Over time, the concept of giftedness has been adopted not only to identify some exceptional performance or impressive achievements, but also to describe in detail certain behavioral characteristics that contribute to this kind of performance or achievements (Westberg, 2012). Early scholars conceptualized giftedness broadly, using terms like “gifted,” “genius,” and “talented” interchangeably. Galton (1869) viewed genius as innate and hereditary, laying a foundation for future inquiry (Kaufman and Sternberg, 2018). Spearman (1904) introduced general intelligence (g) as an inborn trait identified through factor analysis, while Binet and Simon (1916) developed assessments for identifying students needing support - later adapted by Terman (1916) into the Stanford-Binet scale. Although not explicitly focused on giftedness, hierarchical theories of intelligence advanced the understanding of giftedness by acknowledging the layered nature of intellectual abilities.

As the field progressed, dissatisfaction with equating giftedness solely with intelligence led to a more multidimensional approach (Smedsrud, 2020). The shift from genetic determinism toward more dynamic perspectives led to developmental theories which regard giftedness as evolving expertise shaped by both innate dispositions and environmental factors. Renzulli’s (1978, 2005) Three-Ring Conception emphasized above-average ability, creativity, and task commitment, advocating for the identification of gifted behaviors rather than individuals. He further differentiated between “schoolhouse” and “creative-productive” giftedness (Kaufman and Sternberg, 2018). Tannenbaum (1986) highlighted the interaction of abilities, psychosocial skills, external support, and chance, while Mönks (1992) expanded on this by incorporating family, school, and peer influences. Sternberg’s (2003, 2005) Wisdom, Intelligence and Creativity, Synthesized model (WICS) conceptualized giftedness as a balance of wisdom, intelligence, and creativity aimed at personal and societal goals. Similarly, Feldman (1992, 2000) and Feldhusen (1986, 1998) proposed multi-dimensional developmental frameworks, incorporating cognitive, emotional, social, and cultural factors. Gagné’s (2004) Differentiated Model of Giftedness and Talent (DMGT) distinguished innate gifts from developed talents, emphasizing the mediating role of personal and environmental factors, learning, and practice. In a like manner, Heller’s Munich Model of Giftedness (MMG) and the Munich Dynamic Ability-Achievement Model (MDAAM) (Heller et al., 2005) recognized the role of creativity and other traits. These models stress the interaction between talent, personality, and environment, influencing domain-specific performance. The Actiotope Model (Ziegler, 2005) extended this systemic approach, conceptualizing giftedness as adaptive behavior within dynamic interactions between individuals and their environments. More recent frameworks adopted a holistic approach. Kaufman’s (2013) theory of personal intelligence integrates motivational, emotional, and cognitive elements in the pursuit of personal growth, while Pfeiffer (2015) proposed a flexible, three-part model encompassing psychometric, developmental, and ecological perspectives. Subotnik et al.’s (2011, 2018) Megamodel synthesized this body of research, framing giftedness as early potential that, with sustained effort and support, can develop into expertise and, ultimately, eminence. Recently, Tordjman (2020) proposed a developmental framework emphasizing the influence of environmental factors on how a child’s high potential emerges and unfolds. This model examines developmental processes shaped by specific environmental elements - such as encouraging motivation, offering enrichment opportunities, recognizing effort, and providing support - across the child’s family, educational/professional, and broader societal contexts. These different environments interact dynamically, consistent with Bronfenbrenner’s ecological systems theory, which explains how a child’s development is influenced by multiple layers of environment and their interconnections (Bronfenbrenner, 1979; Bronfenbrenner and Evans, 2000; Crawford et al., 2019). By examining various developmental paths that can result in similar high-potential outcomes, the above-mentioned model presents diverse pathways for nurturing and expressing children’s high abilities (Tordjman et al., 2021).

