The Internet plays an increasingly noticeable role in adolescents’ daily lives. It has become the perfect tool for learning and information retrieval, for personal and academic developing, for communication and interacting, and for self-expression and leisure. The recent data show the high level of Internet use in adolescents. In Spain 92.0% of 10–15 year-old-children use Internet (National Statistics Institute, 2014). This figure rises to 96% in Germany (Statistisches Bundesamt, 2012), and 95% of adolescents aged 12–17 years are Internet users in the US (Madden, Lenhart, Duggan, Cortesi, & Gasser, Reference Madden, Lenhart, Duggan, Cortesi and Gasser2013). While recognizing the great benefits of Internet, adolescent Internet use is an issue of concern for professionals, researchers and administrations (Choi et al., Reference Choi, Son, Park, Han, Kim, Lee and Gwak2009; Muñoz-Rivas, Fernández, & Gámez-Guadix, Reference Muñoz-Rivas, Fernández and Gámez-Guadix2010; Ybarra, Alexander, & Mitchell, Reference Ybarra, Alexander and Mitchell2005).

On the one hand, adolescents can be particularly vulnerable to some dangers and risks associated with the Internet use (Huang et al., Reference Huang, Unger, Soto, Fujimoto, Pentz, Jordan-Marsh and Valente2014; Kormas, Critselis, Janikian, Kafetzis, & Tsitsika, Reference Kormas, Critselis, Janikian, Kafetzis and Tsitsika2011). According to the European study EU Kids Online, 14% of 9–16 years old have seen pornography online, 15% of children aged 11–16 say that they have seen or received sexual messages on the Internet, and 3% say they have sent or posted such messages. In relation to online bullying, 6% have been sent nasty or hurtful messages online, and 3% have sent such messages to others (Livingstone, Haddon, Görzig, & Ólafsson, Reference Livingstone, Haddon, Görzig and Ólafsson2011).

On the other hand, the possible problematic or addictive Internet use has attracted significant research interest since Young (Reference Young1996) presented the first paper referring to Internet addiction. The vast volume of scientific literature published bears witness to the interest shown in the subject, but more than 15 years later, there is no consensus on prevalence data of this phenomenon in adolescents (Chou, Condron, & Belland, Reference Chou, Condron and Belland2005; Moreno, Jelenchick, Cox, Young, & Christakis, Reference Moreno, Jelenchick, Cox, Young and Christakis2011). Values differ considerably in light of recent European studies results. For instance, the EU Kids Online survey showed that 1% of children aged 11–16 could be said to show pathological levels of Internet use (Smahel et al., Reference Smahel, Helsper, Green, Kalmus, Blinka and Ólafsson2012). Tsitsika, Tzavela, Mavromati, & EU NET ADB Consortium (Reference Tsitsika, Tzavela and Mavromati2012) found that 1.2% of adolescents aged 14–17 years presented with Internet Addictive Behavior (IAB), and 12.7% with at risk IAB. In the meanwhile, the percentage of adolescents with pathological Internet use was an estimated 4.4%, and maladaptive Internet use prevalence among adolescents was estimated at 13.5% by Durkee et al. (Reference Durkee, Kaess, Carli, Parzer, Wasserman, Floderus and Waserman2012). In the Asian countries the Internet addiction prevalence among adolescents ranges from 1.2% in South Korea, 2.2% in China, 3.1% in Japan to 4.9% in Philippines (Mak et al., Reference Mak, Lai, Watanabe, Kim, Bahar, Ramos and Cheng2014). A recent meta-analysis, including 89,281 participants from 31 nations across seven world regions, tries to amalgamate data to give more conclusive evidence, and estimates an Internet Addiction global prevalence of 6.0% (Cheng & Li, Reference Cheng and Li2014).

Limitations of both the academic-scientific and institutional studies hamper the comparability of results, the data integration and the correct current situation diagnosis. The variability in prevalence rates across the studies may be attributed to variations in definitions of the concept, diagnostic instruments and the samples used. First of all, there is a wide terminological heterogeneity to overcome: Internet addiction (Young, Reference Young1996), compulsive Internet use (Greenfield, Reference Greenfield1999), pathological Internet use (Morahan-Martin & Schumacher, Reference Morahan-Martin and Schumacher2000), problematic Internet use (Caplan, Reference Caplan2002), Internet dependency (Rahmani & Lavasani, Reference Rahmani and Lavasani2011) are some of the most commonly used terms. Since such disorders are not nowadays neither listed in the International Classification of Diseases (ICD-10) nor in the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5), a current prudent solution is the use of the term problematic Internet use, for the sake of the rigor and while it is waiting for the necessary consensus (Gómez, Rial, Braña, Varela, & Barreiro, Reference Gómez, Rial, Braña, Varela and Barreiro2014; Thatcher & Goolam, Reference Thatcher and Goolam2005). However it is interesting to mention that the Internet Gaming Disorder was included in the Section III of the DSM- 5 (American Psychiatric Association [APA], 2013) – describing clinical conditions possibly to be defined as formal disorders in future editions of DSM. Moreover some review of definitions of diagnosis and international attempts to standardize that diagnosis have been recently carried out (King & Delfabbro, Reference King and Delfabbro2014; Petry et al., Reference Petry, Rehbein, Gentile, Lemmens, Rumpf, Möβle and O’Brien2014).

