Teacher–Student Relationship Quality and Social, Academic, and Behavioral Adjustment are Associated Within- and Between-Persons from Kindergarten to Grade 6

Abstract

Although the significance of teacher–student relationships for children’s development has long been of interest to developmental scientists, few prior studies have used modeling approaches that explicitly document the intraindividual, dynamic processes that link teacher–student relationships with children’s adjustment. The present study used bias-adjusted Student-Teacher Relationship Scale (Pianta, 2001) scores to document the within-person developmental processes that link teacher–student relationship quality with key developmental outcomes (including social competence, academic competence, and externalizing and internalizing symptoms) from kindergarten through Grade 6. Data were drawn from the NICHD Study of Early Child Care and Youth Development (N = 1,041; 80.4% White, 12.9% Black, 6.1% Hispanic; 31.5% of mothers had a high school diploma or less). Within- and between-person bivariate associations were estimated using a series of latent curve models with structured residuals. Findings revealed consistent within-year associations with both school and home adjustment. Longitudinal findings suggested that within-person variations in teacher–student relationship quality are associated with subsequent academic functioning and externalizing symptoms. Furthermore, analyses suggested that social competence, academic competence, and externalizing symptoms are associated with subsequent teacher–student relationship quality. Results highlight the dynamic, bidirectional interactions between teacher–student relationships and children’s social, academic, and behavioral trajectories over time. Future studies are needed to explore teacher–student relationships and adjustment at different time scales and to understand the extent to which teacher–student relationships are impactful for different students at various levels of developmental risk.

Positive engagement with the academic and social demands of school has been highlighted as a key priority for children’s well-being (U.S. Department of Health and Human Services, n.d.). Therefore, understanding the mechanisms that promote positive school experiences and outcomes is a pressing concern. Theory and a well-developed empirical literature indicate that the relationships children form with their teachers likely play an important role in children’s school adjustment. Indeed, meta-analytic work has suggested that children who have more positive relationships with their teachers (i.e., experience more closeness and less conflict and dependency) also tend to have better relationships with peers and social adjustment (Endedijk et al., 2022; Magro et al., 2023), higher academic achievement and engagement (Lei et al., 2018; Roorda et al., 2011, 2017), and fewer behavioral and emotional problems (Lei et al., 2016; Roorda et al., 2020). Thus, promoting positive teacher–student relationships may improve children’s school adjustment and thereby increase long-term well-being.

Although it is clear that children who form better teacher–student relationships also tend to be better-adjusted to the school environment, the nature by which these processes unfold over time within individuals remain unclear. In other words, the established correlations between teacher–student relationship quality and children’s adjustment reveal little about the underlying developmental processes occurring within individuals—a level of analysis that is arguably most central to questions concerning developmental dynamics. In the present study, we leveraged panel data from the NICHD Study of Early Child Care and Youth Development (SECCYD) and contemporary methods for estimating longitudinal, within-person dynamics to test the extent to which the quality of children’s relationships with their teachers and their level of positive adjustment show reciprocal, self-organizing properties across the elementary school years.

Theoretical Models for Developmental Processes Linking Teacher–Student Relationship Quality with Children’s Adjustment

The Potential Role of Teacher–Student Relationships in Children’s Adjustment

Developmentalists have long applied the attachment perspective to clarify the nature and implications of teacher–student relationships (Spilt & Koomen, 2022). From this view, teachers who behave sensitively toward children in their classrooms serve as a secure base for children and increase children’s feelings of emotional security in the classroom. The provision of a secure base allows children to explore the classroom environment more freely and engage more in social and academic learning opportunities at school, thus facilitating a child’s social and academic development. Additionally, when children feel emotionally secure in the classroom, they are better adjusted to the classroom environment, thus lowering symptoms of psychopathology in that context and beyond (Sabol & Pianta, 2012; Verschueren & Koomen, 2012). Teachers may also serve as a safe haven, providing a supportive relationship to which distressed children can turn when faced with challenges too great to solve on their own. The availability of a safe haven can contribute to emotion regulation and problem-solving skills in the long term through co-regulatory processes. In contrast, children with high levels of conflict may be less openly receptive to comfort from a teacher, perhaps meeting such attempts with anger or resistance, thus reducing the amount of comfort offered over time (Spilt & Koomen, 2022).

Social-motivational theories have also been used to explain the importance of teacher–student relationships for healthy child adjustment (H. Davis, 2003). According to this perspective, when children’s needs for relatedness, autonomy, and competence are met through a close and nonconflictual relationship with a teacher, they will be more motivated to engage with their academic environment, which in turn contributes to more learning and higher academic performance. Social-motivational theory has primarily been applied to understand the ways that teacher–student relationships can support academic achievement (H. Davis, 2003). However, the theory may be relevant to other domains like externalizing behaviors, which children may be more motivated to reduce when such behaviors are in direct contrast with the “good student” identity that teachers are motivating students to internalize.

Finally, behavioral theorists have described the ways that modeling and social referencing processes may influence adjustment, particularly in the domain of social competence (Endedijk et al., 2022). Specifically, a teacher’s behavior toward a student can be a model for both the target student and their classmates. That is, a child’s peers may observe how a teacher treats the child and model their behavior after that of the teacher—behaving with more acceptance toward children who are liked by their teacher and avoiding those who experience a great deal of conflict. For example, experimental work (White & Jones, 2000; White & Kistner, 1992) has demonstrated that a teacher’s response to a student’s behavior in the classroom influences the attitudes of peers toward that student. Thus, students who have better relationships with teachers may experience more social acceptance, have more positive interactions with classmates, and be better emotionally adjusted in the classroom. In contrast, students who are in frequent conflict with their teachers may also be viewed more negatively by peers and have difficulty forming positive relationships with classmates.

The Potential Role of Children’s Adjustment in Relationships with Teachers

Attachment theorists have also posited that a child’s history of interpersonal relationships influences their way of interacting with new individuals through their internal working models (Fearon et al., 2016). Children who have a history of conflictual, ineffective, or exclusionary social interactions with others may form negative expectations for new social relationships. In contrast, children with a history of more positive social interactions may be more likely to interact with new relationship partners (e.g., teachers) in a prosocial manner that facilitates greater closeness and less conflict in the relationship.

According to socialization theory (Meija & Hoglund, 2016), student-teacher relationships are partially influenced by the extent to which students are emotionally and behaviorally well-adjusted to school. Children who are engaged in learning activities, interact well with peers, and are better able to regulate their emotions and behaviors may be better liked by teachers, who in turn invest more time and energy into building relationships with those children. Conversely, when children display difficult behaviors at school like fighting, crying, or withdrawal, teachers may be less motivated to engage with children or have fewer emotional and cognitive resources of their own to spend on building a positive relationship with a child. Furthermore, within a school, teachers may communicate with one another about “problem students,” thus potentially increasing the extent to which student behaviors affect relationships not only within one classroom and school year, but from one year to the next by influencing the expectations of a child’s future teachers.

Dynamic Associations Between Teacher–Student Relationship Quality and Children’s Adjustment

Multiple theoretical perspectives have suggested that there are developmental mechanisms that reciprocally link teacher–student closeness and conflict with children’s adjustment to one another over time. What remains unclear, however, are the dynamics of these associations as they unfold over time. Whether the documented meta-analytic associations noted earlier between teacher–student relationship quality and social competence, academic competence, and symptoms of psychopathology reflect actual influences of one construct on another within an individual over time is unknown. Thus, the theoretical perspectives outlined above regarding whether teacher–student relationships influence adjustment, adjustment influences teacher–student relationships, or both, have not been explicitly tested. Statistical models that explicitly test the extent to which these constructs are dynamically related within individuals, rather than simply estimating between-individual associations, are required to better understand the nature of these unfolding developmental processes.