In conclusion, giftedness - once viewed as an inherent and stable trait - is now increasingly understood as a dynamic construct shaped by environmental influences and developmental context (Wiley, 2020). Contemporary perspectives have moved beyond static, IQ-based definitions, favoring multidimensional models that conceptualize giftedness as a collection of distinct constructs. These include not only cognitive abilities but also motivation, task commitment, creativity, social competence, and sustained effort. Such characteristics are now recognized as equally essential in the development and expression of giftedness (Renzulli, 2016; Sternberg, 2015; Subotnik et al., 2011, 2018; Smedsrud, 2020).

Naturally, the evolving definition of giftedness has shaped the assessment criteria, enabling a more accurate and efficient identification of gifted students. Identification procedures have evolved from traditional approaches focused solely on IQ and achievement tests to contemporary approaches involving a wide variety of instruments and strategies such as creativity, motivation (Renzulli, 2011), teacher, parent self-ratings and nominations and portfolios used together or in combination (Acar et al., 2016; Dai, 2018; Smedsrud, 2020). They can be categorized into two main types: performance-based and non-performance-based methods. Performance-based methods often rely on intelligence testing and achievement scores (Cao et al., 2017; Renzulli et al., 2010). The non-performance methods represent all other gifted identification approaches, not in-volving any performance-based assessment. Examples of forms of assessment in this category are teacher and parent rating scales, self-ratings and peer-ratings (Hodges et al., 2018). With respect to teacher rating scales, these instruments can be used to gather structured information from teachers on all areas of giftedness. Moreover, they can be used to assess a wide range of specific characteristics, related to giftedness, that are difficult to assess with other methods and they are not captured by cognitive ability tests (Benson and Kranzler, 2017). Well-constructed rating scales help transform a teacher’s subjective impressions into more objective evaluations by linking them to specific behaviors. The objectivity of judgments on complex performances improves with the quality of the scoring or rating rubric used and the expertise of the rater, who has been thoroughly trained to identify key qualities (Renzulli et al., 2010). Therefore, using teacher rating scales is considered valuable, as these scales can offer diverse insights into students’ characteristics and behavioral patterns associated with giftedness (Makel et al., 2015). Indicative scales for teachers that have been used to assess giftedness in students include the Behavioral Characteristics of Superior Students (SRBCSS; Renzulli et al., 2010). According to Westberg (2012), this is the most commonly used scale by teachers in the U. S to identify gifted students. Other evaluation approaches for teachers that have been used to assess giftedness in students include the Gifted Rating Scales (GRS; Pfeiffer and Jarosewich, 2003), the Gifted and Talented Evaluation Scale (GATES; Gilliam et al., 1996), the Scales for Identifying Gifted Students (SIGS; Ryser and MacConnell, 2004) and the HOPE Teacher Rating Scale (the HOPE Scale; Gentry et al., 2015).

Parents’ nomination allows the evaluation of children’s characteristics, behaviors, interests, and attitudes in different circumstances including extra-scholastic situations (Milic and Simeunovic, 2022). The Gifted Rating Scales (GRS; Pfeiffer and Jarosewich, 2003) can be utilized for parent nominations, while the Gifted and Talented Evaluation Scale (GATES; Gilliam et al., 1996) may be completed by parents or other individuals familiar with the student. Similarly, the Scales for Identifying Gifted Students (SIGS; Ryser and McConnell, 2004) include two complementary rating forms - a Home Rating Scale and a School Rating Scale - which can be administered together or independently.

In the context of self-assessment, students evaluate their characteristics relative to specific domains such as intellectual and learning abilities, motivation, and creativity (Tirri and Nokelainen, 2007). Several validated instruments have been developed for self-nomination purposes. These include the Perceived Competence Scale for Children (Harter, 1982), the Chinese Student Multiple Intelligences Profile, a self-report measure that focuses on gifted students’ activities or preferences that reflect their self- perceived multiple abilities or intelligences (SMIP-24; Chan, 2001), the Revised Screening Scales for the Evaluation of Multiple Intelligences (Hernández-Torrano et al., 2014), the Children’s Perceived Competence Scale, designed to measure children’s self-perceived competence across three domains of Cognitive (C), Social (S), Physical (P) and the fourth domain of General self-worth (G) (CPCS; Nagai et al., 2014), and the Sumida Checklist on Giftedness in Science, used specifically with regard to science giftedness, validated in Japan (Sumida, 2010) and in the Philippines (Larroder and Ogawa, 2015). Finally, the Self Nomination Inventory for Gifted and Talented Adolescents in a Mexican adolescent sample has been validated by Zavala Berbena and de la Torre García (2021).