Different tests, tools and scales have been developed to try to measure the magnitude of the problem, according to different conceptual frameworks (Chen, Weng, Su, Wu, & Yang, Reference Chen, Weng, Su, Wu and Yang2003; Demetrovics, Szeredi, & Rózsa, Reference Demetrovics, Szeredi and Rózsa2008; Young, Reference Young1996), although some of them do not have adequate psychometric properties (Aboujaoude, Reference Aboujaoude2010). Furthermore, the generalization of the results from the sample to the population is affected by small sample sizes or samples taken from a very specific environment (for instance, a city or a university in particular). Additionally, the samples are extracted from different target population: young people, adolescents, students, minors… and the data collection procedure used in any study varies from personal interviews, household surveys, face-to-face questionnaires to online forms.

It is worth noting that, although the currently available data suffer from the above-mentioned limitations, much progress has been made in the understanding of the phenomenon in general terms (Ólafsson, Livingstone, & Haddon, Reference Ólafsson, Livingstone and Haddon2013; Spada, Reference Spada2014). However, irrespective of these percentages or average values estimated by different studies and research projects, there is room for thinking that all adolescents do not behave online in the same way but there are different use patterns and users profiles. There is a growing need of better understanding of the issue, which means that it would be interesting to introduce a new focus of interest that has hardly been explored until now: the identification and characterization of different adolescent Internet users’ profiles.

As a result of the literature review about this topic, some needs and lacks have been detected. First of all, there are not many studies, and especially they did not use a Spanish adolescent sample. Some a priori or classic segmentations have been carried out, based upon socio-demographic variables such as gender, age or cultural context (Aslanidou & Menexes, Reference Aslanidou and Menexes2008; Li & Kirkup, Reference Li and Kirkup2007; Lin & Yu, Reference Lin and Yu2008). However, fewer studies have stated as their aim the identification of Internet users’ profiles (Hill, Beatty, & Walsh, Reference Hill, Beatty and Walsh2013; Lee, Reference Lee2010), and they are usually focused on characterizing and defining only those users who have problems. Thus, for instance, Tsitsika et al. (Reference Tsitsika, Tzavela and Mavromati2012) based on the qualitative information from 124 interviews of adolescents with signs of Internet Addictive Behavior (IAB), found four types of Internet users at risk: the stuck online, the juggling it all, the coming full cycle, and the killing boredom. Roberts, Foehr, Rideout, and Bordi (Reference Roberts, Foehr, Rideout and Bordi2004) differentiated among six adolescent users’ typologies by using cluster analyses: Media Lite, Interactor, VidKid, Restricted, Indifferent, and Enthusiasts. Chen and Tzeng (Reference Chen and Tzeng2010) identified three profiles for the college female heavy Internet users and three profiles for the college male heavy Internet users, and Lee (Reference Lee2010) found three groups of college students depending on their online life-styles: a Social group, an Individual group, and a Loner group. A recent meta-analysis and review of the research literature on media-user typologies (Brandtzæg, Reference Brandtzæg2010) suggested eight different user types: Non- users, Sporadics, Debaters, Entertainment users, Socializers, Lurkers, Instrumental users, and Advanced users.

This paper seeks to take a step forward in this field and proposes an ad hoc segmentation, to combine classic socio-demographic variables with others, such as adolescents’ attitudes and perceptions. It is of utmost importance to know the here-and-now data, considering the dynamism of the digital reality and the possible sociocultural connotations which could affect the adolescent’s online behavior. The objective of this study is therefore to identify and characterize the different adolescents’ categories or profiles in relation to Internet, nowadays in Spain. To achieve this goal means to reach three specific objectives. Firstly, the existence of different adolescents segments in terms of attitudes and perceptions toward Internet and Social Networks will be confirmed. Secondly, given that attitudes towards use are related to actual behavior (Ajzen, Reference Ajzen1991) it will be checked to what extent those different attitudinal segments are also different in relation to their Internet use. Thirdly, the characterization of each segment will be further developed, adding new variables that result in a more comprehensive analysis and characterization.

This type of segmentation conducted through an identification of variables associated (firstly from an attitudinal perspective, and then from behavioral and socio-demographic viewpoints) will yield two benefits. It will result in a correct differentiation and a precise characterization of each and every profile, and it will result in a deeper understanding of adolescents’ Internet usage, especially their risk-taking behaviors.

