Teacher Efficacy Beliefs: A Multilevel Analysis of Teacher- and School-Level Predictors in Mexico

Abstract

All individuals hold beliefs about their ability to successfully perform specific tasks. These beliefs, known as self-efficacy, play a key role in guiding and motivating human behavior. In education, both teachers’ self-efficacy beliefs and the collective efficacy shared by teachers within a school have been linked to improved performance, well-being, and job satisfaction among students and educators. While these constructs have been widely studied in various countries and contexts, little is known about them in Mexico, the country with the largest Spanish-speaking population worldwide. This study is the first to examine the relationship between teacher self-efficacy (TSE), collective efficacy, and other teacher- and school-level variables in Mexico. Given the absence of psychometrically robust instruments to assess collective efficacy among Spanish-speaking teachers, the Collective Teacher Beliefs Scale (CTBS) was first adapted into Spanish, and its psychometric properties were evaluated. Subsequently, multilevel analyses incorporating teacher- and school-level factors revealed that professional development on multicultural communication, classroom autonomy, and collaboration, at the teacher level, and collective efficacy and stakeholder participation, at the school level, were significant predictors of TSE. Finally, implications for future practice and policy are discussed.

1. Introduction

Everything we do is permeated by our beliefs about how capable we are of performing a certain action. In this sense, we tend to engage in tasks in which we see ourselves as more competent, avoiding those in which we do not feel confident enough to succeed. These beliefs of personal competence are what we know as self-efficacy beliefs, which have been described as the “beliefs in one’s capabilities to organize and execute the courses of action required to produce given attainments” (Bandura, 1997, p. 3).

Within the educational field, teacher efficacy beliefs have been an essential topic for practitioners and researchers because of their close relationship with positive student and teacher outcomes. At the individual teacher level, higher self-efficacy beliefs are linked to improved student achievement (Guo et al., 2012; Mojavezi & Tamiz, 2012), better teacher–student relationships (Hajovsky et al., 2020; Perera & John, 2020), and greater job satisfaction (Kasalak & Dağyar, 2020; Zakariya & Wardat, 2024). At the school level, collective teacher efficacy (CTE)—the shared perceptions among teachers from the same school about their ability to positively influence student learning (Salas-Rodriguez & Lara, 2023)—has been associated with school trust (Lee et al., 2011; Wu et al., 2013), a strong sense of belonging among teachers (Donohoo, 2017b; Skaalvik & Skaalvik, 2021), and higher teacher commitment (Cansoy et al., 2020; Ware & Kitsantas, 2007).

Despite extensive research on these beliefs across various countries such as the United States (R. D. Goddard & Goddard, 2001; Tschannen-Moran & Woolfolk Hoy, 2001), Turkey (Turkoglu et al., 2021), Iran (Mohamadi & Asadzadeh, 2012; Rezaeian & Abdollahzadeh, 2020), China (Ma & Trevethan, 2020; Peng, 2020), Israel (Schechter & Tschannen-Moran, 2006), and Norway (Skaalvik & Skaalvik, 2007, 2019), there is a remarkable gap in studies focused on Latin American contexts.

With this in mind, the purpose of the current study is twofold. First, given the lack of a valid and reliable instrument to assess CTE among Spanish-speaking teachers, we intend to examine the psychometric properties of the Collective Teacher Beliefs Scale (CTBS; Tschannen-Moran & Barr, 2004). Second, we want to explore which teacher-level (e.g., gender, subject taught, school grade, teaching experience, professional development [PD], classroom autonomy, collaboration, job satisfaction [JS]) and school-level (e.g., CTE, participation among stakeholders, instructional leadership) variables predict teacher self-efficacy (TSE) beliefs in a sample of Mexican teachers from private schools. A better understanding of the variables related to TSE might help create educational environments that foster these competence beliefs among the whole faculty. Thus, our research questions are as follows:

RQ1. Does the Spanish CTBS present the same internal structure as the original version of the CTBS?

RQ2. Which teacher-level variables significantly predict TSE?

RQ3. Which school-level variables significantly predict TSE?

RQ4. Which set of teacher and school-level variables better explain TSE?

2. Theoretical Framework

2.1. Teacher Self-Efficacy

Since its inception, TSE has been strongly related to student achievement (Armor et al., 1976; Bal-Taştan et al., 2018; Caprara et al., 2006). Teachers with higher self-efficacy beliefs tend to be more committed to the teaching profession (Chesnut & Burley, 2015; Klassen & Chiu, 2011; Mokhtar et al., 2021) and develop more challenging learning activities for their students (Bandura, 1977; Burić & Kim, 2020). Teachers with more confidence in their own competence are more willing to take risks and innovate (Cai & Tang, 2021), as well as collaborate with colleagues and parents (Skaalvik & Skaalvik, 2010). Furthermore, teachers with higher self-efficacy tend to enjoy their work more and report greater levels of JS and well-being (Burić & Moè, 2020; Wang et al., 2015).

Despite its relationship with numerous positive outcomes, there is a lack of studies that delve into how these beliefs are developed in practice. From a theoretical perspective, Tschannen-Moran et al. (1998) proposed an integrated model of teacher efficacy, emphasizing the cyclical nature of these beliefs. First, teachers value and interpret the information gathered through the four sources of efficacy described by Bandura (Bandura, 1977; Usher & Pajares, 2008): mastery experiences, vicarious experiences, verbal persuasion, and physiological states. Then, teachers put this information in context by considering two more factors: the analysis of the teaching task at hand and their teaching competence. The inclusion of these two factors in the model reveals self-efficacy beliefs are context-specific. Consequently, when reflecting on their self-efficacy beliefs, teachers consider their own competence with respect to the perceived demands of a specific teaching task in a certain learning context (Tschannen-Moran et al., 1998). That is to say, TSE may vary under different circumstances.

Following the model, the way teachers perceive themselves as more or less capable of achieving a certain goal will affect their actions, attitudes, and efforts, thus also affecting their general performance. This performance provides new information to value and interpret to shape future efficacy beliefs. The cyclical nature of self-efficacy has the potential to grow or diminish these competence beliefs since “greater efficacy leads to greater effort and persistence, which leads to better performance, which in turn leads to greater efficacy. [Though] the reverse is also true” (Tschannen-Moran et al., 1998, p. 234).

Based on this integrated model, Tschannen-Moran and Woolfolk Hoy (2001) proposed the Teachers’ Sense of Efficacy Scale (TSES). This measure is composed of three factors—instructional strategies, classroom management, and student engagement—reflecting the multifaceted nature of self-efficacy. Thus, in the present study, we define TSE as the teacher’s self-assessment of their teaching competence, considering their teaching context and tasks related to instructional skills, classroom management, and engagement of the students.

The TSES has been the most used questionnaire to assess TSE beliefs. The scale has been translated into different languages and adapted to diverse learning contexts (Koniewski, 2019; Mastrothanasis et al., 2021; Ninković & Knežević-Florić, 2018; Salas-Rodriguez et al., 2021; Tsigilis et al., 2010; Valls et al., 2020). In 2013, the OECD (Organization for Economic Co-operation and Development) made a few minor changes to the TSES and included it on the teachers’ questionnaire in the Teaching and Learning International Survey (TALIS). The results obtained by TALIS have allowed us to better understand the Mexican educational context regarding TSE beliefs. However, TALIS has not included CTE in any of its previous studies. Thus, more research is needed on this topic since schools’ success is linked to both self- and collective teacher efficacy beliefs.

