Can the Sustaining Environments Hypothesis Be Sustained? Testing Moderation of Sustained Public Preschool Benefits by Kindergarten Classroom Quality

Abstract

Mixed evidence over whether public preschool – Head Start and school-based public pre-k – confers an academic advantage beyond kindergarten has given rise to several explanations of variation in findings across studies. The “sustaining environments” hypothesis posits that for preschool attenders to maintain an advantage over preschool non-attenders, they must experience kindergarten classrooms of sufficiently high quality. Several studies have evaluated this hypothesis by testing whether preschool attenders benefit more than non-attenders from higher quality in their kindergarten classroom. They have produced mostly null findings but have commonly conceptualized the environment as instructional quality in kindergarten classrooms. We expand on this evidence base by testing for moderation of preschool impacts by instructional quality, along with the quality of two other key dimensions of kindergarten classrooms: the self-regulatory environment and the teacher-child relational environments. Moreover, we conduct this test using data on a diverse sample of students from low-income households who attended public preschool in Tulsa, OK, where preschool attendance has been associated with benefits that are sustained through elementary school. Findings suggest that associations between preschool attendance and first-grade outcomes are robust and mostly do not vary by subsequent kindergarten environments. Further tests of this hypothesis should examine variation in kindergarten environments between, rather than within, preschool evaluations. Researchers should also consider other reasons why some public pre-k programs produce more lasting impacts than others.

While it is largely accepted that contemporary, widely-available public preschool programs like Head Start and state-funded public pre-kindergarten (pre-k) promote kindergarten readiness, there is weaker evidence supporting the proposition that these benefits last after kindergarten entry. This is because many studies find that initial benefits of public preschool dissipate over the kindergarten year as children who did not attend preschool catch up to preschool attenders (Ansari et al., 2020; McCormick et al., 2021; Puma et al., 2012). The few studies that have lasted beyond the end of kindergarten report mixed findings: some find that preschool benefits remain relatively stable into early elementary school (Frede et al., 2009; Johnson et al., 2023), whereas others report disappearance (Puma et al., 2012), or in the case of one study, reversal, such that early benefits became deficits in elementary school (Lipsey et al., 2018). This pattern of mixed results has raised questions about what might explain why some studies find a sustained advantage of preschool attendance while others do not.

One popular theoretical explanation centers around the quality of children’s kindergarten classroom environments. This theory, dubbed the “sustaining environments” hypothesis, suggests that “the long-term success of early educational interventions [like preschool] is contingent on the quality of the subsequent learning environment” (Bailey et al., 2020b, pg. 1). A number of studies have tested this theory, but they too have produced a confusing mix of results. Moreover, they have typically defined sustaining environments according to facets of instructional quality – either at the school level (e.g., average achievement) or via observable measures of classroom structural (e.g., proportion of pre-k peers in the classroom) or process (e.g., teacher-child interaction during instruction) quality. This ignores other key dimensions of children’s educational experiences in kindergarten that may facilitate or undermine sustained preschool impacts, such as the classroom self-regulatory environment and the quality of individual student-teacher relationships.

The present study provides a robust test of the sustaining environments hypothesis by testing whether three distinct measures of the quality of children’s kindergarten learning environments – classroom-level instructional support, classroom-level self-regulatory support, and individual teacher-student relationship quality – moderate associations between public preschool attendance and academic outcomes in first grade. We undertake this test with data from a renowned public preschool program that has produced positive impacts on a range of children’s academic skills in first grade in a diverse sample of students from low-income households.

Public Preschool Impacts

A large body of literature confirms that public preschool attendance has an immediate positive effect on children’s early language, literacy, and math skills – effects detected most often at the end of the preschool year or at the start of kindergarten (Phillips et al., 2017; Yoshikawa et al., 2013). This literature includes results from studies of scaled-up contemporary public preschool programs in Boston, Tulsa, Tennessee, New Mexico, Georgia, and North Carolina (Gormley et al., 2008; Henry et al., 2006; Hustedt et al., 2021; Jenkins et al., 2016; Johnson et al., 2022; Lipsey et al., 2018; Peisner-Feinberg & Schaaf, 2011; Weiland & Yoshikawa, 2013), as well as several multi-state pre-k studies (Barnett et al., 2018; Wong et al., 2008) and the national Head Start Impact Study (HSIS; Puma et al., 2012).

The evidence is more mixed, however, around whether and how long public preschool benefits persist after preschool, because relatively few studies have followed children beyond the fall of kindergarten (Phillips et al., 2017). Those that have produce inconsistent results, with some studies showing persistent positive effects through the spring of kindergarten (Ansari et al., 2020), first grade (Johnson et al., 2022), second grade (Frede et al., 2009; Weiland et al., 2021) and even third grade (Johnson et al., 2023), and others finding that effects fade (Lipsey et al., 2018; Puma et al., 2012) or even reverse sign (Lipsey et al., 2018; Puma et al., 2012) as children progress through elementary school. Even among the studies that find continued significant effects after the fall of kindergarten, some have demonstrated that preschool impacts lessen considerably over time, often by more than half the original effect size during the kindergarten year alone (Ansari et al., 2020; Weiland et al., 2021). In most cases, this is because children who did not attend preschool begin to catch up to the children who did during kindergarten and continue to do so in subsequent years, leading to convergence on outcomes between attenders and non-attenders.

This assortment of findings, in which some preschool programs demonstrate rapid fade-out while others demonstrate sustained impacts, has given rise to new questions about what is happening in kindergarten year classrooms (as well as later grades) that may sustain – or reduce – the preschool boost. In the last few years, the sustaining environments hypothesis has captured researchers’ attention as a promising theoretical explanation for variation in preschool impacts.

The Theory of Sustaining Environments

The sustaining environments hypothesis as put forward by Bailey, Duncan, et al. (2020a) posits that the persistent benefits of an early childhood intervention – including preschool – depend on the quality of the subsequent learning environment children experience (Bailey et al., 2020b, pg. 2). Specifically, it is thought that any initial advantage in skills preschool attenders enjoy relative to non-attenders at kindergarten entry will fade if those skills are not nurtured and enhanced during the kindergarten year. That is, without support aimed at preschool attenders’ starting skill level, those skills will plateau and, over the course of the kindergarten year, as non-attenders’ skills rise, the two groups will converge. Scholars have interpreted this proposition as implying that higher quality subsequent environments differentially boost the skills of preschool graduates, thereby sustaining preschool benefits over time. This theory – that kindergarten classroom environment features may differentially benefit preschool graduates – is one of moderation and has most often been tested as such in the extant literature (see Bailey et al., 2020b).

Sustaining Environments: Instructional Quality

Most prior studies testing whether kindergarten classroom quality is a sustaining environment for preschool benefits have conceptualized classroom quality as a function of the instructional support provided to students. The hypothesis has been tested by interacting whether children attended public preschool (typically a state- or city-wide public school-based pre-k or a Head Start program) with measures of kindergarten instructional quality to predict sustained preschool effects. Instructional quality has been conceptualized in a variety of ways. Some studies have used school-level proxies for instructional quality, such as average student test scores. For example, the positive effects of North Carolina’s pre-k program on attenders’ language and working memory skills were more likely to persist into elementary school when pre-k graduates attended elementary schools with higher average levels of academic proficiency (Carr et al., 2021). Graduates of Boston’s pre-k program demonstrated gains on standardized literacy and math test scores in third grade only when they attended schools with higher average standardized test proficiency rates (Unterman & Weiland, 2020). Pre-k graduates in Tennessee had higher third-grade standardized test scores when they experienced higher-quality schools but only if they also had higher-rated teachers (using state standards; Pearman et al., 2020). Yet other studies have found no evidence of differential sustained pre-k benefits by school-level quality measured by percent of students who are proficient on state achievement tests (Burchinal et al., 2022) or teacher value-added (Carr et al., 2024) in North Carolina.

