Skip to main content
NCBI home page
Wiley Open Access Collection logoLink to Wiley Open Access Collection
. 2024 Dec 29;95(Suppl 1):S15–S31. doi: 10.1111/bjep.12732

‘No words’—Machine‐learning classified nonverbal immediacy and its role in connecting teacher self‐efficacy with perceived teaching and student interest

Rebecca Lazarides 1,2,✉, Jonas Frenkel 1,2, Uroš Petković 2,3, Richard Göllner 4, Olaf Hellwich 2,3
PMCID: PMC12427166  PMID: 39737730

Abstract

Background

Much is known about the positive effects of teachers' self‐efficacy on instruction and student outcomes, but the processes underlying these relations are unknown.

Aims

We aimed to examine the effects of teacher self‐efficacy for student engagement (TSESE) before a lesson on teachers' nonverbal immediacy (NVI) and their enthusiastic teaching. Furthermore, we examined how NVI and enthusiastic teaching affected students' interest after the lesson, controlling for prior interest.

Sample

We used data from the German TALIS video study in the context of the international TALIS study. The study included 50 teachers (46% women) and their 1140 students (53% girls; ageM = 15 years).

Methods

We developed a computational model to assess teachers' NVI on classroom video data. Using a multimodal longitudinal approach, we tested sequential processes with multilevel path models.

Results

TSESE before the lesson (Time 1) was positively and significantly related to teachers' NVI during the lesson (Time 2). Teachers' NVI (Time 2) was positively related to class‐level enthusiastic teaching behaviours, reported after the lesson (Time 3). Student‐reported enthusiastic teaching behaviours (Time 3) were significantly and positively associated with students' interest (Time 3) when controlling for students' prior interest (Time 1). Students' interest after the lesson (Time 3) was significantly and positively related to students' interest 6 weeks later (Time 4).

Conclusions

Nonverbal behaviours of the teacher are central to classroom instruction by promoting students' perceptions of the teachers' enthusiastic teaching behaviours.

Keywords: machine learning, motivation, nonverbal behaviours, student interest, teacher self‐efficacy

INTRODUCTION

Teachers' motivations to teach are central to high‐quality instruction and students' academic outcomes (Fives & Buehl, 2016; Lazarides & Schiefele, 2021; Watt & Richardson, 2008). One important component of teacher motivation is teachers' self‐efficacy, that is, their judgement of their own capability to bring about desired outcomes of student engagement and learning, even when students are difficult or unmotivated (Tschannen‐Moran & Woolfolk Hoy, 2001). Although much is known about the positive effects of teacher self‐efficacy on instruction and student outcomes (Kim & Seo, 2018; Klassen & Tze, 2014), it remains to be answered through which processes teacher self‐efficacy is transmitted to students (Lauermann & ten Hagen, 2021). Based on the assumptions of social cognitive theory (Bandura, 1997), in this study, we examine sequential effects from teacher self‐efficacy for student engagement (TSESE) to teachers' in‐class behaviours to students' interest. Concretely, we examine whether TSESE positively affects teachers' nonverbal immediacy (NVI), objectively assessed via machine learning and student‐reported teachers' enthusiastic teaching behaviours, ultimately resulting in increased student interest. Nonverbal immediacy (NVI) can be conceptualized as the entirety of behaviours of individuals that convey positive signals to communication partners, such as sympathy and warmth, thereby reducing the psychological distance between them and fostering a sense of closeness and accessibility (Mehrabian, 1968; Richmond et al., 2003). Teachers' NVI is closely related to enthusiastic teaching behaviours (Keller et al., 2016). Using a multimodal longitudinal methodological approach, we assess teachers' NVI using a computer vision‐based computational model (Petković et al., 2024) and combine it with teacher reports of TSESE and student reports on teachers' enthusiastic teaching behaviours and students' interest.

Teacher self‐efficacy for student engagement: Relations to instructional processes

Teacher self‐efficacy for student engagement (TSESE) assesses the teachers' own evaluations of their capability to help students value learning as well as to motivate students, particularly those who show low interest in learning (Tschannen‐Moran & Woolfolk Hoy, 2001). Tschannen‐Moran and Woolfolk Hoy (2001) differentiate three dimensions of teacher efficacy—for student engagement, for instructional strategies and for classroom management. These dimensions are independent subdimensions of teachers' general self‐efficacy, and thus, can be seen as a task‐specific assessments that reference a particular area of teaching (Lauermann & ten Hagen, 2021). In our study, we refer to TSESE because this dimension of teacher self‐efficacy refers to the domain of motivating students, and we were interested in motivational processes in class and the prediction of student interest in this study.

TSESE has been shown to positively relate cross‐sectionally to teacher‐reported classroom climate (Fackler et al., 2021) and to positively relate longitudinally both to teacher‐reported relationships with students (Lazarides, Watt, & Richardson, 2023) and to students' class‐level interest via student‐perceived emotional support in class (Lazarides, Schiefele, et al., 2023). However, the processes underlying such relations are unknown. On a theoretical level, it has been proposed that TSE guides teachers' goal direction, their decision‐making and their verbal and nonverbal engagement in aiming to realize their instructional goals (Tschannen‐Moran et al., 1998). Thus, it could be assumed that the link between teachers' TSESE and teaching styles or student outcomes is the teachers' motivating nonverbal and verbal actions (Richmond et al., 2003). However, existing work has not directly addressed the theoretically assumed mechanisms through which teachers' self‐efficacy finds its correspondence on the side of students. Most importantly, the existing approaches may not be sophisticated enough to capture all the nuances of teacher behaviours during a lesson.

Teacher non‐verbal immediacy, teaching behaviours and student motivation

In the context of education, teacher immediacy encompasses highly situation‐specific verbal and nonverbal behaviours, such as maintaining eye contact with students, using positive facial expressions, employing a relaxed body language or optimizing physical proximity to engage with students (Andersen et al., 1979; Chesebro & McCroskey, 2001). In our study, in accordance with Witt and Wheeless (2001), we define NVI as a composite of the teachers' proximity (approaching students vs. keeping distance), body posture (open, relaxed vs. closed, formal), gestures (using hands and arms vs. stiff, infrequent or ineffective) and facial expression (pleasant, smiling vs. frowning or lack of expression). Meta‐analytical evidence showed on average moderate effect sizes for associations between NVI and students' affective learning (r = .48 in Harris & Rosenthal, 2005; r = .49 in Witt et al., 2004) and small effects on cognitive performance (r = .11 in Harris & Rosenthal, 2005; r = .17 in Witt et al., 2004).1 A recent literature review (Liu, 2021) also revealed that NVI is a critical factor for learning processes, particularly in the motivational‐affective domain. It is, however, unclear which processes underly the relations between NVI and student outcomes.

