It has been widely recognized that STEM (Science, Technology, Engineering, and Mathematics) education plays a crucial role in preparing students with scientific literacy and critical thinking skills necessary for success in a rapidly evolving world (English, 2017). STEM fields are key drivers of innovation and economic growth as investing in STEM education is vital for cultivating a highly skilled workforce that develop cutting-edge technologies, driving entrepreneurship, and maintaining a competitive edge in the global economy (Rockland et al., 2010). Teachers’ self-efficacy plays an essential role in teacher effectiveness (Bandura, 1997; Holzberger et al., 2013) that leads to optimal student learning outcomes such as attitudes towards STEM and academic achievement in STEM (Han et al., 2021). Research examining self-efficacy has primarily focused on teachers’ confidence in subject matter knowledge, pedagogical skills, or classroom management (e.g., Lee et al., 2019; Mobley, 2015). As the classrooms in the US become more and more diverse with students coming from diverse cultural backgrounds, ethnicities, languages, and abilities (Howard & Rodriguez-Minkoff, 2017), all teachers including STEM teachers need to navigate this multicultural landscape and ensure inclusive and equitable learning experiences for all students. Without a thorough understanding and confidence in their abilities to address the cultural and linguistic diversity, teachers may face challenges in fulfilling this goal.

Despite the significant value of teachers’ self-efficacy in multicultural classrooms (TSMC), research in this vein has been limited, with most studies focusing on teachers as a whole rather than specifically examining STEM teachers’ self-efficacy in multicultural educational settings (Choi & Lee, 2020; Choi & Mao, 2021; Grimberg & Gummer, 2013; Parkhouse et al., 2019). Generally, research has shown that when teachers have acquired high self-efficacy and implemented culturally relevant practices in their multicultural classrooms, students developed more positive interracial relationships (Banks, 1995b), the attendance rate and academic performance of at-risk high school students were also improved (Dee & Penner, 2017). Nevertheless, the relatively diminished self-efficacy in multicultural classrooms of mathematics and science teachers as identified by Choi and Mao (2021) highlights the specific challenge these teachers encounter in meeting the needs of a diverse student body in K-12 classrooms, while at the same time underscores a significant area for improvement in their teaching capacities (Howard & Rodriguez-Minkoff, 2017).

Given the identified differences and the limited research comparing STEM and non-STEM TSMC, it is imperative to examine whether the observed difference identified by Choi and Mao (2021) holds true for a broader range of STEM teachers. Additionally, it is crucial to identify the potential factors that could contribute to this disparity. Accordingly, this study aims to compare STEM and non-STEM TSMC and the associated factors that might contribute to this important construct. By comparing STEM and non-STEM TSMC and factors related to TSMC, researchers can gain better insights into the factors associated with instructional practices, classroom climate, and student engagement, which may subsequently help inform the design of teacher preparation and professional development programs to better equip STEM and non-STEM teachers for diverse classroom environments.

Theoretical and Empirical Perspectives

Teacher Self-Efficacy in Multicultural Classrooms

TSMC as a theoretical construct originated from two theories: self-efficacy (Bandura, 1997) and multicultural education (Banks, 2001; Gay, 2002; Ladson-Billings, 1995). According to Bandura’s social cognitive theory, self-efficacy refers to the confidence in one’s ability to plan and carry out the actions necessary to achieve specific goals (Bandura, 1997). In the teaching context, teacher self-efficacy refers to teachers’ confidence in their capacity to carry out the actions or behaviors required to successfully complete a particular educational goal (Guskey & Passaro, 1994; Tschannen-Moran & Hoy, 2001; Tschannen-Moran et al., 1998). Since the increasingly diverse classrooms in the US bring challenges to teachers to effectively accommodate students from different backgrounds ethnically and linguistically (Howard & Rodriguez-Minkoff, 2017), researchers in the field of multicultural education have attempted to formulate various theories to help teachers mitigate such challenges. Notably, Banks’ multicultural educational theory (1995a), Ladson-Billings’ Culturally Relevant Pedagogy (1995), and Gay’s Culturally Responsive Teaching (2002) emphasized the importance of incorporating students’ cultural backgrounds, knowledge, and perspectives into the teaching curriculum and pedagogy. Banks’ theory also underscores the importance of prejudice reduction in the curriculum used by teachers to help students develop positive attitudes toward peers from different racial, ethnic, and cultural backgrounds. Gay (2002) stresses the significance of incorporating students’ cultural identities and perspectives into the curriculum and instructional practices to enhance student engagement, motivation, and academic achievement.

First introduced in the conceptual framework of TALIS 2018 (Ainley & Carstens, 2018), TSMC was a term bridging Bandura’s self-efficacy (Bandura, 1997), teacher self-efficacy (Guskey & Passaro, 1994; Tschannen-Moran & Hoy, 2001; Tschannen-Moran et al., 1998), and multicultural education (Banks, 2001; Gay, 2002; Ladson-Billings, 1995). Recognizing its significance in the global society, TALIS 2018 included the new theme of equity and diversity; consequently, a distinct scale named “Teacher Self-Efficacy in Multicultural Classrooms” (TSMC) was incorporated in its conceptual framework. This inclusion was implemented due to the particular importance of teachers’ self-efficacy in diverse settings, as it plays a crucial role in equipping educators for the evolving global societies (Ainley & Carstens, 2018). In TALIS 2018, TSMC consists of five items that assess teachers’ level of confidence in addressing the challenges of a multicultural classroom (Ainley & Carstens, 2018). These items include teachers’ ability to adapt their teaching to the cultural diversity of students, promote collaboration between students with and without a migrant background, foster awareness of cultural differences among students, and mitigate ethnic stereotyping in a culturally diverse classroom.