However, studies that attempt to capture instructional quality at the school level may be less precise than those measuring it more proximally at the classroom level. And yet, a number of recent studies – and a recent meta-analysis (Bailey et al., 2020b) – find limited or no evidence in support of the sustaining environments hypothesis even when instructional environments are measured at the classroom level. This seems to be the case whether classroom instructional quality is measured using process or structural quality dimensions. Process quality is defined as the quality of observed interactions between students and teachers (NICHD ECCRN, 1998), and has most commonly been measured using the Classroom Assessment Scoring System (CLASS; Pianta et al., 2008). The CLASS generates three subscales, capturing classroom instructional quality (e.g., language modeling; quality of feedback), classroom organization (e.g., routines; behavior management), and classroom emotional support (e.g., teacher sensitivity; emotional climate).

Multiple recent studies examining the sustaining environments hypothesis have tested the CLASS as the key theorized moderator of sustained preschool impacts. Burchinal and colleagues (2022) found no evidence that kindergarten classrooms’ scores on the CLASS moderated the impacts of North Carolina’s public pre-k program on end-of-kindergarten outcomes. Likewise, Ansari and colleagues (2023) found no significant moderation by kindergarten CLASS scores of Virginia’s public pre-k on academic impacts in first grade (Ansari et al., 2023). In a recent study of the Boston pre-k program (McCormick et al., 2022), kindergarten classrooms’ CLASS scores did not moderate preschool impacts at the end of kindergarten. However, all three studies used the total CLASS score, so the contribution of the individual subscales – including those tapping non-instructional dimensions of the classroom – remains unknown. Jenkins and colleagues (2018) found no moderation of Head Start effects by the quality of kindergarten and first grade instruction assessed using both teacher reports of the frequency of language/literacy practices and an observational measure of math instructional quality and dosage (the COEMET; Jenkins et al., 2018).

Fewer studies have considered other aspects of process quality, such as time teachers spend teaching specific content areas, or the rigor of the instruction teachers provide, but these too do not appear to moderate pre-k impacts. For instance, in the Virginia pre-k evaluation, there was no evidence of moderation of pre-k impacts on first grade outcomes by amount of time spent on academic instruction, or academic rigor in kindergarten (Ansari et al., 2023). Similarly, in the Boston pre-k evaluation, the degree of advanced academic content provided by kindergarten teachers did not moderate the impact of pre-k attendance on end-of-kindergarten skills, but there was suggestive evidence that time spent teaching different skill types (constrained or foundational, versus unconstrained) might (McCormick et al., 2022).

Other studies have tested dimensions of structural quality as moderators of the preschool boost. Structural quality is defined as easily regulable features of preschool environments like teacher qualifications or class size. Again, there is little evidence that structural features of instructional quality moderate pre-k effects. In the national HSIS, class size did not differentially relate to pre-k impacts in first grade (Jenkins et al., 2018). Likewise, in a national sample, small class size did not differentially benefit pre-k graduates (Bassok et al., 2019).

One reason for these weak and inconsistent findings could be the measures of sustaining classroom instructional environments used. Scholars have struggled to identify to a single scale of instructional quality that predicts more than a modest amount of student academic growth (Burchinal, 2018). The present study uses the CLASS but also experiments with a new measure of instructional quality that includes several features of instructional support that have been found to individually predict children’s growth in language, literacy, or math in early education and elementary school. These include quantity, content, and level of instruction (e.g., Burchinal, 2018; Claessens et al., 2014; McCormick et al., 2022). Specifically, we measure time spent on instruction by observing not only the amount of time teachers delivered instruction across key academic areas but also the amount of time children were engaged in academic instruction: that is, instructional quality as it was delivered by the teachers, as well as how it was received by the students. We include a measure of the amount of instructional time spent specifically on math, in light of previous evidence indicating its importance for growth in math skills (Christopher & Farran, 2020; Wang, 2010), and a measure of advanced versus basic content to capture the level of academic instruction, which has also been linked to outcomes (Claessens et al., 2014). We also measure the time children spend on sequential activities (tasks with a clear goal and sequence of steps for reaching it, like board games, puzzles, or math worksheets; see Christopher & Farran, 2020), which is believed to support goal-based mastery of skills (Bronson, 1994) and has been linked to better outcomes in pre-k and kindergarten (Christopher & Farran, 2020).

Potential Sustaining Environments: Looking Beyond Instructional Quality

Emerging theory and research suggest that there are at least two other dimensions of children’s classroom experiences, particularly in kindergarten, that hold the potential to sustain the advantage in skills conferred by preschool attendance. An innovation of the current study is the inclusion of two of these dimensions, separate and apart from instructional quality, as moderators of the association between preschool attendance and child outcomes.

The Classroom Self-Regulatory Environment.

The classroom self-regulatory environment comprises the features of a classroom that support positive self-regulation, which in turn facilitates learning (McClelland & Cameron, 2012). These features include predictable routines and organization, detailed and inclusive instructional plans that keep children engaged in learning, and clear expectations of children’s behavior (Barnes et al., 2022; Pianta et al., 2012).

Many of these features are captured by the CLASS (Pianta, 2008) Classroom Organization subscale. This scale assesses the teacher’s behavior management techniques, transitions between activities, preparedness, and routines. Several studies have found that students in early education classrooms that score higher on Classroom Organization show better executive functioning (Choi et al., 2016; Hamre et al., 2014; Nguyen et al., 2020), language (Xu et al., 2014) and math (Downer et al., 2012) skills but others – including a recent meta-analysis – find no consistent relationship between Classroom Organization and children’s outcomes (Guerrera-Rosada et al., 2021; McDoniel et al., 2022; Moffett et al., 2021; Perlman et al., 2016). Previous studies testing the CLASS in kindergarten as a moderator of pre-k attendance have used a single score combining the Classroom Organization subscale with the Instructional Quality and Emotional Support subscales (e.g., Burchinal et al., 2022). The present study tests each one individually.

In addition to using the CLASS Classroom Organization subscale as a measure of self-regulatory environment, we include an alternative measure as well – one designed specifically to capture the classroom self-regulatory environment: specifically, we draw on items from the Adapted Teaching Style Rating Scale (ATSRS; Raver et al., 2012) designed to measure classroom structure and management. This version of the Classroom Structure and Management Scale has recently been found to predict children’s literacy and math skills at the end of kindergarten, even after controlling for baseline (fall of kindergarten) abilities (Martin et al., 2024).

To date, no studies have tested whether the classroom self-regulatory environment moderates associations between public preschool attendance and children’s outcomes, in kindergarten or beyond. We are the first to undertake this test, expanding the scope of our investigation into promising sustained environment factors beyond the instructional environment. We draw on both the CLASS Classroom Organization Scale as an existing measure of the self-regulatory environment, as well as the ATSRS as a new and promising measure.

The Classroom Relational Environment.

Another understudied potential moderator of the preschool boost is the relationship students have with their teachers. Substantial research suggests that student-teacher conflict and closeness predict children’s outcomes across a variety of domains. Specifically, the quality of the student-teacher relationship in early education has been found to predict children’s academic (e.g., Burchinal et al., 2002; Hamre & Pianta; 2001; Hernández et al., 2017; Maldonado-Carreño & Votruba-Drzal, 2011; McCormick et al, 2013; Pianta & Stuhlman, 2004) and self-regulatory growth (Berry, 2012; Cadima et al., 2016; de Wilde et al., 2016; Lee & Bierman, 2015; Li & Lau, 2019; McKinnon & Blair, 2019; Nguyen et al., 2020; Vitiello et al., 2022). Researchers have proposed that positive, high-quality relationships with teachers may provide students with a secure base that enables them to safely explore their environment and more effectively engage with learning opportunities in the classroom (Howes et al., 2000; Verschueren & Koomen, 2012).

