Early Maternal Sensitivity and Teacher-Student Relationship Quality Across Grade School: Enduring or Transient Associations?

Abstract

Although teacher-student relationships are assumed to in part reflect early caregiving quality, their social provisions also undergo notable normative change over the course of primary school, shifting from a secure base for social exploration to an instrumental relationship centered on achieving academic goals. This report leveraged prospective, longitudinal data from the Study of Early Child Care and Youth Development (N = 1306, 52% male, 77% White/non-Hispanic) to investigate whether the association between early caregiving and subsequent teacher-student relationship quality remains stable or diminishes in magnitude over time. Associations between early maternal sensitivity and teacher-student closeness faded from Kindergarten to Grade 6. In contrast, associations between early caregiving and teacher-student conflict endured and were partially accounted for by child externalizing problems.

Children between the ages of 6 and 12 years spend the largest portion of their waking time in school (Hofferth & Sandberg, 2001), and the experiences to which they are exposed there—academic and interpersonal—appear to have lasting implications for their cognitive and social development (Lei, Cui, & Chiu, 2016; Roorda, Jak, Zee, Oort, & Koomen, 2017). Teacher-student relationships in particular play an important role in both academic and nonacademic development, for better and for worse. Children who develop relationships with teachers that are more supportive and emotionally close evince higher levels of academic, behavioral, and social competence (De Laet et al., 2014; Pianta & Steinberg, 1992; Pianta, Steinberg, & Rollins, 1995). Teacher-student relationships marked by high levels of conflict and negativity, in contrast, foreshadow a variety of negative outcomes for children, including increased risk for behavior problems (Hamre & Pianta, 2001; Pianta et al., 1995; Pianta & Steinberg, 1992; Skalická, Belsky, Stensen, & Wichstrøm, 2015), learning problems and academic underachievement (Hamre & Pianta, 2001; Pianta & Steinberg, 1992; Portilla, Ballard, Adler, Boyce, & Obradović, 2014; Spilt, Hughes, Wu, & Kwok, 2012), and below average social skills (Hughes & Im, 2016; Pianta & Steinberg, 1992).

Although these studies suggest that the quality of teacher-student relationships is predictive of children’s adjustment, it is not entirely clear whether these associations derive from the unique influence of the teacher-student relationship or instead are confounded by characteristics of a child, including dispositional factors and early interpersonal history. Indeed, it is likely that such factors contribute both to the quality of children’s interactions with teachers (e.g., Howes & Matheson, 1992) as well as their developmental outcomes (e.g., Fraley, Roisman, & Haltigan, 2013), thereby potentially confounding associations observed between teacher-student relationship quality and child outcomes at school.

Insights from Attachment Theory

The extent to which correlations observed between teacher-student relationships and child outcomes are a result of selection versus bona fide teacher-student socialization effects can be evaluated in part by examining the predictive significance of the qualities and experiences that children bring with them into interactions with their teachers. One frequent area of investigation of such factors that predate the teacher-student relationship is focused on the role of the quality of the early caregiving environment, driven by hypotheses derived from attachment theory (Bowlby, 1988). From an attachment perspective, the characteristics of a child’s developmental history—particularly the quality of early relationships with primary caregivers—are assumed to make important contributions to fostering a productive or problematic teacher-student relationship. The relational schemas that children abstract from their early caregiving experiences are in turn brought to bear on new caregiving relationships and, thus, are expected to contribute to the way that children interact with and respond to non-parental caregivers, such as teachers (Bowlby, 1988).

Much like parents, teachers in the early grades are seen by children as caregivers who serve as a secure base not only for academic but also social, emotional, and behavioral exploration and support (Davis, 2003). For this reason, relationships between teachers and the children in their classrooms are often assessed along dimensions that parallel variation in early caregiving, including closeness and conflict (Pianta, Hamre, & Stuhlman, 2003). Importantly, however, the evidence that early attachment security predicts subsequent teacher-student relationship quality is mixed, with some investigators observing an association (Howes & Matheson, 1992; Veríssimo et al., 2017), but others finding either conditional or no associations between these variables (Buyse, Verschueren, & Doumen, 2011; DeMulder, Denham, Schmidt, & Mitchell, 2000; Howes & Hamilton, 1992; Jerome, Hamre, & Pianta, 2009; O’Connor, 2010; O’Connor & McCartney, 2006; Rydell, Bohlin, & Thorell, 2005).

It may be that previous studies have failed to observe robust associations between early caregiving and the quality of teacher-student relationships because of the shifting social provisions of teacher-student interactions over time. In the early grades, students use teachers as a secure base from which to explore social relationships in the classroom and to buffer occasional rejection. As children approach adolescence, however, teachers normatively transition to providing instrumental help to their students and supporting them in achieving primarily academic, rather than social, goals (Bokhorst, Sumter, & Westenberg, 2010; Furman & Buhrmester, 1992). Given these changes in the function of teacher-student relationships over the course of primary school, it is important to consider how early caregiving experiences with primary caregivers predict the quality of teacher-student relationships as children progress through their early years of education. That is, selection into teacher-student relationships might follow a dynamic developmental course, with the predictive significance of early caregiving looming largest in the early grades but waning over time for some aspects of teacher-student relationships.

The Enduring Versus Revisionist Effects Framework

The present study expands upon previous work by examining whether associations between the early caregiving environment and subsequent teacher-student relationship quality endure over the course of primary school or fade as children progress through the elementary grades. The study of early caregiving experience has long been framed by the generative tension between researchers who hold that the effects of early experience endure versus those who argue that these effects are transient. “Enduring Effects” perspectives emerged in part from attachment theory and suggest that early experiences with primary caregivers are internalized in a relatively permanent manner and continue to anchor adjustment, potentially across the lifespan (e.g., Sroufe, Egeland, & Kreutzer, 1990). “Revisionist” perspectives on human development, in contrast, assume that such early experiences may impact concurrent development, but any lasting associations between early experience and subsequent adaptation can be accounted for by stability in the outcome of interest (e.g., Kagan, 1980, 1996; Lewis, 1997). Importantly, the predictions of these two contrasting theoretical perspectives can be formally modeled, and their fit to data can be evaluated and compared (Fraley et al., 2013).

The present report is the most recent in a programmatic set of studies focused on the enduring versus transient predictive significance of early maternal caregiving (e.g., Fraley et al., 2013; Haltigan, Roisman, & Fraley, 2013; Raby, Roisman, Fraley, & Simpson, 2015). More specifically, each of these prior studies has leveraged prospective, longitudinal data on the observed quality of maternal sensitivity assessed during the first three years of life, rather than attachment security, as a more direct measure of the quality of the early caregiving environment. Attachment security is not highly stable in the early years (Meins, Bureau, & Fernyhough, 2018). Maternal sensitivity, in contrast, is not only relatively stable over the early life course (e.g., Fraley et al., 2013), but has also been demonstrated in this prior work to show enduring associations with academic and social skills in the normative-risk Study of Early Child Care and Youth Development (SECCYD) sample through age 15 years (Fraley et al., 2013) and in the higher-risk Minnesota Longitudinal Study of Risk and Adaptation sample through age 32 years (Raby et al., 2015). Haltigan and colleagues (2013) likewise demonstrated in the SECCYD that maternal sensitivity showed enduring associations with teacher-reported symptoms of externalizing and internalizing problems for children through grade school, though these associations notably faded over time for mother-reported symptoms of psychopathology, particularly in the case of internalizing problems (see Roisman & Fraley, 2012).

