Socio-Emotional Learning among Low-Income Prekindergarteners: The Roles of Individual Factors and Early Intervention

Abstract

Research Findings:

Previous research has indicated that low-income children are at increased risk for socio-emotional problems, which may contribute to socioeconomic disparities in wellbeing and academic achievement. The present study examines socio-emotional learning (SEL) across the prekindergarten year in a low-income, racially and ethnically diverse sample of Chicago Public School students (N=2,630). The sample included participants of the Child-Parent Center early educational intervention program (N=1,724) and a propensity-score matched comparison group (N=906). At the beginning of the prekindergarten year, teachers rated boys and lower income participants as having relatively lower SEL skills, and CPC participants and older children as having slightly higher SEL skills. Over time, CPC participants exhibited significantly greater rates of SEL growth, ending the prekindergarten year with teacher-rated SEL scores that were an average 10.30% higher than control participants. There were no significant differences in SEL growth over time by sex or family income.

Practice and Policy Implications:

Multicomponent, school-based early intervention programs (e.g., CPC) have the potential to promote SEL among at-risk populations.

Prekindergarten participation has steadily increased in the United States over the last two decades, with 40% of three-year-olds and 68% of four-year-olds enrolled in programs in 2017 (National Center for Education Statistics, 2019). However, closer investigation of enrollment statistics reveals persistent socioeconomic and racial disparities. Children from low-income families, African American and Latino children are less likely to be enrolled in prekindergarten programs than their peers, and are also less likely to attend high-quality programs (U.S. Department of Education, 2015). These disparities in prekindergarten access leave “too many children enter[ing] kindergarten a year or more behind their classmates in academic and socio-emotiaonl skills…[which can lead to] a cycle of continuous catch-up in their learning” (U.S. Department of Education, 2015, p. 1). In response to this issue, several prekindergarten programs for low-income children have been developed and implemented at large scales, often with funding from the Elementary and Secondary Education Act (e.g., the Child Parent-Center (CPC) program, Project Head Start). Further expansion of these programs will be contingent on evaluations of how effectively they promote school readiness among diverse populations of children.

Researchers, educators, and policymakers increasingly agree that one of the most critical aspects of school readiness is developing strong socio-emotional skills (Denham & Brown, 2010; Domitrovich, Durlak, Staley, & Weissberg, 2017; Rimm-Kaufman, Pianta, & Cox, 2000). Prekindergarten-aged children face several critical socio-emotional learning (SEL) tasks, including learning how to regulate emotions and behaviors, form positive relationships with others, and appropriately engage in group activities (Newman & Dusenbury, 2015). Strong skills in these domains have been linked to greater academic success (e.g., Blair & Razza, 2007; Bronson, 2000; Carlton & Winsler, 1999; Jones, Greenberg, & Crowley, 2015; Rhoades, Warren, Domitrovich, & Greenberg, 2011), greater motivation and positive attitudes about school (e.g., Birch & Ladd, 1997), and more positive interactions with peers and teachers (e.g., Berk, 2002; Denham, Blair, Schmidt, & DeMulder, 2002; Engle, McElwain, & Lasky, 2010). On the contrary, children who lag behind in SEL are at increased risk for academic, social, and behavioral problems (e.g., Denham et al., 2003; Fantuzzo et al., 2007; Jones et al., 2015; Kochenderfer & Ladd, 1996; McClelland, Morrison, & Holmes, 2000; Tremblay, Pagani-Kurtz, Masse, Vitaro, & Pihl, 1995).

Researchers have been studying children’s development of various SEL skills (e.g., self-regulation, social skills) for decades, and in the early 1990s, the Collaborative for Academic, Social, and Emotional Learning (CASEL) drew on this body of work to create an integrative SEL framework (CASEL, 2012; Newman & Dusenbury, 2015). This framework has since been widely referenced by researchers, and informed the development of educational legislation and learning standards across the United States. Consequently, interest in assessing the effects of educational programs (e.g., prekindergarten) on children’s SEL has grown, and a number of classroom-based SEL measures have been developed. This body of work is still relatively young, and few studies have specifically examined SEL trajectories among prekindergarten-aged children, especially ones from low-income backgrounds.

The present study contributes to the literature by: (a) examining prekindergarten SEL trajectories in a low-income, racially and ethnically diverse sample (as measured using the GOLD® system); and (b) comparing the SEL trajectories of children enrolled in a comprehensive prekindergarten intervention (CPC) versus typical public prekindergarten programming. The next sections of this Introduction will briefly review the extant literature in both of these areas, highlighting gaps that the present study aims to address.

Group differences in SEL

Family poverty.

Previous work has shown that low-income children typically enter prekindergarten with higher cumulative risk (e.g., greater exposure to maltreatment and neighborhood violence; lower material resources) than their more affluent peers, and are more likely to exhibit emotional and behavioral difficulties (e.g., Comeau & Boyle, 2018; Duncan, Brooks-Gunn, & Klebanov, 1994; Eamon, 2001; Qi & Kaiser, 2003; Knapp, Ammen, Arstein-Kerslake, Poulsen, & Mastergeorge, 2007; Li, Johnson, Musci, & Riley, 2017; Yoshikawa, Aber, & Beardslee, 2012). However, it is important to note that not all experiences of poverty are equal (Roy & Raver, 2014). Even in settings that primarily serve low-income children (e.g., the CPC program, Project Head Start), significant inter-individual variation is evident in children’s depth of experienced poverty (e.g., how close a family’s income is to the poverty threshold, as measured by income-to-poverty ratio or income deficit/surplus) and constellations of risk and protective factors (Roy & Raver, 2014; Semega, Kollar, Creamer, & Mohanty, 2019). As such, it is important not only to compare the SEL of low-income versus more affluent children, but to also examine potential subgroup differences within low-income samples.

Sex.

Researchers have also documented potential sex differences in early childhood SEL. For example, Lambert, Kim, and Burts (2014) examined the SEL of young boys versus girls (ages one to four years) within a growth curve model, drawing on data from a nationally representative sample (N=21,952). Teachers used the observation-based GOLD® assessment system to rate children’s skills in various socio-emotional domains at three time-points. Results indicated that boys received significantly lower socio-emotional ratings than girls at baseline, and that boys also exhibited significantly lower rates of SEL growth over time.

In another study, Barbarin (2013) examined children’s SEL across the prekindergarten year, using repeated-measures ANOVA and controlling for family poverty. Data was drawn from two multi-state evaluations of state-sponsored prekindergarten programs (N=2,458). Teachers rated children’s socio-emotional skills at two time-points using the Teacher-Child Rating Scale questionnaire (TCRS; Hightower et al., 1986). Results indicated that boys received significantly lower socio-emotional ratings at baseline than girls, but that there were no significant gender differences in SEL growth over time. Among boys, there were no significant racial differences in scores either at baseline or over time.

There is also a need for further investigation of potential sex differences in low-income children’s SEL. Lambert, Kim and Burts (2014) and Barbarin (2013) reported that boys had lower teacher-rated socioemotional skills at the start of prekindergarten, compared to their female classmates. Similar findings have been reported in several other general population studies (e.g., Chen, 2008; Lavigne et al., 1996; Matthews, Pointz, & Morrison, 2009) and in research with low-income samples (e.g., Raikes, Robinson, Bradley, Raikes, & Ayoub, 2007). Among low-income boys, neurobiological and environmental risk factors may interact to shape socio-emotional development (Golding & Fitzgerald, 2017; Schore, 2017). For example, research has shown that early-developing stress-regulation circuits mature more slowly in male versus female brains, increasing males’ susceptibility to environmental toxins and social stressors (Schore, 2017). Meanwhile, low-income boys and boys of color are disproportionately exposed to environmental toxins and stressors (e.g., historical trauma, attachment disruptions, neighborhood violence, environmental contaminants) (Golding & Fitzgerald, 2017; McKinney, Fitzgerald, Winn, & Babcock, 2017). In this way, neurobiological vulnerabilities may interact with systemic social disparities to affect boys’ socio-emotional development before prekindergarten. Research has also demonstrated that low-income boys and boys of color face unique environmental challenges when they enter school – including potential implicit bias on the part of educators. Prekindergarten teachers are overwhelmingly female; and boys of color are frequently racially/ethnically mismatched with their teachers (Downer, Goble, Myers, & Pianta, 2016; Whitebook, McLean, & Austin, 2016). Past work has shown that, within this context, teachers may perceive the behavior of boys (especially boys of color) as excessively problematic or aggressive (DiCarlo, Baumgartner, Ota, & Jenkins, 2015; Gilliam, Maupin, Reyes, Accavitti, Shic, 2016). To this end, prekindergarten boys are suspended at three times the rate of their female classmates (U.S. Department of Education Office for Civil Rights, 2014). These findings highlight the importance of identifying effective strategies for supporting low-income boys and boys of color as they transition into prekindergarten.