Peer evaluations can also serve as a valuable component in the assessment of gifted students, particularly when used alongside other methods. This is because (a) peer ratings may be less influenced by social desirability bias, (b) peers often have unique insights into their classmates’ intellectual strengths and behaviors due to daily interactions, and (c) collecting a large volume of peer ratings is often feasible (Kaya, 2013; Paunonen and O’Neill, 2010). However, there is a limited body of research focusing on peer ratings in this context (Marsili and Pellegrini, 2022). Concerns regarding peer nomination tools often center around their limited evidence for reliability and validity, with some critics pointing out the absence of even basic psychometric data (Gagné et al., 1993). One notable exception is a peer referral form developed by Udall (1987), aimed at identifying gifted minority students in Grades 4 to 6. Cunningham et al. (1998) assessed this tool’s reliability and construct validity and concluded that it is generally reliable for identifying gifted Hispanic students, though they also offered recommendations for its effective application. More recently, Kaya and Delen (2014) explored the usefulness of peer assessments in identifying giftedness by creating a computer-based, psychometrically sound peer rating tool known as the Guess Who: Peer Nomination Form (GWPNF). Designed specifically for younger elementary students, the GWPNF features a game-like format and uses simple, age-appropriate language.

Researchers in the field suggest assessing giftedness in stages that include an initial detection stage followed by a thorough assessment of students (Almeida et al., 2016). As such, teacher rating scales, which have been found to be the most commonly employed method for identifying gifted students (Marsili and Pellegrini, 2022), are often used as brief assessment tools, followed by individually administered standardized tests for a more accurate identification (Renzulli and Gaesser, 2015).

One of the major challenges in gifted education in Greece is the accurate identification of gifted and highly talented students through teachers’ and/or parents’ rating scales as part of a screening test. The National Educational Curriculum lacks any reference to giftedness, making it difficult to tailor learning experiences to students with gifted traits. Additionally, the Greek educational system does not have a formal process for identifying gifted children, as no guidelines exist within the curriculum or syllabus documents. Furthermore, the absence of official gifted identification programs and structured enrichment initiatives highlights the urgent need for effective evaluation of gifted learners. Given that SRBCSS scales were initially designed in a different educational and cultural context, it is unavoidable to question their relevance and suitability in the Greek context. Moreover, to our knowledge, no Greek published research has tested in Greek population some of the psychometric properties – factor structure, internal consistency reliability, convergent validity – of the SRBCSS. Since, according to Westberg (2012), the first four scales of the SRBCSS – Learning, Motivation, Creativity, and Leadership – have been among the most widely used teacher judgment rating scales for gifted programming in the U. S., and have also been translated and studied in several countries, and given that the authors of the scales did not conduct a confirmatory factor analysis to examine the relationships between the judgmentally developed categories and empirically derived constructs (as noted in the SRBCSS Technical and Administration Manual; Renzulli et al., 2010), the aims of this study were: (a1) the test/confirmation (at item-level data) of the unifactorial structure for each of the first four scales – Learning, Creativity, Motivation, and Leadership – of the Greek version of SRBCSS, as well as (a2) the test of their content validity, by examining the four scales’ inter-correlations. Considering the findings of Srour (1989) and Caroff et al. (2006), who identified one-factor and two-factor solutions, respectively, to account for most of the variance captured by the first four scales of the SRBCSS (as measured variables at scale-level), another aim of this study (a3) was to examine these two alternative possibilities in a Greek sample. Moreover, another aim of this study was (b) the evaluation of the internal consistency reliability of each of the four scales of the Greek version of the SRBCSS, as well as (c) the test of the convergent validity of the Greek version of the first four SRBCSS scales with the Greek version of the six Gifted Rating Scales – School form (GRS-S), which, according to Pfeiffer and Jarosewich (2003), have been found to have high correlations with ratings on Creativity, Motivation and Leadership SRBCSS scales in the GRS-S authors’ American sample.