Although previous studies in this line of work have demonstrated that the early caregiving environment has lasting predictive significance for child adjustment (e.g., Fraley et al., 2013; Raby et al., 2015), all the domains studied to date have arguably been at the individual level of analysis, focusing on adjustment (e.g., social skills, symptoms of psychopathology) or achievement (e.g., ratings of academic skills). Given that the core predictions of attachment theory suggest that early caregiving should have an enduring impact on relationships via internalized working models (Bowlby, 1988), it is important to assess whether early relationships have a lasting impact on subsequent relational domains.

Teacher-student relationships are an ideal domain for this next step of investigation because early in development they represent caregiving relationships with attachment-related functions, whereas later in grade school, they normatively transition to non-caregiving interpersonal contexts focused on the provision of instrumental help (Furman & Buhrmester, 1992). This dynamic feature of teacher-students relationships allows for a test not only of whether early care is associated with any relational outcomes later in childhood, but also of whether early maternal sensitivity has a relatively greater association with other caregiver-like relationships (i.e., teacher-student relationships in the early grades) compared to other types of social relationships (i.e., teacher-student relationships later in primary school and beyond that are focused on academic goals). Furthermore, teacher-student relationships are somewhat unique in the context of the longitudinal analysis of human development because the identity of a given child’s teacher generally changes from year to year. That is, the same child develops a different relationship with his or her teacher in each grade, allowing for a test of whether a child’s early caregiving history is consistently contributing to the quality of teacher-student relationships over time, largely without the potentially confounding influence of stable teacher effects.

Because the quality of teacher-student relationships is moderately stable from year to year, it is important to compare the predictions of the enduring and transient effects models and determine whether any associations observed between maternal sensitivity and subsequent relationship quality over time are attributable to the continued influence of the early caregiving environment, or are better accounted for by stability in teacher-student relationships over time. For instance, if transient associations for early caregiving are observed for teacher-student relationship quality, it would suggest that the shifting role of teachers over the course of grade school may reduce a child’s application of earlier working models of caregiving over time, at least in this interpersonal context. If, on the other hand, enduring associations are observed, it would be consistent with the suggestion that early caregiving has a notable influence on the relational patterns that children use in interacting with a range of adult figures in their life, even as the normative function of the teacher-child relationship shifts.

Addressing these questions will allow researchers who study teacher-student relationships to better understand the extent to which: (a) those relationships are driven by the early interpersonal experiences a child brings to the teacher-student relationship, and (b) evidence of this form of selection into higher versus lower quality teacher-students relationships follows a dynamic developmental course. Importantly, our models allow us to examine not only whether associations between the quality of early maternal care and teacher-student relationship quality stabilize or fade over time, but whether such associations are robust to: (a) demographic covariates and (b) transactional processes that help sustain continuity in the quality of teacher-student relationships across grade school (for additional details, see Methods).

The Mediating Role of Externalizing Behaviors

Additionally, previous work on the enduring versus transient effects of early caregiving has not yet attempted to identify the specific mechanisms of enduring effects, when observed. For example, although it is clear that early maternal sensitivity has enduring predictive significance for academic achievement (Fraley et al., 2013; Raby et al., 2015), it is not clear why. One of the key innovations of the present research was to explicitly focus on specific factors that could serve as mediators of the enduring association between early experience and student-teacher relationships. In addition to introducing this new extension to models examining enduring versus transient effects of early experiences, we provide an initial example focused on whether the enduring predictive significance of early maternal sensitivity can be accounted for by child behavior problems.

We focused on externalizing behaviors as a potential mediator in the present analyses for two primary reasons. First, according to attachment theory, insensitive caregiving early in life can contribute to the development of negative relational schemas that then influence future social interactions (Bowlby, 1988). Children who develop these negative schemas may engage in more frequent negative interactions that lead to spillover of externalizing behavior problems into a variety of developmentally salient social contexts, including the classroom. These problems, in turn, increase the likelihood of ineffective or hostile relationships with teachers. Indeed, previous work has documented that externalizing problems are correlated with both teacher-student conflict and teacher-student closeness in childhood (Fowler, Banks, Anhalt, Der, & Kalis, 2008; Murray & Murray, 2004; Silver, Measelle, Armstrong, & Essex, 2005). Furthermore, maternal sensitivity has been longitudinally associated with externalizing problems in early (Bakermans-Kranenburg & Van IJzendoorn, 2006) and middle childhood (van Aken, Junger, Verhoeven, van Aken, & Deković, 2007), and an experimental manipulation intended to increase mothers’ awareness and use of sensitive caregiving resulted in reduced externalizing problems (Van Zeijl et al., 2006). Overall, children who experience more insensitive maternal care appear to have greater behavior dysregulation, thus potentially limiting their ability to connect with, and eliciting negative reactivity from, their teachers. If enduring effects of early caregiving on externalizing problems in turn partially mediate enduring effects of early caregiving on teacher-student relationship quality, this pattern of results would contribute additional support for these models of human development.

As noted earlier, previous work in the SECCYD sample has already identified enduring associations between early maternal sensitivity and behavior problems through mid-adolescence (Haltigan et al., 2013). Because sensitivity has been shown to have enduring significance for externalizing problems, it is alternatively possible that any observed enduring effects of early caregiving for teacher-student conflict might be empirically redundant with those previously documented associations. That is, teacher reports of externalizing problems and teacher-student conflict may in fact represent essentially identical outcomes. Testing whether any enduring effects of maternal sensitivity for teacher-student conflict in particular can be fully accounted for by child externalizing problems would help determine whether any such effects on teacher-student relationships are simply redundant with those already published. Of course, it is also possible that enduring effects of maternal sensitivity on child behavior problems do not account, even in part, for any enduring effects of early caregiving on teacher-student relationship qualities, in which case there would be value in exploring other potential mediators in future research.

The Present Study

The present study uses the large and well-characterized SECCYD sample to answer two primary research questions. First, we studied whether associations between maternal sensitivity (as assessed over the first 3 years of life, thus preceding the onset of formal schooling) and teacher-student closeness and conflict in Kindergarten through Grade 6 are better accounted for by an Enduring Effects or Revisionist model of development. Notably, teacher-student closeness is less stable than conflict (Jerome et al., 2009; Spilt et al., 2012). This difference could reflect a more prominent contribution of a child’s developmental history for conflict than for closeness, given that children typically are placed with different classroom teachers each year. Furthermore, given the shifting role of teachers from secure bases of social exploration to instrumental supporters of academic progress over the course of primary school, we hypothesized that we would observe enduring effects of early maternal sensitivity for conflict, but not necessarily for closeness.

