Servicios personalizados
Inglés (pdf)
Articulo en XML
Referencias del artículo
Traducción automática
Enviar articulo por email
Citado por SciELO 
Citado por Google 
Similares en
SciELO 
Similares en Google 
Permalink
Introduction
During the last two decades, research has tried to conceptualize and understand the impact that the use of the Internet has had on human beings. Since Young's (1998) influential conception of "internet addiction" to describe the impulse-control disorder that accompany the use of the Internet, different interchangeable terms have been proposed, such as "Internet dependence" (Scherer, 1997), "compulsive use of the Internet" (Greenfield, 1999), "pathological use of the Internet" (Morahan-Martin & Schumacher, 2000), or "generalized problematic Internet use" (Caplan, 2010).
According to Akhter & Khalek (2020), two main perspectives were emerged from the literature to measure problematic Internet use (PIU, onwards). Firstly, Internet addiction perspective, which refers to a clinical disorder or psychopathology and an impulse control disorder (Fioravanti et al., 2013; Pontes et al., 2016; Young, 1998). Most of these formulations consider a pattern similar to that of substance addictions (Echeburúa et al., 2010; Kim & Davis, 2009; Young et al., 2011), although the use of the term "addiction" is still controversial (Carbonell et al., 2012; Gámez-Guadix, Orue, & Calvete, 2013). Some authors criticized this approach on the ground that PIU should not be characterized as a clinical disorder or a disease (Pontes et al., 2016) or a psychological dependency on a substance (Davis, 2001) rather a distinct pattern of excessive Internet use related to cognitions and behaviors (Pontes et al. 2016). Thus, in most cases, it failed to measure problems and the consequences related to excessive Internet use and also missing theoretical framework in this regard (Fioravanti et al., 2013).
Secondly, multidimensional perspective which characterize PIU more than a behavioral addiction (Fioravanti et al., 2013). Davis (2001) introduced a cognitive-behavioral model of PIU, asserting PIU results from problematic cognitions coupled with behaviors that either intensify or maintain the maladaptive response, which result in negative life outcomes (Caplan 2002; Pontes et al. 2016). Davis further classified PIU as specific and generalized PIU. Drawing on the Davis’ theory of cognitive behavior of generalized PIU Caplan (2010) developed a theory-driven, valid instrument to measure generalized PIU.
In the present study, Caplan’s (2003, 2010) research line is followed, and the generalized problematic Internet use (GPIU) term is used to refer to the pattern that includes urgency to connect to the Internet, the need to be connected often, failed attempts to disconnect, the replacement of social and family relationships with time spent connected, using the Internet to avoid problems, and the appearance of negative outcomes in daily life due to Internet use. In summary, it has been proposed, on the basis of this model, that problematic Internet use involves a set of cognitive processes (e.g., distortions or ruminative thoughts) and dysfunctional behaviors (such as Internet use to alleviate emotional distress or compulsive use) that result in a series of negative consequences in different aspects of individual life (Gámez-Guadix, Orue & Calvete, 2013).
Caplan (2010) updated the cognitive-behavioral model of problematic Internet use (Caplan, 2002, 2003; Davis, 2001) after integrating the research carried out up to that year, using four central components: the Preference for Online Social Interaction, which would influence Mood regulation through the Internet, and Deficient Self-Regulation, which is a second-order factor determined by Cognitive preoccupation and Compulsive Use. Such Deficient Self-Regulation will cause Negative outcomes of Internet use. The components of the model are described below.
Firstly, Preference for Online Social Interaction (POSI) refers to the belief that relationships through the Internet are safer, more comfortable and more effective, and less threatening than face-to-face interactions, which would, in turn, be associated with a greater problematic use of the Internet (Caplan, 2003; Caplan & High, 2011). In this vein, various studies have found that people with poor social skills, social anxiety, or isolation are more likely to make an inappropriate use of the Internet (Caplan, 2007; Cheng et al., 2015; Meerkerk et al., 2010).