A total of 145 Greek primary and secondary education school teachers participated in the study. Specifically, 101 (69.7%) were women, and 42 (29%) were men, while 2 (1.3%) did not declare their gender. Regarding the length of time, they had known the student, whose behavioral and learning characteristics they were going to evaluate, 21 (14.5%) reported they knew the student from 1 to 3 months, 40 (27.6%) from 4 to 6 months, 7 (4.8%) from 7 to 12 months and 75 (51.7%) more than a year. Concerning how well they believed they knew the student, 81 (55.9%) teachers stated they feel they know the child fairly well, 46 teachers stated (31.7%) they feel they know the child very well, while only 16 teachers (11%) stated that they do not feel they know him/her very well. A total of 1.4% (N = 2) did not respond to the last two relevant questions. An equal number (N = 145) of students, randomly selected from the student population, were rated by their teachers. More precisely, 83 (57.2%) were girls and 62 (42.8%) were boys. The composition of the students’ sample in terms of grade level was as follows: 40 students were from the 5th grade of Greek primary education school (5th grade of American elementary school), 56 from the 6th grade of Greek primary education school (1st grade of American middle school), 24 were from the 1st grade of Greek secondary education school (2nd grade of American middle school), and 25 from the 2nd grade of Greek secondary education school (3rd grade of American middle school).

Each teacher evaluated one of his/her students by completing the translated Greek version of the first four SRBCSS and the GRS-S. The parents of the students were also provided with a Greek-translated version of the scales, along with a consent form, to ensure they were adequately informed about the measurements used in the study, without participating in the evaluation process.

In specific, each teacher and parent received the following from the researcher: (a) an information letter explaining the research objectives and a consent form, (b) a demographic data form, (c) the translated first four scales – Learning, Creativity, Motivation, and Leadership – of the Greek version of the SRBCSS, and (d) the translated six scales of the Greek version of the GRS-S. All participants completed the assessment individually. Teachers had the flexibility to choose both the location and timing for filling out the questionnaire. If the completion occurred in the presence of the researcher, assistance and clarifications were provided when necessary. Teachers were encouraged to respond truthfully to ensure the reliability of the results. The sample was convenient, and the participants were recruited from schools in various regions of Greece, with data collected randomly by students from the Department of Early Childhood Education at the University of Ioannina, as part of a voluntary task during an Introduction to Psychology course, under the supervision of one of the study’s authors. Since the data collected were considered personal, the European Union law on personal data protection, effective since May 28, 2018, was strictly followed. According to this regulation, the use of sensitive personal data is permitted exclusively for research purposes. The study protocol adhered to the ethical principles outlined in the Helsinki Declaration and received approval from the Scientific and Ethics Committee of the University of Ioannina (25,847/01/06/2021).

In line with Horn’s (1965) critique (as cited in Courtney, 2013) that the Kaiser criterion (K1 rule) is not appropriate for sample-based research due to sampling error in the computation of latent roots, a parallel analysis was carried out in SPSS 21.0 with R-menu installed (Basto and Pereira, 2012) to assess the unifactorial structure of each of the four scales of the SRBCSS.

