Can Peers Help Sustain the Positive Effects of an Early Childhood Mathematics Intervention?

Abstract

Our study assessed whether the peer environment in kindergarten and first grade affected student learning following an early mathematics intervention. We leveraged longitudinal data from a cluster-RCT to examine whether math achievement in kindergarten (n = 1,218) and first grade (n = 1,126) was affected by either the share of high-achieving classmates or the proportion of classroom peers who received a preschool math curriculum intervention. Analyses indicated that exposure to treated peers in first grade, but not kindergarten, was significantly associated with small gains in end-of-year achievement. Some analyses also suggested that average peer math achievement was generally positively related to children’s kindergarten and first-grade achievement across conditions, though these results were less robust. We did not find consistent evidence to suggest that the proportion of treated peers coincided with better teaching practices. Taken together, these findings suggest that classroom peer effects may play only a limited role in sustaining early intervention effects.

Early childhood education (ECE) research has found evidence of immediate, positive effects of ECE interventions on developmental outcomes (e.g., Bailey et al., 2017; Duncan & Magnuson, 2013; Elango et al., 2015). Despite these strong initial benefits, longitudinal evaluations often find evidence fadeout in the years following participation (e.g., Bailey et al., 2017, Clements et al., 2013; Lipsey et al. 2018). Many researchers have proposed that fadeout may result from deficiencies in the early elementary school learning environment. Coined the “sustaining environments hypothesis,” this theory asserts that children may reap the benefits of successful ECE interventions if the quality of the subsequent learning environment adequately promotes opportunities for continued skill-building (Bailey et al., 2017). However, much of the research on the sustaining environments hypothesis has reported mixed (e.g., Ansari & Pianta, 2018; Claessens et al., 2014; Jenkins et al., 2018) or null effects (Bailey et al., 2020). Thus, it remains unknown whether unexplored characteristics of the elementary school environment may contribute to the sustained benefits of early childhood interventions.

Although most studies examining the sustaining environments hypothesis have focused on instructional content, few studies have examined the peer environment as a potential mechanism through which early intervention effects could be sustained (but see Burchinal et al., 2023 for recent evidence). As more programs scale up, children may enter elementary school with a greater number of peers who similarly attended high-quality preschool programs. Continued exposure to high-performing classmates may have both direct and indirect effects on program children’s early elementary school achievement. Our study sought to evaluate this possibility by examining whether the peer environment in kindergarten and first grade serves as a sustaining environment for a preschool mathematics curriculum intervention. Importantly, the efficacy of this intervention was evaluated using a cluster-RCT design among a sample of children from racially minoritized groups (55% Black; 21% Hispanic) growing up in primarily low-income families. In our analysis, we examine whether peer exposure within an intervention group helps to boost later achievement, providing evidence from an experimental context on the likelihood that peer achievement could provide a sustaining environment.

ECE Intervention Fadeout

Early childhood education can promote the development of early skills and capacities and reduce educational disparities (e.g., Duncan & Magnuson, 2013; Elango et al., 2015). Despite the promise of cascading developmental effects, much of the literature on early childhood interventions finds that impacts of high-quality programs exponentially decrease following the program’s conclusion (e.g., Bailey et al., 2017). Various theories have been proposed to reconcile the apparent dissonance between the mounting evidence of fadeout and the underlying expectation of compounding ECE benefits. One plausible explanation is the “sustaining environments hypothesis,” which predicts that the benefits of early childhood programs may persist if the quality of the subsequent learning environment is sufficient to build upon the gains made during preschool (Bailey et al., 2017). In other words, the cognitive and behavioral benefits of ECE programs may be maximized for children who are afforded opportunities for continued skill building and diminished for children attending classrooms of poor quality. Although there is some evidence to suggest that exposure to schools with greater funding may help to mitigate fadeout (Johnson & Jackson, 2019), evidence of the sustained benefit of general school and classroom quality (e.g., school proficiency scores, class size, exposure to advanced material) has been fairly mixed (e.g., Ansari & Pianta, 2018; Bailey et al., 2020; Bassok et al., 2019; Claessens et al., 2014; Jenkins et al., 2018). Therefore, it remains unclear which features of children’s elementary school environments, if any, may help facilitate the sustained benefits of early childhood interventions.

The lack of evidence for instructional characteristics (i.e., school and classroom quality) as sustaining environments has led to an increased interest in the possibility that peer composition may be an overlooked component of early elementary school classrooms. Much of this interest is guided by large-scale evaluations of ECE programs, as some have found evidence suggesting that cognitive and behavioral effects may “spillover” to children who did not attend such programs (e.g., Ladd et al., 2014; List et al., 2019; Neidell & Waldfogel, 2010). Though the exact mechanism is unclear, some researchers have posited that spillover effects may result from observational learning that takes place during direct interactions with peers (e.g., Bandura, 1986; Burchinal et al., 2023; Foster et al., 2020; Justice et al., 2011; Vygotsky, 1978). Conversely, others have argued that exposure to preschool attendees may indirectly impact children’s cognitive and behavioral development through changes in classroom-level processes, such as classroom behavior and teacher instruction (e.g., Bierman et al., 2008; Burchinal et al., 2023).

Peer interactions may facilitate the acquisition of cognitive and behavioral skills by providing children with an opportunity to observe and practice skills that were modeled, scaffolded, and encouraged by their high-achieving peers (Bandura, 1986; Vygotsky, 1978). Due to the dynamic nature of these interactions, much of the direct peer effects research has focused on developmental skills more overtly transmitted through social interactions (i.e., language, behavior, and social-emotional skills). These studies typically report a positive association between children’s exposure to high-performing peers in preschool and early elementary school and their end-of-year language/literacy skills (e.g., Burchinal et al., 2023; Foster et al., 2020; Justice et al., 2011; Mashburn et al., 2009) behavior (e.g., Choi et al., 2018; DeLay et al., 2016), and social-emotional functioning (e.g., Choi et al., 2018; Montroy et al., 2016). Perhaps not surprisingly, relatively few studies have examined the relation between peer achievement and children’s development of less socially oriented skills, such as their early math capability (e.g., Coley et al., 2019; Henry & Rickman, 2007; Moiduddin et al., 2012).

Although children may reap direct benefits from spending time with more academically adept peers, there is less understanding of how variation in peer achievement may relate to early elementary school classroom instruction. This indirect peer pathway proposes that teachers may recognize and respond to the average ability level of their students by tailoring their instruction to focus on more (or less) advanced material (e.g., Sacerdote, 2011). Unfortunately, descriptive work on teacher instructional practices indicates that this may not be the case, as kindergarten teachers frequently target skills that many children have already acquired prior to grade entry (e.g., Engel et al., 2013; Engel et al., 2016). Although there is some evidence to suggest that classroom processes are differentially affected by peer cognitive and behavioral skills (Bierman et al., 2008; Burchinal et al., 2023), many other studies have failed to substantiate these claims (e.g., Coley et al., 2019; Moiduddin et al., 2012; Weiland & Yoshikawa, 2014).

