Abstract
The educational attainment of participants in the Chicago Longitudinal Study (93% Black and 7% Hispanic) was examined. Cluster analysis on measures of human capital resources, family dynamics and demographics was used to identify four distinct profiles of families. In general, children who had family profiles characterized by higher levels of human capital resources and more favorable scores on indictors of family functioning were more likely than other children to have higher educational attainment. Moreover, children who experienced a positive change in family profile characteristics between the ages of 8 and 12 were more likely than other children to have higher educational attainment. Implications for intervention and research are discussed.
High school completion is a milestone of adolescence that serves as an entryway to post secondary education and nearly all career paths. Besides representing the transition to adulthood, graduation from high school has strong and direct impacts on employment, lifetime earnings, health, and criminal justice involvement (Cohen, 1998; Haveman & Wolfe, 1983, 1994, 1995; Hill & Sandfort, 1995; Reynolds, Temple, Ou, Robertson, Mersky, Topitzes, & Niles, 2007; Reynolds, Temple, Robertson, & Mann, 2002). Earning a high school diploma represent a crucial step in gaining access to a college education that will enhance an individual’s work prospects in an increasingly competitive society. School failure has significant costs to society and youth. The unemployment rate for dropouts is approximately 80 percent higher than the unemployment rate for high school graduates (National Center for Educational Statistics, 2000, Figure 24). The annual cost to society of school dropout linked to forgone earnings and tax revenues is estimated to be $250 billion (National Science and Technology Council, 1997).
Despite a variety of educational reforms designed to increase high school graduation rates, little progress has been made (Kaufman, Kwon, Klein, & Chapman, 2001). In 1999, 85.9% of 18 to 24 year olds had completed high school by earning a regular diploma (76.8 %) or GED (9.1%). This represents a net increase of only about 3 percentage-points since 1972 (82.8%). The number of high school graduates who achieve a degree via a GED is growing and when GEDs are classified as dropouts, the high school graduation rate has fallen or remained relatively stable since 1972 (Kaufman, Alt, & Chapman, 2004). The economic payoff for people who complete high school by alternative means (e.g., GED) can be significantly less than what it is for people who earn a traditional high school diploma (Carneiro & Heckman, 2003).
Students in large cities, especially minorities and low-income youth are particularly at-risk for school failure. Large city school districts are more likely than mid-sized cities, large towns and rural districts to report four-year high school completion rates of less than 60 percent (Sable & Hoffman, 2005). Relative to White students, Hispanic and Black students are less likely to complete high school. Nationally, since 1972 the dropout rate for Blacks and Hispanics has been nearly 2 to 4 times higher than the dropout rate for Whites. Equally disturbing is that children from low-income families are 5 to 8 times more likely to dropout than children in high-income families (Laird, DeBell, & Chapman, 2006).
The powerful and long-term importance of early family experiences on the educational attainment of children is suggested by human capital theory (Cairnero & Heckman, 2003; Heckman, 2000; Karoly, 2001). Brooks-Gunn, Duncan, Leventhal & Abner (1997) define human capital as “skills, knowledge, and capabilities acquired by individuals” (p.290). Guided by the human capital framework of Becker (1981, 1993), Haveman and Wolfe (1995) suggest that parents make choices and investments that influence their children’s success by developing and impacting the environment in which children are raised.
The present study builds from a human capital framework and adds to a growing body of research which has established the heuristic value of pattern-oriented analyses. Pattern-centered approaches have been used to examine parenting and grand parenting, family functioning, dropout, homelessness and school performance (Baydar & Brooks-Gunn, 1998; Belsky & Fearon, 2004; Brenner & Fox, 1999; Cairns, Cairns & Neckerman, 1998; Danesco & Holden, 1998; Janosz, Blanc, Boulerice & Tremblay, 2000; Johnson, 2003; Mandura & Murray, 2002; Ramey, Ramey & Lanzi, 1998).
Few longitudinal studies of low-income children have used pattern-oriented methods to examine variation in families in relation to school outcomes. This may be because poor children and families are considered relatively homogeneous (Ramey, Ramey & Lanzi, 1998). Most standard analytic approaches assume that measures of association are interrelated similarly and linearly across all families. However, a systems’ view of families warns against embracing “the prevailing model of family influences that focuses on the specific, independent influence of relationships and assumes that identical processes exist in all families” (O’Connor, Hetherington, & Reiss, 1998, p.354).
According to a developmental science perspective, development occurs as part of an integrated system rather than the result of individual constituent parts (Magnusson, 1998). Examining developmental characteristics collectively within the context of a correlated system (e.g., family) can be more informative than investigating developmental characteristics as singular influences (Cairns, 2000; Magnusson & Cairns, 1996; Roeser & Peck, 2003). Within this context, the goal of the present study was to identify distinct family profiles for low-income minority children participating in the Chicago Longitudinal Study.
The study was guided by three primary aims. The first aim was to use cluster-analytic methods on measures of human capital resources, family dynamics and demographics to develop family characteristic profiles and describe a typology of families in the Chicago Longitudinal Study (CLS). The terms profile, typology, configuration, cluster, and classification are used interchangeable throughout the text. The second aim was to examine whether adolescents’ educational attainment (e.g., eighth grade reading scores, high school completion, 4-year high school graduation, college attendance) differed by family profile. Youth whose family profiles were characterized by low levels of human capital resources and low levels of family functioning were expected to show less favorable outcomes on measures of educational attainment compared to youth whose family profiles were more advantaged. The third aim was to examine the degree to which family characteristic profiles changed over time and to estimate the effect of change (positive and negative) on educational attainment. Change was expected to be related to school performance. Youth whose family profiles showed gains on measures of human capital and indicators of positive family functioning were expected to have higher educational attainment than other children.
Method
Sample and Design
The Chicago Longitudinal Study (CLS, 2005) is a prospective investigation that examines the social and educational development of a same-age cohort of 1,539 economically disadvantaged minority children (93% Black and 7% Hispanic) born in 1980. The original Chicago Longitudinal Study sample included the entire cohort of 989 children who completed preschool and kindergarten in one of the 20 Chicago Child-Parent Center sites. The CPC preschool group further participated for up to three years in the school-age component of the CPCs during grades 1 to 3. The preschool comparison group of 550 children in this quasi-experimental design participated in alternative full-day kindergarten programs for low-income children emphasizing educational enrichment but they did not enroll in CPC preschool.
The study sample consisted of 1,340 CLS children and families for which educational attainment, family human capital, family dynamics and demographic data were available. By child age 23, data were available for 88.2% of the original CLS sample. This represents 87.1% of the original sample for the preschool group (n = 872) and 85.1% for the comparison group (n = 468). In the CPC preschool comparison group, 15.6% of children participated in Head Start preschool and 100% participated in government-funded full-day kindergarten programs.
Educational Attainment Indicators
Educational attainment indicators were developed from school administrative records from Chicago Public and Parochial Schools, and the Illinois Shared Enrollment and Graduation Consortium (ISEGC). The ISEGC maintains college enrollment and credit records collected by the Illinois Community College Board (ICCB) and the Illinois Board of Higher Education (IBHE). These records were supplemented by written and phone interviews with the study participants and other family members.
Eighth grade reading scores
Eighth grade reading scores were assessed from the multilevel battery of the Iowa Test of Basic Skills (Hieronymus & Hoover, 1990; Hieronymus, Lindquist, & Hoover, 1980). The reading comprehension subtest has 58 items designed to measure comprehension of text passages (see Reynolds, 2000). Scores were based on the 1988 national average equating to a mean of 160 in the fall of eighth grade. Test scores ranged from a low score of 77 to a high score of 212.