2.2. Collective Teacher Efficacy

As with TSE, the interest in CTE arose from its strong relationship with student achievement (Bandura, 1993; R. D. Goddard, 1998; Hattie, 2012). Some studies have shown CTE has a stronger impact on student achievement than students socioeconomic status (Bandura, 1993; Erdogan et al., 2022; R. D. Goddard et al., 2000, 2015). Furthermore, CTE promotes a sense of shared responsibility regarding students’ learning and achievement (Prieto, 2007; Tschannen-Moran & Barr, 2004). That is, teachers do not blame external factors for their students’ performance, but rather propose solutions and are more open to change. Teachers with higher collective efficacy beliefs are characterized as persevering amid challenging situations and/or students (R. D. Goddard et al., 2000; R. D. Goddard & Goddard, 2001; Mosoge & Challens, 2018), setting higher learning expectations for their learners (Donohoo, 2017a; R. D. Goddard et al., 2004b), establishing trustful relationships with their students and colleagues (Lee et al., 2011; Zhang & Yin, 2017), and presenting higher levels of commitment towards shared goals (Hallinger et al., 2018; Hosseingholizadeh et al., 2020; Ross & Gray, 2006).

TSE and CTE operate through similar processes and influence performances in similar ways. However, they differ in their forms of agency, namely, personal and collective. Unlike TSE, CTE is an emergent group-level property that results from the interactive dynamics among a school’s faculty (R. D. Goddard et al., 2000; Tschannen-Moran & Barr, 2004). These beliefs are part of the school culture and can enhance or harm how a school works (Bandura, 1993). In this sense, CTE beliefs shape the social norms of a school, thus affecting teachers’ expectations, actions, and performances (R. D. Goddard & Goddard, 2001).

Unlike TSE, CTE has been much less studied due in part to the complexity of the construct (R. D. Goddard, 1998). Although CTE refers to an emergent property of schools, the only way to assess it is through its individuals. However, when measuring CTE, items should ask teachers about their perceptions of their faculty competence by following a group-level orientation. According to previous studies (Eells, 2011; Salas-Rodriguez & Lara, 2020), the two more popular scales to measure CTE have been the Collective Teacher Efficacy Scale (CE-Scale; long form: R. D. Goddard et al., 2000; short form: R. D. Goddard, 2002) and the Collective Teacher Beliefs Scale (CTBS; Tschannen-Moran & Barr, 2004). Both scales have their theoretical foundations on the integrated model of teacher efficacy (Tschannen-Moran et al., 1998) and their items are worded following a group orientation. Nevertheless, the CE-Scale has received several criticisms due in part to its explicit measure of task difficulty, thus impacting CTE scores in schools with a more challenging context (Tschannen-Moran & Barr, 2004).

In response to these concerns, the CTBS was developed as a new measure to assess CTE beliefs. This scale is an adaptation of the TSES (Tschannen-Moran & Woolfolk Hoy, 2001) at the group level. Instead of measuring three factors, the CTBS focuses on two: teachers shared instructional strategies and student discipline in the school (Tschannen-Moran & Barr, 2004). To our knowledge, no previous study has explored the CTE beliefs of Mexican teachers. For this reason, we decided to include CTE as an independent school-level variable in this study. This will allow us to know how CTE, among other teacher- and school-level variables, helps predict TSE within a sample of Mexican teachers.

2.3. Teacher Self-Efficacy Predictors

As previously mentioned, TSE has been related not only to CTE but also to different teacher- and school-level variables. In this study, several variables from the TALIS 2018 framework (Ainley & Carstens, 2018, p. 29) were used to further examine their relationship with TSE beliefs in the Mexican context.

2.3.1. Teacher-Level Variables

Regarding teacher-level variables, TSE has been studied in relation to different demographic factors. As for gender, previous studies have yielded inconsistent results. In some cases, female teachers have reported higher levels of TSE (Dilekli & Tezci, 2020); in others, male teachers did (Klassen & Chiu, 2010). However, some studies claim TSE does not vary regarding teacher gender (Ninković & Knežević-Florić, 2018). Since TSE is context-specific, another factor that may affect efficacy levels is the subject taught. In this regard, TSE beliefs may change depending on the difficulty of the teaching field at hand (Haverback & McNary, 2015).

TSE may also vary according to school grade. Primary school teachers tend to present higher levels of self-efficacy than secondary teachers (Fives & Buehl, 2009). In the Mexican context, secondary teachers have reported lower levels of self-efficacy for student engagement than primary teachers (Backhoff Escudero & Pérez-Morán, 2015; Salas-Rodriguez et al., 2021). As for teaching experience, previous studies have found a curvilinear relationship with TSE. That is, teachers’ self-efficacy increases during their first years in the profession but after about 20 years of teaching experience, efficacy beliefs tend to decline as experience increases (Klassen & Chiu, 2010).

A key factor in improving the quality of teaching is PD, namely, “those processes and activities designed to enhance the professional knowledge, skills, and attitudes of educators so that they might, in turn, improve the learning of students” (Guskey, 2000, p. 16). Tschannen-Moran and Chen (2014) asserted teachers’ ongoing professional learning needs to attend to their efficacy beliefs, so as to drive their motivation and effort when learning and implementing new instructional practices. Y. Liu and Liao (2019) found that the types of PD activities—i.e., format—and the content offered in these activities, are two aspects that can influence TSE beliefs. In the present study, we included both aspects of PD—format and topics—as teacher-level predictors.

Regarding the types of PD, Tschannen-Moran and McMaster (2009) explored four different PD formats to work on the same teaching strategy. Results indicated that the teachers who participated in activities that supported mastery experiences through follow-up coaching presented higher levels of TSE for reading instruction and implementation of the new strategy. Similarly, Gümüş and Bellibaş (2021) explored the extent to which different formats of PD activities predicted TSE in 32 countries. In Mexico, they found that teachers who participated in conferences and seminars, networking activities, mentoring and coaching programs, and research activities presented higher levels of TSE.

Studies exploring how PD topics relate to TSE have found that contents for curriculum, instructional skills, school management, and technology were statistically related to TSE (Y. Liu & Liao, 2019). PD topics are important for teaching quality since teachers who have been trained in certain areas feel more confident and report higher levels of TSE.

Another factor related to TSE is teacher autonomy, or “the freedom to choose goals, teaching methods, and educational strategies that are concordant with the teacher’s personal educational beliefs and values” (Skaalvik & Skaalvik, 2014, p. 69). Autonomy has been considered a universal psychological need and should not be confused with independence or individualism (Deci & Ryan, 2000). When teachers have greater autonomy, their TSE increases because they realize that they can accomplish what they set out to do. In TALIS 2018, 47 of the 48 participating countries obtained a positive and significant correlation between class autonomy and TSE (OECD, 2020).

