Educational Growth Trajectories in Adulthood: Findings from an Inner-City Cohort

Abstract

Educational attainment is typically examined as a static status. As adult learners have become the new trend in higher education, the changes in educational attainment in adulthood warrant more attention. Using data from the Chicago Longitudinal Study (CLS), an ongoing panel investigation of 1,539 children, predictors of educational growth trajectories in adulthood were investigated. Of the study sample (N = 1,418), 51.8% were women, 93.2% were Black, 6.8% were Hispanic, 83.4% were eligible for free lunch between birth and age 3. The average age of the study sample in June 2015 was 35.1, ranging from 34.4 to 36.6. Hierarchical linear modeling (HLM) was used to analyze the changes in educational attainment between ages 24 and 35. Findings indicate that mothers not completed high school by child’s age 3 and days of absence at school were significantly associated with lower educational attainment at age 24. Classroom adjustment, student college expectations, 8^th grade reading scores, and on-time high school graduation were significantly associated with higher educational attainment at age 24. Classroom adjustment, 8^th grade reading score, and on-time high school graduation were significantly associated with a positive growth of education between ages 24 and 35. Findings suggest that improving academic achievement and socioemotional learning skills in elementary and middle school and promoting on-time high school graduation are likely to increase one’s chances to continue pursuing higher education in adulthood for Black low-income children.

Introduction

Educational attainment is essential because it is the major pathway to lifelong well-being, such as stable employment, financial security, and healthy life (e.g. Baker, 2014; Heckman et al., 2018; Mazumder, 2014; Roy et al., 2020; Zajacova & Lawrence, 2018). The importance of educational attainment is accentuated by its role in promoting cardiovascular health and healthy aging, which leads to not only improved economic well-being but reduced disability and disease over the life course (Community Preventive Services Task Force, 2015; Infurna et al., 2020; Roy et al., 2020). Education can be an ongoing process after young adulthood. However, education has been examined as a static status rather than a changeable status. In the past decades, adult learners, or non-traditional students, have increased rapidly in higher education. In 2017, students over 25 years old account for 40% of students enrolled in all of the postsecondary institutions, including 23.5% of full-time students and 65.6% of part-time students (National Center for Education Statistics, 2018). The growing number of adult learners warrants further consideration.

Continuing higher education for adult learners is especially important for historically vulnerable and economically disadvantaged populations because they are more likely to experience unequal educational opportunities than others while they advance in the education system (Mazumder, 2014; Sawhill & Reeves, 2016; Sawhill et al., 2012; Timothy, 2016). Education is the major gateway for a child’s upward mobility. However, success begets further success in the education system. The structure of the education system in the U.S. has a large effect on who succeeds and who does not (Corak, 2016; Timothy, 2016). Children born advantaged retain a large advantage at the end of early childhood, and the pattern persists in subsequent stages (Sawhill & Reeves, 2016). Some evidence also indicates that the gaps in levels of educational attainment by parental income grow in the early years, through K-12, and into higher education (Duncan & Murnane, 2016).

Black and economically disadvantaged populations stand out in the vulnerable populations because of the large and persistent gap in economic status between Blacks and Whites in the United States (Mazumder, 2014). Studies found that almost 50 percent of Black children born into the bottom 20 percent of the income distribution were in the same position as adults, but that only 23 percent of White children born in that quintile were (Timothy, 2016). Educational attainment can be particularly crucial for economic mobility for Black and economically disadvantaged populations because higher education might increase an individual’s job opportunities and chances of promotion, and then lead to better income, health, and other positive lifelong outcomes (Heckman et al., 2018; Herd, 2010; Mazumder, 2014; Sawhill et al., 2012; Timothy, 2016). Improving educational attainment for disadvantaged populations may prevent further income and health disparities because education is greatly connected to income and health (Reardon, 2011; Sawhill et al., 2012; Zajacova & Lawrence, 2018). Using a longitudinal data set of a Black low-income cohort from an inner-city, the present study investigated childhood predictors of educational growth trajectories in adulthood.

Educational Attainment as a Long-Term Process

Bronfenbrenner’s bioecological model of human development, the process–person–context–time (PPCT) model, emphasizes the dynamic interplay of processes across levels of analysis, contexts, and time to explain how an individual develops over time (Bronfenbrenner & Evans, 2000; Magnusson & Cairns, 1996). Based on this framework, it is important to examine how the proximal processes occurred between a developing child and their nested spheres of environmental influences (micro-, meso-, exo-, and macrocontext) to jointly influence and shape the development of educational attainment over time. Based on the PPCT model, educational attainment is a developmental process rather than simply an intellectual achievement. In addition to individual characteristics, the factors contributing to educational attainment come from different levels of contexts, such as school, ethnicity, culture, neighborhood, economic systems, and social policy (Bronfenbrenner & Evans, 2000; Cassells & Evans, 2020).

Although educational attainment is conceptualized as the result of a long-term process, the majority of the empirical evidence on its determinants is derived from longitudinal data beginning in adolescence instead of birth or early childhood (Dupéré et al., 2015). Longitudinal studies have indicated that early family circumstances and school experiences play important roles in placing some children on a high-risk trajectory for school failure (Alexander et al., 2001; Duchesne et al., 2008; Magnuson et al., 2016; Ou & Reynolds, 2008). The bioecological model, as one of the frameworks of developmental science, recognizes the developmental foundation of educational trajectories that begin early in life. This framework can better explain how these developmental foundations early in life interplay with later proximal factors. In addition to integrating factors that may contribute to educational attainment, it is also important to investigate how changes in educational attainment occur over time in adulthood. Lack of prospective longitudinal data may be one of the reasons that the long-term process of educational growth trajectories have not been addressed in the literature.

Predictors of Education

There are numerous studies on the predictors of academic achievement, school dropout, and educational attainment (Magnuson et al., 2016; Rumberger, 2011; Rumberger & Lim, 2008). We briefly review relevant studies and discuss them in the context of the bioecological model.

Sociodemographic Factors

The associations between sociodemographic factors (e.g. gender, race, and SES) and education are well established (Alexander et al., 2014; Timothy, 2016). These factors correspond to the person's characteristics in the bioecological model and they are also closely connected to the macrocontexts. The gaps in educational attainment by gender, race, and family income have been discussed widely (Corak, 2016; Duncan & Murnane, 2016; Fortin et al., 2015; Sawhill & Reeves, 2016). In terms of gender differences in education, studies have found that girls have more advanced reading skills, have advantages in social skills and classroom behavior, and obtain higher grades in school than boys (Fortin et al., 2015; Loveless, 2015). Race/ethnicity and socioeconomic status are powerful predictors and they can shape risks and opportunities and moderate how proximal processes influence development (Cassells & Evans, 2020). Whites and Asians have higher educational attainment compared with Blacks and Hispanics (de Brey et al., 2019). The differences by race/ethnicity are highly correlated with the differences by family SES (Timothy, 2016). Children from higher SES families have higher educational attainment than those from lower SES families (Duncan & Murnane, 2016; Reardon, 2011; Rumberger, 2010). The social class background was found to have a strong effect on college completion (Rumberger, 2010). Recognizing the inequality of economic and educational opportunities resulting from family income, early childhood interventions are targeting to improve low-income children's academic skills and human capital development (Magnuson & Duncan, 2016). Sociodemographic factors, especially race/ethnicity and family income, are embedded in macrocontexts and have direct and indirect powerful effects on educational attainment. Due to their connection with cultural and structural inequality, such factors are usually treated as covariates in the analyses.

Cognitive and Noncognitive Skills

Individual factors are the "developmentally structuring characteristics" in the bioecological model, which influence an individual’s interactions with the environment. Individual factors can be examined via cognitive and noncognitive skills. Cognitive skills, such as Intelligence Quotient (IQ) and academic performance, have received the greatest attention historically (e.g. Rumberger, 2011). For example, academic preparation and performance are associated with college attendance (Eccles et al., 2004; Martinez & Klopott, 2005). Noncognitive skills, such as socioemotional learning skills and behavioral indicators, have gained attention increasingly in the past decade (Lee & Stankov, 2018; Levin, 2012; Ou & Reynolds, 2016; Rosen et al., 2010). Socioemotional learning skills are found to be positively associated with academic achievement, education (Farrington et al., 2012; Gutman & Schoon, 2013; Lee & Stankov, 2018; Magnuson et al., 2016; Rumberger & Lim, 2008), earning (Lleras, 2008), and healthy behaviors (Chiteji, 2010). Behavioral indicators, such as antisocial behavior and risky behavior, are found to be negatively associated with education (Evensen et al., 2016; Lee et al., 2008; Lleras, 2008; Owens, 2016; Rosen et al., 2010). Students’ plans for college and education aspirations are positively associated with college attendance and macrocontexts (e.g. culture, economic environment, and lack of diversity in the education system) can play important roles in influencing those factors (Browman et al., 2017; Eccles et al., 2004; Fan & Wolters, 2014; Lee & Stankov, 2018; Lee et al., 2008; Phinney et al., 2006),.

