Abstract
Identity is one core developmental task of adolescence. Although Marcia’s model, comprising of the dimensions of exploration and commitment, has dominated identity research for decades new models have recently been proposed. Luyckx and colleagues’ model poses that identity is a process consisting of five aspects: Exploration in Breadth, Commitment Making, Ruminative Exploration, Exploration in Depth and Identification with Commitments. The Dimensions of Identity Development Scale (DIDS) is a 25-item instrument developed to assess those five aspects. The goal of this study is: (a) to test the applicability of DIDS in a sample of Greek adolescents, and (b) to investigate the longitudinal measurement invariance of the scale. The results support the use of DIDS in Greek context and show that strong measurement invariance holds longitudinally in the course of 12 months. Echoing recent studies, the six-factor model showed significantly better fit, with Exploration in Depth splitting to Exploration in Depth and Reconsideration of Commitment. The scale is suitable for studies of longitudinal change in identity development.
Introduction
Identity broadly refers to the sense of personal uniqueness and sameness across time and contexts (Erikson, Citation1968). Identity synthesis is one of the core developmental tasks of adolescence (Motti-Stefanidi, Citation2015). Erikson theorized about the different aspects of identity, one of which is personal identity (Schwartz, Luyckx, & Vignoles, Citation2011). The most prominent model which translated Erikson’s theory into measurable constructs has been Marcia’s model of identity exploration and commitment (Marcia, Citation1980). According to this model, adolescents form an identity through a combined process of identity exploration and commitment. The presence or absence of these two dimensions results in a typological model of four statuses (Achievement, Foreclosure, Moratorium, and Diffusion).
Despite this model’s prominence for decades, there have been major developments in the field of identity the last years (Schwartz, Citation2001). New models have been proposed which tackle the major critiques of Marcia’s model, like the fact that this model presents a static rather than a dynamic picture of identity development (Côté & Levine, Citation1988). One such model is the five-dimensional model proposed by Luyckx and colleagues (Citation2008).
This model posits that identity is a dynamic process, and captures four aspects of this process: Exploration in Breadth, Commitment Making, Exploration in Depth, Identification with Commitments, whereas Ruminative Exploration is considered a risk factor for identity development (Beyers & Luyckx, Citation2016; Luyckx, Goossens, Soenens, & Beyers, Citation2006; Luyckx et al., Citation2008). Those aspects are used to describe two phases in identity dynamics, hence this model is called the ‘Dual-cycle model’ (Luyckx et al., Citation2006). During the first cycle, commitment formation, adolescents explore different options and make some initial commitments, and during the second cycle, commitment evaluation, adolescents explore these particular options in depth, and either identify with the initial commitments or start a new commitment formation cycle.
Importance of the DIDS Model
Luyckx and colleagues’ model of identity development has proved fruitful in developmental research. Distinguishing between different aspects of identity has led to a better understanding of theoretically relevant relations. For example, longitudinal relations between identity and self-esteem have been shown to be different for those dimensions (Luyckx et al., Citation2013), leading to a finer understanding of both phenomena and to possible theoretical refinements.
Furthermore, this model has been tested in different cultural contexts. Zimmermann, Lannegrand-Willems, Safont-Mottay, and Cannard (Citation2015) tested this model in France and the French-speaking part of Switzerland. Skhirtladze, Javakhishvili, Schwartz, Beyers, and Luyckx (Citation2016) tested the model in Georgia. Both studies generally supported the utility of the DIDS, although both showed that a six-factor solution might be more appropriate.
The Greek context
Adolescent development in the Greek context presents challenges both common with those in other European countries, and unique (Efklides & Moraitou, Citation2007). For example, Greece is one of the most westernized countries of South-Eastern Europe, and therefore western values affect youth development to a large extent (e.g., lifestyle, ways to spend free time). On the other hand, the strong influence of family relationships in domains such as education and career choice even through emerging adulthood and beyond, is a rather unique aspect of youth development in Greece. Finally, the financial crisis with the increasing unemployment which affects mostly young people and, thus, may seem to destroy their future prospects, is another aspect unique to the Greek context. For example, according to Eurostat unemployment rates for youth under age 25 were 55.3 and 58.3% during 2012 and 2013, whereas at the same time the European-28 unemployment rates were 23.3 and 23.7% respectively.
