Initially, descriptive statistics of the EAGrIt-LP items are presented. The minimum and maximum values, in addition to the means and standard deviation, are shown in Table 1. Furthermore, the asymmetry and kurtosis coefficients of the distribution of results in each item are presented.

Table 1 shows a positive trend in the students’ responses, as the averages are close to the maximum value of the assessment scale used (4 points or totally in agreement), translating positive self-assessments by the sample. The item with the lowest average was ‘Whenever I think about my goals, I feel that they are increasingly consolidated’, with item 9 “I try to commit to achieve my goals”, a higher average. As for the other items, they presented averages close to each other. On the other hand, in terms of response variability, we can point out that the values were between 0.65 and 0.87 (standard deviations), being considered acceptable values because there were only four response levels in the Likert-type scale used in the responses. On the other hand, the asymmetry and kurtosis coefficients are very low, which suggests normality of the results distribution in each item.

Subsequently, a parallel analysis was carried out to determine the number of factors to be retained. The Hull method and Parallel Analysis indicated retention of one factor (Lorenzo-Seva et al., 2011). To assess whether the data matrix was factorable, KMO (0.935) and Bartlett’s sphericity test (χ²= 4034.1, p < 0.001) were used, both coefficients being adequate. Although there is no consensus in the literature on the importance of the sedimentation graph, we chose to use it in order to help define the number of factors to be retained in our analyses (Figure 1).

Figure 1 allows us to understand that one or two factors can be considered in the analyses regarding the dimensionality of the scale. Considering that the construction of the scale indicated the presence of two factors, Consistency of Interests and Perseverance of Effort, it was then decided to organize the data into two factors. Together they explain 68% of the variance in the results of the 12 items. The distribution of the factor loadings of the items in the two factors (oblimin rotation) is shown in Table 2.

Regarding distribution of the items among the factors, it is worth noting that one of them did not load the theoretically predicted factor (I try to commit myself to achieving my goals). On the other hand, the saturation indices were higher than 0.40, even though in two items (item 1 and item 7) the variance indices explained by the set of two factors (communality) have moved away from 0.50 (respectively 0, 39 and 0.29).

Then, the Confirmatory Factor Analysis was carried out, with the purpose of analyzing the adequacy of the theoretical model used as a subsidy for the construction of the items. The results pointed to an adequate fit (χ²/gl = 1.83, GFI = 0.999, CFI = 0.994, RMSEA = 0.035 [0.024-0.046]. To assess precision, in the consistency of interest’s dimension, α = 0.85 was found, and in the effort perseverance dimension α = 0.86. Figure 2 presents the factor loadings of the factors, with no modification having been introduced, for example correlation between errors, to improve the fit indices. Finally, a high correlation coefficient is recorded between the two dimensions of the scale (0.84), which may suggest future developments in the dimensional analysis by bringing together the 12 items of the scale.

The factor loadings ranged from 0.46 (“I try to commit to achieving my goals” and “I persevere in my efforts to achieve my goals”) to 0.63 (“Whenever I think about my goals, I feel that they are increasingly consolidated”). In this way, the two-dimensional structure recommended for the instrument was confirmed, which are strongly correlated. The first factor was named effort perseverance, and it is composed of items 2, 6, 8, 10, 11 and 12. The second factor was called consistency of interests and the items that compose it are: 1, 3, 4, 5, 7 and 9. Finally, Table 3 presents the results of the scale two-dimensional model of the analysis of invariance of the two countries.

Regarding the Configuration model which assesses whether the dimensional structure of the instrument applies to both groups, the results indicated that there was invariance. Thus, the internal structure of the two factors, namely consistency of interests and perseverance of efforts, does not differ between Brazilian and Portuguese students. Subsequently, the Metric model was tested, which in turn analyses the equivalence of the items’ regression weights, which are equivalent to their factor loadings. The findings indicated invariance, which means that the relationships presented by the items regarding the construct are similar for both groups (Xia & Yang, 2019). Continuing with the comparison analysis between the groups (Brazilian and Portuguese students), the scalar invariance analysis was carried out. This analysis assesses whether the scores obtained are associated with the latent trait level of the study participants, regardless of their group. The results confirmed the invariance. Finally, the Thresholds and Residual were tested, and in both cases the measurement invariance was confirmed.