However, no studies to date have explored the potential of teacher-student relationship quality – or what we term the classroom relational environment – to be a sustaining environment of the preschool boost. One study did examine whether a sustained pre-k impact from fall to spring of kindergarten was reduced by controlling for teacher-student relationships, and found that it was not, but this study did not test moderation of sustained pre-k effects (Ansari et al., 2020). To the extent that the relational environment is another aspect of the quality of children’s experiences in kindergarten classrooms, it may represent an as-yet-overlooked mechanism that moderates the extent to which advantages conferred by preschool attendance are sustained over time. We test this mechanism using the CLASS Emotional Support scale as well as with a direct measure of teacher-student relationship quality.

The Current Study

The current study pursues answers to one central research question: are positive associations between public preschool attendance and children’s language, literacy, and math skills in first grade larger for – or only evident among – children who experienced higher-quality kindergarten instructional, self-regulatory, and teacher-child relational environments? We conduct this test using data from an evaluation of Tulsa, OK’s renowned universal public preschool program, which has produced evidence of lasting positive associations between preschool attendance and outcomes beyond kindergarten (Johnson et al., 2023; 2024). This is important, as prior explanations for the null results of recent tests for sustaining environments have included a lack of main effects of preschool on outcomes past the end of kindergarten (Bailey et al., 2020b).

Additionally, we aim to capture a broader range of sustaining environments than have been explored in prior work. In addition to the instructional environment, we examine the roles of the self-regulatory and relational environments. Of particular value to the field is our ability to test not just newer measures, but also the most commonly relied upon tool used to measure classroom quality in early education, and the most widely-used measure of the sustaining environments in prior tests of moderation of the pre-k boost: the CLASS. For each dimension of the kindergarten classroom environment that we examine as a moderator of the association between preschool attendance and first-grade outcomes, we test both a new measure as well as the relevant CLASS subscale capturing analogous features of classrooms. With respect to child outcomes, the current study goes beyond limited measures of sustained skill advantages to assess associations with a wide assortment of language, literacy, and math skills including those tapping specific skills on which preschool benefits have been found to persist (e.g., “unconstrained” skills; see Johnson et al., 2023; McCormick et al., 2021). Finally, the children in our sample attended one of the nation’s two primary public preschool programs available to children from low-income households – Head Start and school-based pre-k – which together serve a majority of low-income 4-year-olds in the U.S. (Friedman-Krauss et al., 2022). Our sample, while exclusively low-income, is highly diverse in terms of child race/ethnicity and language status (e.g., nearly 80% Hispanic/Latinx or Black and more than 50% from Spanish-speaking households) and is thus reflective of the increasingly diverse population of young children in public preschool and elementary school settings in the U.S.

Method

Data Source and Sample

Data are drawn from the Tulsa School Experiences and Early Development (SEED) Study, an ongoing longitudinal study about the preschool and elementary school experiences of children from low-income households. Children in the Tulsa Public Schools (TPS) district who were low-income (family income below 185% of federal poverty level or received public benefits in the last year) were enrolled in the study at age 3 (2016; n = 611), age 4 (academic year 2017– 2018; n = 687), or kindergarten (academic year 2018– 2019; n = 130). Recruitment strategies differed slightly across waves to capture a range of preschool experiences. At age 3, children were recruited from publicly funded center-based programs feeding into TPS elementary schools that housed pre-k programs and served primarily low-income families (at least 80% of the student body qualified for free or reduced-price lunch). These programs included Head Start, which was administered by the largest Head Start grantee in Tulsa— the Community Action Program (CAP); all three of Tulsa’s Educare centers, which are birth-5 early intervention programs that align with Head Start standards and receive Early Head Start funds; and community-based child care centers serving child care subsidy recipients.

TPS’ universal school-based pre-k program begins at age 4; approximately 82% of Tulsa’s 4-year-olds attend preschool in a mix of TPS school-based pre-k classrooms and CAP Head Start classrooms. Most children who were in our 3-year-old sample remained in their age 3 arrangements in the 4-year-old year (what we refer to as the “preschool year”), some moved away or enrolled in private programs, and many enrolled in TPS school-based pre-k. In the preschool year, we recruited additional TPS pre-k attenders who were not already in our sample. Research staff went to all TPS elementary schools with pre-k classrooms to provide study information, recruit parents based on a screener questionnaire used to determine eligibility based on family income, and obtain parents’ consent to participate.

The following year, at kindergarten entry, we recruited one final group of children: those who had not attended TPS pre-k, CAP Head Start, Educare, or another center-based preschool in Tulsa the previous year. Candidate families were identified based on school records, and research staff sent recruitment materials home in backpacks and went to school events to meet parents in person. A parent-reported screener questionnaire collected information about children’s prior year care arrangements and household income. The University of Oklahoma - Tulsa IRB approved all study protocols.

This analysis draws on the 1093 children in Tulsa SEED who stayed in the TPS district for kindergarten. We excluded 37 children who had attended TPS pre-k, CAP Head Start, or Educare for fewer than 50% of the school days offered; 24 children who attended other center- based care settings; 27 children whose preschool arrangements were unknown; and 4 children who were cared for in a home by a non-relative. We also excluded 43 children who had entered the study at age 3 but by the preschool year had a family income that exceeded 185% of the federal poverty line, and another 145 children because they were not assessed in first grade (fall of 2019). Finally, we excluded 21 children whose kindergarten classrooms were not observed and whose kindergarten teachers did not rate the quality of their relationship with the child. The current study thus includes 792 children. During the kindergarten year, there were 6 study children on average per kindergarten classroom (SD = 3, Range = 1-14). The following year, there were 5 study children on average per first grade classroom (SD = 3, Range = 1-17). Compared to those dropped from the analytic sample, children in the analytic sample were more likely to be Hispanic/Latinx, less likely to be Black, non-Hispanic or White, non-Hispanic, and more likely to be dual language learners. Children included in the analytic sample also had less educated parents, mothers who were more likely to have been married at their birth, and larger households with lower monthly incomes.

The overall sample of 792 children (Table 1) includes 717 children (91%) who attended public preschool via TPS pre-k (76%) and Head Start or Educare (15%), and 75 children (9%) who were preschool “non-attenders,” meaning they stayed at home with a parent or relative during the preschool year. The relatively small sample of preschool non-attenders potentially underpowers our analyses. Children in the analytic sample were racially and ethnically diverse: 51% were Hispanic/Latinx; 20% were Black, non-Hispanic; 12% were White, non-Hispanic; 9% were multiracial; 7% were American Indian or Alaskan Native; 1% were Asian American/Pacific Islander; and <1% belonged to another racial/ethnic group. Roughly half (49%) of children were dual language learners, meaning they lived in a household where a language other than English was spoken. Half (50%) of mothers were unmarried at the time of the child’s birth, and households had 4–5 members on average. Most mothers (65%) were employed during the child’s preschool year, and 38% had some education beyond high school/GED. On average, families in the sample earned $21,354 per year, which is less than the federal poverty level for a family of four in 2018 ($25,000).

Table 1.

Sample Descriptive Statistics

	Preschool Attenders			Preschool Non-Attenders
	M/%	SD	n	M/%	SD	n
Child Characteristics
Age in months at fall of preschool year	54.81	3.66	717	55.14	3.80	75
Female*	51%	—	717	36%	—	75
Child’s race/ethnicity
Hispanic/Latinx***	53%	—	717	32%	—	75
Black, non-Hispanic**	21%	—	717	8%	—	75
White, non-Hispanic*	12%	—	717	20%	—	75
Other race***	14%	—	717	40%	—	75
Dual language learner	50%	—	717	39%	—	75
Parent/Household Characteristics
Mother’s age at child’s birth*	26.24	5.97	597	24.49	5.35	53
Mother was single at child’s birth	49%	—	604	61%	—	51
Parent had more than a HS education	37%	—	611	40%	—	57
Parent is employed (full or part time)	66%	—	496	.57	—	53
Household size	4.78	1.65	673	4.60	1.83	55
Monthly household income ($)	1,782	993	580	1,747	1032	51
Kindergarten Sustaining Environments
COPG/TOPG instruction*	0.00	0.58	601	−0.17	0.44	65
ATSRS self-regulatory quality**	3.60	1.03	632	3.20	0.95	70
STRS teacher-child relationships	4.30	0.67	611	4.25	0.57	63
First Grade Academic Outcomes
Letter-word identification*	394.98	36.17	713	383.64	42.24	73
Phonological awareness***	15.10	5.9	716	12.55	6.43	74
Expressive vocabulary	19.48	9.46	716	17.49	10.8	75
Sentence structure***	20.22	4.29	716	17.59	5.74	75
Applied problems***	440.18	17.58	713	431.12	25.90	73
Numerical fluency**	17.04	8.79	712	13.63	11.70	75
Conceptual counting**	0.74	0.24	714	0.65	0.27	73

Notes. Descriptive statistics are not imputed data. HS = High school.

