Family-School Connections and The Transitions of Low-Income Youths and English Language Learners From Middle School To High School

Developmental Psychology © 2009 American Psychological Association
2009, Vol. 45, No. 4, 1061–1076 0012-1649/09/$12.00 DOI: 10.1037/a0016131
Family–School Connections and the Transitions of Low-Income Youths

and English Language Learners From Middle School to High School
Robert Crosnoe
University of Texas at Austin
The theoretical and policy focus on parental involvement in education has evolved into a consideration
of 2-way connections between families and schools. Working from a social capital perspective empha-
sizing the importance of information in periods and domains of uncertainty, the author tested a specific
application of this reconceptualization in this study. Multilevel models of the National Education
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Longitudinal Study (n ⫽ 17,899) revealed that youths started high school in higher level math when
This document is copyrighted by the American Psychological Association or one of its allied publishers.
parents, middle school personnel, and high school personnel were in contact with each other and when
middle school personnel bridged middle school and high school. The observed effects of other family–
school patterns on math and of all family–school patterns on science were driven by selection, including
the personal characteristics that select adolescents into different family–school configurations. Multiple
forms of family–school communication were related to reduced income and language disparities in math
and science coursework, regardless of their associations with coursework in the general population.
Keywords: education, family, poverty, immigration, policy
On the basis of extensive research in developmental psychology framework into a statistical model, however, requires careful at-
and other disciplines (Comer, 2004; Eccles & Harold, 1993; N. E. tention to the threats to causal inference that often reduce the
Hill et al., 2004), parental involvement in education is widely policy relevance of developmentally oriented studies (Duncan,
viewed as integral to boosting academic progress and reducing Magnuson, & Ludwig, 2004).
academic disparities in American schools. This viewpoint, how- To this end, the National Education Longitudinal Study
ever, is not without its critics, some of whom argue that it focuses (NELS) is analyzed to determine whether communication
too much on parents at the expense of schools and others who among parents, middle school personnel, and high school per-
question the underlying assumption of causality (Hoover-Dempsey sonnel reduces discontinuity in math/science coursework as
& Sandler, 1997; Thurston, 2005). Sorting out these issues is of students transition from middle to high school, with special
vital importance, given the continued centrality of parental in- attention to two student populations (low-income youths and
volvement in numerous large-scale educational policies, including English language learners [ELLs]) widely considered to be at
No Child Left Behind (NCLB; No Child Left Behind Act of 2001 risk for truncated rates of educational attainment. Because
(2002); Organization for Economic Cooperation and Develop- students in this transition are adolescents developmentally ca-
ment, 1997). pable of providing their own input on curricular positioning
Working from a social capital perspective that emphasizes ex-
(Barber & Olsen, 2004), I attempt to account for the selection
changes of information in a social system (Coleman, 1990), I take
role that youths play in the communication patterns between
up this challenge in this study. I view parental involvement in the
home and school. To promote causal inference, propensity
context of two-way streams of communication, between home and
scores and robustness indices are coupled in a strategy likely to
school, intended to promote the academic prospects of youths,
prove valuable to future developmental research (Frisco, Mul-
especially those from historically disadvantaged groups, by en-
ler, & Frank, 2007). The goal in this study is to combine theory
abling more informed decision making by all parties involved. The
focus on information in this conceptual framework highlights the and methods in ways that inject needed specificity into the often
domains and stages of schooling—and related developmental con- vague parental involvement provisions in major policies, by
siderations—in which information gaps are especially conducive identifying concrete actions that some schools can take to create
to disparities in long-term educational attainment. Translating this working models of family–school connections.
Parental Involvement in Education

This research was supported by a William T. Grant Foundation faculty
as an Evolving Concept
scholar award, by National Institute of Child Health and Human Develop-
ment Center Grant R24 HD042849 to the Population Research Center, and
As an aspect of parenting that shapes and responds to the
by an AERA Grants Program small grant. I thank Ken Frank for his advice.
Correspondence concerning this article should be addressed to Robert changing needs of youths, parental involvement in education has
Crosnoe, Department of Sociology and Population Research Center, long interested developmentalists (Eccles & Harold, 1993;
University of Texas at Austin, 1 University Station A1700, Austin, TX Grolnick, Benjet, Kurowski, & Apostleris, 1997; Harvard Family
78712-1088. E-mail: [email protected] Research Project, 2001; N. E. Hill, 2001) and other researchers
1061
1062 CROSNOE
(Alexander, Entwisle, & Olson, 2007; Hoxby, 1998). Generally, effects? Second, how can the concept of family–school connec-
young people achieve more when parents are engaged in their tions be translated into concrete action?
education, including providing cognitive stimulation, meeting with
school personnel, and participating in school activities (Alexander Identifying a Specific Application of the Family–School
et al., 2007; Christenson, 2004; Hoover-Dempsey & Sandler, Connections Concept
1997; Ma, 1999; Raver, Gershoff, & Aber, 2007; Sui-Chu &
Willms, 1996). Of course, these patterns vary by children’s age, The purpose of this study is to answer both of these questions.
school level, and social and academic skills (Eccles & Harold, I do so by using the emphasis on information exchange in the
1993) and are stronger in historically at-risk groups (N. E. Hill, social capital perspective to identify a specific, developmentally
2001; Okagaki & Frensch, 1998). This evidence, as well as per- appropriate form of family–school connections that might achieve
ceptions that involvement is more amenable to large-scale inter- the intended academic goals of the general family–school connec-
vention than are other aspects of parenting, contributes to the tions theory and policy strategy.
greater incorporation of parental involvement into educational Accurate information is more likely to support effective and
policy than many other family dynamics studied by developmen- equitable decision making during periods of uncertainty and when
talists (Crosnoe & Huston, 2007). multiple options are available. When rules, expectations, and re-
Over time, the prevailing conceptual model of parental involve- quirements are unclear or differentially clear across groups, stu-
ment has evolved from being something that parents do to being an dents (and entire groups) can lose ground to peers in ways incom-
exchange between home and school, reflecting awareness that the mensurate with their actual aptitude (Crosnoe & Huston, 2007;
substance and effectiveness of parental involvement is related to Schneider, 2007). For example, school transitions are periods of
how parental involvement is elicited and received by schools uncertainty because they represent entry into new social and or-
(Christenson & Sheridan, 2001; Epstein et al., 2002; Patrikakou, ganizational environments run by people with little knowledge
Weisberg, Redding, & Walberg, 2005). This exchange view is about students’ histories. By virtue of their own educational ex-
advocated in many developmental perspectives on education periences, personal networks, and social standing, some parents
(Hoover-Dempsey et al., 2005; Pianta & Walsh, 1996), especially know what is needed to get ahead in the new environment or to
those rooted in the larger trend in developmental theory (e.g., attract that information from others. This phenomenon contributes
Bronfenbrenner, 1979; Sameroff, 1983) toward capturing transac- to the academic disruptions that occur during transitions and their
tions among contexts. role in academic disparities (Alexander & Entwisle, 1998; Barber
One perspective, social capital, offers insight into why the & Olsen, 2004; Benner & Graham, in press). As another example,
exchange dimension of parental involvement matters (Lareau, coursework decisions are made from a range of options that
2004). Social capital refers to the resources for getting ahead in a proliferate across school levels. To make appropriate decisions,
system that flow through relationship ties. Information that pro- school personnel must know students’ specific abilities and prom-
vides a competitive advantage is one socially traded resource, ise, and parents must know what courses are available, what they
although certainly not the only one (Coleman, 1988). In the realm represent, and what their short- and long-term consequences may
of education, such information could pertain to, for example, be. Indeed, the availability of choice is directly linked to inequality
which schools offer some curriculum, how to secure entry into (Lee, Smith, & Croninger, 1997; Powell, Farrar, & Cohen, 1985).
special programs, or best strategies for home-based skill develop- This link is more pronounced in secondary math/science course-
ment (Furstenberg, Cook, Eccles, Elder, & Sameroff, 1999). When work than in other subjects because it becomes increasingly dif-
family–school connections are formed, they allow the flow of ferentiated and optional as schooling progresses, is more hierar-
information in both directions—information to parents about what chical and cumulative, and is a particularly powerful predictor of
to do for children and information to schools about a child’s college matriculation and completion (Riegle-Crumb, Farkas, &
special needs, talents, and circumstances. This flow promotes Muller, 2006; Shettle et al., 2007).
academic progress by constructing a broader information base for Following theory, then, family–school connections may bring
the formulation of decisions and plans. To the extent that academic returns if they encompass communication about coursework op-
disparities are rooted in differences in parents’ pool of practical tions and consequences during the key secondary school transition:
information about both stated and unstated rules and expectations middle school to high school. The optimal situation would involve
of the educational system, not in their values or aspirations, communication among all three sides of the triangle around this
family–school connections will support efforts to achieve a more transition—parents learning about the available options for stu-
equitable distribution of academic progress (Bryk & Schneider, dents in high school from high school personnel, high school
2002; Farkas, 1996; Lareau, 1989). personnel learning more about incoming students and what they
This conceptualization of family–school connections is evident need from parents, and middle school personnel, who are more
in the policy arena. As a prominent example, NCLB mandates that likely to have established ties to both sides, representing high
school personnel and parents come together to devise compacts— schools and advocating for families. With such triangulation, stu-
and accompanying organization plans—for accomplishing the dents would be less likely to experience “slippage” in their initial
twin goals of boosting achievement and reducing achievement high school math/science level (e.g., lower than prior history and
disparities (Epstein, 2005; Porter & Polikoff, 2007). The incorpo- ability would suggest), which is so important to their curricular
ration of family–school connections into policy, the enormous pathways into college. Less ideal but still potentially valuable
amounts of money involved, and the vagueness of direction that is would be one of the three actors bridging the other two, for
the norm, however, raise critical questions for research. First, how example, middle school personnel consulting with parents and
confident can one be that family–school connections have positive high school personnel, even if the latter two do not consult each
FAMILY–SCHOOL CONNECTIONS 1063
other. Most problematic would be no communication at all in this connections concept in this study, therefore, I use multiple tech-
triangle. niques to address a range of confounds, both known and unknown.
In sum, in this study I attempt to make causal inferences about
Development, Disparities, and Causal Inference the link between family–school communication patterns and math/
science coursework levels of students transitioning from middle
This study, therefore, assesses how patterns of communication school to high school with different backgrounds and histories,
among families, middle schools, and high schools predict students’ paying special attention to any potential added value for low-
initial math/science course level after the transition into high income and ELL youths. Doing so will contribute to the case for
school. This basic model is further specified in three ways. the continued incorporation of family–school connections into
First, working down to such a specific application of the family– educational policy and offer guidance about a specific form such
school connections concept brings up a developmental issue. The connections can take.
young adolescents involved demand more autonomy in their aca-
demic pathways than do children but still require more guidance
and support than do older adolescents. This tension is one reason Method
why transitions into and out of middle school can be so difficult

