The labor market in times of economic uncertainty makes it increasingly difficult for young people to find a job. In this situation, the number of career transitions, both intentional or unintentional, and career choices that an individual must make have increased [International Labor Organization (ILO), 2020]. The coronavirus disease 2019 (COVID-19) pandemic has compounded these difficulties, raising a serious risk for young people that are the most at-risk category (ILO, 2020; Masenya, 2021).

Young people who remain excluded from the labor market, for the short or long term, are defined as not in education, employment or training (NEET). The NEET group is conceived as dynamic because young people are moving in and out of employment or education/training conditions in a short time and often involuntarily. Being excluded from the labor market represents a risk condition at various levels with serious consequences for the individual and society (Eurostat, 2021). On the social level, the failure to move successfully from school to work can harm society in terms of costs and problems related to unemployment (Caliendo and Schmidl, 2016; Morrison, 2005). On the individual level, the literature showed a strong relationship between the NEET condition and mental health (for a review, Bartelink et al., 2019), such as increased stress (Bjarnason and Sigurdardottir, 2003), risk of depression (Crowe and Butterworth, 2016), anxiety (Parola et al., 2022; Virtanen et al., 2016), and decreased mental health later in life (Strandh et al., 2014).

Young people during the pursuit of first employment need to be able to prevent such mismatches and wretched career starts (Koen et al., 2012). Although proper career preparation is useful to successfully find employment (van Hooft et al., 2021), the evidence of stagnating economy can interfere with the school-to-work transition (Fusco et al., 2021). In light of this, it would be helpful to enhance the needed resources to cope with the school-to-work transition to prevent prolonged unemployment (Koen et al., 2012).

Despite the increase in studies that bring attention to the NEET condition, too few have focused on the personal resources that would enable this group to cope with this uncertain era. According to Rossier (2015), regulation skills and adaptability resources, i.e., career adaptability, are needed for young people in this unstable and challenging environment for mastering unpredictable and changing tasks and demands. Career adaptabilities are crucial competencies because they allow the self-extension of the individual in the social environment (Savickas, 2013). The implications are twofold: measuring and monitoring adaptability resources are useful for planning career interventions both for prevention of NEET condition, supporting young people for mastering career transitions, and on the other hand to increase these competences in young NEET to enhance opportunities for entering the world of work. Career adaptabilities, as adaptability resources, can be key competencies for mastering career transitions throughout the lifespan (Rossier, 2015) handling the changing context. Indeed, several studies have found that career adaptability might help remedy the experience of job mismatch and underemployment (Koen et al., 2012).

Career adaptability, as a pivotal construct in career construction theory (CCT; Savickas, 1997, 2005, 2019, 2020), refers to a psychosocial construct that denotes an individual’s resources for coping with current and anticipated tasks, transitions, and traumas in their occupational roles (Savickas & Porfeli, 2012, p. 662). In the last years, career adaptability has been conceived as a needed resource for successful career development (Johnston et al., 2016; Zacher, 2015). This psychosocial construct allows to adaptively respond to challenges in the current era characterized by job uncertainty (Blossfeld et al., 2005). Career adaptability as a multidimensional and hierarchical structure includes four main resources (4 Cs): concern, control, curiosity, and confidence. Concern reflects a planful attitude toward the future. Control indicates belief about personal responsibility in career construction and the perceived control over their future. Curiosity is related to the personal ability to explore career opportunities. Confidence regards the perception of influencing the environment and overcoming obstacles in career paths through personal actions.

Literature showed that career adaptabilities were related to several dimensions of the work, i.e., job search self-efficacy (Guan et al., 2013), employability skills (de Guzman and Choi, 2013), work engagement (Yang et al., 2019), career calling (Douglass and Duffy, 2015; Guo et al., 2014). Furthermore, in the face of unemployment, career adaptability can increase the odds of finding employment (Zikic and Klehe, 2006). Indeed, career adaptability may represent a key resource to cope with stressful situations, such as unemployment, because it seems to mediate the relationship between personal dispositions, environment, and work outcomes (Johnston et al., 2013).

Career scholars agreed on the operationalization of career adaptability and to measure the construct the Career Adapt-Abilities Scale (CAAS, Savickas and Porfeli, 2012) was developed. The CAAS measure was developed and validated in several countries (Chen et al., 2020a; Rudolph et al., 2017). The CAAS, composed of 24 selected items, demonstrated the validity across countries and these led to a rapid increase of research on career adaptability in the last years. Validation of the Italian version of the CAAS was conducted by Soresi et al. (2012) and used among adolescents and young adults in several studies.