For those reasons, it is important to extend knowledge on identity development in this context.
The present study
Since identity is embedded in context and influenced by societal processes it is important to test whether instruments work in similar ways across cultures. Additionally, identity is in flux during adolescence, hence longitudinal studies are appropriate and warranted, if we want to better understand this significant aspect of individual uniqueness.
Furthermore, the fact that new identity models followed by new instruments have proliferated in the last years has led to theoretical discussions regarding the utility of those new models and instruments for better understanding identity as Erikson (Citation1968) sketched it. For example, Waterman (Citation2015) expressed the concern how close these new models and instruments remain to the original ideas regarding identity. He elaborated on the downsides of each of the major new instruments of identity proposing future directions of investigation. He suggested that whenever results using new models of identity run contrary to theory, we should consider the possibility that the instruments and/or methodologies are flawed, along with the possibility that the theory needs to be updated.
Hence, although the focus of the present study is more methodological rather than theoretical, its results can add to the extant literature in two ways. First, by testing the utility of the instrument per se; second, by adding to the understanding of the model behind DIDS in an understudied cultural context. The use of a longitudinal design is a further asset, as, to our knowledge, there hasn’t been previous longitudinal research on the structural properties of this instrument. Therefore, the results of this examination of methods, can contribute to the ongoing discussion about the modern conceptualizations of identity.
In sum, the goals of the present study are to test the psychometric properties of the scale in a sample of Greek adolescents, as well as to test its longitudinal measurement invariance.
Methods
Sample
The analytic sample consisted of 437 Greek students (49.3% boys), 15.7 years old (SD = .76) at Time 1. For more information regarding procedure, see Appendix 1. The majority of the students took part in all three waves (N = 291), and fewer took part in two (N = 87) or only one (N = 59) waves. Students participating in one, two, or three waves were compared, by means of Univariate ANOVA’s on all 25 variables (5 identity dimensions and 3 adaptation indices at three time-points, plus SES) and were not found to differ statistically significantly (see Supplementary material, Table ).
Table 1. Fit indices for the 5- and 6-factor DIDS CFA models, T1–T3 (N = 437).
Measures
Identity
The Dimensions of Identity Development Scale (DIDS) is a 25-item self-report questionnaire developed by Luyckx and colleagues (Citation2008). Each of the five dimensions is measured by 5 items, assessed on a 5-point Likert scale. Example items for each of the dimensions are: I have decided on the direction I want to follow in life (Commitment Making – CM), I think about the direction I want to take in my life (Exploration in Breadth – EB), I keep looking for the direction I want to take in my life (Ruminative Exploration – RE), Plans for the future offer me a sense of security (Identification with Commitments – IC), and I think about the future plans I have made (Exploration in Depth – ED).
The original questionnaire was obtained in early 2012 directly from the authors who published it. Translation in Greek was done by three bilingual psychologists, and the three translations were compared to each other. Disagreements were discussed upon (see Supplementary material, Table ).
Psychological symptoms
The Greek version of the Symptoms Checklist 90 – Revised (Donias, Karastergiou, & Manos, Citation1991) was used to measure symptoms of depression and anxiety. Depression was measured with 13 items, on a 5-point Likert scale (e.g., ‘How much were you bothered by feeling low on energy or slowed down’). Anxiety was measured with 10 items, on a 5-point Likert scale (e.g., ‘How much were you bothered by nervousness or shakiness inside’).
