Adaptation of the Bergen Social Media Addiction Scale (BSMAS) in Spanish

The impact of social networks on people ’ s daily lives is worrisome, particularly in adolescents and young people, who seem to exceed the limits of normal use. Constant excessive use can lead to pathological behaviors linked to social media addiction (SMA). Our objectives were to 1) adapt the Bergen Social Media Addiction Scale (BSMAS) to Spanish and 2) evaluate its psychometric properties in a young population. The BSMAS was adapted to Spanish, involving experts on social media addiction and people from the target population during the adaptation process. For the psychometric evaluation, 650 Peruvian college students responded to the Spanish version (53.5 % women aged 18 to 40, M = 21.5 SD = 2.7). The one-dimensional measurement model proposed for the original BSMAS was confirmed for our version ( X 2(9) = 23.9315, CFI = 0.994, TLI = 0.990, SRMR = 0.032, RMSEA = 0.061). The reliability was good ( α = 0.863; 95 % CI: 0.848 – 0.870; ω = 0.864; 95 % CI: 0.846 – 0.844), and the measurement invariance was confirmed for sex and age by fitting models. The concurrent validity with external social media addiction and mental health indicators was also confirmed. This study provides new and relevant information on the BSMAS validity and allows its application to Spanish-speaker college students from Peru and similar countries.


Background
The use of social media platforms such as Facebook, Instagram, Twitter and YouTube has become a massive activity in contemporary society.Approximately 50 % of the world's population uses this type of platforms (Beveridge, 2022), where people -virtually-connect with other people via portable devices (Décieux et al., 2019).In early 2023, two global social media consultants (Meltwater and We Are Social) reported that 4.76 billion people around the world use social media, showing an increase of 137 million new users within the last year ("Digital 2023("Digital ", 2023)).At the individual level, the time spent on social media has become extreme, with a daily average > 2.5 h that mostly compromises adolescents and young adults (Andreassen, 2015;Andreassen et al., 2017;Kuss & Griffiths, 2011).This excessive exposure to social media platforms can cause addiction ("Digital 2023("Digital ", 2023;;Beveridge, 2022;Décieux et al., 2019;Kuss & Griffiths, 2011), which implies a systematic disruption of normal activities (e.g., work, personal and family life) because of the time dedicated to social media (Andreassen, 2015), bringing other problematic behaviors (Bányai et al., 2017;Paakkari et al., 2021).
A better understanding arises from the biopsychosocial model of Griffiths (2005) that was based on the comparison of people addicted to substances and people who interact with technologies, as shared and reinforcing behaviors.These addiction criteria are uniformly adjusted based on the six core addictive behaviors identified as follows: a) salience, denoting the significance attributed to a specific activity, which interferes across various facets of individuals' lives; b) mood change influenced by attachment to a particular behavior; c) tolerance, reflecting the process of escalating engagement to achieve anticipated effects; d) withdrawal symptoms or the withdrawal syndrome, typically emerging due to a reduction or cessation of specific behaviors; e) conflict, whether interpersonal or intrapersonal, originating from addictive conduct; f) relapse or a tendency to revert to problematic behaviors following a period of abstinence (Griffiths, 2005(Griffiths, , 2017;;Kuss & Griffiths, 2017).
Although the BSMAS is a robust psychometric tool, there are two aspects of its structural validity that merit special revision in any transcultural adaptation.First, three SMA indicators included in the BSMAS have shown very high correlations with each other (i.e., salience, tolerance and relapse, and conflict) (Balcerowska et al., 2022;Huang et al., 2021;Monacis et al., 2017;Yam et al., 2019), an issue called overlapping.This issue can, among other things, lead to wrong interpretations (e.g., based on an incorrect measurement model) and inflate reliability estimates (Balcerowska et al., 2022;Yam et al., 2019).Second, the measurement invariance (Putnick & Bornstein, 2016), which relates to how the BSMAS measurement model and some of its properties are stable across subgroups, has been barely studied.Some studies have confirmed the measurement invariance across groups by gender (Chen, Strong, et al., 2020;Stȃnculescu, 2022;Yue et al., 2022), although others just found partial invariance (Lin et al., 2017;Monacis et al., 2017).We are particularly interested in the invariance according to age groups, as adult users spend less time on social media and, therefore, may be less likely to develop SMA compared to younger users (Chen, Strong, et al., 2020;Lin et al., 2017;Yam et al., 2019).
The BSMAS studies evaluated for their concurrent validity with other instruments related to social media such as facebook addiction (BFAS); social network intensity or forms of engagement with social networks (SNI or SMEQ) and fear of missing out (FOMO) obtaining moderate relationships (Bakioglu et al., 2022;Chen, Strong, et al., 2020;Islam et al., 2022;Stȃnculescu, 2022).These variables have shown importance because they share characteristics between the functionalities and possible maintenance conditions in SM (Bakioglu et al., 2022;Balcerowska et al., 2022).Likewise, other psychological factors have shown small and moderate relationships, especially with psychopathological variables such as anxiety and depression using widely studied screening (i.e, PHQ and GAD) ( Žmavc et al., 2022).These conditions can generate escape/avoidance behaviors in situations perceived as harmful, maintaining the connection pattern by problematic users (Cerniglia et al., 2019).
Our study objectives were to 1) perform a cultural adaptation of the BSMAS to Spanish, 2) determine the internal structure validity and reliability of the Spanish version, 3) analyse the BSMAS measurement invariances across groups by age and sex, 4) determine the concurrent validity of the BSMAS with other variables related to SMA (i.e., facebook addiction, Fear of missing out, social network intensity and social media engagement).In addition, the inclusion of other psychological factors (i.e., depression and anxiety scale).