School and Family Functioning

School experiences (e.g. school absence) and family functioning (e.g. parent's expectations) are identified as important predictors of education. The effects of such factors can be explained through proximal processes in micro- and mesocontexts in the bioecological model. Absenteeism (Ansari et al., 2020; Balfanz & Byrnes, 2012; Morrissey et al., 2014) and school mobility (Han, 2014; Herbers et al., 2013; Sandstrom & Huerta, 2013) have been found to have negative effects on academic achievement and education. However, the effects of some school-related factors on education, such as school quality and school resources, are mixed (Autor et al., 2016; Rumberger & Lim, 2008). Further studies might help to untangle the relations.

Family functioning and resources (e.g. parental expectations and parental involvement) are found to be positively associated with educational outcomes (Castro et al., 2015; Hill & Tyson, 2009; Huat See & Gorard, 2015; Jeynes, 2007; Rumberger & Lim, 2008). Family resources are closely tied to family income, and they are powerful predictors of higher educational attainment (Duncan & Murnane, 2016; Rumberger, 2011). Parental involvement has been studied extensively, and the positive impacts of parental involvement on a child’s education are reported (Castro et al., 2015; Huat See & Gorard, 2015). There are intervention programs targeting parents to improve outcomes for children living in poverty (Morris et al., 2017).

Some limitations in the literature are evident. First, educational attainment results from a long-term process of interactions among an array of factors (Cabrera et al., 2005). It is critical to acknowledge that the overall impact emerges from the dynamic balance among all levels of influences in the environment. However, the majority of the studies in education have focused on one or two domains and overlooked the influences of multiple domains. Second, most studies focus on factors that occurred in high school. Few studies have assessed early factors that might predict educational attainment (Eccles et al., 2004). Finally, educational attainment has been examined as a stable status rather than a changeable status. As adult learners have increased rapidly in higher education, the change of educational attainment in adulthood warrants further investigation. Particularly it might be beneficial for historically vulnerable and economically disadvantaged populations. Longitudinal studies spanning from early childhood to adulthood may provide insights on the determinants of educational growth trajectories in adulthood. Educational attainment in adulthood has been collected at multiple waves in a few longitudinal studies, such as the Chicago Longitudinal Study (CLS), the Abecedarian Project, and the High/Scope Perry Preschool program (Cannon et al., 2017; Reynolds et al., 2018). However, changes in educational attainment in adulthood have not been examined.

The Present Study

Using data from the Chicago Longitudinal Study (CLS), the present study is unique in several respects. First, educational growth trajectories (changes in educational attainment) in adulthood are investigated. Growth trajectories of academic performance, such as reading and math achievement, have been examined in studies (Aikens & Barbarin, 2008; Choi et al., 2016; Stipek & Valentino, 2015), but growth trajectories of educational attainment have never been examined. Although education is considered a long-term process, education has been examined as a constant rather than a variable. The present study contributes to the field by advancing knowledge of the predictors of educational growth trajectories by including the “time” in the analyses, and whether the predictors of initial or final educational attainment are different from those of changes of educational attainment in adulthood. Different from previous studies in understanding educational attainment at one point in time, the present study examined the change of educational attainment between ages 24 and 35.

Second, we used a prospective longitudinal cohort design spinning from early childhood to adulthood. Educational attainment is the accumulation of learning experiences over a lifetime. This prospective longitudinal design provides a unique opportunity to explore how the growth of education in adulthood might be affected by early life experiences. Moreover, we examined factors from multiple contexts using the process–person–context–time (PPCT) model. This is different from other studies that focused on one domain, such as academic achievements or school engagement. Including all aspects of the PPCT model would provide a more comprehensive view of the educational growth trajectories. Although educational gaps result from structure inequality are unlikely to completely close, it is possible to reduce the disparities in education via intervening determinants of education (e.g. socioemotional functioning) (Metsäpelto et al., 2010; Morris et al., 2017). Identification of a wider range of predictors of educational growth trajectories may contribute to the improvement of educational opportunities for children most in need. Findings may inform policymakers on what components to include in intervention programs to promote educational growth in adulthood.

Third, the study sample of the CLS is Black low-income youth growing up in high-poverty neighborhoods. The educational attainment of the CLS participants is lower than the average Black. For example, 3.8% of the CLS participants completed Bachelor degrees by age 24 in 2004, and 11.9% of the CLS participants completed Bachelor degrees by age 35 in 2015, compared to 6.2% of the Black of 20 to 24 years old completed Bachelor degrees in 2004 (U.S. Census Bureau, 2004) and 25.3% of the Black of 25 to 34 years old completed Bachelor degrees in 2015 (U.S. Census Bureau, 2015). The CLS participants are vulnerable and disadvantaged because they faced multiple barriers to economic opportunity while being Black and came from low-income families (Timothy, 2016). There is a primary focus of prevention efforts to reduce achievement gaps and improve well-being over the life course for this at-risk group (Duncan & Murnane, 2016; Magnuson & Duncan, 2016). The uniqueness of the study sample provides an opportunity to identify factors that might reduce educational disparities for this at-risk group, which might further mitigate income and health disparities. Findings may provide insights into narrowing educational gaps by family income and race/ethnicity and further promoting healthy development.

Finally, the study examined the long-term effects of the Child-Parent Center (CPC) program on educational growth in adulthood for the first time. Long-term effects of the ECE programs in education in adulthood are reported previously (Cannon et al., 2017; Magnuson & Duncan, 2016; McCoy et al., 2017). However, the effects of ECE programs on educational growth in adulthood have never been examined before. Different from other early childhood intervention studies (e.g. High/Scope Perry Preschool program and the Abecardian Project), the CPC program is the only large-scale public program that reported long-term effects into adulthood (Reynolds et al., 2018; Reynolds et al., 2011), and the CPC program is still implemented nowadays. Only a few studies of contemporary state prekindergarten programs have reported enduring effects on middle school and beyond (Bailey et al., 2020). Understanding the effectiveness of the CPC program in educational growth in adulthood may provide insights on how early childhood intervention might help educational growth beyond initial educational attainment for this at-risk population.

Method

Sample and Design

The data were drawn from the Chicago Longitudinal Study (CLS) (Chicago Longitudinal Study, 2005), an on-going investigation of the well-being of a disadvantaged cohort of 1,500 children who attended kindergarten in the Chicago Public Schools in 1985-1986 (Reynolds, 2000). The original sample (N = 1,539) included 989 children who attended the Child-Parent Center (CPC) preschool program and 550 children who participated in alternative public early childhood kindergarten programs.

The Child-Parent Center (CPC) Program

The Child-Parent Center (CPC) program is described fully in previous reports (Reynolds, 2000). A summary of the main features is provided here. The CPC program provides comprehensive educational and family support services from age 3 to age 9 in co-located schools that emphasize small classes, an instructional balance between teacher- and child-initiated learning, a leadership team to support learners and teachers, and family engagement activities in school and at home. All teachers have bachelor’s degrees and are certified in early childhood education. Major components of the intervention include small class sizes; a parent program that includes parenting education, parent room activities, classroom volunteering, and home visitation; and health and nutrition services, including screening and diagnostic services, meal services, and referral by program nurses. Parents are expected to participate in the program for up to half of a day per week through various supported activities. The CPC preschool program is followed by a full-day or part-day CPC kindergarten. The school-age CPC program is open to any child in the school in either first through second grades in 14 sites or first through third grades in 6 sites. Findings have shown significant benefits of CPC participation on multidimensional well-being, such as educational attainment, crime, and public aid participation (Reynolds et al., 2018; Reynolds et al., 2007; Reynolds et al., 2011).