Cross-lagged panel models (CLPM) have been used extensively in developmental psychology to estimate reciprocal, prospective pathways, such as the associations between teacher–student relationship quality and adjustment from one grade to the next. For example, McKinnon and Blair (2019) showed that teacher–student conflict was associated with lower subsequent reading skills and executive functioning over the course of kindergarten. Similarly, Pakarinen and colleagues (2020) showed in a sample of Finnish kindergarteners that teacher–student conflict was predictive of lower pre-literacy skills, literacy interest, pre-math skills, and math interest. Others have documented associations only in the reverse direction, including associations between academic performance and symptoms of psychopathology and subsequent teacher–student relationship quality (Hajovsky et al., 2017; Mejia & Hoglund, 2016; Pakarinen et al., 2018). Finally, some CLPM research has shown reciprocal associations between teacher–student relationship quality and psychopathology symptoms (de Jong et al., 2018; Roorda & Koomen, 2020), academic achievement (Košir & Tement, 2014; Zee et al., 2020), and school motivation (Zee et al., 2020).

Disaggregating Between- and Within-Person Associations to Create Interpretable Estimates

The traditional CLPM approach used in the studies reviewed above, however, is limited by its aggregation of between- and within-person variance, which muddles the interpretation of coefficients (Berry & Willoughby, 2017; Hamaker et al., 2015). That is, the CLPM fails to disaggregate stable, between-person differences (i.e., comparing individuals to one another), from within-person, time-varying associations (i.e., comparing individuals to themselves over time). This is problematic because we cannot assume that the valence and magnitude of findings at the between-person level is equivalent to those from within-person analyses. Indeed, within- and between-person effects are frequently of different magnitudes and sometimes even of different valences (Snijders & Bosker, 2012). Thus, estimates that do not disaggregate variance at both levels are challenging to interpret. Given that between- and within-person effects have substantively different meanings, they cannot be interpreted unless separated or, in rare cases, demonstrated to be equivalent (i.e., “convergent”; Bryk & Raudenbush, 1992).

For example, Serdiouk and colleagues (2016) found that at the within-person level, positive teacher–student relationship quality was associated with higher levels of peer victimization, whereas at the between-person level, positive teacher–student relationship quality was associated with lower levels of victimization. These results suggest that children who generally tend to have positive relationships with their teachers tend to generally experience lower levels of victimization. However, when children become closer than is typical (for that child) with their teachers (e.g., seeking out more time with or comfort from teachers), this may be a warning sign that they are also experiencing an uptick in peer victimization. Thus, understanding both the within- and between-person relations between teacher–student relationship quality and peer victimization may help practitioners identify children who are chronically versus acutely at risk for adverse peer experiences. In short, separating between- and within-person effects statistically makes results both more interpretable and transferable to applied contexts.

Using Within-Person Analyses to Rule Out Time-Invariant Between-Person Confounds

The CLPM and meta-analytic findings reviewed above suggest that teacher–student relationship quality and child adjustment are statistically related (Lei et al., 2016; Magro et al., 2023; Roorda et al., 2011, 2017, 2020). However, the internal validity of these findings is often unclear because it is challenging to rule out all the numerous variables that may confound these relations (e.g., intelligence, social motivation). An added benefit of focusing on within-person processes is that such models hold the individual—and, thus, all time-invariant confounds—constant. For example, when an individual’s relationship with their teacher is more positive than is typical for that individual, and adjustment is healthier than would be expected for that individual, stable covariates that vary between individuals like birth sex and genetic makeup can be ruled out as potential explanations for an observed association.

Some researchers have used multilevel modeling to examine within-person associations between teacher–student relationship quality and developmental outcomes. For example, Maldonado-Carreño and Votruba-Drzal (2011) found within-person associations between overall teacher–student relationship quality and teacher-reported academics, mother- and teacher-reported internalizing symptoms, and teacher-reported externalizing symptoms. McCormick and O’Connor (2015) observed within-person associations between teacher–student closeness and reading achievement, such that children who were experiencing higher-than-typical levels of closeness were also demonstrating higher-than-expected levels of achievement at the same time point. These within-person findings suggest that contemporaneous associations between teacher–student relationship quality and children’s adjustment are robust after holding time-invariant confounds constant.

Importantly, multilevel models have typically been used to estimate concurrent associations rather than prospective relations. When attempting to understand the directionality of associations between teacher–student relationships and adjustment, however, it is also of interest to estimate prospective associations, as is done in CLPM. Thus, an ideal analytic approach is one that both separates within- and between-person associations (as in the case of traditional multilevel models) and estimates concurrent as well as longitudinal associations between constructs of interest (as in the case of traditional CLPM).

The Present Study

The present study employed the latent curve model with structured residuals (LCM-SR, Curran et al., 2014; also known as the autoregressive latent trajectory model with structured residuals, e.g., Berry et al., 2017) to assess the within-person cross-lagged relations between teacher–student relationship quality and child adjustment. LCM-SR models allow for the separation of within- and between-person relations by fixing an effect to each individual, thus estimating within-person deviations from any individual’s trajectory. Thus, the model produces distinct estimates of the intra- and interindividual relations between two constructs.

Two recent studies have employed the LCM-SR to answer questions about the within-person dynamics between teacher–student relationship quality and adjustment. Murray and colleagues (2021) assessed a sample of Swiss children at ages 11, 13, and 15. Within-person, cross-lagged effects were observed between higher quality teacher–student relationships and lower internalizing symptoms (Murray et al., 2021). Additionally, aggression was predictive of lower subsequent teacher–student relationship quality (Murray et al., 2021). Similarly, in a sample of Canadian preschoolers, Zatto (2018) observed reciprocal cross-lagged effects between teacher–student conflict and child depressive symptoms, as well as an effect of depression on subsequent teacher–student dependency. These studies demonstrate the potential of the LCM-SR model to answer questions about the connections between teacher–student relationships and child adjustment within individuals over time.

The present study builds upon previous research in this area (Murray et al., 2021; Zatto, 2018) by examining the within-person, reciprocal relations between teacher–student relationship quality (i.e., conflict and closeness) and key indicators of child adjustment (i.e., social competence, academic competence, externalizing symptoms, and internalizing symptoms) from kindergarten through Grade 6. This age range was not explored in prior research using an LCM-SR modeling approach, which represents a significant gap in the literature given that teacher–student relationships may serve different functions at different age ranges. Furthermore, Murray and colleagues (2021) and Zatto (2018) explored only associations with internalizing symptoms, aggression, and peer relationship quality, leaving open questions about the dynamics that link teacher–student relationship quality with academics as well as social competence and externalizing symptoms more broadly.

Given that teacher–student conflict and closeness are related, but separate, dimensions of teacher–student relationship quality, they were modeled independently. In general, we expected teacher–student conflict to be negatively associated with social and academic competence and positively associated with symptoms of psychopathology, whereas we expected teacher–student closeness to be positively associated with social and academic outcomes and negatively associated with symptoms of psychopathology at the within-person level. Preregistered hypotheses included expectations to observe: (H1) Reciprocal, cross-lagged associations between teacher–student conflict and each indicator of adjustment; (H2a) Null or small (relative to conflict) cross-lagged paths from teacher–student closeness to each indicator of adjustment; and (H2b) small (relative to conflict) cross-lagged paths from adjustment to teacher–student closeness.

Method

Transparency and openness

All analyses were preregistered using the OSF Secondary Data Preregistration template. Preregistration and analysis scripts for replication can be found on the project’s OSF page (Magro et al., 2022). Data are available to qualified individuals at https://www.icpsr.umich.edu/web/ICPSR/series/00233.

Participants

The present study used data collected from families recruited as part of the SECCYD (NICHD Early Child Care Research Network, 2001).¹ Sample children were recruited beginning in 1991 from ten sites around the U.S. All participating families were required to satisfy inclusion criteria for both mothers (spoke English, over age 18, no known substance abuse, not planning to move in the next three years, no medical illness, not enrolled in another study, lived in a neighborhood that was deemed safe for home visitation by the original study researchers, lived within one hour of a study site) and children (no known disability or serious medical complications, no birth complications requiring extended hospitalization). Of the 8,996 newborns screened, 5,416 were identified as eligible and 1,364 were randomly selected to participate in the study. Primary caregivers for the included sample of study children reported on the birth sex, race, and ethnicity of their children. The original sample of study children was 48.3% male. Most study participants were identified as White (80.4%), whereas 12.9% were Black, 1.6% were Asian, and 0.4% were Native American. Furthermore, 4.7% of study children were identified as “Other race/ethnicity” by their caregivers; most of these children were multiracial. Additionally, mothers identified 6.1% of the participants as Hispanic. The original sample had an average annual household income of $37,781 when study children were one month of age, whereas the average annual household income in the U.S. at that time was $36,875. At the beginning of the study, 10.4% of mothers had not completed high school, 21.1% had a high school diploma or equivalent, 33.2% had completed some college, 20.8% had earned a bachelor’s degree, and 14.5% had earned an advanced degree.

