With all the items, the Parallel Analysis indicated the existence of six factors, however, the distribution of items in the factors did not correspond to the projected facets for the construction of the items, leaving the factors, apparently, without theoretical adequacy. Therefore, a decision was taken to reduce the number of items and factors seeking a parsimonious model consistent with the theoretical aspects recommended for the locus of control.

The scale, which had 69 items, had its quantity reduced to 29 items (13 items for the external locus of control and 16 for the internal locus of control), after the EFA results. Exclusion criteria for the items were those that did not have factor loadings above 0.45, items with cross loadings (i.e., having loadings above 0.30 in the two retained factors) and negative items, that is, an item that was constructed for the external locus of control negatively loading in the internal locus of control and vice versa. From this, the 29 items that made up the final version of the scale were obtained. For the final version, Bartlett’s test of sphericity (15110.7; df = 406; p < 0.001) and the KMO (0.938; 95% CI = 0.937 - 0.939) indicated interpretability of the correlation matrix of the items. The Parallel Analysis in this configuration indicated two factors as representative of the data (Table 1). In this analysis with 29 items, the use of bootstrapping was considered.

Table 1 Table 1 Result of the Parallel Analysis Note.

CI = Confidence Interval.

The items presented high factor loadings in the factors, above 0.45, without the cross-loading pattern, and composite reliability, Cronbach’s alpha, McDonald’s omega and H indices (latent and observed) above their cutoff points (0.70 for the internal consistency coefficients and 0.80 for the H indices). In the case of the H indices, their values suggest that the factors are likely to be replicable in future studies. These data are presented in Table 2.

Table 2 Table 2 Factorial Structure of the Locus of Control Scale (ELOCUS) Note.

LOC E = external locus of control; LOC I = locus of internal control; bolded loads indicate values above 0.45.

The factor loadings of the internal dimension items ranged from 0.680 to 0.861 and for the external items these ranged from 0.494 to 0.757. The correlation between the factors was negative and of moderate magnitude (-0.421). The fit indices of the instrument in question were adequate, with RMSEA = 0.066 (95% CI = 0.062 – 0.067), CFI = 0.977 (95% CI = 0.974 – 0.981), and TLI = 0.973 (95% CI = 0.969 – 0.978) indicating a good fit of the model to the data. Regarding dimensionality, all coefficients indicated that there was no unidimensionality: UniCo = 0.808 (95% CI = 0.766 - 0.850); ECV = 0.755 (95% CI = 0.735 - 0.777); MIREAL = 0.321 (95% CI = 0.304 - 0.334). Discrimination parameters and item thresholds were checked using the Item Response Theory. Table 3 presents the discrimination parameters.

Table 3 Table 3 Discrimination Parameters of the ELOCUS Items Note.

^* indicates the most discriminative item for the factor; MDISC = multidimensional discrimination; a LOC E = discrimination parameter for external locus of control; a LOC I = discrimination parameter for internal locus of control.

According to Table 3, the most discriminative item of the external locus of control factor was item V43 (“Destiny controls me”; a = 1.077), followed by item V34 (“Luck is fundamental in my life”; a = 1.026) with both items referring to the chance/luck/destiny descriptor. Regarding the locus of internal control factor, the item with the highest discriminative power was V63 (“I’m struggling in my tasks”; a = 1.649), item from the proactivity/effort descriptor, in addition to two other items with close scores: item V21 (“When something is difficult, I dedicate myself to getting it”; a = 1.467), which is a proactive/effort item, and item V46 (“It is through my ability that I am able to finish my activities”; a = 1.457), which is a self-efficacy item. It should be highlighted that item V63 was the most discriminating item for the scale as a whole, and that the internal items showed higher discriminative values in general than the external items. Table 4 illustrates the thresholds of the ELOCUS items according to Likert scale.

Table 4 Table 4 Thresholds of the ELOCUS Items Note.

b₁, b₂, b₃, b₄ = difficulty of the parameters (thresholds).