The study sample included 1,418 participants (92.1% of the original sample) with valid data on educational attainment. Of the study sample, 51.8% were women, 93.2% were Black, 6.8% were Hispanic, 83.4% were eligible for free lunch between birth and age 3. The average age of the study sample in June 2015 was 35.1, ranging from 34.4 to 36.6. The high rate of sample retention is owing to the use of multiple sources including administrative records, self-report, and follow-up tracking. Attrition analysis was conducted to compare sociodemographic factors between the study sample and the attrition sample. There were no significant differences between the two groups on most of the factors except some. The attrition group was more likely to be male, receive AFDC by age 3, have missing values on any family risk indicators by age 3, and was less likely to have CPC extended participation than the study sample (See Table 1). Data were collected from participants, parents, teachers, and schools from birth to adulthood through surveys, assessments, and various administrative records. All data collection procedures have been approved by the Institutional Review Board at the University of Minnesota (IRB ID 0511S77508; project title: Longitudinal effects of extended early childhood intervention). All participants provided written or oral informed consent, and all data were de-identified for analysis.

Table 1.

Characteristics by attrition status

Characteristics	N	Original sample (n=1539)	Study sample (n=1418)	Attrition sample (n=121)
Percent females	1531	50.2	51.8	29.2***
Percent Black	1539	93.0	93.2	90.1
Family risk index (0-7) by child’s age 3	1539	4.52	4.50	4.81
Percent four or more risk factors by child’s age 3	1531	72.9	72.5	77.7
Percent mother not completed high school by child’s age 3 1	1475	54.0	53.6	58.8
Percent single parent by child’s age 3 1	1482	75.6	75.5	77.1
Percent mother not employed by child’s age 3 1	1342	62.9	62.4	69.9
Percent ever reported receiving free lunch by child’s age 3 1	1445	82.7	82.5	85.3
Percent ever reported receiving AFDC by child’s age 3 1	1440	62.2	61.5	72.0*
Percent having 4 or more children at home by child’s age 31	1482	17.3	17.5	14.3
Percentage children in school area in which 60% or more of children reside in low-income families	1531	76.0	76.2	73.6
Percent mother was teen at child’s birth 1	1493	16.7	17.0	12.1
Percent missing on any family risk indicators	1539	16.2	14.4	38.0***
Percent any child welfare case history by child’s age 3 1	1411	4.1	4.0	6.0
Percent low birthweight (< 2500grams) 1	1456	12.4	12.6	10.2
Percent CPC preschool participation	1539	64.2	64.8	57.9
Percent CPC school-age participation	1539	55.2	55.6	50.4
Percent CPC extended participation	1539	35.9	36.8	25.6*

Note. Note.

^1.Means reported before imputation for missing data.

^***p < .001

^**p < .01

^*p< .05

Measures

Educational Attainment

Data on educational attainment were obtained from administrative records, including National Student Clearinghouse and school youth attended, and were supplemented with self-reports at age 22/24 and age 35. Educational attainment was measured through the total year of education. The total year of education was coded as a continuous variable, ranging from 7 to 22. Obtaining a GED and high school completion were coded 12. College attendance was coded based on credits earned. Thirty credits were treated as one year of college attendance. Associated degrees were coded as 14. Bachelor's degrees were coded as 16. Master's degrees were coded as 18. Doctoral degrees were coded as 22.

The present study used the data collected by August 2004, August 2009, and June 2015 when participants were about ages 24, 29, and 35. The sample sizes were 1,367, 1,380, and 1,396, at ages 24, 29, and 35, respectively. The educational growth between ages 24 and 35 was examined because adult surveys were conducted at age 24 and age 35. Self-report educational attainment was available in addition to administrative records. The data at those ages were more comprehensive than data at other ages. To examine educational growth, we included age 29 as the middle point between ages 24 and 35. Of the study sample, 13.4% of whose total year of education increased from ages 24 to 29, and 26.8% of whose total year of education increased from ages 29 to 35.

CPC Participation

CPC participation was assessed by six groups: 1) no intervention or kindergarten only (None or K only; reference group, 9.8% had K only), 2) kindergarten and/or school-age participation (K-3; 18.7% had 1^st to 3^rd grade only), 3) preschool and kindergarten participation (P-K), 4) preschool through 1^st grade participation (P-1), 5) preschool through 2^nd grade participation (P-2), and 6) preschool through 3^rd grade participation (P-3). Because children began preschool at ages 3 or 4, the total duration of CPC participation varies slightly. The None or K only group was the reference group.

Table 2 shows sociodemographic characteristics and recovery rates by CPC groups. They were measured from administrative records and family surveys between birth and age 3 years. The comparability between the program and comparison groups is well documented (Reynolds et al., 2018; Reynolds et al., 2007; Reynolds et al., 2011). Variables from many dimensions were used to account for potential selection bias in previous studies (Ou et al., 2019; Reynolds et al., 2011). Those sociodemographic factors were included in the models as covariates.

Table 2.

Comparability of Preprogram Attributes by CPC Groups

Variables	Overall	P-3	P-2	P-1	P-K	K-3	None/K only1
Characteristics
Female	50.2	54.3	52.8	50.0	50.0	42.2	49.3
Black, %	93.0	95.4	91.8	91.0	93.1	97.6*	91.7
Mother did not complete HS, child age 0-3, %	54.3	39 9***	52.0*	56.0	53.8	61.4	59.6
Child eligible for subsidized meals, child age 0-3, %	83.8	88.4*	80.1	83.6	87.2*	86.1	81.5
Mother under age 18 at child birth, %	16.2	13.3	15.1	18.7	16.1	20.5	15.9
Four or more children in family, child age 0-3, %	16.6	14.5	17.5	13.4	16.1	18.1	17.7
Participate in AFDC program, child age 0-3, %	62.8	60.1	63.7	60.4	65.2	69.3*	59.1
Mother not employed, child age 0-3, %	66.3	63.6	66.3	65.7	71.5**	71.7*	61.5
Single parent family status, child age 0-3, %	76.5	74.6	77.2	79.1	76.4	75.3	76.3
Missing any risk factors, child age 0-3, %	16.2	9 2**	10.3***	18.6	21.6	16.9	19.8
Reside in high poverty neighborhood, %	76.0	57.8***	85.4***	74.6	80.7*	71.7	73.4
Number of family risk index, child age 0-3	4.5	4.1*	4.6	4.5	4.7	4.7	4.5
Change in family risk index from child age 0-3 to age 8	−.27	−.27	−.37*	−.19	−.32	−.13	−.23
Family conflict, child age 0-5	5.8	8.1	4.8	8.2	4.3	4.8	6.4
Family financial problems, child age 0-5	7.0	5.8	5.6	11.2	7.9	7.8	6.4
Low birth weight (<2500g), %	11.8	12.1	10.1	12.7	10.5	15.7	12.2
Age in months at kindergarten	63.3	63.8	63.2	63.2	62.6**	64.1	63.5
Original study sample size	1531	173	377	134	304	166	377
Sample recovery rate
% with education at age 24	89.2	92.5	93.1**	85.1	87.5	88.0	87.0
% with education at age 29	89.8	92.5	93.4**	85.8	88.9	88.6	87.8
% with education at age 35	90.8	92.5	93.1	86.6	90.8	89.8	89.8

^***p < .001

^**p < .01

^*p< .05

The original sample of 1,539 was reduced to 1,531 due to 8 participants not having identifying information for matching.

¹Reference group

Individual Attributes

Cognitive skills at kindergarten entry were measured through scores on the basic composite of the Iowa Tests of Basic Skills (ITBS; Level 5) (Hieronymus et al., 1982) in October of the kindergarten year (age 5). This was used as an indicator of developed abilities. The kindergarten test was a composite scale measuring listening and word analysis skills, vocabulary, language, and mathematics concepts. Internal consistency reliability is .94 (Hieronymus et al., 1982).

Academic achievement at age 14 was measured through the reading comprehension subtest of the ITBS, including 58 items and emphasized understanding of text passages (Level 14 or 13; Hieronymus & Hoover, 1990). The test was group-administered as part of a larger battery in April of 1994 of the eighth grade. The reliability of the scale was .92 (Reynolds, 2000).