Taken together, peer effects research suggests that classroom composition may provide a sustaining environment for high-quality ECE attendees through both direct peer interactions and indirect effects on classroom instruction (though the evidence for indirect effects has been less robust). Unfortunately, few studies (e.g., Unterman & Weiland, 2020) have examined the potential for the elementary school peer environment to sustain the effects of early childhood programs. A notable exception is a recent paper by Burchinal and colleagues (2023) that evaluated whether the share of kindergarten peers who attended preschool could abate fadeout effects. Using data obtained from a district-wide cohort study of a preschool program, the authors found limited support for both the “direct” and “indirect” peer pathways. Indeed, having more preschool-attending peers during kindergarten was related to gains in children’s vocabulary and inhibitory control skills, as well as small changes in teacher instructional quality. However, null effects were reported for most other child outcomes (e.g., reading, math, executive function) and classroom process measures. Importantly, the authors found no evidence to suggest that this dimension of peer composition could provide a sustaining environment for ECE effects, as both preschool attenders and nonattenders benefitted from having more preschool peers in their kindergarten classrooms. Given that this study lacked any exogenous variation in ECE attendance, it is difficult to know how selection bias might affect these results; selection effects could impact the share of pre-k attenders in a classroom, and the comparison between preschool attenders and non-attenders.

To address these issues, we examined experimental data from an intervention evaluation known as the TRIAD study. The evaluation used a cluster randomized design to test the efficacy of the Building Blocks curriculum, a preschool mathematics program that utilizes evidence-based learning trajectories to promote children’s foundational understanding of math (Clements & Sarama, 2008 for more information). Initial analyses found that the intervention produced large gains (i.e., ~0.70 SDs) in children’s math achievement at the end of preschool (Clements et al., 2011), though differences between the intervention and control group ultimately converged by the middle of elementary school (see Bailey et al., 2017 for impacts through fifth grade). Interestingly, the TRIAD study incorporated a sustaining environments treatment arm that exposed intervention children to enhanced instruction in kindergarten and first grade. Children in this “follow-through” condition had stronger impacts through first grade than those who received only the preschool intervention (Clements et al., 2013). Though these results suggest that instructional alignment contributed to the persistence of effects, Jenkins and colleagues (2018) found little indication that the quantity and quality of teachers’ math instruction explained any sustained intervention impacts in the Building Blocks study. Therefore, the current study extends previous work by further examining whether peer composition in kindergarten and first grade could moderate trajectories of fadeout and persistence in TRIAD.

Present Study

Our study contributes to the growing body of literature on peer effects in children’s early learning environment. Using data obtained from the TRIAD Study (described below), we examined whether the peer environment in kindergarten and first grade influences children’s subsequent achievement following the implementation of the Building Blocks preschool mathematics curriculum intervention. Several aspects of the TRIAD study make it uniquely suited for the examination of peer effects. First, the Building Blocks curriculum focuses on integrating mathematical learning into preschool classroom activities (e.g., music, art, play), and encourages peer interactions around math by promoting small and whole group discussions of children’s mathematical strategies and reasoning (e.g., “Turn to your partner and explain what the answer is and why”; Clements & Sarama, 2007). We examine the extent to which these peer dynamics – which were explicitly encouraged during the preschool period for children in the intervention group – extend into kindergarten and first-grade learning environments. Second, the TRIAD study’s inclusion of an extended treatment arm, in which kindergarten and first-grade teachers were encouraged to build upon the pedagogical approach implemented in the preschool intervention, allows us to test if peer interactions coupled with additional professional support for teachers may lead to stronger achievement gains. Third, the TRIAD study’s intentional targeting of children growing up in low-income communities provides a crucial context for understanding the dynamics of sustained early intervention effects, as ECE programs are often designed to provide support to children who may be at risk for underachievement early in school (e.g., Duncan & Magnuson, 2013). Fourth, the incorporation of classroom observational measures may help to elucidate the mechanism through which classroom composition may affect achievement outcomes. Finally, the initial large, positive, impacts on children’s math achievement should allow us to better observe peer effects, as classroom composition is likely to be meaningfully affected by the share of “treated” classmates.

Although previous studies have examined both the intervention effects of the TRIAD study (e.g., Bailey et al., 2018; Clements et al., 2011; Clements et al., 2013) and the role of teacher instruction (e.g., Jenkins et al., 2018), we are the first to our knowledge to investigate direct and indirect effects of the peer environment. We operationalized children’s early elementary school peer environments using two key measures of classroom composition; the share of “treated” peers and the share of “high math achievers.” Further, we measured indirect peer effects through differences in classroom characteristics, which we defined as the number of observed math activities and the quality of teacher math instruction. Thus, the present study sought to answer three research questions:

1. To what extent is the share of “treated” peers in kindergarten and first-grade classrooms related to individual gains in end-of-year achievement?

2. Does the proportion of “high math achieving peers” in early elementary school classrooms predict children’s subsequent math achievement in kindergarten and first grade?

3. To what extent does the variation in children’s classroom peer environment relate to changes in teachers’ instructional practices in kindergarten and first grade?

Methods

TRIAD Study Design

The current study relies on data obtained from an evaluation of a scale-up model based on learning trajectories, called TRIAD (Technology-enhanced, Research-based, Instruction, Assessment, and professional Development), which used the Building Blocks curriculum in preschool. The preschool math program was designed to boost early achievement among marginalized children from low-resource communities in an effort to mitigate the effect of early inequity on subsequent educational and achievement outcomes (Clements et al., 2011). As a result, the TRIAD cluster-randomized trial involved 42 public elementary schools serving low-resource communities in Buffalo, NY and Boston, MA. Importantly, these schools had a high number of self-contained preschool classrooms, and a history of children remaining from preschool through elementary school (Clements et al., 2011). Participating schools were randomized within blocks to one of three conditions: 1) the “preschool-only” intervention group, 2) the “follow-through” intervention group, or 3) the treatment-as-usual control. Both intervention conditions (“preschool-only” group and “follow-through” group) provided preschool teachers with professional development and coaching sessions to facilitate the implementation of the Building Blocks curriculum. In addition to preschool teacher training, schools assigned to the “follow-through” intervention condition received supplementary professional development sessions to help kindergarten and first-grade teachers build upon the gains children made in preschool. Though we do not rely on impact estimates in this study, it should be noted that random assignment was compromised by researcher-approved additions (n = 4) and post-randomized switching (n = 6). As a robustness check, we excluded these ten schools from our analyses and determined that our interpretation remained unchanged (see Tables A9-A10). Therefore, we include all schools in our analyses and examine peer effects based on each school’s experimental condition at the time of data collection.