English Language Learners (ELLs).

Under federal law, students who require support to develop English proficiency are broadly referred to as “ELLs”. In other policies and research, several other terms have also been used to delineate developmental differences in language acquisition (Halle et al., 2014). For example, the term “dual language learners” is sometimes used to describe young children who are simultaneously learning English as a second language while continuing to master their home language(s) (Williams, 2014). The present paper, while recognizing the diversity of terminology in this area, will use the term “ELLs” for ease of reading.

Young ELLs in the U.S. face a number of unique risk and protective factors for SEL (Oades-Sese & Esquivel, 2006; Oades-Sese et al., 2012). ELLs often receive strong support and supervision from their nuclear and extended families, and from their broader cultural and religious communities (Cardoso & Thompson, 2010). However, as noted by Oades-Sese and colleagues (2012, p. 279), “the transition from home to school [e.g., at prekindergarten entry] for these children brings with it a sudden encounter with cultural and linguistic discontinuities that [can] affect emotional and academic adjustment”.

Several empirical studies have examined the SEL of ELLs in early education contexts. In a narrative review of 14 studies of dual-language learners (birth to age five), Halle and colleagues (2014) reported that teachers generally rated ELLs’ socio-emotional skill as equal or even better than monolingual English speakers. This review included both cross-sectional and longitudinal studies, making it difficult to discern whether teaechers’ ratings of ELLs changed over the course of the academic year.

In another study, Lambert, Kim, and Burts (2014) empirically investigated the SEL of a nationally representative sample of one- to four-year-old children (N=21,952) using growth-curve modeling. Teachers rated children’s socioemotional skills at three time-points using the observation-based GOLD® assessment system. Results indicated that ELLs received significantly lower socio-emotional ratings at baseline, but exhibited significantly higher growth rates over time than non-ELLs. The authors hypothesized that these results partially reflected increased accuracy of teacher ratings over the course of the academic year, as ELLs gain English proficiency and teachers become more acquainted with children and their families.

Special education.

According to the federal Individuals with Disabilities Education Improvement Act (IDEA, 2004), all children ages three years and older have the right to special education services in U.S. public schools. School districts and/or health and human service agencies conduct eligibility evaluations and plan early intervention services, with some variation across states. Sample groups of children who may be eligible for special educational services under IDEA include those with developmental delay (e.g., cognitive, physical, communication, social/emotional, adaptive), and those who are diagnosed with physical or mental conditions (e.g., deaf-blindness, traumatic brain injury).

Previous research has highlighted heightened rates of socio-emotional problems among young children with developmental delays (e.g., Emerson, Einfeld, & Stancliffe, 2010; Gallagher & Lambert, 2006; Kim, Carlson, Curby, & Winsler, 2016; Merrell & Holland, 1997). Similarly, caregivers of young children in special education have been shown to endorse greater concerns about their children’s behavior and communication skills than caregivers whose children are in general education (e.g., McIntyre, Eckert, Fiese, Reed, & Wildenger, 2010). These findings suggest that young children who are eligible for special education may be good candidates for socio-emotional screening and intervention upon prekindergarten entry.

Intervention-Focused Studies.

The previous section highlighted the current knowledge base on SEL across early childhood, including potential high-risk groups (e.g., children from low-income families, boys). Findings underscored the need for additional prospective investigations of SEL over time, which might inform the development of tailored interventions for high-risk groups. To this end, the next section of this Introduction will review what is currently known about the relationship between prekindergarten intervention and SEL, with a particular focus on the CPC program (which will be empirically investigated later in the present study).

Public prekindergarten.

In a notable study, Weiland and Yoshikawa (2013) examined the SEL of four- and five-year-old children in Boston (N=2,018). Intervention group participants were recent graduates of Boston Public School’s (BPS) prekindergarten program, which includes evidence-based curricula and a teacher coaching component. Comparison group participants had a variety of experiences during the prekindergarten year, ranging from relative care to center-based prekindergarten programming. Trained study assessors rated children’s socio-emotional functioning in Fall of the kindergarten year, based on their performance on the Task Orientation Questionnaire (TOQ; Smith-Donald, Raver, Hayes, & Richardson, 2007) and Emotion Recognition Questionnaire (ERQ; Ribordy, Camras, Stefani, & Spaccarelli, 1998). Results indicated that BPS prekindergarten graduates exhibited significantly greater emotion regulation and inhibitory control than non-graduates. Among BPS prekindergarten graduates, those who qualified for free or reduced lunch exhibited greater improvements in inhibitory control than those who were ineligible. Some differences by race and ethnicity were also evident. This study did not utilize a randomized or longitudinal design; however, its results point to the potential SEL benefits of prekindergarten intervention, particularly among low-income children.

The Child-Parent Center (CPC) program.

Several studies have examined children’s SEL in the context of the CPC intervention program (the same program that will be empirically investigated in the present study). CPC was founded in Chicago in the 1960s with funding from the Elementary and Secondary Education Act. In recent years, the program has been scaled-up to serve thousands of children and families living in high-poverty neighborhoods around the Midwest (Reynolds, Hayakawa, Candee, & Englund, 2016).

CPC centers are embedded in public school settings and emphasize six core elements: (a) collaborative leadership; (b) effective learning experiences; (c) alignment between evidence-based curricula and instructional practices; (d) parent involvement and engagement (e) professional development for educators; and (f) continuity and stability from prekindergarten through third grade (Reynolds, 1994; Reynolds, Hayakawa, Candee, & Englund 2016; Reynolds & Mondi, 2016; Reynolds, Temple, & Ou, 2003.) The aforementioned CPC elements are designed to enhance children’s holistic development and school readiness, including SEL (Reynolds, Hayakawa, Candee, & Englund, 2016).

Reynolds, Richardson, Hayakawa, Englund, and Ou (2016) examined the effects of CPC on prekindergarteners’ SEL using a quasi-experimental, matched-group cohort design. Intervention group members attended CPC prekindergarten (n=1,724), while matched comparison group members attended alternative prekindergarten programs (n=906). Teachers rated children’s SEL across the pre-kindergarten year using the GOLD® assessment system. Results indicated that, when controlling for baseline scores, CPC participants received significantly higher SEL ratings at the end of the year than comparison group members (mean scores of 57.0 vs. 51.8, respectively, p<0.001). These results parallel those of Weiland and Yoshikawa (2013), underscoring the potential of pre-kindergarten intervention to support low-income children’s SEL. However, Reynolds and colleagues did not examine whether the relations between CPC participation and SEL differed for key subgroups of children (e.g., by sex, race/ethnicity, income, or child age).

Richardson, Reynolds, Temple, and Smerillo (2017) further investigated the impacts of CPC participation on two subgroups of children: children who spoke Spanish at home, and children who were eligible for free lunch (a proxy for family poverty). Their study sample was based in Chicago and consisted of both CPC prekindergarten participants (n=1,289) and children who attended public prekindergarten at matched comparison group sites (n=584). Teachers rated children’s SEL across the pre-kindergarten year using the GOLD® assessment system. Results indicated that, when controlling for baseline scores, CPC participants received significantly higher SEL ratings at the end of prekindergarten than comparison group members, regardless of Spanish language or family poverty level. These results further highlight the benefits of CPC participation, and are consistent with other research indicating that children with the highest levels of sociodemographic risk often benefit the most from early intervention services (e.g., Ou & Reynolds, 2010; Reynolds, Temple, White, Ou, & Robertson, 2011).