TSE has also been linked to higher levels of teacher collaboration (Sehgal et al., 2017; Vangrieken et al., 2015). When teachers collaborate they have more opportunities to share their knowledge and experiences, thus learning from each other and promoting their instructional improvement (Y. L. Goddard et al., 2007). In TALIS 2018, 61% of teachers reported discussing the learning development of specific students at least once a month, whereas only 21% of teachers participated in collaborative professional learning, and 9% observed other teachers’ classes and provided feedback. Teachers who engaged in deeper forms of collaboration regularly tended to report higher levels of TSE (OECD, 2020).

JS refers to the sense of fulfillment derived from enjoying one’s work (García Torres, 2019; Klassen, 2010; Virtanen et al., 2019). It is considered a multidimensional construct since workers may experience diverse emotions—positive or negative—regarding the different aspects of their job (Locke, 1969). TSE and JS have been considered antecedents of teacher retention (Aldridge & Fraser, 2016). Teachers who believe they have what it takes to teach their students tend to experiment more positive affect and, consequently, are more satisfied with their job (Moè et al., 2010). In this sense, several studies have found TSE is strongly related to JS (Burić & Kim, 2021; Caprara et al., 2003; Zakariya, 2020; Zee & Koomen, 2016). According to TALIS 2018, 98% of Mexican teachers reported being satisfied with their jobs, heading the list of the 48 participating countries (OECD, 2020).

2.3.2. School-Level Variables

Regarding school-level variables, TSE has been strongly related to CTE. Some studies have even shown that self- and collective teacher efficacy have a reciprocal relationship (R. D. Goddard & Goddard, 2001). For example, a teacher with high self-efficacy might act differently in a school with low CTE. Though the reverse is also true, a teacher with low self-competence beliefs might try harder and be more motivated to teach in a school with higher levels of CTE. “The sense of collective efficacy in a school can affect teachers’ self-referent thoughts and, hence, their teaching performance and student learning” (R. D. Goddard et al., 2004a, p. 8). Kurt et al. (2011) found CTE was the strongest variable having a direct effect on TSE as well as having a mediating role between transformational leadership and TSE.

Another school-level variable related to TSE is participation among stakeholders. TALIS defines this construct as the opportunities that faculty, parents, and students have to actively participate in school decisions, along with the school’s culture of shared responsibility and collaboration (OECD, 2020). Some studies have equaled participation among stakeholders to distributed leadership since both terms refer to how different individuals within an educational organization enact leadership (García Torres, 2019). In this sense, Sun and Xia (2018) found that teachers who perceived distributed leadership more positively (i.e., higher levels of participation among stakeholders) tended to report higher levels of TSE and JS. In the words of García Torres (2019): “When teachers are granted greater control over their work conditions through distributed leadership opportunities, they experience greater self-efficacy to collaborate with peers, which is also associated with greater job satisfaction levels” (García Torres, 2019, pp. 120–121).

Finally, several studies have found that principals’ instructional leadership is related to higher levels of TSE. In the present study, instructional leadership is defined as the actions taken by the principal to promote student learning and improve the instructional quality of teachers (Ainley & Carstens, 2018). According to Y. Liu et al. (2020), school leadership focused on instructional improvement helps develop teachers’ instructional beliefs and, consequently, has a greater impact on TSE than distributed leadership. Çalik et al. (2012) found that the instructional leadership of principals had a significant effect on TSE. Also, a meta-analysis by Alanoglu (2021), concluded that the overall effect size for the relationship between instructional leadership and TSE was medium.

Taking all this into account, we expect that most teacher- and school-level variables included in this study may significantly help predict the variance in TSE in our data sample. Figure 1 illustrates the conceptual framework for the present study.

3. Methods

3.1. Participants

Two different samples were used in this study since some schools did not provide the necessary data to perform multilevel analyses. First, we describe the sample used to assess the psychometric properties of the CTBS (i.e., Sample 1). Then, we depict the sample used to test the different multilevel models (i.e., Sample 2). A convenience sampling method was chosen for both samples. Data were collected between May and June of 2020 through a Google Forms questionnaire. Participants were informed of the purpose of the studies and answered the survey anonymously, knowing that the confidentiality of the provided data was guaranteed. Ethical approval was obtained by the Research Ethics Committee of the authors’ affiliated university (Project ID: 2020.042).

3.1.1. Sample 1

This sample encompassed 190 in-service teachers (120 females, 70 males; age: M_age = 40.89, SD = 10.05) from 25 private Mexican K-12 schools. The participating teachers taught in 4th grade (n = 45), 5th grade (n = 39), 6th grade (n = 34), 7th grade (n = 34), and 8th grade (n = 38). Years of teaching experience oscillated between 1 and 41, with a mean of 16 years (SD = 9.98).

3.1.2. Sample 2

This sample included 162 in-service teachers (94 females, 68 males; age: Mage = 41.19, SD = 10.33) and 22 school principals (12 females, 10 males; age: Mage = 48.37, SD = 7.67) from 22 private K-12 schools in Mexico. These 22 schools represent a subsample of those included in Sample 1. The participating teachers taught Spanish (47.5%) and mathematics (52.5%) in 4th grade (n = 40), 5th grade (n = 32), 6th grade (n = 29), 7th grade (n = 29), and 8th grade (n = 32). Years of teaching experience ranged from 1 to 41, with a mean of 16.4 years (SD = 10.29).

It is important to emphasize the rationale behind our choice of private schools in Mexico. Since the first part of the study is dedicated to examining the psychometric properties of the Spanish CTBS, we aimed to uphold the characteristics of the sample used in the original study conducted with middle school teachers (5th to 8th grade; Tschannen-Moran & Barr, 2004). However, in the Mexican context, public schools are not K-12 and they are divided into primary (1st to 6th grade), secondary (7th to 9th grade), and high school (10th to 12th grade). Therefore, K-12 private schools serve as the only means to compare efficacy beliefs among teachers within the same school but across different educational levels.

Unlike public schools, Mexican private schools rely on tuition rather than government funding, and they mainly serve students from middle and high socioeconomic backgrounds. They have more resources than public schools and more flexibility regarding teaching and learning approaches while still following the national curriculum established by the Secretariat of Public Education (INEE, 2019c; Pozas et al., 2021). Although quality and equity have been a priority in the Mexican context (OECD, 2019a), based on national evaluations, students in private schools present significantly higher levels of achievement throughout the different school levels (INEE, 2019a, 2019b).

3.2. Variables and Measures

This section highlights the variables and measures used for this research. Appendix A presents the instruments used to assess the independent variables.

3.2.1. Dependent Variable

TSE was measured using the Spanish version of the TSES (Salas-Rodriguez et al., 2021). This scale comprises 12 items grouped into three subscales with 4 items each: instructional strategies, classroom management, and student engagement. Items are rated on a 9-point unidirectional scale, ranging from 1 (Nothing) to 9 (A great deal). The global score was obtained by averaging the mean score of the three factors. The overall reliability was good (α = 0.91), as was the consistency of its subscales (ranging from 0.78 to 0.85; Salas-Rodriguez et al., 2021).

3.2.2. Independent Variables

Teacher-Level Variables

Socio-demographic variables included teachers’ gender, subject taught, school grade, and years of teaching experience.

