Sample

The original sample comprised 251 respondents, and 19 of them were outliers that were removed. The final sample comprised 232 respondents, most of whom were female. The average age was 30.03 years. Regarding the marital status, the majority were single or married. Concerning qualifications, most had completed secondary education or had graduated. In the professional context, most were active. Table 1 presents the sample sociodemographic characterization.

Table 1 Characterization of the sample (n = 232) 

Note: SD = Standard deviation.

Measures

The BFAS was designed by ^{Andreassen et al. (2012}). It originally included 18 items, three for each of the six core criteria of addiction (i.e., salience, mood modification, tolerance, withdrawal, conflict, and relapse) experienced during the past year. Each item is scored on a five-point Likert scale using anchors of 1: Very rarely to 5: Very often. Higher scores indicate greater FB addiction. The authors found a unifactorial model with 6 items and considered that the factor structure of the scale was good and Cronbach’s alpha coefficient was 0.83. The 3-week test-retest reliability coefficient was 0.82. The scores converged with scores for other scales of FB activity. They were also positively related to neuroticism and extraversion, and negatively related to conscientiousness (Andreassen et al., 2012). However, the six items resulting from the adaptation made by Andreassen and colleagues (2012) seemed insufficient to assess the complexity of the phenomenon of adding to FB (as representative of a social network), because each dimension includes only one item, and so it was decided to use the original version of 18 items.

The Brief Symptom Inventory (BSI) (^{Derogatis, 1975}, ¹⁹⁹³), adapted by ^{Canavarro (Canavarro, 1999}), is a 53-item self-report inventory in which participants rate the extent to which they have been bothered (0 = "not at all" to 4 = "extremely") in the past week by various symptoms. The BSI has nine subscales designed to assess individual symptom groups: somatization, obsessive-compulsive, interpersonal sensitivity, depression, anxiety, hostility, phobic anxiety, paranoid ideation, and psychoticism. BSI also includes three indices of distress (Global Severity Index, GSI, Positive Symptom Distress Index, PSDI, and Positive Symptom Total, PST). The BSI Cronbach’s alpha coefficient in the current sample is 0.97.

The UCLA Loneliness Scale (^{Russell et al., 1980}), adapted by ^{Neto (1989}), is an 18-item scale designed to measure one’s subjective feelings of loneliness as well as feelings of social isolation. Participants rate each item as O (“I often feel this way”), S (“I sometimes feel this way”), R (“I rarely feel this way”) or N (“I never feel this way”). The UCLA Cronbach’s alpha coefficient in the current sample is 0.40.

The Beck Depression Inventory (BDI) is a 21-item, self-report rating inventory that measures characteristic attitudes and symptoms of depression (^{Beck et al., 1961}), adapted by ^{Vaz-Serra and Pio-Abreu (1973}). The BDI Cronbach’s alpha coefficient in the current sample is 0.89.

The Zung Self-Rating Anxiety Scale (SAS) is a method of measuring levels of anxiety in patients who have anxiety-related symptoms (^{Zung, 1971}), adapted by ^{Ponciano et al. (1982}). The scale focuses on the most common general anxiety disorders: coping with stress typically causes anxiety. The SAS test is self-administered, with each response using a 4-point scale, from ‘none of the time” to “most of the time.” There are 20 questions with 15 increasing anxiety level questions and 5 decreasing anxiety questions. The SAS Cronbach’s alpha coefficient in the current sample is 0.80.

The Freiburg Personality Inventory Revised (FPI-R) (^{Fahrenberg et al., 1984}), adapted by ^{Soares et al. (2005}) is a structured, verbal, omnibus measure of personality traits, which was constructed to measure normal personality characteristics. The questionnaire consists of 138 dichotomous items designed to measure 12 personality traits: Life satisfaction; Social orientation; Achievement orientation; Inhibition, Irritability; Aggressiveness; Strain; Somatic complaints; Health concerns; Frankness; Extraversion; and Emotionality. FPI-R objectively describes the manifestation of important dimensions of personality, allowing inter-subject comparison. The FPI-R Cronbach’s alpha coefficient in the current sample is 0.84.

