeling (http://bit.ly/NNCxju).
Portuguese GPIUS2 at several levels using different statistical indicators. Finally, but importantly, LPA and Wald’s chi-square tests were performed on the data to ascertain participants’ latent proﬁles and establish their potential risk of PIU based on the GPIUS2 scores. As suggested by the results of the CFA, the best ﬁtting factor solution found for the GPIUS2 in the present study included preference for online social interaction, mood regulation, negative outcomes, and deﬁcient self-regulation as ﬁrst-orders factors. Thus, deﬁcient self-regulation was operationally deﬁned in this model as a single latent factor comprising the indicators pertaining to the compulsive Internet use and cognitive preoccupation factors and not as a second-order latent factor that includes compulsive Internet use and cognitive preoccupation separately. Furthermore, the factor structure of the GPIUS2 found in the present study replicates the one reported in a similar Italian study (Fioravanti & Casale, 2015). There may be two possible, non-mutually exclusive, explanations for this ﬁnding. At the sampling level, the main characteristics of both the Italian and Portuguese samples are more similar in nature (i.e., Latin) than the sample included in Caplan and High’s (2010) study (i.e., US students). Furthermore, at the methodological level, several validation studies of the GPIUS2 (see Barke et al., mez-Guadix, Orue, & Calvete, 2013; Ga mez2014; Caplan, 2010; Ga Guadix et al., 2012) found very high correlations (i.e., close to 1) between the deﬁcient self-regulation factor and compulsive Internet use and cognitive factors when deﬁcient self-regulation was operationalized as second-order factor. This suggests that deﬁcient self-regulation may be best conceptualized as combination of compulsive behavior and cognitive preoccupation. For mez-Guadix et al. (2013) found in a CFA that the instance, Ga regression path from deﬁcient self-regulation leading to cognitive preoccupation was 0.99. Moreover, Barke et al. (2014) found that deﬁcient self-regulation had an association of 0.90 with cognitive preoccupation and 0.99 with compulsive Internet use respectively. These could also be potentially indicative of multicollinearity between these factors due to the overlap between them Thus, operationalizing deﬁcient self-regulation as a ﬁrst-order factor might not only be best from a statistical point of view, but also from a conceptual standpoint as this factor structure appears to reﬂect a simpler and less complicated structure of the GPIUS2. This ﬁnding both adds to and extends the existing literature on PIU by providing empirical evidence concerning questions that recent studies have raised about how to best conceptualize deﬁcient self-regulation mez-Guadix et al., 2013). (e.g., Barke et al., 2014; Ga Regarding the validity of the GPIUS2, the results provided highly robust evidence in terms of convergent, discriminant, concurrent, and criterion-related validity. Speciﬁcally, convergent validity was
warranted both at the subscale (i.e., each latent construct of the GPIUS2 had an AVE 0.50) and overall construct level (i.e., as per the positive association between the GPIUS2 and the IAT). Moreover, discriminant validity was demonstrated at the subscale level (i.e., the square root of the AVE for each latent construct of the GPIUS2 was higher than the correlations between it and the other constructs). Additionally, the results demonstrated concurrent validity with the positive associations found between the GPIUS2 and the other scales (i.e., depression, anxiety, and stress scales). Finally, the scale demonstrated criterion-related validity with a positive association found between the GPIUS2 and daily and weekly Internet use (as presented in Tables 5 and 7). Taken as a whole, these results are in line with recent studies that found the GPIUS2 to be correlated with the IAT (Barke et al., 2014; Fioravanti & Casale, 2015), as