Student data collection began during the 2006–2007 academic year following a training year of intervention implementation. Children’s math achievement was assessed in the fall (pretest) and spring (posttest) of preschool using the Research-based Early Mathematics Assessment (REMA; Clements et al., 2008). Follow-up assessments of children’s achievement occurred at the end of kindergarten (n = 1,218) and first grade (n = 1,126), along with classroom observation to assess the quantity and quality of math instruction (Clements et al., 2008).

Children were eligible for inclusion in our analyses if they completed all assessments through kindergarten or first grade, and neither skipped a grade nor were retained. Of the original sample (N = 1,305), 93% of participants met the inclusion criteria in kindergarten (n =1,214) and 84% of participants met the criteria in both kindergarten and first grade (n = 1,095). Table 1 presents baseline characteristics for each analytic sample. As can be seen in Table 1, the demographic breakdown of the kindergarten sample was nearly identical to that of the first-grade sample. Approximately 55% of participants identified as Black, 21% identified as Hispanic, and 85% qualified for free or reduced-price lunch. There was no evidence of systematic differences between children included and excluded from either analytic sample (see Table A1), and little indication of baseline imbalance across the three experimental conditions (see Table A2).

Table 1.

Sample Baseline Characteristics

	Kindergarten Sample	First Grade Sample
	M	M
Pre-K Entry Rasch Score	−3.23 (0.83)	−3.20 (0.82)
Pre-K Posttest Rasch Score	−1.97 (0.71)	−1.93 (0.68)
Female	0.51	0.51
African American	0.54	0.55
Hispanic	0.21	0.21
Ethnicity- Other	0.06	0.06
Limited English Proficiency	0.16	0.16
Free/Reduced Price Lunch	0.84	0.85
Age (years) at Pretest Assessment	4.34 (0.35)	4.34 (0.35)
Buffalo Site	0.74	0.73
Stayed in Pre-K School Through Kindergarten	0.79	0.80
Stayed in Pre-K School Through First Grade		0.69
Observations	1214	1095

Note: Means and standard errors (in parentheses) are presented for the study sample within each grade. Children were included in the sample if they had completed all assessments, and neither skipped a grade nor were retained.

Follow-Up Data Considerations

Our study aimed to use measures that captured children’s early elementary school peer environment following a preschool intervention. However, some discussion is warranted about how we used the original RCT data, and how we addressed the complex nature of children’s post-randomized school environments. As with other preschool intervention evaluations, children in the TRIAD study were not randomly assigned to their kindergarten or first-grade classrooms. Rather, participating schools were randomized to an experimental condition during the preschool year, which they maintained for the rest of the study. Given that treatment and control groups were predominantly self-contained, variation in the share of “treated” children was largely limited to the intervention groups (i.e., “preschool-only” and “follow-through”). This is because most children attended preschool at their districted neighborhood elementary school; thus, approximately 80% of study children matriculate into kindergarten at their preschool site. Therefore, the majority of control children had no exposure to “treated” children in kindergarten or first grade. This lack of variation across conditions prompted the examination of peers within each condition rather than testing how peer composition may moderate intervention impacts.

Although we attempt to capitalize on the within-group variation described above, it is important to recognize that kindergarten and first-grade study classrooms were not solely composed of Building Blocks and/or control group children. In addition to study-consented children (observed Building Blocks or control group children), classrooms also contained non-study children (unobserved Building Blocks or control group children) and children who did not attend preschool at a study school. In other words, the “non-treated” peers in a given classroom cannot necessarily be considered control peers (i.e., having no exposure to ECE), and non-control peers cannot always be deemed “treated” peers. Further variation in classroom composition was derived from school switching during a child’s kindergarten or first-grade year. We attempted to comprehensively measure the composition of treatment and control classrooms by incorporating administrative data from non-study children. Using these data (which were obtained from one district), we were able to capture all children who attended a Building Blocks or control preschool classroom and construct more precise measures of peer exposure in kindergarten and first grade.

With these sources of classroom variation in mind, we examined whether the share of “treated” children in a given kindergarten and first-grade classroom predicted children’s math achievement at the end of the year across experimental groups. We also tested whether the classroom share of “high math achievers” predicted end-of-year gains in achievement. Finally, we assessed whether these classroom composition measures predicted variation in the quantity and quality of teachers’ math instruction.

Measures

Children’s mathematical knowledge.

Children’s preschool and elementary school math ability was measured using the Research-based Elementary Math Assessment (REMA; Clements et al., 2008; Sarama & Clements, 2011), a 225-item assessment that is ordered by Rasch item difficulty and concludes after 4 consecutive errors. Math ability is measured according to the developmental progressions that underlie the Building Blocks learning trajectories, including number (e.g., verbal counting, object counting, subitizing, addition and subtraction) and geometry progressions (e.g., shape recognition, shape composition and decomposition, spatial imagery). The REMA is conducted as an individual interview and is informed by standardized procedures for administration, recording, coding, and scoring. Overall test reliability ranged from 0.93 to 0.94, with subtest reliability ranging from 0.75 to 0.94, (see Clements et al., 2008 for details on validity). The reliability for the present study population was 0.92.

Classroom instructional quality.

Math instructional environment was assessed using the Classroom Observation of Early Mathematics Environment and Teaching (COEMET), a researcher-developed observational measure of math instruction. The COEMET is designed to be administered over a three-hour period, wherein trained observers rate the quality of each structured math activity using a series of 28 Likert-scaled items (e.g., “the teacher began by engaging and focusing children’s mathematical thinking”). We generated an overall instructional quality score by averaging these items for each observed activity and used the total number of observed math activities to construct a measure of instructional quantity. COEMET observations also provided us with data on class size, which was used as the denominator in our key peer environment measures described below.

Notably, COEMET data was only available for a sub-sample of study children in kindergarten and first grade. As explained in Clements et al. (2013), student movement between schools proliferated as the study years progressed. The number of schools and classrooms grew from 42 schools and 106 classrooms in preschool, to 140 schools and 275 classrooms in kindergarten, and finally reaching 175 schools and 374 classrooms in first grade. While children who moved or transferred schools were largely retained, the increasing number of classrooms prohibited the collection of classroom-level data for all study children. COEMET observations were therefore limited to 93 classrooms, which constrained the collection of data on teacher instructional practices and classroom characteristics (i.e., class size). In response to these constraints, we adjusted class size information for children who were enrolled in a study school but were missing classroom data due to COEMET limitations in kindergarten (n = 67) and first grade (n = 203). We relied on Boston Public Schools administrative data to fill in the missing classroom characteristics for Boston study schools. Because were unable to access Buffalo administrative records, missing class size information in first grade (n = 156) was imputed using the average class size among Buffalo study schools (i.e., M = 20.38).

Classroom composition

Peer exposure.

