The importance of teacher-student relationships in classrooms with ‘difficult’ students: a multi-level moderation analysis of nine Berlin secondary schools

ABSTRACT

Meta-analyses suggest that instructional quality in the classroom and the quality of teacher-student relationships (TSR) predict positive social-emotional and achievement-related outcomes. Psychoanalytic theory asserts that positive teacher-student relationships are particularly important for outcomes in classrooms with more students with severe psychosocial difficulties. Hence, this study tests whether classrooms with more students with severe psychosocial difficulties have better social-emotional and achievement-related outcomes when teachers have been able to establish more positive relationships with their students. Hierarchical linear regression models use nested student survey data from 32 classrooms. Results only partially support the hypothesis and suggest that too many students with psychosocial difficulties might overwhelm even teachers with strong relationship-building skills, leading to detrimental outcomes.

KEYWORDS:

• Teacher-student relationships
• psychosocial difficulties
• student social-emotional outcomes
• instructional teaching quality
• student achievement
• studentwell-being

Introduction

While the role of elementary school teachers is often seen as that of both nurturer and knowledge conveyor, many teachers at the secondary school-level interpret their job as only the latter (Garcia-Moya, Moreno, and Brooks 2019). These teachers’ traditional views have been increasingly challenged, most recently by those advocating for student-centred learning approaches that more strongly focus on students’ individual needs, including the need for close relationships with adults (Garcia-Moya, Moreno, and Brooks 2019; Kaput 2018).

Empirical research supports the student-centred learning approach, suggesting that it has a small but significant positive effect on achievement compared to more traditional teacher-centred learning approaches (Cornelius-White 2007; Wilson et al. 2019). It is unclear, however, how exactly these positive effects are attained, because the greater emphasis on relationships is only one of several differences between student-centred learning and more traditional teaching styles (Keiler 2018).

Meta-analyses have confirmed that both the quality of teacher-student relationships (Roorda et al. 2011) and teaching quality (Kraft, Blazar, and Hogan 2018) are significant predictors of positive student outcomes. However, these studies typically do not differentiate between the two constructs, which are most likely correlates (Ferguson et al. 2015). Thus, it is largely unclear to what degree and how the quality of teacher-student relationships improves student outcomes beyond instructional quality.

Instructional versus relational dimensions of teaching quality

Leading theoretical frameworks of teaching quality differentiate between teachers’ instructional teaching skills and their ability to establish positive relationships with students. For example, Panorama (2020) distinguishes between pedagogical effectiveness (‘the quality of teaching and amount of learning students experience from a particular teacher’) and classroom teacher-student relationships (‘How strong the social connection is between teachers and students within and beyond the classroom’) in their teacher evaluations.

Similarly, the Tripod 7Cs Framework of Effective Teaching (Tripod 2016, 2) is composed of six dimensions of classroom-related instructional teaching skills,¹ and one dimension going beyond the classroom, Care (‘show concern for students’ emotional and academic well-being’). A large study of N = 16,300 American classrooms using the Tripod framework shows that the instructional dimensions of the 7Cs tend to be more strongly related to students’ achievement-related outcomes, while Care most strongly and positively correlates with emotional and relational outcomes, i.e. feelings of happiness and help-seeking (Ferguson et al. 2015).

Teachers’ impact on students’ academic and social-emotional development

Widely regarded studies (Hattie 2009; Cantrell and Kane 2013; Kane and Cantrell 2010) and metanalytic calculations (Kraft, Blazar, and Hogan 2018) show that the instructional teaching quality in the classroom is a reliable and valid predictor of students’ learning progress, controlling for confounding student background characteristics. Theoretically, it is assumed that better instructional teaching quality improves achievement in part because it positively impacts achievement-related student factors, including feelings of efficacy, sense of purpose, and joy of learning (Ferguson et al. 2015), and because it reduces academic-status insecurity (Dietrich and Hofman 2019).