^*p < .05;

^**p < .01;

^***p < .001

Procedures

Information about family and household demographic characteristics used as covariates in the current analysis was drawn from a survey parents completed in the spring of the preschool year (2018). For children who were recruited in kindergarten, this information was drawn from the survey their parents completed in kindergarten but refers to their status as of the previous year. Surveys were distributed via text message, email, and in children’s backpacks; parents received a $30 gift card for completing this survey. Observations of children’s kindergarten classrooms were conducted in early 2019 by a team of coders, one of whom scored the CLASS and the other who scored the other observational measures (ATSRS and COPG/TOPG). Each observation began at approximately 8:30 am and lasted for about 4.5 hours. Observers included both English monolingual and Spanish-English bilingual professionals. Most observers held a degree in child development, education, psychology, or a related field. Also in the spring of kindergarten (2019), kindergarten teachers were sent an anonymous Qualtrics survey link for each participating child in their classroom and asked to rate the quality of their relationship with each child. Each survey took approximately 6 minutes to complete and teachers received $20-$60 in gift cards as compensation, depending on the number of participating children in their classroom. To measure developmental outcomes in the fall of first grade (fall 2019), children were assessed by trained assessors in a quiet area free of distraction; if a child was a Spanish-speaking dual language learner, their assessor was bilingual in English and Spanish to ensure children could comfortably communicate with the assessor and fully participate in the assessment regardless of their English skills. Assessments occurred over 2 days to reduce child fatigue.

Measures

Preschool Attender Status

School and program administrative records, cross-referenced with parent report when appropriate, were used to identify preschool attenders and non-attenders based on their preschool arrangement during the preschool year. Children were coded as preschool attenders if they attended a TPS public school or TPS-affiliated charter school-based pre-k program or a CAP-Tulsa Head Start or Educare program for more than 50% of the offered days that school year. Children were coded as preschool non-attenders if they stayed home with a parent or relative during their preschool year.

Kindergarten Classroom Quality

Classroom Instructional Environment: COPG/TOPG Instruction.

The kindergarten classroom instructional environment was assessed during observations using the Child Observation in Primary Grades and Teacher Observation in Primary Grades (COPG/TOPG; Peabody Research Institute, 2017a, 2017b), which focus the observer’s attention on the behavior of individual children in the classroom as well as the classroom teacher. Both measures are based on a series of 3-second snapshots or “sweeps” across each individual child in the classroom as well as the lead teacher. After each sweep, the observer records detailed information about the person being observed, including their current academic focus, what tasks or activities they were engaged in, their interaction state (e.g., to whom, if anyone, they were speaking or listening), and—for teachers—their tone and level of instruction. The COPG collects data on every child in the classroom (not just children participating in the study), and scores are then averaged across sweeps and aggregated to the classroom level. The TOPG records data on the teacher, and scores are averaged across sweeps. Observers were trained on the COPG/TOPG during a 2-day in-person training by the tool’s developers at Vanderbilt University. They were certified as reliable if they reached an average of 80% agreement with a gold standard coder across all dimensions and counts, with no single dimension or count falling below 70% agreement. After each observation, gold-standard coders reviewed all codes and met with the observer in cases of disagreement. Weekly meetings were also held to reach consensus on how to code new situations as they arose.

The current study selected items from the COPG/TOPG to capture features of the instructional environment that have shown the most promise in predicting child outcomes in recent research. These include items that capture time spent in instruction and the content and level of that instruction, both delivered by the teacher and received by the child (e.g., Cheung & McBride, 2017; Christopher & Farran, 2020; Claessens et al., 2014; McCormick et al., 2022). Specifically, we selected three items from the COPG: the average proportion of sweeps where children’s academic focus was math; the average proportion of sweeps where children’s academic focus was English, science, or social studies; and the average proportion of sweeps where children were engaged in sequential activities—defined as activities or tasks that have a predetermined sequence of steps, such as completing a puzzle, reading a book, or completing a worksheet. We selected an additional three items from the TOPG: the proportion of sweeps where the teacher was engaged in instruction; the proportion of sweeps where the teacher’s academic focus was math; and the teacher’s average level of instruction on a scale from 1-4 (1 = low, 4 = high inferential learning) during sweeps where they were engaged in instruction. To enable the combination of items on different metrics (proportions vs. 1-4 scale), all items were standardized and averaged to construct a composite of the kindergarten classroom instructional environment (α= .61). We experimented with a reduced version of the COPG/TOPG that prioritized instruction and did not include TOPG proportion of time on instruction or math, but the alpha was reduced (α= .30) and results were unchanged (available upon request).

Classroom Instructional Environment: CLASS Instructional Support.

The CLASS (Pianta et al., 2008) Instructional Support subscale was also used to capture instructional quality. Trained classroom observers assessed teacher-child interactions across four 15-minute cycles of observations, generating three domains of teacher-child interaction quality (instructional support; classroom organization; emotional support) on a 7-point rating scale (1 = low to 7 = high). Following standard practice, we constructed one score for each domain by averaging its component items. The Instructional Support subscale is based on 14 items which feed into three dimensions capturing the quality of teachers’ feedback to students, techniques used to develop learning concepts, and teachers’ modeling of language and higher-order thinking skills (α = .91). All CLASS observers were certified through a 2-day training with a TeachStone certified trainer, using TeachStone’s video training process, which includes self-study of materials, discussion, and practice scoring. Ongoing reliability was maintained by a gold-standard coder who reviewed every set of scores and supporting notes from each live observation; for disagreements, gold-standard coders and observers met to come to consensus on the score. There were no situations in which observers could not provide sufficient evidence to reach consensus with a gold-standard coder.

Classroom Self-Regulatory Environment: ATSRS Self-Regulatory Quality.

The kindergarten classroom self-regulatory environment was assessed via live observation using the Adapted Teaching Style Rating Scale (ATSRS; Barnes et al., 2022; Raver et al., 2012). Observers were trained and certified by one of the ATSRS’ authors. They were certified as reliable if they scored within one point of a gold-standard coder on 80% of items across two live practice visits. After every classroom visit, each observer’s scores and detailed notes supporting the scores were reviewed by a gold-standard coder to ensure that supporting evidence for the scores matched the ATSRS manual and codebook. Any score disagreements were resolved through discussion within the pair to reach consensus. Protocols are described in more detail elsewhere (see Johnson et al., 2021).

The ATSRS rated the quantity and quality of 13 teacher practices related to classroom structure and management and support for social-emotional skills using a 5-point scale (1 = low, 5 = high), with item-specific descriptors for each level. The current study drew on 4 items: Consistency/Routine (providing clear rules, routines, and expectations), Preparedness (clear, detailed instructional plans keep children engaged), Classroom Awareness (awareness of the classroom environment and reliance on proactive behavior management), and Attention Support (different gestures, cues, and signals maintain students’ attention). These items were averaged (α = .93) into a single scale which has recently been shown to predict improvements in children’s math and literacy skills during the kindergarten year (Martin et al., 2024). We conceptualize this as a measure of the quality of the kindergarten classroom self-regulatory environment and consider it as a possible moderator of the association between public preschool attendance and children’s academic outcomes in first grade.