TSMC was later formally defined by Choi and Lee (2020) as teachers’ confidence in their capacity to apply a variety of strategies to enhance instruction and learning in a multicultural classroom. Further developed subsequently by other researchers (Choi & Mao, 2021; Mo et al., 2021), TSMC specifically refers to teachers’ belief in their ability to effectively teach and support students from diverse cultural backgrounds, and it encompasses teachers’ confidence in understanding and valuing cultural differences, adapting instructional strategies to meet the needs of diverse learners, and creating an inclusive and respectful classroom environment. In our study, we define STEM teachers’ self-efficacy in multicultural classrooms (STEM TSMC) as STEM teachers’ belief in their capacity to apply a variety of strategies to enhance their instruction and facilitate student learning in STEM subjects within a multicultural classroom where the diverse cultural and linguistic background of students presents challenges to STEM teachers. STEM teachers are those who teach the individual subject areas of science, technology, engineering, and mathematics, not necessarily the integrated STEM subject.

Difference between STEM and Non-STEM TSMC

A review of literature revealed a scarcity of studies on this topic, with most focusing primarily on the concept of teachers’ self-efficacy instead of TSMC (Grimberg & Gummer, 2013; Parkhouse et al., 2019; Sakız et al., 2020), or investigating TSMC without differentiating between STEM and non-STEM teachers (Choi & Mao, 2021). In a study that used general teacher self-efficacy instead of TSMC, Sakız et al. (2020) found that math teachers’ self-efficacy was lower than that of language arts teachers, and science teachers’ self-efficacy was lower than that of social science teachers in Turkish K-12 schools. One notable study that did focus on TSMC and classify teachers according to their subject areas was conducted by Choi and Mao (2021). Based upon their analysis of TSMC across 47 education systems in TALIS 2018, they reported that mathematics and science teachers exhibited lower levels of TSMC compared to their counterparts in the reference group established in their analysis, which included teachers of ancient Greek and/or Latin, technology, arts, physical education, religion, and vocational skills. While focusing on TSMC, Choi and Mao (2021) did not specifically compare between STEM and non-STEM TSMC. Given the paucity of relevant studies and the significance of TSMC, it is crucial to explore the differences in TSMC between these two groups of teachers.

Factors Associated with STEM and Non-STEM TSMC

While there have been notable strides in studying TSMC, our understanding of the factors contributing to TSMC remains limited. Previous studies have focused on factors such as elements of initial teacher preparation, in-service teacher professional development, gender, school instructional leadership, and multicultural climate. In the following, we will review studies in these areas.

Initial Teacher Preparation

Regarding the potential association of the initial teacher preparation with TSMC, some studies have focused on analyzing multicultural coursework and student teaching experience. While some earlier studies found that multicultural coursework had no significant association with preservice teachers’ attitudes toward diversity (Locke, 2005; Nadelson et al., 2012), a recent study by Paulick et al. (2023) revealed that pre-service teachers who participated in the hands-on anti-bias community engagement module during their student teaching reported more significant growth in self-efficacy for culturally responsive teaching than those who did not participate in the module. Mahon and Cushner (2002) analyzed 50 participants’ overseas student teaching experience that lasted between 8 and 15 weeks and found that such an experience positively impacted their teaching self-efficacy that enables them to work better with different types of students and become sensitive to the needs of children with different backgrounds. Similarly, Mo et al. (2021) found a positive relationship between teachers’ study abroad experiences during their initial teacher education program and their TSMC by analyzing the Finnish sample of TALIS 2018.

Although these findings regarding multicultural coursework and overseas student teaching experience seem useful, none of these studies specifically compared STEM and non-STEM student teaching in a mixed ability, multicultural, or multilingual setting. Teaching in these settings can provide pre-service teachers with valuable opportunities to cultivate their self-efficacy in multicultural classrooms. Consequently, further research is warranted in this area.

In-Service Teacher Professional Development

Research suggests that educators working with students from diverse backgrounds should possess professional expertise in employing culturally relevant tactics (Siwatu, 2011). Considering the wide range of students in classrooms and their unique learning styles, effective professional development programs targeted at addressing the varying needs of students can significantly contribute to students’ academic success. Many studies have identified a positive link between in-service professional development in multicultural education and TSMC. For example, by analyzing the data of 86 U.S. elementary school teachers, DeJaeghere and Cao (2009) found increased intercultural competence among these teachers after their participation in district and school-based professional development programs. Malinen et al. (2013) compared teacher efficacy of Chinese, Finnish, and South African teachers teaching various subjects in grades 1–9 and found that the amount of teacher training related to inclusive education had a positive and significant effect on teachers’ self-efficacy in the Finnish sample, though this relationship was not identified among the Chinese or South African samples. Dixon et al. (2014) found that professional development training for both STEM and non-STEM teachers in U.S. K-12 schools enhanced their teaching efficacy in multicultural setting. Two recent studies both using TALIS 2018 data arrived at similar results. Choi and Lee (2020) identified a significant positive relationship between professional development in multicultural education and TSMC both for Korean and U.S. STEM or non-STEM teachers. Another study by Schwarzenthal et al. (2023) also drew on TALIS 2018 data but focused extensively on 91,768 teachers in grades 7–9 across 11,523 schools and 46 education systems, and they found that intercultural professional development was positively related to teacher intercultural self-efficacy.