In order to verify the one-factor structure, identified through the EFA application, for each of the four SRBCSS scales, a set of four CFA was conducted. These analyses were applied to the data corresponding to each of the four SRBCSS scales – Learning (11 items), Creativity (9 items), Motivation (11 items), and Leadership (7 items) – using the Maximum Likelihood estimation method. Each CFA was carried out in EQS 6.1 (Bentler, 2005), based on the covariance matrix of items measured on a six-point scale. Model fit was primarily assessed using the chi-square (χ²) test, where a non-significant result indicates that the hypothesized model adequately reproduces the observed variance–covariance structure (Brown, 2015; Kline, 2003). The χ²/df ratio – which was introduced by Wheaton et al. (1977) in attempt to use χ² in a manner that would foster more realistic model evaluation – was also calculated. Given the test’s sensitivity to sample size, additional fit indices were employed. The Root Mean Square Error of Approximation (RMSEA) – which, according to Fan and Sivo (2007), stood out as having ideal behavior patterns expected of good model fit index – was used to evaluate how well the model would fit the population covariance matrix. According to established guidelines, RMSEA values ≤ 0.05 indicate a close fit, values between 0.05 and 0.08 suggest a reasonable fit, and values ≥ 0.10 indicate poor fit (Brown and Cudeck, 1993; Kline, 2003). Furthermore, the Comparative Fit Index (CFI), a widely used incremental fit index, was utilized to assess the improvement in model fit compared to a baseline model. CFI values between 0.95 and 1.00 reflect a good fit, while values between 0.90 and 0.95 represent an acceptable fit (Bentler, 1990; Schweizer, 2010). Lastly, the Standardized Root Mean Square Residual (SRMR), which reflects the average discrepancy between observed and predicted correlations, was also examined. An SRMR value below 0.08 is considered indicative of a good model fit (Bentler, 2005; Brown, 2015; Hu and Bentler, 1999). However, it is worth noting at this point, that we should be cautious, as according to Fan and Sivo (2007), for most fit indices including CFI and SRMR, it would be difficult to establish cut-off criteria that would be generally useful in SEM applications. Additionally, although the χ²/df ratio has become very popular in applied research, according to Brown (2006) its use is strongly discouraged.

The fit indices of the initial CFA models’ set, that was conducted to test the one-factor structure of each of the four SRBCSS scales, ranged from not accepted to marginally accepted:

Learning Scale, χ² (44, N = 144) = 118.06, p < 0.001, χ²/df = 2.683, CFI = 0.93, SRMR = 0.05, RMSEA = 0.11 (CI90% 0.08–0.13);

Creativity Scale, χ² (27, N = 144) = 92.45, p < 0.001, χ²/df = 3.424, CFI = 0.91, SRMR = 0.06, RMSEA = 0.13 (CI90% 0.10–0.16);

Motivation Scale, χ² (44, N = 143) = 101.19, p < 0.001, χ²/df = 2.999, CFI = 0.92, SRMR = 0.06, RMSEA = 0.10 (CI90% 0.07–0.12);

Leadership Scale, χ² (14, N = 145) = 27.45, p = 0.017, χ²/df = 1.960, CFI = 0.97, SRMR = 0.04, RMSEA = 0.08 (CI90% 0.03–0.13).

All parameters of the aforementioned models were found to be statistically significant (p < 0.05). Standardized Root Mean Squared Residual (SRMR) values were below 0.08, also indicating a good fit for the models tested. Similarly, the χ²/df ratio of one model fell to the range of 0.00–2.00 and was indicative of a marginally good model fit, the χ²/df ratios of two models fell to the range of 2.00–3.00 and were indicative of a marginally accepted model fit, while the fourth model’s χ²/df ratio fell to the range of χ²/df > 3.00 indicating a poor model fit (Schermelleh-Engel et al., 2003). Furthermore, the chi square goodness-of-fit test was statistically significant for all the initial models resulting in a rejection of the null hypothesis of good fit. In addition, the Comparative Fit Index (CFI) values of the three models fell to the lowest boundary of the marginal range of 0.90–0.95 and were indicative of a marginally accepted model fit (Bentler, 2005; Brown, 2015; Hu and Bentler, 1999). Finally, the Root Mean Squared Error of Approximation (RMSEA) values were above 0.08, indicating a poor fit for the models tested (Brown, 2015; Brown and Cudeck, 1993).