well as several other studies showing that PIU is positively associated with depression (Barke et al., 2014; G amezGuadix et al., 2012), anxiety (Akin & Iskender, 2011; G amezGuadix et al., 2012), stress (Akin & Iskender, 2011; Jun & Choi, , Benka, Sebena, & Gajdosova , 2014), and time 2015; Orosova mez-Guadix et al., 2012; Siciliano et al., 2015). The spent online (Ga results also indicate the GPIUS2 is reliable and internally consistent, as assessed by several indicators such as the Cronbach’s alpha, composite reliabilities, and factor determinacies. In order to characterize participants’ potential risk of PIU, an LPA analysis was conducted. Interestingly, the results revealed three latent classes according to participants’ risk of PIU and their responses to the GPIUS2. Participants in the “low risk” class accounted for almost half of the sample (46.7%) and were characterized as experiencing very few PIU cognitions, behaviors, and/or negative outcomes. Participants with “medium risk” of PIU represented 40.7% of the total sample and tended to use the Internet more often as a way of enhancing their mood, as demonstrated by the markedly high scores on the mood regulation subscale of the GPIUS2. The third and ﬁnal class comprised 12.6% of the total sample and featured participants showing “high risk” of PIU cognitions, behaviors, and negative outcomes due to Internet use. Overall, these results illustrate distinct underlying patterns and differences of Internet use among participants with varying degrees of PIU severity. The current paper advances the literature on PIU by employing LPA analysis to shed further light on how the GPIUS2 captures varying degrees of PIU severity and how the scale may be used to identify the latent proﬁles of problematic Internet users. Procedures such as Latent Class Analysis (LCA) and LPA are gaining popularity in the gaming studies ﬁeld. For instance, Wartberg et al. (2015) conducted a large study in a representative quota sample of 1723 young adolescents from Germany in order to characterize the
H.M. Pontes et al. / Computers in Human Behavior 63 (2016) 823e833
sample based on their latent proﬁles and establish the prevalence rate of PIU among German adolescents. In their study, ﬁve latent proﬁles of adolescents Internet users were found and a class of “high risk” Internet users comprising 3.2% of the total sample (57.1% male) was used to establish the prevalence rate of PIU. The authors further reported that Internet users classed as “high risk” scored signiﬁcantly higher in the Compulsive Internet Use Scale (CIUS) (Meerkerk, Van Den Eijnden, Vermulst, & Garretsen, 2009) in comparison to the other proﬁles identiﬁed in the LPA, and showed that length of time spent using the Internet as well as the highest frequency of difﬁculties were associated with PIU. In another recent study using LCA and the CIUS to establish the prevalence of PIU in a large representative sample in Germany (i.e., Rumpf et al., 2014), the authors found six latent classes of Internet users. Furthermore, Rumpf and colleagues found that the small proportion of participants belonging to the sixth class scored signiﬁcantly higher on the CIUS and spent more hours on the Internet during the week in comparison to the remaining classes. These ﬁndings are similar to those in the present study in which the overall score on the GPIUS2 increased in the “high risk” group in comparison to the other two groups. Moreover, future studies could expand on these ﬁndings by further examining how well the GPIUS2 identiﬁes and replicates the classes found here in other samples and cultural contexts. However, more studies, including Portuguese samples, should be carried out in order to better characterize how PIU may be impacting on the Portuguese society as a whole since little is known about PIU in this context. The present study makes an important contribution to advancing research on the cognitive-behavioral model of problematic