High school graduation
High school graduation indicates that the study participant graduated from high school with a regular diploma in four years.
High school completion
High school completion indicates that the study participant completed his/her secondary education by obtaining either an official diploma or a General Education Diploma (GED).
Years of schooling
Years of schooling measures the last grade of school that the youth officially completed. Years of schooling ranged from a low score of 7 years to a high score of 16 years (bachelor’s degree).
College attendance by age 23
College attendance indicates that a study participant had been enrolled for one or more credits in a degree- or certificate-bearing program at an accredited two- or four-year college by the summer term of 2003.
Family Typology Measures and Family Characteristic Profiles
Family typology measures
Cluster analysis was used on the measures of maternal age, maternal education, number of children living in the home, number of adults living in the home, family demographic index, child maltreatment scale, parental involvement and parental expectations to develop family profiles for children at 8 and 12 years of age. Data were obtained from parent and participant interviews, teacher interviews, and administrative records from the Illinois Department of Health and Human Services, the child protection division of the Illinois Department of Children and Family Services, Chicago Public and Parochial Schools, and the Cook County Juvenile Court.
Maternal age at child’s birth
Maternal age at child’s birth indicates the age of the mother in years at the time of the study participant’s birth.
Maternal education
Maternal education is the number of years of education completed by the study child’s mother.
Number of children
Number of children signifies the number of children including the participant living in the immediate family context (children were defined as any household member less than 18 years of age).
Number of adults
Number of adults indicates the number of adults living in the immediate family or home context (adults were defined as any household member more than 17 years of age).
Family demographic index
This four-item index consisted of the sum of four dichotomous indicators (parent employed full- or part-time, child ineligible for free lunch, no AFDC/TANF participation, and two-parent family status) shown to be related to positive child and family functioning (Bendersky & Lewis, 1994; Rutter, 1987).1
Parent employed full- or part-time indicates that one or more of the participant’s parents reported full- or part-time employment.
Child ineligible for free lunch indicates that the study child was reported ineligible for the subsidized school lunch program because their family income exceeded 130% of federal poverty line.
No AFDC/TANF participation indicates that the study child’s mother was never the principal grantee of an AFDC/TANF grant.
Two-parent family status indicates that the participant was living with two biological parents.
Child maltreatment scale
This four-level scale indicates whether child maltreatment was identified as a problem or potential problem in the participant’s family. Quasi-absolute scaling (see Bergman, Magnusson, & El-Khouri, 2003) was applied to child maltreatment and foster care records. This four-level scale consisted of the following values: 1 = two or more substantiated reports of child maltreatment and/or two or more out-of-home placements because of child maltreatment (pronounced problem in family functioning), 2 = one substantiated report of child maltreatment and/or one out of home placement because of child maltreatment (presence of a problem), 3 = one or more unsubstantiated reports of child maltreatment and/or families with child protection service records not directly related to the study child (tendency to problem), and 4 = no child protection service records (no visible problem).
Parental involvement in children’s schooling
This measure is the sum of the number of years a teacher rated parent involvement average or better (1 = average or above, 0 = below average or poor) between first and sixth grade. The original scale values were as follows: 1 = poor, 2 = below average, 3 = average, 4 = above average and 5 = excellent.
Parental expectations for children’s educational attainment
This measure is the sum of the number of years a teacher and one or more parents rated parent expectations average or better between second and sixth grade (1 = average or above, 0 = below average or poor). The original scale values for the teacher reported measure were: 1 = poor, 2 = below average, 3 = average, 4 = above average and 5 = excellent.2
Changes in Family Characteristic Profiles between Ages 8 and 12
Two dichotomous indicators were used to examine the effect of change in family characteristic profiles during early adolescence.
Positive change
Positive change indicates that children showed gains on family measures of human capital and family functioning between the ages of 8 and 12.
Negative change
Negative change indicates that children showed declines on family measures of human capital and family functioning between the ages of 8 and 12.
Covariates
CPC preschool
CPC preschool includes all children who participated in the half-day preschool component of program for 1 or 2 years (ages 3 to 4 years).
CPC school-age
CPC school-age includes all children who participated in the school-age component of the program (ages 6 to 9).
CPC program sites
Twenty dichotomous variables were used to indicate the sites of CPC program participation. These indicators were used to control for the local unobserved influences associated with attending a particular CPC program site (see Reynolds, 2000).
Sex of child
Sex of child is the gender of the study participant.
Race/ethnicity
Race/ethnicity is the race/ethnicity of the study participant.
School mobility
School mobility indicates the number of times a participant changed schools from kindergarten to sixth grade (ages 5 to13). School mobility from kindergarten to third grade was used in child age 8 models. School mobility from first to eighth grade was used in child age 12 models.
Grade retention
Grade retention indicates whether a child was ever retained from kindergarten to fourth grade (age 8 models) and kindergarten to sixth grade (age 12 models).
Kindergarten school achievement
Kindergarten school achievement indicates the word analysis score results at the end of kindergarten on the Iowa Test of Basic Skills (Hieronymus, Linquist & Hoover, 1980). The word analysis contained 35 items assessing prereading skills and letter-sound recognition (see Reynolds, 2000).
Data Analysis
Family characteristic profiles were identified from the family typology measures using the SLEIPNER II statistical package for pattern-oriented analysis (Bergman & El-Khouri, 1998) following methods outlined by Bergman, Magnusson & El-Khouri (2003). Five procedures were employed: (a) IMPUTE, a procedure that imputes cases with missing data using a twin approach;3 (b) CLUSTER, a clustering program that identifies initial classifications based on a clustering algorithm (with CLUSTER, initial configurations were determined using Ward’s method (1963) on standardized measures with the similarity among cases indicated by squared Euclidean distance);4 (c) RELOCATE, a relocation procedure that repositions ill fitting cases to alternative classifications if doing so reduces the error sums of squares of the classification; (d) EVALUATE, a procedure that evaluates a cluster solution by examining the explained error sums of squares (ESS) and other indices such as the Point-biserial correlation that measure the goodness-of-fit between input data and the resulting classification (see Milligan, 1981; Milligan & Cooper, 1988); and (e) RANDOM, a procedure that draws a random sample without replacement from the population.5
Following previous analyses in the CLS, hierarchical probit regression analysis was used to estimate effects for dichotomous outcomes (e.g., high school completion and graduation, college attendance). To enhance interpretability, probit coefficients were transformed to marginal effects using STATA version 8 (Stata, 2003) which can be interpreted similar to the metric coefficient in ordinary regression analysis.6 Hierarchical linear regression analysis and univariate analysis of covariance (ANCOVA) were used on outcomes that approximated normal distributions (e.g., eighth grade reading scores, years of completed schooling).
