1 | Introduction

In teacher education, field practice constitutes a pivotal component, serving as a bridge between theoretical knowledge and practical application (Caires & Almeida 2005). It also serving as a real-life arena, where skills are practiced and learned in their own right (Youngs et al., 2022). Across a wide array of European teacher education programs, field practice forms a significant portion of the time in the program, although cross-country differences exist (Oberhuemer, 2014; Weisdorf, 2020). Typically, professional education programs consist of multiple field practice placements in schools, where students practice core skills and engage in profession-specific reasoning that becomes more complex as they progress through their studies. Field practice placements offer student teachers unique opportunities to engage themselves in the real-life complexities of the classroom while learning to master teacher and teaching-related practices through observation, own practice and receiving feedback (Nielsen, 2021). It is in these authentic settings that the theoretical concepts learned through on-campus teaching are brought to life, allowing student teachers to not only apply and refine the core skills of their profession, but also to situate and deepen their understanding.

            Research indicates that positive field practice experiences are essential in setting teachers on a successful professional trajectory. Rots et al. (2007) demonstrated that a teacher’s decision to remain in the profession hinges on initial commitment and the quality of early teaching experiences. They highlight the role of hands-on training in teacher education, emphasizing its significant impact on teachers quitting. Other work emphasizes the crucial role of field practice in mitigating the 'practice shock' often experienced by new teachers (Haugen & Hestbek, 2015). According to this account, field practice – although not always sufficiently realistic – provides vital insights into the challenging realities faced in school environments, such as a lack of time for preparation, parent collaboration, conflict management, and so on (Böwadt & Vaaben, 2021), as well as the need to be adaptable to the teaching context (Danmarks Evalueringsinstitut, 2022).

1.1. Academic self-efficacy

A key concept in education is academic self-efficacy (Bandura, 1997; Khine & Nielsen, 2022). Self-efficacy can be generally defined as "a belief in one's ability to plan and execute the necessary actions to achieve specific goals" (Bandura, 1997, p. 3). Academic self-efficacy has been defined in different ways and with different degrees of specificity. For example, part of the field (what we might refer to as the Schwartzer school of self-efficacy) has argued that the various and numerous cases of failure or success over time in different academic domains may lead to a more general sense of self-efficacy, referring to a global ability to master challenges which also cover education (e.g. Scholz et al., 2002; Schwarzer & Jerusalem, 1995). From this follows that academic self-efficacy does not need to be measured in itself. Another part of the field (i.e., the Bandura school of self-efficacy) has advocated that academic self-efficacy is task-specific and should be assessed against specific tasks in a specific context or against tasks in a somewhat broader domain to be termed specific self-efficacy (e.g. Bandura, 1997; Scherbaum et al., 2006). However, both schools of thought on self-efficacy agree that differing demands through education imply that one’s sense of academic self-efficacy might vary depending on the specific educational context and the point in time during the course of education (Nielsen, et al., 2017).

            Academic self-efficacy in higher education has mainly been studied in relation to the academic content of education, and has, in meta-analyses, been found to be the strongest non-ability predictor of grades, above other non-ability constructs such as personality traits, motivation, and various learning strategies (Bartimote-Aufflick et al., 2015; Richardson et al., 2012). In addition, students who feel efficacious when learning or performing a task, participate more readily, work harder, persist longer when they encounter difficulties, and achieve at a higher level of academic performance (Schunk & Pajares, 2002). Research on academic self-efficacy in relation to field practice is less prominent, particularly research taking into account the special position of the student teachers as learners having to engage in teaching practices that they do not yet master, in order to learn and move towards becoming full-fledged teachers. In this situation, where students are learning professional skills while practicing them, their identity will invariably be split between being a student and a teacher (e.g., Anspal et al., 2018; Alsup, 2005). This highlights the need to attend to self-efficacy in relation to field practice taking into account this unique learner’s position in the relatively unsafe, psychologically, high-risk learning environment compared to the on-campus learning environment. During field practice the consequences of “failure” can extend to other people, i.e., the children and their parents, as well as “teacher colleagues”. The personal psychological risk is also higher in the field practice learning environment, compared to the on-campus environment, since student teachers face increased communicative, emotional, and psychological demands while learning (Anspal et al., 2018; Caries & Almeida, 2005). These circumstances of the field practice learning environment factors may contribute to student teachers doubting whether they have what it takes to engage in the profession’s core practices, i.e. affecting their self-efficacy for practice negatively (Nielsen et al., 2024).

1.2. Research on self-efficacy and field practice in teacher education

Previous research on student teachers’ self-efficacy in field practice has mainly been focused on examining the potential of field practice to foster teacher self-efficacy. Studies show that authentic classroom experiences, through school-based field practice, significantly enhance student teachers’ teacher self-efficacy (e.g. Klassen & Durksen 2014; Flores, 2015; Pfitzer-Eden 2016; Rupp & Becker 2020; Brown et al., 2021; Symes et al. 2023). For example, Klassen and Durksen (2014) tracked the self-efficacy of Canadian student teachers during an 8-week school-based field practicum. This period included weekly mentoring sessions and daily informal support from mentor teachers to help with planning and managing their teaching. Using only three items from the Teachers' Sense of Efficacy Scale (TSES; Tschannon-Moran & Woolfolk Hoy, 2001), they reported that student teachers’ teacher self-efficacy increased during their field practice, coupled with declining levels of stress.