The previously discussed intervention studies (Reynolds, Richardson, et al., 2016; Richardson et al., 2017) highlighted the potential SEL benefits of CPC prekindergarten for low-income children, including children from the highest poverty and Spanish-speaking families. Notably, the primary aim of both studies was to quantify the average effect sizes of CPC participation on child outcomes. Both studies utilized regression-based approaches to examine point-differences in CPC versus comparison group members’ SEL scores at the end of the prekindergarten year, controlling for baseline scores. This approach provided valuable insights into the overall impacts of CPC; however, important questions remain about the patterns of SEL growth among different subgroups of children, from baseline to the end of prekindergarten and beyond.

Present Study

The previous review highlighted the need for prospective studies examining: (a) SEL trajectories in early childhood, particularly among low-income and racially/ethnically diverse children; and (b) the effects of prekindergarten intervention on young children’s SEL, including investigations of potential differential impacts for high-risk subgroups. The present study begins to address both of these gaps by examining SEL across the prekindergarten year in a large (N=2,630), low-income, racially and ethnically diverse group of children. The study sample includes both CPC prekindergarten participants and a demographically comparable comparison group. Multilevel linear mixed effects modeling is employed (rather than a regression-based approach), allowing for more nuanced investigation of individual differences in baseline SEL and children’s SEL over time.

Our first set of hypotheses focuses on individual differences in SEL in this diverse, low-income sample. Based on the previously reviewed research, we hypothesized that boys, students from the highest poverty families, special education students, and ELLs would receive significantly lower SEL ratings at baseline (prekindergarten entry) relative to their counterparts. We also hypothesized that relatively older children and children attending programs in higher achieving schools would enter prekindergarten with higher SEL skills, and that there would be no differences in baseline SEL based on race/ethnicity. Meanwhile, regarding change in SEL over time, we posited that boys and special education participants would exhibit less SEL growth over time, and that participants from the highest poverty families, ELLs, and participants attending programs located within higher achieving schools would exhibit greater growth over time than their counterparts. We posited that there would be no differences in SEL growth over time based on age at prekindergarten entry or race/ethnicity.

Our second set of hypotheses focuses on differences in SEL based on intervention status. Based on the previously reviewed research, we hypothesized that CPC participants would exhibit significantly greater SEL growth over time than comparison group members.

Method

Sample

The present study analyzes secondary data from the Midwest Longitudinal Study (MLS), which began in 2012 as a federally funded evaluation of the Midwest CPC Expansion Project (Reynolds, Hayakawa, Candee, & Englund, 2016). The final study sample included 2,630 children who were enrolled in prekindergarten in Chicago Public Schools (CPS) in Fall 2012. The majority of these participants (n=1,724) were enrolled in the CPC prekindergarten program at 16 different sites. As previously described, the CPC program serves children who attend schools that are located in high-poverty areas, and which receive Title I funding from the Elementary and Secondary Education Act. The program provides comprehensive, center-based services, including educational, health, social support, and family support services (Reynolds, 1994; Reynolds, Hayakawa, Candee, & Englund, 2016; Reynolds & Mondi, 2016; Reynolds, Temple, & Ou, 2003). One of the primary goals of the CPC program is to “enhance social adjustment and psychological development in the early grades, including socio-emotional learning, school commitment, and self-control” (Reynolds, Hayakawa, Candee, & Englund, 2016, p. 5). This goal is accomplished through a variety of learning experiences for children, educators, and caregivers, including classroom-based SEL curricula (e.g., grade-specific interactive SEL lessons), professional development programming on facilitating children’s SEL, and SEL programming for parents. The CPC program also prioritizes small class sizes, consistency in learning environments across grades, and referrals to health/social services for children with identified social or emotional needs.

The remaining participants (n=906) were enrolled in state-funded Chicago Public School prekindergarten programs. These participants were enrolled at 14 matched comparison sites, which were selected on the basis of having similar demographic characteristics as the CPC sites. The primary selection criterion was that sites were located in high-needs areas where student achievement was a high priority. Study researchers, along with independent evaluation partner SRI International, used school-level propensity score matching to identify comparison schools offering the usual preschool services for four-year-olds. The matching variables were drawn from state report card data and included eligibility for subsidized lunches (family income proxy), percentage of English Language Learners served, percentage of minority students served, and percentage of students meeting the third-grade reading proficiency benchmark on state achievement tests. Several secondary indicators were utilized to help select among similar matches and to strengthen precision (percentage of black, Hispanic, and Asian students; mobility and special education rates; third grade math proficiency; and presence of other interventions). The selected comparison schools provided the closest matches and were also located in close proximity to the CPC sites (to control for local history).

Inclusion criteria and imputation.

Across both the CPC and comparison groups, participants had to be enrolled in their respective program for at least four consecutive months to be included in analyses. Some participants enrolled later in the school year, resulting in increased sample size across time-points. In order to meet the four-month-minimum requirement, participants had to be enrolled in their respective program by January 2013. Overall, 2,630 participants met this inclusion criteria.

The Chicago Public Schools district provided administrative data (including demographic information and outcome measures) to the study researchers, as part of an ongoing evaluation of the CPC program. Extensive efforts were made to obtain consent from participants’ parents; ultimately, affirmative consent forms were returned for approximately 80% of the current study’s participants. CPS, in coordination with the research team’s Institutional Review Board, determined that consent could be waived for the remaining 20% of cases. As such, the study sample represents all of the children who were enrolled in their respective program for at least four consecutive months.

During the data collection period, Chicago Public Schools was requiring all teachers to assess their students using the GOLD® measure (this study’s primary outcome measure; see description below). However, this requirement was not stringently enforced, resulting in a substantial proportion of missing data. Additionally, as noted above, some students left or joined the school district mid-way through the school year, resulting in missing data at some time-points. Empirical and visual inspection of this data (drawing on extensive participant demographic information made available by the district) supported a missing at random pattern. Withhin- and across-wave data were highly correlated. Multiple imputation based on the expectation-maximization algorithm was consequently conducted to increase the sample size (Little & Rubin, 1987; Schafer & Olsen, 1998). The input variables for multiple imputation included: school, age, CPC participation, full-day participation, race/ethnicity, special education, eligibility for subsidized lunch, and fall to spring TS GOLD scores. Imputation was conducted on the total SEL scores (versus individual items) at each time-point within a single dataset. No statistically significant differences were evident in unimputed versus imputed SEL scores, increasing confidence in the imputation process (see discussions in Reynolds et al., 2014; Richardson, Reynolds, Temple, & Smerillo, 2017).

Table 1 describes the demographic characteristics of the final imputed sample. There were no significant differences in the proportion of males versus females, or in participant age across groups. The intervention group had a higher proportion of African American participants (64.1% versus 45.6%, p<0.001), and was more likely to attend a high-achieving school than the comparison group (47.4% versus 40.7%, p<0.001). The comparison group had a higher proportion of Hispanic/Latinx participants (53.8% versus 34.1%, p<0.001) and English Language Learner participants (49.8% versus 29.2%, p<0.001) than the intervention group. CPC participants were more likely to come from single parent families than comparison group participants (56.0% vs. 44.0%, respectively; p< 0.001). CPC participants also received slightly higher teacher ratings on the Wave 1 literacy subscale of the GOLD® than comparison group participants (mean = 33.7 vs. 31.4, respectively; p<0.01).

Table 1.

Baseline Characteristics of Intervention and Comparison groups

	CPC	Control	p	Std. Error Difference
Age in months on September 1, 2012	48.4	48.6	>0.05	0.02
Female, %	51.6	50.2	> 0.05	0.02
Black/African American, %	64.1	45.6	< 0.001***	0.02
Hispanic/Latinx, %	34.1	53.8	< 0.001***	0.02
English Language Learner (ELL), %	29.2	49.8	< 0.001***	0.02
Special education, %	9.6	9.2	> 0.05	0.01
Eligible for free lunch, %	85.4	83.2	> 0.05	0.02
High-achieving school, %	47.4	40.7	< 0.001***	0.02
Single parent family status, %	56.0	44.0	< 0.001***	0.02
Fall score on Literacy subscale, mean (SD)	33.7 (15.3)	31.4 (13.0)	< 0.01**	0.60
Fall score on Math subscale, mean (SD)	22.6 (8.5)	23.2 (7.2)	> 0.05	0.33

^*p<0.05

^**p <0.01

^***p<0.001

Measures

Socio-emotional competencies.