Classroom Self-Regulatory Environment: CLASS Classroom Organization.

We also used the CLASS Classroom Organization subscale to capture the self-regulatory environment. The Classroom Organization subscale is based on 12 items which feed into three dimensions capturing teachers’ management of children’s behavior, teachers’ productivity, and teachers’ use of instructional learning formats to maximize child learning through classroom routines (α= .86).

Classroom Relational Environment: STRS Teacher-Child Relationships.

The quality of the relational environment experienced by each child in kindergarten was assessed using the Student-Teacher Relationship Scale (STRS; Pianta, 2001). Kindergarten teachers endorsed 15 statements regarding their relationship with the child (e.g., “I share an affectionate, warm relationship with this child”) using a 5-point scale (1 = definitely does not apply, 5 = definitely applies). The seven negatively valanced items capturing student-teacher conflict (e.g., “This child and I always seem to be struggling with each other”) were reverse coded and averaged alongside eight items capturing student-teacher closeness (α = .89). We refer to this STRS total positive relationship scale as a measure of the relational environment.

Classroom Relational Environment: CLASS Emotional Support.

Additionally, as with the other indicators of classroom quality, we also assessed the kindergarten relational environment using an analogous CLASS subscale: Emotional Support. The Emotional Support subscale is based on 16 items which feed into four dimensions capturing positive and negative emotional climate, teacher sensitivity, and teachers’ regard for student perspectives (α = .89).

First Grade Outcomes

Letter-Word Identification.

Letter-word identification skills were assessed using the Woodcock-Johnson Letter-Word Identification subtest (Woodcock et al., 2001), which asks children to accurately identify letters and correctly pronounce sight words. Only English answers were accepted, but bilingual children were prompted once per item if they responded in Spanish. Our analyses used W scores, a Rasch transformation of raw scores that adjusts for the difficulty of individual items.

Phonological Awareness.

We administered the Clinical Evaluation of Language Fundamentals— Phonological Awareness supplement (CELF-PA; Semel et al., 2003) to assess syllable blending, syllable segmentation, rhyme detection, phoneme identification, and phoneme blending. Syllable blending captures children’s abilities to identify a complete word after the assessor reads the word segmented by syllables (e.g., assessor says “ta,” pauses, and says “ble”; child responds “table”). Syllable segmentation captures children’s ability to identify the syllables in a word, for example by clapping twice as the assessor reads “sweatshirt.” Rhyme detection refers to children’s ability to identify if two words rhyme. Phoneme identification captures children’s abilities to identify the sound— not the letter—at the start of the word. For example, if the assessor says “toy” and the child responds with a “t” sound, that would be correct but pronouncing the letter like “tee” would be incorrect. Finally, phoneme blending captures children’s ability to blend sounds into a word. The assessor says a word slowly, separating each phoneme like “f- i- n” and the child must provide the entire word (“fin”). Each skill test included a trial and five items; children received a point for each correct item. Within each skill score, children who failed trials were assigned a score of 0. We used raw total scores that summed scores across all five skills (range: 0–25; α= .64).

Expressive Vocabulary.

The CELF— Expressive Vocabulary (CELF- EV) assesses expressive language by asking children to label pictures (e.g., a tree branch, with target responses of branch, tree limb, or limb). Only English answers were accepted, but bilingual children were prompted once per item if they responded in Spanish. Raw scores are used in all analyses; for children who failed trial items, we assigned a score of 0. For children in our study’s age range, the reported reliability coefficients for internal consistency are r = .85– .91 (Semel et al., 2003).

Sentence Structure.

The CELF Sentence Structure (CELF-SS) subtest gauged children’s understanding of both syntax (grammar) and semantics (meaning). Here, the assessor reads the child a sentence and the child is asked to point to an image that corresponds to the sentence (e.g., “Point to the girl who is standing in the front of the line wearing a backpack”). Raw scores were used; for children who failed trial items, we assigned a score of 0. For children in our study’s age range, the reported reliability coefficients for internal consistency are r = .62– .77 (Semel et al., 2003).

Applied Problems.

The Woodcock-Johnson Applied Problems subtest (Woodcock et al., 2001) measures children’s mathematical problem-solving ability by asking them to solve computational word problems (e.g., “how many dogs are in this picture?”), and to perform basic addition and subtraction. Our analyses used W scores.

Numerical Fluency.

Numerical fluency was measured with the Symbolic Numeral Comparison subtest from the Numeracy Screener (Nosworthy et al., 2013), a paper-pencil task adapted by Lyons et al. (2018). This task measures the efficiency with which children can access the underlying meaning of number symbols. Each item in this task consisted of two symbolic numerals (1– 9) presented side-by-side; 48 items (12 per page) were presented. Participants were told, “In this task, your job is to decide which of the two numbers is bigger. Mark the box with the number that means the most things.” Children completed as many items as they could within 1 minute. Scores were calculated as the number of correct responses minus incorrect responses to adjust for guessing.

Conceptual Counting.

Concepts underpinning counting were captured using the modified short-form Research-Based Early Mathematics Assessments (SF-REMA), based on an instrument designed by Johnson et al. (2019), which is itself an adapted version of the REMA- SF (Weiland et al., 2012). Children were given 31 unorganized pennies and asked to count them. The child was scored on knowledge of 1:1 correspondence (i.e., that each item should only be counted once) by the assessor for pennies separately in sets of 1–5, 6–10, 11–20, and 21–31. The total possible score ranged from 0% (never followed 1:1 correspondence) to 100% (followed 1:1 correspondence for all four sets of numbers). The child was considered to have knowledge of the cardinality principle (i.e., that when counting, the last number mentioned represents the total number in the set) if the child answered correctly when the assessor asked “How many?” after the child finished counting. The child was scored on knowledge of stable order according to the highest number counted to without any errors, expressed as a percentage of the total possible score. These three items were averaged.

Covariates

Child race/ethnicity was coded as Hispanic/Latinx, Black, White, or multiracial/another race (which, due to small sizes, combined children who were Asian American/Pacific Islander, Native American, or multiracial). Child gender was coded as male or female. The child was coded as a dual language learner if their household spoke a language besides English at home. We also included binary indicators for maternal education (e.g., whether the mother had any post- secondary education), maternal employment status, and whether the mother was unmarried at child’s birth. The log of household income was used in all analyses. Other covariates included household size, mother’s age at child’s birth, and child age in months at the date of the respective assessment.

Analytic Strategy

The current study was designed to test whether associations between public preschool attendance and children’s language, literacy, and math skills in first grade vary by key dimensions of the kindergarten classroom environment: the instructional environment, the self-regulatory environment, and the teacher-child relationship (relational) environment. To accomplish this, we used the “mixed” command in Stata 17 to estimate 2-level random intercept models (Level 1: Child; Level 2: Kindergarten Classroom) predicting each of the 4 language/literacy and 3 math outcomes measured in first grade; separate models were estimated for each outcome. For each outcome, we tested each measure of the kindergarten environment (instructional, self-regulatory, and relational) separately, in a stepwise fashion. Our primary models included an indicator variable capturing public preschool attendance, the respective measure of the kindergarten classroom environment, and an interaction between the two. (Supplementary models tested for main effects of preschool attendance and main effects of the separate measures of the kindergarten classroom environment; see Appendix A). Models were estimated separately for each measure of environmental quality (e.g., once for the ATSRS self-regulatory quality scale and once for the CLASS Classroom Organization scale as measures of the Self-Regulatory Environment). Coefficients from all models can be interpreted as effect sizes as all continuous predictors were standardized. We note that these models are correlational and thus results should be interpreted as such, and not as causal estimates.

Finally, we used multiple imputation to address missing data on covariates in the current study (we did not impute missing data for moderators or dependent variables, and thus sample sizes vary by model and are denoted in the tables accordingly). We used the ice command in Stata to impute 25 data sets via a series of chained, multivariate normal regressions. Estimates and standard errors are combined via the mi estimate command and averaged across data sets using Rubin’s Rules.

