If we take into consideration the social and educational characteristics described above, we can argue that science education (SE) is still dominated by technical rationality. These ideas reveal the importance of approaching SE from a transformative vision.

Therefore, teacher development must be understood as a continuous process of theoretical-practical integration in which the teacher is conceived as a learner, innovator and researcher who participates in the construction of the science of teaching science (Furió and Carnicer, 2002).

In this line, the studies initiated by Shulman (1986) on the professional knowledge of teachers stand out. Specifically, from this research we focus on the ability to transform an object of knowledge into an object of teaching (Chevallard, 1985), in other words, in the pedagogical content knowledge (PCK).

The concept of PCK has been studied by various authors and teams. In the case of SE, PCK has been considered a useful theoretical framework since its origin and, therefore, we can find various proposals, such as Magnusson et al. (1999) and Gess-Newsome (2015) or the Refined Consensus (Carlson and Daehler, 2019). These models, given that they are based on advances in cognitive psychology, do not include aspects linked to affective variables, such as attitudes, in the components to be considered in teacher training and development. Other models, such as the hexagonal model (Acevedo and García-Carmona, 2016) or the ReCo model (Loughran et al., 2004), consider affective variables, but from an exclusively cognitive perspective, and the conative or emotional component of attitude is overlooked. In this sense, it is necessary to reflect on the role of these affective variables in the teaching of NoST in order to address the factors that prevent the implementation of these contents in the classroom, given their relevance in the creation of an authentic STL.

Taking the previous ideas into account, it becomes clear that teacher education must be based on a process that fosters didactic change (Mellado, 2003). This change involves not only professional transformation but also personal development. Teaching is not merely a cognitive process; it requires emotional commitment, engaging both the head and the heart, as it tests personal emotional and cognitive competencies as well as professional practice (Day, 1999). Therefore, learning changes who we are and what we are capable of doing—it is an experience of identity (Wenger, 1998). Identity is particularly relevant as it highlights the need for teacher learning and development to go beyond knowledge and skills (Avraamidou, 2014).

“Emotions influence the goals that teachers set and indicate the intensity of their relationships with ideas, their beliefs about science, others, and science teaching” (Rivera Maulucci, 2013, p. 137). This idea underscores that the development of Pedagogical Content Knowledge (PCK) cannot be limited to cognitive aspects alone if it is to contribute meaningfully to the construction of professional teaching identity.

It is important to note that evidence shows changes are more likely to become embedded when they incorporate attitudes and values (Sanmartí, 2001). Therefore, the focus on attitudes must be twofold: they should be seen both as a tool to achieve positive learning outcomes, given their influence on learning and behavior, and as essential and valuable content in science education. However, at present, emotional aspects remain largely absent from teacher education programs and, consequently, from student training plans (Bisquerra and Pérez, 2007).

With regard to knowing how to be, it is important to pay attention to the concept of attitude as a construct. Attitudes can be positive and negative, i.e., they are evaluative predispositions with a motivational component that implies affectivity and a tendency to action (Sarabia, 1992). Moreover, they have a learned character (Morales, 2000) which makes them a fundamental factor to take into account in the educational field due to their influence on behavior and learning.

There are multiple models that study the structure of attitudes. Following the three-dimensional model (Martín-Baró, 1983; Hogg and Vaughan, 1995), we can identify a clear parallelism with the structure of competences and, therefore, with educational objectives. These three components are: the cognitive component, which refers to the subject’s knowledge, beliefs and opinions about the object, taking a position in favor or against; the affective component, which refers to feelings, to the emotions of acceptance or rejection, liking or disliking, that are activated when faced with a stimulus; and the conative component, which refers to the declarations of intentions before the object of attitude. These components differ from one another, but converge intensely in that they share a common basis (the representation of the same attitude).