Second, our study builds upon the methods developed by Fraley and colleagues (2013) by examining, where applicable, a directly assessed mechanism that might account for enduring associations between early sensitivity and later teacher-student relationship quality—child problem behaviors. Doing so allows us to explore the possibility that enduring effects, if observed, might be mediated by enduring effects of behaviors that are ill-suited for the classroom as a result of immersion in an insensitive early caregiving environment. In particular, we examined the role that behavior problems might play in mediating any enduring associations observed between early maternal sensitivity and subsequent teacher-student relationship quality.

Method

Participants

In 1991, mothers were recruited from 24 hospitals in the surrounding areas of 10 research sites (Charlottesville, VA; Durham, NH; Fayetteville, AR; Irvine, CA; Lawrence, KS; Madison, WI; Morganton, NC; Philadelphia, PA; Seattle, WA; State College, PA) around the United States for the SECCYD. Participating families met the following inclusion criteria: the mother was at least 18 years old; the mother spoke English; the family intended to stay in the area for at least 1 year; the infant was born without serious medical complications; the mother was not engaging in substance abuse; the mother was not ill; the mother was not intending to place her child for adoption; the mother did not live more than 1 hour from the lab site; the family was not enrolled in another study; and the mother did not live in an area deemed unsafe for home visits (NICHD Early Child Care Research Network, 2005). Of the 8,996 mothers screened for participation, 5,416 met these selection criteria. A random sample of 1,364 mothers were then selected for participation in the study. Selected families had an average annual income of $37,947—slightly higher than the U.S. average at the time ($36,875). Approximately half (56%) of families were living above the poverty line, 23% were living near poverty, and 21% were living in poverty in 1991 (NICHD Early Child Care Research Network, 2005). These families participated in data collection beginning when the study child was 1 month old, and the present study used data collected from mothers and teachers through 2004, when the study children were in Grade 6. Analyses for this study were conducted with participants who had any maternal observational data available that was collected at 6, 15, 24, or 36 months (N = 1,306). In the analysis sample, 52% of children were male and 77% were White/non-Hispanic. Descriptive statistics for and correlations among all study variables are presented in Tables 1 and 2.

Table 1.

Correlations, Means, and Standard Deviations of Early Maternal Sensitivity, Teacher-Student Closeness, Externalizing Problems, and Covariates.

	1	2	3	4	5	6	7	8	9	10	11	12	13	14
1. Early Maternal Sensitivity	—
2. Teacher-Student Closeness, K	.11	—
3. Teacher-Student Closeness, G1	.09	.31	—
4. Teacher-Student Closeness, G2	.14	.26	.37	—
5. Teacher-Student Closeness, G3	.09	.28	.33	.29	—
6. Teacher-Student Closeness, G4	.10	.18	.24	.24	.37	—
7. Teacher-Student Closeness, G5	.12	.17	.27	.26	.33	.33	—
8. Teacher-Student Closeness, G6	.08	.15	.23	.24	.25	.22	.33	—
9. Teacher-Rated Externalizing	−.27	−.12	−.11	−.11	−.12	−.08	−.00	.05	—
10. Mother-Rated Externalizing	−.24	−.09	−.08	−.08	−.08	−.03	−.02	.00	.36	—
11. Child Gender	.09	.14	.17	.24	.22	.24	.17	.21	.00	−.01	—
12. Child Ethnicity	.38	.09	.11	.11	.05	.11	.10	.05	−.17	−.07	.00	—
13. Maternal Education	.50	.10	.12	.13	.10	.11	.13	.04	−.18	−.24	.04	.22	—
14. Income-to-Needs Ratio	.41	.08	.08	.09	.11	.13	.08	.05	−.15	−.19	.04	.23	.54	—
N	1306	1006	1006	934	971	909	923	854	1012	1046	1306	1306	1306	1304
M	−0.02	10.60	10.92	10.94	11.63	11.15	11.42	11.05	50.65	50.70	0.48	0.77	14.28	3.45
SD	0.77	5.36	5.18	5.41	6.04	5.75	5.71	5.63	8.41	7.45	—	—	2.50	2.71

Note. Gender was coded as 1 = female, 0 = male. Ethnicity was coded as 1 = White/non-Hispanic, 0 = non-White. Bolded values are associated with p < .05.

Table 2.

Correlations, Means, and Standard Deviations of Early Maternal Sensitivity, Teacher-Student Conflict, Externalizing Problems, and Covariates.

	1	2	3	4	5	6	7	8	9	10	11	12	13	14
1. Early Maternal Sensitivity	—
2. Teacher-Student Conflict, K	−.15	—
3. Teacher-Student Conflict, G1	−.18	.40	—
4. Teacher-Student Conflict, G2	−.27	.42	.46	—
5. Teacher-Student Conflict, G3	−.27	.35	.47	.55	—
6. Teacher-Student Conflict, G4	−.30	.35	.42	.47	.53	—
7. Teacher-Student Conflict, G5	−.30	.27	.44	.42	.46	.50	—
8. Teacher-Student Conflict, G6	−.22	.22	.42	.35	.43	.45	.53	—
9. Teacher-Rated Externalizing	−.27	.66	.60	.62	.56	.46	.36	.29	—
10. Mother-Rated Externalizing	−.24	.24	.26	.32	.26	.26	.23	.22	.36	—
11. Child Gender	.09	−.13	−.17	−.17	−.20	−.19	−.18	−.23	.00	−.01	—
12. Child Ethnicity	.38	−.10	−.15	−.19	−.18	−.22	−.25	−.14	−.17	−.07	.00	—
13. Maternal Education	.50	−.09	−.15	−.16	−.19	−.21	−.24	−.20	−.18	−.24	.04	.22	—
14. Income-to-Needs Ratio	.41	−.07	−.12	−.16	−.18	−.21	−.23	−.18	−.15	−.19	.04	.23	.54	—
N	1306	1006	1005	934	971	909	923	854	1012	1046	1306	1306	1306	1304
M	−0.02	34.23	33.96	33.67	33.08	32.53	31.86	30.31	50.65	50.70	0.48	0.77	14.28	3.45
SD	0.77	5.34	5.04	5.17	5.16	5.12	5.36	5.74	8.41	7.45	—	—	2.50	2.71

Note. Gender was coded as 1 = female, 0 = male. Ethnicity was coded as 1 = White/non-Hispanic, 0 = non-White. Bolded values are associated with p < .05.

Measures

Early maternal sensitivity.