Teaching Strategies Gold® (GOLD® ).

GOLD® is an observation-based assessment system for children from birth through Kindergarten (Lambert, Kim, & Burts, 2013). Teachers document children’s developing knowledge, skills, and behaviors in the following domains: socio-emotional, cognitive development, physical, language, literacy, mathematics, and English-language acquisition. This information can be used to assess children’s readiness for elementary school, to inform instructional practice, and to facilitate communication with stakeholders.

During the data collection period, the Chicago Public Schools district required all teachers to undergo routine training in the GOLD® system. Total training time was approximately two-and-a-half days, and covered the GOLD® skill areas, performance-based data collection, item scoring, and use of the GOLD®’s interactive computer-based system. Following training, teachers observed students for four to six weeks before submitting scores for Wave 1 (Fall 2012). At the Wave 1 checkpoint, teachers evaluated the information that they had collected on children’s skills, knowledge, and behaviors, and gave them overall ratings in the aforementioned domains. Ratings were on a 10-point scale, ranging from Level 0 (“Not Yet”) to Level 9 (“Beyond Kindergarten Expectations”). The GOLD® system included “indicator levels” which provided teachers with examples of what evidence might look like at different Levels for the majority of children, as well as particular subgroups (e.g., what evidence could a teacher look for when deciding whether to rate a child as being at Level 5 or 6 for “manages feelings”). This process was repeated for two additional timepoints in Winter 2012/2013 (Wave 2) and Spring 2013 (Wave 3).

The present study’s outcome data is drawn from the socio-emotional domain of GOLD®. This domain parallels seminal work on SEL (see overview in CASEL, 2012) and assesses children’s functioning in regards to three main objectives: (a) regulating emotions and behaviors (three items: “manages feelings”, “follows limits and expectations”, “takes care of own needs appropriately”); (b) establishing and sustaining positive relationships (four items: “forms relationships with adults”; “responds to emotional cues”; “interactes with peers”; “makes friends”); and (c) participating cooperatively and constructively in group situations (two items: “balances needs and rights of self and others”, “solves social problems”) (Table 3). These objectives are grounded in empirical research on child development and are aligned with Common Core State Standards, early learning guidelines, and the Head Start Child Development and Early Learning Framework (National Governors Association Center for Best Pracices & Council of Chief State School Officers, 2010). Each objective measures several dimensions, for a total of nine dimensions.

Table 3.

GOLD® individual item descriptive statistics (unimputed)

	Wave 1				Wave 2				Wave 3
	Full sample		CPC	Control	Full sample		CPC	Control	Full sample		CPC	Control
	Mean (S.D.)	Unimputed sample size	Mean (S.D.)	Mean (S.D.)	Mean (S.D.)	Unimputed sample size	Mean (S.D.)	Mean (S.D.)	Mean (S.D.)	Unimputed sample size	Mean (S.D.)	Mean (S.D.)
Regulating emotions and behaviors
1a. Manages feelings	4.48(1.76)	1,031	4.43(1.85)	4.70(1.31)	5.37(1.53)	1,099	5.40(1.57)	5.26(1.32)	6.20(1.59)	1,130	6.30(1.62)	5.84(1.42)
1b. Follows limits and expectations	4.66(1.68)	1,002	4.64(1.73)	4.75(1.38)	5.54(1.40)	1,124	5.57(1.45)	5.43(1.15)	6.34(1.42)	1,128	6.45(1.45)	5.92(1.23)
1c. Takes care of own needs appropriately.	5.12(1.64)	1,008	5.10(1.68)	5.22(1.44)	5.93(1.35)	1,116	5.94(1.38)	5.86(1.23)	6.72(1.34)	1,107	6.81(1.31)	6.35(1.37)
Establishing and sustaining positive relationships
2a. Forms relationships with adults.	5.68(1.80)	1,006	5.69(1.86)	5.65(1.45)	6.46(1.35)	1,124	6.49(1.37)	6.35(1.24)	7.23(1.21)	1,106	7.34(1.18)	6.82(1.20)
2b. Responds to emotional cues.	4.31(1.83)	979	4.31(1.91)	4.32(1.25)	5.23(1.55)	1,098	5.27(1.61)	5.07(1.24)	6.07(1.55)	1,100	6.22(1.55)	5.43(1.34)
2c. Interacts with peers.	4.28(1.63)	1,046	4.31(1.71)	4.14(1.20)	5.19(1.60)	1,122	5.28(1.66)	4.84(1.31)	6.12(1.64)	1,138	6.31(1.63)	5.44(1.49)
2d. Makes friends.	4.41(1.90)	1,032	4.38(2.02)	4.51(1.19)	5.32(1.64)	1,101	5.38(1.70)	5.06(1.27)	6.23(1.59)	1,136	6.39(1.61)	5.65(1.36)
Participating cooperatively and constructively in group situations
3a. Balances needs and rights of self and others.	3.98(1.78)	1,053	3.96(1.87)	4.04(1.26)	4.97(1.60)	1,108	5.03(1.65)	4.65(1.26)	5.91(1.64)	1,145	6.08(1.66)	5.26(1.38)
3b. Solves social problems.	4.19(1.60)	1,367	4.15(1.65)	4.39(1.31)	4.95(1.42)	1,411	4.99(1.47)	4.78(1.15)	5.81(1.55)	1,384	5.90(1.59)	5.39(1.28)

Note: Wave 1= Fall 2012, Wave 2=Winter 2012-2013, Wave 3=Spring 2013

Reliability for the un-imputed SEL scale (sum of the nine items) was excellent in the current study (Cronbach’s alpha = 0.97). This is consistent with previous research reporting strong psychometric properties for the GOLD® (e.g., concurrent validity with other socio-emotional measures, reliability across time-points; Lambert, Kim, & Burts, 2015). One previous study assessing inter-rater reliability between the ratings of teachers and master GOLD® trainers reported a correlation of 0.90 for the socio-emotional domain (Lambert et al., 2015). Previous research has also demonstrated the reliability and validity of GOLD® for use with ELL children (Kim, Lambert, & Burts, 2013).

Covariates.

The following information was drawn from school administrative records.

CPC.

A dummy coded variable was created to indicate group membership (0=comparison group, 1=CPC intervention group).

Gender.

A dummy coded variable was created to indicate gender (0=male, 1=female).

Race/ethnicity.

A dummy coded variable was created to indicate whether participants were black (e.g., African American or African immigrant; 1=yes). An additional dummy coded variable was created to indicate whether participants were Hispanic (1=yes).

Free lunch.

A dummy coded variable was created to indicate ineligibility (0) or eligibility for free lunch (1).

Free lunch status is a proxy for family poverty. Although all the school sites in the present study were located in low-income neighborhoods, individual families’ financial resources varied. During the 2012-2013 academic year, students qualified for reduced or free lunch if their family’s total household income fell below 130 percent of the federal poverty line, or within 130 and 185%, respectively.

Notably, although the number of children who were eligible for free lunch outnumbered those who were not eligible (Table 1), the respective sample sizes were adequately powered to detect small effect sizes (e.g., Cohen’s d < 0.2).

Age.

A continuous variable indicating children’s age in months at baseline was created.

Special education.

A dummy coded variable was created to indicate whether participants received any special education services during the prekindergarten year (1=yes). Eligibility for these services was determined by the school district, based on local and federal guidelines.

ELL.

A dummy coded variable was created to indicate language status (0=English native language, 1=ELL).

School-level achievement.

A dichotomous variable was created to indicate whether 70% or more of the students at each school met statewide benchmarks for third grade reading (1=yes/high-achieving school, 0=no/lower achieving school).

Baseline submission date.