The quality of teacher-student-relationships (Roorda et al. 2011) and the overall school relationship climate (Dulay and Karadağ 2017) have also been found to impact student achievement. This can be explained with attachment theory, which proposes that developing a more or less secure attachment style as a young child, which is based on primary caregivers’ nurturing and emotional responsiveness, can impact behaviour and developmental outcomes throughout one’s entire lifetime (Ainsworth and Bell 1970; Riley 2010). While people with secure attachment – i.e. secure inner emotional representations of positive relationships – are able to sustain positive relationships, those with insecure attachments struggle with this task.

Mentalisation theory adds that a secure attachment style is closely linked to developing the skills for understanding others’ thoughts and emotions impacting behaviour (Fonagy et al. 2004; Gingelmaier, Taubner, and Ramberg 2018; Taubner 2015). These skills can be improved at any developmental stage with the help of ‘relevant others’, such as teachers (Fonagy et al. 2004; Gingelmaier, Taubner, and Ramberg 2018; Taubner 2015). Quantitative-empirical findings indicate that better teacher-student relationships can improve students’ social-emotional skills, which in turn lead to higher achievement (Skinner et al. 2008).

The positive link between teacher-student relationships and students’ social-emotional well-being has also been repeatedly confirmed (Durlak et al. 2011; Holt-Lunstad, Smith, and Layton 2010; Waldinger et al. 2014). While there is most certainly a direct effect, positive teacher-student relationships might also indirectly improve students’ emotional well-being, because how teachers treat students affects how students treat each other (Chory and Offstein 2018; Dietrich and Cohen 2019; Hughes, Cavell, and Jackson 1999). Teacher support, a concept closely related to teacher-student relationships, has been positively linked to students’ school-related feelings of anger, anxiety (Lei, Cui, and Chiu 2018), and school-connectedness (Joyce 2018).

Instructional teaching skills might also influence students’ social-emotional well-being. A significant link has been found between teachers’ ability to explain content material and students’ feelings of anger (Ferguson et al. 2015). Similarly, students of worse teachers are academically less successful (Cantrell and Kane 2013; Kane and Cantrell 2010), and lower academic success is associated with a decrease in emotional well-being (Bücker et al. 2018).

Students with severe psychosocial difficulties

High quality teacher-student relationships might be more important for student outcomes when there are more students with severe psychosocial difficulties in the classroom, which include those with overly externalising or internalising behaviours (Zimmermann 2018). Many of these students have had traumatising early childhood experiences and are particularly in need of positive relationships (Zimmermann 2016). Psychoanalytic theory assumes complex transference – countertransference processes in which students unconsciously re-enact early childhood experiences in subsequent relationships (Würker 2007). Most teachers are overwhelmed by the resulting severe misbehaviour and emotional outbursts of such students, and tend to react with punishment and feelings of helplessness (Zimmermann 2018). This reaction only worsens misbehaviour and escalates conflict, which re-traumatises students and threatens teachers’ mental health (Aloe et al. 2014; Gavish and Friedman 2010; Ozkilica and Kartal 2012). As a result, internalising and externalising students who experience conflict with their teachers show worse school adjustment and behaviour (Baker, Grant, and Morlock 2008).

However, psychoanalytic theory also asserts that teachers can learn to establish corrective emotional experiences (Alexander and French 1946) in their relationships with students who identify with them due to their parental roles. This provides teachers a unique opportunity to positively influence or ‘correct’ these students’ internal representations of close relationships, leading to better behaviour (Hanifin and Appel 2000; Aichhorn 2005) and thus, more positive classroom environments, which improve student outcomes (Roorda 2011; Durlak et al. 2011).

Although most teachers might not know a great deal about their students’ early childhood experiences, they can often find out enough information to establish a professional attitude that can bring about these corrective experiences. Here, the concepts of holding by Donald Winnicott (1986) and containment by Wilfred Bion (1962) suggest that teachers can facilitate students’ processing of otherwise overwhelming negative emotions. Students’ emotional trust (Müller 2017) and positive transference (Aichhorn 2005) are considered preconditions for corrective relationship experiences (Müller 2017).