When the study children were 6, 15, 24, and 36 months old, mothers were filmed during a series of 15-minute interactions with their infant (NICHD Early Child Care Research Network, 1999). At the 6-month assessment, mothers played with their children first with a set of toys brought from home and then with a standard set of toys. At the 15-, 24-, and 36-month assessments, mothers were provided with three sets of age-appropriate toys and were instructed to play with their infants in a standard order. At the 6-, 15-, and 24-month assessments, videotapes were rated on 4-point scales for positive regard (i.e., quality and frequency of mother’s expressions that connote positive feelings toward the child), sensitivity to nondistress (i.e., mother’s prompt and appropriate responses to child’s behavior), and intrusiveness (i.e., degree to which the mother imposes her own will on the child). At the 36-month assessment, videotapes were rated on 7-point scales for respect for autonomy (i.e., the degree to which the mother provides a sense of control to the child), supportive presence (i.e., the degree to which the mother provides a sense of control to the child), and hostility (i.e., the frequency and intensity of negative affect directed toward the child). The intrusiveness and hostility scales were then reverse-scored so that higher scores represented more sensitivity. Intercoder reliability scores were .87, .83, .84, and .84 and internal consistency ratings were .75, .70, .79, and .78 for the 6-, 15-, 24-, and 36-month assessments, respectively (NICHD Early Child Care Research Network, 1999). These internal consistency scores are above minimum values considered acceptable (Hammill, Brown, & Bryant, 1992). Therefore, scores were summed and standardized to create a measure of maternal sensitivity at each of the four time points. These four standardized scores were subsequently averaged to create a composite measure of observed maternal sensitivity over the first 3 years of life (α = .73).

Teacher-student relationship quality.

Classroom teachers completed the Student-Teacher Relationship Scale, Short Form (STRS-SF; Pianta, 2001) each year between Kindergarten and Grade 6. Teachers self-reported their gender when the study children were in Kindergarten (97.3% female) and Grade 1 (95.7% female); data on teacher gender were not available for Grades 2–6. The majority of teachers were White/non-Hispanic (Kindergarten: 87.6%; Grade 1: 94.0%; Grade 2: 94.8%; Grade 3: 92.2%; Grade 4: 94.6%; Grade 5: 93.1%; Grade 6: 94.5%). At each year of assessment, teachers reported a wide range of years of teaching experience (Kindergarten: M = 15.45, SD = 9.06; Grade 1: M = 14.52, SD = 9.48; Grade 2: M = 12.38, SD = 10.07; Grade 3: M = 11.88, SD = 10.54; Grade 4: M = 11.34, SD = 10.26; Grade 5: M = 12.04, SD = 10.63; Grade 6: M = 11.52, SD = 10.43).

The STRS-SF consists of fifteen items on which teachers rated their agreement with a number of statements regarding their relationship with the study child on a Likert-type scale (1 = “Definitely does not apply” to 5 = “Definitely applies”). Scores on seven items assessing closeness (e.g., “If upset, this child will seek comfort from me”) were summed to create a scale of teacher-student closeness. Scores on eight items assessing relational problems (e.g., “The child remains angry or is resistant after being disciplined”) were summed to create a scale of teacher-student conflict. The STRS-SF has demonstrated high test-retest and internal reliability (Pianta, 2001) and both concurrent and predictive convergent validity (Hamre & Pianta, 2001). Cronbach’s alphas suggested adequate reliability (closeness: .84 < α < .86; conflict: .88 < α < .91).

Externalizing problems.

Each year from Kindergarten to Grade 6, teachers completed the Teacher Report Form (TRF; Achenbach, 1991b). The TRF consists of 113 items regarding students’ problem behaviors that are rated on a 3-point scale (0 = “Not True (as far as you know)” to 2 = “Very True or Often True”). To obtain a measure of externalizing problems, we used age- and sex-standardized T scores applied to the raw scores from the externalizing behaviors scale of the TRF (e.g., “Defiant, talks back to staff”, “Destroys property belonging to others”). In a parallel fashion, mothers completed the Child Behavior Checklist (CBCL; Achenbach, 1991a) when the study children were 24, 36, and 54 months; in Kindergarten; and in Grades 1, 3, 4, 5, and 6. T-scores from the externalizing behaviors scale were again used. The TRF and CBCL are widely-used measures of the psychological well-being of children, and have demonstrated good test-retest reliability and reliability between raters (Achenbach, 1991a, 1991b). To estimate externalizing problems at 54 months, we used a growth curve model in which externalizing problems at each measurement point were used to estimate each child’s intercept and slope across time. Time was scaled such that the intercept represented externalizing problems at 54 months. Those intercepts were then used in the mediational models (see below).

Control variables.

We selected four control variables that have been frequently used in other research with this data set and are known correlates of both maternal sensitivity and teacher-student relationship quality (e.g., Fraley et al., 2013; Jerome et al., 2009). At 1 month, mothers provided information on each child’s sex, race, and ethnicity, her own education level, and household income during an in-home interview. Child gender was coded in a binary fashion (1 = female, 0 = male). Because the majority of the study children were not from a minority group, race and ethnicity information were collapsed to create a dummy variable (1 = White/non-Hispanic, 0 = other). Maternal education was recoded to represent the approximate number of years of education (e.g., “High school graduate or GED” = 12, “Bachelor’s degree from college or university” = 16, etc.). Years of education ranged from 7 to 21. Household income was additionally reported at ages 6, 15, 24, and 36 months. Poverty thresholds were determined for each participant by considering the year in which income was earned, number of members in the household, and number of children living full-time in the household. Household income was then divided by the poverty threshold to calculate an income-to-needs ratio for each of the five time points. These values were then averaged to create an index for early childhood (α = .94). Income-to-needs ratios ranged from 0.14 to 18.76, with higher values indicating greater income relative to economic need.

Missing data.

Analyses were conducted for the 1,306 participants for whom observational data was available from at least one of the 6, 15, 24, or 36 month assessments. Independent samples t-tests revealed that participants included in the analysis sample did not differ from those without maternal observational data on gender (p > .050). Participants included in the analysis sample did, however, differ on race (slightly more likely to be White/non-Hispanic; Cohen’s d = 0.24, p = .046), maternal education level (slightly more likely to have more years of education; d = 0.45, p < .001), and income-to-needs ratio (much more likely to have higher income-to-needs; d = 1.05, p < .001).

