Differentiation of self (DoS) is a pivotal construct of the Bowen Family Systems Theory (BFST) that elucidates the interplay between DoS and the growth and preservation of intimate partnerships (Kerr & Bowen, 1988). Bowen (1978) defined DoS as the degree to which individuals are able to balance emotional and rational functioning as well as intimacy and independence in relationships. Skowron and Friedlander (1998) further refined this definition, conceptualizing DoS as the ability to strike a balance between achieving an independent sense of self and maintaining close ties with significant others, particularly those within the nuclear family. The development of DoS is a continuous process that begins in infancy and progresses throughout childhood and adolescence until it reaches a fundamental level in early adulthood (e.g., Bowen, 1978; Lampis et al., 2017). However, it is possible for individuals to increase their level of DoS to a certain extent over their lifespan (Bowen, 1978; Knauth & Skowron, 2004). During childhood and adolescence, DoS takes shape and is influenced by the level of DoS exhibited by parents (Peleg et al., 2023). Individuals with high levels of DoS are better equipped to adhere to their feelings, thoughts, and needs, and effectively manage stressful situations and social pressures. Conversely, those with low levels of DoS may experience greater difficulty in regulating their emotions, behaving authentically, and coping with stress and crises (Bowen, 1978; Peleg et al., 2023). Furthermore, research has demonstrated a significant association between low levels of DoS and various negative outcomes among adolescents, including high levels of anxiety (Peleg et al., 2015), depression (Choi & Murdock, 2017), eating disorders (Peleg et al., 2022), problematic coping styles, and physical health problems (Hooper & Doehler, 2011).

The Differentiation of Self Inventory (DSI) is a self-report instrument developed by Skowron and Friedlander (1998) to operationalize Bowen's concept of DoS. The original version of the DSI consisted of 43 items and encompassed four subscales that captured the intrapersonal dimension and interpersonal dimension of the differentiation of self construct: Emotional Reactivity (ER), which assesses the inclination to react to tension with irrational, passionate flooding; taking an I-position (IP), which measures one's ability to assert oneself and fulfill personal desires (intrapersonal dimension); Emotional Cutoff (EC), which explores the tendency to detach and sever relationships as a coping mechanism for symbiotic dynamics; and Fusion with Others (FO), which evaluates the inclination to form dependent and symbiotic relationships with significant others (interpersonal dimension). Various studies provided evidence in support of its construct validity (Skowron, 2000; Skowron & Friedlander, 1998). However, concerns have been raised about the internal consistency and construct validity of the FO subscale, and the questionnaire was revised by Skowron and Schmitt (2003) to enhance its FO subscale, resulting in a 46-item revised version (DSI-R).

With these revisions and since 2003, the international use of the DSI-R has proliferated. For example, the DSI-R and its adaptations are now the most widely used measures to assess DoS (Calatrava et al., 2022). The DSI-R has been validated in several other languages beyond English, including Portuguese (Major et al., 2014), Chinese (Lam & Chan-So, 2015), Turkish (Işık & Bulduk, 2015), Italian (Lampis et al., 2017), and Filipino (Antazo, 2020). We selected the DSI-R in this study because of its widespread use in research on BFST, particularly in cross-cultural research, as is the focus of the current study.