However, no studies so far have validated any versions of the CAAS, in NEET youth who are not in education, employment, or training. Only one study, in the Chinese version, has tested the CAAS measure in the NEET population (Su et al., 2021). The study showed that the original version of CAAS does not apply to NEET. The study proposed a two-factor structure drawn from 17 items of 24-CAAS to operationalize the career adaptability of NEETs. These two factors were reconceptualized as career exploration (eight items) and career self-efficacy (nine items). Career exploration refers to self-exploration and environment exploration for the sake of exploring possible opportunities for one’s future career. Career self-efficacy refers to a self-belief about one’s ability to make decisions, to be responsible for one’s own choices, and to perform tasks efficiently. According to the CCT, both dimensions are conceptualized as important dimensions of career adaptability (Chen et al., 2020b; Creed et al., 2009; Solberg et al., 1995).

Consistent with Su et al. (2021), it remains unknown whether the four-factor structure of CAAS applies to the target group of NEETs, given their higher level of vulnerability. For this reason, the present study aims to examine the factor structure of the Italian version of the CAAS in the NEET population. Specifically, the study proposes to test the original four-factor structure of the Italian version of the CAAS with the NEET population.

Method

Participants and procedure

Sample size calculation was based on the recommendations of quality criteria for measurement properties of questionnaires (Terwee et al., 2007), using at least ten subjects per item. Participants were recruited through the network with the Italian youth associations, respecting some inclusion and exclusion criteria. Inclusion criteria were: young people aged 18–34, and young adults who were not studying nor working for at least 4 weeks from the survey, following the recommendation of the NEET categorization of Eurofound (2012). Exclusion criteria were: not being a native Italian speaker, young people engaged in education, employment, or training programs. The sample is composed of 318 young NEET, 169 males (53.15%) and 149 females (46.85%), 155 aged 18–24 (48.7%) and 163 aged 25–34 (51.3%) [mean (M)age 24.45, standard deviation (SD)age 3.23].

Participation was voluntary, and respondents did not receive remuneration. The questionnaire was completed entirely online. Participants were informed about the aim of the study and were reassured of the confidentiality of their answers. The initial page contained a detailed description of the study, and the inclusion and exclusion criteria. Participants were then asked to acknowledge that they had read the terms and conditions by signing an informed consent form. This research was approved by the Ethical Committee of Psychological Research of the University of Naples Federico II.

Measure

The Career Adapt-Abilities Scale (CAAS)—Italian Form

The study used the 24-item Career Adapt-Abilities Scale—Italian form (Soresi et al., 2012). The participants were asked to answer on a five-point Likert scale from 1 (not strong) to 5 (strongest). The scale assesses the four dimensions of career adaptability: concern (six items), control (six items), curiosity (six items), and confidence (six items). The 24 items combine into a total score indicating adaptability. The original version of CAAS reported reliability of 0.83 for concern, 0.74 for control, 0.79 for curiosity, and 0.85 for confidence. The total score has reported reliability of 0.92. The CAAS—Italian form for the general population reported reliability of 0.80 for concern, 0.74 for control, 0.77 for curiosity, and 0.85 for confidence. The total score has reported reliability of 0.91.

Statistical analyses

Statistical analyses were performed with R software (v.4.0.5) and using the “lavaan” (v. 0.6-8) package (Rosseel, 2012). To test the factorial structural model of CAAS, a confirmatory factor analysis (CFA) was performed. Considering the nature of the response scale and distribution of the items, the robust maximum likelihood (MLM) estimator was used as an estimator for the CFA (Brown, 2015; Kline, 2016; Pietrabissa et al., 2020; Simpson et al., 2018). The MLM is a robust variant of maximum likelihood (Bentler, 1989) that provides robust standard errors and is also referred to as the Satorra–Bentler chi-squared test (S–Bχ2) to assess the model fit. Moreover, the comparative fit indexII (CFI), the Tucker–Lewis index (TLI), the root mean square error of approximation (RMSEA), the standardized root mean square residual (SRMR), and the ratio of S–Bχ2 to the degrees of freedom (df) were also used to assess model fit.

The following cut-off criteria were used to evaluate the goodness of fit: CFI and TLI with values between 0.90 and 0.95 (Brown, 2015; Kline, 2016), RMSEA and SRMR with values between 0.05 and 0.08 (Byrne, 2016; Browne and Cudeck, 1992; Pietrabissa et al., 2020), and a χ2/df ratio value of 3 or less.