Missing data in the outcome variables (i.e., teacher-student closeness and conflict) varied by assessment point (see Table 1, 2). Independent samples t-tests were conducted for each time point. Participants with data were more likely to be female in Grade 1 (d = 0.22, p = .014), Grade 3 (d = 0.21, p = .009), Grade 5 (d = 0.18, p = .018), and Grade 6 (d = 0.14, p = .029). Participants with data were more likely to be White/non-Hispanic in Kindergarten (d = 0.37, p < .001), Grade 1 (d = 0.27, p = .004), Grade 2 (d = 0.36, p < .001), and Grade 3 (d = 0.18, p = .036). Participants with data were more likely to have better-educated mothers in Kindergarten (d = 0.34, p < .001), Grade 1 (d = 0.39, p < .001), Grade 2 (d = 0.51, p < .001), Grade 3 (d = 0.34, p < .001), Grade 4 (d = 0.36, p < .001), Grade 5 (d = 0.29, p < .001), and Grade 6 (d = 0.16, p = .018). Finally, participants with data were more likely to have higher income-to-needs ratios in Kindergarten (d = .23, p = .014), Grade 1 (d = .28, p = .002), Grade 2 (d = 0.32, p < .001), Grade 3 (d = .026, p = .002), and Grade 4 (d = 0.27, p < .001). Data were treated as missing at random and analyses were conducted using full information maximum likelihood estimation, selected to minimize bias and optimize model fit (Enders & Bandalos, 2001).

Statistical Analyses

All analyses were conducted in R, version 3.4.3, using the package lavaan (Yves, 2012), with the bootstrapping option utilized for the estimates of standard errors (B = 1000). We estimated the parameters of a series of models, depicted in Figure 1, using early maternal sensitivity to predict: (a) teacher-student closeness over time and (b) teacher-student conflict over time. Because the formal models that result from the enduring effects and revisionist perspectives are nested, we were able to compare their respective fits with difference in chi-square tests (Fraley et al., 2013). Additionally, we compared two specifications of the enduring effects model: one that allowed the b paths (i.e., the paths representing the enduring effect of early experiences) to vary over time, and another that constrained them to be equal at each time point. The effect of these constraints on model fit was also assessed with difference in chi-square tests. We calculated additional estimates of model fit, including the Comparative Fit Index (CFI), Root Mean Square Error of Approximation (RMSEA), and Standardized Root Mean Squared Residual (SRMR) for each model. Values that indicate acceptable model fit for these measures are: CFI > .95, RMSEA < .06, and SRMR < .08 (Hu & Bentler, 1999).

Figure 1.

Conceptual models of the influence of early maternal sensitivity on teacher-student relationship quality through Grade 6. Models were fit separately for teacher-student closeness and conflict. (1a) The most basic model. Enduring b paths were constrained to 0 (revisionist) or were allowed to vary over time (enduring effects). (1b) Builds on the initial model by including the covariates of child sex, child race, maternal education, and income-to-needs ratio. (1c) Builds on the previous model by including second-order transactional pathways.

Following the comparison of the basic revisionist model to the basic enduring effects models, we increased the complexity of the models to include covariates (Figure 1b). All covariates were included and retained, regardless of their effect size, consistent with prior work (Fraley et al., 2013). Finally, we estimated models that included transactional pathways (Roisman, Fraley, Haltigan, Cauffman, & Booth-LaForce, 2016). These increasingly complex models represent a more conservative test of the basic model results, and served as an important robustness check given the nonexperimental nature of the research design of the SECCYD.

Mediational processes.

To attempt to explain enduring associations observed between early sensitivity and subsequent teacher-student relationship quality, we estimated the parameters of a series of models that tested the assumption that the effect of sensitivity on teacher-student relationship quality at each wave was accounted for by a mediator. Specifically, we evaluated the evidence that enduring associations (b paths) between early sensitivity and teacher-child conflict from Kindergarten through Grade 6 could be accounted for in part or completely by enduring effects of externalizing problems on teacher-student conflict.

We first examined the possibility that the effect of sensitivity on teacher-child conflict could be partially accounted for by enduring effects of externalizing problems on teacher-student conflict (Figure 2a). The partial mediation model is the most inclusive model tested and is therefore possible to directly compare to both no- and full-mediation models (see below). If partial mediation is observed, it would suggest that early maternal sensitivity elicits problems in both behavior (i.e., externalizing) and in relational contexts (i.e., teacher-student relationship quality) over time. We then examined no mediational model (Figure 2b), which assumed that the enduring effects of early sensitivity on teacher-student conflict were not accounted for by enduring effects of externalizing problems on teacher-student conflict. Finally, we examined the possibility that the enduring predictive significance of early sensitivity on teacher-student conflict could be fully accounted for by the enduring effects of externalizing problems on teacher-student conflict (Figure 2c). If externalizing problems fully mediate the relation between early sensitivity and subsequent teacher-student relationship quality, such findings would suggest that the present results may be redundant with previous findings in this sample, and that teacher reports of conflict may be equivalent to behavior problems in the classroom (i.e., Haltigan et al., 2013).

Figure 2.

Conceptual mediational models of the mediating effect of child externalizing problems. (2a) A partial mediation model, in which child externalizing problems and maternal sensitivity jointly have enduring predictive significance for teacher-student conflict over time. (2b) A no mediation model, in which child externalizing problems are assumed to not have enduring associations with teacher-student conflict. (2c) A full mediation model, in which enduring effects of child externalizing problems on teacher-student conflict are assumed to fully account for the enduring effects of maternal sensitivity previously established.

Results

We present results first for teacher-student closeness, followed by teacher-student conflict. Finally, we investigate the extent to which enduring effects observed for early maternal sensitivity on teacher-student conflict could be accounted for (i.e., was partially or completely mediated) by child behavior problems, as reported by mothers and teachers (see Figure 2). All reported path estimates are standardized, except where noted.

Teacher-Student Closeness

The basic model.

Correlations between early maternal sensitivity and teacher-student closeness are presented in Table 1. These correlations are generally of small magnitude (rs < .14), with the smallest correlation at the Grade 6 assessment. Model fit statistics and model comparisons are presented in Table 3. In the basic model (Figure 1a), the enduring effects b paths ranged from 0.04 to 0.11 when allowed to vary; constraining the paths to be equal at each time point did not significantly decrease model fit (Δχ² = 3.39, p = .640). When constrained, these paths were equal to 0.07. Fixing the b paths to equal 0 (i.e., as implied by the revisionist model) resulted in a significant decrease in fit over the enduring effects model (Δχ² = 37.81, p < .001).

Table 3.

Model Fit Estimates for the Influence of Early Maternal Sensitivity on Teacher-Student Closeness Across Childhood

Model	Model fit						Nested model comparisons
	χ2	df	p	CFI	RMSEA	SRMR	Δχ2	df	p
Basic Model
A. Revisionist	293.04	21	< .001	.70	.10	.12	A – B = 37.81	6	< .001
B. Enduring	255.23	15	< .001	.74	.11	.11	C – B = 3.39	5	.640
C. Enduring: Equality Constraint	258.62	20	< .001	.74	.10	.11	A – C = 34.42	1	< .001
Basic Model with Covariates
A. Revisionist	241.29	41	< .001	.81	.06	.07	A – B = 5.14	6	.526
B. Enduring	236.15	35	< .001	.81	.07	.07	C – B = 3.40	5	.639
C. Enduring: Equality Constraint	239.55	40	< .001	.81	.06	.07	A – C = 1.74	1	.187
Transactional Model with Covariates
A. Revisionist	107.75	36	< .001	.93	.04	.04	A – B = 4.08	6	.665
B. Enduring	103.67	30	< .001	.93	.04	.04	C – B = 2.63	5	.757
C. Enduring: Equality Constraint	106.29	35	< .001	.93	.04	.04	A – C = 1.46	1	.228

Note. CFI = comparative fit index. RMSEA = root mean-square-error of approximation. SRMR = standardized root-mean-square residual. N = 1306.