            However, it is difficult to assess how effective field practice is in fostering student teachers’ teacher self-efficacy or their instructional self-efficacy. While some studies report no or even negative effects (e.g., Savasci & Tuna 2018; Elstad & Christophersen 2017; Atay, 2007; Lin & Gorrell 2001; Pendergast et al., 2011), variations in study designs - such as setting (school-based vs. lab-based), support activities (feedback on lesson plans, lessons or modeling), target group (beginners or advanced students), length of field practice and educational programs across different countries - complicate comparisons. To address these differences, Mok et al. (2023) conducted a meta-analysis of 28 studies focusing on pre-service and beginning teachers. They reported a medium overall effect size on teacher self-efficacy, indicating a generally positive impact.

            Previous studies have also suggested that receiving feedback on lesson planning (e.g. Mok et al. 2023) may amplify the positive impact of field experiences on teacher self-efficacy. In addition, the duration of the field experience contributes positively to its impact (Charles et al. 2023), whereas field practice seems to produce similar positive effects among beginner and advanced student teachers (Pfitzner-Eden 2016). Other research has linked shifts in self-efficacy during field practice to direct engagement with teaching activities (e.g. Martins et al. 2015; Flores 2015; Pfitzer-Eden 2016; Weber & Greiner 2019; Brown et al. 2021), observation of experienced teachers (e.g. Flores 2015; Pfitzer-Eden 2016), the opportunity to link theoretical knowledge to practice (e.g. McDonnough & Matkins 2010), and feedback from mentors, peers and students (e.g. Klassen & Durksen 2014; Flores 2015; Karaka? & Erten 2021).

1.3. What is missing?

Several studies have explored the constructs of teacher self-efficacy and instructional self-efficacy in the context of field practice. However, a significant limitation arises from their dependence on generic teacher self-efficacy measures, such as Teachers' Sense of Efficacy Scale (TSES) (Tschannen-Moran & Woolfolk Hoy, 2001), derivations thereof, or the Student Teachers’ Instructional Self-Efficacy in Technology-Rich Classrooms questionnaire (Elstad & Christophersen, 2017). The first is concerned with self-efficacy for teaching practices, but it was developed purely by experienced teachers and for use within this target group. Thus, while it contains three self-efficacy subscales: self-efficacy for instructional strategies, self-efficacy for classroom management, and self-efficacy for student engagement, these subscales were not intended for use with student teachers. The latter was indeed developed for student teachers. It only contains two 3-item scales concerned with self-efficacy: self-efficacy for maintaining discipline and self-efficacy for influencing students’ use of ICT in the service of learning, but both are concerned only with teaching practices related to the use of ICT. Thus, such measures fall short in capturing the unique beliefs and challenges experienced by student teachers during their field practice training. Typically, such studies frame questions within a general teaching context (e.g., “for teachers in their school context”), neglecting the unique and constrained context of student teachers’ experiences. Unlike professional teachers, who teach in an independent environment, student teachers’ experiences are likely confined to a controlled, learning-focused, and supervised setting (Cohen & Berlin, 2020). While the Instructional Self-Efficacy questionnaire does not suffer from this framing issue, it fails to tap into teaching practices not connected to the use of ICT.

            This lack in specificity is critical. According to Bandura (1997), self-efficacy only has predictive power when it is evaluated against specific tasks in a specific context or domain. Pajares (1996) further emphasized that self-efficacy judgments need to be domain- and task-specific to be predictively relevant. When questions are specifically tailored to the context of field practice, they are more likely to draw out responses that genuinely reflect student teachers' perceptions of their abilities in that particular environment. In contrast, using a general teaching context may result in responses that, at best, inaccurately represent student teachers' self-efficacy, either overestimating or underestimating their perceived capabilities, or at worst, rather than relevant context-specific self-efficacy beliefs tap into a generalized personality trait (Pajares, 1996).

            Despite the risk of misinterpretation inherent in using general measures among student teachers, there are currently no instruments specifically designed to measure student teachers' self-efficacy in field practice. This gap underscores the need for specialized, context-specific measurement tools that can accurately assess student teachers' self-efficacy, taking into account the unique challenges and tasks they encounter. A recent study by Nielsen and colleagues (2024) used focus groups and individual interviews with student teachers and their field practice teachers/mentors to identify 33 teacher and teaching-related practices that student teachers might doubt they have sufficient abilities and skills to engage in during field practice, i.e. that they might not feel efficacious towards.

1.3.1. The current study

The overall aim of the study was to fill this gap in the research on field practice-related self-efficacy in teacher education by developing a first and multi-dimensional instrument for the measurement of the various aspects of field practice self-efficacy for student teachers. This aim was fulfilled by:

• Collecting qualitative information on the teacher and teaching-related practices that student teachers might doubt they have the necessary abilities to engage in during field practice from a broad selection of teacher education stakeholders and formal curriculum sources, analyzing these data, and designing items based on the results.
• Collecting survey data with the developed instrument and conducting a first investigation of its psychometric properties using Rasch family Item Response Theory models, specifically considering local independence of items and differential item functioning relative to background variables that would plausibly be used in group comparison studies.