Results
Descriptive Results
As shown in Table 1, the overall economic disadvantage of the study families was reflected by the measures used to develop family profiles. On average, mothers were likely to have less than twelve years of education. The majority of mothers at both child age periods were not employed full- or part-time, received AFDC/TANF benefits, had three or more children and were single-parents. As global indicators of family dynamics and functioning, overall scores on the child maltreatment scale, parental involvement and parental educational expectations, suggested that many families in the study were not well-functioning. For example, before age nine, 100 children (7.5%) had experienced child maltreatment. By age thirteen, 139 children (10.4%) had substantiated reports of maltreatment. The majority of children in the study were reported as having below average parental expectations and parental involvement. The at-risk status of the children in the study was also reflected by relatively high percentages of children who had been retained or experienced multiple and atypical school moves. By sixth grade, 331 children (24.7%) had been retained for at least one year of schooling. Between first and eighth grade, 496 children (17.0%) had changed schools 3 or more times. The correlations between like-item indicators between child age periods were relatively high ranging from r = .51 (mother employed full- or part-time) to r = .95 (no TANF/AFDC participation). The correlations among key study indicators for the age 12 sample are provided in Appendix 1.
Table 1.
Descriptive Statistics, Sample Sizes and Correlations between Age Periods for the Family Profile Measures and School-level Covariates
| Age 8 |
Age 12 |
Pearson correlation |
|||
|---|---|---|---|---|---|
| n = 1338 | n = 1340 | ||||
| Family profile measures | Mean | N | Mean | N | R |
| Maternal age at child’s birth, years+ | 22.7 | 1283 | -- | -- | -- |
| Maternal education, years+ | 11.3 | 1329 | 11.4 | 1332 | .68 |
| Family demographic index (0–4), mean+ | 1.3 | 1338 | 1.3 | 1340 | .77 |
| No TANF/AFDC participation, %# | 41.2 | 1338 | 38.7 | 1340 | .95 |
| Mother employed (full- or part-time), %# | 43.6 | 1338 | 48.5 | 1340 | .51 |
| No Free lunch eligibility, %# | 12.6 | 1338 | 16.8 | 1340 | .75 |
| Two-parent status, %# | 40.2 | 1338 | 32.2 | 1340 | .54 |
| Number of children (1–7), mean+ | 3.2 | 1313 | 3.1 | 1322 | .78 |
| Number of adults (1–6), mean+ | 1.6 | 1305 | 1.8 | 1305 | .91 |
| Number of Children (1–7), mean+ | 3.2 | 1313 | 3.1 | 1322 | .78 |
| Number of adults (1–6), mean+ | 1.6 | 1305 | 1.8 | 1305 | .91 |
| Child maltreatment scale (1–4), mean+ | 3.8 | 1338 | 3.7 | 1340 | .89 |
| No visible problem (4), % | 89.6 | 1338 | 87.2 | 1340 | .70 |
| Tendency toward problem (3), % | 2.9 | 1338 | 2.4 | 1340 | .72 |
| Presence of problem (2), % | 3.5 | 1338 | 3.3 | 1340 | .62 |
| Pronounced problem (1), % | 4.0 | 1338 | 7.1 | 1340 | .63 |
| Parental involvement (0–3, 0–6), mean+ | 1.1 | 1338 | 2.0 | 1340 | .82 |
| Parental educational expectations (0–3, 0–5), mean+ | 1.5 | 1338 | 2.3 | 1340 | .85 |
| School-level covariates | |||||
| Number of school moves (0–5, 0–9), mean | 1.4 | 1338 | 2.3 | 1340 | .91 |
| Grade retention, % | 20.0 | 1338 | 24.7 | 1340 | .87 |
Note. Variable is included in cluster analysis as a stand-alone indicator.
Variable is included in cluster analysis as part of the family demographic index. Maximum n for Pearson correlation was 1338.
Overall, 64.0% of the youth in the study completed high school by age 23 (see Table 2). Slightly more than half of the study participants graduated from high school with a regular diploma in 4-years (50.2%). Females outperformed relative to males on all educational attainment indicators. For example, females were more likely than males to complete high school (71.6% vs 56.0%, p < .001), and complete more years of schooling (11.73 vs 11.08 years, p < .001).
Table 2.
Descriptive Statistics of Educational Attainment by Key Study Indicators for the Age 12 Sample
| CPC Preschool |
Extended |
||||
|---|---|---|---|---|---|
| Factor | Total Sample |
Any | None | 4 to 6 | 1 to 3 |
| High school completion, % | 64.0 | 67.3 | 57.9 | 69.5 | 60.6 |
| Girls | 71.6 | 72.8 | 69.4 | 76.3 | 67.3 |
| Boys | 56.0 | 61.1 | 47.6 | 61.4 | 54.2 |
| Mother HS completer | 72.5 | 75.5 | 66.3 | 76.1 | 71.9 |
| Non completer | 53.1 | 55.6 | 49.4 | 59.2 | 46.4 |
| No TANF/AFDC | 74.4 | 77.5 | 68.7 | 80.7 | 72.1 |
| TANF/AFDC participation | 57.5 | 60.9 | 51.1 | 62.7 | 53.9 |
| Child maltreatment | 47.5 | 49.3 | 45.2 | 47.2 | 44.4 |
| No visible problem/ tendency to problem | 65.9 | 69.1 | 59.9 | 71.2 | 62.9 |
| Low parent involvement | 57.1 | 60.7 | 52.7 | 61.2 | 56.2 |
| Average or better | 75.9 | 78.3 | 70.3 | 78.2 | 75.4 |
| Low parent expectations | 56.5 | 59.9 | 51.1 | 61.8 | 53.7 |
| Average or better | 73.2 | 75.3 | 68.5 | 74.9 | 73.0 |
| Highest grade completed, years | 11.41 | 11.53 | 11.17 | 11.64 | 11.24 |
| Girls | 11.73 | 11.80 | 11.58 | 11.97 | 11.60 |
| Boys | 11.07 | 11.23 | 10.80 | 11.26 | 10.97 |
| Mother HS completer | 11.74 | 11.85 | 11.49 | 11.91 | 11.60 |
| Non completer | 10.99 | 11.08 | 10.84 | 11.22 | 10.78 |
| No TANF/AFDC | 11.84 | 11.94 | 11.64 | 12.05 | 11.71 |
| TANF/AFDC participation | 11.14 | 11.28 | 10.87 | 11.40 | 10.94 |
| Child maltreatment | 10.68 | 10.70 | 10.65 | 10.50 | 10.71 |
| No visible problem/tendency to problem | 11.50 | 11.61 | 11.26 | 11.73 | 11.31 |
| Low parent involvement | 11.10 | 11.22 | 10.90 | 11.21 | 11.04 |
| Average or better | 11.99 | 12.06 | 11.79 | 12.10 | 11.90 |
| Low parent expectations | 11.05 | 11.19 | 10.81 | 11.22 | 10.96 |
| Average or better | 11.95 | 11.90 | 11.74 | 11.93 | 11.85 |
Note. Child maltreatment indicates that by child age 12 there was 1 or more substantiated report of maltreatment. No visible/tendency to problem indicates that there were no child protection service records or unsubstantiated reports of maltreatment by age 12.
Other sizable differences existed for participants based on family subgroups derived from rates of parent education, TANF/AFDC participation, child maltreatment, and parental involvement and expectations. For example, children whose mothers completed high school were more likely to complete high school (72.5% vs 53.1%, p < .001) and complete more years of schooling (11.74 vs 10.99, p <.001). Relative to children in families with one or more substantiated reports of child maltreatment (pronounced, presence of a problem), children in families without substantiated reports of child maltreatment (tendency to problem, no visible problem) were more likely to complete high school (65.9% vs 47.5%, p < .001) and complete more years of schooling (11.50 vs 10.68, p < .001). Relative to low levels of parent involvement, average or above parent involvement was significantly associated with higher rates of high school completion and more years of competed schooling. Similarly, relative to low levels of parental expectations, average or above parental expectations was significantly associated with higher rates of high school completion and more years of competed schooling.
