About this paper symposium

---

Panel information
Panel 18. School Readiness/Childcare

Paper #1
A Systematic Evaluation of Historical Changes in Head Start’s Population
Author information	Role
Ana Whitaker, Ph.D., University of Delaware, United States	Presenting author
Cara L. Kelly, University of Oklahoma, United States	Non-presenting author
Abstract
Introduction Given the expansion of publicly-funded preschool programs in the U.S., there has been an increased focus on factors of programs that promote children’s development. However, a large body of work suggests program effectiveness may be changing over time, with attendance in modern preschool programs having smaller associations with children’s outcomes than programs prior to 2000 (e.g., Whitaker et al., 2023). While other studies have examined societal factors that may be contributing to smaller effects for modern preschool programs (Duncan & Magnuson, 2013), our study aims to fill a gap in the field by systematically examining whether Head Start’s population, in terms of background and demographic characteristics, has changed over time. Hypotheses Study research question: To what extent have the demographics and background characteristics of children served by Head Start changed over time? Are there differential changes in demographics by program characteristics? Given shifts in the U.S. population generally (Frey, 2020), we expect to see the Head Start population grow more diverse in terms of race/ethnicity and home language over time. We also expect to see children become more advanced in terms of the skills they are entering with into Head Start programs. Study Population We relied on the Head Start Family and Child Experiences Survey (FACES) data collected between 1997 and 2019, which collected nationally representative data on a sample of Head Start programs, centers, classrooms, and children. Our sample includes all children and centers in each wave of the data at the fall time point. Methods To answer our research question, we conducted descriptive analyses, including computing means, standard deviations, ranges, and frequencies, for children’s demographic and background characteristics and program characteristics. Depending on the wave of data, children’s demographic characteristics included parent-report of child age, race and ethnicity, home language, and disability status whereas background characteristics include program entry academic and social-emotional skills as reported by parents or collected through direct assessments. Program characteristics were reported by the teacher and/or center director and include type of Head Start classroom, characteristics of children enrolled, and language status of families and teachers. Results Preliminary results for children’s demographic characteristics using FACES 2014, 2009, and 2006 waves of data suggest that children are older when they enroll in Head Start (Table 1). Additionally, more White and Hispanic children enrolled in Head Start over time, but fewer Black children enrolled in Head Start in 2014 compared with 2009 and 2006. Compared with 2009 and 2006, more children with an IEP/IFSP were enrolled in Head Start in 2014. Results from all waves of data (i.e., 2019-1997) will be finalized by the conference. Implications Understanding how the population of children may have changed over time may allow programs to alter instructional practices and content to more closely align with the cultural, linguistic, and racial diversity represented in the children being served by Head Start. Additionally, systematically documenting how Head Start’s population has changed over time can provide policy guidance on what culturally responsive teaching practices and curricula should be required and implemented in programs.