The values were organized in Table 4 from the lowest to the highest value of parameter b.₁ In terms of thresholds, there was no unexpected response pattern, indicating increasing theta levels in all items. There was a clear division between the two dimensions of the locus of control. In this case, the easiest item to endorse was V63 (“I’m struggling in my tasks”). For external locus of control, the hardest item to endorse was V35 (“most tasks are solved by other people”). According to the patterns presented, on the one hand, it is possible to identify that the items allocated in the internal dimension need a lower level of latent trait to be endorsed. On the other hand, a higher level of latent trait is needed to mark the higher points of the Likert scale (i.e., 4 and 5) in external locus of control factor.

The results obtained with the analysis from a restricted model (CFA), aiming to verify the two-dimensionality for the final configuration of the scale, indicated adequate but borderline indices, with RMSEA = 0.082 (90% CI 0.080 - 0.085), CFI = 0.903 and TLI = 0.895. The factor loadings for the ELOCUS items, from this analysis, are presented in Table 5.

Table 5 Table 5 Factor Loadings from the CFA for the ELOCUS Note.

LOC E = external locus of control; LOC I = internal locus of control.

According to Table 5, the factor loadings for the external locus of control ranged from 0.375 to 0.767 and for the internal locus of control from 0.694 to 0.846, values close to those found in the unrestricted model, except for item V60 (“I think that my fate is already determined”), an item of the external dimension, with the descriptor chance/luck/destiny, which presented the lowest loading of 0.375. The correlation between the factors in CFA was negative and of moderate magnitude (-0.457).

The main focus of this study was to present the validity evidence based on the internal structure and reliability for a new scale that aims to measure the generalized expectations of locus of control in adults, designed to overcome the lack of tools of this type in the Brazilian culture. The scale in question presented a two-factor structure, conforming to what was theoretically expected: the locus of control has an internal dimension, representing the internal aspects of individuals that make them attribute any consequence as arising from the action itself, such as self-efficacy, autonomy, proactivity and responsibility; and the external locus of control expressed by individuals who interpret the results of their actions as consequences of external agents, such as people who exercise power over the person’s action, luck or fate leading the outcomes of events, passivity and disengagement directed toward the actions.

In terms of the theoretical model, although it was expected that the locus of control would present two factors, due to the elaboration of the descriptors for the composition of the internal and external locus, it was not known how this model would behave. Therefore, it was first decided to test the model through EFA, as the literature indicates that EFA is suggested to test models in which it is not known which factorial structure will result, as a way of tracking the model and providing a general view regarding the structure of the items, making it possible to indicate a potential model (Ferrando & Anguiano-Carrasco, 2010; Hair et al., 2018).

As this is a new instrument, starting the analyses with an EFA is acceptable, however, as the instrument, although new, was constructed considering a theoretical framework, it is desirable that it be minimally testable by a restricted model, as initial hypotheses were constructed based on a theoretical model for the construction of the measure (Ferrando & Anguiano-Carrasco, 2010). Therefore, the use of both analyses, unrestricted at first and subsequently restricted, is justified, even though the same sample was used (Hauck-Filho, 2019).

Of the nine descriptors defined and used for the semantic construction of the items, only one of them, “other powerful deities”, was totally excluded after the series of factor analysis were conducted. Its exclusion was due to criteria mentioned above, however, this was an unexpected result, mainly because the relevance that religion exerts in the formation of Brazilian cultural thought was considered (Silva, 2009). It is therefore suggested that, despite religiosity being something significant for most Brazilians, it does not affect the type of locus of control that they will tend toward. This may indicate that people with high internality scores may have characteristics focused on religiosity/spirituality, contrary to what was intended in their construction. In this sense, Author et al. (2021) found the internal locus of control more closely associated with expressions of religion/spirituality than the external locus of control. Therefore, further research needs to be conducted to better understand the relationship between the influence of these “hidden forces” (i.e., God, deities, energies, saints, and others) and the locus of control.