Measures

Descriptive statistics for all study variables are reported in the Online Supplement (Supplemental Table 1). Correlations between covariates, teacher–student relationship quality, and each adjustment indicator are reported in Supplemental Tables 2 through 4. Developmentally relevant indicators of adjustment and covariates were selected to be consistent with prior work based in the SECCYD (e.g., Fraley et al., 2013; Haltigan et al., 2013; Kunkel et al., 2022; Magro et al., 2020; Roisman et al., 2012).

Teacher–Student Relationship Quality

Every year from kindergarten to sixth grade, participants’ classroom teachers completed the Student-Teacher Relationship Scale, Short Form (STRS; Pianta, 2001). Teachers rated participants on 15 items on a Likert-type scale (1 = “definitely does not apply” to 5 = “definitely applies”). The STRS includes two subscales relating to teacher–student conflict (seven items) and teacher–student closeness (eight items). Teacher–student conflict is conceptualized as the degree to which a teacher perceives their relationship with the child as negative, conflictual, and characterized by struggle (e.g., “Dealing with this child drains my energy”). Teacher–student closeness, on the other hand, refers to the degree to which a teacher perceives their relationship with the child as warm, affectionate, and characterized by open communication (e.g., “It is easy to be in tune with what this child is feeling”). Initial tests of configural invariance supported a two-factor structure in every grade, with standardized factor loadings ranging from .62 to .83 for conflict and .39 to .79 for closeness. Given the prior documented measurement non-invariance in STRS items according to child characteristics like age, race, and sex in other samples (e.g., Aboagye et al., 2019; Koomen et al., 2012; Webb & Neuharth-Pritchett, 2011), STRS factor scores were extracted using moderated nonlinear factor analysis models (Bauer, 2017; Curran et al., 2014) via the aMNLFA R package (Cole et al., 2021), simultaneously adjusting for potential item bias in age, gender, race/ethnicity, maternal education level, and family income-to-needs ratio (we used the approach described in and producing identical results to Magro et al., 2024). The resulting factor scores were highly correlated with raw mean scores for both conflict (Pearson’s r = .92 to .93) and closeness (Pearson’s r = .96 to .98) across grades.

Social Competence

Participants’ classroom teachers reported on their social competence in the classroom setting annually from kindergarten through Grade 6 using the Social Skills Rating System – Teacher Form (SSRS; Gresham & Elliott, 1990). Items related to cooperation (e.g., putting away work materials properly), assertion (e.g., starting conversations with peers), and self-control (e.g., responding to teasing from peers appropriately). Teachers rated how often the target child exhibited each of 30 behaviors on a Likert-type scale (0 = “never” to 2 = “very often”). Items were summed within each grade (Cronbach’s α = .90 to .94) to create a total social skills score, with higher scores indicating greater social competence in the classroom.

Children’s mothers completed the SSRS – Parent Form (Gresham & Elliott, 1990) to report on children’s social behaviors in the home. Scores from Grades 1, 3, and 5 were used in the present analyses. Items related to cooperation (e.g., helping members of the household), assertion (e.g., starting conversations with others), self-control (e.g., controlling temper), and responsibility (e.g., showing appropriate regard for property). Mothers rated how often the target child exhibited each of 38 behaviors on a Likert-type scale (0 = “never” to 2 = “very often”). Items were summed within each grade (Cronbach’s α = .88 to .90) to create overall social skills scores, with higher scores indicating greater social competence in the home.

Academic Competence

A second subscale from the SSRS – Teacher Form (Gresham & Elliott, 1990) was used to assess teachers’ perceptions of target participant’s academic competence. Nine items were rated on a Likert-type scale (0 = lowest 10% of the class to 5 = highest 10% of the class) annually from kindergarten through Grade 5. Items related to reading skill, math skill, academic motivation, parental encouragement to succeed academically, and intellectual abilities. Items were summed within each grade (Cronbach’s α = .95 to .96) to create an overall score of teacher perception of academic competence, with higher scores indicating greater academic competence in the classroom.

Additionally, target participants were administered reading and mathematics subtests from the Woodcock-Johnson Psycho-Educational Battery – Revised (Woodcock, 1990; Woodcock & Johnson, 1989) in Grades 1, 3, and 5. In Grade 1, the Letter-Word Identification, Word Attack, and Applied Problems subtests were administered. In Grade 3, the Letter-Word Identification, Word Attack, Passage Comprehension, Applied Problems, and Calculation subtests were administered. In Grade 5, the Letter-Word Identification, Passage Comprehension, Applied Problems, and Calculation subtests were administered. Within each grade, standard scores for each subtest were averaged (Cronbach’s α = .85 to .89) to create academic achievement scores.

Internalizing and Externalizing Symptoms

Each year from kindergarten to Grade 6, teachers completed the Teacher Report Form (TRF; Achenbach, 1991b). Teachers responded to 120 items regarding students’ problem behaviors on a Likert-type scale (0 = “Not True (as far as you know)” to 2 = “Very True or Often True”). Items assessed a range of behavior problems, including externalizing symptoms (e.g., “Destroys property belonging to others”) and internalizing symptoms (e.g., “Complains of loneliness”). Raw scores were created by summing items for the externalizing (Cronbach’s α = .94 to .95) and internalizing (Cronbach’s α = .85 to .89) scales. Given that symptoms of psychopathology are highly skewed in typically developing samples like the SECCYD, they are not appropriate for inclusion in models that assume a normal distribution of errors. Skewness of raw scores ranged from 2.10 (Grade 1) to 2.59 (Grade 6) for externalizing symptoms and from 1.60 (Grade 4) to 2.31 (kindergarten) for internalizing symptoms. As such, TRF scores were transformed into categorical variables to represent low (0 ≤ x < 1 SD), borderline (1 SD ≤ x < 2 SD), and clinical (2 SD ≤ x) levels of psychopathology in the classroom context for each year. See Supplemental Table 5 for a summary of TRF categorical distributions across grades.

In a parallel fashion to teachers, mothers of target participants completed the Child Behavior Checklist (CBCL; Achenbach, 1991a). Scores from Grades 1, 3, and 5 were used in the present analyses. Mothers responded to 120 items regarding their child’s problem behaviors on a Likert-type scale (0 = “Not True (as far as you know)” to 2 = “Very True or Often True”). Raw scores were created by summing items from the externalizing (Cronbach’s α = .89 to .90) and internalizing (Cronbach’s α = .81 to .86) scales. Skewness of raw scores ranged from 1.20 (Grade 3) to 1.44 (Grade 5) for externalizing symptoms and from 1.57 (Grade 3) to 2.02 (Grade 5) for internalizing symptoms. In a similar fashion to TRF scores, CBCL scores were transformed into categorical variables to represent low (0 ≤ x < 1 SD), borderline (1 SD ≤ x < 2 SD), and clinical (2 SD ≤ x) levels of psychopathology in the home context for each year. See Supplemental Table 5 for a summary of CBCL categorical distributions across grades.

Covariates

When study children were one month old, mothers provided information on child birth sex (coded as 0.5 = male, −0.5 = female), child race/ethnicity (coded as White/non-Hispanic = 0.5, Other = −0.5), and their own educational level (coded to represent approximate years of education, e.g., “bachelor’s degree” = 16; scores were then mean-centered). Mothers reported on household income when the study child was 1, 6, 15, 24, and 36 months old. Income-to-needs ratios were calculated to represent the ratio of the family’s income to the poverty threshold for that family (calculated based on the number of adults and children living full-time in the household). These values were then averaged (α = .97) and mean-centered to create an overall income-to-needs ratio variable. Notably, all covariates were treated as time-invariant and only used at the between-person level.