If we transfer this construct to Science Education, the literature reveals that there are problems derived from the lack of precision of the attitude, the attitude object and the absence of a single construct (Bogdan and Meneses, 2021; Manassero Mas and Vázquez, 2002; Vázquez and Manassero, 1995; Vázquez and Manassero Mas, 1997).

In this sense, the literature puts together taxonomies to organize and support the study of this construct and to define more accurately and precisely the attitudinal objects. One of the widely accepted classifications is the one proposed by Gardner (1975): attitudes toward science (a more affective and evaluative approach toward the elements involved in learning science) and scientific attitudes (with a cognitive character and linked to scientific thinking). However, the objects of attitude in science education cannot be limited to the attitude toward learning science. Hodson (1988) reveals that school is a key element in the shaping of a balanced attitude and, for this reason, he includes the attitude toward SE in his taxonomy. This category is the focus of the present study.

In summary, taking into account the ideas gathered, the manuscript is based on multiple theoretical perspectives that complement each other to support, contextualize, and operationalize the attitudinal approach (which constitutes the central construct).

To achieve the objective of designing and validating the proposed tool, a process comprising five stages was established, as summarized in Figure 1 and detailed below. Firstly, a review of existing instruments was conducted, and an initial draft of the tool was developed based on the reviewed literature. Secondly, content validity was assessed through expert judgment. Thirdly, the length of the scale was reduced by means of item correlation analysis and a second round of expert review. Fourthly, construct validity was examined through confirmatory factor analysis and goodness-of-fit indices. Finally, the internal consistency (reliability) of the questionnaire was evaluated using McDonald’s Omega coefficient. For the psychometric analysis, items related to sociodemographic variables were excluded, and the software programs Microsoft Excel, SPSS v.29, and Jamovi v.24 were used.

Initially, a series of scales were reviewed in order to analyze whether they could be applied according to the needs of the study or used as a reference. Those scales intended for students were discarded.

After the selection and review of scales aimed at teachers, several instruments were identified (Aikenhead and Ryan, 1992; Manassero et al., 2003; Rubba et al., 1996; Tyler-Wood et al., 2010; Vázquez and Manassero Mas, 1997; Villafañe and Lewis, 2016; etc.), but these were few in number and had limitations, such as the language in which they were written (predominantly English) and the non-holistic approach to the attitude object and construct (opinions toward science, attitudes toward scientific research, interest in science, etc.).

Therefore, as they did not respond to the characteristics of the study itself and the context in which it is carried out, a four-point Likert scale was designed (totally disagree, disagree, agree, strongly agree). The different items were written using the traditional method, with the criteria compiled by Morales (2000): Relevance, clarity, discrimination and bipolarity, and taking into account the three attitudinal components. The cognitive component was constructed by focusing on measuring the value, usefulness or importance that teachers attach to the discipline. These were formulated as cognitive judgments centered on the views of NoST as a discipline with value for the present and future of students. In the affective component, a series of adjectives (emotions) with a strong connotative meaning were selected to focus the measurement on the direction of emotions (positive/pleased or negative/unpleased) and their intensity (high or low polarization of the score). Emotions were selected from the Atlas de significados afectivos del español-mexicano by Urbán (1980), the taxonomy proposed by Fernández-Abascal et al. (2001), and the work of Goleman (1996). The conative component was constructed using verbal formulas that denoted intention or predisposition to perform the behaviors linked to the construct. In summary, the wording of the items was carefully designed to avoid expressions related to perceived ability or teaching competence (e.g., “I am able to…”) and instead emphasized evaluative judgements, affective responses, and intentional orientations toward teaching NoST.

For the drafting of the items, reference was also made to the educational models previously analyzed: the questions of NoST according to the STS Tradition (Acevedo and García-Carmona, 2016), Project 2061 (American Association for the Advancement of Science [AAAS], 2009) and the National Research Council (1996), as they contain the appropriate topics to characterize NoST in detail. This process led to a scale of 37 items that were then submitted to expert judgment.