Another aspect that can be considered for this result is that the convenience sample influenced the scores as almost a third of the participants reported not having a religion (people who declared themselves to be atheists, agnostics or without religion), which justifies the fact that these items did not have explanatory power for the locus of control. Additionally, the level of formal education of participants may have influenced the outcomes. According to some research, more years spent in formal education, less likely one is to adhere to a religion or identify with a religious practice, or to continue exercising individual religious acts (e.g., praying), as well as decreasing the frequency of religious expression. (Dilmaghani, 2019; Hungerman, 2013; Masuda & Yudhistira, 2020). Further research is needed to verify the influence of educational level on the rise in power and influence of "hidden forces" on the establishment of the personal locus of control.

The final scale consisted of six proactive items, five self-efficacy items, four responsibility items and one autonomy item, forming the internal locus of control factor. For the external locus of control, five passivity items were obtained, four of luck/destiny (items referring to chance were excluded), two of “other powerful people” and two of disengagement. The meticulous choice of descriptors to theoretically compose each factor may have contributed to the good results regarding the fit indices presented by the scale in this study. The ELOCUS, from the EFA, obtained adequate fit indices of the data to the model, with the RMSEA value being considered good and for CFI and TLI values very good. Initially, the composition of each locus of control with a series of descriptors would be one of the differentials of the scale, that is, expanded and theoretically updated descriptors, which possibly provides a better theoretical explanation for each locus.

It can be seen that the factor loadings were similar for the unrestricted model and for the restricted model. With the exception of item V26 in the unrestricted model (“I always need someone’s help to resolve things”) and item V60 (“I think my fate is already determined”), all the other factor loadings were above .50 in the analyses, fulfilling the recommendation of Hair et al. (2018), in which the estimate for standardized loadings should be above .50, preferably above .70.

Another point regarding the factor loadings is that the loadings for the internal items were generally higher than those observed for the external items. It is assumed that the higher factor loadings for the internality factor are linked to social desirability, as the internal items express more socially accepted behaviors than the external items. Another hypothesis is linked to the sample composition that may have biased the results due to it being mostly female. It is recommended that future studies conduct investigations in relation to the invariance of sex, education, age and socioeconomic level, in order to test this hypothesis, an objective not covered in this study, as well as construct items that not only overcome, but assess the social desirability.

With regard to reliability, locus of control scales have historically presented internal consistency coefficients low or close to the suggested cutoff value of .70, except for the OLCS (Suárez-Álvarez et al., 2016) which presented good Cronbach’s alpha values (.87 for the internal locus and .85 for the external locus). In the case of the ELOCUS, all values of the internal consistency coefficients were higher, with even more adequate and robust indices, such as CR and McDonald’s omega, being proposed in this analysis. There is a discussion about the inadequacy of the alpha coefficient to measure the reliability of measures, however, this coefficient is still the most used in research. Some scholars indicate that other internal consistency coefficients should be presented in addition to alpha (Revelle & Zinbarg, 2009; Sijtsma, 2009). Based on these discussions, Hayes and Coutts (2020) state that the alpha coefficient is less accurate to measure reliability than the omega, and in this same study, they point out the difficulties for researchers to migrate from one coefficient to others and complement this by illustrating a series of tutorials in order to assist this transition. Both composite reliability (CR) and McDonald’s omega are coefficients that are not linked to the assumption of tau-equivalence, being more robust and adequate when compared to Cronbach’s alpha (Hayes & Coutts, 2020; Valentini & Damásio, 2017).

Concerning the item discrimination parameters (parameter a), the item that best represents the external locus of control is an item that addresses fate, that is, people that more attribute beliefs in fate to outcomes in their daily lives tend to be more “external”. The item that was most representative regarding the internal locus of control was a proactive item which attributed personal effort as the main quality for the expression of internality. In other words, this means that the more a person perceives effort in carrying out daily tasks, the more they tend to be “internal”. This same item was also appointed as more representative in the multidimensional discrimination, which means that it is an item of which the discrimination indicates the complete set of the scale, taking into account all the factors existing in the model (Damásio et al., 2021).