Results

Table 1 presents descriptive statistics for all covariates, measures of the kindergarten environment, and first-grade outcomes, disaggregated by preschool attender status. Preschool attenders differed from non-attenders in several ways: preschool attenders were more likely to be female, Hispanic/Latinx, or Black, and less likely to be White or an “other” race. We adjust for covariate differences in all of our regression models, which reduces but does not eliminate the possibility that these background differences bias observed associations between preschool attendance and first-grade outcomes. Preschool attenders also experienced slightly higher-quality kindergarten instructional and self-regulatory classroom environments than non-attenders, but there was no difference in the relational environment; they also outscored non-attenders on all first-grade outcomes, with the exception of expressive vocabulary.

We also examined correlations between preschool attendance, multiple measures of the kindergarten classroom instructional, self-regulatory, and relational environment, and child outcomes in first grade (Appendix Table 1). Preschool attendance was weakly correlated with all measures of the kindergarten classroom environment (rs ranging from .09 to .17), except for the relational environment as measured using the STRS (r = .03), confirming the appropriateness of tests of moderation. Our new measure of the instructional environment was moderately correlated with our new measure of the self-regulatory environment (r = .51) and moderately correlated with the CLASS Classroom Organization (CO) subscale (r = .35), but only weakly correlated with the CLASS Instructional Support (IS; r = .08) and Emotional Support (ES; r = .10) subscales. Our new measure of the self-regulatory environment was weakly correlated with the STRS-based measure of the relational environment (r = .09), moderately correlated with the CLASS IS (r = .39) and ES (r = .46) subscales, and more strongly correlated with the CLASS CO subscale (r = .65). The relational environment as measured using the STRS was not correlated with our new measure of the instructional environment or any of the CLASS subscales (rs range from .02 to .08). Finally, all child outcomes were significantly correlated, with the weakest correlation emerging between expressive vocabulary and conceptual counting (r = .29) and the strongest correlation emerging between sentence structure and applied problems (r = .64).

Tables 2–4 present results of our primary regression models, which tested whether associations between public preschool attendance and children’s language, literacy, and math skills in first grade vary by key dimensions of the kindergarten classroom reflecting the quality of the instructional environment, the self-regulatory environment, and the teacher-child relational environment, respectively. (Results of main effects models appear in Appendix Tables 2–4). As a reminder, each outcome was predicted in a separate regression model that controlled for all covariates and tested moderation of preschool attendance by measures of each dimension of the kindergarten classroom environment; we first tested our new measure of the classroom environment (i.e. COPG/TOPG instruction; ATSRS self-regulatory quality; STRS student-teacher relationship quality), and then repeated the same models but with the analogous CLASS subscale (Instructional Support; Classroom Organization; Emotional Support). Because our models include interactions between a binary (preschool attendance) and continuous (kindergarten classroom environment) variable, and because our measures of the kindergarten classroom environment are standardized, the coefficients for preschool attendance can be interpreted as the association between preschool attendance and child outcomes for those in kindergarten classrooms with average (mean=0) levels of the respective classroom environment variable. The coefficients for each kindergarten classroom environment variable can be interpreted as the association between that measure of the classroom environment and child outcomes for preschool non-attenders (preschool attendance = 0).

Table 2.

Associations between preschool attendance and child outcomes moderated by the kindergarten classroom instructional environment

	Letter-word ID		Phonological awareness		Expressive vocabulary		Sentence structure		Applied problems		Numerical fluency		Conceptual counting
	B	SE	B	SE	B	SE	B	SE	B	SE	B	SE	B	SE
COPG/TOPG instruction
PK	0.22	0.13	0.48***	0.13	0.30*	0.13	0.63***	0.13	0.57***	0.13	0.27*	0.14	0.32*	0.14
COPG/TOPG	0.21	0.15	−0.08	0.15	0.08	0.14	−0.1	0.15	0.06	0.15	0.04	0.15	0.22	0.16
PK x COPG/TOPG	−0.02	0.15	0.14	0.15	−0.04	0.15	0.12	0.15	0	0.15	0.09	0.16	−0.15	0.16
n	661		665		666		666		661		662		660
CLASS Instructional Support
PK	0.25	0.13	0.54***	0.13	0.40**	0.12	0.67***	0.13	0.70***	0.13	0.31*	0.13	0.39**	0.14
CLASS IS	0.16	0.15	−0.11	0.14	−0.15	0.14	−0.04	0.14	−0.25	0.15	0.2	0.15	0.14	0.16
PK x CLASS IS	−0.15	0.15	0.18	0.15	0.19	0.14	0.02	0.14	0.26	0.15	−0.24	0.15	−0.18	0.16
n	694		698		699		699		694		696		693

Notes. All models control for child race/ethnicity, gender, age at date of assessment, dual language learner status, mother’s age and marital status at child’s birth, parent education, employment status, household size, and natural log of household income during the child’s 4-year-old year. Continuous predictors and dependent variables are standardized. Models include random intercepts to account for the nesting of children in kindergarten classrooms. Missing covariates are multiply imputed. PK = Public preschool. COPG = Child Observation in Primary Grades. TOPG = Teacher Observation in Primary Grades. IS = Instructional Support.

^*p < .05;

^**p < .01;

^***p < .001

Table 4.

Associations between preschool attendance and child outcomes moderated by the kindergarten classroom relational environment

	Letter-word ID		Phonological awareness		Expressive vocabulary		Sentence structure		Applied problems		Numerical fluency		Conceptual counting
	B	SE	B	SE	B	SE	B	SE	B	SE	B	SE	B	SE
STRS teacher-child relationships
PK	0.22	0.13	0.41**	0.13	0.36**	0.12	0.63***	0.13	0.52***	0.13	0.21	0.13	0.41**	0.13
STRS	0.27	0.14	0.23	0.15	0.08	0.13	0.21	0.13	0.26	0.15	0.58***	0.14	0.16	0.15
PK x STRS	−0.15	0.15	−0.1	0.15	0.03	0.14	−0.1	0.14	−0.18	0.15	−0.51***	0.14	−0.03	0.15
n	668		671		672		672		668		668		668
CLASS Emotional Support
PK	0.32*	0.13	0.56***	0.13	0.42***	0.13	0.71***	0.13	0.73***	0.13	0.38**	0.14	0.46***	0.14
CLASS ES	−0.05	0.11	−0.1	0.1	−0.13	0.09	−0.12	0.1	−0.22*	0.1	−0.03	0.1	−0.06	0.11
PK x CLASS ES	0.06	0.11	0.1	0.1	0.11	0.1	0.1	0.1	0.17	0.1	−0.06	0.11	0.04	0.11
n	694		698		699		699		694		696		693

Notes. All models control for child race/ethnicity, gender, age at date of assessment, dual language learner status, mother’s age and marital status at child’s birth, parent education, employment status, household size, and natural log of household income during the child’s 4-year-old year. Continuous predictors and dependent variables are standardized. Models include random intercepts to account for the nesting of children in kindergarten classrooms. Missing covariates are multiply imputed. PK = Public preschool. STRS = Student-Teacher Relationship Scale. ES = Emotional Support.

^*p < .05;

^**p < .01;

^***p < .001

Table 2 displays results testing the instructional environment measured two different ways as a moderator of preschool attendance. First, when using our new measure of COPG/TOPG instruction, preschool attendance was a significant predictor of phonological awareness (β = .48, SE = .13, p < .001), expressive vocabulary (β = .30, SE = .13, p < .05) sentence structure (β = .63, SE = .13, p < .001), applied problems (β = .57, SE = .13, p < .001), numerical fluency (β = .27, SE = .14, p < .05), and conceptual counting (β = .32, SE = .14, p < .05). Preschool attendance was not associated with letter-word ID. COPG/TOPG instruction was not significantly associated with child outcomes. The interaction terms were also never significant. Patterns were similar for CLASS Instructional Support (IS) as an alternative measure of the instructional environment. CLASS IS was never associated with any first-grade outcomes, and it did not moderate associations between preschool attendance and child outcomes.