Study purpose

Previous research suggests that teachers with stronger relationship-building skills improve student outcomes (e.g. Roorda 2011; Durlak et al. 2011), and psychoanalytic theory proposes that such skills might be particularly important for student outcomes when there are students with severe psychosocial difficulties in the classroom (Zimmermann 2018). Hence, this study tests the following hypothesis:

Classrooms with higher ratios of students with severe psychosocial difficulties have better social-emotional and achievement-related outcomes when teachers have been able to establish more positive relationships with their students – in terms of students’ emotional trust in the teacher, students’ identification with the teacher, and overall teacher-student relationship climate in the school– holding instructional teaching quality and students’ background characteristics constant.

This study focuses on the classroom-level, because the key interest of this study is in teachers’ skills, most reliably measured at the classroom-level using student survey data (Cantrell and Kane 2013), and teachers’ relationships with entire classrooms and their complex group dynamics.

Methods

Participants

For the study’s data collection, virtually all of the approximately 200 public and private secondary schools in Berlin were initially contacted via email and phone. Of these, a convenience sample of a total of nine schools, including 452 students and 39 teachers, decided to participate and filled in student and teacher surveys. Very low school response rates are common in Berlin, where schools tend to be overburdened and battle with teacher shortages (Tagesspiegel 2020), which is one reason why the Berlin school system is considered to be among the worst in Germany (Statista 2020). In return for their participation, all schools received a summary report of their results, including recommendations for school development and teacher training.

In the data collection process, each class was assigned to only one teacher in order to avoid cross-classification problems in the analysis. According to informal feedback, students typically needed about 30–45 minutes to fill in the questionnaire. Parents of students under the age of 16 were given an information sheet about the study and required to sign a written permission statement for their children’s participation. All students were informed that their participation is voluntary. Parents typically had several weeks to consider giving their consent.

Answering socioeconomic background questions required all parents to sign a permission statement, independent of students’ age. Students of parents who did not sign this permission, but a permission to participate in the survey, received a shortened version of the questionnaire that omitted the socioeconomic questions.

Some classrooms/teachers were excluded from the analysis because fewer than five students completed the questionnaire, which can lead to reliability problems with Tripod survey instruments (Tripod 2015). The final dataset contained 32 classrooms from nine schools and its student composition is presented in Table 1.

Table 1. Student characteristics

Variable	%
At least one parent has a Bachelor’s degree	55
100+ books at home	48
At least 3 computers at home	44
Male	46
Internalising psychosocial difficulties	10
Internalising psychosocial difficulties	9
Non-EU family background	39

Procedures

All schools were offered to conduct the surveys online in computer rooms via internet links generated with the online software tool LimeSurvey. However, only five teachers made use of this option, all others chose pen and paper versions of the questionnaires. These were sent to the schools in exact numbers via mail. Teachers were required to put the completed surveys in a sealed envelope, in order to assure confidentiality.

Instruments

This study only uses data from the student survey, not the teacher survey. All indices used in this study are generated as follows: first, the individual items were aggregated to the classroom-level; second, the indices were standardised.

Social-emotional and achievement-related student outcomes

Achievement-related efficacy

Students’ sense of efficacy is a 4-item index drawn from the Patterns of Adaptive Learning Scales (Midgley et al. 2000), and has established validity and reliability (Midgley et al. 1998). The items are measured on a 5-point Likert scale, ranging from Totally True to Totally Untrue. An example item is ‘I can do almost all the work in this school if I don’t give up’. In this study the efficacy index has an internal reliability of α =.81.

Sense of school purpose

Students’ sense of school purpose is a 4-item index drawn from the Student Subjective Well-Being Scale (Long et al. 2012). The items are measured on a 5-point Likert scale, ranging from Totally True to Totally Untrue. An example item is ‘I feel like the things I do at school are important’. In this study the school purpose index has an internal reliability of α = .80.

Joy of learning. Students’ joy of learning is a 4-item index from the Student Subjective Well-Being Scale (Long et al. 2012). The items are measured on a 5-point Likert scale, ranging from Totally True to Totally Untrue. An example item is ‘I get excited about learning new things in class’. In this study the joy of learning index has an internal reliability of α = .86.

School-connectedness

Students’ school-connectedness is a 4-item index from the Student Subjective Well-Being Scale (Long et al. 2012). The items are measured on a 5-point Likert scale, ranging from Totally True to Totally Untrue. An example item is ‘I feel like I belong at this school’. In this study the school-connectedness index has an internal reliability of α = .86.