Additionally, some studies focused specifically on the association of professional development in multicultural education with STEM teachers’ self-efficacy in multicultural classrooms. For example, by analyzing the effect of a two-year professional development program designed for science educators in grades 3–8 near or on American Indian reservations in Montana, Grimberg and Gummer (2013) found that science teachers gained better self-efficacy in teaching science. A recent study by Choi and Mao (2021) analyzed the data from 47 education systems in TALIS 2018 and found that professional development in multicultural education facilitated teachers in developing their TSMC. A more recent study by Zhou et al. (2024) also used TALIS 2018 but specifically focused on a sample of 985 U.S. STEM teachers of English language learners. After analyzing the inter-relationship among STEM teachers’ professional development participation in multicultural education, perceptions of school climate, teaching efficacy, and job satisfaction, Zhou and colleagues found that teachers’ professional development participation in multicultural education for differentiated instruction and teaching students with special needs directly contributed to STEM teachers’ self-efficacy in teaching English language learners. However, they used general teacher self-efficacy instead of teacher efficacy in multicultural classrooms in their analysis. Therefore, whether such a result holds true for TSMC needs to be further investigated.

Results from the above review echo with the findings of a systematic review (Parkhouse et al., 2019) and a meta-analysis study (Zhou et al., 2023). However, most of the studies included in Parkhouse et al.’s (2019) systematic review focused on teachers of various subjects while only 30% of the studies focused on teachers in the STEM subject area. Furthermore, while Zhou et al. (2023) meta-analysis study offered robust evidence regarding the positive effect of professional development programs on improving STEM teachers’ self-efficacy in K-12 classroom teaching, the study’s focus on STEM TSMC remains limited. Consequently, these gaps and limitations underscore the need for further exploration into the role of professional development programs in enhancing STEM TSMC.

Gender

Gender plays a significant role in shaping teachers’ self-efficacy, but the results were mixed. While Cheung (2006) found that female in-service primary school teachers in Hong Kong had higher levels of self-efficacy than their male counterparts, Klassen and Chiu (2010) found the opposite was true, i.e., female teachers scored lower in self-efficacy than their male colleagues teaching K-12 in western Canada, and this pattern was also identified among the pre-service teachers by Sak (2015) who found male pre-service teachers had higher levels of self-efficacy than their female counterparts. Furthermore, when focusing specifically on TSMC, Choi and Lee (2020) found that female teachers in TALIS 2018 had higher TSMC levels than their male counterparts. The inconclusive nature of the result regarding the relationship between gender and TSMC necessitated further investigation.

School Instructional Leadership

We included instructional leadership instead of other leadership models in our analysis due to its prominence as one of the most extensively studied leadership models in educational research (Daniëls, et al., 2019; Gumus et al., 2018). A plethora of research studies reported the positive correlation between instructional leadership and teachers’ self-efficacy. Some of these studies relied on data collected directly by researchers in various countries (Cansoy & Parlar, 2018; Chen & Rong, 2023; Khan et al., 2024), while others utilized TALIS large-scale datasets (Bellibas & Liu, 2017; Duyar et al., 2013; Fackler & Malmberg, 2016; Liu et al., 2021) to arrive at similar findings.

However, to date, there has been no specific research exploring the connection between instructional leadership and TSMC. Nonetheless, a recent study by Choi (2023) uncovered a positive relationship between another leadership style, distributed leadership, and TSMC. Choi’s study, based on data gathered from 2,050 teachers in 165 U.S. schools as part of the TALIS 2018 survey, revealed that distributed leadership is indirectly associated with TSMC through a feedback network. Our research addresses this gap in the literature by providing crucial empirical evidence on the relationship between instructional leadership practices and TSMC, encompassing both STEM and non-STEM teachers.

School Multicultural Climate

Previous research has identified associations between general school climate and teachers’ self-efficacy (e.g., Aldridge & Fraser, 2016; Almessabi, 2021; Hosford & O’Sullivan, 2016; Pas et al., 2012). However, when it comes to the relationship between school multicultural climate and teachers’ self-efficacy, particularly in the context of STEM or non-STEM TSMC, research is limited. While the findings identified in Zakariya’s study (2020) are valuable in uncovering the potential link between school climate and general self-efficacy among 3,951 lower secondary school teachers predominantly in STEM fields in Norway from TALIS 2018, the researcher overlooked the examination of the potential relationship between multicultural aspects of school climate and STEM teachers’ self-efficacy in multicultural classrooms.

With the multicultural aspect of school climate and TSMC included in their analysis, a recent study conducted by Schwarzenthal et al. (2023) provided important insights. Drawing on data from TALIS 2018 with 91,768 teachers in 11,523 schools across 46 education systems in their analysis, Schwarzenthal et al. (2023) found a positive association between teacher perceptions of a multicultural school climate and their self-efficacy in multicultural classrooms at the school level. Although their analysis focused on multicultural school climate, Schwarzenthal and colleagues did not distinguish between STEM and non-STEM teachers within the sample they analyzed. Consequently, further investigation is warranted to ascertain whether such a positive relationship remains consistent across both STEM and non-STEM teacher cohorts.

