In recent years there has been a growing interest in research on teacher self-efficacy (TSE), defined as an individual teacher’s perception of their capability to accomplish the tasks required to be an effective teacher (Kleinsasser, 2014; Mok and Moore, 2019; Pawlak, 2022; Wyatt, 2022). TSE has been associated with improvements in teacher effectiveness (Boulden et al., 2021; Noorollahi, 2021) and reductions in teacher attrition (Klassen and Tze, 2014). TSE could also significantly predict students’ learning motivation, learning enjoyment and academic performance across varying school environments (Fackler and Malmberg, 2016; Hettinger et al., 2021).

Research in this field has shown significant advancements, including in the measurement of TSE; however, certain methodological limitations exist (see Zee and Koomen, 2016; Ma et al., 2021b). One predominant characteristic of the commonly applied TSE scales (e.g., Teacher Sense of Efficacy Scale [TSES], Tschannen-Moran and Hoy, 2001) has been a focus on classroom teaching (Bin Khairani and Bin Abd Razak, 2012). Such measures might be incapable of covering all essential domains of being a teacher (Avanzi et al., 2013; Karbasi and Samani, 2016) or even teaching. For example, integrating technology into teaching (Gomez et al., 2022) and enhancing democratic cooperation among students (Wheatley, 2005), which could truncate TSE’s predictability (Bandura, 2012). Recent research has also identified certain deficiencies in validating procedures used in the most commonly applied TSES. For instance, the usage of orthogonal rotation when the subdomains were correlated (Ma et al., 2020) and reliance on eigenvalues and scree plots instead of methods like parallel analysis that could elicit more accurate estimation of the number of factors (Lim and Jahng, 2019). These might have inflated the number of factors, which has advocated for rigorous research on TSE measurement (Koniewski, 2019; Salas-Rodríguez et al., 2021). Furthermore, the call for research from a more international perspective has been made (Klassen and Durksen, 2014). Research in the measurement of TSE in the Chinese context is limited (Ma et al., 2020; Yin et al., 2020); therefore, the current study aims to validate a comprehensive TSE scale among Chinese pre-service teachers, incorporating recent advancements in TSE measurement.

Although Bandura’s social cognitive theory has guided research on TSE (Pfitzner-Eden, 2016), questions have been raised about the theoretical accuracy, item identification, and specificity of TSE measurement (Chesnut and Burley, 2015; Kim et al., 2019).

Self-efficacy beliefs reflect an individual’s future-oriented anticipation of their capability to accomplish specific tasks (Bandura, 1993). It differs from constructs like confidence, with the latter being a generalized trait and the former being context-specific (Bandura, 2012). The wording or expression of self-efficacy items, therefore, should be specific regarding the tasks to be accomplished and include wording such as “able” and “capable” (Bandura, 2012). However, it is common for items in TSE measures to contain wording like “I feel confident” (Pfitzner-Eden, 2016), which unnecessarily conflate the differing concept of confidence with the “can do” notion of capability (Bandura, 2006). Meanwhile, the necessity to differentiate self-efficacy and outcome expectancy has been raised with the latter being less predictive in human behaviors (Bandura, 2012). Concerns have been raised about the predictability of TSE items without considering beliefs in the effectiveness of such behaviors (Kim and Kim, 2010; Chesnut and Burley, 2015) as individuals might not be motivated without considering individual beliefs about the outcome (Bai et al., 2022). However, such opinions might have failed to resonate with the suggestions of Bandura (2012) to not confound the measure of self-efficacy with related effects.

The most commonly applied TSE scale is the TSES (Tschannen-Moran and Hoy, 2001), which emphasises the measurement of TSE in three classroom teaching domains: maintaining classroom discipline, involving students in teaching, and employing various instructional strategies (Perera et al., 2019). The domain specificity was intended to increase the predictability of the construct and increase the practical applicability of research findings to the specific contexts of teachers’ work (Avanzi et al., 2013). However, individuals’ perception of their capability varies across different domains and dimensions within each domain (Yang et al., 2022); hence, if TSE is measured in too narrow a domain, educators may be deterred from applying it in practice (Bandura, 2012). As noted by Bandura (2012), individuals tend to evaluate their self-efficacy towards specific tasks simultaneously with an assessment of their general ability. For example, teachers might hold a broader general view of their capability as a teacher in the school context, alongside their self-efficacy for specific tasks. These theoretical assumptions have been confirmed by variations in TSE for subscales and the integrated information on the higher-order overall TSE (Perera et al., 2019).