We propose that teachers' NVI increases students' interest by fostering their perception of higher levels of enthusiastic teaching behaviours. This assumption is based on a social cognitive perspective on learning and teaching proposing learning as a reciprocal interaction between a person's cognitive and affective processes and behavioural and environmental influences (Bandura, 1986). Thus, the individual's perception of environmental factors, such as the teachers' behaviours, matters for motivational processes of the individual (and vice versa). Accordingly, research has shown that the perspective of students about their teaching are an important intermediary between the actual behaviours of teachers and the actual performance of learning activities by each student (den Brok et al., 2006). Accordingly, we examined how the teachers' behaviours—conceptualized as machine‐learning assessed nonverbal immediacy, affect the students' perceptions of enthusiastic teaching, which in turn was assumed to affect the students' interest in the content taught.

Whereas immediacy refers to teachers' verbal and nonverbal behaviours that enable teachers to increase psychological closeness with students (Mehrabian, 1968), enthusiastic teaching behaviours have been defined as a culmination of expressive behaviours including, for example, vocal delivery, eye contact, demonstrative gestures or body movements that signal an overall high level of energy (Collins, 1978). A decade ago, Keller et al. (2014) distinguished teacher enthusiasm into the perceived inner state of the teacher (‘positive affect’) and the displayed behaviours (‘positive emotional expressivity’) of the teacher. Following theoretical concepts of emotional contagion (Hatfield et al., 1993), teachers convey their emotional states to their students through their behavioural expression as individuals have ‘the tendency to automatically synchronize expressions, vocalizations, postures, and movements with those of another person's’ (page 96). Applying such theoretical assumptions to the context of teaching, our assumption is that teachers use nonverbal social cues in a lesson to communicate, and that students perceive and process such social signals, in turn, influencing their own motivations. Thus, in our study, we focus on lesson‐specific interactions, which occur in specific situations and learning settings and are part of long‐term interactions that may span several weeks, a school year or even multiple school years.

Aims and research questions

Based on conceptual models of TSE and its effects on teaching performance (Tschannen‐Moran et al., 1998), we assumed that TSE before the lesson (Time 1) would positively and directly relate to teachers' nonverbal immediacy during the lesson (Time 2) (Hypothesis 1).

In line with the theoretical approaches examining nonverbal immediacy (Mehrabian, 1968; Richmond et al., 2003), we expected that teachers' NVI during the lesson (Time 2) would positively and directly relate to students' perceptions of the teachers' enthusiastic teaching behaviours during the lesson, assessed directly after the lesson (Time 3) (Hypothesis 2). The hypothesised model along with the time sequence of the tested variables is depicted in Figure 1.

FIGURE 1.

FIGURE 1

Open in a new tab

Schematic depiction of the hypothesised model.

Referring to research revealing the positive effects of teachers' enthusiastic teaching behaviours on student motivation (Frenzel et al., 2018), we expected that student‐rated enthusiastic teaching behaviours (Time 3) would be positively associated with students' mathematics interest directly after the lesson (Time 3) when controlling for students' interest before the lesson (Time 1). Additionally, we aimed to examine whether these expected effects would also persist for students' interest 6 weeks later (Time 4) (Hypothesis 3).

We included teachers' years of experience as a covariate because empirical work indicates that teaching experience is significantly and positively related to teachers' self‐efficacy for student engagement (Lazarides, Watt, & Richardson, 2023), and negatively related to teachers' enthusiastic teaching style (Burić et al., 2023).

METHOD

Sample

We used data from the German TALIS video study (Grünkorn et al., 2020), which was funded by the Leibniz Association (2017–2020) in the context of the international TALIS study (Teaching and Learning International Survey; OECD, 2020). The study included 50 teachers (46% women; ageM = 43 years) and their 1140 students (53% girls; ageM = 15 years) who were distributed across grades 8 (10%), 9 (79%) and 10 (11%). The majority of the 38 participating schools were academic‐track schools (‘Gymnasien’; n = 30), followed by comprehensive schools (‘Gesamtschulen’; n = 4), secondary schools (‘Realschulen’; n = 2), an upper secondary school (‘Oberschule’) and a vocational school (‘Berufsschule’). The TALIS video study focused on mathematics lessons in grades 8–10 in public schools in Germany. We used data from the longitudinal pre‐, post‐ and follow‐up tests of the study. The pretest included standardized surveys for students and teachers and was conducted before the lesson (Time 1). Three lessons selected by the 50 participating teacher were recorded on video (Time 2). The post‐test was conducted immediately after the lesson (Time 3). The follow‐up survey was conducted approximately 6 weeks after the lesson (Time 4).

Measures

NVI model

The NVI of the teachers was assessed using a computer vision‐based machine learning model. Reflecting the multimodal nature of nonverbal social communication, the architecture of the model comprised submodels assessing relevant behavioural dimensions (gesture intensity, perceived physical distance, facial expression), as well as an overarching main model that computationally integrated the individual ratings into an overall NVI score. Our reason for developing a machine learning approach to study teachers' NVI was to make the assessment more automated and efficient. The pipeline summary of the development of the NVI score is depicted in Figure 2 (black rectangles are only included for reasons of anonymization in this manuscript).

FIGURE 2.

FIGURE 2

Open in a new tab

Visualization of the preprocessed video inputs. (a) Red, green, blue (RGB) video input; (b) video segmentation, identification of the teacher and students in the video data; (c) estimation of facial emotional display; (d) generated depth image of the classroom situation.

From the 50 teachers in the original sample, four classrooms were excluded from the analysis, either because no video material was available or because the video quality was deemed insufficient for computer vision‐based analysis. Consequently, videos from a total of 46 teachers were utilized. For N = 43 teachers, three suitable video recordings of one lesson each were available for analysis. For the remaining n = 3 teachers, only two videorecorded lessons were available, resulting in a total of 135 videorecorded lessons used for the analysis. The video‐recorded lessons focused on quadratic equations taught to students in grades 9 and 10. In the first step, segments that were suitable for computer vision analysis were identified. Included were segments without excessive camera movement where the teacher delivered frontal presentations, remained fully visible and did not leave the video frame. Following these criteria, a total of 403 30‐s segments were used to train and validate the model.

