Sociology of Education

2023, Vol. 96(1) 62–79

Income-Based Gaps in Ó American Sociological Association 2022
DOI: 10.1177/00380407221138051
College-Going Activities: journals.sagepub.com/home/soe

High School Classes of 1992

and 2004

Preeya Pandya Mbekeani1

Abstract
There has been widespread concern about widening disparities in parental investments that may be asso-
ciated with widening gaps in educational attainment. Using data from the National Education Longitudinal
Study of 1988 and the Education Longitudinal Study of 2002, this study examines parents’ investments and
engagement in the college-going process for two cohorts of high school students, focusing on adolescents
from low- and high-income families. Between the high school classes of 1992 and 2004, income gaps wid-
ened in financial preparation for postsecondary education expenses and student college-admissions test-
preparation. In contrast, the income gap in parents-child conversations about the college-going process
narrowed, due to a larger increase among low-income parents. I examined potential explanations for
growing gaps and found evidence supporting both rising income inequality and changing associations
between income and college-going activities. Implications for educational attainment gaps are discussed.

Keywords
income inequality, parental involvement, class inequality, higher education, savings for college/
education

Scholars have documented widening gaps by fam- There has been substantial discussion about
ily income in educational attainment in the United changes in parental behaviors that might be associ-
States over the past several decades (Bailey and ated with changes in the income gap in educational
Dynarski 2011; Chetty et al. 2014; Reardon outcomes (Reardon 2011a). Existing research on
2011a; Ziol-Guest and Lee 2016). Gaps in college trends by income in parental behaviors, however,
enrollment and completion between adolescents is limited. First, few studies of socioeconomic
from high- and low-income families have gaps in families’ college-going activities, which
increased (Bailey and Dynarski 2011). These dis- document substantial differences by class, have
parities are of concern because completing college examined trends in these behaviors over time
is a key pathway for social mobility and bachelor’s (Buchmann, Condron, and Roscigno 2010; Flaster
degree attainment is associated with a host of pos-
itive life outcomes (Autor 2014; Baum, Kurose,
and Ma 2013). Moreover, these gaps have wid- 1
Brown University, Providence, RI, USA
ened over a period of rising income inequality
(Piketty and Saez 2003, 2020) and growing differ- Corresponding Author:
ences in the life experiences of children from dif- Preeya Pandya Mbekeani, Annenberg Institute for School
Reform, Brown University, Box 1985, Providence, RI
ferent parts of the income distribution (Cooper and
02912, USA.
Pugh 2020; Duncan and Murnane 2011). Email: [email protected]
Mbekeani 63

2018; Hillman, Gast, and George-Jackson 2015; the period of rising income inequality (Bailey
Quadlin and Conwell 2021). Second, studies that and Dynarski 2011; Bastedo and Jaquette 2011;
have examined trends over time have focused on Chetty et al. 2014; Reardon 2011a; Ziol-Guest
aggregate parental investments of time or money and Lee 2016). Reardon (2011a) and Bailey and
rather than education-focused activities (Guryan, Dynarski (2011) sounded an alarm of widening
Hurst, and Kearney 2008; Kornrich 2016; Kornrich income-based gaps in educational achievement
and Furstenberg 2013; Ramey and Ramey 2010; and attainment over the past several decades.
Schneider, Hastings, and LaBriola 2018). These gaps have grown even as college enrollment
In this study, I extend the research on income- and completion rates have increased for low- and
based gaps in parental behaviors and educational high-income students (Bailey and Dynarski
attainment by examining specific measures of paren- 2011; Bastedo and Jaquette 2011; Chetty et al.
tal support and involvement aimed at adolescents’ 2014; Ziol-Guest and Lee 2016). The gap in
college-going process. I examine activities in three enrollment between students from top and bottom
domains: parents’ financial preparation for postsec- income quartile families grew by 12 percentage
ondary education expenses, conversations with their points, from 39 percentage points to 51, between
adolescents about the college-going process, and stu- cohorts that graduated high school around 1980
dents’ college-admissions test-preparation and test- and 2000, and the gap in bachelor’s degree com-
taking. (Hereafter, I refer to these as financial prep- pletion grew by 14 percentage points for these
aration, conversations, and test-taking, respectively.) cohorts (Bailey and Dynarski 2011).
Using a method developed by Reardon (2011b) and In response to these findings, scholars have
data from the National Education Longitudinal sought to understand to what degree widening
Study of 1988 and the Education Longitudinal Study income-based educational gaps may be the direct
of 2002, I document family income-rank-based result of increasing income inequality or growing
trends and gaps in these domains for two cohorts differences between high- and low-income chil-
of high school students, the high school classes of dren in neighborhoods, schools, families, and
1992 and 2004. I also examine the extent to which parental behaviors. Specifically, scholars have
increasing income inequality, diverging family char- hypothesized that widening income-based educa-
acteristics, and changing associations between tion gaps may be due to widening gaps in parental
income and college-going activities explain widen- behaviors (Reardon 2011a).
ing income-based gaps.
I find that some income-based gaps in college-
going activities narrowed and others widened. Spe- Class Divides in Parental Investments
cifically, parental support in the domain of conver-
sations increased for low-, median-, and high-
and Involvement
income students. Parental conversations increased Between 1980 and 2010, as college costs rose con-
the most for low-income students, resulting in a nar- siderably, most U.S. adults maintained the belief
rowing of the gap between high- and low-income that families should bear the responsibility of pay-
families between 1992 and 2004. However, ing for college (Quadlin and Powell 2022). Over
income-based gaps in parent financial preparation the approximate period of this study, the average
for college and student test-taking widened. Diverg- cost (including tuition, room, and board) of attend-
ing family incomes and changing associations ing an in-state public four-year college increased
between income and college-going activities, rather by 27 percent after adjusting for inflation, from
than widening differences in family characteristics, $10,760 (2021 dollars) in 1991–92 to $13,710
account for much of the widening of the gaps. a decade later (Ma and Pender 2021).
It is perhaps unsurprising, then, that children’s
education is increasingly a focus of families’
financial activities (Bandelj and Grigoryeva
BACKGROUND 2021). The introduction in 1996 of 529 College
Class Divides in Educational Savings Plans expanded the set of education-
related financial products available to parents. It
Attainment joined other savings and investment products
A substantial body of work has examined changes already available, including mutual funds, educa-
in the income gap in educational outcomes over tion IRAs (now known as Coverdell Education
64 Sociology of Education 96(1)