Internet use by validating a measure to be used on Portuguese populations as an alternative to IA measures, which has been criticized for failing to fully operationalize the construct (see Tokunaga & Rains, 2016). Although the present study provides preliminary evidence for the validity and reliability of the Portuguese GPIUS2, as well as offering a number of important contributions to the literature, it is not without its limitations. The current study used a convenience sample of Portuguese-speaking Internet users that was not necessarily representative of all Internet users in Portugal. Therefore, the present ﬁndings should be cautiously interpreted and not
generalized to a broader population. Additionally, the use of selfreport questionnaires may have resulted in biases such as social desirability biases and short-term recall biases. Despite these limitations, the ﬁndings of this study provide robust evidence of validity and reliability of the Portuguese GPIUS2. The results suggest that the GPIUS2 is a valid alternative measure of PIU that can be used in the Portuguese cultural context. Furthermore, practitioners working with clients struggling with PIU may beneﬁt from the ﬁndings of the present study as they can be potentially useful in providing further information regarding clients’ patterns of Internet use and the implicit factors associated with their motivation and/or consequences for using the Internet excessively. Put simply, while some people engage in maladaptive Internet use because they use it extensively as a means to cope with negative mood states or as a result of their overwhelming preference for socializing online, other types of Internet users may experience negative outcomes due to their preference for online social interaction, cognitive preoccupation, and compulsive use. Overall, the present study provides further evidence that PIU is both a cognitive and behavioral phenomenon, and that it warrants further study in Portugal. An additional contribution of the study is that it tested the relationships hypothesized by the cognitive-behavioral model of PIU among a population that has not previously been examined, and the results also provide further support to previous studies that have examined some of these relationships. Acknowledgements This work was supported by the Portuguese Foundation for ~o para a Cie ^ncia e Tecnologia Science and Technology (Fundaça [FCT]) with a grant (number: SFRH/BD/112857/2015) that was awarded to Halley M. Pontes. The supporting agency had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; and preparation, review, or approval of the article. Appendix A
Escala do Uso Generalizado Problem atico da Internet 2 (EUGPI2)1,2,3 Items Item Item Item Item Item Item Item Item Item Item Item Item Item Item Item
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
~o ~o face-a-face. Preﬁro a interacç~ ao social online em relaça a comunicaça Usei a Internet para falar com outras pessoas quando me senti sozinho(a). ~o estou online por algum tempo, começo a preocupar-me com a ideia de me conectar. Quando na Tenho diﬁculdade em controlar a quantidade de tempo que passo online. Tenho diﬁculdades em gerir a minha vida por causa da Internet. vel com a interacç~ Sinto-me mais conforta ao social online do que a interacç~ ao face- a-face. Usei a Internet para me sentir melhor quando estava em baixo. ~o me pudesse conectar a Internet. Sentir-me-ia perdido(a) se na difícil controlar o meu uso da Internet. Sinto que e Perdi compromissos ou actividades sociais por causa do meu uso da Internet. -face. Preﬁro comunicar-me com as pessoas online em vez de face-a Usei a Internet para me sentir melhor quando estava chateado(a). ~o estou na Internet. Penso obsessivamente em estar online quando na ~o estou na Internet, e difícil resistir ao impulso de me conectar. Quando na O meu uso da Internet criou problemas na minha vida.