In the second step, human raters received an intense training on rating body posture, gesture intensity, perceived physical distance between the teacher and the nearest student and NVI in the videos. A total of three raters were involved in this step and participated in a rater training. The NVI was evaluated based on the unprocessed RGB videos and each video was evaluated by each rater. The raters were students and employees of a German university and were familiar with the typical teaching situations that occur in German educational institutions. The data were labelled by the human raters in terms of these criteria variables. To this end, we randomly selected 3056 frames from the total sample of 403 video segments (302,250 frames). Each of the raters evaluated each of the 3056 frames on a continuous slider scale with regard to body posture, gesture intensity and perceived distance. In addition, raters judged teachers' overall NVI based on the full 30‐s segments using a continuous slider scale. Each of the raters evaluated teachers' NVI on a slider scale for all 4002 video segments (the prompt was ‘Please rate the teachers’ nonverbal immediacy). To test the interrater reliability of the human ratings, correlations among the judgements of the three raters were analysed with a two‐way random model (ICC(2,3)). In a two‐way random model, raters are assumed to be randomly selected from a larger population, and each rater rates each individual frame (Shrout & Fleiss, 1979). The rating process revealed a low reliability of ratings for body posture in terms of the teacher's relaxation (ICC(2,3) = .59), even after joint discussions and several rounds of ratings, which is why we excluded this variable from the subsequent analyses. There was an acceptable interrater agreement for distance with an ICC(2,3) of .68 and for NVI with an ICC(2,3) of .68, and a good interrater agreement for gesture intensity, ICC(2,3) of .89 (Cicchetti, 1994). In the third step, the labelled data were used to train and validate the model. For this process, the original 403 video segments for the human rating process were split into a training (324 segments) and a validation (79 segments) data set. Segments lacking sufficient agreement among human raters were excluded from the training data set. Consequently, the following segments/frames were included: gesture intensity and perceived distance, 324 segments (1604–2121 frames); NVI, 316 segments. For the validation data set, all segments were retained. To increase the validity of the model as well as to ensure the generalizability of the NVI assessments, a cross subject (X‐Sub) evaluation protocol was employed, meaning that each teacher was either part of the training data set or part of the validation data set, but never in both. This procedure, commonly used in computer vision for human behaviour analysis (Koelstra et al., 2011; Lin et al., 2020; Shahroudy et al., 2016), is crucial to ensure that the computer vision‐based analysis was actually based on the relevant non‐verbal cues, rather than memorizing individual teachers' behaviours. The division thus allows for assessing NVI in an unseen data set, meaning that it can generalize to new data without prior exposure to the individual teachers. For preprocessing, segmentation masks of the video material were generated to identify the position of both teachers and students. Furthermore, a fine‐tuned ResNet architecture (He et al., 2016) was developed to improve the video quality for the facial expression analysis as well as the gesture intensity and distance models. To allow for reliable distance estimations from 2‐D video data, a depth image was created from the RGB frame (Oquab et al., 2023), enabling the model to recognize spatial relationships between teachers and students. The regressor models for gesture intensity and perceived distance were trained using the human ratings. Additionally, facial expression analysis was performed on the basis of teachers' segmentation masks for the complete sample of 403 segments using the HSEmotion library (Savchenko, 2021). Facial expressions were operationalized via theory‐based dimensions of facially expressed emotional states, including anger, contempt, disgust, fear, happiness, neutrality, sadness and surprise. Finally, the overall NVI model integrated the outputs of the gesture intensity and distance models, as well as the facial expression analysis into a 10‐dimensional vector space, which was then fed into a multilayer perceptron model (Rumelhart et al., 1986) with three fully connected layers. The preprocessed video inputs are depicted in Figure 3.

FIGURE 3.

FIGURE 3

Open in a new tab

Pipeline summary. The system processes RGB videos to perform tracking, segmentation and depth estimation. Teacher's facial expressions and gesture intensity are estimated from segmentation mask data. Estimates of perceived distance are generated using depth images and segmentation masks of both teachers and students.

The subsequent validation of the NVI model demonstrated its potential in nonverbal behaviour analysis, approximating the accuracy of individual human raters. The ICC value that represents the correlation among the human ratings of the overall NVI of all three human raters and the NVI algorithm using the validation data set was .690. For further information about the NVI model and its subcomponents, see Petković et al. (2024).

Having confirmed the model's ability to accurately assess teachers' NVI, it was then applied to analyse an additional 220 30‐s segments from the TALIS video corpus. Consequently, NVI data from a total 623 video segments were used for the subsequent path analyses.

Survey measures

Teacher self‐efficacy for student engagement (Time 1)

The four‐point Likert scale ranged from 1 (not at all) to 4 (very much) and assessed class‐specific teacher self‐efficacy for student engagement in line with Tschannen‐Moran and Woolfolk Hoy (2001). The scale instruction was ‘How much can you do the following in your class?’ followed by four items. An example item is ‘Convincing the students in this class that they can achieve good performance in school.’ Reliability was acceptable, ω = .73.

Teachers' enthusiastic teaching behaviours during the lesson (assessed after the lesson; Time 3)

The four‐point Likert scale ranged from 1 (never or almost never) to 4 (always or almost always) and assessed students' perceptions of enthusiastic teaching behaviours in class during the lesson in regard to the topic of the lesson. The scale instruction was ‘How would you describe your teacher during the unit on quadratic equations?’ followed by eight items. Example items are ‘Our mathematics teacher was totally involved in the topic of quadratic equations’ and ‘Our mathematics teacher expressed that the topic of <quadratic equations> is important to him/her’. Reliability at Time 2 was good, ω within = .86, ω between = .98.

Students' mathematics interest (Time 1, Time 3, Time 4)

The four‐point Likert scale ranged from 1 (strongly disagree) to 4 (strongly agree) and assessed at Time 1 (pretest before the lesson) students' personal interest in mathematics classes with the current teacher using the scale instruction ‘How do you think about mathematics?’ followed by three items. An example item is ‘I often find what we cover in math class to be really exciting’. At Time 2 (posttest directly after the lesson), the items referred to students' interest in the content of the [quadratic equations] lesson. An example item is ‘I was interested in the topic of <quadratic equations>’. At Time 3 (follow‐up; 6 weeks after the lesson), items again addressed students' general interest in mathematics classes with the current teacher (e.g. ‘I often find what we cover in math class to be really exciting’). The scale reliabilities were good, see Table 1 (L2) and Table S2 (L1).

TABLE 1.

Descriptive statistics and manifest correlations at the classroom level (L2).

1 2 3 4 5 6 7
(1) Teacher SE T1
(2) NVI T2 .26
(3) Enthusiastic behaviours T3 .37*** .34***
(4) Student interest T1 .41*** .21c .58***
(5) Student interest T3 .25c .33*** .59*** .67***
(6) Student interest T4 .31* .47*** .47*** .80*** .66***
(7) Teacher experience .19 −.05 .02 −.01 .18 .20
M 2.79 .55 2.99 2.40 2.17 2.30 1.94
SD .42 .05 .30 .29 .27 .24 1.04
Min. 1.00 .44 1.00 1.00 1.00 1.00 1.00
Max. 4.00 .71 4.00 4.00 4.00 4.00 4.00
ICC(1) – – .22 .11 .10 .10 –
ω between – – .98 .98 .97 .86 –
Open in a new tab

Note: T1 = Time 1; T2 = Time 2; T3 = Time 3. Correlations are calculated at L2.