Analytic Approach

The LCM-SR model (Curran et al., 2014) was used to estimate the associations between teacher–student relationship quality and each developmental outcome of interest at both the between- and within-persons levels.

Univariate Analyses

Prior to estimating associations across variables, we first established univariate growth models for each of the respective variables (i.e., teacher–student closeness and conflict; teacher-reported social competence, academic competence, internalizing symptoms, and externalizing symptoms; mother-reported social competence, internalizing symptoms, and externalizing symptoms; and Woodcock-Johnson scores), as shown in Figure 1. We estimated taxonomies of univariate LCM-SR models, gradually adding fixed effects and between-person variances to the growth trajectories across nested models (Singer & Willett, 2003). Comparisons of nested models were based on likelihood ratio tests (or adjusted likelihood ratio tests for the ordinal externalizing and internalizing variables). Time was centered at the kindergarten assessment (or the Grade 1 assessment, when kindergarten data were not available) and scaled as grade. With continuous variables, we made typical normality assumptions about the distributions of the within-person residual (co)variances and estimated each model using the FIML estimator in Mplus version 8.3 (L. Muthén & B. Muthén, 2017). With ordinal variables, we used a weighted least squares mean and variance adjusted (WLSMV) estimator with a theta parameterization to allow for the structured residual terms to have defined variances. In all models, we constrained the respective autoregressive parameters and all but the initial timepoint structured residuals to equality over time. In the WLSMV-estimated models, the initial structured residual was constrained to equal one to identify the model (i.e., required given theta parameterization). We managed our workflow using the R package MplusAutomation version 1.0 (Hallquist & Wiley, 2018).

Figure 1. Theoretical Latent Growth Model with Structured Residuals.

Note. The variance of the initial structured residual term (ε_K) is freely estimated for models with continuous observed variables and constrained to equal 1 for models with ordinal variables. For continuous variable models, intercepts for observed variables are constrained to equal 0. For ordinal variable models, the mean of the latent intercept is constrained to equal 0 and observed variable thresholds are constrained to be equal across time points. K = kindergarten. G1 = Grade 1. G2 = Grade 2. G3 = Grade 3. G4 = Grade 4. G5 = Grade 5. G6 = Grade 6. rv = residual variance.

Bivariate Analyses

To estimate the within-person cross-lagged associations between teacher–student relationship quality and developmental outcomes, we integrated pairs of univariate models using the LCM-SR specification (Curran et al., 2014). Specifically, we fitted a total of sixteen LCM-SR models to assess the cross-lagged associations between the two teacher–student relationship variables (i.e., closeness and conflict) and the eight outcome variables of interest (i.e., teacher- and mother-reported social competence, teacher-reported academic competence, academic achievement, teacher- and mother-reported externalizing symptoms, and teacher- and mother-reported internalizing symptoms). Figure 2 shows an example theoretical model for the relations between teacher–student relationship quality and (continuous) developmental outcomes at the between- and within-person levels.

Figure 2. Theoretical LCM-SR Model for the Within- and Between-Person Associations Between Teacher–Student Relationship Quality and Developmental Competence.

Note. All model constraints from Figure 1 are present, but not displayed for readability. Square brackets indicate that all random between-person latent variables (intercepts, linear slopes, and quadratic slopes) are allowed to covary with one another. Covariates (birth sex, race/ethnicity, maternal education, and income-to-needs ratio) are not shown but are allowed to predict each of the random between-person latent variables. TSRQ = teacher–student relationship quality. COMP = developmental competence. K = kindergarten. G1 = Grade 1. G2 = Grade 2. G3 = Grade 3. G4 = Grade 4. G5 = Grade 5. G6 = Grade 6.

At the between-person level, random intercepts, linear slopes, and quadratic slopes (where indicated, as determined from the univariate growth models) were allowed to covary across the two longitudinal panels (i.e., closeness or conflict with the child outcome). Between-person effects were regressed on the four covariates. At the within-person level, residual covariances across the panels were estimated at each time point to capture contemporaneous relations. These within-timepoint covariances were constrained to equality for all but the initial timepoint (kindergarten). We also added cross-lagged relations between each panel to capture the bi-directional, lagged relations between teacher–student relationship quality and the child outcome variable over time. Regression parameters representing the same directional relations were constrained to equality over time for parsimony.

Results from each model were interpreted separately for between- and within-person effects. Standardized parameter estimates were calculated using between-person residual variances for between-person estimates and within-person residual variances for within-person estimates. Parameters associated with p < .05 after a Benjamini-Hochberg multiple tests correction within each model were considered statistically significant (Benjamini & Hochberg, 1995).

Missing Data

Of the 1,364 participants selected for inclusion in the SECCYD, 1,064 participants had at least three STRS forms completed between kindergarten and Grade 6 (78.0%). Furthermore, participants missing data on any covariates were dropped, resulting in a final sample size of 1,041 (76.3%) for the teacher-report analyses. Included participants were more likely to be White/non-Hispanic compared to those not included in the analyses (t = 3.08, d = 0.28, p < .001). Furthermore, included participants had, on average, higher levels of maternal education (t = 6.01, d = 0.53, p < .001) and higher income-to-needs ratios (t = 5.62, d = 0.49, p < .001). There were no differences between the groups in terms of participant birth sex (t = 1.54, d = 0.13, p = .124).

For the mother-report and Woodcock-Johnson analyses, 781 participants (57.3%) had STRS forms available in Grades 1, 3, and 5 and no missing covariate values and could thus be included. Included participants were more likely to be female (t = 3.04, d = 0.17, p = .002) and White/non-Hispanic (t = 4.49, d = 0.27, p < .001) compared to those not included in the analyses. Furthermore, included participants had, on average, higher levels of maternal education (t = 4.80, d = 0.28, p < .001) and higher income-to-needs ratios (t = 3.22, d = 0.20, p = .001).

Results

Univariate Analyses

Univariate model fit estimates and results of likelihood ratio tests comparing nested univariate models for each variable of interest are summarized in Supplemental Table 6. A narrative review of the univariate analyses is available in the Online Supplement.

Bivariate Analyses

Model fit estimates for all bivariate models are presented in Supplemental Table 7. A summary of key results from all bivariate models is displayed in Table 1. Detailed parameter estimates are provided in Supplemental Tables 8 through 23. Key findings are described throughout the following sections.

Table 1.

Summary of Within-Person and Between-Person Results for Relations Between Teacher–Student Relationships and Children’s Developmental Competencies

Variable	Conflict & Teacher Reports		Conflict & Mother Reports / WJ		Closeness & Teacher Reports		Closeness & Mother Reports / WJ
	Effect	Stand. Est.	Effect	Stand. Est.	Effect	Stand. Est.	Effect	Stand. Est.
Within-Person Results
Social competence	Con ~ Soc	−0.47	Con ~ Soc	−0.19	Clo ~ Soc	0.42	Clo ~ Soc	0.12
	Con → Soc		Con → Soc		Clo → Soc		Clo → Soc
	Soc → Con		Soc → Con	−0.11	Soc → Clo		Soc → Clo
Academic competence	Con ~ Aca	−0.26	Con ~ Aca	−0.16	Clo ~ Aca	0.23	Clo ~ Aca	0.16
	Con → Aca		Con → Aca	−0.13	Clo → Aca		Clo → Aca	0.12
	Aca → Con		Aca → Con		Aca → Clo		Aca → Clo	0.10
Externalizing problems	Con ~ Ext	0.48	Con ~ Ext	0.16	Clo ~ Ext	−0.20	Clo ~ Ext
	Con → Ext	0.11	Con → Ext	0.21	Clo → Ext		Clo → Ext
	Ext → Con	0.18	Ext → Con		Ext → Clo	−0.07	Ext → Clo
Internalizing problems	Con ~ Int	0.24	Con ~ Int		Clo ~ Int	−0.16	Clo ~ Int
	Con → Int		Con → Int		Clo → Int		Clo → Int
	Int → Con		Int → Con		Int → Clo		Int → Clo
Between-Person Results
Social competence	Conint ~ Socint	−0.77	Conint ~ Socint	−0.28	Cloint ~ Socint	0.59	Cloint ~ Socint	0.30
	Conint ~ Socslope				Cloint ~ Socslope
	Socint ~ Conslope	0.34			Socint ~ Closlope	−0.58
	Conslope ~ Socslope	−0.77			Closlope ~ Socslope	0.65
Academic competence	Conint ~ Acaint	−0.36	Conint ~ Acaint		Cloint ~ Acaint	0.41	Cloint ~ Acaint
	Conint ~ Acaslope				Cloint ~ Acaslope
					Acaint ~ Closlope	−0.43
					Closlope ~ Acaslope
Externalizing problems	Conint ~ Extint	0.88	Conint ~ Extint	0.28	Cloint ~ Extint	−0.21	Cloint ~ Extint
	Extint ~ Conslope				Extint ~ Closlope	0.24
Internalizing problems	Conint ~ Intint	0.41	Conint ~ Intint		Cloint ~ Intint	−0.46	Cloint ~ Intint
	Intint ~ Conslope	−0.30	Conint ~ Intslope		Intint ~ Closlope	0.52