Teacher PD was measured in two different ways according to TALIS 2018 (OECD, 2019c). First, teachers were asked if, during the last 12 months, they have participated in diverse types of PD activities, such as “Courses attended in person”, and “Education conferences”. Then, teachers indicated which topics were included in those PD activities. A list of different topics was displayed, including options such as “Knowledge of the curriculum” or “Student behavior and classroom management”. Teachers’ response categories for each PD activity and topic were codified as 0 for “No” and 1 for “Yes”.

Satisfaction with classroom autonomy was assessed using the scale included in TALIS 2018 (OECD, 2019c). This scale comprises five items focused on whether teachers have control over certain aspects of their planning and teaching regarding their target class. Participants indicated their response on a 4-point scale, ranging from 1 (Strongly disagree) to 4 (Strongly agree). A higher score on this scale reflects higher satisfaction with their classroom autonomy. The reliability of this TALIS scale used in Mexico was good (ω = 0.83; OECD, 2019c, p. 296).1

Teacher collaboration was measured using the proposed scale from TALIS 2018 (OECD, 2019c). Two subscales form this scale—Exchange and coordination among teachers, and Professional collaboration in lessons among teachers—with four items each. The items are rated on a 6-point scale, ranging from 1 (Never) to 6 (Once a week or more). The higher the total score, the greater the level of teacher collaboration. The internal consistency of this scale used in TALIS Mexico was good (α = 0.81); however, its subscales displayed lower values (ω = 0.78 and ω = 0.64, respectively; OECD, 2019c, p. 242).

JS was evaluated as a multi-dimensional construct by the composite scale from TALIS (OECD, 2019c). This measure comprises two subscales with four items each: JS with work environment, and JS with profession. Items were coded using a 4-point response scale ranging from 1 (Strongly disagree) to 4 (Strongly agree). A higher score on this scale corresponds to a higher level of teacher JS. The overall reliability of this TALIS scale used in Mexico was good (α = 0.79); nonetheless, the internal consistency of its subscales presented lower, though still acceptable values (ω = 0.75 and 0.64, respectively; OECD, 2019c, p. 296).

School-Level Factors

CTE was assessed using the CTBS (Tschannen-Moran & Barr, 2004) with the approval of one of its authors (MTM). This measure comprises two subscales: Instructional strategies (CIS; 6 items) and Student discipline (CSD; 6 items). The 12 items were rated on a 9-point unidirectional scale, ranging from 1 (Nothing) to 9 (A great deal). The overall reliability of the original scale was good (α = 0.97), as was the consistency of its subscales (α_CIS = 0.96; α_CSD = 0.94). Since CTE is conceptualized as a group-level variable, individual teacher responses were aggregated at the school level to obtain a single score for each participating school.

The CTBS was not previously available in Spanish, so it was translated from English following the translation and back-translation procedure. First, the CTBS was translated into Spanish by a native Spanish-speaking scholar. Next, a native English professional re-translated the questionnaire from Spanish to English. Then, the authors and the professionals revised both versions item by item to detect semantic and/or conceptual discrepancies between the original and translated versions. Any disparities were discussed, and a consensus was reached for each item. Finally, a Mexican scholar and two Mexican educators revised the CTBS Spanish version to guarantee the neutrality of the vocabulary used in the adaptation. It was concluded that no further changes were needed due to the standard register of the language used in the questionnaire.

Participation among stakeholders was evaluated using the scale proposed by TALIS 2018 (OECD, 2019c). This scale consists of five items, which asks teachers whether the school has a culture of shared responsibility where teachers, parents, and students keenly participate in school decisions. Items were coded using a 4-point scale ranging from 1 (Strongly disagree) to 4 (Strongly agree). A higher score on this scale corresponds to a higher level of participation among stakeholders. The internal consistency of this TALIS scale used in Mexico was also good (ω = 0.83; OECD, 2019c, p. 331).

Instructional leadership was assessed using the School leadership scale proposed by TALIS (OECD, 2019c). The principals of the participating schools answered this 3-item scale following a 4-point response scale ranging from 1 (Never or rarely) to 4 (Very often). A higher score on this scale reflects a higher level of instructional leadership. This TALIS scale used with Mexican teachers yielded adequate internal consistency (ω = 0.84; OECD, 2019c, p. 387).

3.3. Data Analysis

The collected data were analyzed using STATA 15. The analysis procedures included two stages. The first stage aimed to examine the psychometric properties of the Spanish CTBS in Sample 1 (RQ1), specifically, its internal consistency—by means of Cronbach’s alpha—and internal validity evidence—employing confirmatory factor analysis (CFA).

First, the suitability of the sample data was examined using the Kaiser–Meyer–Olkin (KMO) test and Bartlett’s test of sphericity. Then, following the structural equation modeling framework (Brown & Moore, 2012), two CFAs were applied to assess the internal validity evidence of the CTBS: a one-factor model—loading the 12 items on a single dimension of CTE—and a two-factor model—as proposed by Tschannen-Moran and Barr (2004). To determine the quality of both CFAs, the goodness-of-fit of the models was tested using different measures: χ²/df ratio—where a ratio ≤ 3 suggests a good fit (Byrne, 2004)—; standardized root mean square residual (SRMR) and root mean square error of approximation (RMSEA)—where a value of ≤0.08 specifies a good fit (Hu & Bentler, 1999)—; and comparative fit index (CFI) and Tucker–Lewis index (TLI)—with acceptable values considered as ≥0.9 (Bentler, 1990).

The second stage aimed to examine what variables explain the variance in TSE beliefs and to determine whether they were at the teacher- or the school-level in Sample 2. First, descriptive statistics analyses were used to summarize data. School-level variables, such as CTE, participation among stakeholders, and instructional leadership, were aggregated at the school-level. Shapiro–Wilk tests were applied to check for normality. Since non-normal distributions were found among all variables, Spearman’s correlation analyses were performed to test for significant associations and detect potential multicollinearity problems between variables. Also, due to the dichotomous nature of the formats and topics of the PD activities, point-biserial correlation analyses were used to test for significant associations.

HLM was used to analyze the relationship between TSE and the independent variables while considering the different levels of variables. In this study, teachers (level 1) were nested within schools (level 2). Thus, several models were fitted to estimate the variation in TSE beliefs regarding variables from both levels of analysis. First, the suitability of the data was tested against HLM assumptions. The unconditional model—without predictors—was used to calculate the intra-class correlation coefficient (ICC) to determine how much variation was associated with the school cluster. Next, four partially conditional models—with only teacher-level predictors—were evaluated in a stepwise manner (RQ2). Then, two models were tested, including only school-level predictors (RQ3). Finally, two fully conditional models were tested, incorporating only the significant predictors determined by previous models (RQ4). Model fits were compared using AIC and BIC indices; the smaller the indices obtained, the better the model fit. R² was determined for all models, and Cohen’s f² (Selya et al., 2012) calculated the effect size of the significant predictors. The significance level was set to p < 0.05.