Much evidence on the three-domain TSE structure, mainly based on the TSES, has been confirmed (Valls et al., 2020); further inspections appear to be needed. On the one hand, regarding the statistical techniques, the orthogonal rotation has been utilized, assuming factors were uncorrelated, in factor structure studies, whereas the correlations between TSES subscales range from 0.58 to 0.70 and from 0.95 to 0.98 for the long and short formats, respectively. This rotation method may have enlarged the number of factors retained (see Burgueño et al., 2019; Perera et al., 2019; Ma et al., 2020). The correlations between these subdomains, calculated in subsequent studies, average to be larger than 0.6 (Koniewski, 2019; Salas-Rodríguez et al., 2021). Then, both eigenvalues and scree plots have been predominantly applied to decide the number of factors; however, parallel analysis has been demonstrated to outperform these traditional methods in producing a more accurate factorial structure (Preacher and MacCallum, 2003). Such an advantage of parallel analysis was based on comparing the eigenvalues of the sample with those received from a random correlation matrix without assuming preliminary factor numbers (Lim and Jahng, 2019). However, such a method has rarely been applied in TSE research since the call was made over two decades ago (Henson, 2002). Furthermore, none of the validation studies of the TSES reviewed by Koniewski (2019) reported sufficient chi-square and the average of RMSEA was 0.077 [0.053, 0.134], which slightly exceeded the cut-off values. On the other hand, in terms of research involving pre-service teachers (PSTs), the use of one overall factor has been recommended (Tschannen-Moran and Hoy, 2001) as there is insufficient evidence of the factor structure being relevant in this population, yet the three-domain structure has been applied among PSTs (e.g., Burgueño et al., 2019). The failure to confirm the three-factor structure could explain findings of PSTs’ incapability to comprehend teaching complexity (Fives and Buehl, 2009; Duffin et al., 2012).

Nonetheless, questions have emerged about the nature and assessment of TSE in domains beyond the classroom walls. Cherniss (1993) proposed a framework combining teachers’ capabilities to perform instruction, disciplining, and student engagement within the classroom, establishing interpersonal connections with other professionals and making an impact on the function of the school. Friedman and Kass (2002) commented that domains other than classroom teaching are “undisclosed” and include the ability to socialize with school staff and students, with similar findings in a study conducted by Skaalvik and Skaalvik (2007). Bandura (2006) initiated seven dimensions of TSE: influencing school administrative decisions, acquiring school resources, instruction, discipline maintenance, engaging parents, involvement with communities, and building a positive school climate. Including domains other than classroom teaching is especially meaningful as, according to Bandura, TSE measurements should be “broad in scope and domain specificity” (2012, p. 17). It is also essential to emphasize such a line of research in initial teacher education, because using TSE scales with a narrow scope on classroom teaching not only fails to ensure the predictability of TSE, but is also not able to examine PSTs’ TSE for tasks other than classroom teaching (Cocca and Cocca, 2021), which were reported to be the main reason for ‘reality shock’ once PSTs transfer to their first year of teaching (Ma et al., 2021a).

Chinese teachers, including PSTs, have been found to have a perception of teaching responsibilities differing from their western counterparts; for instance, their preferences for teacher-oriented instruction, exercising parental authenticity, and ethnical care for students (Lin and Gorrell, 2009; Liu et al., 2022). Such differences might be rooted in the Chinese paternalism perspective of teacher morality that addresses teacher authority and moral guidance, together with teaching competence (Ho and Hau, 2004; Ye and Zhou, 2020; Kutuk et al., 2022). Consider the Chinese teacher creed, for instance, Great learning makes a teacher and moral integrity makes a model. Chinese PSTs have easier access to parental assistance than PSTs in the US and do not consider receiving help from parents indicates lower TSE (Cheung, 2006). Meanwhile, Chinese teachers influenced by collectivistic cultures tend not to be as positive about their teaching capability, compared to those from individualistic backgrounds (Vieluf et al., 2013), which has been consistently reported in Asian countries (Ruan et al., 2015). However, the likelihood of teachers to rate their TSE lower in collectivist countries, especially in Asia, could, paradoxically, exhibit more effective teaching strategies in their practice (Vieluf et al., 2013).