For these reasons, we proceeded with the identification of the areas of the initial models that contributed the most to the misfit. To account for possible correlations between variables, we extend the baseline model by incorporating covariances. A residual analysis was conducted, and the Wald test was performed. Different models were tested and the modifications indicated by the aforementioned tests were included in the model being tested each time. The modifications were correlated errors of measured variables (items) that were manifested as large standardized residuals. The modifications are presented after the indices of each of the four final models, and were covariances (correlations) between errors of indicators (items), which were mainly similarly worded (Bandalos, 2021) or differentially prone to social desirability (Brown, 2006). The modifications improved the fit of the final models on all indices:

Learning Scale, χ² (41, N = 144) = 82.73, p = 0.00012, χ²/df = 2.017, CFI = 0.96, SRMR = 0.04, RMSEA = 0.08 (CI90% 0.06–0.11); Covariances (correlations) among errors of measured variables (items) of the Greek version of Learning Scale (standardized solution): E5 – E4 = 0.406, Ε9 – Ε8 = 0.212, Ε11 – Ε10 = 0.245; All correlations presented indicate statistically significant associations (p < 0.05) among errors of measured variables and their z values were 4.026, 2.183, and 2.545, respectively.

Creativity Scale, χ² (23, N = 144) = 43.79, p = 0.00558, χ²/df = 1.903, CFI = 0.97, SRMR = 0.04, RMSEA = 0.08 (CI90% 0.04–0.11); Covariances (correlations) among errors of measured variables (items) of the Greek version of Creativity Scale (standardized solution): Ε6 – Ε1 = −0.483, Ε8 – Ε1 = −0.312, Ε6 – Ε5 = −0.392, Ε9 – Ε8 = 0.192; All correlations presented indicate statistically significant associations (p < 0.05) among errors of measured variables and their z values were −4.706, −3.438, −3.696, and 2.102, respectively.

Motivation Scale, χ² (42, N = 143) = 77.68, p = 0.00067, χ²/df = 1.849, CFI = 0.95, SRMR = 0.05, RMSEA = 0.08 (CI90% 0.05–0.10); Covariances (correlations) among errors of measured variables (items) of the Greek version of Motivation Scale (standardized solution): Ε3 – Ε2 = 0.259, Ε11 – Ε2 = 0.306; All correlations presented indicate statistically significant associations (p < 0.05) among errors of measured variables and their z values were 2.918, and 3.561, respectively.

Leadership Scale, χ² (13, N = 145) = 22.65, p = 0.04608, χ²/df = 1.742, CFI = 0.98, SRMR = 0.04, RMSEA = 0.07 (CI90% 0.01–0.12); Covariance (correlation) between errors of measured variables (items) of the Greek version of Leadership Scale (standardized solution): Ε6 – Ε4 = −0.229; The correlation presented indicates statistically significant association (p < 0.05) between errors of measured variables and its z value was −2.259.

In the second set of the CFA, all parameters of the final models were also found to be statistically significant (p < 0.05). The Standardized Root Mean Squared Residual (SRMR) values were below 0.08, indicating a good fit for the models tested. Additionally, the Comparative Fit Index (CFI) values of the four models exceeded 0.95, indicating a reasonably strong fit across the models. Furthermore, the chi square goodness-of-fit test was not statistically significant for almost all of the final models resulting in the acceptance of the null hypothesis of good fit. Similarly, the χ²/df ratio of three of the final models fell to the range of 0.00–2.00 and was indicative of a good model fit, while the χ²/df ratio of the fourth model fell to the range of 2.00–3.00 and was indicative of a marginally accepted model fit (Schermelleh-Engel et al., 2003). The marginally accepted Root Mean Squared Error of Approximation (RMSEA) values (between the reasonable fit of 0.05 and 0.08), that were found in the four final models, could be interpreted in accordance with the acceptable model fit. Since N was somewhat small, they may be of less concern if all other indices were in a range suggesting a “good” model fit. It is important to mention that the RMSEA is known to produce artificially large values for models with few degrees of freedom and small sample sizes (Bentler, 2005; Brown, 2015; Hu and Bentler, 1999).