The Current Study

The preceding review of the relevant literature indicates that the concept of STEM teachers’ self-efficacy in the context of socio-culturally and linguistically diverse classrooms is an understudied topic. While the prior studies provided valuable insights into the difference between STEM and non-STEM teachers’ TSMC as well as various factors that can contribute to these teachers’ TSMC, their limited attention to STEM education and the inclusion of a scant number of factors in their analysis made it imperative to conduct additional research on this important concept. Drawing on U.S. lower secondary (grades 7–9) teacher and principal data from TALIS 2018, carried out by the Organization for Economic Cooperation and Development (OECD), this study has three purposes. First, it aims to determine whether a significant difference exits between STEM and non-STEM TSMC. Second, it seeks to identify the factors that may significantly associate with STEM or non-STEM TSMC. Thirdly, it endeavors to explore the similarities and differences among the factors that may significantly associate with STEM or non-STEM TSMC. The inquiry is guided by the following three research questions:

  • RQ1: Is there a significant difference between STEM and non-STEM TSMC in U.S. grades 7-9?

  • RQ2: Which factors exhibit significant associations with STEM and non-STEM TSMC, respectively, in U.S. grades 7-9?

  • RQ3: What similarities and differences can be identified among the factors associated with STEM and non-STEM TSMC in U.S. grades 7-9?

Methods

Data Source and Participants

The data of this study were retrieved from the TALIS 2018 U.S. public-use files (https://nces.ed.gov/surveys/talis/publications.asp). As a collaborative effort of the OECD and participating countries or education systems, TALIS has its primary focus on exploring and analyzing the working conditions and learning environments of lower secondary education teachers (grades 7–9 in the US) and their principals in different countries and regions. A total of 48 countries and education systems participated in the data collection in TALIS 2018.

In the United States, the National Center for Education Statistics of the Institute of Educational Science and U.S. Department of Education conducted TALIS 2018. TALIS 2018 adopted a stratified two-stage probability sampling design (Kastberg et al., 2021). A sample of 220 U.S. schools were selected using systematic random sampling with probability proportional to a measure of size (PPS). Teachers were sampled within schools with at least 20 teachers of these schools (Kastberg et al., 2021). Ultimately, 165 schools with 50% or more responses among teachers were included in the 2018 TALIS U.S. public dataset. A total of 2,560 teachers from 165 schools, each covering grades 7, 8, and 9, were included in the U.S. teacher database. Of these lower secondary teachers, approximately 66% were females and 98% worked full-time.

The dataset retrieved from TALIS 2018 public data contained missing values, necessitating a systematic approach for their management. We adopted a multi-step strategy in handling the missing data by taking into consideration the nature of the variables and their relevance to the study. First, concerning the primary subjects taught by teachers, 1,992 teachers provided answers to this question, while 568 teachers did not. Given the pivotal role of this variable in categorizing STEM and non-STEM teachers, we opted not to include these missing values, since this information could not be imputed (Cheema, 2014). Second, for the dependent variable, TSMC, which comprised only five items, participants with missing responses on two or more items were excluded from analysis (n = 283), as their inclusion would not provide meaningful information for the analysis (Downey & King, 1998; Mazza et al., 2015). Following the deletion of the cases with missing values on TSMC, the sample size was decreased to 1,709. Table 1 illustrates the step of this strategy.

Table 1 Management of Missing Values on Variables of Interest
Full size table

Finally, in handling missing values on the independent variables, we employed a listwise deletion approach, which is commonly utilized in educational research (Graham, 2009). While this approach may lead to a reduction in sample size, it ensures completeness of the analyzed data and minimizing potential bias due to missing values. Graham (2009) advocates for listwise deletion when the loss of cases is minimal (typically less than approximately 5%). Therefore, we systematically removed the cases with missing values on each independent variable step-by-step. It is not feasible to replace missing values on categorical variables with binary response options such as “Yes” or “No”, for variable like gender, elements included in formal education, topics included in professional development, or diversity practices implemented in school. Thus, we employed listwise deletion, removing cases with missing values on these variables. For example, we deleted 4 cases with missing values on gender (TT3G01), which fell below the 5% threshold of the valid sample size, 1,709, making listwise deletion an acceptable approach. This step reduced the valid sample size to 1705. Similarly, we removed 11 cases for TT3G06E1 and TT3G06F1, maintaining consistency with the 5% threshold relative to the valid sample size, 1705. Consequently, the valid sample size changed to 1,694. Followed the same rationale and approach, we excluded cases with missing values on other variables listed in Table 1.

After conducting data mining and cleaning processes, the final sample consisted of 1,450 teachers. Among them, there were 326 English teachers teaching reading, writing, literature in English, and English as a Second Language, 244 mathematics teachers, 210 sciences teachers, 174 social sciences teachers, 75 modern foreign language teachers, 57 technology teachers, 152 arts teachers, 119 physical education teachers, 34 teachers teaching practical and vocational skills, 52 others, and a small number of teachers teaching subjects such as Religion or Ancient Greek. We classified math, science, technology teachers as STEM teachers, while the remaining teachers were categorized as non-STEM teachers. In accordance with the guidelines for merging data from TALIS 2018 (Kastberg et al., 2021), the teacher and principal datasets were merged using the school identification variable, allowing for the connection of teacher and principal responses from the same school. The final sample consisted of 511 STEM teachers within 144 schools and 939 non-STEM teachers nested within 154 schools who have answered questions related to TSMC and other questions of interest. Table 2 presents the characteristics of both STEM and non-STEM teachers.