For the content validity of the questionnaire, a procedure was carried out based on the judgment of three experts with recognized expertise in science education, NoST and attitudes. This number is consistent with methodological literature indicating that a small panel of highly qualified experts can provide robust qualitative validation in the early stages of scale development (Hernández-Sampieri et al., 2010). For the evaluation of the instrument, a descriptive scale was developed with four criteria (Escobar-Pérez and Cuervo-Martínez, 2008): clarity, coherence, relevance and sufficiency and four levels (with indicators) plus a section for “other comments.”

Based on the experts’ evaluation, the Content Validity Coefficient (CVC) of Hernández-Nieto (2002), which indicates the degree of agreement between the judges’ evaluations for each item, was calculated using Microsoft Excel.

The results obtained regarding the degree of agreement and content validity (Table 1) for the scale items indicate a predominantly excellent (>0.90) and good (>0.80 and < 0.90) level.

The feedback provided by the experts highlighted items that could be improved. Specifically, modifications were made to items 1, 4, 5, 6, 7, 11, 14, 20, 21, 22, 24, 25, 26, 31, 32, and 34 related to the clarity criterion, i.e., to improve understanding.

Once the necessary modifications had been made following the expert judgment validation, the instrument was administered using two types of sampling. On the one hand, a non-probabilistic snowball sampling method was employed, given that the target population was small and specialized. On the other hand, an accidental/convenience non-probabilistic sampling approach was used, as the selected sample consisted of student groups to which the researchers had access. In total, the instrument was administered to 219 Spanish-speaking participants.

Based on the results obtained from this pilot test, the scale was reduced through a Spearman correlation analysis between items. Items that did not show significant correlations with the others were submitted to a second round of expert judgment in order to assess the relevance of their removal. This decision was based on both the statistical outcome (which indicates whether the items are measuring the same dimension or diverging into unrelated aspects) and relevant literature in the field. As a result of this process, the scale was reduced from 37 to 27 items (Table 2). The Spanish version can be found in Table A1.

To assess the construct validity of the questionnaires, a Confirmatory Factor Analysis (CFA) was performed using the SEM (Structural Equation Modeling) package (Gallucci and Jentschke, 2021; Rosseel et al., 2019) in Jamovi v.24. This type of analysis was chosen because, as previously mentioned, the scales were developed based on a theoretically pre-established structure derived from the literature on the attitude construct. Consequently, the resulting path diagram (Figure 2) reflects a hierarchical model in which a second-order factor represents the overall attitude, expressed through its three components: cognitive, affective, and conative. These components, although distinct, converge strongly due to a shared foundation (i.e., the representation of the same underlying attitude) among individuals with consistent attitudinal profiles.

The factor loadings reveal that, in general, the relationship between observed variables and latent factors is both statistically significant and consistent, with most loadings being moderate (≥0.50) or strong (≥0.70). This suggests that the model’s structure is appropriate. For those items with lower factor loadings, their inclusion was deemed conceptually relevant by the experts involved in the second round of expert judgment. These items provide meaningful insights into the construct and help maintain balance across the different aspects represented (epistemology, internal sociology, and external sociology). For example, removing item 15 would result in a loss of valuable information concerning the affective dimension of teaching’s external sociology.

Model fit indices were analyzed (Table 3) to determine whether the proposed models fit the data satisfactorily. The Robust Weighted Least Squares estimator (WLSMV) was used, as it does not assume normality of variables and is especially suitable for CFAs with dichotomous or ordinal data (Xia, 2016). This estimator corresponds to a robust correction of DWLS. The interpretation of the results was based on scaled indices. First, absolute fit indices were considered: root mean square error of approximation (RMSEA), standardized root mean square residual (SRMR), and chi-square (χ²). Second, incremental fit indices were examined: comparative fit index (CFI), normed fit index (NFI), and Tucker–Lewis index (TLI). Third, parsimonious fit was assessed using the parsimonious normed fit index (PNFI).