Classroom adjustment was measured on the 6-item Social-Emotional Maturity Scale (SEMAT). It was rated yearly by teachers from grades 3 through 6 (ages 9 to 13) (Reynolds, 2000). The six items are “child concentrates on work,”, “child follows directions,” “child is self-confident,” “child participates in group discussions,” “child gets along with others,” and “child takes responsibility for his/her actions”. These items were coded on a Likert-type scale from 1 (poor/not at all) to 5 (excellent/very much). The reliabilities were 0.91, 0.91, 0.89, and 0.91 for grades 3, 4, 5 and 6, respectively. The average scores from grades 3-6 were used (range: 6 to 30).

School and Family Functioning

Absenteeism was measured by average days of absence during the school year by age 12. It was created based on teacher ratings in fifth and sixth grades, and supplement with parent ratings in fourth grade. Teachers were asked to rate participants regarding school attendance in fifth grade (age 11) on a 5-point scale, (1) poor/not attend at all, (2) below average/some, (3) average/satisfactory, (4) above average/good, and (5) excellent/much. This item was reversely recoded, so that (1) meant excellent/much and (5) meant poor/not attend at all. Teachers were asked to rate participants on days of absence during the school in sixth grade (age 12) on a 5-point scale, (1) 0 to 3, (2) 4 to 7, (3) 8 to 12, (4) 13 to 20, and (5) more than 20. Parents’ rating at fourth grade is based on the item “How often does your child stay home from school?”, and the scale is (1) never, (2) once a month, (3) once a week, (4) 2 or 3 times a week, and (5) nearly every day. The scale was recoded into (1) never, (3) once a month, (4) once a week, (5) 2 or 3 times a week, or nearly every day. The ratings among teachers in fifth and sixth grades and parents in fourth grade are significantly correlated (p < .01). The measure was constructed based on the average teacher rating of the fifth and sixth grades. If both teacher ratings were missing, then the parents’ rating was used. All ratings were on 5-point scales. The measure (ranging from 1 to 5) was then converted into days of absence based on the following rules: 1 equals 2 days, 1.5 to 2 equals 6 days, 2.5 to 3 equals 10 days, and above 3 equals 17 days.

Parents’ participation in school was assessed yearly for children between the ages of 7 and 12 (grades 1 through 6). Teachers rated “parent’s participation in school activities” from poor or no involvement (1) to excellent or much involvement (5). The scale reflected the number of times teachers rated parents as average or better than average at participation in school activities. On this six-point scale, average parents’ participation in school was rated as 3; the lowest rating was zero; the highest (i.e., better than average) rating for parents’ participation in school was 5.

Parent expectations on the child’s educational attainment were measured by parent rating in 4^th grade (age 10). Parents were asked about their expectations of the highest level their child will reach. Responses ranged from grade 8 (1) to graduate degree (7), which were recoded into a four-point scale: some high school (1); completed high school (2); some college (3); and completed 4-year college (4). The 4-point scale was transformed into years of education: 10 years (1); 12 years (2); 14 years (3); and 16 years (4). Responses from the parent survey in the 2^nd grade were used if responses were missing in 4^th grade. If responses were missing from both grades, responses from the same item in the 11^th grade parent survey was used.

Student expectations were measured through a dichotomous variable indicating whether students expected to go to college. This measure was based on the item, “How far in school do you think you will get?” from a survey in participants’ fourth grade. If students’ scores were missing in 4^th grade, 10^th grade survey responses on the same item were used. The correlation between 4^th and 10^th grade responses for those who have responses for both was 0.146 (p = .001)

On-time high school graduation was measured by a dichotomous variable indicating whether participants graduated from high school with an official diploma by 1998 or not.

Sociodemographic Factors

Sociodemographic characteristics were included as covariates, including gender, race/ethnicity, and family risk indicators. These were measured from birth to age 3. For gender, females were coded 1, and males were coded 0. For race/ethnicity, Black children were coded 1, and Hispanic children were coded 0. Family risk indicators include eight items, including mother not completed high school, mother unemployed, single-parent status, teen parenthood status, four or more children in the household, participation in the Aid to Families with Dependent Children (AFDC) program, eligibility for subsidized meals, and attendance at a kindergarten program in a school in which 60% or more of children in the attendance area reside in low-income families neighborhood poverty. These risk indicators are consistently associated with school success and educational attainment in studies (Alexander et al., 2001; Masten & Garmezy, 1985).

Missing data were handled after determining that data were missing at random via multiple imputation (MI) with the Expectation-Maximization (EM) algorithm. MI narrows uncertainty about missing values by calculating different options to impute the missing values. See Patrician (2002) and Sinharay et al. (2001) for more information on MI. Because missing values were imputed for family risk indicators, a missing index was created. If they were missing from any of the indicators, they were coded 1 for the missing index. Otherwise, they were coded 0. The missing index was included as a covariate.

Data Analysis

In our longitudinal data, observations were nested within individuals (Bryk & Raudenbush, 1987; Raudenbush & Bryk, 2002). We used hierarchical linear modeling (HLM) (Raudenbush et al., 2004) to analyze this nesting design of the longitudinal data. We used a two-level HLM. The level 1 units of analysis are three waves of educational attainment, and the level 2 units of analysis are the individual-level explanatory variables. Individual changes in educational attainment (level 1) are nested within individuals (level 2). HLM (i.e., multilevel modeling) has the flexibility to handle missing data, as it treats time as a continuous variable. All available data were included in our analysis (Kwok et al., 2008).

We investigated the changes in educational attainment (total year of education) as a function of sociodemographic factors, CPC participation, individual attributes, and school and family functioning. The level 1 model reflects the relationship between time and years of education. Because participants’ years of education were measured at ages 24, 29, and 35, time was operationalized by subtracting 24 from the measured age for a true zero point. The level 2 model reflects the relationship between predictors and the individual’s growth trajectories. The intercept (π_0i) represents individual i’s years of education at age 24. The slope (π_1i) indicates the growth rate of years of education between ages 24 and 35. In the present study, we assumed that both intercept (π_0i) and growth-rate (π_1i) of years of education vary among individuals. The unconditional growth model at level 1 and 2 is as follows:

$$\mathsf{Level}\text{1:}{Y}_{ti}={\pi }_{0i}+{\pi }_{1i}(age-24)+e_{ti}$$

$$\begin{gathered} \mathsf{Level}\text{2:}{\pi }_{0i}={\beta }_{00}+r_{0i} \\ {\pi }_{1i}={\beta }_{10}+r_{1i} \end{gathered}$$

With the unconditional model, we first confirmed our assumption on a random effect (r_1i) of the slope in two ways. First, a deviance test was used to compare the random-intercept unconditional model to the random-intercept and -slope unconditional model. The result of the deviance test (χ²(2) = 606.12, p < .001) supported our assumption of the random intercept and slope specification. Second, we investigated the structure of the unconditional growth model. The estimated intercept and slope were significant (p < .001), which supported the necessity of both parameters to describe the mean growth trajectory. The estimates of reliabilities for intercept and slope were .898 and .683, respectively. These estimates from the unconditional model allow us to investigate systemic relationships between level 2 factors and these growth parameters at level 1 (Bryk & Raudenbush, 1987; Raudenbush & Bryk, 2002). Additionally, the correlation between the intercept and slope was 0.057, indicating a weak relation between years of education at age 24 and growth rates of years of education between ages 24 and 35.