            With regards to initial criterion validity, it was expected that teaching experience prior to commencing the teacher education program would be positively correlated to all sub-constructs.

2 | Method

2.1. Initial instrument development (stage 1)

2.1.1. Data collection and sources of information

To uncover a sufficiently broad and detailed basis of information for the development of the field practice self-efficacy instrument, we collected qualitative data on the teacher and teaching-related practices student teachers could have difficulty believing they have sufficient abilities and competencies to enact/engage in, while in field practice, from a multitude of stakeholders and sources of information. Thus, we conducted two focus-group interviews with a total of 12 first year student teachers prior to their first field practice placement in schools (six students in each focus group). Students were recruited by on-campus teachers from classes, which includes preparing and following students through field practice. The focus groups were gender balanced, as one group had 50% female students and the other had 67% female students. Students’ ages were not recorded. The students had a varied educational background: from entering teacher education directly from high school to having a previously completed bachelor’s degree. In the focus groups, students were prompted to talk about teaching and teaching-related tasks that they thought they would find difficult or not difficult to believe they could engage in, while in field practice. Interviewers had a simple semi-structured interview guide as a backup if needed but were instructed to try to refrain from giving examples, and if they did so, to only mention an obvious activity as the actual teaching Each focus group interview was approximately 45 minutes long. We also conducted four single-person interviews with field-practice teachers from schools where students would be placed in field practice at all levels of the teacher education program. Three of the field practice teachers were female, and their experience as teachers ranged from one to 38 years. The field practice teachers were prompted to talk about teaching and teaching-related tasks that they thought the student teachers found difficult or not difficult to believe they could engage in during field practice. Again, the interviewers were instructed to interfere with the respondents’ responses as little as possible, only prompting further with a simple example of an obvious teaching task. Each of these interviews was approximately 15 minutes long.

            In addition to the interviews (above), we collected written information from four on-campus teachers in the first year of the teacher education program. The on-campus teachers were three female and one male teacher. Their ages ranged from 31 to 54 years. Three had an educational master’s degree on top of either the four-year teacher training program or a bachelor’s degree in educational science. The fourth campus teacher had a full bachelor and master’s degree in English language and history and a teaching degree on top. They had between two to 24 years of experience teaching in primary or secondary school and had six months to three years of experience teaching in the teacher education program. The on-campus teachers were asked to briefly list the teacher and teaching-related practices student teachers could have difficulty believing they have sufficient abilities and competencies to enact/engage in during field practice. A written account of the same was also elicited from a field practice coordinator, who was responsible for maintaining the pool of field practice schools, evaluating field practice with both students and field-practice teachers, and who thus had detailed knowledge of issues and insecurities. The field practice coordinator in question had ten years of experience in this role. 

            As the last source of information, we collected the national skills objectives for the field practice in the Danish teacher education programme (Act on the Education of Teachers of 2015). The objectives are included as Supplemental file 1 in an English translation.

2.1.2. Information extraction and content analyses

All interviews were recorded and transcribed. The interview data, the written accounts, and the skills objective document were analysed for the purpose of extracting all statements on teacher and teaching-related practices. The extracted statements were marked by source. The extracted statements were then subjected to content analysis, and statements with the same meaning were combined to form statements on unique practices.

            To obtain systematic and easily interpretable knowledge on the overlap and discrepancies of teacher and teaching-related practices across the twelve sources of information, the unique practices were recorded in a data matrix linking practices to information source (coded 1, if practice was mentioned by a source and 0 if not). This matrix was then used to generate an innovative co-mention circle using an adapted-for-purpose version of the R code provided by Fried (2016; 2020). The data matrix for generating the co-mention circle is publicly available at Zenodo.org (https://doi.org/10.5281/zenodo.12788471).

            The method for constructing the co-mention circle followed the approach outlined in Fried (2016; 2020), however, with two key adaptations in order to enhance functionality of the code. Firstly, Fried’s code, which utilized the data.table package (Dowle & Srinivasan 2023), was transitioned to the tidyverse package (v2.0.0; Wickham et al. 2019) to enhance clarity and readability of the code and make it more accessible to other users, as the tidyverse eco-system is well-documented and supported by a large, active community. Secondly, we introduced a new approach to determine the order of practices within the circle and across sources of information. Thus, practices were arranged anti-clockwise based on their frequency, starting with those most prevalent across sources and concluding with those mentioned by the fewest sources. In cases where practices shared equal frequency of co-occurrence, we applied a sorting criterion based on their presence in the outermost circle and moving inwards. In this way, we would first plot the practice with the highest frequency of co-occurrence, then the practice(s) with the second-highest co-occurrence, starting with the practice occurring most towards the outer rim of the co-mention circle, and so forth. In this manner, we achieved a co-mention circle with an inward-moving plotting of practices. The R code for the co-mention circle is publicly available at Zenodo.org (https://zenodo.org/doi/10.5281/zenodo.12788451).

