Sample
The total sample comprised 12,753 individuals ranging in age from 16 to 70 years (M = 21.48, SD = 6.95); 8 452 (66.27%) of the participants were women. The mean age of female and male participants was comparable (female: M = 21.30 years, SD = 6.27; male: M = 21.85 years, SD = 8.11). We invited individuals of secondary and high schools, and colleges or universities in several main cities of Poland to take part in the study. Eligibility criteria included participants having a Facebook account. It should be noted that, despite the age requirement for setting up a Facebook account, there were individuals who were under 16 years of age who had their own Facebook account. These individuals were included in the study.
For the factor structure and reliability analyses, the total sample was used. For criterion-related validity analyses, the five subsamples from the total sample were used. More specifically, the criterion-related validity analyses included the following number of participants: subsample 1–662 participants (422 women; age: M = 21.06, SD = 6.26 ); subsample 2–290 participants (234 women; age: M = 22.95, SD = 2.26); subsample 3–857 participants (785 women; age: M = 19.61, SD = 2.58); subsample 4–131 participants (88 women; age: M = 21.05, SD = 2.05) who played video games in the last year; and subsample 5–47 participants (20 women; age: M = 20.32, SD = 4.07) who were receiving alcohol and drug treatment. In subsample 5, individuals receiving alcohol and drug treatment included young individuals with alcohol and drug diagnoses according to the ICD 10 (WHO, 1993) and no other diagnosis indicating psychotic and neurological disorders. These individuals were under the care of an addiction treatment centres in Mazowieckie and Lubuskie voivodships. In addition, individuals of similar age and gender as individuals in the treatment group were selected from total sample for the control group (N = 47, 20 women; age: M = 20.36, SD = 4.01). Individuals in control group currently used no drugs and did not have addiction diagnosis according to the ICD 10 (WHO, 1993). Participation in the study was voluntary and participants were assured that their responses were anonymous. All the procedures applied were in accordance with the ethical standards of the responsible committee on human experimentation (institutional and national) and with the Helsinki Declaration of 2005.
Other measures
The Problem Videogame Playing Questionnaire (Tejeiro et al., 2015), which has been adapted to Polish was used to assess gaming addiction. It comprises nine statements rated by subjects using a dichotomous scale. A greater number of positive responses provided by a subject corresponds with a stronger compulsion to play games. Because of its high clinical accuracy, the questionnaire is one of the best tools currently applied in research into Internet Gaming Disorder (King et al., 2013). The questionnaire has adequate psychometric properties: Cronbach’s alpha equals 0.69.
The Problematic Internet Use Test (Hawi et al., 2015) which is a polish adaptation of Kimberly Young’s Internet Addiction Test was used to assess Internet addiction. It consists of 22 items in which subjects provide answers on a 6-point scale. It has good psychometric properties: Cronbach’s alpha equals 0.935; discriminatory power of the items in the range from 0.40 to 0.70; and split-half reliability of 0.95 with correlation between the halves amounting to 0.91.
The Facebook Intensity Scale (Ellison et al., 2007) consists of eight items which measure the intensity and frequency of Facebook usage as well as emotional attitude to Facebook and its impact on daily activities. Responses are given on a Likert scale from 1 – strongly disagree to 5 – strongly agree. Higher scores indicate greater involvement in Facebook use. Cronbach's alpha for the Polish version was 0.78.
The Bergen Facebook Addiction Scale (BFAS) (Andreassen et al., 2012) has been validated by (Atroszko et al., 2017). The scale includes six items which are based on addiction components described by (Griffiths, 2005). Responses are given on a Likert scale from 1 – very rarely to 5 – very often. In a previous study (Atroszko et al., 2017), the Cronbach’s alpha reliability coefficient was 0.86. In current study, the Cronbach’s alpha was 0.85.
Objective measures of Facebook use included the number of Facebook friends (9-point scale, from 1–0-100 friends, to 9 – above 800 friends), hours spent using Facebook per week and using Facebook apps on a person’s smartphone device.
Data analytic approach
First, descriptive analysis of the characteristics of the FIQ items was conducted (i.e., mean, standard deviation, skewness, kurtosis, intercorrelations). Second, dimensionality of the FIQ via exploratory factor analysis (EFA) and confirmatory factor analysis (CFA) was examined. The total sample was randomly split into two samples; one was used for the EFA and the other for the CFA. CFA was based on the maximum likelihood method with Sattora-Bentler adjustment(Satorra & Bentler, 1994). This adjustment was used because there was a violation of multivariate normal distribution (Doornik–Hansen omnibus test: χ2(df=16) = 12133.40, p < 0.001; Henze–Zirkler’s consistent test: χ2(df=1) = 187000.00; p < 0.001; Mardia’s multivariate kurtosis test: χ2(df=1) = 6232.64, p < 0.001 and Mardia’s multivariate skewness test: χ2(df=120) = 10521.90, p < 0.001). The following statistics were used to determine model fit: χ2, Root Mean Square Error of Approximation (RMSEA), Standardized Root Mean Square Residual (SRMR), comparative fit index (CFI), Tucker-Lewis index (TLI) ( Hu & Bentler, 1999; Kline, 2011). RMSEA lower than 0.08 and SRMR lower than 0.08 indicates a good fit of the model. Also, values of CFI and TLI higher than 0.90 allow a conclusion that a model fits well to a data (Kline, 2011; Hu & Bentler, 1999). Additionally, the unidimensionality of FIQ was further examined using different coefficients such as explained common variance (ECV), mean of item residual absolute loadings (MIREAL) and unidimensional congruence (Ferrando & Lorenzo-Seva, 2018).