Building on the unconditional model, we proposed a series of growth models to examine changes in years of education as a function of sociodemographic factors, CPC program participation, individual attributes, and school and family functioning factors. A set of multivariate growth models were developed using Hox’s (2010) incremental improvement procedure considering measured times of each variable: (1) sociodemographic factors; (2) CPC program participation; (3) individual attributes, and school and family functioning variables. The equations at level 2 of the final model are:

$$\begin{gathered} {\pi }_{0i}={\beta }_{000}+{\beta }_{001}(Female{)}_i+{\beta }_{002}(Black{)}_i+{\beta }_{003}(Freelunch{)}_i+{\beta }_{004}(Mothern.e.{)}_i \\ +{\beta }_{005}(TFpart.{)}_i+{\beta }_{006}(60\%poverty{)}_i+{\beta }_{007}(Singleparent{)}_i \\ +{\beta }_{008}(Mother(<18y){)}_i+{\beta }_{009}(MotherHS{)}_i+{\beta }_{010}(4morechild{)}_i \\ +{\beta }_{011}(Missingrisk{)}_i+{\beta }_{012}(K-3{)}_i+{\beta }_{013}(CPCP-K{)}_i+{\beta }_{014}(CPCPK-1{)}_i \\ +{\beta }_{015}(CPCPK-2{)}_i+{\beta }_{016}(CPCPK-3{)}_i+{\beta }_{017}(ReadinessK{)}_i \\ +{\beta }_{018}(Parentinv{)}_i+{\beta }_{019}(Classroomadj{)}_i+{\beta }_{020}(Parentexp{)}_i \\ +{\beta }_{021}(Studentexp{)}_i+{\beta }_{022}(Absence{)}_i+{\beta }_{023}(8thReading{)}_i \\ +{\beta }_{024}(Ontimegrad{)}_i+r_{0i} \end{gathered}$$

$$\begin{gathered} {\pi }_{1i}={\beta }_{100}+{\beta }_{101}(Female{)}_i+{\beta }_{102}(Black{)}_i+{\beta }_{103}(Freelunch{)}_i+{\beta }_{104}(Mothern.e.{)}_i \\ +{\beta }_{105}(TFpart.{)}_i+{\beta }_{106}(60\%poverty{)}_i+{\beta }_{107}(Singleparent{)}_i \\ +{\beta }_{108}(Mother(<18y){)}_i+{\beta }_{109}(MotherHS{)}_i+{\beta }_{110}(4morechild{)}_i \\ +{\beta }_{111}(Missingrisk{)}_i+{\beta }_{112}(K-3{)}_i+{\beta }_{113}(CPCP-K{)}_i+{\beta }_{114}(CPCPK-1{)}_i \\ +{\beta }_{115}(CPCPK-2{)}_i+{\beta }_{116}(CPCPK-3{)}_i+{\beta }_{117}(ReadinessK{)}_i \\ +{\beta }_{118}(Parentinv{)}_i+{\beta }_{119}(Classroomadj{)}_i+{\beta }_{120}(Parentexp{)}_i \\ +{\beta }_{121}(Studentexp{)}_i+{\beta }_{122}(Absence{)}_i+{\beta }_{123}(8thReading{)}_i \\ +{\beta }_{124}(Ontimegrad{)}_i+r_{1i} \end{gathered}$$

Continuous variables were grand-mean centered. Further, to confirm the random effects of the growth rates across models 1 to 6, we used the deviance test described above. Results of the deviance test, χ²(2) ranging from 494.31 to 564.86, p < .001 in all models, support our assumption of the random effects (r_1i) on the slope. HLM 6.08 was used for data analysis (Raudenbush et al., 2004).

Results

Tables 3 and 4 present the means, standard deviations, and correlations of the variables. Years of education at ages 24, 29, and 35 are significantly correlated with each other. The average year of education increased from ages 24 to 29, and from ages 29 to 35. Standard deviations of the year of education increased from ages 24 to 29, and from ages 29 to 35. Years of education at ages 24, 29, and 35 were significantly correlated with most predictors as expected (p < .05).

Table 3.

Descriptive statistics of key variables ^a

Variable	M	SD	N
Female	.502	.500	1,531
African American	.929	.256	1,531
Eligible for subsidized meals, child age 0-3	.837	.370	1,531
Mother not employed, child age 0-3	.662	.473	1,531
AFDC Participation, child age 0-3	.626	.484	1,531
Reside in high poverty neighborhood	.761	.427	1,531
Single parent family status, child age 0-3	.764	.425	1,531
Mother under age 18 at child birth	.163	.369	1,531
Mother not completed high school, child age 0–3	.540	.499	1,531
Four or more children in family, child age 0-3	.167	.373	1,531
Missing any family risk indicators	.158	.365	1,531
K-3	.108	.311	1,531
CPC P-k	.199	.399	1,531
CPC P-1	.088	.283	1,531
CPC P-2	.246	.431	1,531
CPC P-3	.113	.317	1,531
School readiness at kindergarten	47.387	8.793	1,531
Parent involvement in school	1.857	1.513	1,531
Classroom adjustment	18.822	4.308	1,531
Parent expections	14.324	1.639	1,531
Stud college expections	.615	.487	1,531
Days of school absence	7.885	4.616	1,531
Reading score at 8th grade	144.970	20.788	1,531
On-time high school graduation	.402	.491	1,462
Year of education at age 24	11.875	1.629	1,367
Year of education at age 29	12.107	1.799	1,380
Year of education at age 35	12.657	2.138	1,396

Note.

^aValues result from individual-level (level-2) analyses

Table 4.

Correlations among key variables ^a

Variable	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18	19	20	21	22	23	24	25	26	26
1. Female	−
2. African American	.042	−
3. Free Lunch	.033	.051*	−
4. Mother not employed	.027	.121**	.349**	−
5. AFDC Part.	.028	.172**	.400**	.694**	−
6. High poverty	.048	.025	.055*	.062*	.067**	−
7. Single parent	.005	.183**	.112**	.197**	.272**	.041	−
8. Mother <18y at birth	−.035	.080**	.061*	.079**	.118**	−.014	.224**	−
9. Mother non-HS	−.031	−.111**	.167**	.213**	.215**	.039	119**	.325**	−
10. 4 or more children	.034	.007	.013	−.009	−.062*	.017	−.233**	−.197**	.038	−
11. Missing risk	−.059*	−.048	.099**	.140**	.124**	−.047	.018	−.031	−.021	−.055*	−
12. K-3	−.056*	.055*	.023	.041	.048	−.036	−.009	.040	.052*	.013	.010	−
13. CPC P-k	−.002	.003	.047	.055*	.026	.052*	.000	−.002	−.004	−.008	.076**	−.174**	−
14. CPC P-1	−.001	−.023	−.001	−.003	−.014	−.011	.020	.020	.012	−.027	.024	−.108**	−.154**	−
15. CPC P-2	.030	−.026	−.055*	.002	.013	.125**	.011	−.018	−.023	.012	−.086**	−.199**	−.285**	−177**
16. CPC P-3	.030	.050	.046	−.019	−.018	−.153**	−.015	−.029	−.101**	−.022	−.064*	−.124**	−.178**	−.111**	−.204**	−
17. Readiness at kind.	.060*	.108**	−.074**	−.088**	−.094**	.059*	−.047	−.061*	−.153**	−.001	.042	−.166**	.036	.006	.191**	.103**	−
18. Parent inv	.130**	−.060*	−.036	−.091**	−.117**	−.101**	−.082**	−.039	−.137**	.001	−.187**	−.080**	−.178**	−.049	.195**	.218**	.217**	−
19. Classroom adj	.267**	−.122**	−.096**	−.110**	−.127**	−.018	−.058*	−.038	−.134**	−.021	−.006	−.046	−.060*	.008	.072**	.103**	.292**	.395**	−
20. Parent exp	.109**	−.014	−.094**	−.111**	−.116**	−.020	−.070**	−.022	−144**	−.080**	.050	.000	−.032	−.030	.072**	.045	.169**	.185**	.229**	−
21. Stud college exp	.139**	.033	−.016	−.024	−.036	−.003	.002	−.040	−.090**	−.005	−.205**	−.039	−.094**	.013	.063*	.138**	.109**	.319**	.224**	.152**	−
22. School absence	−.096**	.016	.061*	.079**	.107**	.030	.073**	.054*	.113**	−.005	.077**	.048	.028	.007	−.078**	−.030	−.074**	−.255**	−.208**	−.076**	−.166**	−
23. Reading 8th grade	.189**	−.069**	−.146**	−.131**	−.133**	−.029	−.087**	−.063*	−.172**	−.055*	.030	−.088**	−.011	−.001	.100**	.107**	.423**	.350**	.542**	.287**	.221**	−.176**	−
24. On time HS grad	.198**	−.029	−.081**	−.150**	−.203**	.001	−.095**	−.049	−.179**	−.022	−.022	−.046	−.029	−.028	.083**	.102**	.149**	.291**	.358**	.230**	.223**	−.204**	.365**	−
25. Years of edu 24	.195**	−.060*	−.091**	−.145**	−.175**	−.027	−.088**	−.062*	−.194**	−.050	−.013	−.103**	−.004	.016	.067*	.070**	.137**	.261**	.372**	.180**	.216**	−.229**	.369**	.511**	−
26. Years of edu 29	.189**	−.079**	−.100**	−.139**	−.171**	−.017	−.100**	−.068*	−.189**	−.043	−.008	−.078**	−.011	.007	.066*	.064*	.145**	.253**	.388**	.189**	.196**	−.235**	.388**	.498**	.901**	−
27. Years of edu 35	.208**	−.076**	−.097**	−.135**	−.154**	−.006	−.118**	−.086**	−.193**	−.027	.002	−.100**	.003	−.004	.092**	.074**	.202**	.247**	.410**	.191**	.177**	−.224**	.407**	.493**	.739**	.825**	−

Note.