Sample-size justification. To account for clustering (162 teachers in 22 schools; ICC = 0.16; m ≈ 7.36), we computed a design effect of 2.02, yielding an effective N ≈ 80. An a priori G*Power 3.1 power analysis (Faul et al., 2009) for linear multiple regression (fixed model, R² increase; α = 0.05; 1 focal predictor; 6 total predictors) indicated N_min = 27 for a large effect (f² = 0.33) and 55 for a medium effect (f² = 0.15). Thus, our sample is well-powered for medium-to-large effects. Full post hoc power details are provided in Appendix B.

4. Results

4.1. Psychometric Properties of the CTBS

The KMO measure and Bartlett’s test of sphericity showed high strength in the relationships among items (KMO = 0.937; χ² = 1980.44, p < 0.001), indicating appropriateness to perform a factor analysis with the present data. As shown in

Table 1. Fit indices derived from the CFAs for the Spanish CTBS.

Model	χ2	p	df	RMSEA (90% CI)	SRMR	CFI	TLI
One-factor	343.622	0.001	54	0.168 (0.151–0.185)	0.058	0.854	0.821
Two-factor	269.481	0.001	53	0.147 (0.130–0.164)	0.054	0.891	0.864

Note. RMSEA = root mean square error of approximation; SRMR = standardized root mean square residual; CFI = comparative fit index; TLI = Tucker–Lewis index.

, the two-factor model represented a better approximation than the one-factor model in terms of fit and goodness. Nonetheless, χ²/df ratio (5.08) and RMSEA value (0.147) were higher than expected, whereas CFI (0.891) and TLI (0.864) values were just below the critical threshold.

The factor structure of the two-factor model is shown in Figure 2. All items showed an excellent factor loading across their target factor, oscillating between 0.70 and 0.90 (p < 0.001). The correlation between factors was strong and positive (0.88).

The Cronbach’s alpha coefficient of the Spanish CTBS was 0.95, whereas the reliabilities of its subscales were α_CIS = 0.94 and α_CSD = 0.91. Thus, the internal consistency of the overall CTBS and its subscales were good. As shown in

Table 2. Descriptive statistics for the Spanish and original versions of the CTBS and their subscales.

	Spanish CTBS				Original CTBS (Tschannen-Moran & Barr, 2004)
	Mean	SD	Median	IQR	Mean	SD
CTE	8.09	0.38	8.18	0.53	7.10	0.46
CIS	8.15	0.41	8.23	0.5	7.10	0.44
CSD	8.03	0.37	8.11	0.64	7.13	0.52

Note. In this first stage of the study: N = 25 schools (Sample 1). CTBS = Collective Teacher Beliefs Scales; SD = standard deviation; IQR = interquartile range; CTE = collective teacher efficacy; CIS = collective instructional strategies; CSD = collective student discipline.

, our sample of Mexican teachers from private schools seems to have higher CTE beliefs than the sample from the original study (Tschannen-Moran & Barr, 2004). The Spanish version of the CTBS is presented in Appendix C.

4.2. Descriptive Statistics and Correlation Analyses2

The descriptive statistics, reliabilities, and the results from the Shapiro–Wilk test are summarized in

Table 3. Descriptive statistics, results from the Shapiro–Wilk tests and reliabilities of the scales measured in sample 2.

	Mean	SD	Median	IQR	Skewness	Kurtosis	Shapiro–Wilk		α
							W	p
Teacher-level variables (N = 162)
Teacher self-efficacy	8.11	0.68	8.17	0.83	−1.3	6.08	0.89	0.001	0.91
Satisfaction with classroom autonomy	3.44	0.51	3.6	0.6	−1.27	4.64	0.9	0.001	0.77
Collaboration	3.86	0.8	3.88	1.13	0.46	2.9	0.98	0.02	0.80
Job satisfaction	3.67	0.4	3.88	0.5	−1.6	5.54	0.83	0.001	0.79
School-level variables (N = 22)
Collective teacher efficacy	8.16	0.35	8.3	0.37	−0.72	2.63	0.93	0.001	0.95
Participation among stakeholders	3.24	0.27	3.26	0.35	−0.56	2.55	0.96	0.001	0.85
Instructional leadership	3.46	0.57	3.67	1	−0.7	2.17	0.96	0.001	0.83

Note. SD = standard deviation; IQR = interquartile range.

. The reliability of the scales was good to excellent ranging from 0.77 to 0.95. As shown in

Table 4. Correlations among study variables and scales measured in sample 2.

Variable	1	2	3	4	5	6	7	8	9	10
1. Teacher self-efficacy
2. Gender	−0.12
3. Subject taught	0.02	−0.01
4. School grade	−0.06	−0.08	−0.05
5. Teaching experience	0.14	−0.07	0.04	−0.002
6. Satisfaction with classroom autonomy	0.44 ***	−0.12	−0.03	0.08	0.05
7. Collaboration	0.38 ***	0.09	0.11	−0.25 **	0.07	0.05
8. Job satisfaction	0.21 **	0.08	−0.16 *	−0.03	0.02	0.24 **	0.22 **
9. Collective teacher efficacy	0.21 **	0.12	−0.04	0.05	−0.24 **	0.17 *	0.01	0.09
10. Participation among stakeholders	0.26 ***	−0.09	−0.03	−0.05	−0.27 ***	0.14	0.08	0.05	0.72 ***
11. Instructional leadership	−0.02	−0.28 ***	−0.05	0.03	0.1	−0.001	−0.06	0.005	−0.12	0.06

Note. * p < 0.05; ** p < 0.01; *** p < 0.001.

, Spearman’s correlation coefficients for the relationships between variables ranged from 0.01 to 0.44, with one exception: a strong correlation between CTE and stakeholder participation (r = 0.72; p < 0.001). Hence, special attention was given to these two variables during the HLM models due to potential multicollinearity concerns.

Table 5. Professional development—format and topics: Percentage of participating teachers and point biserial correlation with teacher self-efficacy.

Formats of PD	%	Correlation with TSE	PD Topics	%	Correlation with TSE
Courses/seminars attended in person	78.4	0.04	Knowledge and understanding of my subject	80.9	0.11
Online courses/seminars	76	0.1	Pedagogical competencies	84	0.24 **
Education conferences	50.6	0.24 **	Knowledge of the curriculum	64.8	0.18 *
Qualification program	17.3	0.05	Student assessment practices	75.9	0.23 **
Visits to other schools	20.4	0.12	ICT skills for teaching	85.2	0.11
Visits to business premises	12.3	0.1	Classroom management	72.8	0.12
Observation and coaching	63.6	0.19 *	School management	34.6	0.13
Participation in a network of teachers	31.5	0.17 *	Approaches to individualized learning	61.7	0.19 *
Reading professional literature	46.3	0.09	Teaching students with special needs	42.6	0.12
			Teaching in a multicultural setting	22.8	0.14
			Teaching cross-curricular skills	67.3	0.21 **
			Analysis of student assessments	73.5	0.18 *
			Teacher-parent cooperation	69.8	0.22 **
			Communicating with people from different cultures	15.4	0.27 ***

Note. PD = professional development; ICT = information communication technology; TSE = teacher self-efficacy. * p < 0.05; ** p < 0.01; *** p < 0.001.

illustrates the different formats and topics included in the PD activities as reported by the participating teachers, as well as the point-biserial correlation between TSE and each PD activity. The topics of the PD activities seem to have a closer relationship with TSE than the formats. Specifically, the strongest correlation was found between TSE and “Communicating with people from different cultures” (r = 0.27; p < 0.001). Surprisingly, only 15.43% of the participating teachers reported receiving PD on this topic.