Participants
The psychometric evaluation was performed with Peruvian adults who lived in Lima, the capital and most populous city of Peru.We included people aged 18 to 40 years who agreed to participate in the online survey.Those who reported a work activity directly related to the use of social media (e.g., community manager) or a psychiatric comorbidity were excluded.We performed a controlled quota sampling (Yang & Banamah, 2014), choosing gender and age as specific categories.The sample size calculation suggests values less than n = 400, for more details see the supplementary material.For this study, a sample of 650 Peruvian youth and adults was used.

Cross-cultural adaptation
Following the back-translation strategy, we performed a linguistic adaptation process (Muñiz et al., 2013).The original English BSMAS (Andreassen et al., 2016) was translated into Spanish by two certificated translators for whom Spanish was their first language.Then, the translated BSMAS was back-translated into English by two other certificated translators, for whom English was their born language.Both translations were reviewed and updated until the translation team concluded that the working Spanish version was semantically equivalent to the original English version.This working version was also reviewed by the research team (i.e, specialists in addictive behaviors) to ensure that all the original BSMAS concepts (i.e., SMA indicators) were maintained in the working version (Gjersing et al., 2010).
Then, the working version was responded to by a group of local university students (focus group, n = 30).The items' comprehension and acceptability were assessed, and respondents were asked to rephrase each problematic item (Gjersing et al., 2010;Boateng et al., 2018).The research team analysed all responses from this group and then integrated them into the working version to generate the final BSMAS in Spanish.This version was used in the following step.
The Fear of Missing Out (FoMO) scale, developed by Przybylski, Murayama, DeHann and Gladwell (Przybylski et al., 2013) assesses the fear of missing news on a social network.It is a one-dimensional scale of ten items that can be responded to according to five Likert-Type options (1 = "not at all true" to 5 = "extremely true").The FoMO's onedimensional model has a good fit (RMSEA = 0.036; CFI = 0.988; TLI = 0.985) as well as good reliability (ω = 0.895) (Correa-Rojas et al., 2020).We expect higher levels of missing out fear to be associated with higher levels of social media addiction (i.e., measured by the BSMAS) (Abel et al., 2016;Alt, 2015).
The Social Network Intensity (SNI) is a brief scale designed to assess the intensity of social networks use (Salehan & Negahban, 2013).The SNI has five items with a 7-point Likert-type response option each (1 = "never" to 7 = "always").The SNI authors propose a one-dimensional model, with the main latent factor explaining 73 % of the items' variance.The model presented good fit (CFI = 0.990; TLI = 0.970; SRMR = 0.030).The reliability of the scale was good (α = 0.86) in the original study.We expect a positive correlation between the BSMAS and SNI, since the latter has shown a similar correlation with other technologyrelated addictive behaviors (e.g., with smartphone addiction that was r > 0.40) (Salehan & Negahban, 2013).
The Social Media Engagement Questionnaire (SEMQ) is a onedimensional scale that evaluates daily participation in social media contexts and this fit was adequate (CFI = 0.98; TLI = 0.96; SRMR = 0.025).It has five items with 8-point Likert-type response options (0 = none to 7 = every day) that can be summed to obtain a reliable measure of social media engagement (α > 0.80) (Przybylski et al., 2013).We expect SEMQ to be positively correlated with BSMAS.
Two demographic variables were also measured, sex and age, to evaluate the BSMAS measurement invariance.