School-related anger

Students’ school-related anger is a 2-item index developed from the Tripod Survey; its construct validity has been previously confirmed via confirmatory factor analysis (Dietrich and Zimmermann 2019). The items are measured on a 5-point Likert scale, ranging from Totally True to Totally Untrue. An example item is ‘Being in this class makes me feel angry’. In this study the school-related anger index has an internal reliability of α = .82.

Academic-status insecurity

Students’ academic-status insecurity is a 4-item index developed by Dietrich and Ferguson (2019). It conceptually derives from the learning goal concept of performance avoidance (Midgley et al. 2000; Midgley and Urban 2001), which has established validity and reliability (Jagacinski and Duda 2001). Construct validity has also been confirmed via confirmatory factor analysis (Dietrich and Zimmermann 2019). The items are measured on a 5-point Likert scale, ranging from Totally True to Totally Untrue. An example item is ‘In this class, I worry what other students think about me’. In this study’s data the academic-status insecurity index has an internal reliability of α = .73.

Teacher-student relationships

Teacher-student relationship climate at school

This is a 3-item index derived from the Tripod survey; its validity has been confirmed via confirmatory factor analysis (Dietrich and Cohen 2019). The items are measured on a 5-point Likert scale, ranging from Totally True to Totally Untrue. An example item is ‘Teachers in the hallways treat me with respect, even if they don’t know me’. In this study the teacher-student relationship climate at school index has an internal reliability of α = .82.

Students’ emotional trust towards the teacher

This 3-item index is derived from the Teacher-Student Relationship Scale (Pianta 2019), and measured on a 5-point Likert scale ranging from Totally True to Totally Untrue. The items are: ‘The teacher in this class is aware of my feelings’, ‘When I am in a bad mood, the teacher in this class knows how to handle me’, and ‘I openly share feelings and experiences with the teacher in this class’. In this study the students’ emotional trust towards the teacher index has an internal reliability of α = .74.

Students’ identification with the teacher

While ‘positive transference’ is difficult to measure directly, the identification of a student with a teacher is used as a proxy. This is because patients ‘transfer onto the analyst or therapist those feelings of attachment, love, idealization, or positive emotions that the patient originally experienced toward parents or other significant individuals during childhood’ (APA 2020). The same processes occur between students and teachers (Hanifin and Appel 2000; Freud 1970).

This 3-item index is derived from the Composite Teacher-Student Relationship Instrument (Barch 2015), measured on a 5-point Likert scale, ranging from Totally True to Totally Untrue. They are, ‘I would feel good if someone said I was a lot like the teacher in this class’, ‘I want to be like the teacher in this class’, and ‘As I get older, I’ll probably try to be a lot like the teacher in this class’. In this study the students’ identification with the teacher index has an internal reliability of α = .86.

Teachers’ instructional teaching skills

This is an index based on the Tripod 7Cs Framework of Effective teaching (Tripod 2016). Most studies use the 7Cs composite score, which is the combined average of the individual 7Cs, to measure teaching quality. Validity and reliability of the 7Cs Framework have been previously confirmed (Cantrell and Kane 2013; Polikoff 2016). The internal reliability of the 7Cs in this study’s dataset are: α = .89 for Care, α = .76 for Confer, α = .84 for Captivate, α = .93 for Clarify, α = .78 for Consolidate, α = .81 for Challenge, and α = .94 for Classroom Management.

For this study, Care was taken out of the 7Cs composite in order to create a 6Cs composite that only measures the overall classroom-related instructional dimension of teaching quality. The resulting 6Cs composite has an internal reliability of α = .92. In addition, Care was entirely excluded from the analyses, because of multi-collinearity issues with the students’ emotional trust towards the teacher index, indicating that these two indices measure essentially the same latent construct.

Students’ psychosocial difficulties

Students’ psychosocial difficulties were measured via the self-reported Strengths and Difficulties Questionnaire (SDQ), which includes 21 items measured on a 3-point Likert-scale ranging from Not True to Certainly True. Validity and reliability of the SDQ have been previously established (Goodman 2001). Final scoring revealed that approximately 23 percent of the students in this study’s sample have psychosocial difficulties, close to the German national average of 20 percent (KiGGs 2014).