2.1.3. Development of items and subscales

The statements on unique teacher and teaching practices were developed into items for the PSEQ questionnaire using a straightforward approach following the guidelines of Irwing and Hughes (2018). Thus, statements were condensed into items so that they could follow naturally and grammatically from the item stem, and so they could all be rated using a common response scale. The last step was to thematically divide the items into subscales and consolidate these with three context and subject experts, prior to collecting data for the initial validation analyses. 

2.2. Psychometric analyses of the initial Field Practice Self-Efficacy Questionnaire (stage 2)

2.2.1. Context and participants

The target population was Danish students in the newly reformed teacher education program (Act on the Education of Teachers of 2023), who had not yet completed field practice placement as part of their four-year long teacher education program. In the Danish teacher education, which is a four-year long programme leading to a Professional Bachelor’s degree, field practice amounts to 40 ECTS points (European Credit Transfer System), which is equivalent to two-thirds of a full-time study year. Field practice placements should take place in each of the four years of the program. In the first year of study, and possibly the second, field practice should consist of a prolonged period of time, for example with a period of time with one-day visits combined with a full-time block of field practice (Act on the Education of Teachers, 2023).

            All participants in the validation part of the study were from one Danish university college. Data were collected using targeted online surveys distributed to first-year students, either immediately prior to school visits (start of field practice) or after a period of time with a school visit day every week and just prior to the first full-time placement. The sample for the current validation study consists of 303 students, who had only completed the PSEQ (c.f. section 2.2.2 on the instrument) once; two-thirds just prior to school visits (no field practice experience) and one-third after school visits and just prior to starting the first full time placement of three weeks (Table 1).

            Students were divided equally on the two admission tracks to higher education in Denmark: grade-based and other qualifications (Table 1). The majority of the students (92.7%) were enrolled in the regular teaching education program, and the majority had chosen Danish as their major teaching subject (59.1%) (Table 1).

            Half of the students had teaching experience prior to starting the teacher education program, with an average of 19 months of experience (Table 1). The most common context for the teaching experience was in the public schools (40.9% of the sample, 80.0% of those with experience). Approximately two-thirds of the sample identified as female, while the mean age was 23.4 years.

2.2.2. Item analysis by Rasch and graphical loglinear Rasch models

The Rasch model (RM, Rasch, 1960) generalized to ordinal items (i.e. the Partial Credit Model; Masters, 1982), also known as the one-parameter logistic model, was chosen for the current study, as it is the simplest possible IRT model (Fischer & Molenaar, 1995) and due to its measurement properties (c.f. the below). The Rasch model for ordinal data adheres to the same requirements for measurement as the original dichotomous Rasch model (Kreiner, 2013; Mesbah & Kreiner, 2013), and thus we simply use the term RM for Rasch model in this article. The basic requirements for measurement by the RM are: unidimensionality, in the sense that the items of a scale measure a single underlying latent construct, e.g. class management PSE. Monotonicity, which means that the expected item scores increase with increasing values on the latent variable. Thus, when class management PSE

Table 1

Characteristics of the Initial Validation Sample (N = 303)

Notes. FP = Field practice.

^a includes STEM program, trainee program and distance (online) program.

^b specific major not provided

^c additional context not reported was peoples’ university.

increases, so should the expected scores on each of the items belonging to this scale. Local independence (or no local dependence; no LD), meaning that the item responses are conditionally independent given the latent variable. In other words, only the level of e.g. class management affects the responses to the questions, not the questions mutually. Absence of differential item functioning (no DIF), meaning that item responses and included exogenous variables are conditionally independent given the latent variable. That is, only the level of e.g. class management affects the responses to the questions and not variables outside the measurement model, such as gender or whether students have previous teaching experience, and so on. Homogeneity, meaning that the rank order of item parameters (item ‘difficulties’) is the same for all persons regardless of their level on the latent variable. Thus, the relative difficulty of the items would be the same for all persons, i.e. the item requiring the least self-efficacy to be endorsed would be the same for all, the hardest item to endorse would be the same for all, and so on. 

            Fulfilment of the first four of the above requirements ensures criterion-related construct validity as defined by Rosenbaum (1989). The additional fulfilment of the fifth requirement of homogeneity, which is only possible by fit to the RM (Kreiner, 2013; Rasch, 1960, 1961), means that the sum score is a sufficient statistic for the person parameter estimates.  Thus, the response profile of a person provides no additional information on the latent variable’s score than does the total score.

            With a newly developed instrument with many subscales, however, it is probable that they will not all fit pure RMs. Thus, we chose to also include graphical log-linear Rasch models (GLLRMs; Kreiner & Christensen, 2002, 2004, 2007), which are extended RMs. In GLLRMs it is possible to include DIF and LD by including interaction terms – between items in the case of LD and items and exogenous variables in the case of DIF – in the model, as long as the DIF and/or LD is uniform^[1]. 

            For models with evidence of uniform LD, we know that this entails somewhat lowered reliability of a scale than if the scale consisted only of conditionally independent items (Marais, 2013). This follows from the reduced variance in the item responses caused by items being more strongly correlated than what can be explained by the latent variable. Thus, if reliability calculations are not done in a manner taking into account any LD between items (e.g. Hamon & Mesbah, 2002), reliability will be spuriously inflated. LD, however, does not affect the sufficiency of the sum score (Kreiner & Christensen, 2007).