The two-factor structure with 17 items proposed by Su et al. (2021) was tested and was not shown as an adequate fit: χ2(118) = 525.888, p < .001, TLI 0.834, CFI 0.856, RMSEA 0.087 (0.079–0.095), and SRMR 0.061.

Following the original 24-item validation study (Savickas & Porfeli, 2012), the model comparison was performed following this order: (A) a first-order four-factor model, (B) a first-order one-factor model, and (C) a second-order model (hierarchical). According to the CAAS literature (Soresi et al., 2012), the second-order model was specified by positing an overarching general latent dimension loaded by the four first-order latent dimensions (concern, control, curiosity, and confidence). Model comparison was assessed by using the test differences in fit indices, with the following criteria as cutoffs for model equivalence: ΔS–Bχ2 (p value > 0.050), ΔCFI (< 0.010), and ΔRMSEA (< 0.015) (Brown, 2015; Bryant and Satorra, 2012; Cheung and Rensvold, 2002; Millsap, 2011). Excess of the cutoff in two out of these three indices is evidence of model worsening (Bryant and Satorra, 2012; Cheung and Rensvold, 2002).

Lastly, measurement invariance was computed to evaluate whether the best structure was invariant across sex (males versus females) and age (18–24 years old versus 25–34 years old) (Manzoni et al., 2020; Vandenberg and Lance, 2000). Four nested models were sequentially performed constraining to equality models parameters across groups. First, factorial structure was constrained to equality (model 1: configural invariance). Second, factor loadings were constrained to equality (model 2: metric invariance). Third, factor loadings and intercepts of measured variables were constrained to equality (model 3: scalar invariance). Lastly, factor loadings, intercepts, and latent means were constrained to equality (model 4: latent means invariance) (Meredith, 1993; Millsap, 2011; Vandenberg and Lance, 2000). For both multigroup CFAs, models were sequentially compared by using test differences in three fit indices, with the following criteria as cutoffs for model equivalence: DIFFTEST (equal to Δχ2; p value > 0.050), ΔCFI (≤ 0.010), and ΔRMSEA (≤ 0.015) (Brown, 2015; Cheung and Rensvold, 2002; Millsap, 2011; Muthén and Muthén, 1998–2017).

Finally, Cronbach’s alpha coefficient was used to measure internal consistency for each CAAS subscale.

Results

Factorial validity and model comparison

The CFA suggests a good first-order four-factor solution, with χ2(246) = 463.515, p < 0.001; TLI 0.917; CFI 0.926; RMSEA 0.060 (0.052–0.069); and SRMR 0.049. The concern factor loadings ranged from 0.633 to 0.779. The control factor loadings ranged from 0.589 to 0.698. The curiosity factor loadings ranged from 0.633 to 0.753. The confidence factor loadings ranged from 0.731 to 0.819.

The second order model (hierarchical) shows an adequate data fit, with χ2(248) = 475.038, p < .001; TLI 0.914; CFI 0.923; RMSEA 0.061 (0.053–0.070); and SRMR 0.051. The fit indices were all adequate in accordance with the cutoff criteria. Factor loadings are reported in Table 1. As reported in Table 2, the factors showed positive and moderate correlations.

Table 1 Item descriptive statistics, item factor loading and explained variance (R2) of the second-order factor model
Full size table
Table 2 Correlation matrix among CAAS dimensions
Full size table

The first order one-factor model shows a non-satisfactory solution, χ2(252) = 728.061, p < 0.001; TLI 0.822; CFI 0.837; RMSEA 0.088 (0.081–0.096); SRMR 0.063. Factor loadings ranged from 0.524 to 0.764. The fit indices were not adequate in accordance with the cutoff criteria.

The four-factor, first-order solution was compared with different models that could better explain the CAAS factor structure (Table 3). The four-factor and the hierarchical models produced a strong fit to the data. The information criteria slightly favored the hierarchical model due to its parsimonious nature (Brown, 2015; Hoyle,2012).

Table 3 Model comparison
Full size table

Due to the model comparison revealing the superiority of the second-order (hierarchical) model, it was chosen to perform successive analyses.

Internal consistency

Reliability indices revealed an acceptable consistency of each scale, 0.862 for concern, 0.799 for control, 0.843 for curiosity, and 0.897 for confidence, as well as 0.944 for the overall dimension.