This study continues the investigation on the construction and validation of the Garra Assessment Scale: International Version in Portuguese Language (Noronha & Almeida, 2022), with Brazilian and Portuguese higher education students. In a previous study, the authors reported the process of construction and selection of items, essentially following the qualitative procedures involved in delimiting a construct and collecting behavioral indicators for its evaluation (Noronha & Almeida, 2022). At this point in time, evidence of content validity with Brazilian and Portuguese judges with expertise in the subject or in the construction of instruments, was carried out and confirmed that the items represented two theoretical factors: consistency of interests and perseverance of effort. The pilot study with samples composed of Brazilian and Portuguese university students was also carried out and, together with the other information, contributed to the development of a scale formed by 12 items distributed equitably by the two dimensions.

In view of the AERA et al. (2014) assertions that the assessment instruments must undergo rigorous and different investigations; in order to seek evidence of validity and precision estimates, this study was undertaken with the objective to explore the data structure, using the Exploratory Factor Analysis (EFA) and confirm the two-factor theoretical framework by the Confirmatory Factor Analyses (CFA). In addition, estimates of results accuracy for the scale were tested, as well as the invariance of its dimensionality in the samples of the two countries.

The parallel analysis and Hull’s method predicted one factor, but the sedimentation graph, although there is no consensus on its usefulness (Damásio, 2013), indicated two factors. The two-factor solution was tested and the items presented high factor loadings, the lowest referring to “I maintain a coherent line of objectives over time” (0.559), but still quite satisfactory. However, one item did not load the factor it was developed for (I try to commit to achieve my goals), which should be in Consistency of Interests. We proceeded to the CFA, which aims to answer whether the structure of the items is in line with the literature. The adjustment indices were good for the two theoretically recommended factors, reiterating evidence of validity for the EAGrIt-LP, now through the internal structure. The alphas were also satisfactory, considering that each factor has only 6 items.

Validity evidence can be confirmed based on the internal structure, since the items got arranged as theoretically predicted. That is, both theoretical factors were found (Duckworth et al., 2007; Duckworth & Quinn, 2009). It should be pointed out that many studies that translated the scales and tested them in normative samples different from the original studies were not successful in their results, as is the case of Sordia (2020), whose adjustment rates were poor, or the case of Tan et al. (2019) for Georgia and Egypt respectively. Another example to compose this list is the research by Arco-Tirado et al. (2018), in which the Grit-S was translated and adapted into Spanish. The correlation coefficients between the items were moderate, with the exception of a pair referring to perseverance. It was in this same factor that the lowest alpha index (α = 0.48) was found, and the adjustment indices were considered only acceptable.

Our results demonstrated a high correlation coefficient between the Perseverance of Effort and Consistency of Interests factors, indicating a substantial degree of commonality between them. Consequently, although the two-factor solution for these measures is pertinent for interventions with university students, as observed in other studies (Collantes-Tique et al., 2021), but it is necessary to further explore the dimensionality of the EAGrIt-LP. In a study involving 7,617 participants from 109 countries as part of the International Wellbeing Study (www.wellbeingstudy.com), researchers employed the Duckworth scale to test three models: a one-factor model, a two-factor model (similar to our study), and a two-factor model with a general factor. The findings revealed that the two-factor model provided the best fit indices. Notably, the correlation between the factors in that sample was r = 0.53, which remained statistically significant (Disabato et al., 2018). Therefore, further analyses of the scale’s dimensionality are recommended, including testing alternative models.

Thus, it is argued that the psychometric qualities of the measuring instruments must obey the rigor and steps already established in the specialized literature. A final observation concerns the invariance findings. According to Peixoto and Martins (2021), it is urgent to invest in estimates of psychological instruments invariance, since it is a unique strategy to guarantee the fairness of interpretations in the testing processes, which is also endorsed by Andrade and Valentini (2018). The authors stated that the investigation of invariance must precede the establishment of specific norms for groups.

New studies must be carried out. The sample was composed of Brazilian and Portuguese students, so that invariance analyses can provide relevant information. Comparisons between EAGrIt-LP and other convergent and divergent measures can inform on theoretical approximations and provide evidence of validity based on related constructs. Finally, analyses with academic outcomes are necessary, as a research agenda. The limitations of the research should be mentioned. Samples composed by convenience restrict the possibility of generalizing the results.