Abbreviations: ICC, intraclass correlation coefficient; NVI, nonverbal immediacy score; Teacher SE, teacher self‐efficacy.

c p < .10, *p < .05, ***p < .001.

Covariates

Teachers were asked ‘How many years of work experience do you have as a mathematics teacher regardless of whether you worked full‐time or part‐time?’ and the response categories were coded as in the TALIS data set, ranging from 1 (1–10 years) to 4 (31–49 years).

Statistical analyses

Intraclass correlation coefficients, reported in Table 1. ICC(1) values indicate how much variance is located at the highest level of analyses (L2) in multilevel analyses, in our case, at the classroom level. Values greater than .05 are considered sufficient for using the multilevel approach (Lüdtke et al., 2008). In our study, 10% of the variance in students' interest at each time point was attributable to students' membership in a specific classroom. A total of 22% of student‐perceived teachers' enthusiastic teaching behaviours during the lesson, assessed after the lesson, were explained by students' membership in the same classroom. Given the substantial amount of variance at L2, we applied a multilevel modelling approach with students (Level 1) nested in classrooms (Level 2).

Prior to the hypothesis testing, we tested metric invariance for student interest across time and across levels of analysis (L1, L2) following the steps described by Byrne (2004) and the cut‐off criteria for measuring non‐invariance in line with Chen (2007). The results of the invariance testing are reported in Table S.1 in the supplemental material. Results showed the metric invariance of student interest a cross time and levels.

Simulation studies on adequate sample sizes in multilevel modelling have led to rules of thumb that help guide choices about necessary sample sizes. For multilevel modelling, a sample size of 30 groups at L2 has been proposed as a minimum requirement to achieve unbiased standard errors (Maas & Hox, 2005), which is important because biased standard errors may lead to overestimates or underestimates of statistical power. In our study, given the expected effect sizes, the sample at L2 is likely at the lower bound for ensuring a sensitive test of the hypothesised relationships. For this reason, we applied manifest multilevel modelling (Marsh et al., 2009), a modelling procedure that makes efficient use of the available data. The procedure has been shown to result into lower standard errors while avoiding biased estimation of the model parameters in question. We tested our hypotheses with two models. Model 1 was a less restrictive model and included all relevant direct and indirect regression path between all study variables. Model 2 was a more restrictive model and included only the sequence of effects depicted in Figure 1. Both models included students' interest at Times 1, 3 and 4 at the student level (L1). At the classroom level, the models included teachers' self‐efficacy (Time 1), teachers' work experience (Time 1), students' interest (Times 1, 3 and 4), teachers' nonverbal immediacy (NVI) (Time 2) and student‐reported teachers' enthusiastic teaching behaviours (Time 3). Student interest (Time 1) was group‐mean centred, meaning that the mean of classroom j is subtracted from the score of student i in classroom j (Marsh et al., 2009). Indirect effects defined as products of regression coefficients were tested with 95% confidence intervals (CIs).

Mplus Version 8.3 was used for all analyses (Muthén & Muthén, 1998–2019), and maximum likelihood estimation with robust standard errors was applied. Goodness of model fit was evaluated using the following criteria: the comparative fit index (CFI), root mean square of approximation (RMSEA) and standardized root mean residual (SRMRwithin, between). As a rule of thumb, CFI values greater than or equal to .95 as well as RMSEA values below or equal to .06 and SRMRwithin/between values below or equal to .10 (Hu & Bentler, 1999) indicate highly satisfactory levels of model fit. However, researchers should not overly rely on the golden rules of fit (Hsu et al., 2015; Marsh et al., 2004), and particularly the SRMR value has been shown to be sensitive for sample size (Marsh et al., 2004). Missing data were handled by means of full‐information maximum likelihood estimation (FIML).

RESULTS

Descriptives

Descriptive statistics for all constructs at the class level are reported in Table 1. Descriptive statistics for Level 1 constructs are reported in Table S2. TSESE before the lesson (Time 1) was positively but marginally significantly (p < .10) associated with teachers' NVI during the lesson (Time 2). Teachers' NVI (Time 2) was positively and statistically significantly associated with aggregated student‐reported teachers' enthusiastic teaching behaviours during the lesson and with students' interest at Times 1 and 3 at the class level, but only marginally significantly and positively with students' interest at Time 4. Student‐reported class‐level teachers' enthusiastic teaching behaviours during the lesson was positively and significantly associated with students' interest at each time point.

TSESE, NVI, enthusiastic teaching and student interest

The model fit of Model 1 that included all direct and indirect paths was acceptable: χ 2 = 681.15, df = 279; CFI = .93, TLI = .92, RMSEA = .04, SRMRwithin = .04, SRMRbetween = .13. Coefficients are depicted in Table 2. TSESE before the lesson (Time 1) was positively and significantly related to teachers' NVI during the lesson (Time 2) and to class‐level enthusiastic teaching behaviours, assessed after the lesson (Time 3). Teachers' NVI during the lesson (Time 2) was positively related to class‐level enthusiastic teaching behaviours, reported after the lesson (Time 3). Neither TSESE before the lesson (Time 1) nor NVI during the lesson (Time 2) or student‐reported enthusiastic teaching behaviours (Time 3) were significantly related to students' interest after the lesson (Time 3), or 6 weeks later (Time 4), when controlling for interest at Time 1. Indirect effects were not significant in this model.

TABLE 2.

Model 1: Standardized coefficients of the latent‐manifest model.

Variable L1 Student int. T3 Student int. T4
β SE p 95% CI β SE p 95% CI
Student int. T1 .60 .03 .001 .535, .661 .43 .05 .001 .336, .530
Student int. T3 .47 .04 .001 .386, .561
L1: R 2 .36*** .66***
Variable L2 NVI T2 Enthu T3 Student int. T3 Student int. T4
β SE p 95% CI β SE p 95% CI β SE p 95% CI β SE p 95% CI
TSESE T1 .38 .18 .034 .029, .738 .28 .13 .029 .113, .721 −.31 .20 .119 −.704, .080 −.27 .38 .475 −1.022, .476
NVI T2 .42 .16 .007 .028, .522 .22 .14 .119 −.057, .503 −.04 .21 .844 −.453, .371
Enthu T3 .26 .14 .070 −.021, .531 .07 .23 .771 −.378, .510
Student int. T1 .75 .12 .001 .524, .975 1.32 .39 .001 .554, 2.076
Student int. T3 −.44 .38 .249 −1.187, .308
Teacher exp −.18 .16 .273 −.488, .138 −.05 .20 .797 −.450, .346 .29 .12 .020 .045, .526 .37 .31 .231 −.232, .964
R 2 .15 .32* .64*** .91***
Open in a new tab

Note: The correlation between TSESE and teacher exp. was φ = .23, SE = .16, p = .170, and between TSESE and students' interest T1 was φ = .58, SE = .14, p = .001.