Specifically, in the current study we examine the use of an adapted version of the DSI-R for Spanish-speaking adolescents to better understand the cross-cultural application of the measure for research on BFST and DoS. In 2015, Rodríguez-González and colleagues developed a Spanish language adaptation of the DSI-R specifically tailored for the Spanish adult population. The results revealed the presence of two distinct subscales, characterized as ER and EC, both comprising 13 items each. These subscales exhibited good internal consistency. It is noteworthy to highlight that two other dimensions of DoS, namely the ability to take an "I" position in relationships and the extent of fusion with others, did not emerge as distinct dimensions in the Spanish sample. These particular dimensions exhibited less cross-cultural stability, suggesting that their manifestation and conceptualization may vary across different cultural contexts. As noted by Rodríguez-González et al. (2015), this finding aligns with previous research conducted on diverse populations, including Russian, Chinese, Portuguese, and Filipino samples, samples for which the IP and FO subscales reported less cross-cultural stability compared to the ER and EC subscales. This underscores the significance of considering cultural factors when examining and interpreting DoS. Moreover, the confirmatory factor analyses conducted in the initial S-DSI validation study further supported and verified that the ER and EC subscales are indeed empirically distinct dimensions of DoS in the Spanish sample. Further research using this adaptation provided additional evidence for the validity and reliability of EC and ER subscales in measuring DoS within a Spanish sample (e.g., Rodríguez-González et al., 2016). These findings align with previous research by Jankowski and Hooper (2012), which also emphasized that the EC and ER subscales can provide a more comprehensive assessment of DoS and guide treatment modalities. Overall, there has been a growing trend in utilizing a two-factor scale to assess DoS throughout the years (Calatrava et al., 2022). Thus, in the current study, we also use a two-factor scale to assess DoS.

In 2004, Knauth and Skowron adapted the DSI for implementation with adolescents, specifically targeting a sample of Americans aged 14 to 19 years. The scale exhibited strong reliability and satisfactory internal consistency. Following this adaptation of the DSI for adolescents as well as the Spanish validation of the DSI-R (Rodríguez-González et al., 2015), Duch-Ceballos et al. (2022) validated the Spanish DSI for dolescents (S-DSI-A). The S-DSI-A consisted of 20 items grouped into two subscales: ER and EC. The original four-factor structure of the DSI was not confirmed in the exploratory factor analysis, leading to a bidimensional structure similar to the validation of the DSI for Spanish adults.

The pursuit of comprehending human psychology and behavior beyond traditionally examined cultures, primarily Western populations, has garnered heightened attention (D'amico et al., 2014; Henrich et al., 2010), mainly through cross-cultural research (Jeong & Lee, 2019; Segall et al., 1998). Nonetheless, the initial crucial step towards conducting cross-cultural research is to establish the adequacy of a questionnaire in producing similar results (i.e., measurement invariance) across distinct countries, languages, or cohorts. The concept of measurement invariance refers to the ability of an assessment to accurately measure a construct across different populations, which is a critical component of construct validity (Putnick & Bornstein, 2016). The assessment of measurement invariance involves determining whether the same construct is being measured in a consistent manner across diverse groups (Van Doren et al., 2021). The establishment of measurement invariance is crucial because it enables researchers to confidently attribute any observed cultural differences in the construct to genuine group differences, rather than to differences in how individuals from different cultures respond to particular items in the assessment (Dimitrov, 2017). Therefore, the validity of intergroup comparisons of findings can only be deemed legitimate when measurement invariance is achieved. Failure to ensure measurement equivalence could result in prejudiced conclusions concerning potential cross-cultural distinctions (Spector et al., 2015). Accordingly, this study aimed to establish the measurement equivalence of the S-DSI-A among adolescents from Spanish-speaking countries, including Argentina, Chile, Ecuador, Mexico, Peru, and Spain, by using measurement invariance techniques. This study represents the first examination of invariance for the S-DSI-A, concurrently serving as the pioneering examination of invariance for the adolescent version of the DSI (Knauth & Skowron, 2004) in any language.

Method

Participants

A total of 5,786 participants were recruited within the Your Life project in high schools in Argentina, Chile, Ecuador, Mexico, Peru, and Spain. Those participants presenting missing values in age or sex, or with ages out of the specified range (12–19) were removed (n = 234). A final sample of 5,552 participants was analyzed. The study used a convenience sample method, wherein schools were extended invitation to partake, and only a subset chose to do so. Additionally, within each participating school, students' involvement was voluntary. The sample was large, exceeding the requirements for the analyses detailed in this manuscript, as it is part of a more-extensive, multi-faceted project. For certain analyses within this overarching project, a large sample is imperative.