Cluster Analysis and Four Distinct Family Profiles
Table 3 illustrates the four family profiles that emerged from the cluster analysis.7 To highlight differences between profiles, post hoc comparisons are shown. Means and percentages in the same row that do not share superscripts differ (p < .05) in follow-up Bonferroni corrected t tests. It should be noted that the term “family profile” and the ensuing labels are heuristic and used for descriptive purposes to facilitate discussion and characterize the heterogeneity of the sample. They are not intended to stereotype behavior traits of families nor are they intended to be used for direct parallels among children with similar family profiles.
Table 3.
Family Characteristic Profile (Child Age 12)
| Family Profile |
||||
|---|---|---|---|---|
| Measure | Low Human Capital/ Low Family Functioning LHC/LFF |
Low Human Capital/ Mod. Family Functioning LHC/MFF |
Moderate Human Capital/ Mod. Family Functioning MHC/MFF |
High Human Capital/ High Family Functioning HHC/HFF |
| Maternal age at child’s birth, mean (SD), yrs+ | 20.3³ (3.9) | 20.4³ (3.9) | 29.3¹ (5.2) | 21.8² (4.4) |
| Maternal education, mean (SD), yrs+ | 10.7³ (2.0) | 10.9³ (1.7) | 11.6² (2.0) | 12.2¹ (1.9) |
| Family demographic index (0–4), mean (SD)+ | 0.9³ (0.9) | 0.9³ (0.8) | 1.9¹ (1.2) | 1.5² (1.2) |
| Mother employed full- or part-time, %# | 40.9² | 37.8² | 55.1¹ | 58.0¹ |
| Child ineligible for Free Lunch, %# | 5.7² | 8.9² | 27.6¹ | 21.7¹ |
| Two-parent family status, %#- | 20.9³ | 20.8³ | 56.01 | 33.1² |
| No AFDC/TANF participation, %# | 24.4³ | 26.3³ | 60.1³ | 43.0² |
| Number of children, mean (SD)+ | 3.9¹ (1.9) | 3.4² (1.7) | 3.1³ (1.9) | 2.64 (1.4) |
| Number of adults, mean (SD)+ | 1.5² (1.0) | 1.4² (0.7) | 3.1 ¹(1.4) | 1.5² (0.7) |
| Child spent time in a non parent family, % | 37.4¹ | 2.5² | 3.6² | 3.4² |
| Parent(s) reported health or disability problem, % | 14.5 | 10.0 | 13.2 | 8.7 |
| Child maltreatment scale (1–4), mean (SD)+ | 1.3² (0.5) | 3.9¹ (0.2) | 3.9¹ (0.2) | 3.9¹ (0.3) |
| Parent(s) involvement in child’s school (1–5), mean (SD)+ | 1.2³ (1.2) | 1.1³ (1.0) | 1.9² (1.2) | 3.4¹ (1.2) |
| Parent(s) expectations for achievement (1–5), mean (SD)+ | 1.7³ (1.2) | 1.5³ (1.1) | 2.3² (1.2) | 3.5¹ (0.9) |
| Maximum n | 131 | 509 | 288 | 412 |
Note. Means and percentages are unadjusted. Rows with superscripts indicate significant differences (p < .001) between typologies for the typology measure based on Analysis of covariance (ANCOVA) for non dichotomous indicators and Chi-square analyses for dichotomous measures. Means and percentages in the same row that do not share superscripts differ by p < .05 in follow-up Bonferroni corrected t tests.
The smallest and seemingly most disadvantaged family profile was low human capital and low family functioning (LHC/LFF), which represented 9.8% of the families in the study (n = 131). The most distinguishing characteristics of the LHC/LFF profile were those pertaining to child abuse and neglect. All children within this family subgroup were involved with the Illinois Department of Child and Family Services or the Cook County Juvenile Court because of one or more substantiated report of child abuse or neglect. Over one-third of these children (37.4%) spent time in relative and nonrelative foster homes and other non-biological parent settings. Other notable profile characteristics were low rates of maternal education (10.7 years), low scores on the family demographic index (0.9), and low rates of parent involvement (1.2) and expectations (1.7).
The largest cluster representing nearly 40% of the families in the study (n = 509) was the low human capital and moderate family functioning (LHC/MFF) profile. Notable profile characteristics were low rates of maternal education (10.9 years), parent involvement (1.1) and parent expectations (1.5) and low scores on the family demographic index (0.9). In terms of most indicators, this family profile was comparable to the LHC/LFF profile, except for numbers of children and child maltreatment. Relative to the LHC/LFF cluster, the LHC/MFF cluster had significantly fewer numbers of children (3.4 vs 3.9). Moreover, none of the children in the LHC/MFF profile had substantiated reports of child maltreatment.
The second smallest cluster representing 21.5% of study families (n = 288) was the moderate human capital and moderate family functioning profile (MHC/MFF) characterized by moderate rates of maternal education (11.6 years), parent involvement (1.9) and parent expectations (2.3). Relative to all other profiles, families within this cluster scored highest on the family demographics index (1.9). Notable family profile characteristics that distinguished this cluster from all others were maternal age and household composition. On average, mothers in this group were 7 to 8 years older than mothers in other family subgroups. Moreover, relative to all other family subgroups, this family cluster was likely to have twice as many adults in the household. Among the four profiles of families, mothers in the MHC/HFF cluster had the lowest rates of single-parent status (44.0%), and AFDC/TANF participation (39.9%).
The second largest and seemingly least disadvantaged family profile was high human capital and high family functioning (HHC/HFF), representing 30.8% of the families in the study (n = 412). Relative to all others, this family profile had higher rates of maternal education, parent involvement, and parent expectations, and significantly fewer numbers of children. In terms of many other indicators, (e.g., AFDC/TANF participation, Free Lunch eligibility, parent employment, child maltreatment), HHC/HFF families were comparable to MHC/MFF families. Notable differences between the two more advantaged family subgroups were related to family size and structure, maternal education, parental involvement, and expectations. Relative to MHC/MFF families, HHC/HFF families were smaller in size with fewer children (2.6. vs 3.1) and adults (1.5 vs 3.1), more likely to be single-parent status (66.9% vs 44.0%), have mothers with more years of schooling (12.2 vs 11.6) and have higher rates of parental involvement (3.4 vs 1.9) and parental expectations (3.5 vs 2.3).
Family Profiles and the Relationship to Educational Attainment
As shown in Table 4, family profiles were related to eighth grade reading scores F (3, 1258) = 13.74, p < .001, high school graduation, χ2 (3, N = 1,340) = 25.20, p < .001, high school completion, χ2 (3, N = 1,340 = 25.56, p < .001, years of completed schooling, F (3, 1318) = 14.65, p < .001, and college attendance, χ2 (3, N = 1,340) = 18.42, p < .001. Differences in youths’ educational attainment between profiles were sizable and most pronounced between the LHC/LFF and the HHC/HFF profiles. Children in the HHC/HFF profile were more likely than children in the LHC/LFF profile to have higher eighth grade reading scores (149.3 vs 140.5, p <.001), complete high school (77.5% vs 54.1%, p < .001), graduate from high school (53.3 vs 29.4, p < .024), complete more years of schooling (11.81 vs 10.83 years, p < .001), and attend college (36.7% vs 20.8, p < .001). In general, children who had family profiles characterized by higher levels of human capital resources and more favorable scores on indictors of family functioning were more likely than other children to have higher educational attainment. For example, children in the HHC/HFF and the MHC/MFF profiles were more likely than children in the LHC/MFF and the LHC/LFF to have higher eighth grade reading scores, graduate from high school, complete more years of schooling, and attend college. A comparison of the two more advantaged family profiles indicated that children in the HHC/HFF profile were more likely than children in the MHC/MFF profile to have higher eighth grades reading scores (149.31 vs 144.40, p < .001), complete high school (77.5% vs 66.1%, p < .001) and complete more years of schooling (11.81 vs 11.52, p < .041). A comparison of the two more disadvantaged family subgroups indicated that children in the LHC/MFF profile were more likely to than children in the LHC/LFF profile to complete high school, graduate from high school, and complete more years of schooling.