For this study, externalising (8 percent of the sample) and internalising psychosocial difficulties (10 percent of the sample) were separately scored (SDQinfo 2016). The reason for this is that internalising and externalising students strongly differ in their behaviour and relationship patterns (Levesque 2014) and thus, are likely to impact groups and group dynamics in classrooms in fundamentally different ways (Jenson, Harward, and Bowen 2012; Ollendick, Shortt, and Sander 2005). The two variables used in the analysis are the percentage of students with internalising and externalising psychosocial difficulties in the classroom.

Students’ socioeconomic status

The SES index includes three items: the number of books at home, the number of computers at home, and the highest educational level among students’ parents. Construct validity of this index has been previously confirmed (Dietrich and Zimmermann 2019d). In this study SES has an internal reliability of α = .86.

Students’ accumulated grade point average (GPA)

This item asked students about their accumulated grade point average in the previous semester. Answers ranged from 1.0 to worse than 4.7; grades in Germany are given on a scale ranging from 1.0 (best) to 6.0 (worst).

Analyses

This study uses STATA 14 to conduct the analyses. Six multivariate regression models are created to predict the six social-emotional and achievement-related outcome variables – i.e. school-related efficacy, sense of purpose, academic-status insecurity, joy of learning, school-connectedness, and school-related anger – separately. The key independent variables of interest in all models are the interaction terms between the three teacher-student relationship variables and the two psychosocial difficulties variables. Each model controls for instructional teaching quality, i.e. the 6Cs composite score of instructional teaching quality (excluding Care), and the student body composition in the classroom in terms of SES-background and students’ GPA from the previous semester.

Equation (1)(1) $\begin{array}{rl}& Y_{ij}={\beta }_{0j}+{\beta }_{1j}{X}_{1j}+{\beta }_{2j}{X}_{2j}+{\beta }_{3j}{X}_{3j}+{\beta }_{4j}{X}_{4j}+{\beta }_{5j}{X}_{5j}+{\beta }_{6j}{X}_{1j}{X}_{4j}+{\beta }_{7j}{X}_{2j}{X}_{4j}\text{break}+{\beta }_{8j}{X}_{3j}{X}_{4j}+\\ & {\beta }_{9j}{X}_{1j}{X}_{5j}+{\beta }_{10j}{X}_{2j}{X}_{5j}+{\beta }_{11j}{X}_{3j}{X}_{5j}+\sum {\beta }_{qj}{X}_{qj}+r_{ij}\end{array}$(1) depicts the HLM applied in models 1 through 6 at level-1 (classroom level):

(1) $\begin{array}{rl}& Y_{ij}={\beta }_{0j}+{\beta }_{1j}{X}_{1j}+{\beta }_{2j}{X}_{2j}+{\beta }_{3j}{X}_{3j}+{\beta }_{4j}{X}_{4j}+{\beta }_{5j}{X}_{5j}+{\beta }_{6j}{X}_{1j}{X}_{4j}+{\beta }_{7j}{X}_{2j}{X}_{4j}\text{break}+{\beta }_{8j}{X}_{3j}{X}_{4j}+\\ & {\beta }_{9j}{X}_{1j}{X}_{5j}+{\beta }_{10j}{X}_{2j}{X}_{5j}+{\beta }_{11j}{X}_{3j}{X}_{5j}+\sum {\beta }_{qj}{X}_{qj}+r_{ij}\end{array}$(1)

In this equation Y_ij represents the model’s respective dependent variable measured for case i nested within school j (level-2). β_0 j is the intercept, β_1 jX_1j through β_3 jX_3j depict the three TSR-predictors and their coefficients, β_4 jX_4 j and β_5 jX_5 j depict the two predictors for students’ psychosocial difficulties and their coefficients, and β_7jX_2jX_4j through β_11jX_3jX_5j the interaction terms and their coefficients; ∑ β_qj X_qj depicts the sum of all remaining predictors of the model. The random error at level-1 is depicted as r_ij.