Paper #2
Using Meta-Analytic Data to Examine Fadeout and Persistence of Intervention Impacts on Constrained/Unconstrained Skills
Author information	Role
Mindy Rosengarten, Teachers College, Columbia University, United States	Presenting author
Emma R. Hart, Teachers College, Columbia University, United States	Non-presenting author
Drew H. Bailey, University of California, Irvine, United States	Non-presenting author
Meghan P. McCormick, Overdeck Family Foundation, United States	Non-presenting author
Benjamin J. Lovett, Teachers College, Columbia University, United States	Non-presenting author
Tyler W. Watts, Teachers College, Columbia University, United States	Non-presenting author
Abstract
Introduction The Constrained Skills View (CSV) from literacy research posits that unconstrained skills develop throughout the lifespan, while constrained skills have a ceiling most students reach (Paris, 2005). ECE researchers have suggested that unconstrained skills may persist following Pre-K because they are not targeted in the counterfactual condition (McCormick et al., 2021). However, the CSV has not been rigorously tested across early childhood interventions. This study investigates the CSV using a meta-analytic dataset of educational interventions. We examine the persistence of constrained and unconstrained skills following 38 early childhood interventions and more broadly among educational interventions implemented during other developmental periods. Note, we presented preliminary findings from this project at AEFP. We have since updated our model and added new analyses focused on interventions that we have not yet presented. Population We use the Meta-analysis of Educational RCTs with Follow-up (MERF; Hart et al., in press), a meta-analytic dataset comprised of 85 RCTs reporting cognitive and social-emotional outcomes. We limit the sample to cognitive impacts reported from the same intervention on the same construct, measure, and subscale at posttest and at least one follow-up (called “aligned groups”). Our analytic sample includes 139 “aligned” impacts from 38 early interventions (e.g., Perry Preschool, Head Start Impact Study, TRIAD) measured 6-12 months after posttest. Interventions largely focused on literacy and cognitive development, averaged 10 months duration, with participants 55 months old at baseline, on average. Methods One PhD student and two ECE experts coded 159 combinations of constructs, measures, and subscales as either constrained, unconstrained, or excluded. Any codes that were not unanimous were discussed until consensus was reached, with unresolved codes excluded. We assessed the differential persistence of constrained/unconstrained skills using a meta-analytic regression with random intercepts and effects. We regressed 6–12-month follow-up impacts on posttest impacts, an indicator for whether the aligned group was constrained/unconstrained, and their interaction. Results The three coders achieved 71% reliability. Of the 139 aligned groups, 79 were coded as constrained, 35 unconstrained, and 25 were excluded. Descriptively, unconstrained skills showed smaller impacts at posttest (0.35 SD) and 6–12-month follow-up (0.17 SD) than did constrained skills (0.44 and 0.20 SD, respectively). Regression analyses (Table 1; Figure 1) indicated that unconstrained skills showed a 12-percentage point lower persistence rate than constrained skills, though this estimate was imprecise (p = 0.22). Discussion Our findings provide insight into examinations of impact fadeout within the CSV. First, coder reliability (71%) highlights disagreement surrounding which measures are constrained/unconstrained, possibly because some measures encompass both skill types. Additionally, although constrained skills are defined to have a ceiling, many measures in our sample were normed into adulthood. Second, the introduction of unconstrained skills outside of the intervention may cause control group catch-up. Unconstrained skills like reading comprehension and vocabulary are likely taught in school, while some constrained skills like phoneme segmentation might be intervention specific. Lastly, while we find evidence of fadeout for aligned groups, it remains possible that impacts on unconstrained skills support transfer to broader domains (i.e., reading ability).

Paper #3
An Examination of the Robustness of Estimated Effects of Preschool Programs on Children's Academic Achievement
Author information	Role
Mi Joung Yu, University of California, Irvine, United States	Presenting author
Drew H. Bailey, University of California, Irvine, United States	Non-presenting author
Jade M. Jenkins, University of California, Irvine, United States	Non-presenting author
Abstract
Introduction A large body of literature estimates effects of preschool programs on academic outcomes. While short-term studies, both correlational and experimental, consistently report positive effects, the evidence on medium-term effects is more mixed (Phillips et al., 2017; Whitaker et al., 2024). Recent RCTs have found null (e.g., Gray-Lobe et al., 2023; Weiland et al., 2020) or even negative (e.g., Durkin et al., 2022) impacts on academic achievement in the medium term, while quasi-experimental or observational studies report positive outcomes for similar age groups (e.g., Early et al., 2019). This discrepancy raises important questions about the validity of the different methodological approaches. One possible explanation is that the heterogeneity of medium-term effects is real and provides important insights into program features that could be scaled up successfully. Another interpretation is that the positive effects in non-experimental studies reflect omitted variable bias. Given that about 75% of the preschool effectiveness literature relies on non-experimental methods, examining the robustness of these estimates to potential omitted variable bias is essential for drawing reliable conclusions about the medium-term impact of preschool programs on achievement. Hypotheses We hypothesized that when positive non-experimental estimates of preschool effects on achievement are obtained, they may be biased by unmeasured selection factors. Models that assume some degree of additional confounding may bring these estimates closer to zero, as observed in recent RCTs. Study Population Data came from the Early Childhood Longitudinal Study Kindergarten Cohort of 1998. We restricted the sample to first-time kindergarteners with non-missing data on outcomes variables (n = 6,625) and conducted analyses on the samples with and without Head Start attendees. Methods We regressed reading and math test scores at kindergarten entry and subsequent grades on preschool attendance and a set of child and family covariates to estimate the effects of preschool on achievement. We then applied Oster’s (2019) bounding method, which leverages information on the effects of observed covariates to quantify selection effects due to unobservables. Specifically, we estimated the amount of unobserved selection bias that would be necessary to explain away an effect and calculated bounding values for the treatment effect assuming equal selection on unobservable and observable confounders. Results We found positive effects of preschool attendance on achievement, but these effects significantly decreased after kindergarten entry. Covariate adjustment reduced the estimated effects by ~30-60% and increased R^2 from ~0 to a range of .12-.24, suggesting selection on observed covariates. The controlled estimates were smaller than those previously reported in non-experimental pre-K studies. Before applying bounding, we obtained larger estimates when excluding Head Start attendees from the sample. However, once bounded for potential unobserved confounding, the estimates converged across the full sample and the restricted sample without Head Start attendees. Assuming small amounts of additional confounding, the estimates of preschool effects at kindergarten entry remained generally robust (.10-.16 SD), but beyond Grade 1, the estimates dropped to values closer to those observed in recent RCTs (.02-.06 SD). Complementary analyses will be conducted using data from pre-K programs that have yielded positive medium-term non-experimental estimates.