Measures

The differentiation of self inventory

Differentiation of self was assessed through the Spanish Differentiation of Self Inventory for Adolescents (S-DSI-A), a 20-item self-report inventory (Duch-Ceballos et al., 2022). Participants responded to the items on a 6-point Likert-type scale, from (1) not true at all, to (6) completely true. The scale contains two subscales: emotional cutoff (EC) and emotional reactivity (ER). Some examples of items are: "People often tell me that I am too emotional" (ER), "When someone close to me disappoints me, I withdraw from him/her for a while" (EC). To generate the final scores, items were reversed and the sum of the scores for each dimension was calculated. Higher scores indicate higher levels of differentiation of self in each dimension. The scale has already been validated in a sample of Spanish adolescents obtaining a good model fit: χ2 = 403.059, df = 141, χ2/df = 2.859, RMSEA = 0.036 (90%CI = 0.032-0.040), SRMR = 0.031, CFI = 0.963, and TLI = 0.950.

Sociodemographic variables

The questionnaire also included questions to collect sociodemographic information such as sex (male/female), age (years), country, religious affiliation, and type of school (public/private/mixed).

Procedure

The research team invited high schools to participate in the study from May 2019 to September 2022. All data were gathered through an online, self-administered, anonymous questionnaire during class time. Schools were provided with all the documentation required to obtain parents’ permission. Each school handled parents’ permission according to their own policies. Teachers and students were informed about the anonymity and confidentiality of the study. Students could leave the survey at any time. They were also told that clicking the “Start” button would imply their consent to participate. Ethical approval was obtained from the Ethics Committee of the Universidad de Navarra.

Data analyses

In pursuit of our study´s central aim -to establish the measurement equivalence of the S-DSI-A- we employed multigroup confirmatory factor analysis (MGCFA) to assess measurement invariance among adolescents across six distinct samples, drawn from Spanish-speaking countries including Argentina, Chile, Ecuador, Mexico, Peru, and Spain. MGCFA procedures followed in this study were guided by Byrne (2008). All analyses were conducted in R (R Core Team, 2022) using the Lavaan package (Rosseel, 2012).

Using MGCFA, we tested the equivalence of models across the six samples by imposing a series of between-group constraints to test configural, metric, and scalar invariance, in that order. Configural invariance establishes that participants across the six groups attribute the same meanings to the latent constructs of ER and EC, as demonstrated by each sample having the same number of factors and similar pattern loadings. Metric invariance establishes that factor loadings are similar across groups. We tested metric invariance by constraining factor loadings across the six groups. In scalar invariance, tests establish whether participants in the six groups have similar item intercepts of ER and EC. Scalar invariance is established by constraining item intercepts across the six groups.

Model fit was assessed using Hu and Bentler’s (1999) criteria for good model fit, including the comparative fit index (CFI), root mean square error of approximation (RMSEA) and its 90% confidence interval, and chi-square (X2). CFI ≥ 0.95 and RMSEA < 0.05 are evidence for a good fitting model (Hu & Bentler, 1999) and CFI = 0.92-0.94 and RMSEA = 0.05-0.08 are evidence for an adequate fitting model (Byrne, 2008). As subsequent models that were examined in comparison to the baseline model can be seen as being nested under one another, we followed recommendations from Cheung and Rensvold (1999) to reject measurement invariance if model fit did not degrade based on cutoff values of ΔCFI ≥ -0.01 and when ΔRMSEA ≤ 0.015. Missing data were assessed and handled through Full information maximum likelihood (FIML) estimation methods in the Lavaan package.

Results

Preliminary analyses

Sample demographics for the six Spanish-speaking samples are presented in Table 1. There were statistical differences on age, gender, religion and education center among the six samples. Means, standard deviations, and correlations for the ER, EC, and DoS Total scores were examined separately for each sample (Table 2).