Measurement invariance

Sex

The second-order model shows an adequate fit to the data in the sample of males: χ2 (248) = 357.273, p < 0.001; TLI 0.922; CFI 0.930; RMSEA 0.060 (0.046–0.074); SRMR 0.055. Also, this model shows an adequate fit to data in the sample of females: χ2 (248) = 398.628, p < 0.001; TLI 0.889; CFI 0.900; RMSEA 0.069 (0.056–0.081); SRMR 0.070. A configural invariance model was specified across groups. Good model fit indices were found: χ2 (496) = 752.169, p < 0.001; TLI 0.907; CFI 0.917; RMSEA 0.064 (0.055–0.074); and SRMR 0.060. Also, metric invariance model still fitted the data well: χ2 (519) = 787.134, p < 0.001; TLI 0.908; CFI 0.914; RMSEA 0.064 (0.055–0.073); and SRMR 0.079. In this case, a non-significant decrease in fit indices was found, with DIFFTEST 34.986, p = 0.052, ns; |ΔRMSEA| 0.000; and |ΔCFI| 0.003, indicating that the items were equivalently related to the latent factor in these groups. Also the scalar invariance model revealed adequate fit indices: χ2 (538) = 841.462, p < 0.001; TLI 0.902; CFI 0.904; RMSEA 0.066 (0.058–0.075); SRMR 0.080. An incomplete significant decrease in fit indices compared with the metric invariance model was found: DIFFTEST 83.449, p < 0.001; |ΔRMSEA| 0.002; |ΔCFI| 0.010. This suggested that groups did not have the same expected item response at the same absolute level of the trait. Lastly, means invariance model also revealed adequate fit indices: χ2 (543) = 853.412, p < 0.001; TLI 0.901; CFI 0.902; RMSEA 0.067 (0.058–0.075); SRMR 0.082. In this case, a non-significant decrease in fit indices compared with the scalar invariance model was found, with DIFFTEST 17.812, p = 0.003; |ΔRMSEA| 0.000; and |ΔCFI| 0.002, suggesting that groups had the same latent means.

Age

The second-order model shows an adequate fit to data in the sample of young individuals: χ2 (248) = 375.220, p < 0.001; TLI 0.910; CFI 0.919; RMSEA 0.065 (0.051–0.078); SRMR 0.060. Also, this model shows an adequate fit to data in the sample of adult participants: χ2 (248) = 375.955, p < .001; TLI 0.899; CFI 0.909; RMSEA 0.065 (0.051–0.078); SRMR 0.065. Then, a configural invariance model was specified across groups. Good model fit indices were found: χ2 (496) = 751.191, p < 0.001; TLI 0.905; CFI 0.914; RMSEA 0.065 (0.055–0.074); SRMR 0.060. Also, metric invariance model still fitted data well: χ2 (519) 773.614, p < 0.001; TLI 0.910; CFI 0.915; RMSEA 0.063 (0.053–0.072); SRMR 0.071. A non-significant decrease in fit indices was found: DIFFTEST 20.571, p 0.607, ns; |ΔRMSEA| 0.002; |ΔCFI| 0.001. This indicated that the items were equivalently related to the latent factor in these groups. The scalar invariance model revealed adequate fit indices: χ2 (538) 808.726, p < 0.001; TLI 0.909; CFI 0.911; RMSEA 0.063 (0.054–0.072); SRMR 0.072. Moreover, an incomplete significant decrease in fit indices compared with the metric invariance model was found, with DIFFTEST 40.008, p = 0.003; |ΔRMSEA| 0.000; and |ΔCFI| 0.004, suggesting that groups had the same expected item response at the same absolute level of the trait. Finally, means invariance model also revealed adequate fit indices: χ2 (543) = 815.819, p < 0.001; TLI 0.909; CFI 0.911; RMSEA 0.063 (0.054–0.072); SRMR 0.075. Moreover, a non-significant decrease in fit indices compared with the scalar invariance model was found, with DIFFTEST 6.716, p = 0.243 ns; |ΔRMSEA| 0.000; |ΔCFI| 0.000, suggesting that groups had the same latent means (Table 4).

Table 4 Measurement invariance
Full size table

Discussion

Career adaptability is a key concept in CCT (Savickas, 1997, 2005, 2013, 2019, 2020). Career adaptability was conceived as a crucial meta-competency in coping with vocational development tasks, such as school-to-work transitions. Moreover, career adaptability facilitates individual adjustment to career-related changes (Savickas, 2005). In the NEET population, which refers to young people not engaged in education, employment or training programs, career adaptability could improve the successful transition to the world of work. Indeed, career adaptability was positively correlated with career success (Pan et al., 2018). Despite an intensive use in the literature of CAAS measure, no previous studies have assessed its structural validity with the NEET population. Thus, the study aimed to cover this lack by testing the psychometric properties of the Italian version of CAAS with young adults NEET.