Secondly, Mood regulation through the Internet refers to the use of such a medium to reduce anxiety, feelings of isolation or negative feelings (Caplan, 2002). In this way, such use acts as a dysfunctional emotional regulator (Aranda et al., 2022; LaRose et al., 2003, Mc-Kenna et al., 2002; Spada et al., 2008). In fact, it has been reported that individuals who excessively use the Internet more often connect with other users to alleviate feelings of sadness, anxiety, or loneliness than those level of use is considered normal (Muñoz-Rivas et al., 2010).
Thirdly, Deficient self-regulation includes two distinct but closely related components: Cognitive preoccupation and compulsive Internet use (Caplan, 2010). Cognitive preoccupation refers to patterns of obsessive thinking in relation to Internet use. On the other hand, Compulsive use refers to the inability to control or regulate Internet connection behavior (Griffiths, 2000; Young, 2005). Numerous previous studies have found that both components of Deficient self-regulation are central aspects of problematic Internet use (Caplan, 2010; Caplan & High, 2006; Davis et al., 2002).
Finally, the model points towards the importance of the appearance of Negative outcomes. This component denotes the extent to which an individual experiences personal, social, academic or work problems as a result of a dysfunctional Internet use. At an empirical level, numerous studies have found that Internet addiction is associated with negative outcomes in personal life, such as academic and work deterioration or absenteeism, family or interpersonal problems, abandonment of offline social activities or physical or health problems (Caplan, 2007; Gámez-Guadix et al., 2015; Gámez- Gámez-Guadix, Orue, Smith et al., 2013; Greydanus & Greydanus, 2012; Morahan-Martin, 2007; Tokunaga, 2014; Young, 2005).
This increasingly problematic behavior has led to the development of various detection tools that measure the different aspects of Internet use, particularly Internet addiction. A review run out by Laconi et al. (2014) indicated that there are at least 45 assessment tools for Internet addiction, although most of them have certain limitations and require further study to confirm their psychometric properties. The main limitations of these instruments are based on a one-dimensional conception of the GPIU (e.g. Internet Addiction Test; Young, 1998) or not following a consensual GPIU model (e.g. Internet-Related Experiences Questionnaire, Beranuy et al., 2009). Laconi et al. (2014) pointed out that the Generalized Problematic Internet Use scale-2 (GPIUS-2; Caplan, 2010) is one of the most promising measures, as it has yielded good psychometric properties, including a very consistent 4-factor cognitive-behavioral model, with only 15 items. Caplan (2002) developed the Generalized Problematic Internet Use Scale (GPIUS) from the cognitive-behavioral model of the GPIU, and later proposed a revised version, the GPIUS2 (Caplan, 2010). The model fit the data well and variables in the model accounted for 27% of the variance in mood regulation scores, 65% of variance in participants’ deficient self-regulation scores, and 61% of variance in the negative outcome scores. GPIUS2 sub-scales demonstrated good reliability ranging .82 to .87 in Cronbach’s alpha index, whereas the overall reliability of the scale was .91.
Up to the present, GPIUS2 has been widely validated through different countries and cultures (Akhter & Khalek, 2020; Assunção & Matos, 2017; Barke et al., 2012; Casale et al., 2016; Fioravanti et al., 2013; Montag et al., 2015) showing robust psychometric properties for adults (ranging .86 to .94 in Cronbach’s alpha index) and replicating Caplan's factorial structure. But these studies do not get deeper into gender differences.
Otherwise, studies in adolescent population are less frequent, when it comes to one of the most sensitive groups to this type of problematic use. Adolescents have been reported to be especially vulnerable to behavioral control, Gámez-Guadix et al. (2016) explained that this increased vulnerability is due to psychobiological and evolutionary factors, which favor the emergence of various risky behaviors (Romer, 2010; Steingerg, 2007).