^aValues result from individual-level (level-2) analyses;

^*p < .05

^**p < .01

Table 5 shows the results of the HLM analyses. Overall, the deviances and the Akaike Information Criterion (AIC) values decreased as each block of variables was added, which supports that each set of predictors contributes to the improvement of the models. We also used a model comparison test in ANOVA that provides evidence for the improvement of nested models (Rouder et al., 2016). The model comparison results indicate that as predictors were added, the models were significantly improved from the previous ones across all models we proposed (p<.001). Thus, the increased complexity of models was justified. Additionally, Pseudo R² increased as each block of variables was added. The final model explained 35.3% of the variance of years of education at age 24 and 10.4% of the variance of the growth in the year of education between ages 24 and 35.

Table 5.

Growth Models for Years of Education ^a

Variables	Model 1		Model 2		Model 3		Model 4		Model 5		Model 6
Intercept	12.627***	(.217)	12.477***	(.232)	12.140***	(.225)	11.927***	(.226)	11.839***	(.225)	11.491***	(.221)
Sociodemographic factors:
Female	.622***	(.084)	.606***	(.084)	.331***	(.083)	.299***	(.083)	.275***	(.082)	.172*	(.077)
African American	−.360*	(.170)	−.343*	(.169)	−.028	(.171)	−.056	(.167)	−.001	(.167)	−.070	(.163)
Free Lunch	−.043	(.117)	−.043	(.118)	.017	(.113)	.028	(.111)	.085	(.110)	.035	(.105)
Mother not
employed	−.085	(.114)	−.093	(.114)	−.069	(.107)	−.083	(.105)	−.057	(.101)	−.066	(.098)
AFDC Part.	−.375***	(.115)	−.375**	(.114)	−.301**	(.109)	−.280**	(.107)	−.304**	(.104)	−.128	(.100)
High poverty	−.065	(.099)	−.074	(.097)	−.022	(.093)	−.032	(.092)	−.008	(.091)	−.053	(.087)
Single parent	−.145	(.104)	−.150	(.104)	−.103	(.101)	−.097	(.101)	−.079	(.101)	−.046	(.095)
Mother <18y	.061	(.130)	.065	(.129)	.021	(.121)	.043	(.118)	.045	(.116)	.012	(.109)
Mother non-HS	−.505***	(.091)	−.474***	(.091)	−347***	(.088)	−.314***	(.087)	−.300***	(.086)	−.216**	(.082)
4 or more children	−.271*	(.115)	−.256*	(.114)	−.197	(.111)	−.183	(.110)	−.153	(.110)	−.131	(.102)
Missing risk	.037	(.133)	.076	(.132)	.101	(.128)	.188	(.128)	.143	(.127)	.096	(.118)
CPC program participation:
K-3			−.152	(.157)	−.213	(.147)	−.203	(.146)	−.212	(.145)	−.238	(.135)
CPC P-k			.118	(.130)	.150	(.123)	.151	(.121)	.122	(.119)	.062	(.112)
CPC P-1			.189	(.158)	.162	(.154)	.134	(.150)	.115	(.147)	.107	(.138)
CPC P-2			.293*	(.114)	.116	(.114)	.106	(.113)	.098	(.112)	.019	(.103)
CPC P-3			.322*	(.151)	.057	(.145)	.031	(.144)	.020	(.144)	−.068	(.137)
Individual attributes, and school and family functioning
Readiness at Kind.					−.005	(.005)	−.005	(.005)	−.013*	(.005)	−.009	(.005)
Parent inv					.114***	(.031)	.070*	(.031)	.050	(.031)	.017	(.029)
Classroom adj					.094***	(.011)	.086***	(.011)	.062***	(.011)	.043***	(.011)
Parent exp					.053*	(.025)	.044	(.024)	.024	(.024)	−.005	(.023)
Student exp							.339***	(.085)	.292***	(.085)	.204*	(.080)
School absence							−.039***	(.009)	−.037***	(.009)	−.028***	(.009)
Reading 8th grade									.015***	(.002)	.010***	(.002)
On-time HS grad											1.157***	(.077)
Slope (age)	.107***	(.021)	.095***	(.021)	.092***	(.021)	.097***	(.021)	.094***	(.021)	.087***	(.021)
Sociodemographic factors:
Female	.026***	(.007)	.025***	(.007)	.014	(.007)	.015*	(.007)	.013	(.007)	.012	(.007)
African American	−.025	(.016)	−.025	(.016)	−.021	(.017)	−.021	(.017)	−.019	(.017)	−.020	(.017)
Free Lunch	−.014	(.011)	−.014	(.011)	−.009	(.011)	−.009	(.011)	−.006	(.011)	−.008	(.011)
Mother not employed	−.007	(.009)	−.007	(.009)	−.005	(.009)	−.005	(.009)	−.004	(.009)	−.004	(.009)
AFDC Part.	.009	(.009)	.008	(.009)	.012	(.009)	.012	(.009)	.011	(.009)	.015	(.009)
High poverty	.009	(.008)	.008	(.008)	.007	(.008)	.007	(.008)	.008	(.008)	.008	(.008)
Single parent	−.019*	(.009)	−.020*	(.009)	−.018*	(.009)	−.017	(.009)	−.017	(.009)	−.016	(.009)
Mother(<18y)	−.005	(.009)	−.004	(.009)	−.005	(.009)	−.005	(.009)	−.005	(.009)	−.005	(.009)
Mother HS	−.014	(.008)	−.012	(.008)	−.006	(.008)	−.006	(.008)	−.005	(.008)	−.004	(.008)
4 or more children	.002	(.010)	.003	(.010)	.005	(.010)	.005	(.010)	.006	(.010)	.006	(.010)
Missing risk	.006	(.011)	.009	(.011)	.003	(.011)	.002	(.011)	.000	(.011)	−.001	(.011)
CPC program participation:
K-3			−.001	(.011)	−.002	(.011)	−.001	(.011)	−.002	(.011)	−.002	(.011)
CPC P-k			.014	(.011)	.009	(.010)	.009	(.010)	.008	(.010)	.007	(.010)
CPC P-1			.003	(.014)	−.003	(.014)	−.003	(.014)	−.004	(.014)	−.003	(.014)
CPC P-2			.024*	(.010)	.013	(.010)	.012	(.010)	.012	(.010)	.010	(.010)
CPC P-3			.026*	(.012)	.012	(.013)	.013	(.013)	.013	(.013)	.011	(.012)
Individual attributes, and school and family functioning
Readiness at Kind.					.001*	(.001)	.001*	(.001)	.001	(.001)	.001	(.001)
Parent inv					−.002	(.002)	−.002	(.003)	−.003	(.003)	−.003	(.003)
Classroom adj					.004***	(.001)	.004***	(.001)	.003***	(.001)	.003**	(.001)
Parent exp					.002	(.002)	.002	(.002)	.001	(.002)	.001	(.002)
Student exp							−.009	(.008)	−.011	(.008)	−.014	(.008)
School absence							−.001	(.001)	−.001	(.001)	−.001	(.001)
Reading 8th grade									.001*	(.000)	.001*	(.000)
On-time HS grad											.024**	(.008)
Pseudo R2
Intercept	.106		.116		.208		.231		.253		.353
Slope (age)	.034		.044		.085		.087		.095		.104
Goodness-of-fit
Deviance	13063.4		13036.9		12831.7		12790.3		12737.9		12523.3
AIC	13121.2		13114.7		12925.5		12892.1		12843.7		12633.1

Note.