The results of the study provided psychometric support for the factorial structure suggested by Soresi et al. (2012). The CFAs performed successfully confirmed the four-factor first-order solution and the second-order (hierarchical) model with the superiority of the second-order (hierarchical) model. Unlike the alternative, uni-dimensional structure was not supported. The information criteria favored the use of hierarchical model due to its parsimonious nature (Brown, 2015; Hoyle, 2012). Thus, the Italian version of CAAS with young adults NEET reported a second-order factor structure with four well-separated (but strongly correlated) first-order factors clearly reflecting the four dimensions of career adaptability, concern, control, curiosity, and confidence. This CFA model revealed a good fit to the data and good reliability for each scale, and its overall dimension provides satisfying results. Research suggests that hierarchical models provide a powerful mechanism for parsing shared and unique components of variance (Markon, 2019). Moreover, it leads to greater conceptual clarity rather than other CFAs model structures (Rossi et al., 2022a, 2022b). The hierarchical model helps researchers to understand the domain subscores (i.e., career concern, control, curiosity, and confidence) and the overarching general factor (i.e., career adaptability). The latter has theoretical and practical implications. Related to the theoretical implication, this evidence confirms that the CAAS dimensions, concern, control, curiosity, and confidence are correlated, but are still conceptually different. As recommended by Savickas and Porfeli (2012), career adaptability resources are multidimensional and hierarchical. The dimensions of the first-order level, namely concern, control, curiosity, and confidence, combine to become an aggregate construct. From a methodological point of view, each dimension has a unique effect and together models the global indicator. Related to practical implications, researchers may also want to consider measuring the effect of each dimension on psychological outcomes. Moreover, career practitioners might plan their career interventions with young NEET improving the lacking resources by focusing interventions even more effectively.

This study confirmed the use of the 24-CAAS measure for young NEET. Overall, based on the definition of career adaptability, the CAAS measure could be useful for assessing these skills, also with youth in critical conditions, as well as in young people not engaged in education, employment, or training paths. Indeed, these results suggest possibilities for future use of CAAS for both researchers and career practitioners for working with NEET in community settings and/or career centers. In addition, invariance tests proposed by the current study, i.e., configural, metric, and scalar invariances, showed acceptable fit to the data collected from different age as well as groups of the sexes.

This study, aside from testing the CAAS for the specific group of NEETs, also covered a lack of career adaptability, argued by a recent systematic review (Johnston, 2018). According to Johnston (2018), researchers should consider measuring career adaptability resources as a latent variable (Hirschi & Valero, 2015), rather than as a total score. The findings support the adequacy of using career adaptability as a latent variable rather than as a total score. From a psychometrical point of view, research recommends latent variables because they allow researchers to adjust for measurement errors in the measured variables (Kline, 2016). The use of the latent variable approach is suggested because in the mediation model, for example, this approach provides better estimates on average across all levels of effect size and reliability. According to Ledgerwood and Shrout (2011), direct and indirect effects were more likely to be better estimated when latent, rather than observed, variables were used. Thus, the use of career adaptability as a latent variable might be the best way to shed light on the true value of the relations between career constructs.

However, there are some limitations with the current study opening a potentially important avenue for future investigation. Firstly, the results need to be replicated in other Italian geographical areas and in areas with different neighborhood characteristics to determine their generalizability. Secondly, further psychometric properties of the measure remain to be examined, such as the test–retest reliability of the measure, providing evidence of its stability over time. Thirdly, its predictive validity on specific measures should be investigated, such as career self-efficacy, career decision-making, and career calling. Finally, it has been considered that the data may have been influenced by a reporting bias due to the selection process (response from people with a certain proactivity and interest in returning their situation).

Despite these limitations, this study extends the previous research of the use of CAAS among different population groups and provides additional support for its application. Previous studies have already underlined that the CAAS measure maintains its psychometric properties, also with vulnerable populations, e.g., intellectual disability (Ginevra et al., 2018), mental disorder (Xu et al., 2020), and substance use disorder (Di Maggio et al., 2021). This study justifies the use of CAAS also for the NEET population.

The good psychometric proprieties of the CAAS measure provide important implications for career counseling as well, because this instrument might help young NEET to regulate goal-pursuing processes (Savickas, 1997) in work transitions. Career counselors could orient their interventions starting from the items scales. In this view, CAAS could be used as an effective tool for self-reflection and improvement (Yu et al., 2019) and to increase the possibility of lifelong employment of at-risk groups (Van der Heijden and De Vos, 2015).