Table 2 Characteristics of STEM and Non-STEM Teacher Participants
Full size table

Variables and Measures

Items from both the TALIS 2018 U.S. teacher and principal questionnaires were utilized to measure the variables of interest. The outcome variable, TSMC, was measured by five items that queried teachers’ confidence level in various activities related to teaching a culturally diverse class, such as “adopting teaching to the cultural diversity of students” (see Table 3). Using confirmatory factor analysis to assess the model fit of the five items constituting TSMC, TALIS 2018 reported that the sampled population in the US demonstrated good fit indices, CFI = 0.99, TLI = 0.95, RMSEA = 0.05, and SRMR = 0.01, which allows it to become a latent construct encompassing the five items (Ainley & Carstens, 2018) and the index score was used in our study. The Cronbach’s alpha for TSMC was 0.846 for the current study, indicating a high level of internal consistency for the TSMC scale (OECD, 2019).

Table 3 Description of Variables Used in the Current Study
Full size table

We investigated the relationship between both teacher-level and school-level variables and the outcome variable TSMC by utilizing questions from both TALIS teacher and principal questionnaires. Teacher-level variables, as outlined in Table 3, encompassed gender, initial teacher preparation (ITP), topics covered in professional development (PDT), professional development needs (PDN), and diversity practices implemented in the school (DPI). Gender (TT3G01) was recoded as 0 for male and 1 for female. Similarly, items within initial teacher preparation (TT3G06E1, TT3G06F1) were recoded as 0 for No and 1 for Yes, based upon the original response options indicated in Table 3. The initial teacher preparation items consisted of elements from formal education or training, as well as the perceived preparedness for teaching-related elements (two items for each). Regarding topics covered in professional development (TT3G23H, TT3G23I, TT3G23J), and diversity practices implemented in the school (TT3G47A, TT3G47B, TT3G47C, TT3G47D), we first recoded the initial responses as 0 for No and 1 for Yes, then summed the responses to a composite score indicating the number of topics covered in professional development (ranging from 0 to 3) or number of diversity practices implemented in school (ranging from 0 to 4), with 0 as the reference group. Professional development needs consisted of three items (TT3G27H, TT3G27I, TT3G27J). To further streamline the number of variables in the multilevel model, we standardized the four items for initial teacher preparation and three items for professional development needs to have a mean of 0 and a standard deviation of 1, following the approach outlined by Finch et al. (2019). Subsequently, these standardized items were combined to create composite variables by utilizing the grand mean of the teacher-level variables. Cronbach’s alphas for the above two scales, i.e., initial teacher preparation and professional development needs, were 0.769 and 0.793 respectively, indicating good levels of internal consistency across the above two scales.

In this study, school-level variables encompassed several factors including characteristics of the school’s background such as the percentage of students whose first language is not English and those from socioeconomically disadvantaged households. Additionally, variables such as current school enrollment, the community where the school is situated, and aspects of school instructional leadership were also included. To represent school-level variables in the models, a matrix format was employed. Each column contained specific items, such as the percentage of students whose first language is not English and those from socioeconomically disadvantaged households, while each row represented an individual teacher’s values on these variables (Lüdtke et al., 2008). Similarly, for school enrollment and location, each column contained items like current school enrollment and the community in which the school is located, with each row representing an individual teacher’s values on these variables (Lüdtke et al., 2008). By consolidating the school-level variables into single matrix-vectors, they could be collectively incorporated into the model, mitigating issues of multicollinearity that may arise if entered separately (Kuhn & Johnson, 2013). Lastly, three items were used to assess school instructional leadership, as detailed in Table 3. TALIS generated an index score, T3PLEADS, specifically for assessing school instructional leadership. This index score was derived by calculating the group mean as a representation of school instructional leadership for each school in the study.

Data Analysis

Prior to data analysis, we examined the distribution of the variables of interest and checked correlations, as well as collinearity, among the predictors to ensure that multicollinearity, which can undermine the reliability of regression analyses, was not present. All the absolute values of correlation coefficients among the variables were smaller than 0.70, and variance inflation factor (VIF) smaller than 10. The examination ensured the collinearity was not present at this study.

To address the first research question, general linear model univariate analysis was conducted, allowing for specification of sample weight in the calculation, to compare the mean difference in TSMC between STEM and non-STEM teachers. Sampling weights was constructed in TALIS 2018 (OECD, 2019). In this case, final teacher weight was used.