Method

Participants

In pursuit of its purpose, a selective methodology and a cross design were used. A survey of the Compulsory Secondary Education students from Galicia was carried out. A two-stage sampling design was used that resulted in the selection of 29 first-stage units and 2,339 sample individuals. The selection of the first-stage units (schools) was done by cluster random sampling with stratification according to type of school (public or private/subsidized), area (rural or urban), and Galician provinces, respecting population quotas. The second stage frame was stratified by gender, and grade. The final sample consisted of 2,339 Compulsory Secondary Education students from Galicia, 1,171 girls and 1,168 boys, between the ages of 11 and 18 (M = 13.77, SD = 1.34). Of these, 1,619 attended public schools and 720 attended private or subsidized schools; 1,239 studied in lower secondary education (1st and 2nd grade) and 1,100 were in upper secondary education (3rd and 4th grade).

Instruments

Data collection was undertaken through the application of a questionnaire developed for a broader study that included three different blocks: (1) a set of questions about Internet usage habits, motivations and barriers, information about the Internet, devices and applications availability, knowledge and use of Social Networks and the parents’ role in this context; (2) a scale of attitudes and perceptions toward Internet and Social Networks (the scale has not a specific name), comprising 12 Likert type items with five answer options ranging from 1 Strongly disagree to 5 Strongly agree, (Rial, Gómez, Braña, & Varela, Reference Rial, Gómez, Braña and Varela2014); and (3) a last block including socio-demographic data (Gender, Age, Grade, School and Province).

The above-mentioned scale is grouped in three factors (Negative exo-attitude, Positive endo-attitude and Risk) confirmed by a Confirmatory Factor Analysis (GFI = .97; AGFI = .96; CFI = .95; NFI = .94; TLI = .93; RMSEA [CI] = .053 [.047 – .058]), and it has adequate internal consistency (α = .80). The first Factor of the attitudes scale (Negative exo-attitude) covered different aspects related to how the subjects see the relationship between adolescents and Internet and Social Networks (see Table 1). The second Factor (Positive endo-attitude) covered the relationship between oneself and these, and finally the third factor (Risk) analyzed the evidence of possible problematic Internet use.

Table 1.

Items of the Attitude Scale

Procedure

Data were collected in their own classrooms through an individual questionnaire, in small groups (no more than 20 individuals), after prior detailed explanation of the corresponding instructions. The information was collected by a group of researchers from the Universidad de Santiago de Compostela with extensive experience in carrying out this type of work. The Bioethics Committee of the Universidad de Santiago de Compostela approved this study. Participants were informed about the purpose of the study, and repeatedly assured of the complete anonymity and confidentiality of their responses. This study was carried out with the consent and cooperation from both the school leadership and respective parents’ associations, and the participation was entirely voluntary.

Data analysis

The initial sample was composed by 2,339 subjects, although 327 had missing data in some of the attitudinal variables. A two-stage cluster analysis was carried out over the 2,012 left, in order to identify profile subtypes. In the first exploratory stage (CLUSTER procedure of the PASW Statistic 18), Euclidian distances were chosen as similarity measures because of the metric character of the data (Hair & Black, Reference Hair, Black, Grimm and Yarnold2002). Three different clustering methods were used. A preliminary single-linkage procedure to identify potential outliers was first performed (Mooi & Sarstedt, Reference Mooi and Sarstedt2011). A total of 16 outliers were identified and eliminated from the analyses, so that the final sample comprised 1,996 subjects. Ward’s algorithm was applied to the final sample, as this is more powerful than other agglomerative clustering techniques that use F values to maximize differences among clusters (Hair & Black, Reference Hair, Black, Grimm and Yarnold2002). The coefficients (within-cluster sum of squares) were checked in the agglomeration schedule in order to look for significant changes at each combination step (Burns & Burns, Reference Burns and Burns2009). The dendrogram gave support to the quantitative criterion, finding that the four-cluster option was the optimal solution. Finally, data were analyzed with the Average Linkage (Within Group) procedure to establish the reliability of the Ward solution (Hair & Black, Reference Hair, Black, Grimm and Yarnold2002). The resulting dendrogram was structurally similar to the previous one, with 67% agreement in terms of assignment of participants to specific clusters.

In the second stage of the cluster analysis, a k-means clustering was conducted (QUICK CLUSTER of the PASW) by using the means of the four-cluster solution as the starting points (seeds). This procedure was used to improve assignment of subjects to clusters and to obtain the final solution. Iterative procedures such as the k-means procedure are more powerful and reliable than hierarchical procedures but need prior specification of the number of clusters and of the initial centers. After the k-means analysis, 32% of individuals were assigned to another cluster, getting groups with greater internal homogeneity and more balanced in size.