(Eccles, Midgley, Wigfield, & Buchanan, 1993). Moreover, such Data and Sample
adolescents have had longer histories in the system—with more
NELS is a representative sample of American 8th graders in
time and opportunity to find niches, develop talents, create perso-
1988. Following a two-stage sampling frame, the National Center
nae, identify trouble spots, and be typed by adults—than have
for Education Statistics (NCES) randomly selected youths within
children, but compared with older adolescents, they maintain more
1,052 schools for the base-year data collection in 1988. Parents,
flexibility in what their future pathways might be and enjoy greater
teachers, school administrators, and adolescents were interviewed,
potential for deflection of their extant trajectories (Alexander et al.,
and youths also took achievement tests. This data collection was
2007; Crosnoe & Huston, 2007). Consequently, this study ac-
repeated in 1990 (first follow-up, which corresponded to 10th
counts for the potential of youths’ academic skills, psychological
grade for most sample members), 1992 (second follow-up, 12th
orientations, behaviors, and future expectations to actively or
grade), 1994 (third follow-up, 2 years out of high school), and
passively contribute to the communication lines that form between
2000 (fourth follow-up, 8 years out of high school). Dropouts were
their homes and schools.
followed in a separate survey, and the main sample was freshened
Second, both theory and policy highlight the potential for
over time to maintain representativeness. The sample for the
family–school connections to reduce academic disparities. Two
present study contained the 17,899 adolescents who were part of
such disparities center on low-income youths and ELLs (typically
first or second generation immigrants), both of whom are targeted the 1990 NELS data release but who were not part of the dropout
by NCLB as at risk because of past underachievement, relative to or freshened samples. This study sample included roughly 75% of
peers, on many academic indicators. This pattern extends to the the full base-year sample with the same proportion of low-income
math/science pipeline and the transition into high school (Brooks- and ELL youths (both statuses defined below).
Gunn & Duncan, 1997; Crosnoe, Lopez-Gonzalez, & Muller,
2004; Fitzgerald, 1995; McLoyd, 1998; Rumberger & Larson, Measures
1998; Stanton-Salazar, 2001). One oft-discussed mechanism for
both disparities is the lesser contact between school personnel Math/science performance. To capture math/science course-
(middle class and rarely immigrants themselves) and low-income work at the start of high school, this study measured courses
and immigrant parents, which blocks the flow of information to completed by the start of 10th grade. Following NELS conventions
parents about how to work the system for their children and (see Schneider, Swanson, & Riegle-Crumb, 1998; Stevenson,
interferes with parents’ expression of their wishes and concerns to Schiller, & Schneider, 1994), I organized enrollment patterns into
schools (Lareau, 1989; Suarez-Orozco & Suarez-Orosco, 1995). hierarchical sequences. For math, the sequence had nine points:
Thus, information flowing from school to home is far less likely to 0 ⫽ none, 1 ⫽ remedial math, 2 ⫽ general math, 3 ⫽ prealgebra,
be redundant for low-income and immigrant parents, and informa- 4 ⫽ algebra I, 5 ⫽ geometry, 6 ⫽ algebra II, 7 ⫽ advanced math
tion flowing from home to school is more likely to bridge cultural (precalculus, trigonometry), and 8 ⫽ calculus. For science, the
differences for these same populations. If so, family–school con- sequence had five points: 0 ⫽ none, 1 ⫽ basic/remedial, 2 ⫽
nections would make more of a difference to youths in these physical or earth science, 3 ⫽ biology, 4 ⫽ chemistry, and 5 ⫽
populations (net of prior history), thereby reducing two major physics or advanced science. Mean scores (see Table 1 for de-
math/science disparities. In this study, I tested this possibility. scriptive statistics for study variables) indicated that the average
Third, research on family–school connections is vulnerable to adolescent had completed algebra I (4.16) and physical or earth
misattributed causality, primarily because the factors that sort science (2.75) by the start of 10th grade.
youths into different types of connections can also affect their Objectively, these measures represent ordered sets of hierarchi-
academic progress. Although true of developmental research in cal categories, not continuous variables. Effectively, however, they
general (Foster & Kalil, 2005), this problem is heightened in this tap a linear, quasicontinuous process because of the highly stan-
case because of the scale of investment in family–school connec- dardized, hierarchical nature of math and science coursework in
tions on the policy level. Unfortunately, causal inferences about American schools. Indeed, ample evidence attests that these
family–school connections are still somewhat shaky (Thurston, coursework measures are most accurately analyzed as continuous
2005). In assessing a specific application of the family–school variables akin to educational attainment scales capturing years of
1064 CROSNOE
Table 1
Descriptive Statistics For Primary Study Variables
Characteristic Frequency (%) M SD
Focal adolescent groups

Low-income youth 24.44
English language learner youth 2.31
Academic indicators
Advanced math program (middle school) 30.69
Algebra I enrollment (middle school) 34.25
Math test score (middle school) 36.77 12.19
Math course level (high school) 4.16 1.84
Advanced science program (middle school) 21.91
Biology I enrollment (middle school) 14.97
Science test score (middle school) 18.98 4.91
Science course level (high school) 2.75 1.13

Family–school connections
Triangulation 23.00
Middle school as bridge 5.26
High school as bridge 30.35
Family as bridge 2.11
Disengagement 17.45
Catchall category 21.83
Sociodemographic characteristics
Gender (female) 50.00
Age (years) 13.81 .55
White 66.41
African American 9.53
Latino and Latina 11.41
Asian American 5.92
Other race/ethnicity 1.03
Parent education 3.10 1.16
Family structure (two parent) 65.17
First generation immigration status 5.50
Second generation immigration status 8.87
Third-plus generation immigration status 85.64
School characteristics (8th grade)
Comprehensive school 16.81
School sector
Public 78.28
Catholic 7.59
Non-Catholic private 9.17
School location
Urban 24.81
Suburban 40.40
Rural 29.85
School region
Northeast 18.13
North central 25.18
South 32.96
West 18.78
Note. N ⫽ 17,899.
schooling (Riegle-Crumb et al., 2006; Stevenson et al., 1994). took into account the difficulty of each test item. Also in 8th grade,
Thus, the math/science coursetaking variables are used continu- adolescents reported whether they were enrolled that year in ad-
ously. vanced math and science, algebra I, or biology I. See Table 1 for
To create a baseline from which to compare adolescents on descriptive statistics for these six measures.
math/science coursework in high school, I controlled for several Focal adolescent groups. The ratio of total family income in
aspects of middle school math/science status, including standard- the past year to household size (both parent reported in 1988) was
ized test performance, program level, and coursework. In 8th compared with federal poverty thresholds. Following a common
grade, each adolescent completed tests designed by the Educa- strategy among states implementing NCLB, I created a binary
tional Testing Service for math (including items on computation, marker of low-income status, differentiating youths at 185% of the
equations, and word problems) and science (including items on life federal poverty line (24%). Also in 8th grade, adolescents reported
science, earth science, and chemistry). To reduce floor and ceiling how well they understood, spoke, read, and wrote English, and
effects, NCES created Item Response Theory (IRT) scores that teachers identified limited English proficient adolescents. NCES
created a marker of limited English proficient status, differentiat- household, ranging from a 1 for no high school to a 5 for post-
ing those who responded not well to any language item and had a graduate education), family structure (1 ⫽ two-parent family, 0 ⫽
teacher identify them as low proficient (2%) from all others. all other family forms), and immigration status (dummy variables
Following convention in recent developmental research (Uchiko- for first, second, and third-plus generation, based on reports of
shi, 2005), I have replaced the limited English proficient label with birthplace of parents and youths).
ELL. School characteristics. Additional controls were measured at
Family–school communication. In the base year, parents re- the school level, including school type (1 ⫽ comprehensive, or a
ported whether they had been contacted by personnel at their school that included middle school and high school grades to-
adolescents’ middle schools about course selection and academic gether, 0 ⫽ separate middle and high schools), sector (public,
programs at adolescents’ future high schools. In the first follow- Catholic, non-Catholic private), urbanicity (rural, urban, subur-
up, receiving high schools reported whether their counselors, ban), and region (Northeast, North Central, South, West).
teachers, and administrators met with their middle school counter-
parts to discuss coursework for incoming freshmen. They also Plan of Analyses
reported whether they had parent orientations and parent–teacher
meetings for the parents of these incoming freshmen. These items Analyses proceeded in four basic steps, which I illustrate with
were collapsed into three binary variables (1 ⫽ yes, 0 ⫽ no): high school math coursework. In the first step, this outcome was
middle school outreach to families, middle school and high school regressed on the family–school communication dummy variables
coordination, and high school outreach to families. Next, I com- and then, in sequence, the low-income and ELL markers, the
bined these three markers to create a family–school communica- controls, and the middle school math status and performance
tion typology: (a) triangulation (1 on each variable), (b) middle markers. Results demonstrated the degree to which adolescents
school as bridge (1 on first and second variables), (c) high school experiencing various family–school communication patterns dif-
as bridge (1 on second and third variables), (d) family as bridge (1 fered in math coursework, net of middle school background and a
on first and third variables), and (d) disengagement (0 on each). A large number of co-occurring personal, family, school, and struc-
sixth dummy variable contained all adolescents whose configura- tural factors. These models were estimated with the mixed (mul-
tion on the three constituent items did not fit any of these patterns, tilevel modeling) procedure in SAS (Singer, 1998), with students
which was used after tests revealed little extra value in breaking as Level 1 and schools as Level 2. This approach allowed for more
this category into its various subcategories. precise estimates of school-level control variables, the correction
Because two of three constituent items were reported by school of NELS design effects (e.g., deflated standard errors due to
administrators for all adolescents, whereas the third was reported school-based clustering of data), and sample weighting. Uncondi-
by parents for specific adolescents, this typology mixed school- tional models revealed that the intraclass correlation—the ratio of
level and individual-level measures, which created an imbalance between-school and within-school variance—was .16 for math
when combining these variables. This imbalance was probably coursework and .14 for science, indicating a preponderance of
why the one category that was based solely on school-level vari- variation in the outcomes at the student level rather than the school
ables (high school as bridge) was modal in the typology. Paradox- level.
ically, it potentially overweighted the family variable in the typol- In the second step, I used propensity score techniques to address
ogy. At the same time, the two high school measures tap whether the impact of measurable confounds on any observed associations
an adolescent attended a school in which school outreach to between family–school communication categories and math
parents or middle schools was possible, not whether any given coursework in the multilevel models described above (for more on
adolescent in that school actually experienced school outreach to the logic and mathematics of propensity score techniques, see
their parents or middle school personnel specifically about them. Heckman, Ichimura, & Todd; Rosenbaum & Rubin, 1983). Basi-
The interpretation of triangulation, then, is a collection of adoles- cally, logistic regressions were estimated, predicting each family–
cents whose parents were in contact with their middle schools and school communication category by a large set of covariates. These
who were moving into high schools with a policy of contacting the covariates included all controls listed above as well as three new
parents and middle schools of incoming students. Finally, of the categories of potential confounds: (a) adolescent characteristics
three items, two were specifically about coursework discussion, and statuses (e.g., grades, activity participation, locus of control),
but the third (high school outreach to families) was not specific (b) parent characteristics and behaviors (e.g., school-based in-
enough to know for sure whether coursework was covered. volvement, discussion of coursework), and (c) aspects of middle
Clearly, this is a less than ideal strategy for measuring family– school and high school student body composition (e.g., size, race/
school communication. Yet, these variables were the only avail- ethnic representation, free lunch rates) and curricular organization
able tools for studying three-way communication during this tran- (e.g., course offerings, modes of course assignment). The appendix
sition in a national sample. Thus, the analyses presented here, describes all extra variables incorporated into the propensity
based on this strategy, must be viewed as preliminary, setting the scores. These logistic regressions generated the predicted proba-
stage for future tests with a more fine-tuned measurement ap- bilities of each adolescent being in the focal category of the
proach. dependent variable (e.g., experiencing some type of family–school
Sociodemographic characteristics. To account for systematic communication), given her or his values on this set of factors.
differences in achievement and social contexts, in this study I These predicted probabilities (or propensity scores) measured the
controlled for gender (1 ⫽ female, 0 ⫽ male), age, race/ethnicity average characteristics in one communication category, to identify
(dummy variables for White, African-American, Latino/a, Asian- adolescents in other categories who “looked” like their peers in
American, and other), parent education (maximum level in the that one category.
1066 CROSNOE
Traditionally, developmentalists have used propensity scores to Results