Paper #4
Teacher-Student Racial-Ethnic Match and Kindergarteners’ Academic and Cognitive Gains: Evidence from Boston
Author information	Role
Ms. Emily Yerington, University of Michigan, United States	Presenting author
Christina J. Weiland, University of Michigan, United States	Non-presenting author
Meghan McCormick, Overdeck Foundation, United States	Non-presenting author
JoAnn Hsueh, MDRC, Foundation for Child Development, United States	Non-presenting author
Jason Sachs, Bill and Melinda Gates Foundation, United States	Non-presenting author
Catherine Snow, Harvard Graduate School of Education, United States	Non-presenting author
Tiffany Wu, University of Michigan, United States	Non-presenting author
Paola Guerrero-Rosada, University of California, Irvine, United States	Non-presenting author
Abstract
Introduction Although the share of Asian, Black, and Hispanic or Latino teachers in the U.S. is increasing over time, the racial-ethnic (RE) demographics of kindergarten teachers do not yet match the diversity of students (Census, 2022). As a result, U.S. kindergartners of color are less likely to have a teacher of the same race or ethnicity as themselves – i.e., to experience a RE match between student and teacher (Gottfried & Fletcher, 2023). Research suggests that students of color may benefit from having a teacher of the same RE as themselves (Gershenson et al., 2021; Redding, 2019). However, there is limited literature on RE match between students and their teachers in the context of kindergarten. Further, most existing studies focus on outcomes for Black and Hispanic students. RE match may impact Asian and White students differently. In the current study, we address this gap by examining data from a diverse sample of 833 kindergarten students in the Boston Public Schools (16% Asian, 24% Black, 34% Hispanic, 26% White). Research Questions We address two research questions: 1. Does the prevalence of RE match differ by student race/ethnicity? 2. Do children demonstrate greater gains in language, literacy, mathematics and executive function skills during the kindergarten year when they experience RE match with their teacher? Do these associations differ by child racial/ethnic group? Study Population and Methods We measured academic gains in language, literacy, mathematics, and executive function using direct assessments collected in fall and spring of the 2017-2018 school year. To address RQ1, we used linear probability models with random intercepts for classroom and school. For RQ2, we used multilevel modeling with random intercepts for classroom and school. We fit models controlling for child-level and then classroom-level covariates (see table notes for a list). Results Rates of RE match by RE group are summarized in Table 1. We found that students of color were less likely to experience RE match with their teacher compared to their White peers. Academic gains results are summarized in Table 2. We found positive, statistically significant associations between RE match and gains in math skills for the full sample of students (d = 0.11; p < .05), but no association between RE match and gains in language, literacy, and executive function skills. Within racial-ethnic subgroups, associations between RE match and math gains were statistically significant for Asian students only (d = 0.62; p < .05). Overall, our findings suggest that the effects of RE match may be more pronounced in certain academic domains and/or for certain RE groups. Still, while RE match has been shown to benefit older students, its impact in early childhood education (ECE) is less consistent and needs further exploration. At SRCD, we will discuss how certain factors in the Boston Public Schools context, such as a culturally responsive curriculum and heterogenous RE groups, may mitigate the influence of RE match, leading to more nuanced or null results.

⇦ Back to session