To examine children’s exposure to treated and untreated peers, we calculated the proportion of children in each kindergarten and first-grade classroom that either received the Building Blocks intervention in preschool (i.e., “preschool-only” and “follow-through” children) or were part of the preschool control group. We generated these proportions uniquely for each child using the “leave-one-out” method, which ensured that the referent child was not included as a “peer” in their own peer environment. In other words, we counted the number of Building Blocks children in each class, reduced that number by one if the child themself were treated, and divided by the class size to generate the proportion of treated peers. Similarly, we measured children’s exposure to control peers by counting the number of control children in each class, subtracting one if the child did not receive Building Blocks, and dividing by the class size to produce the proportion of control peers within each class.

Although the original study recruited a large percentage of students within each of the 42 study schools (~70%), roughly one-third of intervention and control group children were left unobserved (this measurement coverage is similar to that of Burchinal et al., 2023). To provide a more accurate picture of children’s peer environment, we used administrative records from Boston Public Schools to locate children who were in a TRIAD preschool classroom but were not recruited to participate in the intervention evaluation. We tracked these non-study children to their subsequent classrooms and constructed a more precise peer exposure measure that included both study and non-study children. Although we were unable to obtain administrative records for Buffalo study schools, the high correlation (r = 0.96–0.99) between the peer measures constructed using TRIAD data alone and those constructed using the comprehensive Boston Public Schools data gives us confidence in the accuracy of the study data.

Additionally, several children transferred schools following the conclusion of the preschool intervention. Specifically, 250 children changed schools between preschool and kindergarten (39 to another study school; 211 to a non-study school), and 143 children transferred schools between kindergarten and first grade (34 to another study school; 109 to a non-study school). Although we were able to reconstruct the peer environments of children who remained within the constellation of study schools, we had little information on the characteristics of non-study school classrooms. However, we were able to track children to their new schools and conclude that they had little to no exposure to other study peers. Therefore, we set the share of “treated” or control peers equal to zero for all children who met sample inclusion criteria but had moved to a non-study school in kindergarten and/or first grade. As we describe below, we tested the sensitivity of our results to different assumptions about these school movers, including analyses that excluded them completely from the sample.

As a whole, our analyses relied on existing variation in the proportion of preschool treated peers within each study classroom and school. Based on the sources of variation noted here, we find that study children are exposed to a wide range of preschool peers during kindergarten and first grade. As illustrated in Figure A1 by grade and study condition (mean and SDs reported in Table 2), one can see that while 15–20% of children have no treated peers in their classrooms, most children have a non-zero value ranging from 5% to over 90%.

Table 2.

Descriptive Statistics for Key Study Variables

	Intervention: Preschool-Only Group		Intervention: Follow-Through Group		Control Group
	M	SD	M	SD	M	SD
	Peer Exposure
Proportion of Treated Peers
Kindergarten (n = 1,214)	0.42	0.24	0.43	0.25	0.02	0.07
First Grade (n = 1,095)	0.30	0.22	0.31	0.24	0.03	0.08
Proportion of Control Peers
Kindergarten (n = 1,214)	0.01	0.04	0.01	0.03	0.32	0.26
First Grade (n = 1,095)	0.01	0.04	0.01	0.06	0.28	0.24
	Math Achievement
Proportion of High Math Achievers
Kindergarten (n = 1,008)	0.13	0.10	0.13	0.09	0.04	0.06
First Grade (n = 830)	0.09	0.09	0.12	0.11	0.08	0.07
	Classroom Characteristi cs
Class Size
Kindergarten (n = 1,015)	1 8.87	2.82	18.22	3.96	18.34	2.96
First Grade (n = 848)	21.45	3.62	19.84	4.48	19.94	4.42
COEMET: No. Math Activities
Kindergarten (n = 948)	4.80	2.61	5.76	2.88	4.76	2.61
First Grade (n = 661)	3.12	1.60	2.36	1.44	2.65	1.28
COEMET: Instructional Quality
Kindergarten (n = 893)	3.79	0.33	3.90	0.47	3.77	0.45
First Grade (n = 661)	3.89	0.44	3.98	0.31	3.91	0.26

Note. High math achievement is defined as performance in the top quartile of math achievement the prior year. First-grade class size was imputed for 144 children at the Buffalo study site using the mean class size for Buffalo. All proportion variables were rescaled for subsequent regression analyses to reflect a ten-percentage point increase. COEMET instructional quality had a maximum value of 5 for both grades, while the maximum number of math activities was 15 in kindergarten and 8 and first grade.

Peer achievement.

Peer achievement was measured by the share of “high math achievers” in each kindergarten and first-grade classroom. “High math achievers” are defined as study children who scored in the top quartile on the REMA math assessment given at the end of the previous year (i.e., end of preschool for kindergarten analyses; end of kindergarten for first-grade analyses). Like our peer exposure measures, we generated the proportion of “high math achievers” by first counting the number of children in each class who scored in the top quartile on the REMA the previous year. We then reduced the number of “high math achievers” by one if the child themself scored in the top quartile and divided by the class size to generate a proportion.

Unlike the peer exposure variables, we were less comfortable making assumptions about the peer achievement within non-study school classrooms. Instead of assuming that children had no exposure to high-achieving peers (i.e., setting the proportion of high achievers equal to zero), we removed them from our peer achievement analyses completely. This meant that 211 children in kindergarten and 265 children in first grade were excluded from the achievement analyses.

Baseline covariates.

Our analyses included child gender, race and ethnicity, limited English proficiency status, age at pretest, and children’s math skills assessed at the beginning (pretest) and end (posttest) of preschool. We also included controls for family income, as defined by children’s eligibility for free or reduced-price lunch.

Analytic Approach

Three sets of analyses were conducted in Stata 16.0 to assess whether children’s peer environments improve math achievement. First, we examined the association between children’s peer exposure at the end of kindergarten and first grade. Next, we assessed the relationship between peer achievement and children’s individual math performance at the end of kindergarten and first grade. Finally, we examined whether our classroom composition measures were associated with key instructional variables in kindergarten and first grade. As mentioned above, school-level intervention restricted the variation in the share of “treated” and control peers to their respective experimental conditions (i.e., the two intervention conditions and the control condition, respectively). Therefore, our analyses focused on understanding how children’s peer environment in kindergarten and first grade affected child and classroom outcomes within each condition. In other words, each set of analyses examined the role of classroom composition separately for children in the “preschool-only” intervention group, the “follow-through” intervention group, and the control group.

The first set of analyses examined the association between children’s peer exposure and their math achievement at the end of kindergarten and first grade. Data were analyzed at the individual level using OLS regression, with standard errors adjusted to account for clustering at the school level (though we also assessed sensitivity to clustering at the classroom level; see Tables A12 through A15). We relied on four model specifications that included increasing sets of covariates to examine the robustness of any associations between peer exposure and end-of-year achievement. Model 1 simply involved a bivariate regression in which a child’s achievement score in kindergarten or first grade was regressed on their peer exposure variable for that specific grade. Next, we introduced baseline covariates (i.e., pretest math score, age at pretest, gender, race/ethnicity, free or reduced-price lunch, and study site) and blocking group fixed effects (Model 2). In the third model, we added preschool fixed effects (i.e., the child’s school at the time of random assignment) to adjust for any between-school differences, such as variation in the share of children recruited for study participation within each school (Model 3). Finally, we included the child’s math achievement from the previous study wave to examine how peer exposure related to gains in achievement above and beyond children’s ability at grade entry (Model 4). This meant that our kindergarten models included preschool posttest as a control in Model 4, and first-grade models included the spring of kindergarten REMA as a control. These achievement measures are endogenous to intervention status, but because our models examine associations within each intervention group, the lagged achievement controls can provide an additional adjustment for each child’s level of achievement at the beginning of the grade in question to better observe growth over the single school year.