match cases—with mean values on an outcome in the sample of
The Composition and Implications of Five Family–School
treatment cases compared with those in the sample of control cases
that had the same or similar propensity scores as the treatment
Communication Categories
cases (J. Hill, Waldfogel, Brooks-Gunn, & Han, 2005). An alter- As seen in Table 2, young people in the triangulation category
native is to use the propensity scores as either model weights or were in general more socioeconomically advantaged and had
covariates, with control cases over- or undercounted in compari- higher math/science achievement than did youths in the other
sons with treatment cases, according to their propensity to be in the categories of the family–school communication typology. In par-
treatment group. Compared with conventional matching, these ticular, they were less likely to be low-income or ELLs. At the
related strategies allow for the full use of data— unmatchable cases opposite extreme were youths in the disengaged category. In
are still retained—and have been proven to significantly increase between, youths in the middle school as bridge category more
model efficiency, thereby answering many critiques of the propen- closely resembled their peers in the triangulated category, youths
sity score approach (Hirano, Imbens, & Ridder, 2003; Morgan & in the high school as bridge category more closely resembled their
peers in the disengaged category, and youths in the family as
Harding, 2006). For a recent application of this nonmatching

propensity technique to a developmental question, see Frisco and bridge category looked like those in the triangulated and middle
colleagues’ (2007) study of family structure. school as bridge categories in some ways (e.g., test scores) and like
The third step involved the calculation of a post hoc robustness those in the disengaged and high school as bridge categories in
index, the impact threshold for confounding variables (ITCV). others (e.g., race, school context).
This step was taken to address another problem with propensity These descriptive statistics suggest that contacts among fami-
scores—they only account for confounds that are known and that lies, middle schools, and high schools co-occurred with family
can be directly observed. Some known confounds (e.g., genetic advantages and academic success. In particular, this pattern was
effects) could not be measured in NELS, and still others may not true for adolescents in categories characterized by the parent-
as of yet be known. Rather than attempting to control for the reported link—parents noted that they had contact with middle
impact of such confounds, as would be done with instrumental school personnel about adolescents’ future coursework—in any
variable analysis (see Gennetian, Magnuson, & Morrison, 2008), combination with other links. This latter pattern could be method-
ological in nature, in that the parent-reported link was the most
calculating the ITCV quantifies just how powerful those con-
concrete (it tapped actual contact, not a policy) and was
founds would have to be to negate causal inference. In a sense, it
adolescent-specific (vs. all adolescents in a school). It likely also
estimates a confidence interval for the inference based on a model
reflects the well-documented socioeconomic gradient in parent-
coefficient (for a complete description of the ITCV, see Frank,
school interaction.
2000). The equation for the ITCV is rxy ⫺ r#xy/1 ⫺ r#xy, where rxy #
⫽ t/关冑共n ⫺ q ⫺ 1兲兴 ⫹ t , t is the critical t value (usually 1.96),

2 To investigate these descriptive patterns further, I turned to a
multivariate strategy. Preliminary multilevel models—without
n is the sample size, and q refers to the number of model param-
controls—revealed the basic associations of the family–school
eters. When covariates are included in the model, the equation
communication categories with the math outcome. The results of
becomes: ITCV no covariates ⫻ 关冑共1 ⫺ R2xg 兲共1 ⫺ R2yg 兲兴, where g is these preliminary models are summarized here and are not pre-
the set of covariates, R2xg is the R2 value from a regression presented in tables.
dicting the focal independent variable by the covariates, and R2yg is With the disengaged category as the reference, all communica-
the R2 value from a regression predicting the outcome by the tion dummy variables were positively associated with math
covariates. coursework level at the start of high school. Rotating the reference
The post hoc calculation of the ITCV gauges the minimum among the various communication dummy variables provided a
product of the correlation between predictor and confound and test of all pairwise differences among the different categories.
the correlation between outcome and confound (rxc ⫻ ryc) Adolescents in the middle school as bridge category had signifi-
needed to make a focal association from the multilevel model cantly higher math levels than did adolescents in the other cate-
(key predictor 3 outcome) just statistically significant. If the gories. Adolescents in the middle school as bridge and family as
actual (if unknown) product of these correlations—which rep- bridge categories had significantly lower levels than did those in
resents the impact of the unobserved confound on the regression the triangulation category but higher levels than did those in the
equation— exceeds this threshold, including that unobserved high school as bridge category and disengaged categories. Ado-
confound in the regression (if it could be observed) would lescents in the high school as bridge category only had signifi-
likely alter the causal inference based on that regression. This cantly higher levels than adolescents in the disengaged category,
information is important because it provides a benchmark for who had significantly lower levels than did adolescents in all other
assessing the strength of a focal association—low values inval- family–school communication categories.
idate the causal inference, whereas high values indicate robust- This preliminary pattern for the family–school communication
ness of that inference. typology replicated the results when the three items that were used
Finally, in the fourth step, all interaction terms between low- to construct the typology were examined as predictors of the math
income and ELL status on one hand and family–school commu- outcome. Links between middle schools and parents and links
nication dummy variables on the other were added to the compre- between middle schools and high schools both predicted higher
hensive multilevel model. These results indicated whether math coursework levels, but links between high schools and par-
disparities in math coursetaking varied across different family– ents did not. Two- and three-way interactions among these three
school connection categories. binary items indicated that adolescents had the highest math
Table 2
Characteristics of Adolescents in Each Family–School Communication Category
Middle school as High school as

Triangulation bridge Family as bridge bridge Disengaged
Characteristic M SD M SD M SD M SD M SD
Focal adolescent groups