The threshold, or threshold between responses (parameter b), is the amount of latent trait (theta) the individual needs to have to go from a response point, in the current response format on the scale, to the next point (Damásio et al., 2021). One of the necessary observations is to verify whether all the items in the scale present a pattern of increasing threshold values, considering the signal presented, being able to start at the smallest points of the negative signal of the response format. Damásio et al. (2021) indicated that what is expected for this indicator is that as the theta level increases, so does the response category level. All items of the ELOCUS presented an increasing gradation of thresholds, indicating a result in line with what is expected. It is important to note that in this study there was no intention of making a complete analysis based on IRT, as it is known that there are other procedures involved and of great complexity, however, with the data from the FACTOR software, it is possible to have a global view on the process of answering the items. That said, it can be inferred that this is another differential of the ELOCUS, that is, the presentation of studies based on the Classical Test Theory (CTT) and Item Response Theory (IRT).

A final point to be addressed is in relation to the statistical procedures used by most locus of control scales to illustrate the internal structure. When these instruments were constructed, it was common to use Principal Component Analysis (PCA) as Exploratory Factor Analysis (EFA). The use of PCA became widespread due to the fact that some statistical packages only offer this method, erroneously naming it EFA (Cudeck, 2000; Watkins, 2018) and due to the lack of computational technology accessible to most researchers, given that the EFA procedure requires more sophisticated software and hardware than was commercially available at the time (i.e., 1960s to late 1980s).

Therefore, PCA was the most used analysis model in studies of locus of control scales, proving inadequate, as it is important to emphasize that PCA and EFA are different mathematical procedures that start from different algebraic assumptions (Cudeck, 2000). It cannot be said that PCA generates factors, but components (Brown, 2015; Damásio, 2012; Watkins, 2018), as the components do not represent latent variables (Cudeck, 2000), being more suitable as indicators (e.g. Human Development Index – HDI). In the case of PCA, the indicators influence the component (Watkins, 2018) and in EFA it is the factor, representing a latent variable, which explains the observable behavior portrayed by the items of an instrument. Furthermore, common and specific variance are undifferentiated in the PCA model (Brown, 2015; Damásio, 2012; Ferrando & Anguiano-Carrasco, 2010).

Currently, there is a series of software programs that are easily mastered by researchers who wish to include EFA in their investigations, while modern and accessible equipment (personal computers and desktops) are available (Watkins, 2018). This technological facility allows us to construct psychometric instruments with better conditions, when compared to those arising from the first scales, in the 1960s. Although much of the mathematical and statistical development was already underway at the time, the lack of technology to process data for this type of analysis did not allow this model, for example, to be frequently used. In fact, from the 1970s onwards, due to the complexity of the EFA, it lost space in research to the CFA (Ferrando, 2021). Therefore, providing validity evidence based on the internal structure to the ELOCUS, based on more up-to-date technological resources, adds to the field as the other scales have not presented studies of this nature.

The present study sought evidence of validity based on the internal structure and reliability for the ELOCUS aiming to add to the field of psychological assessment and generate a new tool that investigates aspects of the relationship of individuals locus of control. The consideration of the various descriptors devised for the construction of the items for each locus and statistical procedures based on more up-to-date models promoted different psychometric qualities for the ELOCUS when compared to other scales developed over more than 50 years of history.

This scale can be of great evaluative use in the most diverse areas of Psychology, as it is an instrument with few items, quick to apply and has been created from a perspective of generalized expectations of control, seeking not to have any context-dependent links, therefore making it as neutral as possible. Furthermore, scales with generalized expectations can provide some advantage in relation to their use in research by allowing some flexibility as they are not linked to a specific context and generate comparable results, broadening the understanding of the construct.

Although the sample had an interesting composition of quantity and included people from practically all Brazilian states, it may have presented some biases, such as being composed by mainly females and people with high levels of education, which can be considered limitations. Furthermore, as mentioned above, almost a third of the sample had no religious ties, which may have influenced the removal of items related to the power exercised by deities.

Accordingly, future research could include invariance studies considering the demographic variables presented in this study, Differential Item Functioning (DIF) studies, also considering the crucial need to present validity evidence in relation to external variables. Providing a quick screening of aspects inherent to control, the ELOCUS could constitute an easy-to-handle instrument for the various Psychology professionals and a practical resource in the daily life of the psychologist who works with psychological assessment and intervention.