Research in the Chinese context has either developed TSE scales initially based on teachers’ responsibilities or they have been adapted from an existing scale (e.g., TSES). In a Hong Kong sample, Chan (2008a) selected 24 items to measure six pre-determined domains: teaching highly capable students, classroom management, providing guidance and counselling, student engagement, teaching in diversified contexts, and teaching for creativity, critical thinking, and problem-oriented teaching. However, the original model fit for confirmatory factor analysis (CFA) was not satisfactory, before the number of items was reduced to 18. In a sequential study, Chan (2008b) added three items to capture interactions with colleagues and parents to make a seven-domain scale, despite the recommendation that there should be at least four items per factor (Fabrigar et al., 1999). It appears unclear whether the seven factors could converge to a high-order factor, due to the high correlations between the different factors of guidance and counselling and student engagement (Chan, 2008b). In both studies, data from both in-service teachers (ISTs) and PSTs were combined, despite the possibility that the factor structure might differ for each.

The TSES does not appear to elicit a consistent factor structure in the Chinese context. Cheung (2008) found one overall factorial structure among the whole sample and confirmed a two-factorial structure among the male group, namely efficacy for student engagement and efficacy for student disciplining. The two factors consisted of items mixed from different original TSES factors. Tsui and Kennedy (2009) confirmed a two-factor structure, efficacy in teaching and assisting students’ learning, by mixing items for instructional strategies, student engagement, and classroom management. Ruan et al. (2015) validated the TSES in China, Japan, and Korea, and found the 24-item version failed to produce a satisfactory model fit and the 12-item version only received an acceptable model fit after being reduced to 11 items, with modification index fixation being applied. Ma et al. (2020) confirmed a unified one-factorial structure among in-service and PSTs, after making adaptations to item wording and response options; however, we raised concerns that the short form of the TSES might be limited in domain breath.

Many improvements, as noted above, have been achieved in TSE measurement, however, certain challenges exist including the validation procedure and domain comprehensiveness. The latter is especially essential considering the wholistic view of teacher professional development which included tasks both within and external classroom teaching. Furthermore, limited agreements have been achieved in validating a solid TSE measurement, especially considering the inconsistencies in the factorial structure of TSES, in Chinese context. Therefore, this study aims to develop a TSE scale to use with Chinese PSTs by incorporating factor analysis methodological recommendations and broader domains of teaching.

The current study aimed to validate a comprehensive TSE scale using an initial 40 items among Chinese PSTs. It confirmed an overarching TSE factor with two subscales, named as Ethos and Teaching. The study also found significant moderate and low correlations between TSE and another two constructs, namely general self-efficacy and agreeableness, respectively. In addition, it found highly efficacious PSTs tended to report higher effectiveness regarding their professional experience and stronger intention to become teachers.