4.3. Multilevel Analyses

HLM was used to examine the proportion of variance in TSE explained by teacher- and school-level variables. As shown in

Table 6. Unconditional model for teacher self-efficacy.

Fixed Effects		Random Effect
	Coefficient (SE)		Variance (SE)
Intercept	8.07 (0.08)	Between-schools	0.08 (0.05)
		Within-schools	0.39 (0.05)
ICC	0.16 (0.09)
AIC	335.11
BIC	344.37

Note. SE = standard error; ICC = intraclass correlation coefficient; AIC = Akaike Information Criterion; BIC = Bayesian Information Criterion.

, the unconditional model indicated that 16% of the total variance in TSE was due to differences between schools (ICC = 0.16), whereas 84% was within schools. Meaning that there is sufficient variance among groups to justify the use of HLM.

As shown in

Table 7. HLM models for teacher self-efficacy beliefs.

	Teacher Self-Efficacy Beliefs
	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
Teacher-level predictors
Gender (female)	−0.23 (0.14)
Subject taught (Math)	−0.02 (0.10)
School grade
5th grade	−0.16 (0.15)
6th grade	−0.04 (0.16)
7th grade	−0.37 (0.16) *
8th grade	−0.07 (0.15)
Teaching experience	0.01 (0.01)
PD formats
Courses		0.04 (0.13)
Online courses		0.16 (0.13)
Conferences		0.18 (0.12)
Qualification program		−0.01 (0.14)
School visit		0.15 (0.13)
Business visit		0.07 (0.17)
Coaching		0.21 (0.11)
Network		0.13 (0.11)
Literature		0.08 (0.11)
PD topics
Knowledge			−0.11 (0.16)
Competencies			0.26 (0.16)
Curriculum			−0.01 (0.14)
Assessment			0.11 (0.15)
ICT skills			0.11 (0.16)
Classroom management			−0.15 (0.14)
School management			0.1 (0.12)
Individualized learning			0.04 (0.13)
Special needs			0.06 (0.12)
Multicultural settings			0.00 (0.13)
Cross-curricular skills			0.03 (0.14)
Assessment use			0.09 (0.15)
Parents			0.14 (0.13)
Multicultural communication			0.36 (0.15) *
Classroom autonomy				0.55 (0.09) ***
Collaboration				0.29 (0.06) ***
Job satisfaction				0.05 (0.12)
School-level predictors
Collective teacher efficacy					0.61 (0.18) ***
Participation stakeholders						0.81 (0.25) ***
Instructional leadership					0.05 (0.12)	−0.01 (0.12)
R2	0.08	0.14	0.17	0.37	0.01	0.01
Wald χ2 (df)	10.65 (7)	20.63 (9)	27.99 (14)	85 (3)	10.76 (2)	10.64 (2)
p-value	0.15	0.01	0.01	0.001	0.005	0.005
Between-school variance	0.08 (0.05)	0.09 (0.05)	0.07 (0.04)	0.07 (0.04)	0.04 (0.03)	0.04 (0.03)
Within-school variance	0.38 (0.05)	0.36 (0.04)	0.36 (0.05)	0.25 (0.03)	0.39 (0.05)	0.39 (0.05)
ICC	0.18 (0.09)	0.20 (0.09)	0.17 (0.09)	0.21 (0.09)	0.09 (0.07)	0.10 (0.07)
AIC	360.62	354.39	366.94	281.81	333.27	332.72
BIC	391.50	391.44	419.43	300.34	348.71	348.15

Note. PD = professional development; ICT = information communication technology; ICC = intraclass correlation coefficient; AIC = Akaike Information Criterion; BIC = Bayesian Information Criterion. * p < 0.05; *** p < 0.001.

, the partially conditional models with teacher-level variables—i.e., Models 1 to 4—revealed the following results: the demographic variables (Model 1) had no significant impact on TSE. Regarding the PD activities in which teachers participated during the last 12 months (Model 2), none of the PD formats predicted TSE. With respect to the different topics covered in those PD activities (Model 3), only “Communicating with people from different cultures or countries” (i.e., multicultural communication) had a significant impact on TSE. Classroom autonomy and collaboration among teachers were also significant predictors of TSE beliefs (Model 4). According to Models 5 and 6, CTE and participation among stakeholders were the two school-level variables that significantly predicted TSE.

Considering these findings, two fully conditional models were tested to find out which set of variables better explains TSE beliefs regarding our Mexican sample. These models included only the significant predictors of TSE (see

Table 8. Fully conditional models for teacher self-efficacy beliefs.

	Teacher Self-Efficacy Beliefs
	Model 7	Effect Size	Model 8	Effect Size
Teacher-level predictors
Multicultural communication	0.28 (0.12) *	0.03	0.27 (0.12) *	0.03
Classroom autonomy	0.54 (0.08) ***	0.30	0.54 (0.08) ***	0.30
Collaboration	0.25 (0.05) ***	0.15	0.24 (0.06) ***	0.15
School-level predictors
Collective teacher efficacy	0.46 (0.16) **	0.003
Participation stakeholders			0.55 (0.23) *	0.001
R2	0.38		0.39
Wald χ2 (df)	103.69 (4)		100.90 (4)
p-value	0.001		0.001
Between-school variance	0.04 (0.03)		0.05 (0.03)
Within-school variance	0.25 (0.03)		0.25 (0.03)
ICC	0.13 (0.08)		0.16 (0.09)
AIC	273.20		274.12
BIC	294.81		295.73

Note. ICC = intraclass correlation coefficient; AIC = Akaike Information Criterion; BIC = Bayesian Information Criterion. * p < 0.05; ** p < 0.01; *** p < 0.001.

). Due to the strong relationship between CTE and participation among stakeholders, we introduced both variables in separate models to prevent multicollinearity effects. Model 7 displayed the smallest indices for both AIC (273.2) and BIC (294.81), indicating a better fit to the data. Likewise, Model 7 showed the lowest ICC value (0.13), meaning that the clustering effect on TSE is reduced by the variables included in this model. However, the predictors included in Model 8 explained 39% of the variance in TSE beliefs, whereas the ones included in Model 7 predicted 38% of the variance.

Regarding the effect size of the significant variables that influenced TSE beliefs, classroom autonomy (f² = 0.30) had the most substantial impact, followed by teacher collaboration (f² = 0.15). These two medium effect sizes were succeeded by the small effect size of multicultural communication (f² = 0.03) on TSE. Surprisingly, CTE (f² = 0.003) from Model 7 and stakeholder participation (f² = 0.001) from Model 8 exhibited trivial effects on TSE beliefs.

5. Discussion

Although TSE and CTE beliefs have been widely studied, they remain as unexplored constructs in Mexico and most Latin American countries. The present research sought to examine these two different but related variables by studying TSE and its relationship with different teacher- and school-level variables with a sample of Mexican teachers.