Table 3 presents results testing the self-regulatory environment as a moderator. Using ATSRS self-regulatory quality, preschool attendance was significantly associated with letter-word ID (β = .31, SE =.13, p < .05), phonological awareness (β = .54, SE = .13, p < .001), expressive vocabulary (β = .38, SE =.12, p < .01), sentence structure (β = .66, SE = .13, p < .001), applied problems (β = .67, SE = .13, p < .001), numerical fluency (β = .36, SE = .13, p < .01), and conceptual counting (β = .40, SE = .14, p < .01). ATSRS self-regulatory quality was not associated with any child outcomes and none of the interaction terms were significant. Results were similar for the CLASS Classroom Organization (CO) scale. The CLASS CO scale did not predict child outcomes, and it did not moderate associations between preschool attendance and outcomes.

Table 3.

Associations between preschool attendance and child outcomes moderated by the kindergarten classroom self-regulatory environment

	Letter-word ID		Phonological awareness		Expressive vocabulary		Sentence structure		Applied problems		Numerical fluency		Conceptual counting
	B	SE	B	SE	B	SE	B	SE	B	SE	B	SE	B	SE
ATSRS self-regulatory quality
PK	0.31*	0.13	0.54***	0.13	0.38**	0.12	0.66***	0.13	0.67***	0.13	0.36**	0.13	0.40**	0.14
ATSRS	0	0.13	−0.11	0.12	−0.11	0.12	−0.08	0.12	−0.13	0.12	0.03	0.13	0.1	0.13
PK x ATSRS	0.14	0.13	0.21	0.13	0.14	0.12	0.11	0.13	0.19	0.13	0.01	0.13	−0.04	0.13
n	697		701		702		702		697		698		696
CLASS Classroom Organization
PK	0.28*	0.14	0.52***	0.13	0.37**	0.13	0.64***	0.13	0.66***	0.13	0.34*	0.14	0.41**	0.14
CLASS CO	0.09	0.12	0	0.11	−0.02	0.1	0.02	0.1	−0.06	0.11	0.06	0.11	0.05	0.12
PK x CLASS CO	0.04	0.12	0.09	0.11	0.04	0.11	0.01	0.11	0.1	0.12	−0.05	0.12	−0.02	0.12
n	694		698		699		699		694		696		693

Notes. All models control for child race/ethnicity, gender, age at date of assessment, dual language learner status, mother’s age and marital status at child’s birth, parent education, employment status, household size, and natural log of household income during the child’s 4-year-old year. Continuous predictors and dependent variables are standardized. Models include random intercepts to account for the nesting of children in kindergarten classrooms. Missing covariates are multiply imputed. PK = Public preschool. ATSRS = Adapted Teaching Style Rating Scale. CO = Classroom Organization.

^*p < .05;

^**p < .01;

^***p < .001

Table 4 presents results testing STRS teacher-child relationships as a moderator. When capturing the relational environment with the STRS teacher-child relationships scale, preschool attendance was significantly associated with phonological awareness (β = .41, SE = .13, p < .01), expressive vocabulary (β = .36, SE = .12, p < .01), sentence structure (β = .63, SE = .13, p < .001), applied problems (β = .52, SE = .13, p < .001), and conceptual counting (β = .41, SE = .13, p < .01). STRS teacher-child relationships was associated with numerical fluency (β =.58, SE = .14, p < .001), but no other outcomes. There was also a significant negative interaction between preschool attender status and STRS teacher-child relationships, indicating that a higher quality relational environment was less advantageous for preschool attenders than non-attenders (β = −.51, SE = .14, p < .001). We then repeated these analyses, replacing STRS teacher-child relationships with the CLASS Emotional Support (ES) subscale. The CLASS ES scale was negatively associated with applied problems (β = −.22, SE = .10, p < .05). However, it was not associated with any other child outcomes, nor did it moderate associations between preschool attendance and outcomes.

Supplemental Analyses

We sought to test the robustness of our results to an additional specification of the relational environment (Appendix Table 5). Some studies have conceptualized student-teacher relationship quality not as a total score on the STRS as we did, but as just the subset of items measuring teacher-child closeness (Burchinal et al., 2002; Cadima et al., 2016; Howes et al., 2008; Lee & Bierman, 2015; Vitiello et al., 2022). We therefore re-estimated our models replacing the STRS total relationship score with the Closeness subscale (α = .86). The results were unchanged with respect to moderation – our main research question – but the Closeness subscale was predictive of phonological awareness, sentence structure, applied problems, and numerical fluency (whereas the total score was only predictive of numerical fluency). We then repeated this using the Conflict subscale (α = .91). As with the STRS total score, conflict only predicted numerical fluency (β = −.33, SE = .13, p < .01). There was also a significant interaction between preschool attendance and student-teacher conflict predicting this outcome, but this time preschool was protective, such that conflict was less disadvantageous for preschool attenders than non-attenders (β = .31, SE = .13, p < .05).

Discussion

The purpose of this study was to test for interactions between public preschool attendance and three dimensions of kindergarten classroom environments that are theoretically promising features of a sustaining environment. We asked whether in first grade, children showed a stronger advantage of having attended preschool if they experienced higher quality classroom environments in kindergarten. We operationalized kindergarten classroom quality more broadly than has been done in most tests of the sustaining environments hypothesis to date, examining the instructional, self-regulatory, and relational environments of kindergarten classrooms separately using both newer measures as well as the most commonly used measure in the field, the CLASS. We found significant associations between preschool attendance and first-grade academic outcomes at average levels of environmental quality. However, all but one of the interactions between preschool attendance and kindergarten classroom quality were null, suggesting the absence of moderation. While disappointing, these results are not surprising, as much recent research has also failed to find evidence of such moderation, albeit without distinguishing instructional quality, self-regulatory quality, and relational quality (Ansari et al., 2023; Burchinal et al., 2022; McCormick et al., 2022).

Robust Associations Between Preschool Attendance and First Grade Outcomes

Although we had hoped that our use of new and more nuanced measures of what might constitute a sustaining environment in kindergarten would illuminate the dimensions of subsequent classroom experiences that help to sustain preschool benefits, we did not find evidence to support this. One encouraging interpretation of these null findings in our and other studies is that it appears that sustained positive associations between preschool attendance and first-grade skills are not dependent on the quality of the environment students experience in their kindergarten year. The findings in this study in particular add to a large body of research on the robustness of public preschool benefits in Tulsa. Several studies now show that participation in this much-examined, long-standing universal program is associated with lasting skill advantages in first grade (Johnson et al., 2023), third grade (Johnson et al., 2024) and – in an earlier cohort of preschool graduates – even through middle and high school (e.g., Amadon et al., 2022; Gormley et al., 2018). The findings are also consistent with past evidence that all students are helped by higher-quality kindergarten classroom environments, regardless of whether or not they attended public preschool (Burchinal et al., 2022).

It is noteworthy that of the three features of the kindergarten classroom environment, the relational environment emerged as the only predictor of first-grade outcomes, and even then, only for math. The STRS teacher-child relationship scale was positively associated with numerical fluency scores, but unexpectedly, the CLASS ES scale was negatively associated with applied problem scores. Accumulating research suggests the central role played by teacher-student relationship quality for early academic development (Pianta & Stuhlman, 2004). Indeed, an earlier analysis of the current sample showed that when considered together, teacher-student relationship quality but not the classroom self-regulatory environment predicted children’s growth in self-regulatory skills over the kindergarten year (Wright et al., 2024).