Self-esteem
The Greek Rosenberg Self-Esteem Scale (Galanou, Galanakis, Alexopoulos, & Darviri, Citation2014) was used to assess self-esteem. The scale consists of 10 items, on a 5-point Likert scale. Example item is “On the whole, I’m satisfied with myself”.
Analytic Plan
Using Mplus 7 (Muthén & Muthén, Citation1998–2012), we followed four steps to data analysis. First, we prepared the data by checking for missing values and outliers. Second, we applied a series of Confirmatory Factor Analyses (ML estimator), to test the factor structure of the Greek DIDS separately for each wave. In this step, we had first opted for an Exploratory Structural Equation Modeling approach (Marsh, Morin, Parker, & Kaur, Citation2014), to test the factor structure of the DIDS in each wave. However, during the review process this approach was dropped in favor of the CFA. The reasons for this choice were (a) the comparability of the current results with extant literature (which used CFA), and (b) the fact that ESEM has been reported to have parsimony problems in large scales (Marsh et al., Citation2014). Third, we performed a series of CFA analyses (ML estimator) to test for the longitudinal measurement invariance of the instrument. During steps 2 and 3, model comparisons were based on CFI, TLI, RMSEA, SRMR, BIC, and AIC. Finally, we computed the bivariate correlations between the DIDS dimensions with depression, anxiety, and self-esteem.
Results
Preliminary analyses
For information regarding outlier detection and handling please refer to Appendix 1. Attrition did not pose a threat to the internal validity of the study (see Supplementary material). Furthermore, we performed a Little’s MCAR test on the level of items, which, although significant, resulted in an acceptable normed χ2: χ2(5633) = 6521, χ2/df = 1.33. However, we employed two-level multiple imputation in order to tackle both simple missingness and attrition (see Appendix 1).
Factor structure of the DIDS within waves
Following recent advances which showed that a six-factor solution fits the data better (Beyers & Luyckx, Citation2016; Skhirtladze et al., Citation2016), we compared the original five-factor with the six-factor models for each wave separately. The ESEM approach supported the five-factor over the six-factor solution.Footnote1 However, the CFA approach supported the six-factor solution, over the five-factor. Results from the CFA approach are summarized in Table (see also Figure 1). For reasons stated above, we based our further analyses on the results from the CFA approach. Descriptive statistics of the six DIDS scales are presented in Table .
Figure 1. Standardized estimates for the DIDS measurement model at T1.
Table 2. Descriptive statistics for the six DIDS scales, T1–T3.
Longitudinal measurement invariance
We tested the configural invariance, where the same model applied to all three waves, covariances between same items across time as well as factor covariances were calculated, but no restriction was imposed. Furthermore, all factors were regressed on gender and SES in all waves. The fit was good (Table ).
Table 3. Fit indices for testing longitudinal measurement invariance of the 6 factor DIDS (N = 437).
Imposing item loadings equalities across time did not worsen the fit significantly, thus metric invariance was attained. Further restricting item intercepts to be equal across time led to small changes in the fit indices (ΔCFI = .002, ΔTLI = .002, ΔRMSEA = .001). Furthermore, this model had the lowest BIC, thus the best trade-off between model fit and complexity (see Table ). Therefore, strong invariance was attained for the DIDS.
Bivariate correlations
After having investigated the factor structure of the DIDS, and the longitudinal measurement invariance of its 6 dimensions, we computed their bivariate correlations with depression, anxiety, and self-esteem. Table presents those bivariate correlations, in all three waves. Four out of six identity dimensions correlate significantly with depression and self-esteem, in theoretically predictable ways.
Table 4. Cross-sectional bivariate correlations between DIDS dimensions, symptoms of depression, anxiety, and self-esteem, T1/T2/T3.