Abbreviations: Enthu, student‐reported teacher enthusiastic teaching; NVI, teachers' nonverbal immediacy; Student int., students' interest; Teacher exp., teachers' teaching experience in years; TSESE, teachers' self‐efficacy for student engagement.

*p < .05, ***p < .001.

The model fit of Model 2 that only included the sequential effects depicted in Figure 1 was acceptable: χ 2 = 707.26, df = 290; CFI = .93, TLI = .92, RMSEA = .04, SRMRwithin = .04, SRMRbetween = .18. Coefficients are depicted in Table 3. TSESE before the lesson (Time 1) was positively and significantly related to teachers' NVI during the lesson (Time 2). Teachers' NVI during the lesson (Time 2) was positively related to class‐level enthusiastic teaching behaviours, reported after the lesson (Time 3). Student‐reported enthusiastic teaching behaviours, assessed after the lesson (Time 3) was significantly and positively associated with students' interest, assessed after the lesson (Time 3) when controlling for students' prior interest before the lesson (Time 1). Students' interest after the lesson (Time 3) was significantly and positively related to students' interest 6 weeks later (Time 4). Given the modelling procedure used, adjusting for students' previous interest means estimating the additional effect of other predictors for students representing an average interest level in their respective classroom at the adjusted measurement time point. As interest at Time 1 was not controlled, interest Time 3 predicted the level in students' interest at Time 4, but not the change in interest from Time 1 to Time 4. The indirect effect from TSESE to student‐rated teachers' enthusiastic teaching behaviours was not significant, β ind = .18, SE = .05, p = .13 [95% CI −.05, .42]. The indirect effects from teachers' NVI (Time 2) to students' interest (Time 3) via student‐reported enthusiastic teaching behaviours (Time 3) (β ind = .13, SE = .05, p = .01 [95% CI .03, .23]) and from student‐reported enthusiastic teaching behaviours (Time 3) to students' interest (Time 4) via students' interest (Time 3) (β ind = .14, SE = .06, p = .03 [95% CI .01, .26]) were significant and positive.

TABLE 3.

Model 2: Standardized coefficients of the latent‐manifest model.

Variable L1 Student int. T3 Student int. T4
β SE p 95% CI β SE p 95% CI
Student int. T1 .60 .03 .001 .535, .660 .43 .05 .001 .336, .529
Student int. T3 .47 .04 .001 .384, .560
L1: R 2 .36*** .65***
Variable L2 NVI T2 Enthu T3 Student int. T3 Student int. T4
β SE p 95% CI β SE p 95% CI β SE p 95% CI β SE p 95% CI
TSESE T1 .40 .20 .048 .003, .793
NVI T2 .45 .11 .001 .235, .672
Enthu T3 .29 .12 .021 .043, .529
Student int. T1 .67 .10 .001 .464, .869
Student int. T3 .47 .15 .002 .179, .767
Teacher exp
R 2 .16 .20* .56*** .22
Open in a new tab

Note: The correlation between TSESE and teacher exp. was φ = .19, SE = .16, p = .220, and between TSESE and students' interest T1 was φ = .50, SE = .13, p = .001. In this model, we correlated teacher exp. with NVI and Enthu, but both coefficients were not significant (NVI: φ = −.15, SE = .16, p = .346, Enthu: φ = .04, SE = .20, p = .828). The residual of an indicator of the latent factor ‘student interest Time 3’ was originally estimated to values below zero, but not significantly different from zero. We fixed this residual to zero following Hox (2002, p. 215).

Abbreviations: Enthu, student‐reported teacher enthusiastic teaching; NVI, teachers' nonverbal immediacy; Student int., students' interest; Teacher exp., teachers' teaching experience in in years; TSESE, teachers' self‐efficacy for student engagement.

*p < .05, ***p < .001.

[Correction added on 5 February 2025, after the first online publication: Table 3 has been updated in this version.]

Our analyses draw on two (minimum) or three (maximum) lessons of the participating teachers, depending on the number of lessons they videorecorded for the TALIS study. However, because a review of the time scheduling of the videorecorded lessons showed that the third lesson was videorecorded in some cases after the posttest, we provide additional analyses of our models based on only the first two videorecorded lessons in Tables S.3 and S.4 in the supplemental material to ensure a meaningful causal order of effects. These additional analyses can be seen as a robustness check and reveal that machine‐learning classified NVI was in Model 1 and 2 positively and significantly related to student‐perceived enthusiastic teaching behaviours. Teachers' self‐efficacy did only in Model 2 significantly and positively relate to machine‐learning classified NVI when only including two lessons.

DISCUSSION

Addressing the paucity of research examining processes that underlie the effects from teacher to student motivation, this study aimed to uncover the links through which teachers' self‐efficacy for student engagement affects their instructional behaviours, which then ultimately relate to student interest. Our key findings revealed that teachers' NVI, objectively assessed via machine learning, was significantly and positively related to the students' ratings of enthusiastic teaching behaviours. Teachers' self‐efficacy for student engagement (TSESE) was related to their NVI across the tested models. However, indirect effects from TSESE to student‐rated enthusiastic teaching via NVI was not significant. Our study was innovative, as we assessed teachers' NVI via objective situation‐specific measures and applied a new theory‐guided AI‐based assessment approach using computational modelling based on classroom video data.

TSESE, teachers' behaviours, and student interest: Is NVI the missing link?

In line with our assumptions (H1), we found that teachers who reported before the lesson that they believed themselves to be capable of motivating the students in their class expressed high levels of nonverbal immediacy. In turn, teachers with low levels of self‐efficacy for student engagement in a target class displayed low levels of NVI during the lesson. It is important to note that the direct effect of TSESE on teachers' NVI was significant across the tested models, However, probably due to the small sample size, the indirect effects from TSESE to student‐rated enthusiastic teaching via NVI was not significant with p = .052. Thus, there was a sequence of effects rather than an indirect pathway. An explanation might be the relatively small sample size at the classroom level of 46 teachers, which led to increased standard errors in our second less restrictive model. Our findings need to be replicated with larger samples particularly at the classroom level in future studies.