Gámez-Guadix et al. (2012) analyzed the psychometric properties of the GPIUS2 with a Spanish speaking adolescents sample from Mexico. Among the first-order factors, internal consistency ranged from .74 (for the Negative outcomes subscale) to .81 (for the cognitive preoccupation subscale). For the second-order factor (Deficient self-regulation), the internal consistency was .88. The internal consistency of the entire scale was also adequate (.90). Regarding convergent validity, time spent using the Internet was significantly associated with all of the subscales of the GPIUS2. More specifically, GPIUS2 revealed the strongest correlations for the time that adolescents spent on Facebook or other social networks (r2 .44, p < .001) and the time they spent chatting or using Messenger (r2 .34, p < .001). The analysis also showed a significant correlation with the scale of interference with daily life activities, and the measures of depression, anxiety, and impulsivity. In addition to the good results for convergent validity and internal consistency, the study revealed no significant differences between male and female adolescents.
Gámez-Guadix, Orue, & Calvete (2013) team validated an adaptation of this instrument on a Spanish adolescent population of the Basque Country, yielding good psychometric properties, but they did no analyze gender invariance. In this study, authors considered the factor Generalized and problematic Internet Use as a third order factor, in place of correlations between factors proposed by Caplan. Internal consistency was adequate for all scales, ranging from .78 to .85 for first order factors. For the second-order factor (Deficient self-regulation), internal consistency was .90. The internal consistency of the entire scale was also adequate (.91) as Caplan’s original instrument (2010).
Attending to a review of recent research (Baloğlu et al., 2020), men are more susceptible to the generalized form of PIU. However, when the literature is investigated on the specific forms of PIU (e.g.: problematic social media use) recent studies found that women are at more risk and score higher than men. In the same line, Su et al. (2019) indicated that men were more likely to exhibit Internet Gaming Disorder than women and less likely to exhibit problematic social media use than women. Nevertheless, despite multiple studies indicate that men possibly have higher prevalence rates and behaviors of PIU, still more conclusive evidence is needed in terms of gender-based variations and its relationships with other variables (Baloğlu et al., 2020).
Consequently, there is a need for knowledge in terms of testing the robustness of a cognitive-behavioral model-based instrument that may be suitable for evaluating problematic Internet use in all types of samples of adolescents, in Spanish language, and regardless of gender, validated to Spanish population. The objective of the present study is threefold: on the one hand, it aims to confirm the factorial structure of the GPIUS2 instrument among an adolescent population from a province of the east of Spain, comparing the original proposal with the proposal of Gámez-Guadix, Orue, & Calvete (2013). Secondly, an analysis of the gender factorial invariance of the instrument is sought, for the first time with a Spanish sample. Finally, Caplan’s theoretical model is tested for the aforementioned sample. The starting hypotheses are the following: 1) the factorial structure of the GPIUS2 will show a good fit; 2) gender invariance of GPIUS2 will be confirmed, without significant differences in metric and scalar invariance measures between boys and girls; and 3) Caplan’s model will be confirmed.
Participants
The sample consisted of 909 students belonging to 7 Secondary Education centers in the province of Alicante, Spain, and aged between 12 and 18 years. Out of the total sample, 538 were boys (59.2%). The total mean age was 14.2 (SD = 1.6). All individuals in the sample were enrolled in one of the courses included between 1st year of Compulsory Secondary Education (CSE) and up to 2nd year of Bachiller (Spanish Baccalaureate)1
A 16.9% of the respondents presented a score at risk of suffering from generalized problematic Internet use (1 SD above the mean) and 4% of individuals obtained a score of up to 2 SD above the mean use of their group.