^aEntries are estimations of fixed effects with robust standard errors. Standard errors are in parentheses. Sample size = 4141 observations (level 1) by 1418 individuals (level 2); Standard errors are in parentheses; Estimation method = Full maximum likelihood

^*p < .05

^**p < .01

^***p < .001.

Education Attainment at Age 24

The intercept section of Table 5 shows the association between explanatory variables and the year of education at age 24. Among sociodemographic factors, gender (i.e., female) and mother’s status of high school completion at child age 3 were significantly associated with the year of education at age 24 in all models. In the final model, females reported 0.17 year (62 days) more of education at age 24 than males (ϐ = .172, p < .05), controlling for other factors. Participants having a mother who did not complete high school by their age 3, reported 0.22 year (80.3 days) less of education at age 24 than participants having a mother completed high school by their age 3 (ϐ = −.216, p < .05).

In Model 2, CPC program participation factors were added. Extended CPC program participation was associated with more years of education at age 24 (P-2, ϐ = .293, p < .05 and P-3, ϐ = .322, p < .05) compared to those who did not participate in any CPC program. CPC P-2 and CPC P-3 were not significantly associated with the year of education when other predictors were added in Models 3 to 6.

For individual attributes, and school and family functioning, factors were added into the model based on the time sequence. Parent involvement (ϐ = .114, p < .001) and parent expectations (ϐ = .053, p < .05) were significantly associated with the year of education at age 24 in Model 3. Parent involvement and parent expectation were not significant as other factors were added into the model. In Model 6, classroom adjustment (ϐ = .043, p < .001), students’ expectation to attend college (ϐ = .204, p < .05), days of absence (ϐ = −.028, p < .001), 8^th grade reading scores (ϐ = .010, p < .001), and on-time high school graduation (ϐ = 1.157, p < .001) were significantly associated with the year of education at age 24. The results indicate that one point increase in classroom adjustment score is associated with 0.04 years (14.6 days) increase in the total year of education at age 24. Students who expect to attend college were associated with 0.2 years (73 days) more of education at age 24 than those who did not expect to attend college. In terms of absenteeism, one day of absence was associated with a decrease of 0.03 year (10.95 days) of education at age 24. Also, one point increase in 8^th grade reading score is associated with 0.01 years (3.65 days) more of education at age 24. Students who graduated from high school on time were associated with 1.16 years (423.4 days) more of education at age 24 than those who did not graduate from high school on time.

Growth of Education between Ages 24 and 35

The slope section of Table 5 shows the association between explanatory variables and the growth of education between ages 24 and 35. Among sociodemographic factors, female (ϐ = .026, p < .001) and single-parent status (ϐ = −.019, p < .05) were significantly associated with the growth rates in Models 1. They were not statistically significant when other factors were added into the models.

CPC P-2 (ϐ = .024, p < .05) and CPC P-3 (ϐ = .026, p < .05) showed a positive association on the growth of education in Model 2. Reading scores at kindergarten (ϐ = .001, p < .05) and classroom adjustment (ϐ = .004, p < .001) were positively associated with the growth of education in Model 3. CPC P-2, CPC P-3, and reading scores at kindergarten were not statistically significant when other factors were entered into the models.

In final model (Model 6), classroom adjustment (ϐ = .003, p < .01), 8^th grade reading scores (ϐ = .001, p < .05), and on-time high school graduation (ϐ = .024, p < .01) were significantly associated with the growth of education, controlling all other factors. For example, one point increase in classroom adjustment was significantly associated with a 0.003 increase in the growth rate of educational attainment controlling for all other variables. On-time high school graduation was associated with a 0.024 increase in the growth rates of educational attainment controlling for all other variables.

To illustrate the influence of the three significant factors on the growth of education, we calculated the expected average years of education at age 35. For example, if the values of all variables are zero, except for on-time high school graduates (x), we can create a simple equation to calculate the year of education at a specific year.

$$Y_{age}=1.157x+0.024x(age-24)$$

The above equation implies students who graduated from high school on time is expected to have 1.4 years more of education at age 35 than those who did not graduate high school on time if other conditions are the same. Similarly, if a classroom adjustment score increases by one point, the year of education at age 35 increases by 0.076 years (27.7 days). One point increase in 8^th grade reading score is associated with 0.021 years (7.7 days) more of education at age 35. Figures 1 and 2 plot educational attainment for participants by classroom adjustment and on-time high school graduation.

Figure 1.

Educational Attainment by Classroom Adjustment

Figure 2.

Educational Attainment by On-Time High School Graduation

Robustness

To assess the robustness of the findings, several additional analyses were conducted. First, the composite family risk score was used instead of the individual family risk indicators (see Table S1). Second, the kindergarten schools were included as a covariate (see Table S2). Third, CPC program participation was examined by component (CPC preschool and CPC school-age programs, see Table S3) and duration (the number of years participated in the CPC program, see Table S4). Finally, the change of family risk status from age 0-3 to age 8 was included as a covariate (see Table S5). The patterns of findings were similar, which supports the robustness of our findings.

Discussion

Our goal was to identify childhood factors that could promote educational growth in adulthood for a vulnerable and disadvantaged population. Several points from the findings are discussed below.

Sociodemographic Factors

Consistent with the literature (Duncan & Murnane, 2016; Morrissey et al., 2014; Reardon, 2011), the negative impact of family adversities was supported in the study. Aid to Families with Dependent Children (AFDC) program participation and mother did not complete high school by child’s age 3 were significantly associated with lower educational attainment at age 24 above and beyond other factors included in most of the models. It suggests that the negative influence of such adversities on education is persistent, although its impact might be reduced to some degree. Children from a disadvantaged background are more likely to fall behind their peers on educational attainment. It is important to note that none of those family adversities was significantly associated with educational growth in adulthood, which suggests that early family adversities are less likely to hinder educational growth in adulthood than the initial educational attainment. The findings added support to the importance of early intervention for children from disadvantaged families (Magnuson & Duncan, 2016; Watts & Raver, 2020). Moreover, based on the findings on sociodemographic factors, subgroups by gender, AFDC participation, and whether the mother completed high school by the child’s age 3 warrant further investigation. For example, gender inequality in education is well known. The findings indicate that women reported 0.17 year more of education at age 24 than men, although gender was not significantly associated with the growth rate of education between ages 24 and 35. It would be interesting to see if the predictors of educational growth differ by gender, given that gender differences have been found on socioemotional learning and education (Aucejo & James, 2019; Oberle et al., 2014; Owens, 2016). By examining the subgroups separately, we might get insights on how predictors work differently by subgroups, which would provide policymakers suggestions on designing effective interventions.

CPC Program Participation

CPC preschool participation and extended CPC participation were found to be associated with higher educational attainment (Reynolds et al., 2018; Reynolds et al., 2007; Reynolds et al., 2011). Relative to the CPC preschool plus kindergarten (P-K) group, CPC participation from preschool through second grade (P-2) and CPC participation from preschool through third grade (P-3) were found to be significantly associated with better academic functioning (Ou et al., 2019). Findings from the present study corroborate with previous CLS studies on the long-term effects of CPC participation on educational attainment. Furthermore, the present study expands the long-term effects of extended CPC participation on educational growth in adulthood. The findings suggest that early childhood education programs not only can promote initial educational attainment, but it is also associated with a better chance that Black low-income students will continue to pursue higher education in adulthood. This extends the potential long-term effects of early childhood education.

The significant associations between extended CPC participation (P-2 and P-3) and age 24 educational attainment and educational growth between ages 24 and 35 disappeared after other factors were added to the models. It suggests that those positive associations might be mediated by other factors added to the models, which corroborate findings on the mechanisms of the long-term effects of the CPC program (Reynolds et al., 2017). Numerous studies have examined the effects of early childhood intervention in academic achievement and educational attainment, but the potential long-term effects of early childhood intervention in educational growth in adulthood have never been explored or examined before. The present study provides a new angle to examine the long-term effects of early childhood intervention.