Low income .20c .40 .24b .43 .23b .42 .24b .43 .32a .47
English language learner .01b .12 .02ab .14 .01b .11 .03ab .16 .04a .18
Female .49ab .50 .47b .50 .50ab .50 .53a .50 .50ab .50
White .77a .42 .80a .40 .69b .46 .61c .49 .56d .50
African American .08c .27 .08c .26 .13a .34 .09bc .29 .11ab .32
Latino and Latina .08b .28 .07b .26 .10b .30 .12a .33 .15a .36
Asian American .06a .24 .03b .17 .07a .24 .07a .26 .06a .23
Third-plus generation .86b .35 .91a .29 .84b .37 .85b .36 .85b .36
Parent education 3.39a 1.13 3.26b 1.25 3.35b 1.16 2.98c 1.14 2.79d 1.18
Family structure (two parent) .73a .44 .73a .44 .69b .46 .61c .49 .58c .50
School characteristics
Public middle school .84a .37 .82a .38 .61c .49 .78b .41 .77b .42
Suburban middle school .49a .50 .40b .49 .28c .49 .38b .49 .38b .49
Public high school .88b .32 .84c .36 .63d .48 .91a .29 .85bc .35
Suburban high school .48a .50 .38b .49 .16c .37 .40b .49 .37b .48
Academic indicators
Math test score (middle school) 39.35a 12.59 38.70a 11.92 38.67a 12.70 36.01b 11.83 33.74c 11.31
Science test score (middle school) 19.86a 5.08 20.03a 4.95 18.93b 4.99 18.62b 4.73 17.90c 4.66
Note. For the triangulation group, n ⫽ 4,117. For the middle school as bridge group, n ⫽ 941. For the family as bridge group, n ⫽ 378. For the high
school as bridge group, n ⫽ 5,432. For the disengaged group, n ⫽ 3,124. Means with different subscripts differed significantly ( p ⬍ .05), as determined
by post hoc analysis of variance tests. An a subscript designates the largest mean, with smaller means designated in descending alphabetical order. Statistics
for sixth, catchall, category are not shown.
coursework level at the start of high school when they had both the causal inference, several additional modeling steps were taken (see
middle school–parent link and the middle school– high school link Table 3 for math results).
(akin to the middle school as bridge category), followed by ado- First, a full set of basic sociodemographic confounds that could
lescents with all three links (akin to triangulation) or with both the be observed in the data (including low-income and ELL status)
middle school–parent link and the high school–parent link (akin to were added to the baseline model (Model 1). This addition had a
family as bridge), followed by those with the high school-parent major impact on the basic pattern of results summarized above.
link and the middle school– high school link (akin to high school Specifically, it substantially attenuated all of the family–school
as bridge). Because the results for a set of dummy variables communication coefficients except the high school as bridge co-
representing the typology are easier to interpret and are more efficient. In particular, the family as bridge coefficient was re-
accessible to a general audience than are three-way interactions, I duced to marginal significance. Still, the basic pattern of pairwise
focus on the typology after this point. differences in math coursework at the start of high school—for
A similar pattern was seen for science coursework at the start of example, middle school as bridge associated significantly more
high school, with a few notable deviations. First, the family as
with the outcome than did all others; the disengaged category, less
bridge partnership category joined the middle school as bridge
so—persisted.
category in having the greatest association with science course
Second, various dimensions of middle school math/science sta-
level. Second, looking at the three items used to make the family–
tus and performance that could influence both family–school com-
school typology revealed no three-way interactions. Main effects
munication and high school coursework were controlled (Model
and two-way interactions, however, indicated that the high school-
family link had the weakest association with the outcome of the 2). As expected, all three middle school math measures signifi-
three and chipped away at some benefits of the other two links cantly predicted initial high school math level, and their addition
when paired with them. further attenuated the coefficients for the family–school commu-
nication dummy variables. Thus, some of the reasons that adoles-
cents with different patterns of family–school communication dif-
Looking More Closely at Family–School Communication
fered in initial math level in high school were due to corresponding
and Math/Science Outcomes differences in their middle school math classes and achievement.
These basic patterns sketch the parameters of coursework dif- With one exception, however, the same pattern of pairwise differ-
ferences by family–school communication patterns, but they do ences among the family–school categories remained the same. The
not allow strong conclusions because they are vulnerable to selec- exception was that the family as bridge coefficient no longer
tion and other biases. To address these vulnerabilities and promote differed significantly from the disengaged coefficient.
1068 CROSNOE
Table 3
Results of Multilevel Models Predicting Early High School Math Coursework
Model
1 2 3d
Unstandardized b Unstandardized b Unstandardized b

Characteristic coefficients SE coefficients SE coefficients SE
Focal adolescent group

Low incomea ⫺.22ⴱⴱⴱ .03 ⫺.11ⴱⴱⴱ .03 ⫺.05 .03
English language learner ⫺.66ⴱⴱⴱ .10 ⫺.41ⴱⴱⴱ .09 ⫺.45ⴱⴱⴱ .10
Family–school connections
Triangulationb .23ⴱⴱⴱ .05 .14ⴱⴱ .05 .11ⴱ .05
Middle school as bridge .40ⴱⴱⴱ .08 .28ⴱⴱⴱ .07 .26ⴱⴱⴱ .07
Family as bridge .16 .10 .03 .09 .04 .09

High school as bridge .13ⴱⴱ .05 .10ⴱ .04 .11ⴱ .05

Gender (female) .12ⴱⴱⴱ .03 .15ⴱⴱⴱ .02 .18ⴱⴱⴱ .02
Age (years) ⫺.39ⴱⴱⴱ .02 ⫺.19ⴱⴱⴱ .02 ⫺.17ⴱⴱⴱ .02
African Americanc ⫺.20ⴱⴱⴱ .05 ⫺.09 .05 .05 .05
Latino and Latina ⫺.13ⴱ .05 ⫺.08 .05 .11ⴱ .05
Asian American .31ⴱⴱⴱ .08 .26ⴱⴱⴱ .07 .42ⴱⴱⴱ .07
Other race/ethnicity ⫺.30ⴱ .13 ⫺.11 .12 ⫺.09 .13
Parent education .27ⴱⴱⴱ .01 .10ⴱⴱⴱ .01 .02ⴱ .01
Family structure (two parent) .23ⴱⴱⴱ .03 .16ⴱⴱⴱ .03 .12ⴱⴱⴱ .03
First generation immigration statusc .33ⴱⴱⴱ .08 .16ⴱ .07 .13 .07
Second generation immigration status ⫺.02 .05 ⫺.12ⴱ .05 ⫺.17ⴱⴱⴱ .05
Prior academic performance
Advanced math program (middle school) .31ⴱⴱⴱ .03 .30ⴱⴱⴱ .03
Algebra I enrollment (middle school) .41ⴱⴱⴱ .03 .39ⴱⴱⴱ .03
Math test score (middle school) .05ⴱⴱⴱ .00 .04ⴱⴱⴱ .00
Intercept 8.23ⴱⴱⴱ 3.98ⴱⴱⴱ 3.25ⴱⴱⴱ
⌬ 2 LL ⫺987e ⫺2,995 ⫺1,339
Note. N ⫽ 15,876. All models controlled for school type, sector, location, and region.
a
The category for those missing income information was also included in analyses. b Disengaged was the reference for family–school connections; the set of
dummy variables also contained a catchall category. c White was the reference for race/ethnicity, third-plus generation for immigration status. d Model 3 added
propensity scores cataloging probability of inclusion in each family–school category (including the catchall category). e Indicates the fit statistic for Model 1
relative to a baseline model containing low-income, English language learner, and the family–school connection dummy variables only.
ⴱ
p ⬍ .05. ⴱⴱ p ⬍ .01. ⴱⴱⴱ p ⬍ .001.
Third, propensity scores indexing the extent to which adoles- was parent education. No covariate had a correlation with the
cents fit the profile of the average adolescent in each family– triangulation category of this strength. Although not foolproof, this
school communication category were entered in Model 3 as co- ITCV boosts confidence that the triangulation coefficient would be
variates. Among other things (see Appendix), these propensity robust for the control for other potential confounds. The associa-
scores cataloged multiple aspects of adolescent adjustment and tion between the middle school as bridge category and the outcome
behavior that could have sorted them into different family–school was less robust, with an ITCV of .03 requiring that some unknown
communication patterns. Once middle school math status was confound be correlated with both predictor and outcome at .17 to
taken into account, the control of these propensity scores had little negate the causal inference. The ITCV for the high school as
added impact on the coefficients for the family–school categories bridge category was less than .01, providing little confidence in
or the pairwise differences among them. causal inference.
Fourth, the potential for the final pattern of results from Model The results for science coursework differed (see Table 4). First,
3 to be robust for the control of unobserved or unknown confounds basic high school science differences by family–school communi-
was assessed by calculating the ITCV for each family–school cation category were attenuated by the control for demographic,
communication category that significantly predicted the math out- family, and school characteristics but persisted in reduced form
come. The analyses indicated that the association between trian- (Model 1). Second, the middle school science status and perfor-
gulation category and math coursework (ICTV ⫽ .06) was the mance and the propensity to be in each kind of family–school
most robust. Assuming that the two correlations (predictor with communication category washed out all but one difference among
confound, outcome with confound) were equivalent, this value the family–school communication categories (Models 2 and 3).
means that each would have to exceed .24 to reduce the triangu- Specifically, adolescents in the family as bridge category—the one
lation coefficient to nonsignificance. The only covariates with a category that did not predict math coursework— had marginally
correlation with math coursework exceeding this value were the higher levels of science coursework at the start of high school than
middle school math measures, and the only covariate matching it did their peers in all other family–school communication catego-
Table 4
Results of Multilevel Models Predicting Early High School Science Coursework
Model
1 2 3d
Unstandardized b Unstandardized b Unstandardized b

Characteristic coefficients SE coefficients SE coefficients SE

Low incomea ⫺.07ⴱⴱ .02 ⫺.04ⴱ .02 .01 .02
English language learner ⫺.31ⴱⴱⴱ .06 ⫺.23ⴱⴱⴱ .06 ⫺.23ⴱⴱⴱ .06
Family–school connection .03
Triangulationb .07ⴱ .04 .03 .02 .03
Middle school as bridge .14ⴱⴱ .04 .09ⴱ .04 .05 .05
Family as bridge .16ⴱ .06 .16ⴱⴱ .06 .11 .06