It should be noted that the aforementioned second set models of CFA at the item-level data –although they verified the proposed by Renzulli et al. (2010) unifactorial structure for each of the first four scales of the Greek version of the SRBCSS – were limited (a) as regards the approach to correlating residuals, that was used, as this approach tends to violate assumptions of regressions and inflate the models’ fit, and (b) as regards the verification (at item-level data) of the SRBCSS organization in four underlying factors/latent variables (scales) and the test of the four factors (scales) correlations, as these correlations depict in confirmatory factor analysis (CFA) (at item-level data) structural model. Therefore, we conducted a set of confirmatory factor analyses, using all items from each of the four 6-point SRBCSS scales (38 items, total), trying to verify (at item-level data) the SRBCSS organization in the proposed – by their authors and Rogalla (2003) as well – four-factor structure. One path from every one of the 4 pertinent factors (scales) to each of every item that constitute it, was freed for each confirmatory factor analysis (CFA) model. Cross-loadings were not permitted. CFA was performed twice: At the first performance, latent factors were defined without any covariances between them (Measurement model). At the second performance, all the latent factors were allowed to freely intercorrelate (Structural model). For both models, the metric was set by fixing factor variances to 1.0.

Although, all parameters of both the measurement and structural model – that were conducted to test (at the item-level data) the SRBCSS organization in four underlying factors/latent variables (scales), as well as the four factors’ correlations – were found to be statistically significant (p < 0.05), the indices of the measurement model – except for the χ²/df ratio, which was accepted – were not accepted: Measurement model, χ² (665, N = 141) = 1700.41, p < 0.001, χ²/df = 2.557, CFI = 0.73, SRMR = 0.37, RMSEA = 0.10 (CI90% 0.10–0.11).

On the contrary, the structural model was found to fit the data better: Structural model, χ² (659, N = 141) = 1296.35, p < 0.001, χ²/df = 1.967, CFI = 0.83, SRMR = 0.07, RMSEA = 0.08 (CI90% 0.08–0.09). All parameters of the structural model were found to be statistically significant (p < 0.05) and the standardized root-mean-square residual (SRMR) value was equal to 0.07 indicating a good fit for the model tested. Similarly, the χ²/df ratio of the model fell to the range of 0.00–2.00 and was indicative of a good fit (Schermelleh-Engel et al., 2003). In addition, the root mean squared error of approximation (RMSEA) value was equal to 0.08 indicating also a marginally reasonable fit for the structural model. However, the comparative fit index (CFI) value – although it was equal to 0.83 and close enough to Rogalla’s (2003) CFI = 0.89 – indicated a non-accepted model fit for the structural model (Bentler, 2005; Brown, 2015; Hu and Bentler, 1999; Kline, 2003), through which we attempted to verify the SRBCSS organization in the proposed by Renzulli et al. (2010) and Rogalla (2003) four-factor structure.

It should be highlighted that this third set of CFAs at the total 38 item-level data was also limited as regards the identification of a second-order underlying factors’ number in which the SRBCSS four first-order factors’ (scales’) could be organized. The aforementioned findings could be explained regarding the sample size, which was somewhat small to permit the confirmation of a CFA structural model, at item level, for the total 38 items of the four scales.

Therefore, given that the second set of CFAs, that were conducted at the item-level, fully confirmed the unifactorial structure proposed by Renzulli et al. (2002) separately for each of the first four scales of the Greek version of the SRBCSS, and considering (a) Srour’s (1989) findings, where most SRBCSS items (29 out of 38) loaded onto a single factor, and (b) Caroff et al.’s (2006) results, which suggested a two-factor structure, rather than the originally proposed four-factor structure, we applied confirmatory factor analyses (CFA), at the scale-level data, to all four scales in order for us to be able to determine the number of the general latent variables (factors) underlying the organization of the first four scales of the Greek version of the SRBCSS. Because these analyses were based on scale-level data, rather than on item-level data, the total scores for the Learning, Creativity, Motivation, and Leadership scales – each previously validated as unifactorial – were treated as observed variables in the CFAs conduction.