Family Functioning

Family support behavior is one of the pathways that explain the long-term effects of early childhood intervention on education and other outcomes (Reynolds et al., 2017). There are programs targeting parents to improve outcomes for children living in poverty (Morris et al., 2017). Meta-analyses show that parental involvement connects to higher academic achievement (Castro et al., 2015; Jeynes, 2007). Parental involvement is a multifaceted phenomenon and includes parents’ participation in all domains of their children’s development. Because of the complexity of parental involvement, it is hard for studies to address all perspectives of parental involvement. Furthermore, there are social class differences in parents’ participation in school activities which are conditioned on SES (Chin & Phillips, 2004). Parents’ participation in school is emphasized in the Child-Parent Center (CPC) Program, and it was examined in the present study. Findings show that parents’ participation in school was significantly associated with education at age 24, but it was not significantly associated with education at age 24 once reading scores in 8^th grade and on-time high school graduation were added into the model. The association between parents’ participation in school and education might be mediated by academic achievement, which is consistent with previous studies on the positive impacts of parents’ participation in school on academic achievement (Huat See & Gorard, 2015).

School Adjustment and Functioning

Classroom adjustment, reading scores in 8^th grade, and on-time high school graduation were significantly associated with higher educational attainment at age 24 and positive educational growth rates between ages 24 and 35. Classroom adjustment and reading scores in 8^th grade represent socioemotional and academic skills, respectively. The predictive power of academic and socioemotional skills in childhood and adolescence on outcomes in adulthood are well documented (Gutman & Schoon, 2013; Jones et al., 2015; Magnuson et al., 2016; Ou & Reynolds, 2008). In addition to the wellestablished positive relations between academic functioning and educational attainment, the positive link between socioemotional skills and future wellness is a newly focused area (Gutman & Schoon, 2013; Jones et al., 2015; Magnuson et al., 2016). Classroom adjustment can be classified as one of the noncognitive skills. Traditionally, cognitive skills gain much more attention than noncognitive skills in educational studies. However, recent studies have found that noncognitive skills might have greater effects on schooling and other outcomes than cognitive skills (Farrington et al., 2012; Kautz et al., 2014; Levin, 2012; Ou & Reynolds, 2016). For example, students who had higher emotional and social competencies are more likely to persist in college (Aryee, 2017; Ou & Reynolds, 2016). Findings from the present study support studies on the significance of noncognitive skills. Moreover, the present study suggests that academic and socioemotional skills in childhood and adolescence are not only important in academic achievement, but they might also play important roles in educational growth in adulthood.

Although student college expectations and days of school absence were not significantly associated with the growth rate of educational attainment in adulthood, they were important factors of educational attainment at age 24. Research indicated that children who expect to attend college matter on their educational attainment (Fan & Wolters, 2014), and its effects are conditioned on family assets and their perceptions of SES mobility in particular for low SES students (Browman et al., 2017; Oyserman, 2013). Consistent with the literature, student college expectation is significantly associated with educational attainment controlling other factors.

Consistent with the literature on the negative impact of chronic absenteeism on educational achievement (Ansari et al., 2020; Gottfried, 2014; Morrissey et al., 2014), days of school absence were found to be significantly associated with educational attainment at age 24. This finding addresses the gap in knowledge of the long-term consequences of absenteeism. One day of absence was associated with a decrease of 0.03 year (10.95 days) of education at age 24. In other words, if a student missed 15 days of school per year, the student will have 0.45 year (164.25 days) less of education at age 24 than those who did not miss any school. Missing school 15 days per year is linked to almost half a year of difference in education at age 24. It is not trivial. It shows how critical absenteeism is to educational attainment. Days of school absence might be connected to educational attainment through a negative influence on achievement and school engagement. Educational growth in adulthood might be a different process and involves factors different from predictors of education. Different predictors for educational growth should be explored in the future.

On-time high school graduation (e.g. 4-year high school graduation) is the leading indicator of social determinants of health based on Healthy people 2020 (United States. Department of Health and Human Services). The findings support the importance of on-time high school graduation through its predictive power in both higher educational attainment at age 24 and positive growth in adulthood.

Strengths and Limitations

The major strengths of the present study include the prospective longitudinal cohort design and large sample size to examine predictors of educational growth trajectories. Very few longitudinal studies collected educational attainment at various ages in adulthood. The change in educational attainment in adulthood has not been examined before. Studies have focused on college attendance for black low-income students, but there is very limited attention to the educational growth in adulthood in the literature, in particular for Black low-income students. In the CLS, we observed the continuing changes in participants’ educational attainment from their 20s to 30s, which is a promising sign given education is the gateway to future well-being for the study population. Moreover, the educational attainment data in the present study are unique and different from other datasets because they were obtained from multiple data sources instead of self-reports only. The study sample of Black low-income youth is a primary focus of prevention efforts to reduce achievement gaps and improve well-being for this vulnerable population over the life course. Educational attainment (college degree) buffers against job loss and underemployment especially in recessions, like the Great Recession of 2008 (Infurna et al., 2020). This is occurring now even worse with the Pandemic recession. Educational growth in adulthood is more important than ever for the Black low-income population because the benefits of education are far more valuable than before.

Several limitations are notable. First, a quasi-experimental design was used in the study, which is more challenging to infer effects compared to randomized experimental design. Causal relationships should not be assumed. Second, the factors examined in the present study are not exhaustive. The final model explained 35.3% of the variance of years of education at age 24 and 10.4% of the variance of the growth in the year of education between ages 24 and 35. There are other factors related to educational attainment that might be omitted due to data availability. For example, peer relationships and neighborhood factors. Third, some factors were defined narrowly or measured through a single item. For example, parents’ participation in school includes a wide range of activities and can be measured differently. Parents’ participation in school was examined in the present study because it is one of the components emphasized in the CPC program and the data are available in the CLS. Moreover, parent expectations and student college expectations were measured through self-reported single items. Selfreports may be biased and affected by the social desirability of respondents (Fadnes et al., 2009; Fulmer & Frijters, 2009). Also, there might be underlying factors that moderate the validity of self-reports (Kuncel et al., 2005).

Fourth, some measures were created by utilizing information from multiple waves of data, such as student expectations, parent expectations, and days of school absence. Although the data collected at different time points are significantly correlated, an individual’s changes over time might be missed when multiple waves of data were combined into one composite measure. For example, the correlation for student expectations for going to college between 4^th and 10^th grade responses is 0.146 (p < .001). Participants’ expectation at 10^th grade are likely to be adjusted based on their understanding of their academic standing and be more realistic.

Fifth, the results need to be interpreted carefully due to the different units of analysis between the intercept and slope. The magnitude of beta coefficients in slope is relatively low compare to the magnitude of beta coefficients in intercept because the units of analysis are different. The unit of analysis in intercept is the year of education at age 24 (M=11.875, SD=1.629), whereas the unit of analysis in slope is the growth rate per year between ages 24 and 35, and the mean difference in year of education between ages 24 and 35 is 0.77 (SD=1.457). Moreover, coefficients (e.g. classroom adjustment, reading scores at 8^th grade, and on-time high school graduation) are statistically significant in slope, but their effects are trivial when effect sizes are calculated (Feingold, 2009). Whereas some coefficients in intercept indicate small or large effects when effect sizes are calculated. For example, CPC P-2 and CPC P-3 in Model 2 (effect sizes = 0.24 and 0.27 respectively) and on-time high school graduation in Model 6 (effect size = 1.03). Finally, generalizability and reproducibility should be further investigated as well as the extent to which findings are generalizable beyond the Black economically disadvantaged population. Moreover, although the retention rate is high (over 92% of the original sample), the attrition analysis indicates that more disadvantaged participants are more likely to be in the attrition sample. Associations between some factors and the outcomes might be underestimated.

Conclusion

In this longitudinal study of educational growth trajectory in an at-risk sample, socioemotional learning, academic achievement, and on-time high school graduation were consistently associated with higher educational attainment at age 24 and positive growth rates of educational attainment between age 24 and age 35. Other than improving academic achievement, findings support the importance of socioemotional learning and on-time high school graduation. It is worth noting that although factors at multiple levels are significantly associated with educational attainment at age 24, only a few factors are significantly associated with educational growth in adulthood. Those factors are closely tied to individual attributes rather than school or family functioning. It suggests that cognitive and noncognitive skills built on early in life can play important roles in whether an individual might pursue higher education in adulthood. Prevention programs that improve academic achievement, strengthen socioemotional learning, and promote on-time high school graduation are more likely to cultivate vulnerable and disadvantaged youth to continue higher education in adulthood. Higher educational attainment may then increase their chance of upward mobility and improve their well-being and health in the long run because education disparities are greatly connected to income and health disparities.