Procedures

A study for the original BFAS adaptation among Portuguese FB users was performed over 4 months in 2018. It is a descriptive and exploratory study. A total of 251 respondents answered a sociodemographic questionnaire and the original version of BFAS (^{Andreassen et al., 2012}) that had been previously translated and back-translated by a group of four psychologists, two of them being native English speakers. Based on an examination of the back-translations, it was concluded that there were no problematic items. The scale was well received and participants did not report any problems in answering the items. The questionnaire and the BFAS were available online and the participants received information through the FB, where a page about the study was created. When accessing the link, the opening page contained the informed consent, which had to be accepted by the participants to continue the survey. Since the respondents could only answer the subsequent question, after having responded to the previous one, the respondents who completed the questionnaire answered all the questions and, therefore, no missing cases exist. The criteria for participation in the study were: to have Portuguese nationality, aged equal to or greater than 18 years and to be a daily FB user.

Statistical analyses

According to ^{Marôco (2010}), the sample size is within the required parameters (n = 200 - 400) concerning the use of the Maximum Likelihood (ML) method in SEM. There is no missing data. The confirmatory factor analysis of the uni and multifactorial BFAS structure was performed with IBM SPSS Amos 22 software (SPSS, version 22.0; IBM Corp, 2013). To assess the assumptions required for the structural equations analysis, values of |sk| < 3 and |ku| < 10 were considered (^{Kline, 2010}) in order to verify the items’ normal distribution, where sk is the coefficient of Skewness and ku is the coefficient of Kurtosis. According to Kline (2010), values of |sk| < 3 and |ku| < 10 are acceptable to SEM. To verify the existence of outliers, the Mahalanobis squared distance (p1 and p2 < 0.001) was used (Kline, 2010), where p1 is the probability of any observation exceeding the squared Mahalanobis distance of that observation and p2 is the probability that the largest squared distance of any observation would exceed the Mahalanobis computed distance. To test the items’ multicollinearity, Spearman coefficients were calculated, according to a reference value of 0.80 (^{Field, 2009}). To evaluate the goodness-of-fit of the model to the global correlation frame, CFI and GFI values greater than 0.9 and PCFI and PGFI values higher than 0.6 were considered indicative of a good fit, where GFI is the Goodness-of-Fit Index, PCFI is the Parsimony-adjusted Comparative Fit Index and PGFI is the Parsimony-adjusted Goodness-of-Fit Index, all considered Parsimonious Fit Indices representing adjustments. Values of (2/df = ~2 and RMSEA < 0.08 (with PRMSEA < 0.05, significant) were considered, indicating a good model fit, where (2 is the Chi-squared index, df is the degree of freedom and PRMSEA is the probability of RMSEA. Residuals were also computed, with a SRMR ≤ 0.05, where SRMR is the Standardized Root Mean Square Residual (^{Schermelleh-Engel et al., 2003}). The refinement of the BFAS original model was performed from the values of the Modification Indices (MI), widely used in structural equation modeling, for the Lagrange multipliers (LM), considering that trajectories and/or correlations with LM > 11 (p < 0.001) indicate significant variation in the quality of the model (Marôco, 2010). To assess the parsimony of each model, AIC, BCC and MECVI Indices were considered, where AIC is the Akaike information Criterion, BCC is the Browne-Cudeck criterion and MECVI is the modified ECVI, that is, the Expected Cross-Validation Index. Lower values of the indices indicate a more parsimonious model (Marôco, 2010).

Finally, the reliability of the construct was assessed with the composite reliability and the validity of the construct was assessed with Average Variance Extracted (AVE) (^{Fornell & Larcker, 1981}). The overall reliability of the instrument was assessed by calculating the Cronbach’s alpha coefficient statistics and the construct validity was also confirmed through the convergent validity with the BSI, the UCLA Loneliness Scale, the BDI, the SAS and the FPI-R. All these instruments were previously assessed for the Portuguese population, with confirmed good psychometric qualities, and thus, no translation process was necessary. Spearman’s Correlation coefficient was used to estimate the convergent validity.