In order to consider CTE as a predictor of TSE, the CTBS (Tschannen-Moran & Barr, 2004) was adapted into Spanish, and its psychometric properties were explored. Based on our results, we may say the Spanish CTBS is a reliable instrument that allowed us to assess valid evidence associated with CTE beliefs in the Mexican context. Regarding our first research question, CFA results indicate the Spanish CTBS is composed of 12 items grouped into two factors—instructional strategies and student discipline—as originally proposed by Tschannen-Moran and Barr (2004). Different goodness-of-fit measures further confirmed this finding: the two-factor model showed better-fit quality than the one-factor model. However, not all reported measures of goodness-of-fit were deemed acceptable (i.e., χ²/df ratio = 5.08; RMSEA = 0.147). Klassen et al. (2008) found similar results with a sample of Canadian (χ²/df = 4.3; RMSEA = 0.11; CFI = 0.88) and Singaporean (χ²/df = 5.7; RMSEA = 0.14; CFI = 0.89) teachers; however, they opted to allow different correlations among error covariances to obtain better goodness-of-fit. In our case, the discrepancy between the observed and expected values regarding χ²/df ratio and RMSEA might be due to the small sample size. These two types of global fit indices are sensitive to sample size; hence, a small sample may lead to false model rejections (Alavi et al., 2020; Hu & Bentler, 1999).

A surprising result was the relatively high levels of CTE obtained by the participating Mexican teachers. The means reported by the original study ranged from 7.10 to 7.13 (Tschannen-Moran & Barr, 2004); however, in our study, means and medians ranged from 8.03 to 8.23 on a scale rated from one to nine. These high scores might indicate a possible ceiling effect. This suggests that the participants of this study, namely, Mexican teachers from private schools, have a greater perception of CTE beliefs in comparison with the original study. Salas-Rodriguez et al. (2021) found similar results regarding TSE beliefs in a similar population in Mexico. Hence, it might be useful to check the scores of both, TSE and CTE beliefs in a larger Mexican sample while keeping in mind the cultural context since a recent cross-cultural study found that teachers’ perceptions of their CTE beliefs may vary according to their cultural values (Da’as et al., 2021).

Regarding our second research question, multicultural communication, classroom autonomy, and collaboration were the teacher-level variables that significantly predicted TSE beliefs. Although overall PD—format and topics—did not seem to predict TSE, participation in activities focused on “Communicating with people from different cultures or countries” significantly predicted TSE. In other words, Mexican teachers who had learned to communicate with people from diverse cultures and countries had higher levels of TSE. This result is in line with what Mexican teachers—and teachers from 40 other countries—reported in TALIS 2018: when teachers participated in at least one training activity focused on multiculturalism, they tended to present higher levels of self-efficacy (Choi & Mao, 2021; OECD, 2019b). These findings might reflect what Mexican teachers value and need to feel more competent when responding to current educational challenges. In this regard, PD focused on teaching in multicultural settings may not only enhance teachers’ competence but also better address the needs of students and families.

Classroom autonomy was the predictor with the greatest impact on TSE beliefs, showing the largest effect size in this sample. This indicates that teachers who perceived they had greater control over determining course content and the amount of homework assigned, as well as selecting teaching methods, assessing students’ learning, and disciplining students, reported significantly higher levels of TSE. In other words, teachers with more control over their instruction tend to feel more confident teaching (OECD, 2020). These findings may be particularly relevant, given that previous studies have found TSE acts as a mediator between teacher autonomy and teaching practice (De Neve et al., 2015). For example, Valckx et al. (2020) found that TSE fully mediated the relationship between teacher autonomy and reflective dialog, and between teacher autonomy and collective responsibility. Thus, autonomous teachers reported higher self-efficacy and willingness to collaborate with colleagues and share responsibility.

Another variable that significantly impacted TSE was teacher collaboration. This finding is in congruence with previous studies which suggest that when teachers engage in collaborative activities they tend to present higher levels of TSE (Çoban et al., 2020; Duyar et al., 2013; OECD, 2020; Sehgal et al., 2017). In this sense, teacher collaboration might promote the development of TSE beliefs by fostering two of the four sources of efficacy: vicarious experience and verbal persuasion. For example, when teachers teach jointly or engage in shared activities, they can learn by observing others’ practices. Likewise, when teachers provide positive feedback to a peer, they may influence his or her levels of TSE (Prilop et al., 2021).

On the other hand, demographic variables such as teachers’ gender, school grade, teaching experience, and subject taught were neither significantly related nor predicted TSE variability. Although previous studies have found different results (Gümüş & Bellibaş, 2021; OECD, 2020), our findings follow what Tschannen-Moran and Woolfolk Hoy (2007) stated: demographic variables are not strong predictors of teacher efficacy beliefs.

As for teachers’ JS, it did not significantly impact TSE beliefs. In other words, teachers who are satisfied with their jobs did not necessarily have higher levels of TSE in this sample. Further research is needed to examine this challenging result in the Mexican context. Perhaps studying both constructs by considering their different subscales might help to better understand the association between these two traditionally related variables. Moreover, longitudinal studies are strongly encouraged since previous research has treated TSE as an antecedent of JS by inferring this assumption through correlational data (Zee & Koomen, 2016). However, a recent longitudinal study by Burić and Kim (2021) found JS predicted TSE and not vice versa.

Regarding our third research question, CTE and participation among stakeholders were both significant predictors of TSE. This means that teachers from schools with higher levels of CTE tend to report higher TSE beliefs. Thus, CTE beliefs affect how teachers perceive themselves as more or less competent. Studies such as R. D. Goddard and Goddard (2001) and Skaalvik and Skaalvik (2007) have found similar results. A possible explanation for these findings might be that CTE beliefs help create a shared culture of perseverance and commitment among the faculty, thus encouraging teachers to excel individually and collectively (R. D. Goddard et al., 2004a). CTE is best understood as an emergent, group-level phenomenon rather than a mere aggregate of individual beliefs (Bandura, 1997; R. D. Goddard et al., 2000). Under this framework, idiosyncratic low scores are treated as measurement error insofar as the school’s shared confidence in its collective capacity benefits all members, regardless of their personal stance. If CTE were purely individual (belief in colleagues’ abilities), discrepancies would translate into differential outcomes for low-efficacy teachers. Stapleton et al. (2016) caution that configural constructs require explicit cross-level invariance testing and may benefit from random-slope or interaction models to reflect variability in how individuals “tap into” collective beliefs.

Similarly, participation among stakeholders significantly predicted TSE beliefs. This result suggests that when the faculty members engage in shared decision-making and a collaborative school culture is present, teachers feel more competent in their teaching tasks. Previous studies have used stakeholder participation to assess distributed leadership from teachers’ perspectives and have reported similar findings: when teachers have a greater voice in school decision-making, they tend to report higher levels of TSE (J. Liu et al., 2023; OECD, 2020; Sun & Xia, 2018).

Lastly, instructional leadership was neither related to TSE, nor significantly predicted these beliefs. This finding challenges previous studies that assured instructional leadership promotes TSE by creating a learning environment among teachers (R. D. Goddard et al., 2015). In fact, some researchers sustained that instructional leadership has a greater impact on TSE than distributed leadership (Y. Liu et al., 2020). A possible explanation for our finding is that instructional leadership was the only data collected from the principals, whereas distributed leadership (i.e., participation among stakeholders) and the rest of the variables belong to data collected from teachers. Hence, instructional leadership data might be biased since principals and teachers tend to have different perceptions regarding responsibilities and tasks (OECD, 2020).