Considering the nested nature of the teacher-level and school-level variables in the present study (teachers nested within schools), analytic approaches that account for the nested structure of the TALIS dataset are essential for accurate and valid analysis. Multilevel modeling allows for the simultaneous analysis of variation at both the individual (teacher) and group (school) levels. It accounts for the nested structure of the data by modeling the hierarchical nature of the data and estimating the effects of predictors at each level while controlling for the dependencies within and between levels (Finch et al., 2019). To address the second research question, we conducted multilevel regression models to examine to what extent teacher-level and school-level predictors significantly relate to TSMC for STEM and non-STEM teachers, respectively. The lme4 package in R was employed to estimate the parameters in linear mixed-effects models (Bates et al., 2015), using restricted maximum likelihood (REML). For both STEM and non-STEM teachers, we estimated a multilevel model:

Level 1 (Teacher Level)

$$TSM{C}_{ij}={\beta }_{0j}+{\beta }_{1j}Gende{r}_{ij}+{\beta }_{2j}IT{P}_{ij}+{\beta }_{3j}PD{T}_{ij}+{\beta }_{4j}PD{N}_{ij}+{\beta }_{5j}DP{I}_{ij}+{e}_{ij}.$$
(1)

Level 2 (School Level)

$${\beta }_{0j}={\gamma}_{00}+{\gamma}_{01}SCH{B}_{j}+{\gamma}_{02}SCHE{L}_{j}+{\gamma}_{03}SCHI{L}_{j}+{u}_{0j}$$
(2)
$${\beta }_{1j}={\gamma}_{10}+{\gamma}_{11}SCH{B}_{j}+{\gamma}_{12}SCHE{L}_{j}+{\gamma}_{13}SCHI{L}_{j}+{u}_{1j}$$
(3)
$${\beta }_{2j}={\gamma}_{20}+{\gamma}_{21}SCH{B}_{j}+{\gamma}_{22}SCHE{L}_{j}+{\gamma}_{23}SCHI{L}_{j}+{u}_{2j}$$
(4)
$${\beta }_{3j}={\gamma}_{30}+{\gamma}_{31}SCH{B}_{j}+{\gamma}_{32}SCHE{L}_{j}+{\gamma}_{33}SCHI{L}_{j}+{u}_{3j}$$
(5)
$${\beta }_{4j}={\gamma}_{40}+{\gamma}_{41}SCH{B}_{j}+{\gamma}_{42}SCHE{L}_{j}+{\gamma}_{43}SCHI{L}_{j}+{u}_{4j}$$
(6)
$${\beta }_{5j}={\gamma}_{50}+{\gamma}_{51}SCH{B}_{j}+{\gamma}_{52}SCHE{L}_{j}+{\gamma}_{53}SCHI{L}_{j}+{u}_{5j}$$
(7)

In Eq. (1), teachers, i, are nested within schools j; β0j is the intercept term, representing the value of the dependent variable when all independent variables are set to zero. β1j, β2j, …, β5j are the regression coefficients associated with each independent variable in the model.

They represent the estimated change in the dependent variable for a one-unit change in the corresponding independent variable, holding all other variables constant. Genderij, ITPij, PDTij, PDNij, DPIij are the independent variables at the teacher level, representing different factors or predictors that may influence the dependent variable, TSMCij.

eij is the usual residual error term (Hox et al., 2018), which captures the variability in the dependent variable, TSMC, that is not explained by the independent variables included in the model.

The school-level factors, denoted as Z, include school background (SCHBj), school enrollment and location (SCHELj), and school instructional leadership (SCHILj), which may contribute to the dependent variable (Hox et al., 2018).

The u-terms in Eqs. 2 through 7 are residual error terms at the school level, assumed to have a mean of zero and be independent from the residual errors, eij, at the teacher level (Hox et al., 2018).

Results

RQ1: Difference between STEM and non-STEM TSMC

For research question 1, our analysis revealed that STEM teachers scored significantly lower in TSMC (M = 10.62, SD = 2.20) than their non-STEM counterparts (M = 11.50, SD = 2.27), F (1, 1,448) = 49.17, p < 0.001.

RQ2: Factors Significantly Associated with STEM or non-STEM TSMC

To address research question 2, we conducted multilevel modeling. Table 4 and Table 5 present the results of multilevel regression analyses for STEM and non-STEM teachers, respectively. The results revealed that, while keeping other variables constant, three factors were found to be significantly related to STEM TSMC as shown in Table 4. These significant factors included gender, professional development needs, and diversity practice implemented in school. Specifically, female STEM teachers scored significantly higher in TSMC than male STEM teachers, β = 0.144, t = 3.691, p < 0.001, partial η2 = 0.026. Moreover, STEM teachers who reported higher professional development needs scored significantly lower in TSMC than their colleagues who reported lower professional development needs, β = -0.179, t = -3.531, p < 0.001, partial η2 = 0.024. Additionally, diversity practice implemented in school is also significantly related to STEM TSMC, β = 0.112, t = 3.627, p < 0.001, partial η2 = 0.025. Surprisingly, none of the school-level variables, including school background, enrollment and location, or school instructional leadership, were significantly associated with STEM TSMC.