BFAS Confirmatory factor analysis

When using the Mahalanobis squared distance (p1 and p2 < 0.001), 19 outliers were found (^{Kline, 2010}), as well as 5 items that remained outside the standard reference (that is |sk| > 3 and |ku| >10) (Kline, 2010). Those 5 items were: item 18 (|sk| = 3.07 and |ku| =9.47), item 17 (|sk| = 3.02 and |ku| = 9.63), item 11 (|sk| = 3.19 and |ku| = 10.10), item 10 (|sk| = 3.43 and |ku| = 11.16) and item 2 (|sk| = 3.61 and |ku| = 14.52). Since the model’s goodness-of-fit would be better without these items (^{Marôco, 2010}), they were excluded from the analysis and the remaining ones meeting sk and ku values within the required standard reference (|sk| < 2.82 and |ku| < 9.47) were kept. The absence of items’ multicollinearity was assessed (0.135 < rsp< 0.808), where rsp is the Spearman correlation coefficient, with the exception of items 13 and 14, correlated with a value of 0.808. Taking into account that the correlation value is very close to the reference value 0.80 (^{Field, 2009}), it was decided to not prematurely exclude items 13 and 14 from the original model (Figure 1). The original BFAS model was tested, following the original unifactorial structure. The model (Figure 1: Model 0), with 18 items, reveals an overall unacceptable fit (Table 2).

Figure 1 Model 0: Original BFAS first-order unifactorial model with 18 items. 

Table 2 Confirmatory Factor Analysis: global adjustment indices (n = 232) 

Note: χ2 = Chi-Square Distribution; df = degrees of freedom; CFI = comparative fit index; PCFI = parsimony-adjusted comparative fit index; GFI = goodness-of-fit index; PGFI = parsimony-adjusted goodness-of-fit index; AGFI = adjusted goodness-of-fit index; RMR = root mean square residual; RMSEA = root mean square error of approximation; PRMSEA = probability of RMSEA. *p < 0.001, **p < 0.05.

The second model tested (Figure 2: Model 1) from which 5 items not in the range of normal distribution were eliminated, also shows an overall mediocre fit, according to data presented in Table 2.

Figure 2 Model 1: BFAS first-order unifactorial model with 11 items and 5 covariances. 

However, and after consulting the MI of the Model 1, there are several examples of covariance between the errors of measurement of several items. From a theoretical standpoint, the ones making sense were set out: items 1 and 3 are examples of that, belonging to the same original theoretical element (salience), and the same happens with the remaining set out covariance, as shown in Model 1 (Figure 2). In the same Model 1, the MI suggests that the measurement error associated with item 8 is correlated with the measurement error of several items belonging to other theoretical elements: item 9 (MI = 142.8), item 16 (MI = 21.8), item 14 (MI = 15.5) and item 13 (MI = 14.2). The same applies to the measurement error associated with item 4, which is correlated with item 13 (MI = 12.6) and item 16 (MI = 13.6). In the absence of a theoretical support, it was decided to exclude item 8 and item 4, considered problematic, and thus not to establish the covariance. When redefining the original BFAS model, and although it is necessary to include 5 examples of covariance (these ones with a theoretical support, because they belong to the same originally considered elements), the model presents a globally acceptable fit (Figure 2). As noted, however, measurement errors are not mostly independent but correlated, thus indicating a common source of unexplained variance of the items, which cannot be explained through the only common factor that is present in the model. According to this, it was decided to test a multifactor model, with 6 of the original theoretical elements, now present in Model 2 (Figure 3) as factors (salience, tolerance, mood modification, relapse, withdrawal and conflict).

Figure 3 Model 2: BFAS first-order multifactorial model with 18 items and 6 factors. 

Aiming to meet the assumptions of normality, the 5 previously referred items - 18, 17, 11, 10 and 2 - were removed from that model, since they presented a non-normal distribution (Figure 4).

Figure 4 Model 3: BFAS first-order multifactorial model with 10 items and 4 correlated factors. 

With these exclusions, the relapse factor (item 10 and item 11, outside the standard reference) and the conflict factor (item 17 and item 18, outside the standard reference) were removed from the model. This first-order multifactorial model (Figure 4: Model 3) reveals a good overall fit (^{Schermelleh-Engel et al., 2003}), as well as factor loadings above 0.74. On the other hand, the correlational values among the various factors are high and significant (ranging from 0.51 to 0.72) and are still in the theoretically expected directions, indicating that these 4 factors seem to measure a common construct of higher order. Accordingly, analyses were run to test a second-order model that sought to measure the overall construct FB Addiction. This final modified BFAS second-order model (with 10 items and 4 factors) (Figure 5: Model 4) showed a good overall fit (Table 2).