            For models with evidence of uniform DIF, the results indicate that item responses fit a RM for each of the subgroups for which one or more items function differentially. It is thus possible to “split for DIF” (Hagquist et al., 2009), which effectively means that person parameter estimates are estimated for each of these subgroups separately. Since these estimates are on the same logit scale (theta), they are still comparable and can be used in subsequent statistical analyses and for individual assessment, and essential validity is maintained (Kreiner & Christensen, 2007). Many practitioners prefer to work with the sum scores and not the person parameter estimates resulting from the RMs. As one of the properties of the RM, and only the RM, is that the sum scores are sufficient for the person parameter estimates, and there is a one-to-one relationship between these (Rasch, 1960), we utilize this property to equate the sum scores for DIF (Kreiner, 2007a) to bring also the sum scores back on the same scale for unbiased subsequent statistical comparisons (Kreiner & Nielsen, 2023).

            Furthermore, GLLRMs can be illustrated using chain graph models, where nodes represent variables and edges represent associations between variables. Thus, missing edges or arrows between nodes means that the variables in question are conditionally independent, given the remaining variables in the model. An arrow connecting two variables refers to a possible causal (at least in the sense of time) relationship as the association between them is directed, while undirected edges illustrate that the variables are conditionally dependent without assuming causality (Lauritzen, 1996, provides details of such models). The same rules apply to the items in a GLLRM (i.e. the measurement part of the GLLRM; Kreiner & Christensen, 2002; Nielsen & Santiago, 2020). Thus, items which are not connected by an edge are conditionally independent given the latent variable (i.e., items are locally independent). In the same manner, items which are not connected to an exogenous variable means that items and exogenous variables are conditionally independent given the latent variable and the other variables in the model (i.e., there is no DIF). The causal relationship between the latent variable and items is illustrated by arrows from the first to the latter, as are associations between exogenous variables and the latent variable. The GLLRM can also be used for assessing criterion validity, as it also contains a graphical model which allows for analysis of association between the exogenous variables (as independent variables) and the latent score (the dependent variable) (Kreiner & Christensen, 2002; Nielsen & Santiago, 2020).

2.2.3. Strategy of analyses and statistics used

The same strategy of analyses was used for each of the seven subscales. First, we tested fit of the set of item responses for a scale to the RM. If this was rejected, we catalogued departures from the RM in the form of LD and DIF in order to define a GLLRM with log-linear interactions for uniform LD and/or DIF and tested the fit to this GLLRM. If this was not successful, the items deemed most problematic were eliminated, and the iterative process started again. Each “test of fit” included several steps in order to test the fit as rigorously as possible:

• Fit of the individual items to the specified model by conditional infit and outfit statistics (Christensen, 2013; Kreiner & Nielsen, 2013) and by comparing the observed item-restscore correlations with the expected item-restscore correlations under the model (Kreiner, 2011).
• Tests of local independence for all item pairs using the same tests as for DIF (see below).
• Overall test of homogeneity of the item parameters across low- and high-scoring groups using Andersen’s (1973) conditional likelihood ratio (CLR) test. 
• Overall tests of invariance relative to exogenous variables considered relevant for group comparisons in further studies using Andersen’s (1973) conditional likelihood ratio (CLR) test. 
• Tests of no DIF for all items relative to the same exogenous variables using Kelderman’s (1984) CLR test of local independence as well as tests of conditional independence given scores using partial Goodman-Kruskal gamma coefficients (Kreiner & Christensen, 2004).

For tests of overall subscale invariance and DIF, the largest subgroup was used as the reference group in each exogenous variable. The exogenous variables were Time of completion of the PSEQ (prior to all field practice, after school visits but before full time field practice), Admission track (grade-based, other qualifications), Prior teaching experience (no, yes), Gender (female, male), and Age (-22, 23+).

            When the final iteration was completed and fit was achieved for a subscale, analyses were concluded with the evaluation of targeting and reliability relative to the validation sample, as well as estimation of person parameters from the final model using weighted maximum likelihood estimation (Warm, 1989), as these are known to provide less biased estimates for the RM than other methods (Kreiner & Christensen, 2013). Hamon and Mesbah’s (2002) Monte Carlo method for reliability of Rasch scales was used, as this method takes into account any LD between items. Targeting of theta was assessed by two indices, which should both have a value close to one: the test information target index as the mean test information divided by the maximum test information for theta, and the root mean squared error (RMSE) target index, as the minimum standard error of measurement (SEM) divided by the mean SEM for theta (Kreiner & Christensen, 2013). In addition, the target and SEM of the observed score was calculated.

            The Benjamini-Hochberg procedure was used to adjust for false discovery rate (FDR) due to multiple testing (Benjamini & Hochberg, 1995). In line with Poulsen et al. (2018), we followed Cox et al.’s (1977) recommendation to abstain from the deterministic decision criterion of 5% critical limit for p-values, and a distinction between weak to moderate evidence against the model (p > .01) and stronger evidence (p < 0.1) was used instead.