1 ~ es: Tendo em conta a seguinte escala, avalie em que medida concorda ou discorda com cada uma das seguintes aﬁrmaço ~es Instruço ~o proﬁssional ou acade mico. Isto e , apenas considere o uso por lazer tanto no computador como em relativamente ao uso da Internet na Internet. qualquer outro tipo dispositivo com acesso a 2 Escala de 7-pontos: 1: Discordo totalmente; 2: Discordo; 3: Discordo um pouco; 4: Neutro; 5: Concordo um pouco; 6: Concordo; 7: Concordo totalmente. 3 ~es: prefere ^ncia pela interacç~ ~o do humor (2, 7 e 12); auto-regulaça ~o deﬁciente (3, 8, 13, 4, 9 e Dimenso ao social online (1, 6 e 11); regulaça ^ncias negativas (5, 10 e 15). 14); conseque
H.M. Pontes et al. / Computers in Human Behavior 63 (2016) 823e833
References Akin, A., & Iskender, M. (2011). Internet addiction and depression, anxiety and stress. International Online Journal of Educational Sciences, 3, 138e148. €ner-Herwig, B. (2014). The German version of the Barke, A., Nyenhuis, N., & Kro Generalized Problematic Internet Use Scale 2 (GPUIS2): A validation study. Cyberpsychology, Behavior, and Social Networking, 17(7), 474e482. http:// dx.doi.org/10.1089/cyber.2013.0706. Bentler, P. M. (1990). Comparative ﬁt indexes in structural models. Psychological Bulletin, 107(2), 238e246. http://dx.doi.org/10.1037/0033-2909.107.2.238. Bentler, P. M., & Bonnet, D. G. (1980). Signiﬁcance tests and goodness of ﬁt in the analysis of covariance structures. Psychological Bulletin, 88(3), 588e606. http:// dx.doi.org/10.1037/0033-2909.88.3.588. Caplan, S. E. (2002). Problematic internet use and psychosocial well-being: Development of a theory-based cognitiveebehavioral measurement instrument. Computers in Human Behavior, 18(5), 553e575. http://dx.doi.org/10.1016/S07475632(02)00004-3. Caplan, S. E. (2003). Preference for online social interaction a theory of problematic internet use and psychosocial well-being. Communication Research, 30(6), 625e648. http://dx.doi.org/10.1177/0093650203257842. Caplan, S. E. (2005). A social skill account of problematic internet use. Journal of Communication, 55(4), 721e736. Caplan, S. E. (2006). Problematic Internet use in the workplace. In M. Anandarajan, T. S. H. Teo, & C. A. Simmers (Eds.), The internet and workplace transformation (pp. 63e79). Armonk, NY: M. E. Sharpe. Caplan, S. E. (2010). Theory and measurement of generalized problematic internet use: A two-step approach. Computers in Human Behavior, 26(5), 1089e1097. http://dx.doi.org/10.1016/j.chb.2010.03.012. Caplan, S. E., & High, A. C. (2006). Beyond excessive use: The interaction between cognitive and behavioral symptoms of problematic internet use. Communication Research Reports, 23(4), 265e271. http://dx.doi.org/10.1080/ 08824090600962516. Caplan, S. E., & High, A. C. (2010). Online social interaction, psychosocial well-being, and problematic internet use. In K. Young, & C. Nabuco de Abreu (Eds.), Internet addiction: A handbook and guide to evaluation and treatment (pp. 35e53). Hoboken, NJ: John Wiley & Sons, Inc. Collins, L. M., & Lanza, S. T. (2010). Latent class and latent transition analysis: With applications in the social, behavioral, and health sciences. Hoboken, NJ: John Wiley & Sons. Davis, R. A. (2001). A cognitive-behavioral model of pathological internet use. Computers in Human Behavior, 17(2), 187e195. http://dx.doi.org/10.1016/S07475632(00)00041-8. Davis, R. A., Flett, G. L., & Besser, A. (2002). Validation of a new scale for measuring problematic internet use: Implications for pre-employment screening. CyberPsychology & Behavior, 5(4), 331e345. http://dx.doi.org/10.1089/ 109493102760275581. Field, A. (2013). Discovering statistics using ibm spss statistics (4th ed.). London: SAGE Publications Ltd. Fioravanti, G., & Casale, S. (2015). Evaluation of the psychometric properties of the Italian internet addiction test. Cyberpsychology, Behavior, and Social Networking, 18(2), 120e128. http://dx.doi.org/10.1089/cyber.2014.0493. Fioravanti, G., Primi, C., & Casale, S. (2013). Psychometric evaluation of the Generalized