A dummy coded variable was created to indicate whether participants’ teachers submitted baseline ratings on-time (by October 31) or late (0=after October 31, 1=late). While this covariate is not related to one of the main research questions, it was included to control for variation in the dates in which teachers completed their evaluations. For example, teachers who submitted their ratings one month late had one additional month of experience with their students compared to students who submitted on time. During this additional month, children gained a month’s worth of experience, and teachers also had an additional month to develop perceptions of their students, which could affect SEL ratings.

Model specification process

Multi-wave research designs allow for assessment of within-individual changes over time, as well as exploration of systematic differences in these changes based on theoretically informed covariates (Fitzmaurice, Laird, & Ware, 2011). Each individual serves as his or her own control, and tests of processes occurring within-subjects, between-subjects, and at different levels can be conducted. When repeated measures are obtained on the same individuals, it is expected that data across time-points will be correlated. It is thus critical to appropriately model the covariance or time dependence among repeated measures, to prevent incorrect estimates of sampling variability or misleading inferences (Fitzmaurice et al., 2011). Both the conditional mean response over time and the conditional covariance on repeated measures obtained on the same individuals must be modeled.

In the present study, we were interested in examining both average SEL patterns and individual-level trajectories. Linear mixed effects (LME) modeling, a form of multilevel or hierarchical modeling, is well-suited for both domains of inquiry (Gelman & Hill, 2007). The random coefficient model (RCM), a form of LME modeling, is particularly appropriate when it is believed that there is significant individual variation in both the slopes and intercepts of a population. RCMs incorporate covariance structures that explicitly acknowledge both between- and within-subject sources of variation. The RCM for simple regression can be expressed as:

$${\mathrm{y}}_i=({\beta }_0+b_{0i})+({\beta }_1+b_{1i})t_{ii}+{\mathrm{e}}_{ij}$$

where β₀ = the population intercept;

$$b_{0i}=\text{the random effect for intercept};$$

$${\beta }_1=\text{the population slope};$$

$$b_{1i}=\text{the random effect for slope};$$

and e_ij = within-subject variability. Based on developmental theory and previous empirical research, we hypothesized that there would be significant individual differences in both baseline SEL scores and SEL growth over time, and that as such, an RCM approach would be appropriate. These hypotheses were confirmed by preliminary analyses, which indicated that an RCM model was a superior fit for the present data relative to other LME models with fixed slopes and/or fixed intercepts.

Each of the theoretically informed covariates described previously in the Method were entered into the final RCM as predictors of individual variability in intercepts. Interaction terms between each of these covariates and time were also entered as predictors of individual variability in slopes (SEL change over time).

Maximum likelihood (ML) estimation procedures were employed to produce unbiased estimates. An autoregressive error structure, in which variance is assumed to be constant across all occasions, was also applied. This error structure is highly parsimonious and is appropriate given that measurements were conducted at approximately equal time intervals (three months apart) in the present study (Fitzmaurice et al., 2011, p. 172). Analyses were conducted in RStudio version 0.99.489 (RStudio Team, 2015) using the nlme package (Pinheiro, Bates, DebRoy, Sarkar, & R Core Team, 2018).

Comparison to latent-curve modeling.

There is substantial theoretical overlap between the mixed-effects and latent-curve modeling frameworks for assessessing growth over time. Both frameworks assume an overall mean trajectory for the entire study sample, with random effect estimates for each participant. The primary practical difference is that, in mixed-effects modeling, random effects are specified within extended regression frameworks, whereas in latent-curve modeling, random effects are specified within confirmatory factor analysis models/SEM frameworks (McNeish & Matta, 2017). In the case of the present study, a major advantage of the mixed-effects approach was the ease of specifying and interpreting interactions across timepoints. For an extensive discussion of this topic and other comparisons of mixed-effects versus latent-curve approaches, readers are directed to McNeish and Matta’s work.

Results

SEL descriptive statistics

Correlations among key demographic and SEL variables are displayed in Table 2. Fall SEL scores were significantly, positively correlated with female sex, Hispanic/Latinx race/ethnicity, age, ELL status, and school-wide achievement; and negatively correlated with Black race/ethnicity and special education participation. Spring SEL scores were significantly, positively correlated with CPC prekindergarten participation, female sex, and age; and negatively correlated with special education participation. As expected, children’s SEL scores across time-points were significantly correlated.

Table 2.

Bivariate correlations among key variables

	CPC Prekindergarten	Female	Black	Hispanic/Latinx	Months old	Eligible for free or reduced lunch	English Language Learner	Special education participation	70%+ of school met third grade reading benchmarks	Wave 1 SEL+	Wave 2 SEL+	Wave 3 SEL+
CPC Prekindergarten	--	--	--	--	--	--	--	--	--	--	--	--
Female	0.01	--	--	--	--	--	--	--	--	--	--	--
Black	0.18***	−0.01	--	--	--	--	--	--	--	--	--	--
Hispanic/Latinx	−0.19***	0.01	−0.91***	--	--	--	--	--	--	--	--	--
Months old	−0.10	0.00	−0.06**	0.08***	--	--	--	--	--	--	--	--
Eligible for free or reduced lunch	−0.04	0.04	0.02	0.04	0.02	--	--	--	--	--	--	--
English Language Learner	−0.20***	0.02	−0.85***	0.83***	0.07**	0.01	--	--	--	--	--	--
Special education participation	0.01	−0.18***	−0.14***	0.13***	−0.01	−0.02	0.12***	--	--	--	--	--
High-achieving school	0.04*	0.01	−0.51***	0.49***	0.04*	−0.05*	0.46***	0.09***	--	--	--	--
F12 SEL+	0.02	0.12***	−0.08***	0.09***	0.65***	−0.02	0.09***	−0.20***	0.04*	--	--	--
W12 SEL+	0.11***	0.13***	−0.08***	0.10***	0.63***	−0.10	0.09***	−0.24***	0.05*	0.93***	--	--
S13 SEL+	0.24***	0.13***	−0.03	0.03	0.62***	−0.01	0.03	−0.21***	0.01	0.86***	0.91***	--

^*p<0.05

^**p <0.01

^***p<0.001

⁺= GOLD® (Imputed)

Note: Wave 1= Fall 2012, Wave 2=Winter 2012-2013, Wave 3=Spring 2013

Table 3 display the unimputed individual item scores of the full sample and various subgroups at baseline/prekindergarten entry (Wave 1). Table 4 and Figure 1 displays both the unimputed and imputed total imputed SEL scores of the full sample and subgroups at baseline/prekindergarten entry (Wave 1). The average total score on the SEL subscale of the TS-GOLD was 40.37 (SD = 11.98) in the full sample at baseline. Table 3 also displays the unimputed individual item scores of the full sample and various subgroups at the end of prekindergarten (Wave 3).

Table 4.

GOLD® descriptive statistics and T-tests by subgroup (N=2,630)

	Full Sample		CPC	Control		Boys	Girls		Special education	General education		Eligible for free lunch	Ineligible for free lunch		ELL	Non- ELL
	Mean (S.D.)	N	Mean (S.D.)		p-value	Mean (S.D.)		p-value	Mean (S.D.)		p-value	Mean (S.D.)		p -value	Mean (S.D.)		p-value
Unimputed total SEL score (sum of nine items)
Wave 1	41.06 (13.72)	1,319	40.91 (14.77)	41.47 (10.61)	0.50	39.96 (14.30)	42.08 (13.10)	0.01**	32.46 (15.74)	41.61 (13.41)	0.00***	40.87 (13.78)	42.88 (13.14)	0.11	42.18 (13.82)	40.59 (13.55)	0.06
Wave 2	48.29 (11.80)	1,474	48.89 (12.35)	46.87 (10.26)	0.00***	47.16 (11.96)	49.30 (11.57)	0.00***	40.54 (14.11)	48.86 (11.41)	0.00***	48.12 (11.87)	50.05 (11.00)	0.07	49.02 (12.71)	47.95 (11.34)	0.10
Wave 3	55.44 (12.26)	1,573	57.32 (12.57)	51.73 (10.70)	0.00***	54.34 (12.37)	56.45 (12.08)	0.00***	48.30 (14.51)	56.00 (11.89)	0.00***	55.33 (12.32)	56.47 (11.62)	0.28	54.78 (12.60)	55.77 (12.08)	0.13
Imputed total SEL score (sum of nine items)
Wave 1	40.37 (11.98)	2,630	40.50 (12.88)	40.12 (10.07)	0.44	38.90 (12.38)	41.77 (11.43)	0.00***	32.84 (12.35)	41.16 (11.67)	0.00***	40.25 (12.19)	40.99 (10.75)	0.26	41.81 (11.61)	39.54 (12.12)	0.00***
Wave 2	47.68 (10.74)	2,630	48.49 (11.24)	46.13 (9.54)	0.00***	46.30 (10.90)	48.99 (10.43)	0.00***	39.78 (11.54)	48.50 (10.32)	0.00***	47.56 (10.93)	48.31 (9.64)	0.20	48.91 (10.92)	46.97 (10.58)	0.00***
Wave 3	54.93 (10.99)	2,630	56.84 (11.19)	51.27 (9.62)	0.00***	53.43 (11.09)	56.35 (10.71)	0.00***	47.92 (11.60)	55.66 (10.67)	0.00***	54.91 (11.19)	55.02 (9.84)	0.85	55.34 (11.00)	54.69 (10.99)	0.14