Measurements
Generalized problematic Internet use. The Generalized and Problematic Internet Use Scale 2 (GPIUS2; Caplan 2010) was used in its translated version, which was validated on the Spanish population (Gámez-Guadix, Orue, & Calvete, 2013). The GPIUS2 assesses different components of the problematic and generalized use of the Internet from a cognitive-behavioral perspective. It consists of 15 items grouped into four differentiated subscales: 1) Preference for online social interaction (POSI) (three items, e.g., "I feel more comfortable communicating with others online than doing it face to face"); 2) Mood regulation through the Internet (three items, e.g., "I have used the Internet to feel better when I have been sad"); 3) Negative outcomes (three items, e.g., "My use of the Internet has created problems in my life") and 4) a second order factor called Deficient self-regulation, which includes a Cognitive preoccupation subscale (three items; e.g. "When I do not connect to the Internet for a while, I start to worry about connecting") and a Compulsive Internet use subscale (three items, e.g., "I have difficulty controlling the amount of time I am connected to the Internet”). The responses are scored on a 6-point Likert- type format, from 1 (Totally disagree) to 6 (Totally agree). The internal consistency of the scale in previous studies (Caplan, 2010; Gámez-Guadix, Orue, & Calvete, 2013) was α = .91.
Procedure
Permission from the Regional Education and Research Secretary office was requested to carry out this project. Afterwards, the management teams of each of the 11 centers included in the proposal (convenience sampling) were contacted, 7 of which agreed to participate. Subsequently, the students wishing to participate in the research were asked to submit an informed consent form signed by one of their parents, with which the ethical guarantees were covered.
For the application of the test, the DetectaWeb platform for the detection of mental health disorders among children and adolescents was used (see complete protocol in Piqueras et al., 2017; for more information, see http://www.detecta-web.com/blog/). The recruited students were instructed in their respective centers to complete the questionnaire in their computer rooms. The tutors responsible for the groups were previously trained in the application of the test, and received telematic guidance.
Data Analyses
The analyses were carried out using the SPSS 22 and EQS 6.2 statistical programs.
Firstly, descriptive measures (M and SD) were included. Secondly, the scale’s factor model was tested with the EQS 6.2 program. The contrasted model included the following four general factors, taking into consideration that they were correlated: Preference for Online Social Interactions (POSI), Mood Regulation, the second order factor Deficient Self-Regulation (which includes Cognitive preoccupation and Compulsive use) and Negative Outcomes.
The analyses were carried out using the robust maximum likelihood (ML Robust) method. For the fit analyses, the following indexes were used: Chi-square χ2, Satorra Bentler chi-square S-B χ2, Robust Root Mean Square Error of Approximation (R-RMSEA), Robust Comparative Fit Index, (R-CFI), Standardized Root Mean Square Residual (SRMR). For RMSEA, values lower than .08 indicate that the model is good, and values equal to or lower than .05 indicate an optimal model (Schumacker & Lomax, 2004). The CFI value indicates a good fit with values greater than or equal to .95 (Bentler, 1990), while the SRMR index is considered good when its values are lower than .08, with values closer to .00 being more acceptable.
In order to verify the best structure for the instrument, following to Anderson & Gerbin (1988), different alterative models were tested. Model 1 will test the original structure of the GPIUS2; Model 2 will group all items with one factor; Model 3 will try the original structure but without factors correlations; Model 4 will analyze no items relations with any factor; Model 5 will test the structure found by Gámez-Guadix, Orue, & Calvete (2013) with GPIUS as a main factor having direct effect to POSI, Mood Regulation, Negative Consequences and Deficient Self-Regulation, with no correlations between factors. After verifying the adjustment of the instrument, the values of total reliability and of each factor are given by McDonald's omega index (1999).
Thirdly, through EQS 6.2, the factorial equivalence was analyzed progressively with the calculation of Factorial Invariance (FI) using the procedure suggested by Dimitrov (2010), whereby the Measurement Invariance refers to: (a) metric invariance (equality of factor loadings between groups) and (b) scalar invariance (equality of intercepts between groups). When a strong invariance measure is reached, the comparison of latent means is justified. Following, calculations were carried out to compare latent means between sexes.
Fourthly, Caplan’s theoretical model, which contains the variables measured with GPIUS2, was tested.
Results
Factorial structure analysis of the GPIUS2
Table 1 presents the descriptive statistics (means and standard deviations) for the variables that compose GPIUS2.