Equations (2)(2) ${\beta }_{pj}={\gamma }_{p0}$(2) and (3) depict the HLM at level-2 (school-level):

(2) ${\beta }_{pj}={\gamma }_{p0}$(2)

(3) ${\beta }_{0j}={\gamma }_{00k}+u_{0j}$(3)

All p independent variables, which include all predictors and interaction terms of equation (1)(1) $\begin{array}{rl}& Y_{ij}={\beta }_{0j}+{\beta }_{1j}{X}_{1j}+{\beta }_{2j}{X}_{2j}+{\beta }_{3j}{X}_{3j}+{\beta }_{4j}{X}_{4j}+{\beta }_{5j}{X}_{5j}+{\beta }_{6j}{X}_{1j}{X}_{4j}+{\beta }_{7j}{X}_{2j}{X}_{4j}\text{break}+{\beta }_{8j}{X}_{3j}{X}_{4j}+\\ & {\beta }_{9j}{X}_{1j}{X}_{5j}+{\beta }_{10j}{X}_{2j}{X}_{5j}+{\beta }_{11j}{X}_{3j}{X}_{5j}+\sum {\beta }_{qj}{X}_{qj}+r_{ij}\end{array}$(1) , are held constant at the school-level as depicted in equation (2)(2) ${\beta }_{pj}={\gamma }_{p0}$(2) ; the classroom-level constant β_0j is allowed to vary at the school-level, as depicted by the random factors u_0j (level-2) in equation (3).

All models exclude outliers with Cook’s Distance > 4/n, with the exception of Model A. An Inter-Correlation-Coefficient (ICC) analysis revealed that school-related efficacy, the outcome variable of Model A, has < 1% variance explained at the school-level, rendering a hierarchical linear analytical approach unnecessary. For this reason, Model A applies STATA 14’s rreg-command for robust regression, which weighs the impact of outliers based on Cook’s Distance and is thus, a more precise and effective way to deal with outliers than the simple elimination of outliers (Rousseeuw and Leroy 2005). Unfortunately, robust regression and HLM cannot be combined.

Missing data

With the exception of the three SES variables and previous semester GPA, all items used in this study have between 5% and 15% missing values. The number of books at home has 23% missing values, the number of computers at home 19%, the highest educational degree among parents 38%. Previous GPA has 19% missing values.

This study used subgroup-mean imputation to treat missing data (Musil et al. 2002), i.e. all missings of students were replaced with their classmates’ average. This approach was considered the most economical solution to handle the higher missing rates of the SES variables. German secondary schools and classrooms tend to be socioeconomically highly homogeneous, which is in part the result of Germany’s early tracking system and advanced placement course system (Georg 2018). This limits the threat of bias due to non-random missing values.

Results

Table 2 displays the regression results for the six models of this study, one for each of the six outcome variables. The focus in these models lies on the interaction-terms. Results partially confirm the hypothesis.

Table 2. Relationships*psychosocial difficulties interaction terms as predictors of student outcomes