Table 4.
Adjusted Mean and Percentages for Measures of Children’s Educational Attainment by Child Age 12 Family Profile
| Family Profile |
|||||
|---|---|---|---|---|---|
| Educational outcome |
Low Human Capital/ Low Functioning LHC/LFF |
Low Human Capital/ Moderate Functioning LHC/MFF |
Moderate Human Capital/ Moderate Functioning MHC/MFF |
High Human Capital/ High Functioning HHC/HFF |
F/ Chi Square |
| Eighth grade reading scores |
140.50³ | 141.56³ | 144.40² | 149.31¹ |
F (3, 1258) = 13.74, p < .001 |
| High school graduation |
29.4³ | 41.9² | 49.9¹ | 53.3¹ | χ2 (3, N = 1340) = 25.20, p < .001 |
| High school completion |
54.1³ | 63.1² | 66.1² | 77.5¹ | χ2 (3, N = 1340) = 25.56, p < .001 |
| Highest grade completed |
10.83³ | 11.13³ | 11.52² | 11.81¹ |
F (3, 1318) =14.65, p < .001 |
| Any college attendance |
20.8² | 24.3² | 31.5¹ | 36.7¹ | χ2 (3, N = 1340) = 18.42, p < .001 |
| Maximum n /total | 131 | 509 | 288 | 412 | 1340 |
Note. Means and percentages have been adjusted for sex of child, race/ethnicity, CPC program participation, CPC program sites, kindergarten word achievement, any grade retention, and number of school moves by child age 12. Analysis of Covariance (ANCOVA) was used to test for differences for eighth grade reading scores and highest grade completed. Probit hierarchical regression was used to test for differences for high school graduation and completion, and any college attendance. Percentages are based on probit coefficients transformed to marginal effects. Means and percentages in the same row that do not share superscripts differ by p < .05 in follow-up Bonferronni t tests.
Changes in Family Characteristic Profiles for Children between Ages 8 and 12
Table 5 provides a crosstabulation of the age 8 and 12 family classifications to examine the extent to which children experienced a change in family characteristic profiles based on the measures used to develop the family typologies.8 The bolded diagonal represents children with stable family characteristic profiles between ages 8 and 12. Overall, 1003 out of 1338 children (75%) had similar family characteristic profiles at both child age periods. The numbers above the diagonal indicate children who experienced a positive change in family characteristic profiles during early adolescence. Overall, 171 children (12.8%) showed beneficial changes in family characteristic profiles. For example, 61 children who showed LHC/MFF profiles at age 8 had HHC/HFF profiles at age 12. The numbers below the diagonal represent children who experienced a negative change in family characteristic profiles. Overall, 164 children (12.2%) showed negative changes in family characteristic profiles. For example, 85 children who had HHC/HFF profiles at age 8 showed LHC/MFF profiles at age 12.
Table 5.
Crosstabulation of the Child Age 8 and Age 12 Family Profile
| Age 12 | ||||||
|---|---|---|---|---|---|---|
| Classification | LHC/LFF | LHC/MFF | MHC/MFF | HHC/HFF | Total | |
| Age 8 | LHC/LFF | 94 | 0 | 0 | 3 | 97 |
| LHC/MFF | 26 | 410 | 51 | 61 | 548 | |
| MHC/MFF | 2 | 13 | 207 | 56 | 278 | |
| HHC/HFF | 8 | 85 | 30 | 292 | 415 | |
| Total | 130 | 508 | 288 | 412 | 1338 | |
Table 6 illustrates estimates of the effect of positive and negative change in children’s family characteristic profiles on educational attainment after controlling for age 8 family characteristic profiles, positive/negative change, sex of child, race/ethnicity, CPC program participation, CPC program sites, kindergarten reading achievement, number of school moves, and grade retention. Children who experienced a positive change in family characteristic profiles showed higher eighth grade reading scores (b = 5.2, p < .001), completed nearly one-third of a year more of schooling (b = .31, p < .038) and were more likely to attend college (b = 9.0, p <.030). On the contrary, children who experienced a negative change in family characteristic profiles were less likely to graduate from high school (b = −11.0, p <.020), less likely to complete high school (b = −10.9, p < .021) and more likely to complete fewer years of schooling (b = −0.46, p < .003). There was some evidence that the effect of positive or negative change in family characteristic profiles on adolescent’s educational attainment was affected by school-level factors (kindergarten reading achievement, number of school moves, and grade retention. In general, the magnitude of positive or negative change was larger in models that excluded school-level factors. For example, the effect of positive change was raised to statistical significance on high school graduation (b = 10.1, p <.034) and high school completion (b = 9.4, p < .015).
Table 6.
Estimates of Family Profile Change on Children’s Educational Attainment
| Positive change |
Negative change |
|||
|---|---|---|---|---|
| b | p-value | b | p-value | |
| Eighth grade reading scores |
5.15 | .001 | −1.25 | .449 |
| High school graduation |
4.9 | .289 | −11.0 | .020 |
| High school completion |
5.2 | .231 | −10.9 | .021 |
| Highest grade completed |
0.31 | .038 | −0.46 | .003 |
| Any college attendance |
9.0 | .030 | −4.6 | .252 |
Note. Means and percentages have been adjusted for sex of child, race/ethnicity, CPC program participation, CPC program sites, kindergarten word achievement, any grade retention by age 12, and number of school moves by child age 12. MANCOVA was used to test for differences for eighth grade reading scores and highest grade completed. Probit hierarchical regression was used for dichotomous outcomes. Percentages are probit coefficients transformed to marginal effects.
Discussion
Through cluster analysis, this study identified four profiles of families for children who participated in the Chicago Longitudinal Study. In general, children who had family profiles with higher levels of human capital resources and more favorable scores on indictors of family functioning were more likely than other children to have higher levels of educational attainment. Children that showed a negative change (i.e., a decrease in human capital resources and family functioning) between the ages of 8 and 12 were less likely than other children to complete high school, graduate from high school and complete additional years of schooling. In contrast, children who experienced a positive change over the same time period were more likely than other children to have higher eighth grade reading scores, complete additional years of schooling and attend college.
The identification of four subgroups of CLS families demonstrates that there are important differences among low-income families with respect to family human capital resources, family functioning and demographics. These differences counter stereotypes about the homogeneity of poor families and early intervention participants. To date, few studies have described different profiles of low-income families. Identifying different profiles of families can contribute to preventative interventions by providing a framework to view variation among low-income families that can affect how children respond to intervention services. This type of insight could be used to help ecological, multicomponent programs more effectively target and tailor services to meet the different strengths and needs of children and families.