The new TSE scale had a two-factor structure, namely Ethos and Teaching. This finding is promising, following the call for a TSE scale that is broader and covers essential domains of being a teacher not only within the classroom but also within the school context (Wheatley, 2005; Cocca and Cocca, 2021). The existence of two factors, especially the Ethos factor, is supported by the low proportion of missing data and infrequent use of the no idea and not applicable response options, suggesting the high relevance of the 40 purposely selected items as revealed in the preliminary relevance rating. The distinct factorial structure could also have benefited by applying the response scale with labels on odd numbers, which was reported to produce better data quality and psychometric results by assisting PSTs to rate TSE more accurately (Trevethan and Ma, 2021). Meanwhile, this structure was supported because, first, two distinct factors were indicated in the parallel analysis; second, there was a totally distinct pattern of loadings on each domain in the EFA; third, the two-factor model fit in the CFA had a high degree of acceptability, and fourth, each factor comprised a meaningful cluster of items that, as noted above, corresponded with the two-factor solutions theorized and identified in other research (e.g., Friedman and Kass, 2002; Skaalvik and Skaalvik, 2007). However, to achieve the satisfied model fit indices in CFAs, this study allowed five pairs of neighboring items to be correlated. This might not be problematic, considering the large number of items increases the challenge of obtaining a satisfactory model fit in CFA (see Matsunaga, 2010; Smolkowski, 2020), particularly the root mean square error of approximation (RMSEA; Schermelleh-Engel et al., 2003). Especially, the model fits for both the two-factor structure and the second-order analysis, based on the 10 items with the highest loadings for each factor, were satisfactory without making any modifications. Such a short 20-item scale tapping broad domains of the teaching profession could be preferrable, considering the convenience it offers researchers. In addition, the model fits could have been more acceptable if extra items were removed, especially retaining only the five items with the highest factor loadings, which was the method used to extract the short form of the TSES (Tschannen-Moran and Hoy, 2001). However, by removing items to obtain more desirable goodness-of-fit indices, we would have jettisoned the main aim of this research, namely, to create a scale with wide content coverage, and we would have done so to satisfy goodness-of-fit rules of thumb that are increasingly being shown to have insubstantial foundations (Brown, 2015).

All items that were either from or bear similar meanings with those in the study of Bandura (2006) had sufficient factor loadings; however, they formed two broad factors with five domains external to classroom teaching, such as being involved with parents, integrated to be one holistic factor. The current factorial structure is more aligned with the personal and professional dimensional structure suggested by Friedman and Kass (2002). The two-factor structure, forming a higher-order structure, might have extended the seven-dimension structure of Chan (2008a,b), who failed to provide an examination of the higher-order structure that tapped domains related to both classroom teaching and activities at the school level. This is especially meaningful considering the high correlations between the two subdomains, namely student engagement and guidance and counselling students in the study of Chan (2008a,b), which might indicate the need to conduct a second-order factor analysis.

TSES items included in the present study divided into two factors. This also occurred with TSES data from ISTs in Hong Kong (Tsui and Kennedy, 2009) and Shanghai (Cheung, 2006), in which the TSES engagement and instruction items loaded on one factor (Teaching, in the current research) and the TSES management loaded items on the other (Ethos, in the current research). Of particular interest, items based on the TSES management factor were among the items with the lowest loadings on Ethos in our research, suggesting that participants in our study regarded management of students to be less salient than other aspects of teachers’ professional experiences and preoccupations. This pattern appears to confirm the finding that Chinese PSTs might have differentiated tasks other than classroom instruction with those associated with moral guidance, which might have given them more authority to correct students’ misbehaviors in, but not limited to, classroom teaching contexts (Ye and Zhou, 2020). This supports a need for considering cross-cultural differences in the foundations of TSE (see, for example, Ho and Hau, 2004).

The varying distributions of TSES items in the current study are also consistent with prior studies (Cheung, 2006) that confirmed a two-factor structure, namely, TSE for enhancing students’ learning and behavior management among Chinese participants. This is also consistent with Labone (2004) who addressed the necessity to extend the engagement domain within the classroom to broader social contexts, including parents and colleagues. Such a finding might be associated with a common understanding of being a teacher among Chinese teachers, including PSTs, that addresses the bond between teachers and students, such as the parenting and obedience relationship, as a separate domain than classroom teaching (Liu et al., 2022). This two-factor structure also aligns with another recent study that validated a revised TSES among both Chinese ISTs and PSTs that identified a one-factor structure, which, as assumed by the Ma et al. (2020), could be attributed to Chinese participants’ holistic view of teaching.