High school as bridge .03 .03 .02 .03 .01 .03

Gender (female) .09ⴱⴱⴱ .02 .13ⴱⴱⴱ .02 .15ⴱⴱⴱ .02
Age (years) ⫺.13ⴱⴱⴱ .02 ⫺.08ⴱⴱⴱ .01 ⫺.07ⴱⴱⴱ .02
African Americanc ⫺.05 .03 .05 .03 .05 .03
Latino and Latina ⫺.03 .03 .03 .03 .07ⴱ .04
Asian American .14ⴱⴱ .05 .16ⴱⴱ .05 .26ⴱⴱⴱ .05
Other race/ethnicity ⫺.06 .08 .02 .07 .10 .08
Parent education .12ⴱⴱⴱ .01 .07ⴱⴱⴱ .01 .01 .01
Family structure (two parent) .10ⴱⴱⴱ .02 .08ⴱⴱⴱ .02 .05ⴱⴱ .02
First generation immigration statusc .15ⴱⴱ .05 .11ⴱⴱ .05 .10ⴱ .05
Second generation immigration status ⫺.06 .03 ⫺.09ⴱ .03 ⫺.13ⴱⴱⴱ .03
Prior academic performance
Advanced science program (middle school) .19ⴱⴱⴱ .02 .16ⴱⴱⴱ .02
Biology I enrollment (middle school) .03 .02 .03 .02
Science test score (middle school) .04ⴱⴱⴱ .00 .03ⴱⴱⴱ .00
Intercept 4.03ⴱⴱⴱ 2.76ⴱⴱⴱ 2.31ⴱⴱⴱ
⌬ 2 LL ⫺485e ⫺615 ⫺1,250
Note. N ⫽ 15,876. All models controlled for school type, sector, location, and region.
a
The category for those missing income information was also included in analyses. b Disengaged was the reference for family–school connections; the set of
dummy variables also contained a catchall category. c White was the reference for race/ethnicity, third-plus generation for immigration status. d Model 3 added
propensity scores cataloging probability of inclusion in each family–school category (including the catchall category). e Indicates the fit statistic for Model 1
relative to a baseline model containing low-income, English language learner, and the family–school connection dummy variables only.
ⴱ
p ⬍ .05. ⴱⴱ p ⬍ .01. ⴱⴱⴱ p ⬍ .001.
ries. Third, the ITCV for this one family–school category that Income, Language, and Family–School Communication
predicted the science outcome was .01, a value indicating a fairly
Moving beyond main effects, the next step was to assess the
low threshold for the correlations of a confound with predictor and
degree to which family–school communication patterns moderated
outcome (approximately .10 for each) to reduce the focal associ-
income and language disparities in initial high school coursework.
ation to nonsignificance. Although this value does not preclude
In other words, did the associations of low-income and ELL status
making a causal inference, it does not inspire much confidence
with the math/science outcomes vary across the different catego-
either.
ries in the family–school communication typology? To test for
To summarize, any link among families, middle schools, and such moderation, I extended the final model from Tables 3 and 4,
high schools was associated with higher level math/science respectively, to include interaction terms between low-income and
coursework for youths transitioning into high school. Most, but ELL statuses on one hand and family–school typology on the other
not all, of these apparent benefits were driven by demographic, (Table 5).
family, and school factors sorting youths into family–school A preliminary model (not shown) including only the income and
communication patterns, especially their prior history in math language markers as predictors revealed a low-income effect size
and science. After controlling for these factors and gauging the of about one third of a standard deviation in the math outcome and
robustness of these observed family–school communication ef- an ELL effect size of about one-half of a standard deviation.
fects for the potential control of unknown confounds, young Moreover, each attenuated the association of the other with the
people tended to do better in math when in the middle school as math outcome, and the two interacted significantly with each other
bridge and triangulation categories. Family–school communi- to predict the math outcome. Basically, that interaction occurred
cation, however, was not related to high school science course- because income-related disparities in math coursework at the start
work in ways robust enough to make sound causal inferences, at of high school were smaller for ELL youths than for non-ELL
least for the full sample. youths. The same patterns were found for science, although the
1070 CROSNOE
Table 5
Selected Results of Multilevel Models Predicting Early High School Math and Science
Coursework With Interactions
Math Science
Unstandardized b Unstandardized b
Characteristic coefficients SE coefficients SE

Low incomea ⫺.04 .06 .01 .04
ELL ⫺.71ⴱⴱⴱ .18 ⫺.45ⴱⴱⴱ .12
Family–school partnership
Triangulationb .12ⴱ .06 ⫺.01 .04
Middle school as bridge .32ⴱⴱⴱ .08 .06 .06
Family as bridge ⫺.11 .11 .05 .07
High school as bridge .14ⴱ .06 ⫺.01 .04

Student ⫻ Partnership interactions

Low Income ⫻ Triangulationb .03 .09 .00 .06
Low Income ⫻ Middle School as Bridge ⫺.06 .13 .01 .08
Low Income ⫻ Family as Bridge .43ⴱ .19 .22ⴱ .11
Low Income ⫻ High School as bridge ⫺.14 .08 ⫺.06 .05
English Language Learner ⫻ Triangulationb .66ⴱ .29 .67ⴱ .20
English Language Learner ⫻ Middle School as Bridge .79ⴱⴱ .32 ⫺.26 .22
English Language Learner ⫻ Family as Bridge 1.19 1.24 1.07 0.83
English Language Learner ⫻ High School as Bridge .58ⴱ .26 .40ⴱ .17
Intercept 3.25ⴱⴱⴱ 2.32ⴱⴱⴱ
⌬ 2 LL (vs. Model 3 in 3–4) ⫺30 ⫺13
Note. N ⫽ 15,876. Models controlled for gender, age, race/ethnicity, parent education, family structure,
immigration status, school type, school sector, school location, school region, middle school test scores/
coursework, and propensity scores cataloging probability of inclusion in each family-school category (including
catchall category).
a
The category for those missing income information was also included in analyses. b Disengaged was the
reference for family–school connections; the set of dummy variables also contained a catchall category, and this
category was also interacted with low-income and English language learner.
ⴱ
p ⬍ .05. ⴱⴱ p ⬍ .01. ⴱⴱⴱ p ⬍ .001.
disparities were smaller in magnitude (one fifth of a standard p ⬍ .01), and high school as bridge (b ⫽ 0.58, p ⬍ .05) categories.
deviation in the science outcome for low-income, one third for The number of ELLs in the family as bridge category was ex-
ELL). The income and language disparities were not attenuated by tremely small, so that no conclusion could be drawn from the
the inclusion of the family–school communication dummy vari- ELL ⫻ Family as Bridge interaction. The predicted math course-
ables as predictors of the math/science outcomes, but they were
attenuated (entirely so for income) by controls for middle school
math status and propensity scores for each family–school commu- 5
nication category. Non-Poor
As for the math moderation models (see Table 5), low-income Low-Income
status only interacted with the family as bridge category (b ⫽ 0.43, 4.5
p ⬍ .05). To interpret this interaction term, I calculated the
predicted math coursework levels for low-income youths in the
Math Coursework Level
family as bridge and disengaged categories (see Figure 1), with all
4
other variables held to their sample means or modes. Low-income
youths and their peers had similar math coursework levels when in
the disengaged category. Low-income students actually had higher
3.5
math coursework levels (22% of a standard deviation) than did
their peers when in the family as bridge category. Thus, the family
as bridge communication pattern appeared to have an added ben-
3
efit for low-income students. This was the only category with a
stronger observed benefit for low-income youths and, therefore,
the only one that would likely improve the relative math course-
2.5
work level (vs. higher income peers), even if other categories Disengaged Family as Bridge
would also likely raise the absolute coursework levels of low- Family-School Communication Categories
income youths.
Also in the math model, ELL status interacted with the trian- Figure 1. Predicted math coursework level at start of high school, by
gulation (b ⫽ 0.66, p ⬍ .05), middle school as bridge (b ⫽ 0.79, family income and family–school communication category.
work levels derived from these interactions are presented in now are tests of whether this reconceptualization has a real and
Figure 2. The language disparities in math coursework were nar- robust academic payoff as well as guidance on how it can be
rower (essentially null) in the triangulation, middle school as translated into concrete actions. Questions of form, timing, targets,
bridge, and high school as bridge categories than they were in the and developmental appropriateness are paramount. Developmen-
disengaged category. This pattern occurred because, although both talists are well-poised to answer these questions.
ELL and non-ELL youths appeared to benefit from all three In this study, I tried to identify an application of the family–
nondisengaged categories, the benefit was greater for ELL youths. school communication concept supported by causal inference.
For example, the observed effect size of the triangulation category Doing so revealed a potential approach to the family–school com-
(vs. disengaged) was 42% of a standard deviation of the math pact of NCLB (see Epstein, 2005) that some schools (e.g., sec-
outcome for ELL youths and about 10% for non-ELL youths. The ondary schools serving diverse populations) may consider as they
corresponding effect sizes were 60%–17% for middle school as try to achieve their performance goals. This application was spe-
bridge and 40%–10% for high school as bridge. cific, not universal, in that it only applied to one specific domain
For science, a similar pattern of interactions was found. The of education in a specific time period. Yet, given the dynamic
exception was the absence of a significant interaction between
nature of development, parenting, and school, universalism is