Next, we examined the relationship between children’s exposure to “high achieving” peers and their math achievement at the end of kindergarten and first grade. As with the peer exposure analyses, student-level data were analyzed using a series of four regressions. However, the peer achievement analyses relied on a more restricted sample. Specifically, children who moved to non-study schools in kindergarten or first grade were excluded from the analyses as we were unable to verify the ability level of their classmates. Apart from these sample distinctions, the four key model specifications remained unchanged.

Our final set of analyses examined whether children’s peer environment was associated with teacher instructional practices in kindergarten and first grade. Unlike the previous analyses, data were analyzed at the classroom level using a series of four OLS regressions for each grade. Models 1 and 2 examined the association between each classroom composition measure and the total number of observed math activities, while Models 3 and 4 assessed the relation between each classroom composition measure and the average quality of classroom instruction. All models included fixed effects for blocking group and site as well as standard error adjustments. No additional controls were included due to the restricted sample sizes, and we considered these models to be exploratory.

Across all analyses, continuous outcomes were standardized so coefficients can be likened to effect sizes. Classroom composition variables were also rescaled to reflect a 10-percentage point increase in the share of peers within each classroom. Though the dataset had very little missing data, mean imputation was used to account for missing data on measures of family income (n = 195; ~16% of the kindergarten sample) and study school class size (n = 156; ~14% of the first-grade sample). This approach is consistent prior peer research (e.g., Burchinal et al., 2023) and has been found to be appropriate for use in group randomized control trial data (Puma et al, 2009). All affected models controlled for imputation using a “dummy variable” that indicated whether data had been imputed for a given variable.

Beyond our three main analyses, several sensitivity analyses were conducted. First, we restricted the sample for both math achievement analyses to only include participants with peer achievement data to ensure that estimates were not driven by children who moved to non-study schools in kindergarten or first grade. Second, we assessed the importance of a consistent peer environment by examining the relationship between children’s end-of-year achievement and the share of kindergarten and first-grade peers that attended the same preschool site. Third, we explored the potential for threshold effects of both classroom composition measures. Fourth, we reexamined the classroom-level analyses that involved count variable outcomes using Poisson regression. Finally, we addressed the nesting of children within classrooms by clustering SEs at the classroom level and testing a mixed model with classroom and school level random effects.

Results

Descriptive Findings

Baseline balance was assessed across the two analytic samples and the three experimental conditions. Table 1 provides baseline characteristics for children eligible for inclusion in the kindergarten and first-grade sample, with a comparison between the full sample and each analytic sample provided in Appendix Table 1 (Table A1). Children included in the kindergarten sample were significantly more likely to remain in the same school from preschool to kindergarten and attend school in Buffalo than those excluded from the sample. Included children were also slightly younger at pretest and were less likely to be Hispanic than excluded peers, though these differences dissipate by first grade. Examination of baseline characteristics across the three experimental conditions revealed little evidence of imbalance apart from small differences in gender and age at pretests (see Table A2).

Table 2 provides descriptive statistics for key study variables for the “preschool-only” intervention group, the “follow-through” intervention group, and the control group. As reflected in Table 2, the proportion of “treated” peers was nearly equivalent across the two intervention groups in kindergarten (M = 0.42–0.43) and first grade (M = 0.30–0.31). This means that approximately 42% of kindergarten classroom peers were treated, on average, for a given treated child in our sample, and approximately 30% of first-grade classroom peers were treated, on average, for the same student. Interestingly, the share of control group peers was similar, but slightly lower in kindergarten (M = 0.32) and first grade (M = 0.28). This is likely due to small sampling differences between the experimental groups.

The share of “high achieving” peers, captured students in the top 25% of the REMA test from the previous measurement wave and was not conditional on treatment status. However, the positive end-of-preschool treatment impact is reflected in this variable, as the two intervention conditions had a higher share of “high achieving” peers in kindergarten (M = 0.13) than the control group (M = 0.04). Reflecting fadeout, this difference decreased in first grade for the control group (M = 0.08), “preschool-only” group (M = 0.09), and “follow-through” group (M = 0.12). Finally, class size and instructional quality were relatively consistent across grade and experimental condition. However, a greater number of math activities were observed in kindergarten for the “follow-through” group (M = 5.76, SD = 2.88) than for the “preschool-only” group (M = 4.80, SD = 2.61) and control group (M = 4.76, SD = 2.61). As with peer achievement, differences in instructional quantity across conditions decreased by first grade.

Main Findings

Estimates from our main analyses are presented in Tables 3, 4, and 5. Associations between children’s end-of-year math achievement and their peer environment in kindergarten and first grade are provided in Tables 3 and 4. Kindergarten (Columns 1–4) and first-grade (Columns 5–8) data were analyzed at the student level using four main model specifications. Results are presented in four panels to reflect examination within the “preschool-only” intervention group, the “follow-through” intervention group, the control group, and a “pooled” intervention group that jointly examines both intervention groups.

Table 3.

Associations Between Peer Exposure and End-of-Year Math Achievement Within Experimental Groups

	1	2	3	4	5	6	7	8
		Kindergarten				First Grade
			“Preschool-Only” Intervention Group
K Prop. of Treated Peers	−0.03 (0.03)	0.04 (0.03)	0.03 (0.03)	0.02 (0.02)
FG Prop. of Treated Peers					−0.02 (0.03)	0.03* (0.02)	0.03* (0.02)	−0.00 (0.01)
N	417	417	417	417	377	377	377	377
			“Follow-Through” Intervention Group
K Prop. of Treated Peers	0.01 (0.03)	0.01 (0.02)	0.02 (0.02)	−0.00 (0.01)
FG Prop. of Treated Peers					0.05 (0.04)	0.05* (0.02)	0.06* (0.02)	0.04* (0.02)
N	436	436	436	436	390	390	390	390
				Pooled Intervention Groups
K Prop. of Treated Peers	−0.01 (0.02)	0.02 (0.02)	0.02 (0.02)	0.01 (0.01)
FG Prop. of Treated Peers					0.02 (0.03)	0.04** (0.01)	0.04** (0.01)	0.02+ (0.01)
N	853	853	853	853	767	767	767	767
				Control Group
K Prop. of Control Peers	−0.06 (0.04)	−0.00 (0.01)	−0.00 (0.01)	0.01 (0.01)
FG Prop. of Control Peers					−0.04 (0.03)	−0.03 (0.03)	−0.04 (0.03)	−0.01 (0.02)
N	356	356	356	356	328	328	328	328
Controls Included
Baseline Student
Controls		Inc.	Inc.	Inc.		Inc.	Inc.	Inc.
Blocking Group /Site Fixed Effects		Inc.				Inc.
Preschool Fixed Effects			Inc.	Inc.			Inc.	Inc.
Prior Year Math Score				Inc.				Inc.