Table 4 Multilevel Modeling Results for STEM TSMC
Full size table
Table 5 Multilevel Modeling Results for Non-STEM TSMC
Full size table

As indicated in Table 5, the results for non-STEM teachers showed that, while controlling for the other factors, several teacher-level factors including gender, initial teacher preparation, professional development topics, professional development needs, and school instructional leadership emerged as the significant predictors for non-STEM TSMC. Female non-STEM teachers reportedly scored significantly higher than their male counterparts, β = 0.203, t = 6.359, p < 0.001, partial η2 = 0.041. Initial teacher preparation was found significantly and positively related to non-STEM TSMC, β = 0.329, t = 7.866, p < 0.001, partial η2 = 0.062. Professional development topics was found significantly and positively related to non-STEM TSMC, β = 0.121, t = 3.949, p < 0.001, partial η2 = 0.016. However, professional development needs were significantly but negatively related to non-STEM TSMC, β = -0.118, t = -3.028, p < 0.01, partial η2 = 0.010, suggesting teachers who have higher level of professional development needs feel less efficacious in teaching in multicultural classrooms. Additionally, at school level, school instructional leadership was significantly and positively related to non-STEM TSMC, β = 0.084, t = 2.119, p < 0.05, partial η2 = 0.005, while school background, school enrollment and location were not significant predictors of non-STEM TSMC (see Table 5).

RQ3: Similarities and Differences among the Factors

In examining factors associated with TSMC among both STEM and non-STEM teachers, some similarities were uncovered. Gender and professional development needs emerged as significant predictors of TSMC for both groups. Female teachers consistently scored significantly higher in TSMC than their male counterparts. Moreover, higher professional development needs were significantly but negatively associated with TSMC, suggesting that teachers with greater professional development needs in multicultural teaching contexts reported lower TSMC scores.

However, differences emerged as well. While the implementation of diversity practices in schools was identified as a significant factor specifically for STEM teachers, other variables including initial teacher training, professional development topics, and school instructional leadership were found to be significant predictors exclusively for non-STEM teachers, as indicated in Tables 4 and 5.

Discussion and Implications

In this study, we sought to achieve a better understanding of the difference between STEM and non-STEM TSMC, the potential factors associated with each of the two groups of teachers, and the similarities and differences among the factors considering the growing diversity observed in U.S. K-12 classrooms. Our analysis revealed several meaningful findings that could be used to extend the theoretical and empirical understanding of the differences between STEM and non-STEM TSMC, as well as shed light on essential factors that could enhance TSMC. Ultimately, these findings have the potential to contribute to the improvement of both STEM and non-STEM teacher education and teacher professional development initiatives.

Deficiency of STEM TSMC

Our analysis revealed that U.S. STEM teachers scored significantly lower in TSMC than their non-STEM counterparts, which partially supports the finding of Choi and Mao (2021) that also utilized TALIS 2018 in their analysis. Their study revealed lower levels of TSMC among mathematics and science teachers compared to teachers in Greek and/or Latin, technology, arts, physical education, religion, and vocational skills across 47 education systems while we found that in the U.S. sample, a wider category of STEM teachers reportedly scored lower in their TSMC relative to non-STEM teachers. Additionally, our finding aligns with that of Sakız et al. (2020) who found math and science teachers’ general teacher self-efficacy was lower than that of their non-STEM peers in Turkish K-12 school settings.

Our finding regarding the deficiency of STEM TSMC in the US is significant as it provides valuable insights into the unique challenges and needs that STEM teachers may encounter. In the multicultural educational setting, communication difficulties arising from cultural and language differences can hinder effective teaching and thus cause STEM teachers to feel not efficacious in conveying complex STEM concepts to students with limited English proficiency, resulting in diminished TSMC. Given the limited research on STEM TSMC, this important finding has the potential to inspire further exploration of this topic, leading to more relevant investigations. Additionally, it can provide insights for teacher education policy makers and designers of in-service teacher professional development programs, aiding in the creation of tailored strategies and support systems that will ultimately enhance STEM TSMC in fostering inclusive and culturally responsive teaching practices.

STEM Teacher Preparation, Professional Development, and TSMC

While it is theoretically assumed that providing pre-service teachers with the opportunity to teach in a mixed ability or multicultural/multilingual setting during their initial teacher preparation program would positively help improve TSMC, our study revealed that this was only the case for non-STEM teachers, not STEM teachers. This finding contributes important empirical evidence to the existing literature, as previous studies often overlooked the distinction between STEM and non-STEM teachers. Some earlier research demonstrated a positive association between TSMC and factors like hands-on anti-bias community engagement modules (Paulick et al., 2023), overseas student teaching (Mahon & Cushner, 2002), or study abroad experiences (Mo et al., 2021), while others did not find such connections (Locke, 2005; Nadelson et al., 2012). The positive but non-significant correlation between STEM TSMC and initial teacher preparation suggests that pre-service teacher education programs should be critically examined so that they will focus more on catering to the unique needs of pre-service teachers in the STEM field. While STEM methods courses and multicultural education courses are typically offered in teacher education programs, it is essential that pre-service teachers be provided with practicum experiences in diverse, mixed-ability, multicultural, or multilingual settings that are vital for fostering their self-efficacy in multicultural classrooms, as demonstrated by the positive and significant correlation between non-STEM TSMC and initial teacher preparation.

Additionally, we found that professional development topics were positively and significantly associated with non-STEM TSMC, which appears to aligh with findings from some prior studies (Dejaeghere & Cao, 2009; Dixon et al., 2014). While the positive relationship between STEM TSMC and professional development topics such as approaches to individualized learning, teaching students with special needs, and teaching in multicultural or multilingual settings resonates with findings from several earlier studies (e.g., Choi & Lee, 2020; Choi & Mao, 2021; Grimberg & Gummer, 2013; Schwarzenthal et al., 2023; Zhou et al., 2024), the lack of a significant association between the two may be attributed to the skeptical view of some experienced STEM teachers towards multicultural-education oriented professional development sessions when they were required to alter their instructional techniques and strategies, especially if they believed their existing methods had proven to be successful (Johnson & Fargo, 2014).