2.2.4. Criterion validity

Two modes of analyses were conducted to assess in a simple manner the criterion validity of the subscales in the PSEQ. First, the criterion variable (i.e. whether the student teachers had teaching experience prior to commencing the teacher education program or not) was included in the analyses by graphical loglinear Rasch models both as an exogenous variable for DIF-analyses in the measurement part of the model and as an endogenous variable in the inference part of the model (Nielsen & Santiago, 2020). Partial gamma correlation coefficients (Goodman & Kruskal, 1954) were used to inform any association between the criterion variable and the seven subscale scores, while taking into account associations with the other background variables in the model (Kreiner & Christensen, 2002, 2004, 2007). Second, mean subscale scores, adjusted for any discovered DIF, were compared for with and without prior teaching experience using Student’s t-tests for independent groups with 1000 bootstrap samples.

2.3. Software

Co-occurrence circles and item maps were created using R. Item analyses were conducted using the DIGRAM software package (Kreiner, 2003; Kreiner & Nielsen, 2013; 2023), and descriptives and Student’s t-test and effect sizes were done using SPSS.

3 | Results

3.1. Instrument development (stage 1)

3.1.1. Extracted information from qualitative sources and content analyses

The extraction of information from the interviews, written statements and the document with skills objectives resulted in a total of 70 raw statements on teacher and teaching-related practice that student teachers might have difficulty believing that they could enact or engage in during field practice. These statements were subsequently subjected to a content analysis. Statements with the same meaning (i.e., redundant statements) were combined, or one was chosen to represent them in the instrument, to form a total of 42 statements on unique practices. For example, the statements “know what it takes to help the weaker pupils” and “scaffold the weaker pupils” were reduced to the former statement, as this conveyed the meaning of both. As the meaning of the raw statements was very clear, this step was done by a single researcher with both content and test development expertise. An overview of how the 42 statements were mentioned by the various sources of information is shown in Figure 1.

3.1.2. Initial item pool and thematic subscales

The items were designed from the 42 unique statements on teacher and teaching-related activities that students might doubt they could engage in during field practice, so that they followed naturally from the item stem “In your upcoming field practice, how confident are you that you have sufficient abilities to …”, and could be rated using the response scale; very confident (3), confident (2), not particularly confident (1), not confident at all (0). Examples are the statements “cope with boundary-pushing children”, which resulted in the item “... deal with pupils who test boundaries”, and the statement “be prepared enough”, which resulted in the item “… be adequately prepared”. The items were developed in Danish and are shown in supplemental file 2, along with an English translation as well as both the numbering in which we refer to them in the article and the order in which they were presented in the questionnaire.

The items were thematically divided into subscales and subsequently consolidated by three context and subject experts. This resulted in seven initial subscales belonging to three domains of practice: Teaching-related PSE with the subscales Planning and preparation (7 items), Teaching in itself (7), Class management (7), and Differentiation (6); Relational PSE with the subscales Pupils (5) and Adult collaborators (parents and co-teachers) (4); and Evaluative and developmental PSE with the single subscale Evaluation and development (6). Figure 1 presents an overview of how the 42 statements and items were mentioned by the various informants and also illustrates the division into subscales.

Figure 1

Overlap of Statements/Items on Teacher and Teaching-Related Practices across the 12 Sources of Information Divided into Subscales

Notes. S-groups = student focus groups, FPT = field-practice teachers, CT = Campus teachers, FPC = Field practice coordinator, Obj = official skills objectives for field practice.

A result that stands out in Figure 1 is the variation in the types of sources that have contributed information on teacher and teaching-related practices that student teachers might have difficulty believing they could engage in during field practice. The Planning and preparation, the Differentiation, and the Pupil subscales have been contributed to by all types of informational sources. At the other end of the spectrum, we see the Adult collaborators, to which neither students nor campus teacher contributed, as well as the Evaluation and Development subscale, where neither students, field-practice teachers, nor campus teachers have contributed information.

At this stage, we chose to eliminate statement 5 (... practice class management) from the Class management subscale, as it generally does not make sense to include a single item referring directly to the latent construct of the scale it belongs to. Thus, the Class management subscale was reduced to six items, and we moved forward with a total of 41 items.

The distribution of subscale scores shows that in the Adult collaborators subscale the full score range is used by students. In the remaining subscales, almost the full score range is being used, with only the lowest two to four not being utilized. While there is a slight skewness towards the higher end of the scales, they approximate a normal distribution well (Figure 2). The most right-skewed scores are in the teaching in itself and the pupils subscales, and the least skewed is the Adult collaborators subscale. 

3.2. Item analyses (stage 2)

3.2.1. The final PSE subscale from psychometric analyses

Initial analyses rejected fit to the pure RM for all but one scale: the Differentiation subscale from the teaching-related PSE domain. Further analyses to discover specific departures from the Rasch model revealed that most departures consisted of items that were locally dependent. This was the case for the Planning and preparation and Teaching in itself subscale from the teaching-related PSE domain, the Adult collaborators subscale from the relational PSE domain, and the Evaluation and development subscale making up its own domain. For a single scale, the Class management subscale from the teaching-related PSE domain, the departure from the pure RM consisted of DIF. In addition, item 13 (... get the pupils to like me) was eliminated from the Pupils subscale from the relational PSE domain, as it was involved in two instances of local dependence and there was weak evidence against fit of this item, none of which could be resolved. Looking at the item content, it was concerned with relations to the pupils, but in a very general manner, which is likely the cause of the mentioned issues.