Note. Standard errors (in parentheses) are clustered at the school level.

⁺p<.10

^*p<.05

^**p<.01

^***p<.001.

Estimates in each panel were derived from separate models that were run using student-level data. Models 1 through 4 examine the association between the share of kindergarten peers and kindergarten math achievement, while models 5 through 8 examine the association between the proportion of first-grade peers and first-grade math achievement. All proportion variables were rescaled to reflect a ten-percentage point increase. For each peer exposure variable, the first model (columns 1 and 5) contains no controls; the second model (columns 2 and 6) includes baseline covariates and fixed effects for blocking group and site; the third model (columns 3 and 7) includes fixed effects for preschool (j = 42); and the fourth model (columns 4 and 8) includes baseline covariates, preschool fixed effects, and prior year math score (i.e., math achievement at end of preschool or end of kindergarten).

Table 4.

Associations Between Peer Achievement and End-of-Year Achievement Within Experimental Groups

	1	2	3	4	5	6	7	8
		Kindergarten				First Grade
			“Preschool-Only” Intervention Group
K Prop. of High Math Achievers	0.27** (0.08)	0.03 (0.04)	0.03 (0.04)	−0.06* (0.03)
FG Prop. of High Math Achievers					0.40** (0.10)	0.14* (0.06)	0.15* (0.06)	0.01 (0.06)
N	345	345	345	345	285	285	285	285
			“Follow-Through” Intervention Group
K Prop. of High Math Achievers	0.35* (0.11)	0.12+ (0.06)	0.13+ (0.06)	0.01 (0.04)
FG Prop. of High Math Achievers					0.50*** (0.07)	0.15 (0.14)	0.21 (0.14)	0.07 (0.10)
N	365	365	365	365	290	290	290	290
			Pooled Intervention Groups
K Prop. of High Math Achievers	0.30** (0.07)	0.08+ (0.04)	0.06 (0.04)	−0.04 (0.03)
FG Prop. of High Math Achievers					0.47*** (0.06)	0.19** (0.06)	0.19* (0.07)	0.05 (0.05)
N	710	710	710	710	575	575	575	575
				Control Group
K Prop. of High Math Achievers	0.56** (0.14)	0.13** (0.04)	0.13** (0.04)	−0.05 (0.05)
FG Prop. of High Math Achievers					0.55*** (0.11)	0.16 (0.15)	0.21 (0.15)	0.07 (0.13)
N	298	298	298	298	255	255	255	255
Controls Included
Blocking Group Fixed Effects		Inc.				Inc.
Preschool Fixed Effects			Inc.	Inc.			Inc.	Inc.
Baseline Student Controls		Inc.	Inc.	Inc.		Inc.	Inc.	Inc.
Prior Year Math Score				Inc.				Inc.

Note. Standard errors (in parentheses) are clustered at the school level.

⁺p<.10

^*p<.05

^**p<.01

^***p<.001.

Estimates in each panel were derived from separate models that were run on student-level data. Models 1 through 4 examine the association between peer achievement and kindergarten math achievement, while models 5 through 8 examine the association between the proportion of high math achievers and first-grade math achievement. The proportion of high math achievers for both grades is defined by the proportion of children scoring in the top quartile of math the previous year and reflects a 10-percentage point increase. For each grade, the first model (columns 1 and 5) contains no controls; the second model (columns 2 and 6) includes baseline covariates and fixed effects for blocking group and site; the third model (columns 3 and 7) includes fixed effects for preschool (j = 42); and the fourth model (columns 4 and 8) includes preschool fixed effects and prior year math score (i.e., math achievement at end of preschool or end of kindergarten).

Table 5.

Associations Between Classroom Composition and Classroom Math Instruction in Kindergarten and First Grade

	No. of Math Activities		Instructional Quality
	1	2	4	5
		“Preschool-Only”	Intervention Group
Kindergarten  Prop. of Treated Peers	−0.11 (0.13)		−0.06 (0.11)
Prop. of High Math Achievers		0.02 (0.14)		0.19 (0.11)
N	50	46	45	41
First Grade
Prop. of Treated Peers	0.09 (0.10)		−0.06 (0.15)
Prop. of High Math Achievers		0.48*** (0.10)		0.43 (0.27)
N	41	36	41	36
		“Follow-Through”	Intervention Group
Kindergarten  Prop. of Treated Peers	0.10+ (0.05)		0.05 (0.08)
Prop. of High Math Achievers		0.16 (0.15)		0.25 (0.16)
N	54	51	52	49
First Grade
Prop. of Treated Peers	0.06 (0.06)		0.01 (0.03)
Prop. of High Math Achievers		0.09 (0.19)		0.09 (0.13)
N	66	55	66	55
		Pooled Intervention Groups
Kindergarten
Prop. of Treated Peers	0.01 (0.07)		−0.02 (0.06)
Prop. of High Math Achievers		0.01 (0.09)		0.13 (0.08)
N	104	97	97	90
First Grade
Prop. of Treated Peers	0.07 (0.04)		−0.00 (0.04)
Prop. of High Math Achievers		0.24* (0.10)		0.26* (0.10)
N	107	91	107	91
		Control Group
Kindergarten  Prop. of Control Peers	0.02 (0.07)		−0.12 (0.07)
Prop. Of High Math Achievers		−0.09 (0.13)		0.23 (0.27)
N	51	48	51	48
First Grade
Prop. of Control Peers	0.06 (0.12)		0.01 (0.09)
Prop. of High Math Achievers		0.49* (0.19)		0.19 (0.15)
N	43	37	43	37
Controls Included
Blocking Group Fixed Effects	Inc.	Inc.	Inc.	Inc.

Note. Standard errors in parentheses.

⁺p<.10

^*p<.05

^**p<.01

^***p<.001.

Columns 1 and 2 examine associations between classroom composition and the number of math activities observed. Columns 3 and 4 examine associations between classroom composition and instructional quality. All models were run on classroom-level data and included blocking group fixed effects. The proportion of “high math achievers” reflects the share of children in each class who scored in the top quartile of math at the end of preschool (kindergarten sample) or kindergarten (first-grade sample). Classroom composition measures are defined in terms of a 10-percentage point increase. Classroom observational measures were standardized, so coefficients can be likened to effect sizes.