Furthermore, our study revealed that professional development needs regarding teaching in a multicultural or multilingual setting are significantly and negatively associated with both STEM non-STEM TSMC, suggesting that teachers who reported higher needs of professional development in these specific areas tend to feel less efficacious in TSMC. STEM fields have traditionally been seen as areas where facts and principles could be applied universally regardless of cultural backgrounds; therefore, emphasizing content knowledge while overlooking or downplaying the importance of cultural diversity and its potential impact on teaching and learning in these subjects has been prevalent (Charity Hudley & Mallinson, 2017; Gutiérrez, 2002; Timmons-Brown & Warner, 2016). Accordingly, both pre-service and in-service STEM teachers may not have sufficient training in culturally responsive teaching strategies or exposure to diverse cultural backgrounds. This lack of preparation can make it difficult for them to effectively connect with and teach students from various cultural contexts, ultimately leading to the deficiency in STEM teachers’ confidence in their ability to work effectively with students from diverse backgrounds (Powell et al., 2016). Based on our findings, it is recommended that more appropriate initial teacher preparation and in-service professional development programs focusing on approaches to individualized learning, teaching students with special needs, and teaching in a multicultural or multilingual setting should be designed and implemented. By providing targeted support in these areas, both pre-service and in-service STEM teachers can improve their self-efficacy to better address the diverse needs of student populations.

School Instructional Leadership, Multicultural Climate, and TSMC

While school instructional leadership was found to demonstrate a positive relationship with both STEM and non-STEM TSMC, this association is only statistically significant for non-STEM teachers. This is valuable empirical evidence, particularly considering that no prior studies have specifically investigated the connection between instructional leadership and TSMC, despite abundant evidence linking instructional leadership to general teacher self-efficacy (e.g., Bellibas & Liu, 2017; Cansoy & Parlar, 2018; Chen & Rong, 2023; Fackler & Malmberg, 2016; Liu et al., 2021). This significant finding suggests that school leaders should critically assess how they can enhance cooperation among STEM teachers to foster the development of innovative teaching practices, while also ensuring that STEM teachers are encouraged to take ownership of improving their teaching skills and feel accountable for their students’ learning outcomes.

Furthermore, our study revealed that although the implementation of diversity practices within the school context has a positive association with both STEM and non-STEM TSMC, the association reaches statistical significance with STEM TSMC only. This finding contributes importantly to the existing empirical evidence in regards to the link between school multicultural climate and TSMC (Schwarzenthal et al., 2023; Zakariya, 2020) and suggests that STEM teachers can particularly benefit from these school-wide multicultural practices, i.e., supporting activities or organizations that encourage students’ expression of diverse ethnic and cultural identities, organizing multicultural events, teaching students how to deal with ethnic and cultural discrimination, and adopting teaching and learning practices that integrate global issues throughout the curriculum. Our finding underscores the importance of implementing these diversity practices within the school environment to cultivate an inclusive and supportive school climate. By doing so, schools can positively influence the multicultural competence of all teachers, especially STEM educators, in diverse settings while simultaneously addressing the needs of diverse student populations.

Limitations and Suggestions for Future Research

We acknowledge two limitations of this study. First, we used the publicly available data from TALIS 2018, which is a cross-sectional study that does not allow for making causal inferences. Thus, results of the study can only be used for correlational interpretations. Second, the sampled teachers in this study were from grades 7–9 in the US; as such, generalization of the results to other grade levels or educational systems should be made with caution. Accordingly, the results from this study should be interpreted with these limitations in mind, and implications discussed in the above are bounded thereof.

Considering the outlined limitations, we suggest that future research endeavors may replicate this study by analyzing samples of teachers from different countries or regions. This replication could help verify whether the findings regarding STEM and non-STEM TSMC hold consistent across different cultural and educational contexts. Additionally, future research efforts could adopt an experimental or quasi-experimental design, enabling a more rigorous investigation of STEM and non-STEM TSMC, as well as the factors associated with TSMC. Moreover, further studies on STEM TSMC could be conducted to examine the effectiveness of more tailored pre-service teacher education programs or in-service teacher professional development initiatives that are designed specifically to assist STEM teachers in enhancing their self-efficacy in teaching diverse student populations. Such research endeavors would contribute significantly to the advancement of knowledge and the development of effective strategies for promoting inclusive teaching practices in STEM education.

Conclusion

By analyzing the U.S. data from TALIS 2018, this study first compared the score difference between STEM and non-STEM TSMC, then investigated the factors contributing to such a difference, and lastly compared the similarities and differences among the factors. The findings of the study contribute to the research literature in identifying the deficiency of STEM TSMC in comparison to their non-STEM counterparts and offers important empirical evidence regarding the factors associated with TSMC for both groups of teachers. The findings also provide crucial implications to initial teacher preparation programs, in-service professional development, instructional leadership, and school multicultural climate for both STEM and non-STEM teachers. Identifying the factors related to TSMC can help educational institutions and policymakers develop targeted strategies to enhance STEM teachers’ competence in multicultural or multilingual settings. Consequently, STEM teachers will effectively address the diverse needs of their students and promote inclusive teaching practices and student learning.