Note. Complete model results are displayed in Supplemental Tables 8 through 23. All included cross-domain pathways are displayed, with black results being associated with a p-value < .05 after Benjamini-Hochberg correction and grey results being non-statistically significant. Standardized estimates are included only for statistically significant results and are calculated using within- or between-person residuals as appropriate. WJ = Woodcock-Johnson. Stand. Est. = standardized estimate. ~ indicates a covariance. → indicates a regression.

Teacher–Student Conflict

Teacher-Reported Social Competence (Supplemental Table 8).

At the between-person level, intercepts were associated, such that higher levels of conflict were associated with lower levels of social competence in kindergarten (${\psi }_{in{t}_{con},in{t}_{soc}}=-2.63$, p = .003, ${\psi }_{stand}=-0.77$). Furthermore, higher social competence in kindergarten was associated with greater instantaneous growth in conflict (${\psi }_{in{t}_{soc},slop{e}_{con}}=0.15$, p = .008, ${\psi }_{stand}=0.34$). Initial levels of conflict were not associated with social competence slopes. Finally, there was a negative relation between the growth rates, such that students showing more positive (instantaneous) linear increases in conflict tended to show more rapid decreases in social competence over time (${\psi }_{slop{e}_{con},slop{e}_{soc}}=-0.04$, p = .003, ${\psi }_{stand}=-0.77$).

Within-persons, there were statistically significant autoregressive relations for conflict (B = 0.05, p = .033, β = 0.05 to 0.06) and teacher-reported social competence (B = 0.07, p = .009, β = 0.07). Although there was no evidence for within-person cross-lagged associations, the statistically significant residual covariances indicated contemporaneous inverse relations in kindergarten (${\sigma }_{{con,soc}_{t0}}=-2.75$, p = .003, ${\sigma }_{stand}=-0.49$) and Grades 1 through 6 (${\sigma }_{{con,soc}_{t1-t6}}=-2.25$, p = .003, ${\sigma }_{stand}=-0.47$).

Mother-Reported Social Competence (Supplemental Table 9).

At the between-person level, intercepts were associated, such that higher conflict was associated with lower social competence in Grade 1 (${\psi }_{in{t}_{con},in{t}_{soc}}=-0.99$, p = .002, ${\psi }_{stand}=-0.28$).

Within-persons, there was a statistically significant autoregressive relation for mother-reported social competence (B = 0.26, p = .002, β = 0.24 to 0.26) but not conflict. Furthermore, cross-lagged paths indicated that higher-than-expected levels of social competence were associated with subsequently lower-than-expected levels of conflict two years later (B = −0.01, p = .044, β = −0.11 to −0.10). The lagged relation from conflict to subsequent social competence was statistically non-significant. Finally, higher-than-expected levels of conflict were contemporaneously associated with lower-than-expected levels of social competence in Grades 3 and 5 (${\sigma }_{{con,soc}_{t1-t2}}=-0.75$, p = .002, ${\sigma }_{stand}=-0.19$).

Teacher-Reported Academic Competence (Supplemental Table 10).

At the between-person level, intercepts were associated, such that higher levels of conflict were associated with lower levels of academic competence in kindergarten (${\psi }_{in{t}_{con},in{t}_{aca}}=-1.06$, p = .002, ${\psi }_{stand}=-0.36$). Conflict in kindergarten was not associated with linear change in academic competence over time.

Within-persons, there were statistically significant autoregressive relations for both conflict (B = 0.07, p = .002, β = 0.07 to 0.08) and teacher-reported academic competence (B = 0.16, p = .002, β = 0.16 to 0.19). There was no evidence for cross-lagged associations between the constructs. Finally, higher-than-expected levels of conflict were associated with lower-than-expected levels of academic competence in the same year in kindergarten (${\sigma }_{{con,aca}_{t0}}=-1.13$, p = .002, ${\sigma }_{stand}=-0.25$) and Grades 1 through 5 (${\sigma }_{{con,aca}_{t1-t5}}=-0.83$, p = .002, ${\sigma }_{stand}=-0.26$).

Directly Assessed Academic Competence (Supplemental Table 11).

At the between-person level, intercepts were not associated.

Within-persons, there was a statistically significant autoregressive relation for directly assessed academic competence (B = 0.19, p = .007, β = 0.19 to 0.27), but not conflict. There were significant cross-lagged paths from conflict to Woodcock-Johnson scores, such that higher-than-expected levels of conflict were associated with lower-than-expected Woodcock-Johnson at the next timepoint (B = −0.97, p = .008, β = −0.13 to −0.12). Woodcock-Johnson scores were not associated with subsequent conflict. There were also contemporaneous within-person inverse associations in Grades 3 and 5 (${\sigma }_{{con,aca}_{t1-t2}}=-0.56$, p = .007, ${\sigma }_{stand}=-0.16$).

Teacher-Reported Externalizing Symptoms (Supplemental Table 12).

At the between-person level, intercepts were associated, such that higher conflict was associated with higher externalizing symptoms in kindergarten (${\psi }_{in{t}_{con},in{t}_{ext}}=0.42$, p = .002, ${\psi }_{stand}=0.88$). Kindergarten externalizing symptoms were not associated with the instantaneous linear slope of conflict.

Within-persons, there were statistically significant autoregressive paths for teacher-reported externalizing symptoms (B = 0.17, p = .002, β = 0.17 to 0.18), but not conflict. Furthermore, higher-than-expected levels of conflict were associated with higher-than-expected levels of externalizing problems one year later (B = 0.15, p = .002, β = 0.11). Conversely, higher-than-expected levels of externalizing problems were associated with higher-than-expected levels of conflict in the subsequent year (B = 0.13, p = .002, β = 0.18 to 0.19). The constructs were also contemporaneously associated in kindergarten (${\sigma }_{{con,ext}_{t0}}=0.29$, p = .002, ${\sigma }_{stand}=0.43$) and Grades 1 through 6 (${\sigma }_{{con,ext}_{t1-t6}}=0.30$, p = .002, ${\sigma }_{stand}=0.48$).

Mother-Reported Externalizing Symptoms (Supplemental Table 13).

At the between-person level, intercepts were associated, such that higher conflict was associated with higher externalizing symptoms in Grade 1 (${\psi }_{in{t}_{con},in{t}_{ext}}=0.15$, p = .026, ${\psi }_{stand}=0.28$).

Within-persons, there were no statistically significant autoregressive relations. There was a significant cross-lagged pathway from conflict to subsequent externalizing symptoms, such that children whose teachers reported higher-than-expected levels of conflict demonstrated higher-than-expected levels of behavior problems at home two years later (B = 0.33, p = 0.14, β = 0.20 to 0.21). The cross-lagged pathway from externalizing symptoms to subsequent conflict was not statistically significant. In terms of contemporaneous associations, higher-than-expected levels of conflict were associated with higher-than-expected levels of externalizing symptoms in Grades 3 and 5 (${\sigma }_{{con,ext}_{t1-t2}}=0.12$, p = .026, ${\sigma }_{stand}=0.16$).