The resulting groups or subtypes were compared on demographic and cognitive variables not included in the original clustering process to validate and adequately characterize each of the clusters. Qualitative or ordinal variables were analyzed with cross-tables, using chi-square analyses to assess global significance and adjusted residuals e ^adj (Haberman, Reference Haberman1973) to assess the significance in each cell. Quantitative variables were analyzed with the ANOVA procedure, choosing Scheffé post hoc test. The values of F and Chi-squared statistical tests, with details of the degree of significance and size of effect (Partial Eta-squared and Cramer’s V) were specified.

Results

Table 2 shows that the clustering solution provides statistically significant distances between the four clusters on most of the attitudinal variables used to define the four clusters. The Group A has the lowest levels in all variables except Hooked on Internet/networks, Excessive/unsuitable use and Accessing more than should, which are slightly lower on Group B. The second group differentiates especially from Group A on Connecting daily, Used to meet people, Like adding friends and Like making comments. Again the third group shows higher levels than Group B on almost all variables, particularly on Accessing more than should and Time flying when connected. The Group C has consistently the highest levels on all variables except Excessive/unsuitable use, higher on Group C. Particularly high with respect to the rest of the groups is Losing hours of sleep.

Table 2.

Means and Standard Deviations of Clustering Variables by Group

Note: ^{a, b, c, d} Clusters with means statistically different (Scheffé Test; α = .05).

All clusters are balanced with respect to their size. Statistically significant differences in the three socio-demographic variables were found. In relation with gender, groups A and D comprise a similar percentage of male and female teenagers (see Table 3). Group B has an almost 10% higher rate in men than women, just the opposite that happens in group C. Nevertheless, major differences exist with regard to age and educational level, variables that can be interpreted in the same line. The average age increases significantly with group number. Means start at 13.4 years old in Group A and end up in 14.2 years old in Group D. Groups A and B are mostly composed of younger adolescents (1^st and 2^nd year) while groups C and D are mostly composed of older subjects (3^rd and 4^th year).

Table 3.

Descriptive Statistics of Demographic Variables by Group

Note: ^{a, b, c, d} Significantly different clusters (Scheffé test; α = .05).

^{+, –} Significant (positive or negative) associations between the cluster and the category of variable (standardized residuals; α = .05).

Further, the four groups are significantly related to the use their members make of Internet and Social Networks (Table 3). Main variables associated with the attitudes towards Internet are parental involvement (Cramer’s V = .455), use of Social Networks (V = .425), accessing time per day (V = .377), reasons why they do not connect more time (V = .376) and frequency of use (V = .371). Adjusted residuals e ^adj are approximately independent and distributed as standard normal. Values greater than 1.96 or less than -1.96 represent a significant departure from the expected value at a 95% confidence level. These residuals are useful in visualizing contingency table data, making it immediately apparent which cells are out of line with expectations, in which direction, and by how much. For instance, Table 4 shows that Group A (37.8% of its members) and Group B (56.0%) have a significantly lower percentage of people accessing the Internet on a daily basis (negative superscript), while Group C (70.2%) and Group D (82.7%) have a significantly higher percentage. These data have been used to profile the four groups and get the most distinct characteristics of each of them.

Table 4.

Descriptive Statistics of Main Usage Variables by Group (α = .05)

Note: * = p < .05; ** = p < .01; *** = p < .001.

^{+, −} Significant (positive or negative) associations between the group and the category of variable (adjusted residuals; α = .05).

The four cluster solution is highly interpretable. Group A teenagers, the youngest ones, can be characterized as First Steppers. They access the Internet very seldom and still do not show much interest on it. They are starting to learn the basics about Social Networks and still hardly used them on a regular basis. Group B teenagers make up the next step in the evolutionary process of familiarity with social media. They access Internet weekly, mainly to be on Social Networks and particularly on Tuenti, a Spanish-based Social Network specifically designed for and fancied by university and High School people. They are starting to get some expertise with social media tools and would like to know a bit more about them. Thus, this stage has been labeled as Trainees.

Group C has been labeled as Sensible Users in order to differentiate them from the following group. Both (Groups C and D) consider they have already a good knowledge about Internet and Social Networking, but this group shows a responsibility in their use which seems to lack the last one. Sensible users access Internet for about 1–2 hours a day and do not spend more time because they have to devote time on their homework. Their parents show a real control over their use of the Internet and do not let them access after sleep time. Finally, Group D reports the highest levels of use and mastery of the Internet, not only on Social Networking (webcam included) but also downloading, watching online contents, buying goods or services or even betting, and from multiple devices. They are the Heavy Users. The key distinctive feature with respect Group C is that there is no real parental control over their use, keeping connected to the Internet till wee hours. The complete description of each of the four groups is summarized in Table 5.

Table 5.

Internet Usage Aspects that Differentiate Best between the Groups