^*p<0.05

^**p<0.01

^***p<0.001

Note: Wave 1= Fall 2012, Wave 2=Winter 2012-2013, Wave 3=Spring 2013

Figure 1. Mean, Imputed Total SEL Scores on the GOLD®, by CPC status.

Note: Wave 1= Fall 2012, Wave 2=Winter 2012-2013, Wave 3=Spring 2013

Table 4 displays the total imputed SEL scores of the full sample and various subgroups at the end of prekindergarten (Wave 3). The average total SEL score was 54.93 (SD = 10.99) in the full sample at year’s end.

RCM

Model Assumptions and Fit.

No deviations from RCM assumptions (multivariate normal distribution of random effects, normal distribution of errors, independence of errors) were evident based on descriptive statistics and visual inspection of residual plots. We also ran general linear models parsing variance components (e.g., variance among students and variance between schools) to assess for potential multicollinearity. Results revealed relatively small intraclass correlations at the school level (r = 0.09 at Wave 1 and r = 0.16 at Wave 3). Notably, the present study’s RCM model accounts for clustering by incorporating school-level variables and estimating random effects for each cluster.

The results of the model estimations for the baseline RCM with no covariates and the final model with added covariates are presented in Table 5. Between the baseline and final models, variance of the intercept decreased from 137.40 to 70.75. Variance of the slope decreased from 9.46 to 6.55. Total error variance increased from 2.76 to 3.75. Overall, these results indicate that the introduction of covariates explained a substantial proportion of the variation in baseline SEL scores and changes in SEL scores over time.

Table 5.

Parameter estimates

Parameter	Baseline RCM+†	Cohen’s d	Final RCM++†	Cohen’s d
Fixed effects
β0	11.72***	--	40.91***	--
β1 – time	3.08***	3.31	5.97***	0.51
β2 – CPC	--	--	0.80*	0.08
β3– Female	--	--	1.95***	0.23
β4– Black	--	--	−0.05	0.00
β5– Hispanic	--	--	0.63	0.03
β6– ELL	--	--	0.75	0.04
β7– Eligible for free lunch	--	--	−2.27***	−0.19
β8– Special education	--	--	−8.39***	−0.56
β9– Age	--	--	1.14***	1.75
β10†– School-level achievement	--	--	1.33***	0.13
β11– Date baseline data was submitted	--	--	−1.11**	−0.12
β12– CPC*time	--	--	2.51***	0.56
β13– Female*time	--	--	0.02	0.00
β14– Black*time	--	--	−0.53	−0.05
β15– Hispanic*time	--	--	−0.37	−0.04
β16– ELL*time	--	--	−0.15	−0.02
β17– Free lunch*time	--	--	0.26	0.05
β18– Special education*time	--	--	0.31	0.05
β19– Age*time	--	--	−0.07***	−0.23
β20– School-wide achievement*time	--	--	−0.55***	−0.12
β21– Date baseline data was submitted*time	--	--	0.42***	0.10
Random effects
b0i	137.40	-	70.75
b1i	9.46		6.55
Model fit
BIC	50663.71		48402.06
Error variance	2.76		3.75

^*p<0.05

^**p<0.01

^***p<0.001

^+:Baseline RCM model: y_ij = β_{0i +} β_1itime_1ij + ε_ij

^++:Final RCM model: y_ij = β_{0i +} β_1itime_1ij + β₂CPC_i + β₃female_i + β₄black_i + β₅Hispanic_i + β₆ELL_i + β₇free_lunch_i + β₈special_education_i + β₉months_old_i + β₁₀school_achievement_i + β₁₁baseline_date_i + β₁₂CPC:time_i+ β₁₃female:time_i + β₁₄black:time_i + β₁₅Hispanic:time_i + β₁₆ELL:time_i + β₁₇free_lunch:time_i + β₁₈special_education:time_i + β₁₉months_old:time_i + β₂₀school_achievement:time_i + β₂₁baseline_date:time_i + ε_ij

^†:Age variable was centered to aid interpretability of the intercept.

SEL at baseline.

Table 5 displays the results of the final RCM, including individual differences at baseline. The average total SEL score at baseline was 40.91 (p< 0.001) for participants who were in the comparison group, male, not black or Hispanic, not in special education, and not eligible for free lunch.

Demographic subgroups.

Readers will recall that our first set of hypotheses focused on individual differences in SEL upon prekindergarten entry. Results indicated that girls were rated 1.95 points higher, on average, than boys at baseline (p<0.001). The latter effect was in the small- to medium-range (Cohen’s d= 0.23; see Cohen, 1988; Durlak, 2009 for effect size interpretation guidelines). Participants who were eligible for free lunch were rated 2.27 points lower, on average, than children who were not eligible at baseline (p<0.05), an effect that was in the small range (Cohen’s d = −0.19). Participants receiving special education services were rated 8.39 points lower, on average, than participants in general education at baseline (p< 0.001), an effect that was in the medium range (Cohen’s d=−0.56). Participants were rated an average 1.14 points higher for every additional month of age at baseline (p< 0.001), an effect that was in the small range (Cohen’s d = 0.13). Participants who attended high-achieving schools were rated an average 1.33 points higher than children from lower-achieving schools at baseline (p< 0.001), an effect that was in the small range (Cohen’s d = 0.13). Finally, participants whose teachers submitted their baseline ratings after October were rated 1.11 points lower, on average, than participants whose teachers submitted their ratings on time (p<0.01), an effect that was in the small range (Cohen’s d =−0.12). There were no statistically significant differences in baseline scores based on black or Hispanic identity, English Language Learner status (p>0.05).

CPC versus comparison.

The average baseline SEL rating for CPC participants was 0.80 points higher than for comparison group participants (p<0.05), an effect that was in the small range (Cohen’s d=0.08).

SEL over the prekindergarten year.

Table 5 displays the results of the final RCM, including individual differences in SEL growth over the course of the preschool year. Teacher-rated SEL increased by an average of 5.97 points (p<0.001) per time-point (Waves 2 and 3) for participants who were in the comparison group, male, not black or Hispanic, not in special education, and not eligible for free lunch. This effect was in the medium range.

Demographic subgroups.

Readers will recall that our first set of hypotheses focused on individual differences in SEL growth over time. Results indicated that, on average, participants whose teachers submitted their baseline ratings after October were rated 0.42 points higher per time-point (Waves 2 and 3; p<0.01) than participants whose teachers submitted their ratings earlier. The latter effect was in the small range (Cohen’s d = 0.10). Participants who attended high-achieving schools were rated an average of 0.55 points lower per time-point (p<0.001) than participants who attended lower achieving schools, an effect that was in the small range (Cohen’s d=−0.12). Older participants also exhibited slightly lower average rates of SELgrowth over time (0.07 points less per assessment for every additional month of age, p< 0.001), an effect that was in the small range (Cohen’s d=−0.23). There were no statistically significant differences in SEL growth over time by gender, black or Hispanic identity, English Language Learner status, free lunch eligibility, or special education status (p>0.05).