Interestingly, the one interaction between preschool attendance and the kindergarten environment applied to the relational environment, and not in the expected direction: a higher-quality relational environment (better STRS teacher-child relationships) predicted higher numerical fluency for preschool non-attenders, not preschool attenders. This finding suggests the possibility that rather than sustain initial advantages conferred by having attended preschool, better relational quality can help narrow gaps between preschool attenders and non-attenders by being more beneficial for children who did not attend preschool. On the other hand, supplementary analyses looking at student-teacher conflict uncovered an opposite finding: conflictual teacher-child relationships were disadvantageous only for non-attenders’ numerical fluency scores. Together, these results suggest the possibility that preschool non-attenders’ first grade numerical fluency is more affected than attenders’ by the relationship they form with their kindergarten teacher, but why this vulnerability should apply only to this outcome is unclear. There is a clear need for further research on whether and how preschool attendance and kindergarten student-teacher quality contribute to emerging math skills.

Our novel measures of instructional and self-regulatory quality were no more strongly associated with first-grade outcomes than the CLASS scales. This is the first use of COPG/TOPG instruction as a measure of instructional quality, but Martin et al. (2024) found in a previous analysis of this sample that the ATSRS self-regulatory environment scale was modestly predictive of children’s growth in literacy and math over the kindergarten year. Perhaps this association does not apply to outcomes measured in first grade. It is also possible that the measure itself does not adequately capture the self-regulatory environment. Yet in our data, neither did the CLASS CO. Newly developed tools designed for greater precision than the CLASS CO subscale are actively being tested in multiple studies around the U.S., but thus far results have been inconsistent. For instance, Moffett et al. (2021) used the Individualizing Student Instruction (ISI) Coding System (Connor et al., 2009) to capture how much time each individual student was exposed to the teacher’s organization strategies but the measure did not predict growth in children’s academic skills. Montoya et al. (2023) recently created observational scales of teachers’ scaffolding of self-regulatory skills in Chilean preschool classrooms, but have not yet tested for associations with children’s outcomes. Clearly, identifying dimensions of the self-regulatory classroom environment that consistently predict outcomes is an ongoing challenge for the field.

Results Do Not Help Explain Variation in Sustained Benefits Across Preschool Programs

Despite consistently robust associations between public preschool attendance and children’s sustained first-grade outcomes, there was no evidence of moderation by kindergarten classroom variables. In sum, this is arguably the most thorough test – and rejection – of the sustaining environments hypothesis to date. The findings are consistent with those from tests of this hypothesis with other preschool programs (Ansari et al., 2024; Burchinal et al., 2022; Carr et al., 2024; McCormick et al., 2022).

One possibility is that the tests of the sustaining environments hypothesis to date – including the current study – have relied on moderation, whereas perhaps the proper test of the sustaining environments hypothesis is one of mediation. That is, preschool attenders may be more likely than non-attenders to be assigned to higher quality classrooms, or to improve the quality of the classrooms they enter, which would in turn perpetuate their preschool advantage, even if non-attenders benefited equally from high-quality environments. To date, however, the sustaining environments hypothesis has not been operationalized to capture such processes.

Limitations

This study has some limitations which should be considered alongside its contributions. First, as stated above, this is an exploratory, descriptive study, and thus all results should be interpreted as correlational associations. Children who attended public preschool in Tulsa not only differed on select demographics from non-attenders, but also experienced higher-quality kindergarten instructional and self-regulatory environments. It is possible that preschool graduates were, on average, assigned to better teachers than other children, or that the neighborhoods they lived in were both more likely to provide preschool and have higher-quality kindergarten programs. Even so, it is difficult to see how such selection would have biased tests of the moderation of preschool impacts by kindergarten classroom quality, as in our data attenders and non-attenders experienced mostly overlapping ranges of quality. It is thus unclear how differential selection of preschool attenders into higher kindergarten classroom quality would have contributed to its non-significant interaction with preschool attendance, but we cannot rule out the possibility that it did. We also cannot dismiss the possibility that there are unmeasured background differences between preschool attenders and non-attenders that could have suppressed an interaction between preschool attendance and kindergarten quality.

Second, our comparison group of non-attenders was small, relative to preschool attenders. While this may have underpowered our analyses, reducing our ability to detect significant effects if effects were there, it reflects the reality of preschool attendance in a city where public preschool has long been universally available. Tulsa’s public preschool program serves approximately 70- 80% of low-income 4-year-olds through school-based pre-k, Head Start, and Educare, all of which are included in our preschool attender group. Even with a larger comparison group, the existing literature suggests we may not find effects: for instance, a recent study using data from North Carolina with a similar overall sample size but larger comparison group also found no interactions (Burchinal et al., 2022).

Third, while the instruments used to capture features of the kindergarten classroom instructional and self-regulatory environments – the COPG/TOPG and the ATSRS – are arguably improvements on earlier tools not designed specifically to assess the most influential aspects of the classroom environment, as with any new instruments, they are relatively untested. More data are needed to assess the reliability and validity of these measures, across different samples and in different locales.

Third, although our sample was highly race/ethnically diverse, it was designed to be exclusively low-income and drawn from a single school district. Thus, the extent to which our results may generalize to other demographic subgroups and contexts is unclear.

Future Directions

The results from the current analysis, alongside findings from numerous other recent studies, suggest that the sustaining environments hypothesis is unlikely to shed light on why preschool benefits sustain in some evaluations but not others. Perhaps it is time to look elsewhere; we have several suggestions.

First, scholars have raised other promising avenues for investigation into the circumstances under which sustained preschool benefits are detected. Chief among those are hypotheses that surround the type and timing of skills built in the preschool years. Specifically, some have argued that preschool programs that focus on foundational or “constrained” skills that all children master (often in kindergarten) – regardless of preschool attendance – are unlikely to demonstrate sustained gains (see Ansari et al., 2020; Johnson et al., 2023; McCormick et al., 2021). Instead, scholars have called for tracing growth on a wider set of skills, including those that are “unconstrained” and continue to develop as children age. Examples of such skills include expressive vocabulary and numerical fluency, which can be introduced early and grow through the lifespan with continued practice and exposure (Johnson et al., 2023; Lyons et al., 2018; Snow & Matthews, 2016). These capacities may benefit from an early skill-boost that cascades into continued higher performance relative to those who did not get such an early boost (Bailey et al., 2017).

And finally, perhaps tests of moderation (or mediation) within a single preschool program – even one like Tulsa’s that has produced sustained benefits – is the wrong way to evaluate this hypothesis altogether. If the mystery the field is trying to solve is one of variation in findings across programs, the answer is likely to lie within characteristics that vary between, not within, programs. It may be that the regions in which preschool programs have been evaluated vary meaningfully in the quality of their kindergarten environments, not just in their overall averages but also in the range of quality they represent; indeed, it would be perfectly reasonable to expect the minimum level of kindergarten quality observed in a region that is on average high to be substantially higher than the minimum quality observed in a region that is on average low. It is entirely possible that a preschool program in a region with higher-quality kindergarten produces more sustained impacts than a preschool program in a region with lower-quality kindergarten, but within each region, the range of kindergarten quality is too restricted to produce evidence of moderation of preschool attendance by quality. In Tulsa, for example, kindergarten quality may simply not drop low enough to eliminate sustained preschool impacts. If moderation of impacts by subsequent environmental quality were to be found within a regional preschool program such as Tulsa’s, it would support the credibility of the sustaining environments hypothesis, but the absence of evidence is not equivalent to evidence of absence. Furthermore, there may be other salient characteristics that vary across preschool programs – for example, in the populations they serve, the content of their curriculum, or the type of care received by non-attenders – that may explain differences in the size and sustainability of their impacts that are impossible to identify without aggregating data across programs. This is a direction ripe for future research.

Research Highlights.

• This study tested the sustaining environments hypothesis

• We examined a broad range of kindergarten classroom quality measures as moderators

• Sustained preschool benefits are robust to subsequent classroom experiences

• Yet there was no evidence to support the sustaining environments hypothesis

• This study joins others highlighting a need for new explanations for sustained benefits