A confirmatory factorial analysis was performed, using the robust estimation method of maximum likelihood (ML Robust), as the Mardia index yielded a value of 117.88, indicating lack of normality in the data. For this reason, when comparing models, the comparative index was adjusted using the Satorra-Bentler Chi-square (S-B χ2) values. Regarding the treatment of cases, there were no missing values as the questionnaire administration was online and the answers were compulsory for all items.
Different models were tested (see table 2). Model 1, which presents the original structure of the GPIUS2 showed the best fit. The fit of the data to the model 1 was adequate: S-B χ2 (82) = 356.07, p < .001, SRMR = .046, R-RMSEA = .066, CI (.054, .067), NNFI = .91, R-CFI = .93. Figure 1 shows the standardized values of the factorial loadings of the model. The factor loadings varied between .55 and .88 (p < .001). Likewise, the correlations between latent variables were above .43 and were statistically significant (p < .001).
Table 2: Goodness of fit of the alternative models.
Note:SBχ2 = Satorra–Bentler scaled chi-square; R-CFI = robust comparative fit index; SRMR = standardized root mean square residual; R-NNFI = robust Non-Normed Fit Index; R-RMSEA = robust root mean square error of approximation; CI = confidence interval; (M.F): Generalized and problematic Internet use as main factor (third order factor, Gámez-Gaudix, Orue & Calvete, 2013)
McDonald's omega index (1999) was used to estimate the internal consistency of each of the subscales, as it is a more reliable measure than Cronbach's alpha (Dunn et al., 2014), all of which were adequate: Preference for online social interactions: Ω = .86; Mood regulation: Ω = .84; Cognitive preoccupation: Ω = .82; Compulsive use: Ω = .84; and Negative outcomes: Ω = .82. For the second-order factor (Deficient self-regulation), the internal consistency was Ω = .89. The internal consistency of the total scale was also adequate (Ω = .90).
Factorial invariance analysis
The factorial equivalence of Caplan’s instrument (2010) (GPIUS2) was evaluated according to gender. In order to begin to evaluate the FI, the Mardia test, which indicates the non-normality of the data, was analyzed, thus, a robust estimation method was used: The Maximum Likelihood (Robust ML). Afterwards, the fit of the model for both samples was calculated, showing an adequate adjustment in both cases (Table 3) and good factorial loadings (Table 4).
Table 3: Adjustment of the GPIUS according to sex.
Note:SBχ2 = Satorra–Bentler scaled chi-square; R-CFI = robust comparative fit index; SRMR = standardized root mean square residual; R-NNFI = robust Non-Normed Fit Index; R-RMSEA = robust root mean square error of approximation; CI = confidence interval
The next step was to test the metric invariance or equivalence of factorial loadings between the two groups (weak invariance). For this reason, the unrestricted multi-group model (Ma) was first calculated. Next, the model was calculated by establishing equality of factorial loadings of the two samples (Mb). Table 5 shows that there were no significant differences in the increase of S-B χ2 between (Ma) and (Mb). Therefore, the analysis of scalar invariance (intercepts of equivalent items between groups) was carried out (Mc). The comparison between the S-B χ2 values indicated that there were significant differences between the two models. Although these differences are not appreciated in the value of R-CFI, which decreased .008, it was decided to observe the changes proposed through the modification indexes calculated with the Lagrange Test.
This test suggested, first of all, releasing the equivalence of the intercept of item 13. With this release, the differences between the previous model (Mb) and the current model remained significant, thus the equivalence between intercepts of item 2 (Mc1) was also released. In this second case, the differences between models were not significant (p = .08). Following to Dimitrov (2010), there is no strict rule as to what degree of partial invariance is acceptable, however less than 20% freed parameters seem acceptable in practical applications. This author considers that if CFI decrease less than .01, and the increase of S-B χ2 is not significant, invariance can be considered. In the present study, both conditions were fulfilled. Therefore, it can be stated that there is metric and scalar invariance, there is strong invariance and it is justified to perform the comparison of latent means.
Table 5: Goodness of fit indexes of the models of invariance by gender.