Extensions of the basic model.

After including all covariates in the model (Figure 1b), the unconstrained enduring effects model no longer showed improved fit over the simpler revisionist model (Δχ² = 5.14, p = .526) or the constrained enduring effects model (Δχ² = 3.40, p = .639). The estimates of the b paths ranged from −0.01 to 0.06 when unconstrained, and were equal to 0.02 in the constrained model. In the revisionist model, sex significantly predicted teacher-student closeness in Kindergarten (b = 0.13, p < .001), such that teachers tended to report more closeness with female students. In Grades 1 through 6, teacher-student closeness was predicted by sex (b = 0.15, p < .001), race (b = 0.04, p = .003; teachers reported more closeness with White/non-Hispanic students), and maternal education (b = 0.05, p = .003), but not income-to-needs ratio (p = .211).

Second-order autoregressive pathways were included in a final step (Figure 1c). Notably, including transactional pathways, in addition to the covariates, resulted in lower χ², RMSEA, and SRMR values and higher CFI values overall, indicating better model fits. Once again, the unconstrained enduring effects model did not fit significantly better than either the revisionist model (Δχ² = 4.08, p = .665) or the constrained enduring effects model (Δχ² = 2.63, p = .757). In the unconstrained model, b path estimates ranged from −0.01 to 0.05; in the constrained model, b paths were estimated to equal 0.02. Taken together, these results indicate that models that allow for enduring effects of early sensitivity on subsequent teacher-student closeness do not provide better fit than those that assume no such enduring effects. That is, the simpler revisionist model was more consistent with these data.

Teacher-Student Conflict

The basic model.

Correlations between early maternal sensitivity and teacher-student conflict are presented in Table 2. Correlations ranged from −.15 to −.30, and do not appear to decrease over time. Model fit statistics and model comparisons are presented in Table 4. In the basic model (Figure 1a), constraining the b paths to equal 0 resulted in a significant decrease in model fit (Δχ² = 160.58, p < .001), as did constraining the paths to be equal over time (Δχ² = 11.14, p = .049), as compared to the unconstrained enduring effects model. The values of the b paths ranged from −0.20 to −0.07 in the unconstrained model, and equaled −0.14 when constrained to be equivalent.

Table 4.

Model Fit Estimates for the Influence of Early Maternal Sensitivity on Teacher-Student Conflict Across Childhood

Model	Model fit						Nested model comparisons
	χ2	df	p	CFI	RMSEA	SRMR	Δχ2	df	p
Basic Model
A. Revisionist	601.26	21	< .001	.73	.15	.18	A – B = 160.58	6	< .001
B. Enduring	440.68	15	< .001	.80	.15	.14	C – B = 11.14	5	.049
C. Enduring: Equality Constraint	451.82	20	< .001	.80	.13	.14	A – C = 149.44	1	< .001
Basic Model with Covariates
A. Revisionist	465.62	41	< .001	.81	.09	.10	A – B = 46.76	6	< .001
B. Enduring	418.86	35	< .001	.83	.09	.09	C – B = 11.68	5	.039
C. Enduring: Equality Constraint	430.54	40	< .001	.82	.09	.09	A – C = 35.08	1	< .001
Transactional Model with Covariates
A. Revisionist	177.46	36	< .001	.94	.06	.05	A – B = 37.00	6	< .001
B. Enduring	140.48	30	< .001	.95	.05	.04	C – B = 12.99	5	.024
C. Enduring: Equality Constraint	153.47	35	< .001	.95	.05	.05	A – C = 24.02	1	< .001

Note. CFI = comparative fit index. RMSEA = root mean-square-error of approximation. SRMR = standardized root-mean-square residual. N = 1306.

Extensions of the basic model.

After including covariates in the model (Figure 1b), model fit was again impaired when the enduring effects b paths were constrained to be equal over time (Δχ² = 11.68, p = .039) or to be equal to 0, as in the revisionist account (Δχ² = 46.76, p < .001). In the unconstrained model, the enduring effects b paths ranged from −0.14 to −0.01. In the constrained model, these paths were estimated to be −0.08. Furthermore, in the unconstrained enduring effects model, sex significantly predicted teacher-child conflict in Kindergarten (b = −0.12, p < .001; teachers reported more conflict with male students). Sex (b = −0.09, p < .001), race (b = −0.06, p < .001; teachers reported less conflict with White/non-Hispanic students), income-to-needs ratio (b = −0.03, p = .008), and maternal education (b = −0.04, p = .006) predicted teacher-child conflict from Grades 1 through 6.

As with teacher-student closeness, we included transactional pathways (Figure 1c). Once again, including these second-order autoregressive pathways improved all five estimates of model fit overall, as compared to the models with only covariates. As before, constraining the pathways to be equal over time (Δχ² = 12.99, p = .024) or to equal 0, as in the revisionist account (Δχ² = 37.00, p < .001), decreased model fit. In the unconstrained model, b path estimates ranged from −0.11 to −0.06 from Kindergarten to Grade 5, while the Grade 6 estimate was a nonsignificant 0.03. When constrained to be equal over time, the enduring effects b paths were estimated to equal −0.07. In contrast to teacher-student closeness, the enduring effects model was more consistent with these data on teacher-student conflict.

Mediational processes.

Mother- and teacher-ratings of children’s externalizing behaviors were included in separate models. In addition, we also evaluated a model that included these two rating sources as dual mediators of the relation between maternal sensitivity and teacher-student conflict. We discuss only the results of the mother-reported externalizing model in detail below. However, similar results were obtained for teacher-ratings of externalizing behavior and in models that included both mother- and teacher-ratings of externalizing behavior (see Table 5).

Table 5.