Table 1 Participants’ sociodemographic characteristics by country
Full size table
Table 2 Intercorrelations, means, and standard deviations (SD) of DSI-R Scores presented by sample
Full size table

The rages of missing values for the 20 items of the S-DSI-A showed variations across the samples: Argentina: 9.32% to 14.34%; Chile: 4.74% to 11.64%; Ecuador: 5.31% to 11.54%; Mexico: 4.49% to 10.82%; Peru: 3.11% to 9.51%; Spain: 4.49% to 10.82%. Missingness analysis showed there were no significant differences in missing data by gender. However, there were significant differences in terms of age-younger students had more missing data than older participants. Full information maximum likelihood (FIML) estimation was used in all subsequent analyses to handle missing data.

Table 2 also presents the correlation matrix, means, and standard deviations for the two S-DSI-A subscales and total score. As expected, the correlations between the two S-DSI-A subscales were strong and positive, ranging between r = 0.78—0.90 across the six samples. Correlations between the S-DSI total scores and subscales were also positive and strong, ranging from r = 0.94—0.98.

Main analyses

Configural invariance

We first specified a MGCFA model that specified configural invariance, which means the same factor structure (same pattern of fixed and free factor loadings) was estimated simultaneously on all six groups with no other between-groups constraints. Results presented in Table 3 suggest a model that fits the data well, with CFI = 0.906, RMSEA = 0.052, 90% CI [0.050, 0.054], χ2(996) = 3390.929 p = 0.001. These results supported the presence of a two-factor S-DSI model containing ER and EC subscales across the six samples. The factor loadings of each of the items are presented in (Table 4).

Table 3 Tests of measurement invariance of a two-factor model of differentiation of self
Full size table
Table 4 Factor loadings of the S-DSI-A scale items
Full size table

Metric invariance

We next tested a model that specified metric invariance, which has been cited as a minimum requirement to assume construct validity across different samples or groups (Kline, 2016). In this model, we placed equality constraints on all factor loadings across the six samples. Metric invariance is met if this model does not result in a deterioration of fit compared with the configural model. Results from the tests of Model 2, which tested for metric invariance, suggest the model fit the data well. Although the chi-square test was significant, X2(1086) = 3579.98, p < 0.001, the CFI (0.901) and RMSEA (0.051) values suggest a good fitting model. Additionally, the ΔCFI and ΔRMSEA values, ΔCFI = -0.004, ΔRMSEA = -0.001, are beneath those cutoff criteria for accepting measurement invariance (see Table 3).

Scalar invariance

In our final model, Model 3, we tested scalar invariance on the S-DSI-A among the six samples. Results indicated that, as a function of the additional constraints on the item intercepts, there was substantial deterioration in model fit, with the CFI value exceeding those that indicate good fit (0.878). Additionally, the ΔCFI (-0.023) was also beyond the recommended cutoff points to reject scalar invariance (i.e., -0.01; See Table 3). We attempted to establish partial scalar invariance; however, after examining the modification indices and making sequential changes to the model (i.e., allowing DSI 18 and DSI 20 intercepts to vary freely among the six samples) model fit did not improve (CFI = 0.880, RMSEA = 0.054, X2(1166) = 4210.17, p < 0.001).

Discussion

The present study aimed to establish the measurement equivalence of the Spanish Differentiation of Self Inventory for Adolescents (S-DSI-A) across Spanish-speaking countries, including Argentina, Chile, Ecuador, Mexico, Peru, and Spain. The primary objective was to determine whether the S-DSI-A could be used reliably and validly across different Spanish-speaking cultural contexts, enabling researchers and practitioners to make meaningful comparisons and draw accurate conclusions about adolescents' DoS. The study employed multigroup confirmatory factor analysis (MGCFA) to test for configural, metric, and scalar invariance of the S-DSI-A across the six samples. The results indicated that the S-DSI-A demonstrated configural and metric invariance, but scalar invariance was not achieved. Thus, partial invariance was obtained for the S-DSI-A for Spanish-speaking adolescents in the sampled countries.