The correlations between general self-efficacy and the two subscales, Ethos and Teaching, were 0.41 and 0.39, lying within the span of 0.35 to 0.45 used by Pfitzner-Eden et al. (2014) to support convergent validity of their TSE scale. Meanwhile, the correlations of the two subscales of 0.25 with the agreeableness scale appeared to be higher than that reported in prior studies (e.g., |r| < 0.07 in Senler and Sungur-Vural, 2013; |r| < 0.14 in Bullock et al., 2015). Also, in absolute terms, these correlations are in the same region as the correlations of −0.31 and −0.21 claimed by Tschannen-Moran and Hoy (2001) as establishing discriminant validity for the TSES, in relation to work alienation and pupil control ideology. However, the influences of scales used for these different constructs on the correlation coefficients shall be taken into accounts. Further exploration of the research literature revealed that agreeableness might not have been ideal for assessing discriminant validity, because not all correlations between TSE and agreeableness are low. For example, Üstüner (2017) obtained a correlation of 0.37 between the same variables with PSTs in Turkey. In addition, known-group validity was well supported, with PSTs with higher TSE for both subdomains being more likely to report their professional practice being effective and a higher intention to enter the teaching profession.

The present study provides evidence of the validity of a comprehensive TSE scale, in both long and short formats, tapping a broad range of teachers’ responsibilities, not limited to classroom teaching. A second-order factor structure with two subscales appears to be sufficiently established, considering the multiple EFA procedures taken, especially parallel analysis and clear factor distribution with satisfactory factor loadings, acceptable model fits for both CFAs and second-order factor analysis (after consideration is given to the effects of larger numbers of items on RMSEA; see Schermelleh-Engel et al., 2003; Matsunaga, 2010), and confirmative results in both convergent and known-group validity. Discriminant validity might not be well supported according to a rigid cutoff; however, the correlations in the present study still fall within the criteria utilized in prior TSE validation studies.

Certain implications could be drawn from the present study. First, researchers could apply the current scale either wholly, as a composite TSE score calculated as confirmation of the second-order structure, or with the subscales used separately. The teaching subscale to determine where teachers’ main interests lie for classroom teaching activities and the Ethos subscale as a measure of a broader domain of teachers’ professional development. Second, teacher educators could try to motivate PSTs to choose a teaching career through assisting them to develop a capable self-image, not limited within classroom teaching domains. Third, differentiated training shall be conducted for PSTs to enhance their TSE for Ethos, perceived as a distinctive construct than that for classroom teaching.

Certain limitations shall be taken into account when researchers interpret the results. First, only PSTs were included in the study, so the validity of this scale with ISTs is unclear. Second, the test–retest reliability of the scale has not been examined, so caution should be taken using this scale to capture changes in TSE. Third, modification indices were employed in the original two-factor and higher-order analysis by allowing errors of several items to correlate. However, the initial model fits that were less satisfactory might have been associated with a large number of items. Fourth, results on the validity were solely reported using the long format of scale considering the necessity to keep the writing consistent and condensed, despite the tests we tested with the short scale producing even satisfactory results. Therefore, future research could be directed at validating the scale with other teacher populations, including in-service teachers, in other cultural contexts, to establish its cross-cultural validity and establish the test–retest reliability of the scale in longitudinal studies. Notwithstanding, the current study provided a new comprehensive TSE scale validated using a relatively rigid procedure, including a large sample size, parallel EFA analysis, multiple reliability indicators, and validity testing with a sound theoretical assumption. Thus, we believe that this scale might serve as a good foundation for further explorations on the source of and influencing factors to teachers’ self-efficacy.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving human participants were reviewed and approved by School of Education, Yancheng Teachers University. The patients/participants provided their written informed consent to participate in this study.

KM designed the research, conducted the analysis and completed the manuscript. JL methodology discussion, revised and proofread the manuscript. MC proofread the manuscript. JD collected the data and proofread the manuscript. MS revised and proofread the manuscript, and funding author. All authors contributed to the article and approved the submitted version.

This project is supported by the National Social Science Funding for Educational Research (CHA220297) and the Beijing Normal University Start-Up Funding (no. 310432119) and Jiangsu Provincial Key Constructive Laboratory for Big Data of Psychology and Cognitive Science (206110067).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.