Analysis of the data indicates that the proposed models, after certain modifications (removal of non-significant paths and addition of correlations between residuals), display an adequate fit. Absolute fit indices show acceptable values for this type of model, with RMSEA ≤ 0.08, SRMR < 0.08, and a χ²/df ratio close to 3 (Browne and Cudeck, 1993; Cho et al., 2020; West et al., 2012).

Similarly, the incremental fit indices (CFI, TLI, NFI, and GFI) yielded values considered acceptable (≥0.90) and optimal (≥0.95), according to specialized literature (Bentler, 1990; Hu and Bentler, 1999; Cho et al., 2020; Lai, 2020; Xia and Yang, 2019).

Finally, in terms of parsimony, the PNFI showed appropriate values, approaching 0.80 (Byrne, 2001; Hu and Bentler, 1999).

In summary, after making adjustments to the model, the results indicate a good fit to the data. Model modifications were implemented following a theory-informed approach. Modification indices were considered only when they were conceptually coherent with the multidimensional attitudinal framework and with established interpretations of NoST-related dimensions. No modifications were introduced that altered the underlying theoretical structure of the model. The various indices used to assess model quality support the consistency and stability of the proposed relationships, thereby confirming the validity of the scale’s underlying structure.

To assess the internal consistency of the questionnaire, reliability was analyzed using McDonald’s Omega coefficient (McDonald, 1999), in order to overcome the limitations associated with Cronbach’s alpha and to obtain more stable estimates (Ventura-León and Caycho-Rodríguez, 2017). The McDonald’s Omega value calculated for the scale was 0.96. The conclusion from this result is that the scale demonstrates high reliability (Campo-Arias and Oviedo, 2008).

As previously indicated, the analysis of the scale was complemented by an examination of the preliminary results obtained through its application. For this purpose, SPSS v.29 was used. This software was employed to compute descriptive statistics for the attitudinal components and for the overall scale. Additionally, comparisons were made based on selected sociodemographic variables (gender, academic/employment status, age, and experience).

Since the Kolmogorov–Smirnov test yielded a p-value of 0.003, indicating that the data did not follow a normal (Gaussian) distribution, and given the ordinal nature of the variables, non-parametric statistical tests were applied: the Mann–Whitney U test for comparisons between two independent groups, and the Kruskal–Wallis H test for comparisons among several independent groups. Effect size (r) was also calculated to assess the strength of the relationships. Furthermore, Spearman’s rank correlation coefficient was used to analyze the relationship between attitudinal components and to assess the external criterion validity of the scale.

Among the main limitations identified is, firstly, the sample size and its non-probabilistic nature. A larger and more diverse sample (both in terms of participants and schools) would yield more robust results across the different phases of statistical analysis. Regarding its non-random nature, it is worth noting that such sampling is common—and often the only feasible option—in educational research, due to practical and contextual constraints.

Secondly, since the instruments were self-administered, there is an increased risk of response bias, which may affect the accuracy of the data collected.

Thirdly, the instrument was designed and adapted for Spanish-speaking contexts, meaning its applicability in other settings cannot be guaranteed without prior translation and cultural adaptation to ensure validity and reliability.

Fourthly, although reducing the length of the questionnaire was necessary to avoid excessive burden, this limited the ability to explore the construct in greater depth and detail.

Finally, it is important to note that this instrument is not used in isolation but forms part of a broader methodological framework involving multiple tools aimed at gaining a deeper understanding of the complex nature of attitudes and avoiding the limitations of a purely positivist perspective in educational research.

Looking ahead, future research could involve expanding the sample size once the scale is fully validated, in order to obtain more conclusive results.

Applying the instrument in various educational programs could help analyze its potential and support further research based on educational interventions. This research should include longitudinal designs to examine attitude change over time, studies exploring the relationship between attitudinal components, and studies investigating the connection between teachers’ attitudes and actual classroom practices.

In addition, cross-cultural validation of the instrument in different educational contexts may be of interest.