Figure 5 Model 4: Modified BFAS second-order multifactorial model with 10 items and 4 correlated factors. 

In this second-order model, all factors are positively correlated with the second-order factor - FB Addiction: salience (r = 0.76, p < 0.001), tolerance (r = 0.91, p < 0.001), mood modification (r = 0.74, p < 0.001) and withdrawal (r = 0.75, p < 0.001).

Regarding the parsimony of the final modified BFAS model (Figure 5: Model 4), it can be observed that indices AICmodifiedBFAS = 107.26, BCCmodifiedBFAS = 109.66 and MECVImodifiedBFAS = 0.48 present values which indicate a more parsimonious model than the original one (AICoriginalBFAS = 131.58; BCCoriginalBFAS = 134.54 and MECVIoriginalBFAS = 0.58). Regarding the construct validity, the AVE was 0.57, 0.61, 0.78 and 0.73, for salience, tolerance, mood modification and withdrawal, respectively. The values of AVE confirm the strong convergent validity of the factors under study, according to ^{Fornell and Larcker (1981}). In terms of the construct’s reliability, it is higher than 0.7 for all factors, being of 0.73, 0.82, 0.88 and 0.76, to salience, tolerance, mood modification and withdrawal, respectively. Finally, according to the recommended value to the Cronbach’s alpha coefficient (( > 0.70; (^{Hair et al., 2009})), the analysis of the obtained values show that the final modified BFAS model (10 items) (Figure 5: Model 4) has an overall value of Cronbach’s alpha of 0.89.

Convergent validity

Additionally, to confirm the construct validity of the final modified model (Figure 5: Model 4) of BFAS, the convergent validity was also analyzed with the BSI, the UCLA Loneliness Scale, the BDI, the SAS and with the FPI-R. The Spearman correlation coefficients in each one of the cases referred to above are presented in Table 3.

Table 3 Spearman correlation coefficients between scores on BFAS and BSI, UCLA Loneliness Scale, BDI, SAS and FPI-R dimensions (n = 232) 

Note: *p < 0.05, **p < 0.01

The results show that there are significant positive correlations for p < 0.01 between the total score of BFAS and all dimensions of BSI, with 0.221 < rsp < 0.357. Positive correlations are also observed with the indices GSI (rsp = 0.350, p < 0.01) and PSI (rsp = 0.310, p < 0.01), where GSI is the Global Severity Index and PSI is the Positive Symptom Index. There are also positive and significant correlations between the total score of BFAS and the total score of BSI for p < 0.01, with rsp = 0.308.

The results also show that there is a positive and significant correlation between the total score of BFAS and a dimension - social isolation - of the UCLA Loneliness Scale (rsp = 0.230, p < 0.01). A positive and significant correlation also exists between the total score of BFAS and the total score of UCLA Loneliness Scale (rsp = 0.157, p < 0.05).

Positive and significant correlations between the total score of BFAS and all dimensions of SAS, except for the anxiety of the central nervous system, for p < 0.01, with 0.200 < rsp < 0.337, can also be found. A positive and significant correlation also exists between the total score of BFAS and the anxiety of the central nervous system SAS subscale (rsp = 0.131, p < 0.05).

Finally, positive and significant correlations are found between the total score of BFAS and the total score of FPI-R (rsp = 0.216, p < 0.01): and also several FPI-R dimensions: the dimension strain (rsp = 0.256, p < 0.01), the dimension aggressiveness (rsp = 0.207, p < 0.01), the dimension excitability (rsp = 0.222, p < 0.01), the dimension somatic complaints (rsp = 0.173, p < 0.01), the dimension frankness (rsp = 0.278, p < 0.01) and the dimension emotionality (rsp = 0.311, p < 0.01). A negative and significant correlation exists between the total score of BFAS and the dimension life satisfaction (rsp = -0.324, p < 0.01).