In line with our assumptions (H2) and following the theoretical assumptions of approaches studying nonverbal immediacy (Mehrabian, 1968; Richmond et al., 2003), we found that teachers' NVI during the lesson positively and significantly related to students' perceived enthusiastic teaching behaviours during the lesson, as reported by students after the lesson. Our findings can be seen as a validation of the machine‐learning assessed NVI as the results of the computational model of NVI corresponded with students' ratings of teachers' enthusiastic teaching behaviours, and thus, with a conceptually closed theoretical construct. Methodologically, our findings point to machine learning as an option to classify teaching behaviours for research in a resource‐efficient way. More research with different data sets and different teaching situations is needed to validate this approach further. However, we believe that classroom observation studies could largely benefit from our approach. On a theoretical level, we were able to show a sequence from TSESE to NVI, and from NVI to teaching behaviours. Thus, nonverbal behaviours of the teacher in class might be an explanation for why motivated teachers provide high‐quality instruction in class. Prior work has pointed out a deficiency of research of the processes linking teacher motivation to actual teaching behaviours (Lazarides et al., 2024). Our study partially closed this knowledge gap by revealing that teachers' nonverbal behaviours during the interactions with their students might be a part of the missing link.

In line with our expectations (H3) and work on teacher enthusiastic teaching behaviours (Frenzel et al., 2018), we further found that student reports of their teachers' enthusiastic teaching behaviours during the lesson (assessed directly after the lesson) was positively associated with students' domain‐specific interest directly after the lesson. However, Model 1 shows that neither TSESE nor NVI or enthusiastic teaching behaviours had a longitudinal effect on students' interest 6 weeks after the lesson (Time 4) when controlling for students' baseline interest (Time 1). One explanation for this finding might be the high stability of students' interest across time. The results of Model 2 show that when baseline interest was not controlled for, student‐reported enthusiastic teaching behaviours (Time 3) were positively related to students' interest 6 weeks after the lesson (Time 4) via students' interest directly after the lesson (Time 3). However, because both, student‐reported enthusiastic teaching behaviours and students' interest (Time 3) were assessed at the same time point, the directionality of effects cannot be disentangled in our models.

Implications for research and educational practice

Our findings show that nonverbal communication in terms of body language, facial expressions and physical proximity foster student perceptions of enthusiastic teaching behaviours and subsequent student interest. Thus, teacher education and training that includes a focus on NVI might benefit both teaching and learning. Instructional strategies for teachers should be highlighted that allow them to express enthusiasm and interact closely with students, thereby creating a more engaging and dynamic learning experience. The present study showed that TSESE was prospectively related to teacher NVI; however, fostering teachers' NVI could potentially also shape their TSESE, which is a question left open for future research. Another question for future work would be in what terms the lesson‐specific interactions in our study are embedded in more long‐term, dynamic effects among teachers and their students. Thus, multi‐wave studies that comprise multiple lessons embedded in multiple measurement occasions assessed over longer time spans are needed to examine how much the relations between teachers' nonverbal behaviours and their students' perceptions thereof are across lessons and across longer time spans.

Strengths and limitations

The longitudinal and multimodal design of our study is a definite strength, as it allows us to examine how the change in mean values of students' group‐level interest across time, from before the lesson to 6 weeks after the lesson, depends on mediational processes that transmit teachers' self‐reported self‐efficacy through their objectively assessed behaviours in class, and students' perceptions of these behaviours. However, our study also has limitations. First, the sample was limited to 50 teachers and their classrooms, and although this sample size is sufficient to test our hypotheses with considerable statistical power, the complex multilevel model that we tested should be validated in larger samples. Second, our findings are limited to the domain of mathematics, and more specifically to lessons on quadratic equations, and teachers' NVI behaviours might differ depending on the subject they teach (Goetz et al., 2013). Therefore, further research is needed to validate the NVI algorithm by applying it to various mathematics topics and subject domains. Third, the different levels of specificity in the measures are a unique feature of the data, enabling us to examine whether and to what extent teachers shape students' learning in particular classroom moments, which may, in turn, lead to more global manifestations in students' motivation. We sincerely believe that this operationalization effectively reflects teacher effects on learning processes and their potential consequences for students. It could be argued that variations in the level of specificity in the measures may explain weak relationships between constructs. For instance, the relationship between student‐rated enthusiastic teaching behaviours (lesson‐specific, topic‐related) and mathematics interest (general, not lesson‐specific; domain‐related) might be stronger if both measures referred to the same reference frame in terms of the teaching situation (lesson‐specific vs. general) and content area (topic vs. domain).

CONCLUSIONS

Our study helped to shed light on the effects from TSESE to student‐perceived teaching behaviours and domain‐related interest by using a rigorous study design and an automated AI‐based assessment of teachers' NVI. Motivational research and classroom research are both currently developing towards situated approaches that include multimodal assessments of motivations and behaviours in instructional situations. Bringing together these fields and using new and innovative AI‐based methods would be promising for the future, enabling researchers to better understand how and why certain instructional settings decisively enhance students' motivation in class.

AUTHOR CONTRIBUTIONS

Rebecca Lazarides: Conceptualization; investigation; funding acquisition; writing – original draft; methodology; visualization; writing – review and editing; formal analysis; project administration; resources. Jonas Frenkel: Data curation; writing – review and editing; project administration; visualization; methodology; validation. Uroš Petković: Methodology; visualization; project administration; data curation; validation. Richard Göllner: Writing – review and editing; formal analysis; supervision; methodology. Olaf Hellwich: Project administration; funding acquisition; investigation; methodology; software; supervision; resources.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest to disclose.

Supporting information

Data S1.

BJEP-95-S15-s001.docx (37KB, docx)

ACKNOWLEDGEMENTS

We would like to thank the Forschungsdatenzentrum Bildung at the Leibniz Institute for Research and Information in Education for the provision of the data base for this study. Open Access funding enabled and organized by Projekt DEAL.

Lazarides, R. , Frenkel, J. , Petković, U. , Göllner, R. , & Hellwich, O. (2025). ‘No words’—Machine‐learning classified nonverbal immediacy and its role in connecting teacher self‐efficacy with perceived teaching and student interest. British Journal of Educational Psychology, 95(Suppl. 1), S15–S31. 10.1111/bjep.12732

Footnotes

1

We report here the weighted correlation coefficients from the meta‐analyses.

2

Three of the total 403 segments were excluded from NVI scoring by the human raters because the video quality did not allow an analysis of the full 30‐second duration. However, since analysis was still possible for the majority of the individual frames without restriction, these segments were retained for use in the respective submodels.

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.