Teacher-Reported Internalizing Symptoms (Supplemental Table 14).

At the between-person level, intercepts were associated, such that higher conflict was associated with higher internalizing symptoms in kindergarten (${\psi }_{in{t}_{con},in{t}_{int}}=0.11$, p = .002, ${\psi }_{stand}=0.41$). Furthermore, higher levels of kindergarten internalizing symptoms were associated with smaller instantaneous linear slopes for conflict (${\psi }_{in{t}_{int},slop{e}_{con}}=-0.01$, p = .012, ${\psi }_{stand}=-0.30$).

Within-persons, there were statistically significant autoregressive relations for both conflict (B = 0.10, p = .002, β = 0.09 to 0.10) and teacher-reported internalizing symptoms (B = 0.08, p = .012, β = 0.07 to 0.08). There was no evidence of cross-lagged associations. However, contemporaneous associations were observed in kindergarten (${\sigma }_{{con,int}_{t0}}=0.15$, p = .002, ${\sigma }_{stand}=0.23$) and Grades 1 through 6 (${\sigma }_{{con,int}_{t1-t6}}=0.18$, p = .002, ${\sigma }_{stand}=0.24$).

Mother-Reported Internalizing Symptoms (Supplemental Table 15).

At the between-person level, there was no evidence for associations between conflict and internalizing symptoms intercepts. Initial levels of conflict were not associated with linear change in internalizing symptoms.

At the within-person level, neither conflict nor mother-reported internalizing symptoms showed evidence for significant autoregressive paths. Furthermore, there was no evidence for cross-lagged or contemporaneous associations.

Teacher–Student Closeness

Teacher-Reported Social Competence (Supplemental Table 16).

At the between-person level, intercepts were associated, such that children with higher closeness also had higher social competence in kindergarten (${\psi }_{in{t}_{clo},in{t}_{soc}}=1.69$, p = .002, ${\psi }_{stand}=0.59$). Furthermore, higher kindergarten social competence was associated with more negative instantaneous growth in closeness (${\psi }_{in{t}_{soc},slop{e}_{clo}}=-0.23$, p = .002, ${\psi }_{stand}=-0.58$). Kindergarten closeness was not associated with linear change in social competence from kindergarten to Grade 6. Finally, instantaneous linear slopes of closeness were associated with linear slopes of social competence, such that a less rapid decrease in closeness in kindergarten was associated with less decrease in social competence (${\psi }_{slop{e}_{clo},slop{e}_{soc}}=0.03$, p = .026, ${\psi }_{stand}=0.65$).

Within-persons, there were statistically significant autoregressive paths for teacher-reported social competence (B = 0.10, p = .002, β = 0.10 to 0.11), but not closeness. There were no significant cross-lagged paths. In terms of contemporaneous associations, higher-than-expected levels of closeness were associated with higher-than-expected levels of social competence in kindergarten (${\sigma }_{{clo,soc}_{t0}}=3.22$, p = .002, ${\sigma }_{stand}=0.48$) and Grades 1 through 6 (${\sigma }_{{clo,soc}_{t1-t6}}=2.28$, p = .002, ${\sigma }_{stand}=0.42$).

Mother-Reported Social Competence (Supplemental Table 17).

At the between-person level, intercepts were associated, such that higher initial levels of closeness were associated with higher initial levels of social competence (${\psi }_{in{t}_{clo},in{t}_{soc}}=0.84$, p = .004, ${\psi }_{stand}=0.30$).

Within-persons, there was a statistically significant autoregressive relation for mother-reported social competence (B = 0.28, p = .004, β = 0.26 to 0.28), but not closeness. There was no evidence for cross-lagged associations. Higher-than-expected levels of closeness were contemporaneously associated with higher-than-expected levels of social competence in Grades 3 and 5 (${\sigma }_{{clo,soc}_{t1-t2}}=0.51$, p = .012, ${\sigma }_{stand}=0.12$).

Teacher-Reported Academic Competence (Supplemental Table 18).

At the between-person level, intercepts were associated, such that closeness was associated with academic competence in kindergarten (${\psi }_{in{t}_{clo},in{t}_{aca}}=1.21$, p = .002, ${\psi }_{stand}=0.41$). Furthermore, higher levels of academic competence in kindergarten were associated with more rapid decreases in closeness over time (${\psi }_{in{t}_{aca},slop{e}_{clo}}=-0.24$, p = .002, ${\psi }_{stand}=-0.43$). The association between kindergarten closeness and change in academic competence was not significant. Linear slopes of closeness and academic competence were also not significantly associated.

Within-persons, there was a statistically significant autoregressive relation for teacher-reported academic competence (B = 0.16, p = .002, β = 0.16 to 0.20), but not closeness. There was no evidence for cross-lagged associations. Closeness and academic competence were contemporaneously associated at the within-person level in kindergarten (${\sigma }_{{clo,aca}_{t0}}=1.37$, p = .002, ${\sigma }_{stand}=0.27$) and Grades 1 through 5 (${\sigma }_{{clo,aca}_{t1-t5}}=0.79$, p = .002, ${\sigma }_{stand}=0.23$).

Directly Assessed Academic Competence (Supplemental Table 19).

At the between-person level, closeness intercepts were not significantly associated with academic competence intercepts.

Within-persons, there were statistically significant autoregressive paths for directly assessed academic competence (B = 0.18, p = .007, β = 0.19 to 0.26), but not closeness. Furthermore, higher-than-expected closeness was associated with higher-than-expected performance on the Woodcock-Johnson two years later (B = 0.84, p = .023, β = 0.12 to 0.13). Additionally, higher-than-expected performance on the Woodcock-Johnson was associated with closer-than-expected relationships with teachers two years later (B = 0.01, p = .028, β = 0.10 to 0.13). Finally, in terms of contemporaneous associations, higher-than-expected levels of closeness were associated with higher-than-expected Woodcock-Johnson scores in Grade 1 (${\sigma }_{{clo,aca}_{t0}}=0.93$, p = .014, ${\sigma }_{stand}=0.16$) and Grades 3 and 5 (${\sigma }_{{clo,aca}_{t1-t2}}=0.60$, p = .007, ${\sigma }_{stand}=0.16$).

Teacher-Reported Externalizing Symptoms (Supplemental Table 20).

At the between-person level, intercepts were associated, such that higher levels of closeness were associated with lower levels of externalizing symptoms in kindergarten (${\psi }_{in{t}_{clo},in{t}_{ext}}=-0.09$, p = .009, ${\psi }_{stand}=-0.21$). Furthermore, externalizing intercepts were associated with the instantaneous closeness slope (${\psi }_{in{t}_{ext},slop{e}_{clo}}=0.01$, p = .047, ${\psi }_{stand}=0.24$).

Within-persons, there were statistically significant autoregressive relations for both closeness (B = 0.06, p = .033, β = 0.06) and teacher-reported externalizing symptoms (B = 0.19, p = .005, β = 0.19). Higher-than-expected levels of externalizing symptoms were associated with lower-than-expected levels of closeness in the subsequent grade (B = −0.05, p = 0.29, β = −0.07). The cross-lagged pathway from closeness to externalizing symptoms was not statistically significant. Finally, higher-than-expected levels of closeness were associated with lower-than-expected levels of externalizing symptoms at contemporaneous time points in kindergarten (${\sigma }_{{clo,ext}_{t0}}=-0.12$, p = .047, ${\sigma }_{stand}=-0.14$) and Grades 1 through 6 (${\sigma }_{{clo,ext}_{t1-t6}}=-0.15$, p = .005, ${\sigma }_{stand}=-0.20$).

Mother-Reported Externalizing Symptoms (Supplemental Table 21).

At the between-person level, intercepts of closeness and externalizing symptoms were not associated.

Within-persons, neither closeness nor mother-reported externalizing symptoms showed evidence of statistically significant autoregressive relations. Furthermore, there was no evidence of cross-lagged or contemporaneous associations between the constructs.

Teacher-Reported Internalizing Symptoms (Supplemental Table 22).