ELL status and middle school as bridge category. likely to be an unrealistic standard. Instead, specific recommen-
For the most part, low-income youths and ELLs demonstrated
dations are what are needed to enable more schools to make
the same observed coursework benefits as their peers with similar
progress on the NCLB mandate in useful ways, and this study
demographic, personal, academic, family, and school profiles of
offers such recommendations for one subset of students and school
having two or three lines of communication among their families,
communities.
middle schools, and high schools as they transitioned into high
In looking at family–school communication and its link to
school. When they demonstrated differences in these observed
math/science coursework across the transition from middle school
benefits, they always appeared to derive greater benefits from
having such lines of communication than did their peers. Thus, to high school, this study found that much of the story was really
building family–school communication would likely have a bigger about the selection of young people into different kinds of con-
impact on these groups (who were less likely to have such com- nections, not the academic significance of the connections them-
munication) than on other groups, thereby producing some reduc- selves. Still, lines of communication among the home, the sending
tion in income and language disparities in math/science course- school, and the receiving school did appear to come with academic
work at this point. benefits for students in general and did demonstrate the potential to
reduce key disparities in academic outcomes. In discussing these
Discussion and Conclusion policy relevant patterns, I focus on the most trustworthy results; in
other words, observed associations between family–school com-
Developmentalists have done a great deal to shift the concep- munication and math/science outcomes that persisted after observ-
tualization of parental involvement from a decontextualized parent able confounds (e.g., middle school status and performance, family
behavior to an exchange between families and schools (N. E. Hill and school factors, propensity to have different kinds of family–
et al., 2004; Hoover-Dempsey & Sandler, 1997), a shift evident in school communication patterns) were taken into account and after
educational policy, especially including NCLB. What is needed unobservable confounds were assessed with the ITCV.
1. In general, triangulated family–school communication

4.5 Non-ELL was associated with adolescents being in higher level
ELL
4.3
math coursework at the start of high school. Because this
pattern was more pronounced for ELLs and because it
4.1
extended to science coursework for ELLs only, such
3.9 communication was associated with smaller language
Math Coursework Level
disparities in math and science coursework during this

3.7
period.
3.5
3.3
2. In general, middle school as bridge communication was
associated with adolescents being in higher level math
3.1 coursework at the start of high school. Because this
2.9 pattern was more pronounced for ELLs, such communi-
cation was associated with smaller language disparities in
2.7
math coursework during this period.
2.5
Disengaged Triangulation Middle School High School as 3. High school as bridge communication was associated
as Bridge Bridge
with higher level math/science coursework for ELLs
Family-School Communication Categories
only. As a result, it was associated with smaller lan-
Figure 2. Predicted math coursework level at the start of high school, by guage disparities in math/science coursework but not
language status and family–school communication category. ELL ⫽ En- with a higher overall level of such coursework for the
glish language learner. full sample.
1072 CROSNOE
4. Family as bridge communication was associated with 2006). Thus, family–school communication should have mattered
higher level math/science coursework for low-income more to science coursework level during the transition into high
students only. As a result, it was associated with smaller school than to math coursework, but the opposite pattern was
income disparities in math/science coursework but not found. At least in the full sample, any observed links between
with a higher overall level of such coursework for the full family–school communication and science coursework were com-
sample. pletely due to selection on observable (especially, middle school
science status as well as other adolescent characteristics indexed in
Based on these findings as well as previous research on the link propensity scores) and unobservable factors. Family–school com-
between initial high school placement and eventual credit accu- munication appeared to have more independent effects on math
mulation and achievement (Baker & Stevenson, 1987; Crosnoe & coursework. One explanation for this unexpected pattern is that
Huston, 2007), the general conclusion is that building lines of math coursework does have an element of uncertainty but also
communication among parents, high school personnel, and espe- provides a more limited choice set at each point of uncertainty. For
cially, middle school personnel may promote progress toward example, parents and schools may still have decisions to make
multiple goals—promoting continuity between middle school about students’ math coursework, but they can make this decision
math/science status in general and reducing disparities in such