	Model A	Model B	Model C	Model D	Model E	Model F
Dependent Variables:	Efficacy	Purpose	Enjoy Learn	Connect	Anger	Academ. Insec.
Interaction Terms
TSR Climate * Ext. PSD	0.598	1.377***	0.631	0.929**	-0.146	-0.0377
	(0.460)	(0.368)	(0.452)	(0.285)	(0.488)	(0.151)
TSR Climate * Int. PSD	-0.488	0.0650	-0.0790	0.322^+	-0.748**	0.827***
	(0.339)	(0.216)	(0.305)	(0.170)	(0.271)	(0.116)
T Emotional Trust * Ext. PSD	0.0633	0.266	0.0706	0.716**	0.282	0.973***
	(0.604)	(0.319)	(0.450)	(0.248)	(0.409)	(0.169)
T Emotional Trust * Int. PSD	-0.725	0.190	0.264	0.136	-1.502**	2.205***
	(0.844)	(0.435)	(0.550)	(0.334)	(0.567)	(0.198)
ST Identification * Ext. PSD	-0.679	-1.864***	-0.357	-1.614***	0.357	-1.986***
	(0.672)	(0.442)	(0.666)	(0.368)	(0.526)	(0.220)
ST Identification * Int. PSD	0.831	0.0887	-0.00188	0.112	1.210^+	-3.020***
	(1.008)	(0.571)	(0.840)	(0.443)	(0.732)	(0.236)
Independent Variables
TSR Climate	0.203	0.973***	0.226	0.683***	-0.0619	0.308***
	(0.273)	(0.210)	(0.301)	(0.164)	(0.258)	(0.0934)
T Emotional Trust	0.00918	0.228	0.174	-0.0879	-0.0715	2.507***
	(0.422)	(0.258)	(0.385)	(0.206)	(0.309)	(0.0956)
TS Identification	-0.279	-0.659	-0.319	-0.0845	0.447	-2.949***
	(0.718)	(0.412)	(0.621)	(0.320)	(0.524)	(0.140)
Externalizing PSD	-0.247	0.123	-0.748**	0.335^+	0.754**	-1.062***
	(0.311)	(0.230)	(0.280)	(0.187)	(0.276)	(0.128)
Internalizing PSD	-0.398	-1.258***	-0.488	-1.154***	0.659*	-0.273**
	(0.354)	(0.274)	(0.386)	(0.231)	(0.336)	(0.0930)
Controls
Instructinoal Quality	0.0177	-0.0579	0.282	0.0220	-0.600*	-0.351***
	(0.276)	(0.212)	(0.339)	(0.167)	(0.264)	(0.0683)
Average SES	-0.218	-0.388*	-0.714*	0.539***	-0.216	1.011***
	(0.236)	(0.189)	(0.321)	(0.155)	(0.198)	(0.0788)
Acc. Grade Average	0.388	0.252	0.552*	0.0489	-0.284	-0.240***
	(0.277)	(0.156)	(0.278)	(0.127)	(0.203)	(0.0486)
Constant	-0.269	-0.0854	-0.219	-0.0266	0.188	-0.180
	(0.157)	(0.146)	(0.220)	(0.128)	(0.136)	(0.284)
Observations	32	30	30	29	30	28
Adjusted R-squared	0.571
Number of groups	-	9	9	9	9	9

Standard errors in parentheses

*** p < 0.001, ** p < 0.01, * p < 0.05, ⁺ p < 0.1

TSR = teacher-student relationships; T = teacher; TS = teacher-student

Five of the interaction terms, between the internalising psychosocial difficulties and teacher-student relationship variables, are significant: the interaction term with the overall teacher-student relationship climate correlates with less class-related anger (Model L: standardised coefficient = −.75, p < .01), which is expected and supports hypothesis (b), but also more academic-status insecurity (Model M: standardised coefficient = .83, p < .001), which is unexpected and contradicts hypothesis (b); the interaction term with students’ emotional trust towards their teacher correlates with less class-related anger (Model L: standardised coefficient = −1.50, p < .01), as expected, and more academic-status insecurity (Model M: standardised coefficient = 2.21, p < .001), which is unexpected; the interaction term with students’ identification with the teacher correlates with less academic-status insecurity (Model M: standardised coefficient = −3.02, p < .001), as expected.

Seven of the interaction terms, between externalising psychosocial difficulties and the teacher-student relationship variables, are significant: the interaction term with the overall teacher-student relationship climate correlates with a stronger sense of school purpose (Model I: standardised coefficient = 1.38, p < .001) and more school-connectedness (Model M: standardised coefficient = .93, p < .01), as expected; the interaction term with students’ emotional trust towards their teacher correlates with more school-connectedness (Model K: standardised coefficient = .72, p < .01), as expected, and unexpectedly more academic-status insecurity (Model M: standardised coefficient = .97, p < .001); the interaction term with students’ identification with the teacher correlates with less academic-status insecurity (Model M: standardised coefficient = −1.99, p < .001), as expected, but also a lower sense of school purpose (Model I: standardised coefficient = −1.86, p < .001) and less school-connectedness (Model K: standardised coefficient = −1.61, p < .001), which are unexpected results.