At the between person level, intercepts were associated, such that higher levels of closeness were associated with lower levels of internalizing symptoms in kindergarten (${\psi }_{in{t}_{clo},in{t}_{int}}=-0.11$, p = .002, ${\psi }_{stand}=-0.46$). Additionally, the intercepts of internalizing symptoms were associated with the instantaneous linear slope of closeness, such that higher levels of kindergarten internalizing symptoms were associated with a smaller decline in closeness in kindergarten (${\psi }_{in{t}_{int},slop{e}_{clo}}=0.02$, p = .002, ${\psi }_{stand}=0.52$).

Within-persons, there were statistically significant autoregressive relations for both closeness (B = 0.07, p = .011, β = 0.07) and teacher-reported internalizing symptoms (B = 0.09, p = .007, β = 0.07 to 0.09). There were no significant cross-lagged associations. Higher-than-expected levels of closeness were associated with lower-than-expected levels of internalizing symptoms contemporaneously in kindergarten (${\sigma }_{{clo,int}_{t0}}=-0.22$, p = .002, ${\sigma }_{stand}=-0.26$) and Grades 1 through 6 (${\sigma }_{{clo,int}_{t1-t6}}=-0.15$, p = .002, ${\sigma }_{stand}=-0.16$).

Mother-Reported Internalizing Symptoms (Supplemental Table 23).

At the between person level, there was no association between closeness intercepts and internalizing symptoms intercepts, nor was there an association between closeness intercepts and linear change in internalizing symptoms.

At the within-person level, neither closeness nor mother-reported internalizing symptoms showed evidence of autoregressive associations. Furthermore, there were no cross-lagged or contemporaneous associations between the constructs.

Discussion

The aim of the present study was to build upon prior research reviewed earlier demonstrating that teacher–student relationships are associated with children’s social competence, academic performance, and symptoms of psychopathology by examining the extent to which within-person, cross-lagged associations are present from kindergarten through sixth grade. Although the first a priori hypothesis predicted reciprocal, cross-lagged associations between teacher–student conflict and each developmental competency of interest, bidirectional associations were observed only for teacher-reported externalizing symptoms. There were, however, unidirectional paths from higher teacher–student conflict to lower Woodcock-Johnson scores and higher mother-reported externalizing symptoms, as well as a unidirectional path from higher mother-reported social competence to lower subsequent teacher–student conflict. In contrast, Hypothesis 2 predicted null or small paths from teacher–student closeness to adjustment and from each adjustment indicator to subsequent closeness. This was generally the case, with higher teacher–student closeness only predicting higher Woodcock-Johnson performance (though it was an association of comparable magnitude to that of conflict). There were also relatively small associations between lower teacher-reported externalizing problems and lower Woodcock-Johnson performance in one grade and lower subsequent teacher–student closeness in the next. Contemporaneous, within-person associations were also explored. Overall, concurrent associations between teacher–student relationship quality and developmentally relevant indicators of child adjustment were observed (except for teacher–student closeness and internalizing symptoms), such that in years when children tended to have better relationships with teachers than would be expected based on their growth trajectories, they also tended to have better adjustment. In general, the magnitudes of the relations were larger for conflict than for closeness and for models using teacher-reported outcomes than for models that relied on mother-reported outcomes or Woodcock-Johnson performance.

Within-Person Fluctuations in Teacher–Student Relationship Quality Are Associated with Children’s Subsequent Academic Competence and Externalizing Symptoms

Results from the present analyses suggest that, both concurrently and prospectively, teacher–student conflict is a risk factor for lower-than-expected academic performance and teacher–student closeness plays a promotive role in academic progress. Interestingly, these associations were observed in the models predicting children’s performance on the Woodcock-Johnson, but not for models that used teacher-reported academic competence as the outcome of interest. This discrepancy could be due to several factors—however, it is likely indicative of differences in the underlying constructs being studied as well as in the reliability and validity of the two measures of academic competence. That is, the Woodcock-Johnson may be a better estimate of true academic achievement, whereas teacher-reported academic competence included both items regarding how a child performs relative to their peers in the domains of reading and mathematics as well as items regarding a child’s academic engagement, classroom behaviors, and social and cognitive skills necessary for academic success. The teacher ratings thus may represent a broader construct than pure academic achievement and may be biased according to teachers’ beliefs about children. The teacher ratings also relied only a small number of items rated on a 5-point scale, whereas the Woodcock-Johnson included multiple subtests with many validated items. Thus, the effects of teacher–student relationships may be specific to true academic achievement rather than classroom-based academic engagement and performance.

Across models relying on teacher-reported and mother-reported data, evidence emerged for the predictive power of teacher–student conflict for subsequent levels of externalizing symptoms. These findings indicate that when children have higher levels of conflict with their teachers than is typical, they tend to demonstrate higher levels of behavior problems both in the same year and in the next year at school and at home. Given that child externalizing symptoms are a risk factor for a variety of adverse outcomes—including involvement with the criminal justice system (e.g., Cauffman et al., 2005), lower academic achievement and educational attainment (e.g., Masten et al., 2005), and adult mental illness and substance use problems (Loth et al., 2015; Meque et al., 2019)—it is important for researchers to identify promising intervention points for the reduction of child behavior problems. The present findings suggest that reducing teacher–student conflict could be a means by which children’s trajectories of externalizing symptoms may be altered.

Some experimental studies have provided evidence in support of the positive effects of reducing teacher–student conflict, including evaluations of the efficacy of Teacher-Child Interaction Training—Universal (TCIT-U). E. M. Davis and colleagues (2023) recently demonstrated that teachers who were randomly assigned to receive TCIT-U engaged in more positive and consistent interactions and fewer negative interactions with children. Preschool children whose teachers received TCIT-U also showed short-term improvements in behavior problems (E. M. Davis et al., 2023). The effectiveness of TCIT-U with older children remains unknown but could be explored as a model for improving teacher–student relationships and children’s behavior in the elementary setting. Importantly, the present findings also highlight important between-person associations, suggesting that targeted approaches (i.e., focusing on improving relationships between a teacher and a specific child with behavior problems) rather than universal approaches (i.e., providing general trainings to teachers on how to form high-quality relationships with children) may provide the greatest returns.

Within-Person Fluctuations in Children’s Social Competence, Academic Competence, and Externalizing Symptoms Are Associated with Subsequent Teacher–Student Relationship Quality

The present results also suggest that children who are more socially and academically competent and behaviorally regulated are likely to form more positive relationships with their teachers. It seems likely that children who possess better social skills with peers may also be more adept at interacting with adults, thus making it easier for them to form positive relationships with teachers. Furthermore, those children who are better adjusted academically may experience less stress and anxiety in the classroom and be more motivated to engage with teachers, whereas children who are struggling academically may be less motivated to build a close relationship with a teacher. Teachers, too, likely prefer to interact with the children in their classrooms who possess better social and academic skills. These findings suggest that teachers may require additional support to build relationships with children who possess lower social skills or are struggling academically. Importantly, doing so may allow practitioners to capitalize on the positive influences of a high-quality relationship with a teacher identified in the present analyses.

The finding that higher levels of teacher-reported (but not mother-reported) externalizing problems in one grade puts a child at risk for greater conflict and lower closeness with a teacher in the next grade is particularly interesting. Given that most children in the present sample attended a consistent elementary school from one year to the next, this finding may reflect a process by which children gain a reputation of being a “problem child” among teachers within a particular school. This finding suggests that providing opportunities for children with behavior problems early in elementary school to strengthen relationships and learn to regulate themselves is particularly important. Notably, there were also large between-person associations between initial levels of behavior problems and initial levels of conflict, initial levels of closeness, and trajectories of closeness. Students demonstrating high externalizing problems overall, then, are at risk for a positive feedback loop, wherein they are likely to have high conflict and low closeness with teachers in the early grades, which puts them at risk for higher behavior problems concurrently and in future, and so on. Those children who demonstrate greater behavior problems than most same-age peers should thus be targeted for early intervention approaches to disrupt this developmental cascade and promote positive adjustment in the classroom and at home.