At first, a CFA was conducted in order to confirm the aforementioned model with a single general factor (Model A). Results revealed that the unifactorial model provided an excellent fit for the first four scales of the Greek version of SRBCSS data [χ² (2, N = 141) = 2.79, p = 0.248, χ²/df = 1.396, CFI = 1.00, SRMR = 0.01, RMSEA = 0.05 (CI90% 0.00–0.18)]. R² values for the single general factor ranged from 0.54 to 0.83. Standardized loadings and residual variances for observed variables are presented in Figure 1.

In addition, we conducted a second CFA to test the alternative correlated two-factor model solution (Model B) proposed by Caroff et al. (2006), whose EFA results suggested that all items from the Learning and Creativity scales loaded onto one factor (F1), while a second factor (F2) was shared by items from the Motivation and Leadership scales. The indices of the correlated two-factor model were as follows: χ² (1, N = 141) = 0.91, p = 0.340, χ²/df = 0.911, CFI = 1.00, SRMR = 0.01, RMSEA = 0.00 (CI90% 0.00–0.22), indicating also an excellent fit for the correlated two-factor model solution. Correlation between F1 and F2 was 0.96. R² values for the correlated two-factor model ranged from 0.54 to 0.87. Standardized loadings and residual variances for the observed variables are presented in Figure 2.

As it was found that both A and B models presented satisfactory fit indices, we proceeded to compare the difference χ²(Δχ²) between the two models. The difference in χ² between Model A and Model B Δχ² (Δdf = 1) = 1.88 was not found to be statistically significant at the p < 0.05 level. Therefore, the two models seem to be considered as statistically equivalent.

At first, Cronbach’s alpha coefficients were used to assess the internal consistency reliability of the SRBCSS scales of Learning, Creativity, Motivation, and Leadership. The Cronbach’s α internal consistency coefficients of the four scales of the Greek version of the SRBCSS, for the corresponding sample of the present study, were good to excellent as they ranged between 0.883 and 0.942.

Except for (a) Cronbach’s alpha, we also took into account the following reliability coefficients that their calculation is available by EQS 6.1.: (b) Reliability Coefficient RHO; (c) Greatest Lower-Bound Reliability (GLB Reliability); (d) Bentler’s Dimension-Free Lower-Bound Reliability; (e) Shapiro’s Lower-Bound Reliability and (f) Hancock’s H coefficient. The specifics are as follows:

In Table 1, the inter-correlations among the four scales of the SRBCSS are presented, as calculated at the aforementioned structural model that was conducted to test (at the item-level data) the SRBCSS organization in four underlying factors/latent variables (scales), as well as the four factors’ correlations. As presented in Table 1, all scale correlations were positive and statistically significant at the p = 0.05 level, indicating a strong interrelationship among the scales. The inter-correlations ranged from 0.70, observed between the Creativity and Leadership Scale, to 0.88 between the Learning and Motivation Scale (Table 1).

Additionally, the inter-correlations among the four scales of the SRBCSS were calculated using the Pearson correlation coefficient (r). As presented in Table 2, all scale correlations were positive and statistically significant at the p = 0.01 level, indicating a strong interrelationship among the scales. The inter-correlations ranged from 0.61, observed between the Creativity and Motivation Scale as well as the Creativity and Leadership Scale, to 0.80 between the Learning and Motivation Scale (Table 2).

Finally, in order to test the convergent validity of the first four scales – Learning, Creativity, Motivation, and Leadership – of the Greek version of SRBCSS with the six scales of the Greek version of the GRS-School form (Pfeiffer and Jarosewich, 2003), the Pearson correlation coefficient (r) was conducted. The application of the Pearson correlation analysis revealed statistically significant mainly moderate correlations between the ratings on SRBCSS and GRS-S scales. Table 3 depicts the correlations between the four scales of the Greek version of the SRBCSS with the six GRS-S scales.