REFERENCES

  1. Andersen, J. F. , Andersen, P. A. , & Jensen, A. D. (1979). The measurement of nonverbal immediacy. Journal of Applied Communication Research, 7(2), 153–180. 10.1080/00909887909365204 [DOI] [Google Scholar]
  2. Bandura, A. (1986). Social foundations of thought and action: A social cognitive theory. Prentice‐Hall, Inc. [Google Scholar]
  3. Bandura, A. (1997). Self‐efficacy: The exercise of control. Freeman. [Google Scholar]
  4. Burić, I. , Butković, A. , & Kim, L. E. (2023). Teacher personality domains and facets: Their unique associations with student‐ratings of teaching quality and teacher enthusiasm. Learning and Instruction, 88, 101827. 10.1016/j.learninstruc.2023.101827 [DOI] [Google Scholar]
  5. Byrne, B. M. (2004). Testing for multigroup invariance using AMOS graphics: A road less traveled. Structural Equation Modeling, 11(2), 272–300. [Google Scholar]
  6. Chen, F. F. (2007). Sensitivity of goodness of fit indexes to lack of measurement invariance. Structural Equation Modeling, 14(3), 464–504. 10.1080/10705510701301834 [DOI] [Google Scholar]
  7. Chesebro, J. L. , & McCroskey, J. C. (2001). The relationship of teacher clarity and immediacy with student state receiver apprehension, affect, and cognitive learning. Communication Education, 50(1), 59–68. 10.1080/03634520109379232 [DOI] [Google Scholar]
  8. Cicchetti, D. V. (1994). Guidelines, criteria, and rules of thumb for evaluating normed and standardized assessment instruments in psychology. Psychological Assessment, 6(4), 284–290. 10.1037/1040-3590.6.4.284 [DOI] [Google Scholar]
  9. Collins, M. L. (1978). Effects of enthusiasm training on preservice elementary teachers. Journal of Teacher Education, 29(1), 53–57. [Google Scholar]
  10. den Brok, P. , Fisher, D. , Rickards, T. , & Bull, E. (2006). Californian science students' perceptions of their classroom learning environments. Educational Research and Evaluation, 12(1), 3–25. 10.1080/13803610500392053 [DOI] [Google Scholar]
  11. Fackler, S. , Sammons, P. , & Malmberg, L. E. (2021). A comparative analysis of predictors of teacher self‐efficacy in student engagement, instruction and classroom management in Nordic, Anglo‐Saxon and east and south‐east Asian countries. Review of Education, 9(1), 203–239. 10.1002/rev3.3242 [DOI] [Google Scholar]
  12. Fives, H. , & Buehl, M. (2016). Teacher motivation: Self‐efficacy and goal orientation. In Wentzel K. R. & Miele D. B. (Eds.), Handbook of motivation at school (2nd ed., pp. 340–360). Taylor and Francis Inc. 10.4324/9781315773384 [DOI] [Google Scholar]
  13. Frenzel, A. C. , Becker‐Kurz, B. , Pekrun, R. , Goetz, T. , & Lüdtke, T. (2018). Emotion transmission in the classroom revisited: A reciprocal effects model of teacher and student enjoyment. Journal of Educational Psychology, 110(5), 628–639. 10.1037/edu0000228 [DOI] [Google Scholar]
  14. Goetz, T. , Lüdtke, O. , Nett, U. E. , Keller, M. M. , & Lipnevich, A. A. (2013). Characteristics of teaching and students' emotions in the classroom: Investigating differences across domains. Contemporary Educational Psychology, 38(4), 383–394. 10.1016/j.cedpsych.2013.08.001 [DOI] [Google Scholar]
  15. Grünkorn, J. , Klieme, E. , Praetorius, A.‐K. , & Schreyer, P. (2020). Mathematikunterricht im internationalen Vergleich. In Ergebnisse aus der TALIS‐Videostudie Deutschland [mathematics instruction in international comparison: Results from the TALIS video study Germany]. Leibniz‐Institut für Bildungsforschung und Bildungsinformation. [Google Scholar]
  16. Harris, M. J. , & Rosenthal, R. (2005). No more teachers' dirty looks: Effects of teacher nonverbal behavior on student outcomes. In Riggio R. E. & Feldman R. S. (Eds.), Applications of nonverbal communication (pp. 157–192). Psychology Press. [Google Scholar]
  17. Hatfield, E. , Cacioppo, J. T. , & Rapson, R. L. (1993). Emotional contagion. Current Directions in Psychological Science, 2, 96–100. 10.1111/1467-8721.ep10770953 [DOI] [Google Scholar]
  18. He, K. , Zhang, X. , Ren, S. , & Sun, J. (2016). Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition.
  19. Hox, J. J. (2002). Multilevel analysis: Techniques and applications. Lawrence Erlbaum Associates. [Google Scholar]
  20. Hsu, H.‐Y. , Kwok, O.‐M. , Lin, J. H. , & Acosta, S. (2015). Detecting misspecified multilevel structural equation models with common fit indices: A Monte Carlo study. Multivariate Behavioral Research, 50(2), 197–215. [DOI] [PubMed] [Google Scholar]
  21. Hu, L. , & Bentler, P. M. (1999). Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling: A Multidisciplinary Journal, 6(1), 1–55. 10.1080/10705519909540118 [DOI] [Google Scholar]
  22. Keller, M. , Goetz, T. , Becker, E. S. , Morger, V. , & Hensley, L. (2014). Feeling and showing: A new conceptualization of dispositional teacher enthusiasm and its relation to students' interest. Learning and Instruction, 33, 29–38. 10.1016/j.learninstruc.2014.03.001 [DOI] [Google Scholar]
  23. Keller, M. , Hoy, A. W. , Goetz, T. , & Frenzel, A. C. (2016). Teacher enthusiasm: Reviewing and redefining a complex construct. Educational Psychology Review, 28(4), 743–769. 10.1007/s10648-015-9354-y [DOI] [Google Scholar]
  24. Kim, K. R. , & Seo, E. H. (2018). The relationship between teacher efficacy and students' academic achievement: A meta‐analysis. Social Behavior and Personality: An International Journal, 46(4), 529–540. 10.2224/sbp.6554 [DOI] [Google Scholar]
  25. Klassen, R. M. , & Tze, V. M. (2014). Teachers' self‐efficacy, personality, and teaching effectiveness: A meta‐analysis. Educational Research Review, 12, 59–76. 10.1016/j.edurev.2014.06.001 [DOI] [Google Scholar]
  26. Koelstra, S. , Muhl, C. , Soleymani, M. , Lee, J.‐S. , Yazdani, A. , Ebrahimi, T. , Pun, T. , Nijholt, A. , & Patras, I. (2011). Deap: A database for emotion analysis; using physiological signals. IEEE Transactions on Affective Computing, 3(1), 18–31. 10.1109/T-AFFC.2011.15 [DOI] [Google Scholar]
  27. Lauermann, F. , & ten Hagen, I. (2021). Do teachers' perceived teaching competence and self‐efficacy affect students' academic outcomes? A closer look at student‐reported classroom processes and outcomes. Educational Psychologist, 1‐18, 265–282. 10.1080/00461520.2021.1991355 [DOI] [Google Scholar]