by weighing one or two alternatives (e.g., algebra I or geometry

continuity related to language status. Other kinds of two-way after middle school prealgebra) rather the more plentiful alterna-
communication among the three actors might promote progress to tives involved in science coursework. Regarding math, therefore,
a single goal—reducing income disparities or language disparities, communication between families and schools may have less
but not both and not affecting the overall coursework level of ground to cover and may focus on more specific recommendations.
students in general. This conclusion brings up several questions. In other words, the exchange of information may matter most in
First, what are the mechanisms involved? Drawing on the social times of uncertainty but not too much uncertainty.
capital perspective, this study postulated that the two-way ex- Third, why would different kinds of family–school communi-
change of new, nonredundant information about coursework and cation be associated with income and language disparities? When
adolescents’ histories between home and school was the primary such disparities (e.g., the observed effect of low-income or ELL
mechanism. This theoretical mechanism, however, could not be status on the outcomes) were moderated by family–school com-
tested extensively. This study measured contact and policies munication, it was always in the form of protection (Masten &
of contact among the home, the sending school, and the receiving Coatsworth, 1998). That is, family–school communication was
school. Because of how the questions were asked, I know that two often more (or only) predictive of coursework among low-income
of these points of contact were specifically about the high school and ELL youths than among their peers, which suggests that
coursework of incoming middle school students. I could not be disparities were narrower in the context of certain types of com-
sure about the third point of contact (high school–family) because munication, even if the overall level of achievement in the sample
the question was about the transition experience in general. More- did not change. Yet, this pattern was less pronounced for income
over, even for the two points of contact specifically about course- disparities than for language disparities. Moreover, families
work, the substance of what was actually said or communicated seemed to be the crucial bridge for low-income youths, the middle
cannot be known. Clearly, much more needs to be done in this school more so for ELLs. A likely explanation concerns differ-
regard, including investigations of other forms of social capital, ences in parents’ familiarity with how American schools work.
such as the construction and enforcement of social norms Most of the low-income parents came up through the American
(Coleman, 1990; Furstenberg et al., 1999). Smaller-scale studies educational system, but a large proportion of the parents of ELL
(Benner & Graham, in press) and/or qualitative studies (Lareau, youths (most of whom were immigrants) did not. Moreover, be-
2004) will likely be more valuable in such future endeavors. cause of language barriers, this latter group of parents faced added
Unfortunately, such studies also come with the nonnegligible obstacles in learning (and working) the system (Suarez-Orozco &
disadvantage that they are less equipped to promote generalizabil- Suarez-Orosco, 2001). Thus, information coming from the school
ity and that they cover less diverse (and internally diverse) groups to home was more likely to be redundant for low-income parents
of adolescents, families, and schools than an analysis of nationally than immigrant parents, many of whom were also low-income.
representative data sets, such as NELS. Both establishing gener- With more familiarity in schools and generally fewer linguistic
alizable, robust patterns and identifying mechanisms are important obstacles, low-income parents would also be in a better position
from a theoretical and policy standpoint, but they are hard to than would immigrant parents to make a case for the children at
accomplish at the same time. Thus, coupling studies with different school. For immigrant parents, having middle school personnel
approaches—such as building on the insights gleaned from this involved was crucial. After all, these personnel typically knew the
study with more intensive investigative approaches—is a way to students better than did receiving high school personnel, and these
advance the field and inform policy for now. personnel were also better informed about high school policies and
Second, why did the results differ for math and science? Ac- practices than were parents themselves. They could make up more
cording to theory, social capital (especially information) matters of the information gap—which is quite different from a gap in
more in times of uncertainty. To some extent, the science portion values or motivation—in this population.
of the math/science pipeline introduces more uncertainty than the Fourth, why was triangulation not always the most important
math portion. Math is more vertically organized, with one class form of family–school communication? If family–school commu-
building on another. Science coursework is not as sequential as nication is good, then the maximum lines of communication
math, and some science courses are more easily substituted for among families and schools should have predicted the highest
others or skipped entirely (Lee et al., 1997; Riegle-Crumb et al., level of math/science coursework. Yet, math coursework was
typically lower (and language disparities larger) in the triangula- placement. These reports, which would need to be done for each
tion category than in the middle school as bridge category, and new group of students each year, would then be shared with
income disparities were typically larger in the triangulation cate- adolescents and their families. A more general schoolwide report
gory than in the family as bridge category. The science pattern has (that would not need to be revised annually) would also be pro-
been addressed above—low-income parents needing to be bridges vided, with information on the normative math/science sequences
during transitions so that they can make a case for their children of students in the high school, data on the sequences that best
and explain their children’s circumstances in ways that overcome predict various educational outcomes (e.g., high school graduation,
institutional barriers (e.g., low expectations, discrimination) and college matriculation) on the state and national level, and data on
inertia (Lareau, 1989). One possible explanation for the math the math/science benchmarks that best prepare students for the
finding is that the high school to family link (the piece of the rigors of higher education. Such state and national data could be
triangulation category missing from the middle school as bridge culled from extant reports from the U.S. Department of Education
category) was a reaction to historical problems in the math/science (Shettle et al., 2007). Parents would then be required to sign off on
pipeline; for example, when slippage between middle schools and the adolescents’ report before the start of the high school year.
high schools in math/science coursework has been a concern for a These plans are simple in nature, but they would require some
number of years or when overall performance levels in math/ effort and investment from schools. Essentially, what they would
science coursework have been low for some time. In this case, the do is allow a more even playing field for parents, schools, and
triangulation pattern might more reflect than predict math/science adolescents to make decisions about short-term actions that have
coursework patterns for students in general, relative to the middle long-term consequences.
school as bridge pattern. Despite the remaining questions, the need to couple this national
Fifth, do the findings about the high school as bridge commu- analysis with a more in-depth investigation of the substance of the
nication pattern suggest that high schools play a less integral role communication being exchanged, and the value of investing in
in the process of interest? In general, the high school as bridge even more strategies to determine causality (e.g., instrumental
category had the weakest associations with math/science course- variable analyses; see Gennetian et al., 2008), this study has served
work of the family–school communication categories, largely be- a useful purpose by identifying one potential way for schools to act
cause of the strong selection effects associated with this category. on a concept of developmental research that has been incorporated
This pattern could have been a measurement issue. This category into educational policy. This strategy is a specific one that applies
was based solely on school policies (e.g., high school policies only to a narrow age range of students in one level of the educa-
about parent outreach and partnering with middle schools), not on tional system. As a result, this study needs to be matched by future
links specific to any one adolescent. As a result, it was the one studies that identify other applications of the family–school con-
category in which the engagement of parents was unknown. Many nections concept that may be useful for other curricula and other
potentially negative forces could have sorted young people into segments of the student population differing in educational and
situations in which their parents were absent from the equation, developmental status.
forces that could overwhelm school policies in their educational
careers. Still, I want to stress that the weaker findings about the References
high school as bridge communication category do not let high Alexander, K. L., & Entwisle, D. R. (1988). Achievement in the first two
schools off the hook for managing the transition process. Recall years of school: Patterns and processes. Monographs of the Society for
that the triangulation, middle school as bridge, and family as Research in Child Development, 53, 1–157.
bridge categories could not exist without high school policies Alexander, K. L., Entwisle, D. R., & Olson, L. S. (2007). Lasting conse-
about communication with families and/or middle schools. Thus, quences of the summer learning gap. American Sociological Review, 72,
the findings indicate that high schools need to be involved in the 167–180.
transitions of middle school youths from various populations but Baker, D., & Stevenson, D. (1987). Mothers’ strategies for children’s
need not be the lead actor in these transitions. Parents and middle achievement: Managing the transition to high school. Sociology of
school personnel, who know the background of young people Education, 59, 156 –166.
Barber, B. K., & Olsen, J. A. (2004). Assessing the transitions to middle
better, need to be the leaders.
and high school. Journal of Adolescent Research, 19, 3–30.
Sixth, how can such an application of family–school connec- Benner, A. D., & Graham, S. (in press). The transition to high school as a
tions be implemented? By taking the general concept of family– developmental process among multi-ethnic urban youth. Child Devel-
school connections and the broad family–school compact provi- opment.
sion of NCLB to a more specific level, this study can provide more Bronfenbrenner, U. (1979). The ecology of human development: Experi-
concrete recommendations. At the heart of this study are the ments by nature and design. Cambridge, MA: Harvard Press.
unequal distribution of information about the sequence of math/ Brooks-Gunn, J., & Duncan, G. (1997). The effects of poverty on children.
sequence coursework best preparing students for entry into and Future of Children, 7, 55–71.
completion of higher education and the unequal distribution being Bryk, A., & Schneider, B. (2002). Trust in schools: A core resource for
able to create an uneven playing field at the start of high school, improvement. New York: Russell Sage Foundation.
Christenson, S. L. (2004). The family–school partnership: An opportunity
with a highly cumulative character. The results of this study
to promote the learning competence of all students. School Psychology
suggest that this information may be made up by contact among Review, 33, 83–104.
different actors involved in students’ transitions into high school. Christenson, S. L., & Sheridan, S. M. (2001). School and families: Cre-
High school counselors could partner with their feeder school ating essential connections for learning. New York: Guilford Press.
counterparts to prepare reports on the math/science history of Coleman, J. S. (1988). Social capital and the creation of human capital.
transitioning students, including a recommendation for high school American Journal of Sociology, 94, S95–S120.
1074 CROSNOE
Coleman, J. S. (1990). Foundations of social theory. Cambridge, MA: school behavior, achievement, and aspirations: Demographic variations
Harvard. across adolescence. Child Development, 75, 1491–1509.
Comer, J. (2004). Leave no child behind: Preparing today’s youth for Hirano, K., Imbens, G. W., & Ridder, G. (2003). Efficient estimation of
tomorrow’s world. New Haven, CT: Yale University Press. average treatment effects using the estimated propensity score. Econo-
Crosnoe, R., & Huston, A. C. (2007). Socioeconomic status, schooling, and metrica, 71, 1161–1189.
the developmental trajectories of adolescents. Developmental Psychol- Hoover-Dempsey, K., & Sandler, H. M. (1997). Why do parents become
ogy, 43, 1097–1110. involved in their children’s education. Review of Educational Research,
Crosnoe, R., Lopez-Gonzalez, L., & Muller, C. (2004). Immigration from 67, 3– 42.
Mexico into the math/science pipeline in American education. Social Hoover-Dempsey, K. V., Walker, J. M. T., Sandler, H. M., Whetsel, D.,
Science Quarterly, 85, 1208 –1226. Green, C. L., Wilkins, A. S., & Closson, K. E. (2005). Why do parents
Duncan, G. J., Magnuson, K., & Ludwig, J. (2004). The endogeneity become involved? Research findings and implications. Elementary
problem in developmental studies. Research in Human Development, 1, School Journal, 106, 105–130.
59 – 80. Hoxby, C. M. (1998). Analyzing school choice reforms that use America’s
Eccles, J. S., & Harold, R. D. (1993). Parent-school involvement during the traditional forms of parental choice. In P. E. Peterson & B. C. Hassel
early adolescent years. Teachers College Record, 94, 568 –587. (Eds.), Learning from school choice (pp. 133–156). Washington, DC:
Eccles, J. S., Midgley, C., Wigfield, A., & Buchanan, C. (1993). Devel- The Brookings Institution.
opment during adolescence: The impact of stage-environment fit on Lareau, A. (1989). Home advantage: Social class and parental intervention
young adolescents’ experiences in schools and in families. American in elementary education. London: Falmer Press.
Psychologist, 48, 90 –101. Lareau, A. (2004). Unequal childhoods: Class, race, and family life.
Epstein, J. L. (2005). Attainable goals? The spirit and letter of the No Child Berkeley, CA: University of California.
Left Behind Act on parental involvement. Sociology of Education, 78, Lee, V., Smith, J., & Croninger, R. (1997). How high school organization
179 –182. influences the equitable distribution of learning in mathematics and
Epstein, J., Sanders, M. G., Simon, B. S., Salinas, K. C., Jansorn, N. R., & science. Sociology of Education, 70, 128 –150.
Van Voorhis, F. L. (2002). School, family, and community partnerships: Ma, X. (1999). Dropping out of advanced mathematics: The effects of
Your handbook for action. Thousand Oaks, CA: Corwin Press. parental involvement. Teachers College Record, 101, 60 – 81.
Farkas, G. (1996). Human capital or cultural capital? Ethnicity and Masten, A. S., & Coatsworth, J. D. (1998). The development of compe-
poverty groups in an urban school district. New York: Aldine de tence in favorable and unfavorable environments: Lessons from research
Gruyter. on successful children. American Psychologist, 53, 205–220.
Fitzgerald, J. (1995). English-as-a-second-language (ESL) learners’ cog- McLoyd, V. C. (1998). Socioeconomic disadvantage and child develop-
nitive reading processes: A review of research in the United States. ment. American Psychologist, 53, 185–204.
Review of Education Research, 65, 145–190. Morgan, S. L., & Harding, D. J. (2006). Matching estimators of causal
Foster, E. M., & Kalil, A. (2005). Developmental psychology and public effects: Prospects and pitfalls in theory and practice. Sociological Meth-
policy: Progress and prospects. Developmental Psychology, 41, 827– ods and Research, 35, 3– 60.
831. No Child Left Behind Act of 2001, 115 Stat. 1425 (2002). Available from
Frank, K. A. (2000). Impact of a confounding variable on a regression U.S. Department of Education, at ww.ed.gov/policy/elsec/leg/esea02/
coefficient. Sociological Methods and Research, 29, 147–194. index.html
Frisco, M. L., Muller, C., & Frank, K. (2007). Parents’ union dissolution Okagaki, L., & Frensch, P. A. (1998). Parenting and children’s school
and adolescents’ school performance: Comparing methodological ap- achievement: A multi-ethnic perspective. American Educational Re-
proaches. Journal of Marriage and Family, 69, 721–741. search Journal, 35, 123–144.
Furstenberg, F., Cook, T., Eccles, J., Elder, G. H., Jr., & Sameroff, A. Organization for Economic Cooperation and Development. (1997). Par-
(1999). Managing to make it: Urban families and adolescent success. ents as partners in schooling. Paris: Author.
Chicago, IL: University of Chicago Press. Patrikakou, E. N., Weisberg, R. P., Redding, S., & Walberg, H. J. (Eds.).
Gennetian, L., Magnuson, K., & Morris, P. A. (2007). From statistical (2005). School–family partnerships for children’s success. New York:
associations to causation: What developmentalists can learn from instru- Teachers College Press.
mental variables techniques coupled with experimental data. Develop- Pianta, R. C., & Walsh, D. J. (1996). High-risk children in schools:
mental Psychology, 44, 381–394. Constructing sustaining relationships. New York: Routledge.
Grolnick, W. S., Benjet, C., Kurowski, C. O., & Apostoleris, N. H. (1997). Porter, A. C., & Polikoff, M. S. (2007). NCLB: State interpretations, early
Predictors of parent involvement in children’s schooling. Journal of effects, and suggestions for reauthorization. SRCD Social Policy Report,
Educational Psychology, 89, 538 –548. 11(4), 3–14.
Harvard Family Research Project. (2001). Concepts and models of family Powell, A. G., Farrar, E., & Cohen, D. K. (1985). The shopping mall high
involvement. Available from www.gse.harvard.edu/hfrp/pubs/pubslist.html school: Winners and losers in the educational marketplace. Boston:
Heckman, J. J., Ichimura, H., & Todd, P. (1997). Matching as an econo- Houghton-Mifflin.
metric evaluation estimator: Evidence from evaluating a job training Raver, C. C., Gershoff, E., & Aber, L. (2007). Testing equivalence of
programme. Review of Economic Studies, 64, 605– 654. mediating models of income, parenting, and school readiness for White,
Hill, J., Waldfogel, J., Brooks-Gunn, J., & Han, W. J. (2005). Toward a Black, and Hispanic children in a national sample. Child Development,
better estimate of causal links in child policy: The case of maternal 78, 96 –115.
employment and child outcomes. Developmental Psychology, 41, 833– Riegle-Crumb, C., Farkas, G., & Muller, C. (2006). The role of gender and
850. friendship in advanced course-taking. Sociology of Education, 79, 206 –
Hill, N. E. (2001). Parenting and academic socialization as they relate to 228.
school readiness: The role of ethnicity and family income. Journal of Rosenbaum, P. R., & Rubin, D. B. (1983). The central role of the propen-
Educational Psychology, 93, 686 – 697. sity score in observational studies for causal effects. Biometrika, 70,
Hill, N. E., Castellino, D. R., Lansford, J. E., Nowlin, P., Dodge, K. A., 41–55.
Bates, J., & Petit, G. (2004). Parent-academic involvement as related to Rumberger, R. W., & Larson, K. A. (1998). Toward explaining differences
in educational achievement among Mexican-American language minor- Stanton-Salazar, R. D. (2001). Manufacturing hope and despair: The
ity students. Sociology of Education, 71, 69 –93. school and kin support networks of U.S.–Mexican youth. New York:
Sameroff, A. (1983). Developmental systems: Context and evolution. In P. Teacher’s College.
Mussen (Ed.), Handbook of Child Psychology (Vol. 1, pp. 237–294). Stevenson, D. L., Schiller, K. S., & Schneider, B. (1994). Sequences of
New York: Wiley. opportunities for learning. Sociology of Education, 67, 184 –198.
Schneider, B. (2007). Forming a college-going community in U.S. schools. Suarez-Orozco, C., & Suarez-Orosco, M. (1995). Transformations: Immi-
Seattle, WA: Bill and Melinda Gates Foundation. gration, family life, and achievement motivation among Latino adoles-
Schneider, B., Swanson, C. B., & Riegle-Crumb, C. (1998). Opportunities cents. Stanford, CA: Stanford.
for learning: Course sequences and positional advantages. Social Psy- Suarez-Orozco, C., & Suarez-Orosco, M. (2001). Children of immigration.
chology of Education, 2, 25–53. Cambridge, MA: Harvard.
Shettle, C., Roey, S., Mordica, J., Perkins, R., Nord, C., Teodorovic, J., et Sui-Chu, H., & Willms, J. D. (1996). Effects of parental involvement on
al. (2007). The nation’s report card: America’s high school graduates eighth-grade achievement. Sociology of Education, 69, 126 –141.
(U.S. Department of Education Publication No. NCES 2007– 467). Thurston, D. (2005). Leveling the home advantage: Assessing the
Washington, DC: U.S. Government Printing Office. effectiveness of parental involvement. Sociology of Education, 78,
Singer, J. D. (1998). Using SAS proc mixed to fit multilevel models, 233–249.
hierarchical models, and individual growth models. Journal of Educa- Uchikoshi, Y. (2005). Narrative development in bilingual kindergarteners:
tional and Behavioral Statistics, 24, 323–355. Can Arthur help? Developmental Psychology, 41, 464 – 478.
Appendix
Table A1
Descriptions of Extra Variables for Propensity Scores
Variable Description
Adolescent factors
Grade point average NCES-constructed measure based on adolescents’ 1988 reports of grades in
math, science, English, and social studies. This measure takes the form of
a standard four point grade point average (M ⫽ 2.98, SD ⫽ 0.73).
Locus of control NCES-constructed measure based on the mean of adolescents’ 1988
responses (1 ⫽ strongly agree to 4 ⫽ strongly disagree) to five items,
including “I don’t have enough control over the direction my life is
taking” (M ⫽ .05, SD ⫽ .71). Constituent items were standardized.
Positive self-concept NCES-constructed measure based on the mean of adolescents’ 1988
responses (1 ⫽ strongly agree to 4 ⫽ strongly disagree) to five items,
including “On the whole, I am satisfied with myself” (M ⫽ .02, SD ⫽
.73). Constituent items were reverse coded and standardized.
Timing of immigration In 1988, parents reported when an adolescent born outside the United State
came to the United States. Reports were dichotomized into less than 5
years (1%) or more.
Middle school athletic participation Adolescent reported involvement with varsity sports in 1988 (57%).
High school athletic participation Adolescent reported involvement with any of 30 school clubs and
extracurricular activities in 1990 (75%).
Middle school activity participation Adolescent reported involvement with any of 7 school sports in 1988 (51%).
High school activity participation Adolescent reported involvement with any of 11 school clubs and
extracurricular activities in 1990 (61%).
Expectations for high school Adolescent reported whether he/she expected to enroll in college-preparatory
program in high school (33%).
Future educational expectations Adolescent reported how far he/she expected to get in the educational
system (1 ⫽ won’t finish high school, 2 ⫽ will finish high school, 3 ⫽
vocational school, 4 ⫽ attend college, 5 ⫽ graduate from college, 6 ⫽
enter higher school after college; M ⫽ 4.68, SD ⫽ 1.23).
Parent–family factors
Home language use NCES-constructed measure, based on adolescent and parent interviews,
dichotomized into non-English is primary or sole language (12%) or not.
(table continues)
(Appendix continues)
1076 CROSNOE
Table A1 (continued)
Variable Description
Parent discussion of coursework In 1988, adolescents reported how often (1 ⫽ never to 3 ⫽ often) they
talked with their parents about course selection (M ⫽ 2.50, SD ⫽ 0.50).
Parent school-based involvement Sum of adolescents’ 1988 reports of whether their parents had attended a
school event, spoken with their teacher, visited their classes, or attended a
school event (M ⫽ 1.90, SD ⫽ 1.23).
Extended family structure typology NCES family structure composite extended to include categories for two-
parent (65%), single parent (16%), stepparent (11%), and other families.
Middle school factors
Size NCES composite based on 1988 school administrator reports, with
categories ranging from 1 (1–99 students) to 7 (1200⫹; M ⫽ 3.75, SD ⫽
1.76).
% Race/ethnic minority students School administrator reports of percentage of non-Whites in student body
(M ⫽ 24.27; SD ⫽ 28.03).
% Students on free/reduced lunch School administrator reports of percentage of student body receiving free
and reduced lunches (M ⫽ 22.28; SD ⫽ 21.93).
Math/science instruction requirements School administrator reports of whether full-year instruction in math and
science required for all 8th graders in school (M ⫽ .86, SD ⫽ .34).
Non-English math/science instruction 1988 school administrator reports of whether non-English math and science
instruction was available in school (M ⫽ .04, SD ⫽ .20).
Mode of course assignment Four variables, based on 1988 school administrator reports of how much
influence (1 ⫽ none, 4 ⫽ a lot) parents (M ⫽ 2.12, SD ⫽ 0.82), teachers
(M ⫽ 2.38, SD ⫽ 0.71), counselors (M ⫽ 2.13, SD ⫽ 0.89), and test
scores (M ⫽ 2.40, SD ⫽ 0.80) had in the assignment and/or selection of
high school courses/programs for 8th grade students.
High school factors
Size See middle school description (M ⫽ 4.71, SD ⫽ 2.40).
% Race/ethnic minority students See middle school description (M ⫽ 27.05, SD ⫽ 26.69).
% Wtudents on free/reduced lunch See middle school description (M ⫽ 19.09, SD ⫽ 19.67).
Grade structure Whether school contained an eighth grade (18%) or not, based on school
administrator reports.
Number of math/science teachers School administrators’ estimates of how many teachers they had in various
departments, ranging from 0 (0–5) to 3 (15⫹). The sum for math and
science was taken (M ⫽ 2.20, SD ⫽ 1.76).
Math/science course offerings Proportion of 16 math courses and 17 science courses offered in school,
based on responses of school administrator to a set list of courses in each
subject (M ⫽ .59, SD ⫽ .11).
Mode of course assignment See middle school description (M ⫽ 2.12, SD ⫽ 0.82 for parents; M ⫽
2.38, SD ⫽ 0.71 teachers; M ⫽ 2.13, SD ⫽ 0.89 for counselors; M ⫽
2.40, SD ⫽ 0.80 for test scores).
Note. All sociodemographic and school controls are also included in propensity scores. NCES ⫽ National Center for
Education Statistics.
Received November 1, 2007