Peer Exposure Analyses

As shown in panels 1, 2, and 3 of Table 3, we found little evidence to suggest that the share of preschool “treated” peers significantly predicted end-of-kindergarten math achievement. While the proportion of “treated” peers was generally positively associated with children’s end-of-kindergarten achievement, these estimates were small in magnitude and were not statistically significant. However, the share of “treated” peers in first grade was significantly associated with achievement at the end of first grade. Once controls were added to the model, a significant association of peer exposure was consistently observed for both intervention groups, such that a 10-percentage point increase in the proportion of “treated” peers predicted a 0.03–0.06 SD gain in first-grade achievement for children in the “preschool-only” and “follow-through” intervention group. After controlling for children’s lagged achievement score (i.e., end-of-kindergarten REMA), the share of “treated” peers continued to significantly predict children’s achievement in the “follow-through” intervention group (β = 0.04; SE = 0.02; P = 0.045), though the association for the “preschool-only” group was near zero (β = −0.00; SE = 0.01; P = 0.836). We ran an additional post hoc model to examine if the effects for the two intervention groups were statistically significantly different from one another by testing an interaction between intervention status and peer exposure using the final model specification. This effect was not significant (p = 0.385).

The last two panels in Table 3 offer additional robustness tests of our primary study hypotheses. In the third panel, we combine the samples of the two treatment groups, analyzed separately in the first two panels, to form a “Pooled Intervention” group to address any sample size concerns. The pattern of results remains similar here, although the final specification reduces the proportion of “treated” first-grade peers to marginal significance. In the last panel, we find that the share of control group peers was slightly negatively related to children’s achievement in both kindergarten and first grade, but the associations were not significant. These null associations for the control group act as a type of “placebo test,” and suggest that the associations for the intervention groups were unlikely to be driven by attending classrooms with a greater number of peers from the same preschool. In other words, any positive associations on achievement for the “peer exposure” variable appeared to be unique to the intervention groups and driven by exposure to the intervention.

Peer Achievement Analyses

Table 4 presents the relationship between the share of “high math achievers” and children’s math achievement in kindergarten (Columns 1 through 4) and first grade (Columns 5 through 8). Overall, peer math ability appeared to be positively related to children’s end-of-kindergarten math achievement across all experimental conditions. Although we observed positive associations across all three intervention samples when no controls were included, adding covariates to the model quickly diminished associations, providing substantial evidence for selection effects on this measure. Sample sizes for these models were smaller than those shown in Table 3 because models were restricted to students who remained in a study school, driving SEs upward. Some substantively important associations were observed once school fixed effects were included in kindergarten and first grade (estimates ranging from ~0.10 – 0.20). However, in both grades, we did not find any indication that additional “high achieving” peers positively affected achievement once the child’s lagged test score was included, coefficients tended to be close to zero (i.e., Models 4 and 8, respectively).

Classroom Instruction

Table 5 presents findings from the classroom-level analyses that examined the association between classroom composition and teacher instructional practices. Model 1 (Columns 1 and 5) contains no controls, Model 2 (Columns 2 and 6) includes baseline covariates and fixed effects for blocking group and site, Model 3 includes baseline covariates and preschool fixed effects (j = 42), and Model 4 contains baseline covariates, preschool fixed effects, and prior year math score. Columns 1 and 2 provide OLS estimates of the relation between classroom composition measures and the number of math activities observed (for kindergarten and first grade, respectively), while Columns 3 and 4 report the associations between each classroom composition measures and teacher instructional quality (for kindergarten and first grade, respectively). Across the models shown in Table 5, we did not find any consistent evidence suggesting that the proportion of “treated” peers related to either the number of math activities observed or the quality of math instruction.

However, we saw more consistent evidence that the share of “high math achievers” was correlated with instructional measures in both kindergarten and first grade. These correlations suggest that teachers with more advanced math students did, perhaps, teach higher levels of math. However, these bivariate correlations are exploratory and should be viewed with some skepticism given the large reductions in correlations shown in Table 4 when student covariates were included in models relating “high achieving” peers to student end-of-year achievement.

Supplemental Analyses

Results of our sensitivity analyses largely supported the findings presented above. First, we re-examined our main classroom composition analyses using a restricted sample that included only children with peer achievement data. Estimates for the proportion of Building Blocks peers suggest some sensitivity to this restriction—suggesting that the effects may have been partially driven by students moving to non-study schools (Tables A3 and A4). Second, we assessed the relation between children’s end-of-year achievement and the share of kindergarten and first-grade peers that attended the same preschool. No new relations emerged (Table A5). Third, we explored the potential for threshold effects of both classroom composition measures within each experimental condition; however, these estimates were too noisy and inconsistent to interpret (Table A6). Fourth, we used a Poisson regression to examine our classroom-level analyses, which confirmed that model fit was not significantly improved when accounting for the use of a count variable outcome (Tables A7 and A8). Fifth, we re-examined our main analyses while excluding schools that either switched conditions or were added after randomization. Adjusting for these post-randomization changes did not alter the interpretation of our findings (Tables A9, A10, and A11). Finally, we addressed the interdependency of children at the classroom level by performing classroom level SE adjustments (Tables A12, and A13) and testing a mixed model with classroom and school level random effects (Tables A14, and A15). Estimates and SEs from these analyses were largely consistent with those reported in our main results.

Discussion

The fadeout of early childhood intervention effects has become an increasingly clear and consistent issue for researchers and policymakers. Many have argued that the learning environment following early childhood programs is integral to the persistence of program effects (e.g., Stipek et al., 2017; Phillips et al., 2017). However, many of the sustaining environments studies have been unsuccessful in their efforts to identify the instructional characteristics that may foster continued program benefits. This study uniquely examined the role of the early elementary school peer environment using data from the TRIAD study, a multi-site preschool mathematics curriculum evaluation with a sustaining environments condition. Using measures of classroom composition, we investigated the extent to which two peer effect mechanisms influenced children’s subsequent achievement following the implementation of a preschool mathematics curriculum intervention.

We found little evidence to suggest that continued exposure to “treated” peers significantly predicted children’s achievement at the end of kindergarten. Although exposure to “treated” kindergarten peers was generally positively associated with kindergarten math achievement, these findings were not statistically significant for either intervention group. These results are surprising given both the proximity to the intervention and our theoretical expectations that having higher shares of treated students may generate opportunities for direct and/or indirect peer effects. However, our findings are somewhat in line with recent work by Burchinal et al. (2023), as they did not find an association between the share of kindergarten classmates with public preschool experience and children’s growth in math skills, though associations between kindergarten exposure to preschool attendees and children’s gains in language and inhibitory control skills were observed. Note that our work differs in a crucial way from the Burchinal et al. study, as all children in our sample attended preschool. Thus, our analysis evaluates whether the peer environment meaningfully changes as a result of exposure to enhanced math instruction during preschool. Yet, our findings converge with Burchinal et al. as both studies show that enhanced peer skill levels driven by ECE exposure do not appear to have strong effects on child achievement during kindergarten.