Problematic Internet Use Scale 2 in an italian sample. Cyberpsychology, Behavior, and Social Networking, 16(10), 761e766. http://dx.doi.org/10.1089/ cyber.2012.0429. Fornell, C., & Larcker, D. F. (1981). Evaluating structural equation models with unobservable variables and measurement error. Journal of Marketing Research, 18(1), 39e50. http://dx.doi.org/10.2307/3151312. mez-Guadix, M., Orue, I., & Calvete, E. (2013). Evaluation of the cognitiveGa behavioral model of generalized and problematic internet use in Spanish adolescents. Psicothema, 25(3), 299e306. http://dx.doi.org/10.7334/ psicothema2012.274. mez-Guadix, M., Villa-George, F. I., & Calvete, E. (2012). Measurement and Ga analysis of the cognitive-behavioral model of generalized problematic internet use among Mexican adolescents. Journal of Adolescence, 35(6), 1581e1591. http://dx.doi.org/10.1016/j.adolescence.2012.06.005. Grifﬁths, M. D. (1996). Internet “addiction”: An issue for clinical psychology? Clinical Psychology Forum, 97, 32e36. Hair, J. F., Black, W. C., Babin, B. J., & Anderson, R. E. (2010). Multivariate data analysis. A global perspective (7th ed.). New Jersey: Pearson Prentice Hall. Harkness, J., Pennell, B. E., & Schoua-Glusberg, A. (2004). Survey questionnaire translation and assessment. In S. Presser, J. M. Rothgeb, M. P. Couper, J. T. Lessler, E. Martin, J. Martin, et al. (Eds.), Methods for testing and evaluating survey questionnaires (pp. 453e473). Hoboken, NJ: John Wiley & Sons. Hooper, D., Coughlan, J., & Mullen, M. R. (2008). Structural equation modelling: Guidelines for determining model ﬁt. Electronic Journal of Business Research Methods, 6(1), 53e60. Hu, L. T., & Bentler, P. M. (1999). Cutoff criteria for ﬁt indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling: A Multidisciplinary Journal, 6(1), 1e55. http://dx.doi.org/10.1080/ 10705519909540118. IBM Corp. (2011). IBM SPSS statistics for windows, version 20.0. New York: IBM Corp. Instituto Nacional de Estatística. (2014). Mais de metade dos utilizadores da internet fazem-no em mobilidade. Lisbon. Retrieved from: http://www.ine.pt/xportal/
xmain?xpid¼INE&xpgid¼ine_destaques&DESTAQUESdest_ boui¼211422849&DESTAQUESmodo¼2&xlang¼pt. Jun, S., & Choi, E. (2015). Academic stress and internet addiction from general strain theory framework. Computers in Human Behavior, 49, 282e287. http:// dx.doi.org/10.1016/j.chb.2015.03.001. Kim, J., LaRose, R., & Peng, W. (2009). Loneliness as the cause and the effect of problematic internet use: The relationship between internet use and psychological well-being. CyberPsychology & Behavior, 12(4), 451e455. http:// dx.doi.org/10.1089/cpb.2008.0327. Kim, S. J., Park, D.-H., Ryu, S.-H., Yu, J., & Ha, J. H. (2013). Usefulness of young’s internet addiction test for clinical populations. Nordic Journal of Psychiatry, 67(6), 393e399. http://dx.doi.org/10.3109/08039488.2012.748826. Kline, R. B. (2011). Principles and practice of structural equation modeling (3rd ed.). New York: Guildford Press. Kuss, D. J., van Rooij, A. J., Shorter, G. W., Grifﬁths, M. D., & van de Mheen, D. (2013). Internet addiction in adolescents: Prevalence and risk factors. Computers in Human Behavior, 29(5), 1987e1996. http://dx.doi.org/10.1016/j.chb.2013.04.002. Laconi, S., Tricard, N., & Chabrol, H. (2015). Differences between speciﬁc and generalized problematic internet uses according to gender, age, time spent online and psychopathological symptoms. Computers in Human Behavior, 48, 236e244. http://dx.doi.org/10.1016/j.chb.2015.02.006. LaRose, R., Eastin, M. S., & Gregg, J. (2001a). Reformulating the internet paradox: Social cognitive explanations of internet use and depression. Journal of Online Behavior, 1(2). No Pagination Speciﬁed. LaRose, R., Lin, C. A., & Eastin, M. S. (2003). Unregulated internet usage: Addiction, habit, or deﬁcient self-regulation? Media Psychology, 5(3), 225e253. http:// dx.doi.org/10.1207/S1532785XMEP0503_01. LaRose, R., Mastro, D., & Eastin, M. S. (2001b). Understanding