Regarding our fourth and last research question, Models 7 and 8 seem to include the set of variables that better explain TSE beliefs within our Mexican sample, i.e., multicultural communication, classroom autonomy, and collaboration—at the teacher level—and CTE and participation among stakeholders—at the school level. In our study, these two school-level variables seem closely related and presented similar effects on TSE beliefs. Thus, further research is needed to better understand how CTE and participation among stakeholders influence and impact one another.

5.1. Implications

Our findings have important implications for policymakers and educational practitioners. As shown in this study, teachers present higher levels of self-efficacy when they can engage in decision-making both in their classroom (i.e., classroom autonomy) and at their school (i.e., participation among stakeholders). One implication for policymakers and principals could be to give teachers more of a say in their professional practice. When teachers perceive they are trusted and feel a sense of ownership over their teaching practice, they have more opportunities to grow their expertise and contribute to the teaching profession.

Regarding PD, policymakers and school leaders should pay special attention to the multiculturalism topic. Our findings and those of TALIS 2018 (OECD, 2019b) show that only a small percentage of Mexican and Latin American teachers have received PD focused on teaching in multicultural environments and communicating with people from different cultures. Surprisingly, this small percentage of teachers also reported higher levels of self-efficacy, thus receiving training on multicultural education is helping teachers deal with more diverse classrooms and develop their own competence beliefs.

Finally, if we reflect on the variables that significantly impacted TSE in the present study—namely, classroom autonomy, collaboration, CTE, and participation among stakeholders—we can see that they are all different pieces of the school culture puzzle. In this sense, promoting teacher efficacy beliefs goes hand in hand with the school culture since teachers feel more competent when they work at schools characterized by a sense of trust, collaboration, participation, and shared responsibility.

5.2. Limitations and Further Research

Despite the contributions, there are a number of limitations that should be taken into account. First, this study is based on a small sample of Mexican private schools; hence, the findings of this research should be considered in the context of these characteristics and caution should be taken when generalizing our findings to other educational contexts. Further research should be conducted to better understand how self- and collective teacher efficacy behave in the various Mexican contexts, including teachers from public schools.

Second, the study was based on a convenience sample, and not all members of each school community participated in the online survey. Therefore, school-level variables may not represent the entire faculty, and the sample may be biased toward those schools and teachers with more open dispositions. Future studies based on more randomly selected samples are needed to further explore our findings in the Spanish-speaking context. Moreover, qualitative or mixed method studies might also help expand our understanding of how self- and collective teacher efficacy develop and relate to the school context. Our Level-2 reliability index (ICC) rests on the assumption that participating teachers constitute a random sample of each faculty. In practice, we employed convenience sampling and surveyed only those present on the data-collection day, which may introduce self-selection bias. As a result, our reliability estimates may not fully capture systematic sources of error such as social-desirability effects or contextual influences beyond sampling variance. A post hoc power analysis (Appendix B) using our observed R²s (small-effect range 0.02–0.07) indicates limited power (<0.50) to detect the smallest effects, though ample power (>0.80) for moderate to large effects. Future studies should increase sample size or employ longitudinal designs to enhance detection of subtle predictor-outcome relationships.

Third, the present study is based on cross-sectional data, and no causal conclusions can be drawn from it. Longitudinal studies are recommended to explore the causal ordering of the relationships between teacher efficacy beliefs and teacher- and school-level variables. This type of study would also allow us to examine the reciprocal causation between self- and collective teacher efficacy, as well as between these beliefs and JS and teacher collaboration.

Fourth, a limited amount of teacher- and school-level variables were included in this study. Hence, future research should include variables such as student achievement, instructional practices, and socioeconomic status. Specifically, more longitudinal studies are needed to explore the causal association between teacher efficacy beliefs and student achievement.

Fifth, although our two-factor CFA yielded acceptable loadings and reliability, some fit indices fell outside recommended thresholds (χ²/df = 5.08; RMSEA = 0.147) (Hu & Bentler, 1999). Similar discrepancies have been reported by Knickenberg et al. (2025), who observed RMSEA = 0.09 in a strict CFA of their inclusive-practice collective efficacy scale. By moving to an ESEM framework, they recovered meaningful cross-loadings and achieved acceptable fit without compromising interpretability. In line with this approach, future CTBS validations should test ESEM alongside CFA or partial-invariance models to allow for cross-loadings and better capture the configuration of shared teacher beliefs.

Regarding the Spanish CTBS, Spanish and Mexican experts have ensured the neutrality of the language used, thus researchers are furthermore encouraged to use this CTBS version in different Spanish-speaking countries. Perhaps the response set might be shortened from a 9-point to a 5-point scale to avoid overestimation and facilitate comparisons with results from other countries and cultures (Da’as et al., 2021). We believe the Spanish CTBS will boost studies regarding CTE among Spanish-speaking countries, ameliorating the lack of studies on this topic in the Latin American context. When studying CTE beliefs, researchers should take into account its aggregated nature in order to avoid falling into fallacies of inference, namely, ecological and atomistic (Chan, 2006). Thus, we strongly encouraged studies with a large sample both at the individual and the group level.

Finally, building on the importance of aggregation, it is also critical to reflect on how that aggregation is performed. Although we operationalized Collective Teacher Efficacy by aggregating individual CTBS scores (Level 1) into school-level means, recent work warns that emergent group constructs may not be fully captured by simple averages. Jak and Jørgensen (2017) demonstrate how testing cross-level measurement invariance and multilevel reliability indices can ensure that items function equivalently within and between clusters, validating that aggregated scores reflect a true school-level phenomenon. Stapleton et al. (2016) further argue that the ‘meaning’ of a construct can shift when moved from individuals to groups, suggesting that arithmetic means may overlook configuration-based dynamics inherent in shared beliefs. Although our ICC (0.16) and design-effect adjusted power analysis support the statistical justification for aggregation, future research should explicitly evaluate cross-level invariance and consider configurational or network-based methods to capture the interplay of individual perceptions in forming collective teacher beliefs.

6. Conclusions

The purpose of this study was to examine which teacher- and school-level variables predict TSE beliefs in a sample of Mexican teachers from private schools. First, the CTBS (Tschannen-Moran & Barr, 2004) was adapted into Spanish due to the absence of a valid and reliable instrument to assess CTE among Spanish-speaking teachers. Then, its psychometric properties were explored and internal validity evidence was obtained, thus supporting the use of the Spanish CTBS among Mexican teachers. Next, multilevel analyses were performed and results revealed teacher—i.e., multicultural communication, classroom autonomy, and collaboration—and school—namely, CTE and participation among stakeholders—level variables significantly predicted TSE beliefs in this Mexican sample. Out of these significant predictors, classroom autonomy (f²= 0.30) and collaboration among teachers (f²= 0.15) presented the greatest impact on TSE. Finally, since this is the first study to explore both teacher self- and collective efficacy beliefs in the Mexican context, more studies are needed to further understand how these variables behave in this Spanish-speaking country.