Item analysis
We described the BSMAS items' distributions by reporting their mean, standard deviation, skewness and kurtosis and by verifying the floor and ceiling effect.The corrected test-item correlation was evaluated as an index of item discrimination.The possible ways to exclude an item were if r ≤ 0.20 (insufficient information to represent the construct) or r > 0.85 (possible overlap).

Internal structure validity
We fitted a one-dimensional model by performing a confirmatory factor analysis with the weighted least squares mean and variance estimator (WLSMV), which allows for handling non-normality and ordinal data (Li, 2016).For assessing the model fit, we calculated: the robust x 2 for which p-values >0.05 are a reflection of good fit (Hu & Bentler, 1999), the comparative fit index (CFI) as an incremental measure for which values >0.95 indicate good fit (Hu & Bentler, 1999), the root mean square error of approximation (RMSEA) as a parsimony measure for which values ≤0.06 are optimal (Brown, 2006;Hu & Bentler, 1999), and the standardized root mean residuals ratio (SRMR) for which values ≤0.08 are optimal (Brown, 2006).The average variability extracted (AVE) was assessed, as an expression of the items' variance proportion explained by the latent factor, using the cut-off point >0.50 as the minimal expected (Fornell & Larcker, 1981).The detection of possible between-item overlapping was determined by modification indexes (MI) that allow models of possible error structures in the model, values below lower <30 (Whittaker, 2012).

Measurement invariance
We performed a multiple group confirmatory factor analysis (MG-CFA) analysis for the measurement invariance evaluation across defined groups (i.e.gender and age).Different progressive restrictions were added and different change criteria were used in a wide repertoire of fit indices (CFI; TLI; SRMR and RMSEA) for the comparison of models with restrictions against models with less restrictions (Ding et al., 2023).First, we assumed configural invariance (similar structure between groups), threshold invariance (i.e., invariant structure and thresholds between groups), metric invariance (i.e., similar factorial structure, thresholds and factor loadings between groups) and finally the scalar invariance (i.e., thresholds, factor loadings, factor structure, and intercepts).The difference x 2 was not taken into account due to its sensitivity to sample size.For analyses, we prefer to examine the ΔCFI; ΔTLI; ΔSRMR and ΔRMSEA.For the initial levels of invariance, more liberal criteria can be taken in the cut-off points (e.g., CFI of ≤0.02 or RMSEA of ≤0.03) (Ding et al., 2023;Rutkowski & Svetina, 2017).However, in the case of the evaluation of scalar invariance, a more rigid criterion ΔCFI <0.01 or ΔRMSEA <0.015 was maintained (Cheung & Rensvold, 2002;Putnick & Bornstein, 2016).

Ethics
The Ethics Committee of the Instituto Peruano de Orientación Psicológica -IPOPS (IPOPS-024-2020) approved the study protocol.The BSMAS was administered only under conditions of voluntariness, anonymity and non-remuneration in people who were previously informed about the aims and purposes of the study.Only participants who fully accepted their participation on a voluntary and non-remunerated basis were counted.

Cross-cultural adaptation
After the full translation process (Fig. 1), the expert panel improved some language details in the Spanish version.For example, the preassessment question for the SMA items was customized and salience component in Spanish was changed "mucho tiempo" to "largo tiempo".Each expert scored each item regarding its content validity, showing an average Aiken's V equal to 0.96.The focus group identified some problems related to temporality and social desirability.For example, the average response time of those evaluated was timed with a maximum of 15 min.To mitigate the possible effect of social desirability regarding SM ("facebook, instagram or twitter") the terms directed to these social media site was removed.Considering these improvements, the Spanish version was updated until its optimal shape.