Discussion
The goal of this study was to test the construct validity of the DIDS in a sample of Greek adolescents, and to test its longitudinal measurement invariance. The results showed that the instrument is appropriate for use in this cultural context, as it shows the same six-dimensional structure it has also shown in other countries (Beyers & Luyckx, Citation2016; Zimmermann et al., Citation2015). Furthermore, the scale showed longitudinal measurement invariance in one year. Finally, the correlations of the six dimensions with indices of adaptation are generally in accordance to what has been found in other studies (Beyers & Luyckx, Citation2016; Ritchie et al., Citation2013).
The contradictory findings from the ESEM and CFA approaches regarding the number of factors are possibly a result of the features of each approach (Marsh et al., Citation2014). For example, we noted that applying the six-factor ESEM models resulted in many items failing to show any significant loadings with any factor. In other words, not only did the six-factor ESEM models not support the presence of a sixth factor consisting of items 23–25, but they led to the erroneous concealment of the significant loadings shown by items 1–22. Keeping in mind that the ESEM approach needs more development (Marsh et al., Citation2014), but also following recent studies using DIDS (e.g., Beyers & Luyckx, Citation2016), we conclude that the CFA approach, which supports the six-factor structure, better reflects the true DIDS properties.
Longitudinal measurement equivalence of a construct is essential in order to investigate its temporal change (Brown, Citation2006). Therefore, this study shows that the DIDS is appropriate for studying the longitudinal relations of identity with other constructs (e.g., in cross-lagged models), as well as for studying the trajectories identity follows through adolescence. Thus, this study contributes to the goal of better understanding identity processes, by showing that a relatively new instrument of identity assessment works well both in a different cultural context (Greece), and in a longitudinal design. However, as Waterman (Citation2015) stressed, the utility of the theoretical model behind the instrument studied in the current article should be evaluated with regards to the degree it helps clarify and/or update the original identity theory (Erikson, Citation1968).
Furthermore, the different significant correlations of most of the DIDS factors with depression, anxiety, and self-esteem, support the scale’s validity. Of note, the 6 vs. 5 factor structure is further supported by the correlations of Exploration in Depth and Reconsideration of Commitment with self-esteem, which have the opposite signs. Recent studies have showed the negative role that Reconsideration of Commitment plays in healthy identity development, as well as adaptation (Beyers & Luyckx, Citation2016). The current results show that Reconsideration of Commitment is positively related with depression, negatively related with self-esteem, and that these correlations are stable in time. On the other hand, Exploration in Depth is related positively to self-esteem, supporting the idea that it is a healthy process of identity development (Beyers & Luyckx, Citation2016).
Conclusion
This study adds to the existing literature by showing that the DIDS is a useful instrument for assessing longitudinal dynamics of identity development in Greek context. The clear support of a sixth factor means that identity dynamics can be studied and understood in more detail, and longitudinally.
Supplemental data
Supplemental data for this article can be accessed at http://dx.doi.org/10.1080/17405629.2016.1241175
Disclosure statement
No potential conflict of interest was reported by the authors.
Supplementary_Material__3_Tables___1_.docx
Download MS Word (32.8 KB)Acknowledgements
Stefanos Mastrotheodoros was supported with a doctoral scholarship by the “Alexander S. Onassis” Public Benefit Foundation. The authors would like to thank Professor Dr. Jens Asendorpf for his comments during earlier drafts of this paper.
Notes
1 Results from the ESEM approach can be obtained from the first author.
References
- Aguinis, H., Gottfredson, R. K., & Joo, H. (2013). Best-practice recommendations for defining, identifying, and handling outliers. Organizational Research Methods, 16, 270–301. doi:10.1177/1094428112470848
- Beyers, W., & Luyckx, K. (2016). Ruminative exploration and reconsideration of commitment as risk factors for suboptimal identity development in adolescence and emerging adulthood. Journal of Adolescence, 47, 169–178. doi:10.1016/j.adolescence.2015.10.018
- Brown, T. A. (2006). Confirmatory factor analysis for applied research. New York, NY: Guilford.