Model Fit Estimates for the Mediating Effect of Child Externalizing Problems on Teacher-Student Conflict Across Childhood

Model	Model fit						Nested model comparisons
	χ2	df	p	CFI	RMSEA	SRMR	Δχ2	df	p
Maternal Report
A. Partial Mediation	155.48	40	< .001	.95	.05	.04
B. No Mediation	211.95	41	< .001	.93	.06	.05	B – A = 56.47	1	< .001
C. Full Mediation	172.13	41	< .001	.95	.05	.04	C – A = 16.65	1	< .001
Teacher Report
A. Partial Mediation	383.24	40	< .001	.90	.08	.05
B. No Mediation	677.72	41	< .001	.81	.11	.08	B – A = 294.49	1	< .001
C. Full Mediation	390.75	41	< .001	.90	.08	.06	C – A = 7.51	1	.006
Maternal and Teacher Report
A. Partial Mediation	381.56	45	< .001	.91	.08	.05
B. No Mediation	694.24	47	< .001	.82	.10	.08	B – A = 312.68	2	< .001
C. Full Mediation	387.15	46	< .001	.91	.08	.05	C – A = 5.59	1	.018

Note. CFI = comparative fit index. RMSEA = root mean-square-error of approximation. SRMR = standardized root-mean-square residual. N = 1306.

In the mother-reported externalizing model, the model that allowed the enduring effects of early sensitivity to be partially accounted for by enduring effects of child externalizing problems on teacher-student conflict (Figure 2a) demonstrated better fit than the model that assumed no mediation (Figure 2b; Δχ² = 56.47, p < .001). Furthermore, fixing the b paths to equal 0 (i.e., assuming externalizing problems could fully account for the enduring effects of teacher-student conflict; Figure 2c) resulted in a significant decrease in fit over the model that assumed externalizing problems only partially accounted for the enduring effects of early sensitivity (Figure 2a; Δχ² = 16.65, p < .001). For the partial mediation model, the indirect path for maternal sensitivity and externalizing behavior was statistically significant (unstandardized b = −0.11, p < .001; 95% CI: −.17 < b < −.05). Despite this indirect effect, maternal sensitivity continued to exhibit direct effects on teacher-student conflict across assessment waves (b = −0.06, p < .001). In summary, although it appears that externalizing behavior partially explains why children with less sensitive caregiving experience more conflict with their teachers, it does not fully explain the enduring nature of the association.

Discussion

Although theory focused on early caregiving suggests that the quality of relationships between infants and their primary caregivers influences the quality of a child’s future relationships with teachers, the evidence for such associations is mixed (e.g., Jerome et al., 2009; O’Connor & McCartney, 2006). The goals of the current study were: (1) to determine whether associations between early maternal sensitivity and subsequent teacher-student relationship quality could be better accounted for by an Enduring Effects or Revisionist model of development and (2) to evaluate the extent to which enduring associations between early maternal sensitivity and subsequent teacher-student conflict were mediated by child externalizing problems. Using data from the Study of Early Child Care and Youth Development (SECCYD), we found evidence for dynamic and stable selection processes. The association between early caregiving and teacher-student closeness was more consistent with a revisionist model, in which associations waned over the course of grade school. In contrast, teacher-student conflict appeared to be consistently shaped by early caregiving experiences through the end of elementary school. These enduring associations were at least partially sustained by externalizing problems that were associated with less sensitive maternal care early in life.

Teacher-Student Relationships: Dynamic and Stable Processes

Contrary to the predictions of attachment theory, early maternal sensitivity did not appear to have lasting predictive significance for children’s closeness with their grade school teachers. These findings, however, are readily explicable in the context of research documenting the shifting social provisions of teacher-student relationships over the course of elementary school (Bokhorst et al., 2010; Furman & Buhrmester, 1992). That is, the fading relation between early caregiving relationships and subsequent teacher-student closeness could reflect the changing role of the teacher as children progress from Kindergarten to Grade 6. As children age, attachment theory suggests that they may develop more diverse working models of caregiving relationships that they are then able to apply independently to a wider variety of contexts (Baldwin, 1992). A teacher in the early years may be more easily assimilated into a child’s working model of a caregiver because teachers serve as a secure base from which children can explore novel social and academic challenges. With time, however, children may come to separate their working models of teachers from that of caregivers as classrooms become more academically-focused and teachers provide less social and emotional support.

Given these fading associations over time, it seems clear that teacher-student closeness in grade school is not simply a reflection of the quality of experiences with caregivers in early childhood. Replication is needed before we can conclude that teacher-student closeness is entirely distinct from parent influences, but the present results suggest that there may be aspects of the teacher-student relationship that play a unique role in human development. Furthermore, the present results suggest that teacher-student closeness is particularly amenable to change from year to year, and may therefore be a good target for interventionists.

In contrast, the predictive significance of maternal sensitivity for teacher-student conflict was, in the present sample, observed to endure through Grade 6. Children are typically paired with a different classroom teacher each year of elementary school, making it particularly striking that early experiences continue to correlate with teacher-student conflict through at least the end of primary school. Given these findings, it may be that the predictive significance of teacher-student conflict for future adjustment observed in previous studies (e.g., Hughes & Im, 2016; Portilla et al., 2014; Skalická, 2015) is confounded by the enduring influence of early caregiving. In other words, children who are exposed to insensitivity early on may form lower quality relationships with teachers and may also tend to perform poorly in school and socially for that same reason. Future work should examine whether teacher-student conflict is itself a mechanism by which early insensitivity shapes maladaptive development in adolescence and beyond, or whether teacher-student conflict demonstrates predictive significance above and beyond the role of early parenting.

Why might we have observed enduring effects for conflict, but not for closeness? Evidence for enduring associations between early sensitivity and teacher-student conflict may have been identified because conflict can emerge in a variety of relationships, regardless of their primary function. Conflict appears to be universal among close relationships, emerging between parents and children, within peer dyads, among coworkers, and in couples (Braiker & Kelley, 2013; Dunn & Herrara, 1997). Even as teacher-student relationships transition from being attachment-like to having a more academic focus, conflict remains possible. Children of insensitive caregivers may develop negative working models of dyadic interactions generally, and approach teachers with greater responsiveness to conflict regardless of grade. Closeness with teachers, on the other hand, may be more likely to be fostered by supportive relationships with primary caregivers early on in grade school, when teachers normatively provide a secure base for their students in a manner that parallels the social provisions of primary caregivers (Furman & Buhrmehster, 1992). As children age, however, it may become more difficult to apply an early model of caregiving to teachers who are becoming less caregiver-like. Overall, more research is needed on the normative trends in teacher-student relationship quality over time to illuminate and confirm these findings.

Behavior Problems as a Link Between Early Sensitivity and Later Conflict

Given the enduring effects observed for teacher-student conflict, we sought to account for these associations by testing the mediational role of child externalizing problems. Externalizing behavior has been previously shown in the SECCYD to be associated with early maternal sensitivity (Haltigan et al., 2013) and is frequently associated with teacher-student conflict (Fowler et al., 2008; Murray & Murray, 2004; Silver et al., 2005). By expanding on the methods developed by Fraley and colleagues (2013), we developed three models to test the role of child externalizing problems as a mediator between early maternal sensitivity and teacher-student conflict. Our analyses revealed that a partial mediation model was most consistent with the data. These results suggest that lower levels of sensitivity early on tend to be associated with higher levels of externalizing problems in early childhood, which in turn have implications for how children interact with and respond to their classroom teachers. Children who are exposed to insensitive caregiving early in life may be more likely to interpret conflictual teacher and caregiver interactions as negative and respond in more coherent ways between caregivers at home and at school.