In addition to the null kindergarten peer effects, we also found little indication that teachers altered their classroom instruction in response to the share of “treated” children. Specifically, we found mostly small and null relations between our measures of classroom composition and indicators of math instructional quality and quantity. These results provide additional support for earlier descriptive work suggesting that kindergarten teacher instruction may not be sensitive to children’s abilities, as multiple studies using the ECLS-K sample have shown that kindergarten math instruction largely focused on redundant instruction of already mastered skills (Engel et al., 2013; Engel et al., 2016). Assuming elementary school teachers can freely alter their instruction, our null effects on child achievement measures may have arisen because teachers neither recognized the change in classroom ability level nor adjusted their instruction accordingly.

Despite the null effects in kindergarten, both intervention groups experienced a small positive benefit associated with the share of “treated” peers in their first-grade classrooms. This benefit was strongest for the “follow-through” group, whose teachers received additional professional development sessions on the program’s learning trajectories. The effect was consistently observed across specifications, though the coefficient in our most robust specification produced a p-value just under the .05 threshold (p = 0.045). Certainly, one could argue that our lack of strong a priori hypotheses would necessitate a p-value adjustment to account for multiple comparisons between the two treatment groups of interest. For example, applying the conservative Bonferroni adjustment would set the alpha level to 0.025. So, the effect observed here should be interpreted with some caution.

With this in mind, it is possible that the additional PD sessions provided to teachers in the follow-through group allowed for more vertically aligned classroom processes, which was particularly beneficial when children encountered more advanced math materials in first grade. Our confidence in these first-grade findings is further bolstered by the placebo tests run on the control group. If this finding were due to selection into classrooms, we would expect a similar pattern of effects among children exposed to control group peers. However, having a higher share of control group classmates was consistently negatively associated with subsequent achievement. Thus, this small peer effect does appear to be unique to treatment schools.

Lastly, we observed some positive correlations between the share of “high math achievers” in kindergarten and first-grade classrooms and children’s subsequent achievement. These findings were observed across the three experimental conditions, suggesting that children may universally benefit from exposure to more academically skilled peers. However, we are less confident in the validity of these findings, as our estimates were noisy and diminished upon the inclusion of controls. Therefore, our findings may be due to selection effects rather than children’s exposure to more cognitively and behaviorally skilled peers, as previously found (e.g., Foster et al., 2020; Justice et al., 2011; Mashburn et al., 2009).

This study adds to the ECE literature by examining whether children’s peers within an intervention group may help to maintain the effects on children’s subsequent achievement. Considering our findings in conjunction with the numerous evaluations of the sustaining environments hypothesis, it seems that the effect of ECE programs is unlikely to be sustained by easily observed components of early learning environments (Bailey et al., 2020). Indeed, the magnitude of our findings indicates that while the peer environment may play a role in the persistence of preschool effects, it is unlikely to be the key component responsible for sustaining the impact of EC programs. However, the observed impact of “treated” peers should be understood within groups of children who participated in the intervention. It is unclear whether exposure to “treated” peers may confer benefits to non-treated children, as the lack of crossover between experimental conditions prohibited us from spillover effects (e.g., Dodge et al., 2017; Ladd et al., 2014). This lack of variation in the measure of “treated” peers in control group classrooms also limited our ability to conduct the traditional sustaining environments interactions; however, the observed first-grade benefit indicates that peer groups could play a role in the persistence of early childhood program effects.

Limitations

Despite this study’s notable strengths, several limitations should be acknowledged. First, we were unable to fully observe the peer environment of all study school classrooms, as approximately 70% of children in each study school were recruited for study participation. Although we were able to use administrative records to improve the precision of our classroom composition measures for Boston study schools, we were unable to obtain the data needed to fully observe the peer environment of Buffalo study schools. Though we were able to verify the accuracy of the TRIAD data using Boston’s administrative records, our classroom composition measures may still underrepresent children’s exposure to treated classmates.

Relatedly, we were unable to fully observe the classroom composition and processes of children who moved to a non-study school. Given the nature of the intervention delivery, we were confident that children would have little to no exposure to “treated” peers outside of TRIAD study schools. Therefore, we set the peer exposure variables equal to zero for all children who met sample inclusion criteria but had moved to a non-study school in kindergarten and/or first grade. Though we were able to track children as they moved, we could not attest to the achievement level of their peers, their class size, or their teachers’ instructional practices. Rather than assuming that children had no exposure to “high achieving” students, we excluded all “movers” from analyses examining peer achievement and classroom instructional practices.

Third, our analyses were limited by school-level random assignment. Because preschools fed into elementary schools, children attended the same school from preschool through first grade. Due to the lack of crossovers, we were unable to examine both spillover effects and sustaining environments interactions. Therefore, we separately examined the role of classroom composition within each experimental condition and included a “placebo check” to uniquely examine the peer environment of children in the control group.

Fourth, our conclusions regarding teacher instructional practices are limited by the strength of the observational measure. Due to the number of kindergarten and first-grade classrooms, only a handful of classrooms were observed. Further, selected classrooms were observed at only one point in the middle of the year, with each observation spanning around 3 hours. The limited availability of classroom observational data caused our classroom-level bivariate correlations to be underpowered and noisy. Additionally, the use of a single, point-in-time observation may not accurately capture differences in teachers’ day-to-day instruction.

Finally, the generalizability of our findings is limited by the demographic characteristics of our sample and the historical context during which the study occurred. We relied on data collected between 2006–2007 from a sample of predominantly children of color who attended public preschool and public school in low-resource, urban communities in the Northeast. Therefore, we cannot ascertain how these peer mechanisms will operate in other populations (e.g., non-urban, White, high-resource sample), or whether these mechanisms from fifteen years ago may continue to persist given the subsequent preschool expansion efforts and the academization of preschool and kindergarten.

Conclusion

Taken together, this study provides some evidence to suggest that variation in the peer environment of early elementary school classrooms may differentially impact academic growth following early childhood interventions when additional supports are provided to teachers to build on academic gains made during preschool. Our findings indicate that children in TRIAD’s “follow-through” condition benefitted from having more treated classmates, particularly in first grade when children are exposed to more challenging material. While the peer environment is unlikely to be the sole factor to sustain the impacts of early childhood intervention efforts, more work is needed to better understand the role that they play in promoting or inhibiting continued academic growth. This may include additional random assignment to elementary school classrooms to disentangle peer effects from selection effects, brief observational measures that allow for complete data collection within all study classrooms, and/or the examination of the peer environment across populations and developmental periods. Given the continued efforts to expand early childhood programs, future work should also examine whether changes in the peer environment following randomly assigned early childhood interventions may confer benefits to both treated and non-treated children.

Research Highlights.

• Direct and indirect peer effects were assessed in the context of a randomized control trial.

• Exposure to treated peers was associated with small gains in first grade achievement.

• Peer ability was positively, but less robustly, related to individual math achievement.

• Teaching practices were not consistently improved by the share of treated peers.

• Peer effects may play a limited role in sustaining early intervention effects.