CPC versus comparison.

Readers will recall that our second set of hypotheses focused on differences in SEL growth over time between CPC and comparison group members. Results indicated that, on average, CPC participants were rated 2.51 points higher per time-point (Waves 2 and 3) than comparison group participants (p<0.001). The effect size for the CPC-by-time interaction term was in the medium range (Cohen’s d=0.56).

Discussion

Over the last several decades, there has been growing interest in interventions to support the holistic development of children affected by poverty, particularly in the domain of SEL. However, few prospective studies have examined how SEL unfolds over time in different subgroups of low-income children. Even fewer have empirically investigated whether multicomponent prekindergarten programs (e.g., CPC, Head Start) boost children’s SEL. The present study aimed to address these knowledge gaps by investigating SEL across the prekindergarten year in a large (N=2,630), racially and ethnically diverse sample of low-income children, some of whom participated in the CPC intervention.

This discussion will cover: (a) findings related to baseline SEL at prekindergarten entry, including differences between demographic subgroups, and between the CPC and comparison groups; (b) findings related to SEL growth across the prekindergarten year, including differences between demographic subgroups, and between the CPC and comparison groups; and (c) strengths, limitations, and future directions. Potential implications for future research, practice, and policy will be discussed throughout.

Demographic subgroups

Our first set of hypotheses focused on subgroup differences in baseline socio-emotional skills and SEL growth over time in this diverse, low-income sample.

Baseline SEL.

Family poverty.

All of the present study’s participants attended prekindergarten programming in low-income school districts, and the majority met criteria for reduced-cost lunch. Nonetheless, there was still inter-individual variation in the depth of poverty that paraticipants experienced. Consistent with past work (e.g., Comeau & Boyle, 2018; Duncan, Brooks-Gunn, & Klebanov, 1994; Eamon, 2001; Qi & Kaiser, 2003; Knapp, Ammen, Arstein-Kerslake, Poulsen, & Mastergeorge, 2007; Li, Johnson, Musci, & Riley, 2017; Yoshikawa, Aber, & Beardslee, 2012) , we hypothesized that children from the highest poverty families (as measured by free lunch eligibility) would receive lower SEL ratings at baseline than children from less impoverished families. Results supported this hypothesis, with an effect size in the small range. This finding underscores the heterogeneity of poverty, even within predominately low-income schools. Exposure to adverse events, environmental and family stressors tends to increase with poverty severity, with potential repercussions for children’s development (Council on Community Pediatrics, 2016; Evans & Cassells, 2014; Evans & English, 2003; Luby et al., 2013; Mersky, Janczewski, & Topitzes, 2016). Conducting family needs assessments at the start of prekindergarten may be an effective strategy for identifying children who are affected by the highest levels of socioeconomic risk, and who may also benefit from additional SEL supports (Reynolds, Hayakawa, Candee, & Englund, 2016).

Sex.

Based on previous research (e.g., (Barbarin, 2013; Lambert, Kim, & Burts, 2014; Raikes, Robinson, Bradley, Raikes, & Ayoub, 2007), we hypothesized that boys would receive lower SEL ratings than girls upon prekindergarten entry. This hypothesis was supported, with an effect size in the small- to medium range. In this sample, sex differences in neurobiological development (e.g., slower maturation of stress-regulation circuits among boys) likely interacted with systemic social inequities to shape boys’ socioemotional skills before they entered prekindergarten (Golding & Fitzgerald, 2017; McKinney, Fitzgerald, Winn, & Babcock, 2017). It is also possible that implicit bias negatively affected teachers’ ratings of boys, especially boys of color (DiCarlo, Baumgartner, Ota, & Jenkins, 2015; Gilliam, Maupin, Reyes, Accavitti, Shic, 2016). Additional research on the latter topic is needed to clarify the extent to which baseline sex differences in SEL reflect true phenomena versus reporter biases.

ELLs, race, and ethnicity.

Drawing on past work with a nationally representative sample (Lambert et al., 2014), we also hypothesized that ELLs would receive lower SEL ratings at baseline than monolingual English speakers. Results did not support this hypothesis. On the contrary, there were no significant differences in ELLs’ versus non-ELLs’ baseline SEL ratings at this low-income sample. Similarly, there were no significant differences in SEL based on whether participants identified as black or Hispanic. These findings may reflect unique protective factors for children from multicultural backgrounds (e.g., strong family ties, extended social support networks, increased cognitive flexibility in the context of multilingualism) (Han, 2010; Oades-Sese, Esquivel, Kaliski, & Maniatis, 2011; Zhou et al., 2012). They also likely reflect the social-ecological contexts of the surveyed school sites, where cultural diversity among students and staff was the norm, not the exception. Both the Chicago Public Schools district and the CPC program frequently facilitated professional development on multicultural education, language, and related topics. This broader context likely enhanced teachers’ abilities to evaluate their ELL students’ SEL skills in an accurate, culturally competent manner at the start of prekindergarten.

Special education.

Our hypothesis that special education participants would enter prekindergarten with lower SEL skills than children in general education was supported, with an effect size in the medium range. This finding reflects the complex socio-emotional needs of children with emergent disabilities, and the importance of providing tailored interventions to this group beginning at school entry (Emerson, Einfeld, & Stancliffe, 2010; Gallagher & Lambert, 2006; Kim, Carlson, Curby, & Winsler, 2016; McIntyre, Eckert, Fiese, Reed, & Wildenger, 2010; Merrell & Holland, 1997).

Age.

As hypothesized, relatively older children entered prekindergarten with higher SEL skills than their younger counterparts, with an effect size in the small range. This finding is consistent with research indicating that early childhood is a period of rapid brain and socio-emotional development, such that even a few months age difference may affect teacher ratings of SEL (Dawson, Ashman, & Carver, 2000; Tierney and Nelson, 2003).

School-level achievement.

As hypothesized, children enrolling in higher-achieving schools (a proxy for community-level advantage and climate) entered prekindergarten with higher SEL scores than children enrolling in lower achieving schools, with an effect size in the small range. This finding likely reflects the impact of neighborhood-level advantages and protective factors on young children’s socioemotional development, even before they enter a formal school environment.

Additional analyses.

Finally, though not related to one of our a priori hypotheses, we investigated whether children’s baseline SEL ratings differed based on when their scores were submitted, after observing that some teachers submitted ratings later in the Fall semester. We found that participants whose teachers submitted their baseline ratings after October 31 received slightly lower SEL ratings than children whose teachers submitted their ratings on-time. It is possible that teachers who submitted their ratings late were more highly stressed or disorganized than teachers who submitted on-time, and thus had more negative views of their students’ baseline SEL skills. It is also possible that other school- or community-level factors affected teachers’ workflows. At any rate, this is is an area that merits investigation in future work.

SEL growth over time.

Family poverty.

Contrary to our hypothesis and the results of previous research (e.g., Lambert et al., 2014; Weiland & Yoshikawa, 2013), children from the highest poverty families (as measured by eligibility for free lunch) did not exhibit greater SEL growth during the prekindergarten year, relative to their peers. The time-by-free-lunch interaction term was statistically trending (p<0.10) but not significant; however, at the end of prekindergarten, there were no significant differences in SEL ratings based on free lunch eligibility. Readers will recall that there was a small (yet significant) difference in eligible versus non-eligible children’s baseline SEL ratings. Take together, these results suggest that prekindergarten participation (either in regularly offered district programming, or multi-component programming such as CPC) is associated with similar SEL gains for children affected by varying levels of family poverty.

Sex.

Contrary to our hypothesis and the results of previous work (e.g., Lambert et al., 2014), boys did not demonstrate less SEL growth over time than girls (e.g., Lambert et al., 2014). Readers will recall that, in this study, boys received lower baseline SEL ratings than girls. No significant sex differences in SEL growth over time were detected, meaning that boys’ SEL ratings were still significantly lower than girls’ at the final time-point. Once again, there is a critical need for additional research to assess the extent to which these findings reflect true sex differences in SEL, versus potential implicit biases towards boys (and especially boys of color) (e.g., Gilliam, Maupin, Reyes, Accavitti, & Shic, 2016). In either case, these findings underscore the need for additional research that identifies the unique risk and protective processes that affect boys’ SEL, as well as school- and home-based intervention strategies.