Given that the enduring predictive significance of early care for behavior problems and teacher-student conflict were relatively modest in magnitude, more work is needed to understand what may elicit negativity between teachers and students. However, clinicians and educators who aim to improve child experiences in school may wish to target behavioral problems given their persistent associations with relational adjustment. Children experiencing insensitive early care are apparently more likely to evince problem behaviors and subsequently engage in conflict with their teachers. As such, targeting behavior problems of children from at risk home environments could reduce teacher-student conflict over time. Researchers implementing programs designed to reduce behavior problems in the classroom may therefore wish to examine teacher-student relationship quality as a potential outcome to measure the effects of intervention.

The present mediational analyses serve as a potential model for researchers who wish to study not only whether a developmental experience (e.g., maternal sensitivity) has enduring or transient predictive significance, but additionally explain why enduring associations may be observed. Although partial mediation was most consistent with the data in the present analyses, future investigations of enduring effects in other domains may be better accounted for by full or no mediation models. By comparing models that assume the presence of a mediator to those that do not, it is possible to evaluate the degree to which an intermediate factor can account for relations between an early experience and a subsequent, and perhaps distal, series of outcomes. Furthermore, although the present analyses focused only on the role of externalizing problems, our models allow for the presence of multiple mediators at once. In the current context, we demonstrated this by including both mother- and teacher-reported externalizing problems in a single model, but in theory could have continued to add additional theoretically-relevant mediators into the model. Future researchers may wish to incorporate multiple mediators at once, comparing their relative effects, or may alternatively add mediators sequentially in an attempt to more fully account for observed enduring associations.

Limitations and Future Directions

Although the present manuscript leveraged data from a large, longitudinal study to estimate two competing models of human development, below we identify six noteworthy limitations and areas in need of further investigation. First, although the theoretical models contrasted in this paper make different causal assumptions about the influence of early caregiving, the non-experimental nature of the design of the SECCYD constrains the causal conclusions that can be drawn from both our basic and mediational analyses. Indeed, the direction of assumed causal influence could be different from that which we have considered here. As a reviewer of our manuscript suggested, we could have conducted reverse mediation analyses in order to examine alternative possibilities. However, given the current debate surrounding the utility and interpretation of reverse mediation analyses (see Lemmer & Gollwitzer, 2017; Thoemmes, 2015), we did not examine this possibility in the present longitudinal dataset. Questions about causal influence thus remain open, and experimental studies targeting maternal caregiving, child behavior problems, and relational adjustment are needed to clarify the likely complex temporal and causal relations between the variables under study.

Second, the SECCYD sample is composed primarily of normative-risk families and while roughly consistent with the demographics of the 1991 US birth cohort, was not designed to be a representative sample of that cohort. The generalizability of these results will thus remain unknown, pending replication attempts in independent samples. Replication attempts in samples that are more diverse in terms of ethnic and racial background, migration and English-language-learner status, and economic advantage will be particularly important. Third, although early maternal sensitivity was measured through observation at several assessments over a relatively short period of time, teacher-student relationship quality was assessed by a single reporter (i.e., the primary classroom teacher) only once per year from Kindergarten through Grade 6. Relying on teacher-report thus limits the validity of the measurement of the construct by excluding child or outside observer perspectives.

A fourth key area for further consideration regards the absolute fit of the models presented in this paper. Overall, our final teacher-student conflict models had good absolute fit (Table 4, transactional models). The final models for closeness, on the other hand, had only acceptable fit (Table 3, transactional models). Our partial mediation model using mother-reported externalizing problems also evinced good fit, but the models using teacher-reported or teacher- and mother-reported externalizing problems did not. This issue has been discussed in more detail by Fraley and colleagues (2013), but is likely a reflection of the limited number of variables used to predict teacher-student relationship quality and the missing mediators reflected by our transactional pathways (Roisman et al., 2016). Importantly, our analysis approach focused on comparisons of nested models using traditional chi-square difference tests rather than interpretation of absolute model fit statistics. Although other methods for comparing fit between models exist (e.g., Chen, 2007), none of the proposed methods are widely accepted enough to be considered “best practice.” In the case of our enduring (equality constraint) and revisionist models, the key difference between the two reduces to the addition (or removal) of a single parameter: The direct b path between a measure of early experience and subsequent outcomes. Thus, the significance test of the change in chi-square is identical to the significance test for the estimate of the enduring path itself.

Fifth, despite it not being a primary aim of the present study, it is clear from our results that child traits, beyond the quality of early caregiving, have an impact on teacher-student relationship quality. For example, teachers reported more closeness with female and White/non-Hispanic students as well as children coming from homes with better-educated mothers. Males, children from poorer and less-educated families, and students from minority groups were rated as having more conflict with their teachers. These consistent associations over time suggest that these along with other child traits merit exploration in future studies seeking to explain between-child differences in teacher-student relationship quality.

Sixth and finally, we importantly did not consider the influence of teacher characteristics or demographics, despite the fact that previous work has shown that these can have implications for teacher-student relationship quality (Thijs, Westhof, & Koomen, 2011; Yiu, 2013). Indeed, teachers and children who perceive themselves as being more alike tend to have closer relationships (Gehlbach et al., 2016). It is possible that the higher rates of conflict and lower rates of closeness observed for minority students in the present sample might be less pronounced when students are rated by teachers with whom they are matched on demographic characteristics. Including time-varying covariates such as teacher characteristics (e.g., quality of instruction), child experiences (e.g., stressors at home), or dyad-level variables (e.g., race- or gender-matching) increases the complexity of these models, but would be a worthwhile pursuit for researchers working with large sample sizes who wish to explain additional variance in teacher-student relationships both between- and within-persons. Given this limitation, our models should be interpreted only as providing relative support for enduring over revisionist effects of early caregiving on teacher-student conflict. Importantly, these results do not suggest that child or teacher characteristics that vary from year to year are unrelated to teacher-student relationship quality, which is by nature a dyadic outcome.

In sum, these data are consistent with the possibility that early maternal sensitivity has enduring predictive significance for teacher-student conflict, but not closeness, through at least Grade 6. The data also suggest that child externalizing problems may play a role in partially accounting for the enduring associations observed. Future studies can expand upon these findings by seeking replication in independent samples, particularly ones that include participants of a different generation, risk level, and/or ethnic composition. Furthermore, replications using alternative methods of assessing teacher-student relationship qualities can also help establish the robustness of these findings. Finally, future investigators would be well advised to examine the influence of teachers in later grades or developmental periods to expand knowledge on the significance of early caregiving for relationships in secondary school environments. Continuing this line of research will thus allow for a clearer understanding of the role of the early parenting context in the development of children and their interactions with adults outside of the home through adulthood.