Third, an item responses theory (IRT) analysis which is concerned with development of test items and accurate test scoring (Hambleton & Swaminathan, 2013) was used. EFA and CFAs are the most common techniques for evaluating the dimensionality of questionnaires. However, these analyses do not constitute an exhaustive analysis at the item-level. Therefore, we used item responses theory (IRT) analysis which is concerned with development of test items and accurate test scoring. Also, IRT analysis reflects more precisely the relationship between the underlying psychological construct being measured and the measurement process (Hambleton & Swaminathan, 2013).
The eight items were analyzed using the graded response model (GRM) which analyze ordinal responses and rating scales and where each response option level is compared to all response options above that level (Samejima, 1997). Specifically, if the FIQ has a 7-point response scale, from 1 – strongly disagree, to 7 – strongly agree, there is six comparisons between levels. Each comparison is described by threshold which indicates the location on theta (θ) at which individuals would be equally likely to indicate above and below comparison response levels. In this context, the theta (θ) reflects a unidimensional latent trait being assessed by the FIQ and has a mean of 0 and a standard deviation of 1 with an arbitrary range that will cover the latent trait that is being measured by this scale. The first threshold (β1) describe location on theta when individuals choosing to respond 1 versus all other responses. Similarly, the second threshold (β2) describe location on theta when individuals choosing to respond 1 or 2 versus all other responses. Other thresholds (i.e. β3, β4, β5, β6) are described analogously. The item discrimination parameter (α) reflects how well item identify individuals at different levels of the latent trait. This parameter has theoretical range from -∞ to +∞. However, items with negative values of α and lower than 1 may considered problematic and consideration should be given to removing them from the scale (Hambleton & Swaminathan, 2013) (Yang & Kao, 2014). In addition, item response category characteristic curve (CCC) was used in order to analyse the eight items more accurately. The item response category characteristic curve presents the probability of individuals choose a certain response on the scale (1–7) at various levels of the Facebook intrusion latent trait.
Fourth, reliability analysis of the FIQ was conducted using Cronbach's alpha, composite reliability, factor determinacy (the correlation between the factor score estimates and true factor scores) and average variance extracted. For the above analyses, the total sample was used.
Fifth, construct validity was assessed. The relationships between FIQ score and Facebook Intensity Scale score, the Bergen Facebook Addiction Scale score, problematic Internet use and problematic video gaming which have been associated with behavioral addiction was assessed. The rho Spearman correlations coefficient and bootstrap method (N = 5000; 95% CI) was used to examine these relationships. Additionally, partial correlations was used to verify the separation between FIQ and problematic video gaming whilst controlling for problematic Internet use. The criterion validity of the FIQ was also analyzed using Multiple Indicator, Multiple Cause (MIMIC) model (Lee et al., 2013) including items-based latent construct of Facebook intrusion and predictors associated with objective Facebook-related behaviors indicated in previous studies: hours spent using Facebook per week, number of Facebook friends (9-point scale, from 1–0-100 friends, to 9 – above 800 friends) and using Facebook apps in smartphone (Salehan & Negahban, 2013) (Kittinger et al., 2012) (Zheng & Lee, 2012) (A. Błachnio et al., 2014). Also, bootstraping method (5000 sample) with bias-corrected percentile method was used to estimate a standardized regression weights, correlations and R-squared value with 95% confidence interval (Byrne, 2010) (Kline, 2011). Taking into account that people who use psychoactive substances are also more likely to have behavioural addictions, such as social networking addiction (Thege et al., 2016) (Pawłowska et al., 2014) (Kuss & Griffiths, 2011), comparisons between the addiction treatment group and the control group on the FIQ were tested. The t Student test was used to verify difference between groups in terms of the FIQ score, and Cohen’s d was calculated as an effect size (Cohen, 1988). The difference between groups was also verified in terms of hours spent using Facebook per week, number of Facebook friends, hours spent using Internet per week, years of having a Facebook profile, and using Facebook apps on smartphone.
Sixth, FIQ normalization was developed. In this context, the standardized sten score was used (Coaley, 2014). Given sex and age differences (Atroszko et al., 2017) (A. Błachnio et al., 2015) (Agata Błachnio, Przepiorka, Bałakier, et al., 2016), we calculated norms based on subgroups samples. The age subgroups were separated due to the educational system in Poland (primary school: up to 15 years old; high school: 16–19 years old; college 20–24 years old). The FIQ score for each subgroups have values of skewness and kurtosis lower than 1, except for the three subgroups (men: 10–15 years; women: 30–39 years; men: 40 and above years) where these values were higher than 1 but lower than 1.2. Therefore, logarithmic data of FIQ score was performed using natural logarithm for these three groups before standardization. The statistical analyses were conducted using IBM SPSS Version 21 with AMOS 22 (descriptive statistics, criterion validation analysis), Stata 14 (CFA, IRT analysis) and Factor 10 (EFA, unidimensionality analysis).