  28. Lazarides, R. , & Schiefele, U. (2021). Teacher motivation: Implications for instruction and learning. Introduction to the special issue. Learning and Instruction, 76, 101543. [Google Scholar]
  29. Lazarides, R. , Schiefele, U. , Daumiller, M. , & Dresel, M. (2024). From teacher motivation to instructional behaviour: A systematic review of the mediating processes . 10.31219/osf.io/u2rxp [DOI]
  30. Lazarides, R. , Schiefele, U. , Hettinger, K. , & Frommelt, M. (2023). Tracing the signal from teachers to students: How teachers' motivational beliefs longitudinally relate to student interest through student‐reported teaching practices. Journal of Educational Psychology, 115(2), 290–308. 10.1037/edu0000777 [DOI] [Google Scholar]
  31. Lazarides, R. , Watt, H. M. , & Richardson, P. (2023). Does school context moderate longitudinal relations between teacher‐reported self‐efficacy and value for student engagement and teacher‐student relationships from early until midcareer? Contemporary Educational Psychology, 72, 102136. 10.1016/j.cedpsych.2022.102136 [DOI] [Google Scholar]
  32. Lin, L. , Song, S. , Yang, W. , & Liu, J. (2020). Ms2l: Multi‐task self‐supervised learning for skeleton based action recognition. In 28th ACM international conference on multimedia .
  33. Liu, W. (2021). Does teacher immediacy affect students? A systematic review of the association between teacher verbal and non‐verbal immediacy and student motivation. Frontiers in Psychology, 12, 713978. 10.3389/fpsyg.2021.713978 [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Lüdtke, O. , Marsh, H. W. , Robitzsch, A. , Trautwein, U. , Asparouhov, T. , & Muthén, B. (2008). The multilevel latent covariate model: A new, more reliable approach to group‐level effects in contextual studies. Psychological Methods, 13(3), 203–229. [DOI] [PubMed] [Google Scholar]
  35. Maas, C. J. , & Hox, J. J. (2005). Sufficient sample sizes for multilevel modeling. Methodology, 1(3), 86–92. 10.1027/1614-2241.1.3.86 [DOI] [Google Scholar]
  36. Marsh, H. W. , Hau, K.‐T. , & Wen, Z. (2004). In search of golden rules: Comment on hypothesis‐testing approaches to setting cutoff values for fit indexes and dangers in overgeneralizing Hu and Bentler's (1999) findings. Structural Equation Modeling, 11(3), 320–341. 10.1207/s15328007sem1103_2 [DOI] [Google Scholar]
  37. Marsh, H. W. , Lüdtke, O. , Robitzsch, A. , Trautwein, U. , Asparouhov, T. , Muthén, B. , & Nagengast, B. (2009). Doubly‐latent models of school contextual effects: Integrating multilevel and structural equation approaches to control measurement and sampling error. Multivariate Behavioral Research, 44(6), 764–802. 10.1080/00273170903333665 [DOI] [PubMed] [Google Scholar]
  38. Mehrabian, A. (1968). Relationship of attitude to seated posture, orientation, and distance. Journal of Personality and Social Psychology, 10(1), 26–30. 10.1037/h0026384 [DOI] [PubMed] [Google Scholar]
  39. Muthén, L. , & Muthén, B. (1998. –2019). Mplus user's guide. Muthen & Muthen. [Google Scholar]
  40. OECD . (2020). Global teaching insights: A video study of teaching. OECD. [Google Scholar]
  41. Oquab, M. , Darcet, T. , Moutakanni, T. , Vo, H. , Szafraniec, M. , Khalidov, V. , Fernandez, P. , Haziza, D. , Massa, F. , El‐Nouby, A. , Assran, M. , Ballas, N. , Galuba, W. , Howes, R. , Huang, P.‐Y. , Li, S.‐W. , Misra, I. , Rabbat, M. , Sharma, V. , … Bojanowski, P. (2023). Dinov2: Learning robust visual features without supervision. arXiv preprint arXiv:2304.07193.
  42. Petković, U. , Frenkel, J. , Hellwich, O. , & Lazarides, R. (2024). Nonverbal immediacy analysis in education: A multimodal computational model. In 17th International conference on the simulation of adaptive behavior (SAB 2024): From animals to Animats. University of California. [Google Scholar]
  43. Richmond, V. P. , McCroskey, J. C. , & Johnson, A. D. (2003). Development of the nonverbal immediacy scale (NIS): Measures of self‐and other‐perceived nonverbal immediacy. Communication Quarterly, 51(4), 504–517. 10.1080/01463370309370170 [DOI] [Google Scholar]
  44. Rumelhart, D. E. , Hinton, G. E. , & Williams, R. J. (1986). Learning representations by back‐propagating errors. Nature, 323(6088), 533–536. 10.1038/323533a0 [DOI] [Google Scholar]
  45. Savchenko, A. V. (2021). Facial expression and attributes recognition based on multi‐task learning of lightweight neural networks. In 2021 IEEE 19th international symposium on intelligent systems and informatics (SISY) (pp. 119–124). IEEE. [Google Scholar]
  46. Shahroudy, A. , Liu, J. , Ng, T. T. , & Wang, G. (2016). Ntu rgb+ d: A large scale dataset for 3d human activity analysis. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 1010–1019). IEEE. [Google Scholar]
  47. Shrout, P. E. , & Fleiss, J. L. (1979). Intraclass correlations: uses in assessing rater reliability. Psychological Bulletin, 86(2), 420–428. 10.1037/0033-2909.86.2.420 [DOI] [PubMed] [Google Scholar]
  48. Tschannen‐Moran, M. , & Woolfolk Hoy, A. (2001). Teacher efficacy: Capturing an elusive construct. Teaching and Teacher Education, 17(7), 783–805. 10.1016/S0742-051X(01)00036-1 [DOI] [Google Scholar]
  49. Tschannen‐Moran, M. , Woolfolk Hoy, A. , & Hoy, W. K. (1998). Teacher efficacy: Its meaning and measure. Review of Educational Research, 68(2), 202–248. 10.3102/00346543068002202 [DOI] [Google Scholar]
  50. Watt, H. M. G. , & Richardson, P. W. (2008). Guest Editorial. Motivation for Teaching. Learning and Instruction, 18(5), 405–407. 10.1016/j.learninstruc.2008.06.009 [DOI] [Google Scholar]
  51. Witt, P. L. , & Wheeless, L. R. (2001). An experimental study of teachers' verbal and nonverbal immediacy and students' affective and cognitive learning. Communication Education, 50(4), 327–342. 10.1080/03634520109379259 [DOI] [Google Scholar]
  52. Witt, P. L. , Wheeless, L. R. , & Allen, M. (2004). A meta‐analytical review of the relationship between teacher immediacy and student learning. Communication Monographs, 71(2), 184–207. 10.1080/036452042000228054 [DOI] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Data S1.

BJEP-95-S15-s001.docx (37KB, docx)

Data Availability Statement

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.

Articles from The British Journal of Educational Psychology are provided here courtesy of Wiley

RESOURCES