- Côté, J. E., & Levine, C. (1988). A critical examination of the ego identity status paradigm. Developmental Review, 8, 147–184. doi:10.1016/0273-2297(88)90002-0
- Donias, S., Karastergiou, A., & Manos, N. (1991). Standardization of the symptom checklist-90-R rating scale in a Greek population. Psychiatriki, 2, 42–48.
- Efklides, A., & Moraitou, D. (2007). Greece. In J. J. Arnett (Ed.), International encyclopedia of adolescence (pp. 354–369). New York, NY: Routledge.
- Erikson, E. H. (1968). Identity and identity diffusion. The self in social interaction, 1, 197–205.
- Galanou, C., Galanakis, M., Alexopoulos, E., & Darviri, C. (2014). Rosenberg self-esteem scale Greek validation on student sample. Psychology, 5, 819–827. doi:10.4236/psych.2014.58093
- Luyckx, K., Goossens, L., Soenens, B., & Beyers, W. (2006). Unpacking commitment and exploration: Preliminary validation of an integrative model of late adolescent identity formation. Journal of Adolescence, 29, 361–378. doi:10.1016/j.adolescence.2005.03.008
- Luyckx, K., Klimstra, T. A., Duriez, B., Van Petegem, S., Beyers, W., Teppers, E., & Goossens, L. (2013). Personal identity processes and self-esteem: Temporal sequences in high school and college students. Journal of Research in Personality, 47, 159–170. doi:10.1016/j.jrp.2012.10.005
- Luyckx, K., Schwartz, S. J., Berzonsky, M. D., Soenens, B., Vansteenkiste, M., Smits, I., & Goossens, L. (2008). Capturing ruminative exploration: Extending the four-dimensional model of identity formation in late adolescence. Journal of Research in Personality, 42, 58–82. doi:10.1016/j.jrp.2007.04.004
- Marcia, J. E. (1980). Identity in adolescence. In J. Adelson (Ed.), Handbook of adolescent psychology (vol. 9, pp. 159–187). New York, NY: Wiley & Sons.
- Marsh, H. W., Morin, A. J. S., Parker, P. D., & Kaur, G. (2014). Exploratory structural equation modeling: An integration of the best features of exploratory and confirmatory factor analysis. Annual Review of Clinical Psychology, 10, 85–110. doi:10.1146/annurev-clinpsy-032813-153700
- Motti-Stefanidi, F. (2015). Identity development in the context of the risk and resilience framework. In K. C. McLean & M. Syed (Eds.), The Oxford handbook of identity development (pp. 472–489). Oxford: Oxford University Press. Retrieved from http://doi.org/10.1093/oxfordhb/9780199936564.001.0001
- Muthén, B. & Muthén, L. (1998–2012). Mplus 7. Los Angeles, CA: Author.
- Reitz, A. K., Motti-Stefanidi, F., & Asendorpf, J. B. (2016). Me, us, and them: Testing sociometer theory in a socially diverse real-life context. Journal of Personality and Social Psychology, 110, 908–920. doi:10.1037/pspp0000073
- Ritchie, R. A., Meca, A., Madrazo, V. L., Schwartz, S. J., Hardy, S. A., Zamboanga, B. L., … Lee, R. M. (2013). Identity dimensions and related processes in emerging adulthood: Helpful or harmful? Journal of clinical psychology, 69, 415–432. doi:10.1002/jclp.21960
- Schwartz, S. J. (2001). The evolution of Eriksonian and neo-Eriksonian identity theory and research: A review and integration. Identity: An International Journal of Theory and Research, 1, 7–58. doi:10.1207/S1532706XSCHWARTZ
- Schwartz, S. J., Luyckx, K., & Vignoles, V. (2011). Handbook of identity theory and research. New York, NY: Springer. doi:10.1007/978-1-4419-7988-9
- Skhirtladze, N., Javakhishvili, N., Schwartz, S. J., Beyers, W., & Luyckx, K. (2016). Identity processes and statuses in post-Soviet Georgia: Exploration processes operate differently. Journal of Adolescence, 47, 197–209. doi:10.1016/j.adolescence.2015.08.006
- Waterman, A. S. (2015). What does it mean to engage in identity exploration and to hold identity commitments? A methodological critique of multidimensional measures for the study of identity processes. Identity, 15, 309–349. doi:10.1080/15283488.2015.1089403
- Zimmermann, G., Lannegrand-Willems, L., Safont-Mottay, C., & Cannard, C. (2015). Testing new identity models and processes in French-speaking adolescents and emerging adults students. Journal of Youth and Adolescence, 44, 127–141. doi:10.1007/s10964-013-0005-7
Appendix 1
Sample and procedure
This study is part of a bigger survey on adolescent adaptation which was conducted by the Department of Psychology, of the University of Athens, Greece.