Note:SBχ2 = Satorra–Bentler scaled chi-square; R-CFI = robust comparative fit index; R-RMSEA = robust root mean square error of approximation; CI = confidence interval; SRMR = standardized root mean square residual; ∆SBχ2 = adjusted SBχ2 difference; Ma = free model (baseline); Mb = Ma with invariant slopes; Mc = Mb with invariant intercepts; Mc1= Mb with partial invariant intercepts (freed intercept Item2 and item 13).
The results of the observed mean for each factor as a function of gender are presented in Table 6. However, the comparison of latent means was better than those observed, as the former are not associated with measurement errors (Brown, 2006).
The EQS 6.2 was used to compare latent means. This program allows to set the values of the intercepts of the factors of one of the groups to 0, thus establishing a base model on which the possible differences between the two groups are calculated. The equivalence between the intercepts of the items is also imposed. In this case, the values of the group of boys were set to 0.
The resulting model showed adequate adjustment values SB(2 (175) = 440.36; R-CFI = .94; NNFI = .90; SRMR = .055; R-RMSEA = .058 (.051- .065). As can be seen in Table 7, the comparison between the results of the two groups showed that there are only significant differences in deficient self-regulation, with their values being higher in girls (0 vs 1.2).
Caplan´s theoretical model
Figure 2 shows the direct standardized coefficients of Caplan’s theoretical model for the general population, boys and girls. The fit for this model was adequate: S-B χ2 (83) = 371.96, p < .001, SRMR = .054, R-RMSEA = .062, CI (.056, .068), NNFI = .91, R-CFI = .93. Indirect effects are offered in Table 8.
Figure 2. Results of the application of Caplan’s theoretical model (2010). Standardized values (total, boys and girls).
Table 8: Indirect Effects.
First value (bold): total sample. Second value: Boys. Third Value (cursive): Girls
The model explained 22% of the variance of mood regulation (R2 = .22), 39% of the variance of deficient self-regulation (R2 = .39), 65% of the variance of negative outcomes (R2 = .65).
Discussion
In the present article, the factorial structure of the GPIUS2 instrument was analyzed on an adolescent sample from the province of Alicante. The results confirmed hypothesis 1, showing that the GPIUS2 exhibits a good internal consistency (Caplan, 2010; Fioravanti et al., 2013; Gámez-Guadix et al., 2012). Even the structure proposed by Gámez-Guadix, Orue, & Calvete (2013) has very similar fit indices to Caplan (2010) original structure, in order to better comparison with other studies, original structure is preferred. Moreover, the adjustment of the theoretical model proposed by Caplan is adequate (hypothesis 3), as has been confirmed in previous studies (Caplan, 2010; Fioravanti et al., 2013; Gámez-Guadix et al., 2012).
It has also been shown that this instrument is equally valid for boys and girls, presenting good adjustment indexes (gender invariance). This result confirms hypotheses 2.
Only one difference between gender in Deficient self-regulation has been found, being greater in girls. It is plausible that this difference is due to the greater probability of girls presenting obsessive-compulsive responses at a community level (Martínez-González et al., 2011; Piqueras et al., 2009; Rodríguez-Jiménez et al., 2014). Machimbarrena et al. (2019), in their study, with 12285 participants (11-18 years old), found that girls had higher scores at cognitive preoccupation, compulsive use, mood regulation and negative consequences. These results are contrary to Bernal-Ruiz et al. (2017), they found that boys had significant higher score in Compulsive Internet Use and in Negative Outcomes. Must be considered that in their study sample is older (16-23), and smaller (132 boys / 178 girls), moreover they only analyzed differences in observed means, rather latent means. With these results, more research is needed in this area.
Regarding the limitations of the study, the type of sampling carried out (convenience sampling and limited to the same geographic area) and the absence of other measures apart from the self-report, should be noted as the main limitations, which hinders any generalization of the results.
It would be of interest hereafter to design longitudinal studies that account for the problematic and generalized use of the Internet in relation to other personality or transdiagnostic variables. Such information is essential when developing predictive models and prevention programs from a community level perspective.