internet usage: A social-cognitive approach to uses and gratiﬁcations. Social Science Computer Review, 19(4), 395e413. http://dx.doi.org/10.1177/089443930101900401. Lovibond, P. F., & Lovibond, S. H. (1995). The structure of negative emotional states: Comparison of the depression anxiety stress scales (DASS) with the Beck depression and anxiety inventories. Behaviour Research and Therapy, 33(3), 335e343. http://dx.doi.org/10.1016/0005-7967(94)00075-U. Meerkerk, G. J., Van Den Eijnden, R. J. J. M., Vermulst, A. A., & Garretsen, H. F. L. (2009). The compulsive internet use scale (cius): Some psychometric properties. CyberPsychology & Behavior, 12(1), 1e6. http://dx.doi.org/10.1089/ cpb.2008.0181. Meerkerk, G. J., Van den Eijnden, R. J. J. M., Franken, I. H. A., & Garretsen, H. F. L. (2010). Is compulsive internet use related to sensitivity to reward and punishment, and impulsivity? Computers in Human Behavior, 26(4), 729e735. http://dx.doi.org/10.1016/j.chb.2010.01.009. nok, K., Berczik, K., Urba n, R., Szabo, A., Grifﬁths, M. D., Farkas, J., et al. (2012). Mo Psychometric properties and concurrent validity of two exercise addiction measures: A population wide study. Psychology of Sport and Exercise, 13(6), 739e746. http://dx.doi.org/10.1016/j.psychsport.2012.06.003. n, L. K., & Muthe n, B. O. (2012). Mplus user’s guide seventh edition. Los Muthe n & Muthe n. Angeles: Muthe , O., Benka, J., Sebena, R., & Gajdosov Orosova a, B. (2014). Gender, optimism, perceived stress on problematic internet use among Slovak university students. In C. Pracana (Ed.), Psychology applications & developments (pp. 306e315). Lisbon: Science Press. € Ozdemir, Y., Kuzucu, Y., & Ak, S¸. (2014). Depression, loneliness and internet addiction: How important is low self-control? Computers in Human Behavior, 34, 284e290. http://dx.doi.org/10.1016/j.chb.2014.02.009. Pais-Ribeiro, J., Honrado, A., & Leal, I. (2004). Contribuiç~ ao para o estudo da ~o portuguesa das escalas de ansiedade, depressa ~o e stress (EADS) de 21 adaptaça itens de Lovibond e Lovibond [Contribution to the Portuguese validation study of Lovibond and Lovibond’s short version of the depression anxiety and stress scale (DASS)]. Psicologia, Saúde & Doenças, 5(2), 229e239. Patr~ ao, I. M., Rita, J. S., & Pontes, H. M. (2013). Internet addiction and loneliness among portuguese elementary school students: An exploratory quantitative study. Atencion Primaria, 45(Supplment 2), 132e188. http://dx.doi.org/10.1016/ S0212-6567(13)70034-9. Pies, R. (2009). Should DSM-V designate “internet addiction” a mental disorder? Psychiatry (Edgmont), 6(2), 31e37. Piguet, C., Berchtold, A., Akre, C., & Suris, J.-C. (2015). What keeps female problematic internet users busy online? European Journal of Pediatrics, 1e7. http:// dx.doi.org/10.1007/s00431-015-2503-y. Pontes, H. M., & Grifﬁths, M. D. (2015). The role of age, age of internet access initiation, and time spent online in the etiology of internet addiction. Journal of Behavioral Addictions, 4(Suppl. 1), 30e31. doi: 10.1556/JBA.4.2015.Suppl.1. ~o, I. M. (2014). Internet addiction and loneliPontes, H. M., Grifﬁths, M. D., & Patra ness among children and adolescents in the education setting: An empirical i de l’Esport, 32(1), pilot study. Aloma: Revista de Psicologia, Ci encies de l’Educacio 91e98. Pontes, H. M., Kuss, D. J., & Grifﬁths, M. D. (2015). Clinical psychology of internet addiction: A review of its conceptualization, prevalence, neuronal processes, and implications for treatment. Neuroscience and Neuroeconomics, 4, 11e23. http://dx.doi.org/10.2147/NAN.S60982. ~o, I. M. (2014). Estudo explorato rio sobre as motivaço ~es perPontes, H. M., & Patra cebidas no uso excessivo da internet em adolescentes e jovens adultos [An exploratory study on the perceived motivations underpinning excessive internet use among adolescents and young adults]. Psychology, Community & Health, 3(2), 90e102. http://dx.doi.org/10.5964/pch.v3i2.93.