The results of the present study also confirm the importance of human capital resources and family processes within low-income families on adolescents’ educational outcomes. The considerable variation in school outcomes among the four profiles of low-income families identified in the present study indicates that family pathways contributing to the educational attainment of at-risk children are not uniform. The development and use of family typologies can serve to clarify variation in family factors, processes and other experiences that have been shown to influence children’s educational attainment and response to preventative interventions (Bendersky & Lewis, 1994; Janosz, Blanc, Boulerice & Tremblay, 2000; Rutter, 1987; Sameroff, Bartko, Baldwin, Baldwin & Seifer, 1998).
Changes in Family Characteristic Profiles and Effects on School Outcomes
A large body of research has shown that family human capital resources and family poverty characteristics (including timing, duration and intensity) affect children’s cognitive development, socio-emotional functioning and educational attainment (Carneiro & Heckman, 2003; Duncan, Brooks-Gunn & Klebanov, 1994; Hill & Sandfort, 1995; Korenman & Miller, 1997). Few longitudinal studies, however, have documented changes in family human capital resources and poverty characteristics for low-income populations post early childhood. In the present study, nearly, 8 out of 10 participants experienced no major change on family measures of human capital resources and functioning between the ages of 8 and 12. Less than 18% of children with disadvantaged family profiles (i.e., LHC/LFF, LHC/MFF) at 8 years of age showed advantaged family profiles at 12 years of age (i.e., MHC/MFF, HHC/HFF). Similarly, less than 16% of children with advantaged family profiles at 8 years of age had disadvantaged profiles at 12 years of age. These findings are consistent with Sameroff et al (1998) that show a high level of stability for family risk factors between child ages 4, 13, and 18 and that children reported living in high-risk environments at 4 years of age were likely to be in high-risk environments throughout their childhood and adolescence.
In the present study, for the minority of children who experienced a change in family characteristic profiles between the ages of 8 and 12, the effect of the change on their educational attainment was sizable and long-term. A positive change was associated with gains in eighth grade reading scores, years of completed schooling and rates of college attendance. Conversely, a negative change was associated with decreases in rates of high school completion and graduation rates and years of completed schooling. Similar findings have been reported for the effect of change in cumulative social and family risk on children’s IQ. In the Rochester Longitudinal Study, Sameroff et al. (1998) found that children who changed from a high-risk to a low-risk group between the ages of 4 and 13 increased in IQ by 13 points. In contrast, children who changed from a low-risk to a high-risk group during the same age period dropped in IQ by 15 points.
Limitations
The findings of this study should be interpreted within the context of two limitations. The first limitation is that the family profiles were in great part dependent on the input variables. A different or a larger set of family typology measures could yield different results. Similar findings based on alternative family measures would strengthen the reliability and validity of the results.
The second limitation is that the generalizability of findings to other populations and contexts is uncertain. While the consistency of findings across different clustering methods was strong and the 4-cluster solution was replicated across subsamples, tests of the model in different social contexts, in different geographic areas, and with samples that are more ethnically and culturally diverse will provide the strongest tests of the external validity of the findings.
Implications
Our findings have practical implications for preventative research and policy and serve to underscore the body of research that indicates that family resources matter for children’s education. This study suggests that increasing the human capital and family functioning of low-income families can improve the odds that children will be successful in school. Moreover, findings suggest a potentially broader time frame for preventative services in contrast to previous studies that indicate low returns for compensatory interventions post early childhood and that the educational trajectories of at-risk children are well-established by the early years of schooling (Alexander, Entwisle, & Horsey, 1997; Carneiro & Heckman, 2003; Dauber, Alexander, & Entwisle, 1996; Ensminger & Slusarcick, 1992).
Among children in this study enrolled in government-funded early childhood programs because they were considered economically at-risk for school difficulties, family human capital resources and measures of family functioning were found to vary substantially and have strong effects on children’s education. The four profiles of families identified in this study were unique and differentially related to children’s educational attainment suggesting that needs and strengths of poor families may not be optimally met with a single model of service delivery (Ramey, Ramey, and Lanzi, 1998). Programs oriented towards fostering school success and healthy behavior for at-risk populations may be more effective when they provide differentiated services that account for child and family variation.
The results of this study demonstrate that typological frameworks can provide valuable insight about children and families which could be used to strengthen “needs assessments” to help programs more effectively target and tailor services. The differential relationship between family subgroups and measures of educational attainment raises important questions regarding further research on intervention services and effects. Identifying children that are more seriously affected by family poverty factors is warranted given that these children and families may need more intensive interventions. Additional evaluations of children’s outcomes for subgroups of low-income families would likely yield useful information relevant for developing more effective service models for at-risk children and their families.
Acknowledgments
This research was supported by grants from the Institute of Child Health and Human Development (R01HD034294) and the U.S. Department of Education’s Office of Education Research and Improvement (R305T99047 and R306F960055). We would also like to thank the Chicago Public Schools, the Illinois Shared Enrollment and Graduation Consortium, the Illinois Department of Child and Family Services and the Chapin Hall Center for Children at the University of Chicago for their invaluable assistance with data collection. The views expressed in this article are ours and do not represent the granting agencies.
Appendix 1
Correlations Among Key Study Indicators (Child Age 12)
| 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1. Maternal age | 1 | ||||||||||||||
| 2. Maternal education | .132*** | 1 | |||||||||||||
| 3. Family demographic index | .228*** | .255*** | 1 | ||||||||||||
| 4. Parent not employed | −.047 | −.203*** | −.646*** | 1 | |||||||||||
| 5. Free lunch status | −.125*** | −.248*** | −.600*** | .229*** | 1 | ||||||||||
| 6. Single-parent status | −.214*** | −.039 | −.620*** | .146*** | .147 | 1 | |||||||||
| 7. TANF/AFDC | −.206*** | −.176*** | −.672*** | .218*** | .295 | .257*** | 1 | ||||||||
| 8. Number of children | −.060* | −.134*** | −.161*** | .152*** | .186 | −.025 | .110*** | 1 | |||||||
| 9. Number of adults | .300*** | .059* | .202*** | .008 | −.077 | −.211*** | −.234*** | .070 | 1 | ||||||
| 10. Child maltreatment | .144*** | .122*** | .106*** | −.042 | −.078 | −.058* | −.088*** | −.139*** | .060* | 1 | |||||
| 11. Parent involvement | .085** | .130*** | .123*** | −.060* | −.058 | −.070* | −.058* | −.110*** | .016 | .149*** | 1 | ||||
| 12. Parent expectations | .075** | .138*** | .115*** | −.039 | −.072 | −.043 | −.040 | −.122*** | .049+ | .133*** | .584*** | 1 | |||
| 13. High school completion | .052+ | .173*** | .177*** | −.080** | −.118 | −.062* | −.171*** | −.136*** | .033 | .110*** | .248*** | .187*** | 1 | ||
| 14. Highest grade completed | .085** | .179*** | .207*** | −.099*** | −.131 | −.094*** | −.182*** | −.116*** | .065* | .134*** | .287*** | .232*** | .831*** | 1 | |
| 15. CPC preschool | −.025 | .097*** | −.002 | −.003 | −.021 | .019 | .006 | −.098*** | −.010 | .072** | .141*** | .101*** | .093*** | .092*** | 1 |
Note. n = 1340.
Footnotes
Unfortunately, dichotomous measures are not well suited for the standard cluster analytic approach based on Ward’s method (Bergman, Magnusson, & El Khouri, 2003). For this reason, as well as, parsimony these measures were not included as stand alone measures.