Characteristics of participants
We analysed Spanish BSMAS responses from 650 college students, mostly women (53.5 %) and aged, on average, 21.1 years old (SD = 2.7).They were mostly in their third year of studies (48.5 %), considering that typical Peruvian college studies take 5 years.Anxiety (31.2 %) and depressive symptoms (40 %) were highly prevalent in this population, while indicators of high social media use were present in >18 % of this group (Table 1).

Item analysis
Table 2 shows the mean (M), standard deviations (SD) and correctedtotal correlation (r itc ) for each BSMAS item.The lowest mean was on item 4, while the highest mean was on item 2.6.The skewness and kurtosis values were within acceptable limits.The discrimination index on the BFAS scale was also acceptable (r itc = 0.581-0.706).

Measurement invariance
For the models by age and sex, ΔCFI; ΔTLI and ΔSRMR were <. 02, while ΔRMSEA were < 0.03 (Table 4).This means that BSMAS measurement is invariant across groups by sex and age.

Concurrent validity
The BSMAS measures showed a positive and large (ρw > 0.70) or moderate (ρw > 0.50) correlation with Facebook addiction (BFAS) and fear of missing out (FOMO), respectively (Table 5).For social network intensity, social network engagement, anxiety and depression, the correlation against BSMAS was also positive but with a small effect (ρw > 0.70).In general, these findings show a good concurrent validity.