ELLs, race, and ethnicity.

We also hypothesized that ELLs would exhibit greater SEL growth over the course of prekindergarten, in keeping with previous research (e.g., Lambert et al., 2014). However, we found no statistically significant differences in growth over time between ELLs and non-ELLs students. Similarly, we found no differences in growth over time based on whether participants identified as black or Hispanic. These findings indicate that children in this sample exhibited similar growth over the prekindergarten year regardless of their ELL status or ethnicity. As previously discussed, these findings also likely reflect the richly multicultural environment in the included school sites.

Special education.

We also posited that special education students would exhibit less SEL growth over time than general education students. In contrast to this hypothesis, we found no statistically significant differences in SEL across groups. As readers will recall, special education students were rated as having lower SEL skills than their peers at baseline; this pattern was evident again at the end of prekindergarten. Taken together, these findings suggest that special education students made socio-emotional gains throughout prekindergarten, and did not fall even further behind their peers – likely due to the intervention services they were receiving. Nonetheless, these intervention services did not completely ‘close the gap’ in SEL that was evident between groups at the start of the year. These findings highlight the importance of early screening and ongoing intervention for young children with disabilities.

Age.

We hypothesized that there would be no significant differences in SEL growth over time based on age at prekindergarten entry. This hypothesis was not confirmed, with results indicating that older children actually exhibited slightly lower SEL growth over time than younger children, an effect that was in the small range. This difference was statistically significant; however, given that the average between-groups difference was less than one point, it was likely not clinically meaningful. Nonetheless, one possible explanation for this finding is that teachers may have geared their classroom-wide SEL practices more toward the needs of younger students, rather than the needs of more developmentally mature students.

School-level achievement.

We also hypothesized that children attending high-achieving schools (a proxy for community-level resources and climate) would exhibit greater SEL growth over time than children attending lower achieving schools. This hypothesis was not confirmed, with results indicating that children from higher achieving schools actually were rated slightly lower per time-point than their counterparts at lower achieving schools, an effect that was in the small range. This difference was statistically significant; however, given that the average difference between groups was less than one point, it likely did not represent a clinically meaningful difference. Nonetheless, one possible explanation for this finding is that teachers at higher achieving schools may have held higher expectations for children’s SEL, rating them more severely over the course of the year than their colleagues at other schools.

CPC versus comparison groups

Baseline SEL.

As previously described, significant efforts were made to match CPC and comparison group sites on key characteristics using propensity score matching procedures. Examination of both groups’ characteristics indicates that these procedures were generally successful, though there were small differences in regards to race/ethnicity, ELL status, single parent status, and baseline literacy scores (Table 1). At baseline, CPC participants received slightly higher SEL ratings than comparison group members (Table 5 and Figure 1). This difference was statistically significant; however, given that the average difference between groups was less than one point, it did not likely represent a clinically meaningful difference.

SEL growth over time.

Based on previous research (e.g., Reynolds, Richardson, et al., 2016; Richardson et al., 2017), we hypothesized that CPC participants would exhibit significantly greater SEL growth across the prekindergarten year than comparison group members. This hypothesis was supported, with an effect size in the medium range. CPC participants’ SEL scores were 10.30% higher, on average, than those of comparison group participants at the end of the prekindergarten year (Table 5 and Figure 1). This finding indicates that multicomponent prekindergarten programs like CPC have the potential to substantially enhance the SEL of low-income children. Multicomponent programs are not necessarily a replacement for more intensive, skills-based SEL interventions (e.g., Prekindergarten PATHS, the Incredible Years), which may be particularly beneficial for children with significant socio-emotional challenges (Domitrovich, Greenberg, Kusche, & Cortes, 2004; Mondi, Giovanelli, & Reynolds, under review; Webster-Stratton & Hammond, 1997). However, multicomponent programs like CPC may be ideal frontline preventive interventions, given their focus on enhancing protective factors at multiple social-ecological levels (e.g., school, home, neighborhood) and promoting children’s holistic development (Mondi, Giovanelli, & Reynolds, under review; Reynolds, 2000). Importantly, previous research has indicated that multicomponent interventions like CPC yield high returns on financial investment, and can be implemented at large scales with high-needs populations (e.g., Reynolds et al., 2011). Thus, the present study’s results suggest that increasing access to multicomponent interventions may be an effective (and ultimately cost-effective) strategy for enhancing children’s SEL and school readiness in low-income contexts.

Strengths, Limitations, and Future Directions

Strengths.

The present study has a number of strengths. First, it utilized a large (N=2,630) sample of low-income, racially and ethnically diverse prekindergarteners, a population that has been understudied in prospective research. The large sample size increases the likelihood that results are valid and generalizable to similar populations.

Second, the intervention and comparison group sites were generally well-matched using propensity score procedures, with minimal group differences in SEL scores at baseline. This increase confidence that the findings related to CPC participation can be attributed to the intervention, rather than demographic differences in the two samples.

Third, unlike many studies that have utilized brief socio-emotional measures in the context of cross-sectional research designs, the present study used repeated measures data from a SEL measure with strong psychometric properties (GOLD®). This allowed for examination of participants’ SEL growth over the course of the prekindergarten year (versus purely focusing on point-differences at the end of the year).

Limitations and Future Directions.

It is also important to acknowledge the limitations of the present study. First, SEL entirely teacher-reported using the GOLD® measure, which is a valid, yet broad measure of children’s developing skills. It would be beneficial for future work to incorporate parent reports and direct observation measures to establish interrater reliability and to obtain a more comprehensive picture of SEL. Future studies should also examine whether informant characteristics (e.g., age, years of teaching experience) impact ratings of children’s SEL, particularly in light of concerns about potential implicit bias.

Second, because all participants attended a prekindergarten program (either CPC or public prekindergarten), it is not possible to evaluate how much of children’s SEL over time is related to typical developmental gains versus prekindergarten participation. This is an especially important question in light of burgeoning public debates about prekindergarten access. Future studies should consider comparing public prekindergarten and CPC participants to a third group of children who are not enrolled in any programming. Furthermore, within studies comparing differnet types of prekindergarten (e.g., typical district programming versus CPC), it would be valuable to incorporate measures of intervention fidelity and classroom quality, to assess similarities and differences in children’s experiences and functioning across settings.

Third, GOLD® data was not collected beyond the prekindergarten year in the present study, making it difficult to ascertain whether prekindergarten SEL gains were sustained over time. Future studies should extend data collection through Kindergarten and beyond. It would be particularly valuable to explore other factors that are associated with maintenance of SEL gains (e.g., intervention dosage, family support) within multilevel modeling or latent-curve modeling frameworks. It would also be beneficial to build on previous research connecting SEL to other domains of wellbeing (e.g., academic achievement) (e.g., Jones et al., 2015).

Fourth and relatedly, this study should be replicated with children from different geographic regions and racial/ethnic backgrounds, to clarify the generalizability of the present findings.

A final important direction for future research is examination of the mechanisms that underlie the effects of CPC participation on children’s SEL (e.g., cognitive advantage, socio-emotional advantage, school support) (Reynolds, 2000). Identifying salient mechanisms of change could inform efforts to tailor the CPC program to the needs of particular populations to maximize SEL gains.

Conclusion

The development of strong socio-emotional skills in early childhood is critical for later school success and psychological wellbeing. Yet, relatively few studies have prospectively studied how SEL unfolds over the course of the prekindergarten year. The present study is one of the first to do so within the context of a large, low-income, racially and ethnically diverse sample of prekindergartners. Overall, results indicated that multicomponent prekindergarten programs (e.g., CPC) have the potential to enhance the SEL of young children affected by socioeconomic adversity. Results also suggested that some groups of children (e.g., boys, children from the highest poverty families) may benefit from socio-emotional screenings, family needs assessments, and more intensive SEL intervention. Moving forward, there is a critical need for additional studies that will examine children’s SEL within rigorous longitudinal designs, with the aim of informing practice and policy initiatives to promote child wellbeing.