Discussion

The majority of the results support the theory that positive teacher-student relationships are particularly important in classrooms with greater numbers of students with severe psychosocial difficulties, in accordance with the hypothesis. However, several of the results also challenge the hypothesis, indicating that stronger relationship-building skills among teachers do not (i.e. linearly) translate into better student outcomes across the board in classrooms with increasing numbers of students with severe psychosocial difficulties, but instead can even lead to worse outcomes.

An explanation for this might be that inner conflicts and difficult group dynamics arise from such constellations (Koch 2014). For example, students with internalising psychosocial difficulties might respond to positive relationship climates with stronger worries about their status among peers. As a result, students with severe internalising difficulties might not be able to participate in positive group dynamics. When they see how well others come along with each other in a cordial climate, they might wonder why they are unable to be a part of it and conclude that they themselves are at fault, triggering extremely negative emotions. This explanation would be in line with a similar phenomenon that has been revealed in the field of bullying research, the so-called healthy-context paradox (Huitsing et al. 2018). Students who get bullied in school climates with comparatively little bullying report less emotional well-being than students who get bullied in schools with more overall bullying. This finding has been explained with the idea that students in healthier school climates are more likely to internalise their bullying experiences, concluding that there is something fundamentally wrong with them, not their environment.

The finding that the interaction effect between higher emotional trust towards the teacher and having more students with psychosocial difficulties (internalising and externalising) in the classroom correlates with more academic-status insecurity might be explained similarly. Students with severe psychosocial difficulties tend to be particularly worried about how peers view their close relationships with teachers (Weiner 2007). Additional research, combining quantitative and qualitative research methods, is required to shed further light on these unexpected findings.

The reason why having more students with externalising psychosocial difficulties in combination with stronger teacher identification correlates with lower feelings of school purpose and school-connectedness might be the following: it is possible that these students do not want to ‘share’, with other needy students, a teacher who has (unconsciously) taken on a parental role (Hanifin and Appel 2000; Freud 1970). Nonetheless, this is what likely happens in a classroom with more students with severe psychosocial difficulties. These students might experience feelings of jealousy and disappointment, and feel neglected when the teacher they feel so closely connected to dedicates significant amounts of time and energy to other students. Externalising students might release their resulting inner emotional upheaval and feelings of frustration as aggressive behaviour (Berkowitz 1989), which leads to worse classroom climates that reduce all students’ positive feelings about school.

Hence, even teachers with highly advanced relationship-building skills may simply become overwhelmed in classrooms with too many externalising students because, as we know from psychoanalytic theory, emotional attention is a limited resource (Bruya and Tang 2018). This interpretation concurs with the results from a depth-hermeneutic analysis of teacher interviews, which revealed that highly committed teachers often face particularly severe feelings of helplessness and alienation when trying to reach students with severe psychosocial difficulties. As one teacher in that study explained: ‘It’s like a wall they form around themselves, that’s how it feels to me’ (Zimmermann 2016, 160).

Education policy leaders need to take note of this study’s results, which suggest that emphasising relationship-building skills in teachers’ professional development training can, overall, improve students’ social, emotional, and academic development. However, it is important to realise that too many students with psychosocial difficulties in the classroom might overwhelm even the best teachers. Thus, schools should prevent a concentration of such students in a single classroom. It might be a helpful guidance for education policy makers to estimate how many externalising students in a classroom are still manageable for teachers of different skill levels without leading to significant reductions in students’ social-emotional and academic outcomes.

Limitations

The main limitations of this study are the comparatively small sample size and cross-sectional nature of the data. As a result, the study does not have sufficient power to reveal small effects, and causalities of the relationships in the proposed models cannot be verified. Future research needs to test the proposed models with larger and longitudinal samples.

Conclusion

This study provides some evidence that strong relationship-building skills are important for teachers in classrooms with more students with severe psychosocial difficulties, above and beyond instructional teaching skills. However, some unexpected results indicate that the relationship is not linear. More quantitative and qualitative research studies are required.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. Confer = ‘encourage and value students’ ideas and views’, Captivate = ‘spark and maintain interest in learning’, Clarify = ‘help students understand content and resolve confusion’, Consolidate = ‘help students integrate and synthesize key ideas’, Challenge = ‘insist that students persevere and do their best work’, and Classroom Management = ‘foster orderly, respectful, and on-task classroom behavior’.