Following the elimination of item 13, and previously item 5, further analyses showed that two subscales fitted pure RMs: the 6-item Differentiation subscale and the 4-item scale Pupils subscale. There was no evidence against global homogeneity and no evidence against global invariance relative to time of completion of the PSEQ, admission track, prior teaching experience, age, or gender, except for weak evidence against gender invariance of the Pupils subscale (Table 2 and Table 3). However, as this evidence was weak, and there was no evidence of gender DIF, we leave this matter to be investigated further in future studies. There was no evidence against the fit of individual items (Appendices 6 and 8). The remaining five subscales all fitted simple GLLRMs.

The Planning and preparation subscale GLLRM only included strong local dependence between items 40 (... make plans for single lessons) and 41(... plan short teaching modules), and between items 41 and 42 (... plan long teaching modules) (Figure 3). There was no evidence against global homogeneity or against global invariance relative to the five exogenous variables (Table 2). There was no evidence of DIF or additional local dependence among items, nor was there evidence against the fit of individual items or the component item resulting from the local dependence between items 40, 41 and 42 (Appendix 3).

Figure 2

Distribution of Raw Sum Scores on each of the Seven PSEQ Subscales Prior to Item Analysis

Notes. Item 5 from the Class management subscale has been eliminated at this point. X-axis covers the score range of each subscale, while the Y-axis frequency range was set to 0-60.

The Teaching in itself subscale fitted a very simple GLLRM, which only included strong local dependence between item 25 (... "manage on my own" without the field-practice teacher) and item 30 (... stand in front of the pupils and teach) (Figure 3, Appendix 4). There was no evidence against global homogeneity and no strong evidence against global invariance relative to the five included exogenous variables (Table 2). There was no evidence against the fit of individual items or the component item representing the two locally dependent items, and no evidence of DIF or further local dependence between items (Appendix 4).

The Evaluation and development subscale fitted an equally simple GLLRM as the Teaching in itself subscale did (c.f. the above), as the only departure from the RM consisted of strong local dependence between items 33 (... evaluate my own teaching) and 35 (... contribute to the development of my fellow students' teaching practice on the basis of our common experiences) (Figure 5). There were no issues with evidence against global homogeneity or global invariance (Table 3), nor evidence against item fit, evidence of DIF or additional local dependence between items (Appendix 9).

The Adult collaborators subscale fitted the most complex GLLRM of all the subscales, as three of the four items were locally dependent and strongly so: items 10 (… write to/with parents) and 11 (… speak to/with parents), and items 11 and 12 (... contribute to parents' evening/parents-teacher meetings) (Figure 4). There was no evidence against global homogeneity and no evidence against global invariance relative to the five exogenous variables included (Table 3). There was no evidence against the fit of individual items or the component item representing the three locally dependent items, no evidence of DIF or further local dependence between items (Appendix 7).

The Class management subscale was the only subscale with evidence of DIF; item 4 (... be able to get the class to be quiet) functioned differentially relative to prior teaching experience, so that students with prior teaching experience systematically indicated greater confidence in being able to get the class to be quiet than students without prior teaching experience independent of their level of Class management self-efficacy. When including this DIF as an interaction in the subscale, there was no further evidence against fit to the GLLRM (Figure 3) of any kind, (Table 2 and Appendix 5).

In addition to documenting item fit and the necessity of any LD or DIF interactions, Appendices 3 through 9contain conversion tables with the necessary information for converting sum scores to weighted maximum likelihood estimates of the person parameters for each subscale. Also, the Appendices show that the bias of these estimates is very small except for at the extremes, where bias is always high (Table A3d though A9d). Thus, the person parameter estimates or the summed raw scores can be used as fitting for purpose.  For the Class management subscale, an additional conversion Table (A5e) was added for the 5-item DIF-free scale, so that all subscales may be used without having to equate sum scores or split the person parameter estimates to overcome DIF.  

Table 2

Global Tests of Homogeneity and Invariance for the Four Teaching-Related PSE Subscales of the PSEQ under the Final RMs or GLLRMs

Notes. RM: Rasch model; GLLRM: Graphical loglinear Rasch model; CLR: Conditional likelihood ratio; df: degrees of freedom; p: p-value. Global homogeneity test compares items parameters in approximately equal-sized groups of high and low scoring students. The critical limits for the p-values after adjusting for false discovery rate were: ⁺ 5% limit p = .0083, 1% limit p=0.0017.

^a There is local dependence between items 40 and 41, and between items 41 and 42.

^b There is local dependence between items 25 and 30.

^c Item 4 functions differentially relative to prior teaching experience.

^d Item 5 was excluded from the final subscale.