The total sample consisted of 685 adolescent students, attending 8 public high schools in Athens, Greece. The schools were selected from the pool of all the high schools in Attiki (the prefecture Athens lies in, with almost half the country’s population). Access to this pool was given by the Greek Ministry of Education. In order to broaden the population of interest, from this pool we selected 8 high-schools from different parts of Athens metropolitan area, corresponding to different socio-economic strata: 3 schools from the center of Athens (low / lower-middle class), 3 schools from middle-class areas (western, southern, and eastern parts of the city), one school from an upper middle-class suburb (north), and one school from a less-urbanized middle class town outside Athens (east).
The students were assessed three times in 12 months (two 6-month intervals), between March 2012–March 2013. The procedures were identical in all three waves. Trained assistant researchers visited the classrooms during school hours. Questionnaire completion took part in two-hour slots, after the school principal’s permission.
For the purposes of this study, only adolescents with both parents born in Greece were included, something which is in accordance with other recent studies in similar samples (e.g., Reitz, Motti-Stefanidi, & Asendorpf, Citation2016. Furthermore, because there was only one vocational school in the original sample, which comes from a different population (very high ethnic diversity in this school) the students of the vocational high school were excluded. The analytic sample of 437 adolescents is the one after the outlier handling (see below).
SES: The socioeconomic status is a composite score comprising of seven variables: mother education, father education, mother employment, father employment, family status, own house (vs. rent), and home density. Higher scores indicate lower socioeconomic status. Those variables were measured in all three waves, resulting in three measures of SES, which were then combined in one general SES variable.
Outlier detection and management
In order to identify and handle possible outliers, based on recent best-practice recommendations (Aguinis, Gottfredson, & Joo, Citation2013) we applied the following: (1) For each composite score of the DIDS we computed the Z-scores for each wave. Cases with scores lower than −2.24 or higher than 2.24 were labeled as Possible Error Outliers (PEO). (2) The boxplots of each of those variables was also inspected, and cases outside the whiskers were also labeled as PEO. (3) Leverage and studentized deleted residuals were also used as means to locate more PEO. Step 1 resulted in 108 cases, whereas steps 2 and 3 resulted only in cases that were already included in those of step 1.
Then, all the 75 items of each of those cases were inspected for coding errors (none was found), or other possible errors. In 11 of those cases the DIDS was obviously answered carelessly, e.g., giving only 5’s in all 25 items. In 6 cases this was done in all three waves, whereas in the rest 5 cases this only happened to one wave. Given that correction was not possible, the former cases were deleted completely, whereas for the latter the items of the problematic wave were considered missing values. The rest 97 cases were not found to clearly be PEO, and therefore were considered Possible Interesting/Influential Outliers (PIO). After multilevel multiple imputation the CFA’s were ran both with and without the PIO cases. No differences in model fit were found.
Dealing with missingness – performing multiple imputation
We performed a multilevel multiple imputation procedure to tackle missingness and attrition. Individual SES, gender, school type, and PIO were entered at level 2, and all other variables (DIDS items) at T1–T3 were nested within individuals, and therefore were entered at level 1. Forty new datasets were asked.