The findings of configural invariance indicated that the factor structure of the S-DSI-A, which includes the Emotional Reactivity (ER) and Emotional Cutoff (EC) subscales, was consistent across the six Spanish-speaking samples. The metric invariance results further supported the construct validity of the S-DSI-A, as the factor loadings of the items were similar across the samples. These findings indicate that the S-DSI-A measures the same underlying constructs of ER and EC in a consistent manner across the Spanish-speaking countries. However, the study did not find scalar invariance, indicating that the means of ER and EC are not comparable across the six samples (Zhao et al., 2023).

The lack of scalar invariance highlights the importance of considering cultural and contextual factors that may influence the manifestation and interpretation of DoS. It is possible that cultural norms, values, and family dynamics vary across these countries, leading to differences in the levels of ER and EC. These results are consistent with those of a prior investigation of the S-DSI on adults (Rodríguez-González et al., 2020). In that study, configural and metric invariance were also established, signifying that the factor structure and factor loading of the DoS measures remained consistent across diverse cultural groups (i.e., Spain, Italy and the U.S.). Such results, in conjunction with those of the current study, suggest that participants from various cultures ascribe comparable meanings to the constructs of ER and EC. However, the non-invariance of latent means across the samples prevents direct comparisons of the means of ER and EC between the Spanish-speaking countries. Similarly to the current study, the study run by Rodríguez-González et al. (2020) also hindered direct comparison of mean levels of ER and EC participants from the US., Spain and Italy.

Future research should explore the specific cultural factors that contribute to variations in DoS across different populations. Moreover, future studies are needed to explore invariance in DoS among adults in Spanish-speaking countries, as this study with adolescents is the first where invariance among Spanish-speaking countries has been tested. According to Bowen (1978), DoS changes during adolescence and achieves stability in adulthood. Although empirical research exploring how this process unfolds is scarce, based on Bowen’s hypotheses, one would expect to find differences in DoS through development from adolescence and into adulthood. In the current study, we did not find any age differences across the six samples. Further, because our data were cross-sectional, we were not able to investigate changes in DoS across different development periods or age. Longitudinal research beginning in adolescences and into early adulthood is best positioned to investigate this theoretical proposition that DoS levels solidify in adulthood.

The findings of this study have important implications for the cross-cultural measurement and understanding of DoS. Establishing measurement invariance is essential to ensure that the assessment tool measures the same construct consistently across diverse populations (Gracia et al., 2019). By demonstrating configural and metric invariance, the study provides evidence for the cross-cultural applicability of the S-DSI-A in Spanish-speaking countries. However, the lack of scalar invariance highlights the need for caution when making comparisons between different cultural groups. Researchers and clinicians should be mindful of cultural variations in DoS and consider the cultural context when interpreting and applying the S-DSI-A. Because partial invariance was obtained on the S-DSI-A across Argentina, Chile, Ecuador, Mexico, Peru, and Spain, future research could examine associations between DoS and other variables of interest across these different groups, as correlation, regression and structural equation modeling are all appropriate statistical techniques when partial invariance on a given measure has been demonstrated (Steenkamp & Baumgartner, 1998).