Revision received November 4, 2008
Accepted February 23, 2009 䡲

Family-School Connections and The Transitions of Low-Income Youths and English Language Learners From Middle School To High School

Uploaded by

Copyright:

Available Formats

Family-School Connections and The Transitions of Low-Income Youths and English Language Learners From Middle School To High School

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Family-School Connections and The Transitions of Low-Income Youths and English Language Learners From Middle School To High School

Uploaded by

Copyright:

Available Formats

Developmental Psychology © 2009 American Psychological Association

2009, Vol. 45, No. 4, 1061–1076 0012-1649/09/$12.00 DOI: 10.1037/a0016131

Family–School Connections and the Transitions of Low-Income Youths

Keywords: education, family, poverty, immigration, policy

Parental Involvement in Education

why transitions into and out of middle school can be so difficult

Characteristic Frequency (%) M SD

Focal adolescent groups

Science course level (high school) 2.75 1.13

Traditionally, developmentalists have used propensity scores to Results

Harding, 2006). For a recent application of this nonmatching

⫽ t/关冑共n ⫺ q ⫺ 1兲兴 ⫹ t , t is the critical t value (usually 1.96),

Middle school as High school as

Focal adolescent groups

Unstandardized b Unstandardized b Unstandardized b

Focal adolescent group

Family as bridge .16 .10 .03 .09 .04 .09

High school as bridge .13ⴱⴱ .05 .10ⴱ .04 .11ⴱ .05

Unstandardized b Unstandardized b Unstandardized b

Focal adolescent group

Family as bridge .16ⴱ .06 .16ⴱⴱ .06 .11 .06

High school as bridge .03 .03 .02 .03 .01 .03

Focal adolescent group

High school as bridge .14ⴱ .06 ⫺.01 .04

Student ⫻ Partnership interactions

nature of development, parenting, and school, universalism is

1. In general, triangulated family–school communication

disparities in math and science coursework during this

math/science status in general and reducing disparities in such

by weighing one or two alternatives (e.g., algebra I or geometry

Received November 1, 2007

You might also like