H.M. Pontes et al. / Computers in Human Behavior 63 (2016) 823e833 ~o, I. M., & Grifﬁths, M. D. (2014). Portuguese validation of the Pontes, H. M., Patra internet addiction test: An empirical study. Journal of Behavioral Addictions, 3(2), 107e114. http://dx.doi.org/10.1556/JBA.3.2014.2.4. Rücker, J., Akre, C., Berchtold, A., & Suris, J.-C. (2015). Problematic internet use is associated with substance use in young adolescents. Acta Paediatrica. http:// dx.doi.org/10.1111/apa.12971. Rumpf, H. J., Vermulst, A. A., Bischof, A., Kastirke, N., Gürtler, D., Bischof, G., et al. (2014). Occurrence of internet addiction in a general population sample: A latent class analysis. European Addiction Research, 20(4), 159e166. http:// dx.doi.org/10.1159/000354321. Sariyska, R., Reuter, M., Lachmann, B., & Montag, C. (2015). Attention deﬁcit/hyperactivity disorder is a better predictor for problematic internet use than depression: Evidence from Germany. Journal of Addiction Research & Therapy, 6(209). http://dx.doi.org/10.4172/2155-6105.1000209. Schembre, S. M., & Geller, K. S. (2011). Psychometric properties and construct validity of the weight-related eating questionnaire in a diverse population. Obesity, 19(12), 2336e2344. http://dx.doi.org/10.1038/oby.2011.96. Shaffer, H. J., Hall, M. N., & Vander Bilt, J. (2000). “Computer addiction”: A critical consideration. American Journal of Orthopsychiatry, 70(2), 162e168. http:// dx.doi.org/10.1037/h0087741. Siciliano, V., Bastiani, L., Mezzasalma, L., Thanki, D., Curzio, O., & Molinaro, S. (2015). Validation of a new short problematic internet use test in a nationally representative sample of adolescents. Computers in Human Behavior, 45, 177e184. http://dx.doi.org/10.1016/j.chb.2014.11.097. Starcevic, V. (2016). Internet addiction: Rarely an addiction and not the Internet that one is addicted to. In Paper presented at the 3rd international conference on behavioral addictions, Geneva. Starcevic, V., & Aboujaoude, E. (2016). Internet addiction: Reappraisal of an increasingly inadequate concept. CNS Spectrums, 1e7. http://dx.doi.org/10.1017/ S1092852915000863.
Tokunaga, R. S. (2014). A unique problem or the manifestation of a preexisting disorder? The mediating role of problematic internet use in the relationships between psychosocial problems and functional impairment. Communication Research, 41(4), 531e560. http://dx.doi.org/10.1177/0093650212450910. Tokunaga, R. S. (2015). Perspectives on internet addiction, problematic internet use, and deﬁcient self-regulation: Contributions of communication research. In E. L. Cohen (Ed.), Communication yearbook, 30 pp. 131e161). New York: Routledge. Tokunaga, R. S., & Rains, S. A. (2016). A review and meta-analysis examining conceptual and operational deﬁnitions of problematic internet use. Human Communication Research, 42(2), 165e199. http://dx.doi.org/10.1111/hcre.12075. Van Rooij, A., & Prause, N. (2014). A critical review of “internet addiction” criteria with suggestions for the future. Journal of Behavioral Addictions, 3(4), 203e213. http://dx.doi.org/10.1556/JBA.3.2014.4.1. Wartberg, L., Kriston, L., Kammerl, R., Petersen, K. U., & Thomasius, R. (2015). Prevalence of pathological internet use in a representative German sample of adolescents: Results of a latent proﬁle analysis. Psychopathology, 48(1), 25e30. http://dx.doi.org/10.1159/000365095. Weinstein, A., Dorani, D., Elhadif, R., Bukovza, Y., & Yarmulnik, A. (2015). Internet addiction is associated with social anxiety in young adults. Annals of Clinical Psychiatry, 27(1), 2e7. Young, K. S. (1996). Psychology of computer use: xl. addictive use of the internet: A case that breaks the stereotype. Psychological Reports, 79(3), 899e902. http:// dx.doi.org/10.2466/pr0.19220.127.116.119. Young, K. S. (1998). Caught in the net: How to recognize the signs of internet addiction e And a winning strategy for recovering. New York: John Wiley & Sons, Inc. Young, K. S. (2011). Clinical assessment of internet-addicted clients. In K. S. Young, & C. Abreu (Eds.), Internet addiction: A handbook and guide to evaluation and treatment (pp. 19e34). New Jersey: John Wiley & Sons.