Figure 3

The Final Rasch and Graphical Loglinear Rasch Models for the Four Teaching-Related PSE Subscales; Planning and Preparation (Top Left), Teaching in Itself (Top Right), Class Management (Bottom Left), Differentiation (Bottom Right)

Notes. γ-correlations are partial Goodman and Kruskal’s rank correlations for ordinal data. Adm = Admission track (grade-based vs. other qualifications), Exp = Prior teaching experience (no vs. yes), Time = time of completion of PSEQ (Prior to school visits (no FP) vs. after school visit, prior to full time FP), Sex (female vs. male), Age (22 years and younger vs. 23 years and older).

Table 3

Global Tests of Homogeneity and Invariance for the Two Relational PSE Subscales and the Evaluative and Developmental PSE Subscale of the PSEQ under the Final RM or GLLRMs

Notes. RM: Rasch model; GLLRM: Graphical loglinear Rasch model; CLR: Conditional likelihood ratio; df: degrees of freedom; p: p-value. Global homogeneity test compares items parameters in approximately equal-sized groups of high and low scoring students. The critical limits for the p-values after adjusting for false discovery rate were: ⁺ 5% limit p = .0083, 1% limit p=0.0017

^a There is local dependence between items 10 and 11, and between items 11 and 12.

^b Item 13 was excluded from the final subscale.

^c There is local dependence between item 33 and 35.

Figure 4

The Final Rasch and Graphical Loglinear Rasch Models for the Two Relational PSE Subscales; Adult Collaborators (Left), Pupils (Right).

Notes. γ-correlations are partial Goodman and Kruskal’s rank correlations for ordinal data. Adm = Admission track (grade-based vs. other qualifications), Exp = Prior teaching experience (no vs. yes), Time = time of completion of PSEQ (Prior to school visits (no FP) vs. after school visit, prior to full time FP), Sex (female vs. male), Age (22 years and younger vs. 23 years and older).

Figure 5

The Final Rasch and Graphical Loglinear Rasch Models for the Evaluation and Development Subscale Making Up the Evaluative and Developmental PSE Domain.

Notes. γ-correlations are partial Goodman and Kruskal’s rank correlations for ordinal data. Adm = Admission track (grade-based vs. other qualifications), Exp = Prior teaching experience (no vs. yes), Time = time of completion of PSEQ (Prior to school visits (no FP) vs. after school visit, prior to full time FP), Sex (female vs. male), Age (22 years and younger vs. 23 years and older).

3.2.2. Effect of DIF in the Class management subscale

To evaluate the effect of the DIF on item 4 in the Class management subscale, the summed scores were adjusted by DIF-equating via the person parameter estimates, thus making the DIF-adjusted scores comparable across prior teaching experience subgroups and facilitate true group comparisons (Appendix 5).

Equating he sum score for the teaching experience DIF in the Class management subscale did not change the conclusion of a significant difference in mean scores dependent on prior teaching experience, only the magnitude of the difference was slightly reduced (Mean difference_observed  = 1.30, Mean difference_adjusted = 1.11) (Table 4).

Table 4

Comparison of Observed and DIF-Adjusted Mean Class Management Scores for Student with and without Prior Teaching Experience

Notes. SE, Standard error.

At the individual level, DIF-equating of the Class management scores meant that scores were reduced by up to half a point for students with prior teaching experience to obtain scores that were on the same scale as those for students without prior teaching experience. With normal rounding rules to achieve integer scores, the lower half of the score range would be reduced with a whole point for students with teaching experience prior to commencing teacher education. Thus, DIF-equating or splitting for DIF is necessary before using the sum scores and the person parameter estimates, respectively, in statistical analyses (Appendix 5).

3.2.3 Targeting and reliability

Six subscales were well-targeted to the study sample, with test information target indices indicating that 74% to 92% of the maximally obtainable information was obtained. The exception was the Planning and preparation subscale, which only had acceptable targeting with approximately 65% information obtained (Table 5). RMSE target indices were close to one for all subscales, indicating that standard error of measurement for theta was close to the lowest possible for the scales. For most of the subscales, the item thresholds covered the same range on the scales as did the person parameter estimates of the students (c.f. the item maps in Figures A3 through A9 in Appendices 3 through 9). The item maps also showed that, for all but the Evaluation and development subscale, the test target value (i.e. where most information is provided) was located at the lower end of the scales and thus lower than the student mean scores on the latent variables. Lastly, the item maps illustrate that the information level is relatively high across the part of the scales where the majority of the students are located.

Person separation indices for theta ranged from 0.72 to 0.81, while reliability (taking into account any locally dependent items) of the scores ranged from 0.66 to 0.80. Only the two Relational PSE subscales, Pupils and Adult collaborators, had reliabilities lower than 0.70 (Table 5).

Table 5

Targeting and Reliability of the Seven Subscales of the PSEQ (N = 303)

Note. q, theta; Targeting and reliability is provided for groups defined by differential item functioning (DIF) variables and subgroups in variables associated with the latent PTSD variable; Tl, test information; RMSE, root mean squared error of the estimated theta score; SEM, standard error of measurement of the observed score; r, reliability; Plan, Planning and preparation; Teach, Teaching in itself; Cman, Class management; Exp, teaching experience; Diff, Differentiation; Adult, Adult collaborators; Eval, Evaluation and development. 

^a In case of differential item function, results are shown for subgroups.

^b Weighted mean PSI across subgroups = 0.77, weighted mean reliability = 0.75.