Because teacher reports were not available after grade 4 for the parent expectation measure, parent reports in grade 6 were used. The original scale values for the parent reported measure were: 1 = eighth grade, 2 = some high school, 3 = complete high school, 4 = some college, 5 = complete 4-year college, 6 = some graduate school and 7 = complete graduate school. Because the scales from the teacher survey and the parent survey were different, the median answer choice in the parent survey (4 = some college) was used to indicate the cut-point for average parent expectations in the parent survey.
In a twin approach, a missing value in a variable is replaced by the value of that variable in a twin subject that has complete data (see El-Khouri & Bergman, 1992). A total of 104 cases (122 variable values) were imputed in the following four variables: maternal age (n = 58), maternal education (n = 9), number of children (n = 20), and number of adults (n = 35). To examine whether the imputed data was likely to bias the results, a missing data dummy variable indicator was constructed where 1 represented a case that was missing a variable data value and 0 represented a case with no missing data. This indicator was not correlated with any of the educational attainment measures used in the study.
The scales, metrics, and standard deviations of measures can have large impacts on cluster solutions. Standardization prevents measures with large standard deviations (e.g., maternal education, maternal age) from dominating cluster results and profiles (Aldenderfer & Blashfield, 1984; Bailey, 1994; Bergman, Magnusson & El-Khouri, 2003).
The procedure was used to examine the sensitivity of the results to sampling variation. The same hierarchical cluster analysis with Ward’s method of linkage and K-means relocation was applied to a random sample of slightly over 50% of the data. There was no evidence that random sample clusters differed from the full sample clusters.
For dichotomous predictors, marginal effects represent the percentage-point difference between groups derived from the partial derivative evaluated at the mean of the explanatory variable (Greene, 1997).
Several criteria were used to determine the appropriate number of family profiles. These included statistical evaluation criteria, such as, the error sums of squares explained by the classification (ESS), the Point-biserial correlation, and subjective criteria, such as, the expectation of patterns established in previous work, the agglomeration schedule, scree plot and dendrogram, and practical considerations including cell size, theoretical interpretability and utility (see Bergman, Magnusson & El-Khouri, 2003; Milligan, 1981; Milligan & Cooper, 1988). For specific details related to the evaluation statistics and procedures used to determine the four family profiles in the Chicago Longitudinal Study, see Robertson (2004).
The age 8 and age 12 family profiles were compared using the CENTROID module in SLEIPNER. CENTROID compares two cluster solutions by providing descriptive information about the centroid means and variances by providing pairwise comparisons based on the squared average Euclidean distance. The mean of the average squared Euclidian distances can be used as an index of similarities between the centroids in different classifications (Bergman, Magnusson & El-Khouri, 2003). Small differences between cluster pairings in terms of squared Euclidian distance indicate that there are minimal differences between two cluster classes.
Contributor Information
Dylan L. Robertson, University of North Carolina-Chapel Hill
Arthur J. Reynolds, University of Minnesota
REFERENCES
- Aldenderfer MS, Blashfield RK. Cluster analysis. Beverly Hills: Sage Publications; 1984. (Sage university papers series No. 07-044) [Google Scholar]
- Alexander KL, Entwisle DR, Horsey CS. From first grade forward: Early foundations of high school dropout. Sociology of Education. 1997;70:87–107. [Google Scholar]
- Bailey KD. Typologies and taxonomies: An introduction to classification techniques. Beverly Hills: Sage Publications; 1994. (Sage university papers series No. 07-102) [Google Scholar]
- Baydar N, Brooks-Gunn J. Profiles of grandmothers who help care for their grandchildren in the United States. Family Relations (Special issue: The family as a context for health and well-being) 1998;47:385–393. [Google Scholar]
- Becker GS. Human capital: A theoretical and empirical analysis, with special reference to education. Chicago, IL: University of Chicago Press; 1993. [Google Scholar]
- Becker GS. A treatise on the family. Cambridge, MA: Harvard University Press; 1981. [Google Scholar]
- Belsky J, Fearon RMP. Exploring marriage-parenting typologies, their contextual antecedents and developmental sequelae. Development and Psychopathology. 2004;16:501–524. doi: 10.1017/s095457940400464x. [DOI] [PubMed] [Google Scholar]
- Bendersky M, Lewis M. Environmental risk, biological risk, and developmental outcome. Developmental Psychology. 1994;30:484–494. [Google Scholar]
- Bergman LR, El-Khouri BM. Sleipner: A statistical package for pattern-oriented analyses (Version 2.0) [Computer software] Department of Psychology, Stockholm University. 1998 [Google Scholar]
- Bergman LR, Magnusson D, El-Khouri BM. In: Studying individual development in an interindividual context: A person-oriented approach. Magnusson D, editor. Vol. 4. Mahwah, NJ: Erlbaum; 2003. Series of Paths through life. [Google Scholar]
- Brenner V, Fox RA. An empirically derived classification of parenting practices. Journal of Genetic Psychology. 1999;160(3):343–356. doi: 10.1080/00221329909595404. [DOI] [PubMed] [Google Scholar]
- Brooks-Gunn J, Duncan GJ, Leventhal T, Aber JL. Lessons learned and future directions for research on the neighborhoods in which children live. In: Brooks-Gunn J, Duncan GJ, Aber JL, editors. Neighborhood poverty. Vol. 1. New York: Russell Sage Foundation; 1997. pp. 279–297. [Google Scholar]
- Cairns RB. Developmental Science: Three Audacious Implications. In: Bergman LR, Cairns RB, Nilsson L-G, Nystedt L, editors. Developmental Science and the Holistic Approach. Mahweh, NJ: Lawrence Erlbaum Associates; 2000. pp. 49–62. [Google Scholar]
- Cairns RB, Cairns BD, Neckerman HJ. Early school dropout - configurations and determinants. Child Development. 1989;60(6):1437–1452. doi: 10.1111/j.1467-8624.1989.tb04015.x. [DOI] [PubMed] [Google Scholar]
- Carneiro P, Heckman J. Human capital policy. NBER working paper 9495. Cambridge, MA: National Bureau of Economic Research; 2003. [Google Scholar]
- CLS. Chicago Longitudinal Study A Compendium of Internal Reports. Madison, WI: University of Wisconsin, Waisman Center; 2005. Data retrieved on Dec. 13, 2006, from http://education.umn.edu/icd/cls/internal.htm. [Google Scholar]
- Cohen MA. The monetary value of saving a high-risk youth. Journal of Quantitative Criminology. 1998;14(1):5–33. [Google Scholar]
- Danesco ER, Holden EW. Are there different types of homeless families? A typology of homeless families based on cluster analysis, Family Relations. 1998;47(2):159–165. [Google Scholar]
- Dauber S, Alexander KL, Entwisle DR. Tracking and transitions through the middle grades: Channeling educational trajectories. Sociology of Education. 1996;69(4) [Google Scholar]
- Duncan GJ, Brooks-Gunn J, Klebanov PK. Economic deprivation and early childhood development. Child Development. 1994;65(Special Issue No. 2):296–318. [PubMed] [Google Scholar]