Discussion
The BSMAS was adapted to Spanish, involving social media addiction experts and people from the target population during the adaptation process.The Spanish version showed psychometric properties similar to the original BSMAS'; for example, we confirmed the one-dimensional measurement model in the adapted version.We also verified that the Spanish BSMAS provides a reliable and invariant measure across groups by sex and age, and that its general score is positively correlated with other external indicators of social media addiction.
The internal structure of the BSMAS (i.e., one-dimensional model) that we verified has been largely confirmed in populations from different countries in Europe (Andreassen et al., 2016;Bányai et al., 2017;Dadiotis et al., 2020;Monacis et al., 2017;Stȃnculescu, 2022), Asia (Chen, Strong, et al., 2020;Leung et al., 2020;Lin et al., 2017;Shin, 2022;Yam et al., 2019) and America (Watson et al., 2020).Unlike a few studies from Italy, Hong Kong and Poland that showed some overlapping between salience, tolerance, relapse, and conflict items (Balcerowska et al., 2022;Huang et al., 2021;Monacis et al., 2017;Yam et al., 2019), our findings did not show any overlap between these or other items.This is a good sign in terms of cultural equivalence to the original English version (Andreassen et al., 2016) because conceptual differences between items/indicators remain clear for Spanish readers.
The BSMAS also showed optimal reliability values, which is consistent with previous findings in non-clinical populations (Andreassen et al., 2016;Bányai et al., 2017;Chen, Ahorsu, et al., 2020;Dadiotis et al., 2020;Monacis et al., 2017;Stȃnculescu, 2022).Despite these populations being culturally different in types of users (e.g., users of one or more social media platforms), average connection time and Internet accessibility, reliability is always reported as optimal (i.e., between 0.82 and 0.91).Even where different reliability coefficients are reported (e.g., alpha or omega), the internal consistency shows similarity among studies.Only one study on schoolchildren from Turkey reported relatively low values of reliability (Demirci, 2019).Preadolescents may have greater supervision when using social media, experiencing SMA in a differentiated way (Bloemen & De Coninck, 2020) or having different risks of suffering it due to their parents' rules.The heterogeneity among preadolescents in terms of SMA symptoms, self-perception of social media use and actual SMA risk might affect the BSMAS internal consistency.For example, when the variance associated with each item is higher than the variance associated with the total score (i.e., in Cronbach's alpha calculation).However, this is certainly not the case in our and most studies on BSMAS worldwide.This implies that the accuracy of the reliability of the Spanish version is as accurate as in other widely used versions.
Our findings coincide with the measurement invariance of the BSMAS across age and sex groups reported elsewhere (Chen, Ahorsu, et al., 2020;Chen, Strong, et al., 2020).Measurement invariance is essential before performing any formal comparisons between these groups with the BSMAS score (Chen, Strong, et al., 2020;Leung et al., 2020;Lin et al., 2017;Stȃnculescu, 2022).We started this study assuming some potential deviations from measurement invariance in our target population.For example, with Peruvian men more focused on sports and other competitive activities and Peruvian women more focused on recreational/social activities (Vallejos-Flores et al., 2018), they could experience social media addictive use in a different way (Yue et al., 2022).These sex differences impact on addictions to psychoactive substances (PAS); thus, it is not unexpected to find similar differences in other addictive behaviors such as those reflected in BSMAS measures (Stein et al., 2021).Regarding differences by age reported previously, children and adolescents spend the least average time on social media (Chen, Strong, et al., 2020;Lin et al., 2017;Yam et al., 2019), reducing the risk of developing social media addiction compared to college students (Chen, Ahorsu, et al., 2020;Lin et al., 2017;Yam et al., 2019).This can be explained by the control that parents have over the way their kids use social media devices before they gain more independence.However,   (Abel et al., 2016;Alt, 2015;Bloemen & De Coninck, 2020).To our knowledge there are few studies related to social network intensity (SNI) and SMA (Cataldo et al., 2022;Stȃnculescu & Griffiths, 2022).Despite this, the SNI is a strong predictor of negative purchasing behaviors that could generate maintenance in SMA (Pellegrino et al., 2022).Studies show that SMA is moderately related to some psychopathological variables such as anxiety or depression (Hussain et al., 2020;Malak et al., 2022), which is consistent with our findings.Our study presents major implications for depressive symptoms associated with problematic use of social media.This is because SM in adolescents and young people are used as an escape/avoidance behavior from situations perceived as harmful by certain users (Cerniglia et al., 2019).In sum, the Spanish BSMAS offers a valid measure of SMA that is consistently related to external indicators of SMA and mental health.
We identified some strengths and limitations of our study.This is the first time where verifiable adaptation process has been undertaken for the Spanish version of the BSMAS, a procedure that has not been previously carried out.We included a number of external measures for exploring concurrent validity that were not simultaneously considered in other studies, making our evidence stronger.However, we only had the opportunity to evaluate college students, which limits the generalization of our findings to similar populations.The measurement invariance was only evaluated for sex and age, which could not cover other important and more diverse groups such as cultural variables (in other Spanish-speaking contexts), differentiated age groups (adolescents or older adults) or other differentiated addictive behaviors.Despite these limitations, this study provides new and relevant information on the BSMAS validity and allows its application in some critical Spanishspeaker contexts.

Conclusion
We successfully adapted the BSMAS to Spanish, confirming the onedimensional measurement model of the original version, good reliability, measurement invariant across groups by sex and age, and concurrent validity with external indicators of social media addiction and mental health.

Table 2
Descriptive statistics of the BSMAS items.

Table 3
Standardized factor loadings and composite reliability of BSMAS in Peruvian sample.

Table 4
Measurement invariance for the variables sex and age. of our concerns about BSMAS measurement invariance related to sex and age were confirmed in the target population.This study evidences the bases for making valid comparisons at the level of sex and age, which makes it important to deepen these comparisons in this cultural group.The BSMAS score shows good concurrent validity with other addictive behavior measures, as could be expected considering previous research on addiction topics.Individuals with SMA can present other addictive behaviors related to technologies (e.g., FOMO is a trigger for an indeterminate connection loop), generating irritability, anxiety, and maladaptive feelings and enhancing abuse/addiction by increasing participation in social media Note. df = degrees of freedom; CFI = comparative fit index; TLI = Tucker Lewis index; RMSEA = root mean square error of approximation; SRMR = standardized root mean residual; ΔCFI = delta of CFI; ΔTLI = delta TLI; ΔRMSEA = delta RMSEA; ΔSRMR = delta SRMR.(n = 650).A. Copez-Lonzoy et al.none

Table 5
Mean scores, standard deviations, robust correlation coefficients.