Although all the countries included in our sample are Spanish-speaking, it is important to acknowledge that there exist important cultural differences among them. For example, Belintxon et al. (2022) found that adolescents from Chile, Mexico, Peru, and Spain differed in the levels of certain internal developmental assets (i.e., interpersonal competence and commitment to learning) and in substance use. Additionally, Orejudo et al. (2022) found that the associations between some family variables and positive youth development were different across Hispanic countries. Furthermore, the findings of a study by van Zyl and Meiselman (2015) provide concrete evidence supporting the notion that cultural differences exist not only across different languages but also within the same language across different countries. Their research examined the emotional responses and language usage related to beverages among Spanish-speaking countries, specifically comparing Mexican and Spanish respondents. The results revealed notable distinctions between the two groups, with Mexican respondents reporting a higher percentage of applicable emotions to beverages compared to Spanish respondents. Intriguingly, there were greater similarities in relevant emotions between Mexican respondents and English-speaking respondents from countries such as the USA, UK, Australia, and New Zealand, than between Mexican and Spanish respondents, despite sharing the same root language. These findings underscore the importance of considering cultural factors, beyond nationality, when examining the responses elicited by research inventories (van Zyl & Meiselman, 2015). In sum, these findings emphasize the need to account for cultural disparities and variations when assessing measurement invariance and making cross-cultural comparisons in research. These variations may have contributed to the lack of evidence documenting full measurement invariance across the six Spanish-speaking samples included in this study. Factors such as social norms, values, educational systems, and historical contexts differ among Spanish-speaking countries (Buré-Reyes et al., 2013). These differences can, in turn, shape individuals' perceptions and responses to the measures used in our study. Moreover, these prior studies together with our results in the current study suggest future research should be conducted to test the DSI-R invariance across English-speaking countries (e.g., UK, USA, Canada, Australia), as (to the best of our knowledge) they have not been conducted so far.

The strengths of this study include the large sample size and the inclusion of multiple Spanish-speaking countries, which enhances the generalizability of the findings. However, there are some limitations to consider. In the current study, we relied on self-report measures, which are subject to potential biases and limitations. Future studies might incorporate a qualitative approach, in addition to self-report, to provide a more comprehensive understanding of DoS. Additionally, the reliance on an anonymous questionnaire administered online during class time introduces certain limitations. The online format may have influenced the participants' responses, potentially leading to response biases. It is also important to note that the acquisition of parental permission varied across schools, as each institution followed its own policies. This variation in obtaining parental consent may have introduced selection biases. Therefore, caution should be exercised when interpreting the results, considering the potential limitations associated with the study's methodology.

Bowen (1978) postulated the cross-culture universality of his theory, including the principles and hypotheses related to the differentiation of self construct. In this regard, this study provides only partial confirmation of this assumption, as scalar invariance was not achieved. According to Bowen’s theory, self-report is not the optimal approach to measure levels differentiation of self, and this partial confirmation through invariance testing could be explained by experts in the theory as a potential measurement effect. Moreover, differentiation is a multidimensional construct and, accordingly, although this is not a statement postulated by Bowen, previous research (e.g., Rodríguez-González et al., 2020), together with the current study, support the idea that some traits or dimensions of differentiation are more stable across cultures than others.

This study is only a first step into investigations that seek to further understand the applicability of BFST to adolescent populations, for example, future research is needed to test the development and stability of DoS during adolescence and into emerging adulthood, and the ability of the individual to differentiate from one’s family of origin and achieve levels of DoS that are either similar or different. Even though the research on the development of DoS in adolescence is lacking, prior research has shown that DoS in adolescence is a relevant factor to adolescent development. For example, Knauth et al. (2006) found that DoS in adolescents aged 14–19 is a cognitive factor that is predictive of health risk behaviors such as substance use through the modulation of adolescent anxiety. This research suggests that DoS, even if not fully developed, may play a contributing role on adolescent risk behavior development.

In conclusion, the present study contributes to the literature by examining the measurement equivalence of the S-DSI-A across Spanish-speaking countries. The findings support the configural and metric invariance of the S-DSI-A, indicating its cross-cultural applicability in assessing differentiation of self. However, the lack of scalar invariance emphasizes the need to consider cultural variations in the interpretation and application of the S-DSI-A, and highlights the need to conduct invariance tests even between countries with the same language (e.g., Spanish, English, Portuguese). Future research should continue to explore the cultural factors influencing differentiation of self and develop culturally sensitive measures to enhance our understanding of this important construct in diverse populations.