- El-Khouri B, Bergman LR. M-prep: a Fortran 77 program for the preparatory analysis of multivariate data. Stockholm, Sweden: Department of Psychology, Stockholm University; 1992. [Google Scholar]
- Ensminger ME, Slusarcick AL. Paths to high-school graduation or dropout: - A longitudinal study of a 1st-grade cohort. Sociology of Education. 1992;65(2):95–113. [Google Scholar]
- Greene W. Econometric Analysis. 3rd. Prentice-Hall; 1997. [Google Scholar]
- Haveman RH, Wolfe BS. Education and economic well-being the role of non- market effects. Madison, Wis: IRP Discussion papers: Discussion papers (University of Wisconsin-Madison, Institute for Research on Poverty; 1983. [Google Scholar]
- Haveman RH, Wolfe BS. Succeeding generations on the effects of investments in children. New York: Russell Sage Foundation; 1994. [Google Scholar]
- Haveman RH, Wolfe BS. The determinants of children’s attainments: a review of methods and findings. Journal of Economic Literature. 1995;33(4):1829–1878. [Google Scholar]
- Heckman J. Policies to foster human capital. Research in Economics. 2000;54(1):3–56. [Google Scholar]
- Hieronymus AN, Hoover HD. Iowa tests of basic skills manual for school administrators (supplement) Chicago: Riverside; 1990. [Google Scholar]
- Hieronymus AN, Lindquist EF, Hoover HD. Iowa tests of basic skills. Early primary battery. Chicago: Riverside; 1980. [Google Scholar]
- Hill MS, Sandfort JR. Effects of childhood poverty on productivity later in life - Implications for public-policy. Children and Youth Services Review. 1995;17(1/2):91–126. [Google Scholar]
- Janosz M, Blank M, Boulerice B, Tremblay RE. Predicting different types of school dropouts: A typological approach with two longitudinal samples. Journal of Educational Psychology. 2000;92(1):171–190. [Google Scholar]
- Johnson VK. Linking Changes in Whole Family Functioning and Children’s Externalizing Behavior Across the Elementary School Years. Journal of Family Psychology. 2003;17(4):499. doi: 10.1037/0893-3200.17.4.499. [DOI] [PubMed] [Google Scholar]
- Karoly L. Investing in the future: reducing poverty through human capital investment. In: Danziger SH, Haveman RH, editors. Understanding Poverty. Cambridge, MA: Harvard University Press; 2001. pp. 314–358. [Google Scholar]
- Kaufman P, Kwon J, Klein S, Chapman C. Dropout Rates in the United States: 1999. Washington, DC: U.S. Government Printing Office; 1999. (NCES 2001–022).U.S. Department of Education. National Center for Education Statistics. [Google Scholar]
- Kaufman P, Alt M, Chapman C. Dropout Rates in the United States: 2001. Washington, DC: U.S. Government Printing Office; 2004. (NCES 2005-046). U.S. Department of Education, National Center for Education Statistics. [Google Scholar]
- Korenman S, Miller JE. Effects of long-term poverty on physical health of children in the National Longitudinal Survey of Youth. In: Duncan G, Brooks-Gunn J, editors. Consequences of Growing Up Poor. New York, NY: Russell Sage Foundation; 1997. pp. 70–99. [Google Scholar]
- Laird J, DeBell M, Chapman C. Dropout Rates in the United States: 2004. Washington, DC: National Center for Education Statistics; 2006. (NCES 2007- 024). U.S. Department of Education Retrieved [date] from http://nces.ed.gov/pubsearch. [Google Scholar]
- Magnusson D. The logic and implications of a person-oriented approach. In: Cairns RB, Bergman LR, Kagan J, editors. Methods and models for studying the individual. Thousand Oaks, CA: Sage Publications; 1998. pp. 33–63. [Google Scholar]
- Magnusson D, Cairns RB. Developmental Science: Toward a Unified Framework. In: Cairns RB, Elder GH Jr, Costello EJ, editors. Developmental Science. Cambridge, England: Cambridge University Press; 1996. pp. 7–30. [Google Scholar]
- Mandara J, Murray CB. Development of an empirical typology of African American family functioning. Journal of Family Psychology. 2002;16(3):318–337. doi: 10.1037//0893-3200.16.3.318. [DOI] [PubMed] [Google Scholar]
- Milligan GW. A Monte Carlo study of thirty internal criterion measures for cluster analysis. Psychometrika. 1981;46:187–199. [Google Scholar]
- Milligan GW, Cooper MC. A study of standardization of variables in cluster analysis. Journal of Classification. 1988;5:181–204. [Google Scholar]
- National Center for Education Statistics. School-level correlates of academic achievement: Statistical analysis report. Washington, D.C: United States Department of Education, Office of Educational Research and Improvement; 2000. (No. 2000-303) [Google Scholar]
- National Science and Technology Council, Office of Science and Technology Policy. Investing in Our Future: A National Research Initiative for America’s Children for the 21st Century. Washington, DC: Executive Office of the President, National Science and Technology Council; 1997. [Google Scholar]
- O’connor TG, Hetherington EM, Reiss D. Family systems and adolescent development: Shared and nonshared risk and protective factors in nondivorced and remarried families. Development and Psychopathology. 1998;10:353–375. doi: 10.1017/s0954579498001643. [DOI] [PubMed] [Google Scholar]
- Ramey CT, Ramey SL, Lanzi RG. Differentiating developmental risk levels for families in poverty: Creating a family typology. In: Lewis M, Feiring C, editors. Families, risks, and competence. Mahwah, NJ, London: Lawrence Erlbaum; 1998. pp. 187–205. [Google Scholar]
- Reynolds AJ. Success in early Intervention: The Chicago Child-Parent Centers. Lincoln: University of Nebraska Press; 2000. [Google Scholar]
- Reynolds AJ, Temple JA, Ou S-R, Robertson DL, Mersky JP, Topitzes JW, Niles MD. Effects of a school-based, early childhood intervention on adult health and well-being. Archives of Pediatric and Adolescent Medicine. 2007;162(8):730–739. doi: 10.1001/archpedi.161.8.730. [DOI] [PubMed] [Google Scholar]
- Reynolds AJ, Temple JA, Robertson DL, Mann EA. Age 21 cost-benefit analysis of the Title I Chicago Child-Parent Centers. Educational Evaluation and Policy Analysis. 2002;24(4):267–303. [Google Scholar]
- Robertson DL. Family typologies and educational attainment in the Chicago Longitudinal Study. Madison: University of Wisconsin; 2004. Unpublished dissertation. [Google Scholar]
- Roeser RW, Peck SC. Patterns and pathways of educational achievement across adolescence: A holistic-developmental perspective. New Directions for Child and Adolescent Development. 2003;101:39–62. doi: 10.1002/cd.81. [DOI] [PubMed] [Google Scholar]
- Rutter M. Psychosocial resilience and protective mechanisms. American Journal of Orthopsychiatry. 1987;57:316–331. doi: 10.1111/j.1939-0025.1987.tb03541.x. [DOI] [PubMed] [Google Scholar]
- Sameroff AJ, Bartko WT, Baldwin A, Baldwin C, Seifer R. Family and social influences on the development of child competence. In: Lewis M, Feiring C, editors. Families, risk, and competence. Mahwah, NJ: Erlbaum; 1998. pp. 161–185. [Google Scholar]
- Sable J, Hoffman L. Characteristics of the 100 Largest Public Elementary and Secondary School Districts in the United States: 2002–03 (NCES 2005–312) Washington, DC: U.S. Government Printing Office; 2005. U.S. Department of Education, National Center for Education Statistics. [Google Scholar]
- Ward J. Hierarchical grouping to optimize an objective function. Journal of American Statistical Association. 1963;58:236–244. [Google Scholar]