Adolescents’ Identity Formation: Linking the Narrative and the Dual-Cycle Approach

The narrative and dual-cycle approach conceptualize and operationalize adolescents’ identity formation in different ways. While the narrative approach focuses on the construction of an autobiographical life story, the dual-cycle approach focuses on the formation of identity commitments. Although these approaches have different emphases, they are conceptually complementary. Yet, their empirical links and distinctions have only scarcely been investigated. Empirical knowledge on these links in adolescence and across time has been especially lacking. In the present research, it was therefore examined whether key characteristics of adolescents’ narration (autobiographical reasoning and agency) were concurrently and prospectively related to engagement in the dual-cycle processes of commitment making, identification with commitment, exploration in breadth, exploration in depth, and ruminative exploration. The findings from a cross-sectional sample of 1,580 Dutch adolescents (Mage = 14.7 years, 56% female) demonstrated that autobiographical reasoning was significantly positively associated with the commitment and more adaptive exploration processes (i.e., in breadth and in depth). In addition, agency was significantly positively associated with the commitment processes and exploration in depth. Yet, these associations between the narrative characteristics and dual-cycle processes were only weak. Subsequently, the findings from a two-year longitudinal subsample (n = 242, Mage = 14.7 years, 62% female) indicated that on average commitment strength remained stable but exploration increased across middle adolescence. A stronger increase in identification with commitment and adaptive exploration (i.e., in breadth and in depth) was predicted by a higher degree of agency in adolescents’ narratives. Overall, these findings indicate that both approaches to identity formation are associated, but the small size of these associations suggests that they predominantly capture unique aspects of identity formation. Both approaches could thus complement and inform each other.

Note. From the adolescents who did not write a turning point narrative (n = 311, 16.4% of total sample), 16 stated that they did not want to share one, 184 stated that they could not think of a turning point event, and 111 did not provide a clear reason. Adolescents were not asked to provide a reason when they did not write a turning point narrative. Note. Differences between adolescents with (Study 2 sample) and without (descriptive statistics shown here) a turning point narrative at T1 in the five identity processes across time were tested with t-tests. For the comparisons at T1, T2, and T3 the n of adolescents without a turning point narrative at T1 was 28, 19, and 18, respectively. If significant, adolescents without a turning point narrative at T1 scored lower (see Table 3). * p < .05; ** p < .01.

Measurement Invariance
Longitudinal measurement invariance of the DIDS subscales was tested following steps described by Widaman, Ferrer, and Conger (2010). A series of longitudinal Confirmatory Factor Analysis (CFA) models was estimated in Mplus 7 (Muthén & Muthén, 1998-2015. Each dual-cycle process was tested separately. For reasons of slightly nonnormally distributed variables, a Full Information Robust Maximum Likelihood estimator (MLR) was used (Satorra & Bentler, 2001). Model fit was considered to be acceptable when the Comparative Fit Index (CFI) was above 0.90 and the Root Mean Square Error of Approximation (RMSEA) was below 0.08 (Byrne, 2013). A significant reduction in model fit was concluded if two of the following three criteria were met: Δχ 2 SB significant at < .050, ∆CFI < -.010, and ∆RMSEA > .015 (Chen, 2007;Cheung & Rensvold, 2002).
First, it was examined whether a longitudinal Confirmatory Factor Analysis (CFA) model fitted the data sufficiently (Widaman et al., 2010). A sufficient model fit would indicate configural invariance, which refers to the validity of the same CFA at every wave (Van de Schoot, Lugtig, & Hox, 2012). In addition to CFA models for one of the identity processes at all three waves, longitudinal CFA models contained covariances between the latent variables and between the residuals of like items across the waves. Only the model for exploration in breadth did not fit the data well, χ 2 SB (72) = 421.88, p < .001, CFI = .68, RMSEA = .14. The Modification Indices indicated that the residuals of the items "I am considering a number of different lifestyles that might suit me" and "I am thinking about different lifestyles that might be good for me" were correlated within every wave. Correlating the residuals of these items was deemed reasonable, as these items were more alike than others. Including these correlations resulted in an acceptable model fit for the model of exploration in breadth. The fit statistics of all models are displayed in Table S2.
Second, metric invariance was tested by constraining all factor loadings of like items to be equal across time. For none of the identity processes this resulted in a significantly worse model fit (see Table S2). Third, scalar invariance was tested by also constraining all intercepts of like items to be equal across time. Again, this did not result in a significantly worse model fit for any of the commitment and exploration processes (see Table S2).
Moreover, all final models in which factor loadings and intercepts of like items were constrained to be equal across time had a sufficient model fit.
Lastly, it was checked whether constraining strict invariance would decrease the model fit by constraining the residuals of corresponding items to be time invariant. Adding these constraints did not significantly worsen the model fit for the model of commitment making, identification with commitment, and ruminative exploration (see Table S2). However, the model fit for the model of exploration in breadth, Δχ 2 SB (10) = 22.09, p = .015, ΔCFI = -.011, ΔRMSEA = .000, and exploration in depth, Δχ 2 SB (10) = 25.09, p = .005, ΔCFI = -.021, ΔRMSEA = .007, became significantly worse. Next, it was checked for both models for which item constraining the residuals resulted in the biggest decrease in model fit, and subsequently which wave resulted in the biggest decrease in model fit. Based on these findings, a model was tested in which the residuals of four items were constrained across time and the residual of one item was constrained across two waves. The residual of this latter item was freely estimated at one wave. For exploration in breadth this was the item "I think about different goals that I might pursue" at T3. For exploration in depth this was the item "I think about whether the aims I already have for life really suit me" at T2. These models with partial strict invariance did not fit significantly worse than the models with scalar invariance (see Table S2). Note. a The change in model fit refers to a comparison with the model in the previous line. b The model of exploration in breadth included correlations between the residuals of two pairs of items within every wave, as described in the text.

Table S4 Comparisons between Adolescents With and Without a Self-event Connection at T1 in the Commitment and Exploration Processes Across the Three Waves (Study 2)
DIDS T1 DIDS T2 DIDS T3 Self-event connection T1: Note. DIDS = Dimensions of Identity Development Scale. Differences between the two groups were tested with t-tests, but were not significant, p ≥ .079.