Within-Person Fluctuations in Teacher–Student Relationship Quality and Children’s Adjustment Are Contemporaneously Associated

Overall, the results from the present study provide support for concurrent associations between teacher–student relationship quality and children’s adjustment at home and at school. Results suggest that when children have a better-than-expected relationship with a classroom teacher, they also tend to have stronger social and academic competence and fewer behavior problems than would be expected of the child given their predicted trajectories. Contemporaneous associations were observed across teacher-reported outcomes, mother-reported outcomes, and Woodcock-Johnson scores, suggesting that they are likely not completely attributable to methods-related biases (i.e., having the same reporter complete questionnaires for multiple constructs of interest). However, standardized estimates were generally larger for teacher-reported outcomes, suggesting that informant and context may have an influence on effect size. Future studies that document children’s social, academic, behavioral, and emotional adjustment in the classroom context without relying on teacher reports can help to clarify the extent to which the relatively larger associations observed in this study for teacher-reported outcomes are due to the same reporter completing multiple measures versus the common context between teacher–student relationships and classroom adjustment. For example, researchers could rely on children’s self-reports, peer nominations, observer ratings, or other teachers’ ratings of classroom adjustment to address this research question.

Interestingly, teacher–student relationship quality and children’s adjustment were more consistently related contemporaneously than longitudinally. Specifically, contemporaneous, but not cross-lagged, associations were observed between teacher–student conflict and teacher-reported social competence, academic competence, and internalizing symptoms. Similarly, teacher–student closeness and teacher-reported social competence, academic competence, and internalizing symptoms, as well as mother-reported social competence, were associated within-persons on a contemporaneous but not a cross-lagged timescale. Part of the mixed findings with regard to these cross-lagged relations may be explained by the non-trivial, year-long lag between observation points. In the case of this secondary data analysis, study design, rather than theoretical motivations, dictated the analytic approach. The true causal time scale of the influence of teacher–student relationships may in fact occur on a smaller time scale, making it particularly unlikely that such effects could be detected in the present model—especially given that cumulative effects of within-person experiences are not represented. In future studies, it could be useful to examine developmental trajectories at a finer timescale (e.g., over the course of a single school year). For example, researchers could assess teacher–student relationship quality and social competence monthly rather than annually (as was done in the present study). Documenting fluctuations in a child’s relationship with a single teacher (rather than comparing the influence of current teachers to past ones) and understanding the extent to which these more fine-grained changes are associated with changes in adjustment can help researchers identify the time scale at which teacher–student relationships are having an influence on children (if at all), as well as answer questions regarding how long effects last.

Limitations

Although the present analyses had many strengths, including use of a large, longitudinal dataset and advanced modeling strategies that allowed for the disaggregation of within- and between-person associations both concurrently and over time, several limitations must be considered when interpreting findings. Importantly, although the SECCYD sample was drawn from and has sociodemographic characteristics comparable to the 1991 U.S. birth cohort, the original inclusion criteria for the study contributed to some important groups of individuals being underrepresented or excluded from the study (e.g., individuals with medical or developmental complexities; individuals living in some high-poverty areas). Furthermore, attrition and the changing demographics in the U.S. since the study began mean that the analysis sample does not reflect the makeup of children enrolled in U.S. schools today. The present analyses should be replicated with younger, older, and more diverse cohorts to clarify the consistency of findings.

Next, although the LCM-SR models utilized in the present study likely provide a better approximation of the true developmental processes underlying the constructs of interest than traditional CLPM, they are far from perfect. LCM-SR models come with a set of complex assumptions (Berry & Willoughby, 2017). Furthermore, although the within-person associations reported in the present analyses adjust for measured and unmeasured time-invariant confounds, these models do not rule out time-varying confounds. For example, other researchers have demonstrated that changes in observed classroom organizational quality from one year to the next are also associated with changes in academic achievement and children’s behavioral problems (Watts et al., 2021). Including all possible covariates is an unreasonable goal for researchers, but testing key potential explanatory variables that vary over time and could theoretically be associated with both teacher–student relationship quality and children’s adjustment (e.g., classroom quality, family stressors, parent-child relationship quality) is an important next step for testing the robustness of the present results.

Future Research Directions

The present findings provide valuable insights into new areas of research that may advance the understanding of the role that teachers play in children’s development. First, the present findings revealed associations at the between-person level between teacher–student conflict and closeness and teacher’s ratings of children’s adjustment, as well as mother-reported social competence and externalizing symptoms. This finding suggests that developmental processes prior to kindergarten are contributing to an association between teacher–student relationship quality and children’s adjustment. Although beyond the scope of this analysis, ample data exist prior to kindergarten regarding children’s development and interpersonal experiences in the SECCYD. Analyses in the domains of interest prior to kindergarten can help to identify these developmental mechanisms and provide valuable information for intervention efforts. It may be useful to employ a similar modeling strategy in the earlier years of the SECCYD data to estimate the within-person associations between relationships with earlier caregivers—including parents, childcare providers, and preschool teachers—and early childhood adjustment. Such results could help explain the associations between kindergarten teacher–student relationship quality and adjustment observed in the present study.

Second, the present analyses revealed relatively fewer significant associations between teacher–student relationship quality and internalizing symptoms relative to the other domains of interest. Given the large sample size of the present analysis, it is unlikely that this result is due to a lack of statistical power; rather, it may be the case that teacher–student relationships are not the most promising point of intervention for reducing symptoms of anxiety and depression in young children, nor are they a strong indicator of which children are struggling to regulate their emotions. Given that schools are an important developmental context and the top source of mental health care for children in the U.S. (National Association of School Psychologists, 2021), it is essential to identify the mechanisms by which school systems can maximally prevent and reduce internalizing symptoms in elementary aged children. Researchers and practitioners may wish to invest in domains in which there is the greatest promise for effective prevention and intervention, such as school-based prevention and early intervention programs specifically designed to reduce internalizing symptoms (Feiss et al., 2019; Herman et al., 2011; Werner-Seidler et al., 2017, 2021), bullying and peer victimization reduction programs (Guzman-Holst et al., 2022), and integrated mental health treatment systems (Hoover & Bostic, 2021). Importantly, however, the present findings are limited in their reliance on parent- and teacher-reported internalizing symptoms. Given that internalizing symptoms are less readily visible to external observers than the other domains explored in the present analysis, future researchers should explore the extent to which teacher–student relationships are associated with self-reported internalizing symptoms to confirm the present findings.

Third, though the present study provided evidence that teacher–student relationship quality is, on average, associated with academic competence and externalizing symptoms, it remains unclear for whom teacher–student relationships have the greatest (or least) influence. Literature in the field of differential susceptibility has accumulated to suggest that some individuals are more susceptible to environmental influences than others (Belsky & Pluess, 2009). Other scholars have argued that teacher–student relationships have varying influence on children according to their level of risk for maladjustment (Sabol & Pianta, 2012). In light of these theoretical perspectives, important work remains to be done to clarify for whom teacher–student relationships are particularly consequential. That is, the strength of the within-person relations between teacher–student relationship quality and key developmental outcomes may vary from one child to the next. Although beyond the scope of the present study, analytic approaches that identify the characteristics of children who are more likely to respond—for better and for worse—to the quality of their relational environment at school are essential for practitioners who wish to identify promising points of interventions.

Finally, the present analysis focused on one domain of teacher–student relationships and one outcome of interest at a time to avoid overly complex models that may be more prone to false positive and negative findings. However, the reality of developmental processes is that they are dynamic and multifaceted. For example, teacher–student conflict and closeness are separate but related dimensions that do not exist in isolation from one another. Likewise, children’s social, academic, behavioral, and emotional adjustment develop together and influence one another. Therefore, the models used in the present analysis cannot fully capture the complex nature of a developing child. Future studies may be able to implement more complex models that can capture multiple relationship domains (e.g., with teachers as well as with peers, parents, siblings, and other important figures in children’s lives) along with multiple key developmental competencies at once. Such models may come closer to approximating the true developmental processes that explain individual differences between children and over time.

Public significance statement:

This study supports the conclusion that the relationships children form with teachers in the elementary years have significance for their current and future academic and behavioral development, and vice versa. Improving teacher–student relationships through school-based interventions may be a useful avenue for improving children’s adjustment.