Savings Accounts), and Uniform Transfers/Gifts Moreover, other parental behaviors that may
to Minors Accounts (UTMAs/UGMAs). affect children’s educational achievement and
There are sizeable cross-sectional differences attainment may not be captured in surveys of
by family income in parental support aimed at time use or spending. For example, conversations
college-going. Scholars have documented large between parents and students and parents’ stated
socioeconomic differences in financial prepara- aspirations for their children’s educational attain-
tion, with high socioeconomic status parents ment are not measured. To date, only two studies
more willing to pay for college and to save have examined specific income-based parental
more, both of which are associated with students’ activity gaps with children over time, and both
college enrollment (Flaster 2018; Hillman et al. focused on young children (Bassok et al. 2016;
2015; Quadlin and Conwell 2021). Bandelj and Kalil et al. 2016). There is a dearth of research
Grigoryeva (2021) show that wealthier families on these questions related to adolescents, particu-
are using 529 College Savings Plans and increas- larly around the college-going process, a key
ingly saving more for their children’s education access point for increased educational attainment
relative to less wealthy families. Adolescents and social mobility.
from higher-income families are more likely to
engage in costly forms of college-admissions
test-preparation (e.g., for the SAT or ACT) than Explanations for Widening Divides
adolescents from lower-income families, and these
‘‘shadow education’’ activities have implications in Parental Behaviors
for selective-college enrollment (Buchmann et al. Why might upper-income families be increasing
2010). Apart from Bandelj and Grigoryeva their involvement more than lower-income fami-
(2021), however, these studies have not examined lies? One possibility is that the growing impor-
gap changes in these activities over time. tance of a college degree may elicit different
Studies of time use and expenditures document responses from higher- and lower-income parents.
growing differences between upper-income and With growing inequality, higher-income parents
college-educated parents versus lower-income may believe there are now larger negative conse-
and less educated parents. Since the 1990s, time quences to not attending a selective college or
spent with children increased sharply among not completing college (Reeves 2017). They may
highly educated families but not among other fam- invest more because they perceive increased com-
ilies (Ramey and Ramey 2010), with some stabili- petition for scarce spots at selective colleges
zation more recently (Schneider et al. 2018). Edu- (Ramey and Ramey 2010), which confer the larg-
cational spending gaps between high- and low- est average labor market returns (Chetty et al.
income families (defined as top- and bottom- 2017; Long 2008). High-income families may
income-decile families) and more and less edu- feel they must increase their investments to main-
cated parents grew between 1972 and 2010 (Korn- tain educational advantage as inequality moves to
rich 2016; Kornrich and Furstenberg 2013). higher levels of the education system, that is, max-
These studies, however, are limited in that imally maintained inequality (Lucas 2001). Put
they aggregate multiple types of expenditures differently, inequality may be perpetuated through
or time-diary responses. They are unable to differences in quality, not quantity, of education.
observe trends in parental activities that are spe- Another explanation is growing income
cifically education-focused, particularly if fami- inequality. As the gap in income between upper-
lies reallocate investments within aggregate cat- and lower-income families has grown, upper-
egories. For example, studies that examine income families may devote their additional
spending on a category such as extracurricular resources to their children’s educational develop-
lessons, which includes both dance and test ment (Schneider et al. 2018). Consistent with
preparation, may miss trends specific to each this argument, decomposition analyses of early
of these activities. Education-focused invest- childhood spending gap changes indicate that
ments likely have a greater effect on children’s changes in the income distribution explain a large
educational outcomes, and failing to observe portion of the gap (Kornrich 2016). In other
them may obfuscate the specific pathways words, spending gaps may be growing mechani-
through which parental behaviors affect educa- cally because higher-income families increasingly
tional outcomes. have more to spend.
Mbekeani 65

Demographic changes over the past several families of all income levels may invest more in
decades may also have contributed to widening their adolescents because they anticipate a larger
gaps. Several trends are of note. First, children benefit.
of more highly educated mothers are increasingly In summary, the research on income-based
more likely to be raised in families headed by two changes in college attainment and parental invest-
married parents than are children of less educated ments over time is limited in several regards. First,
mothers (McLanahan 2004). Two-parent families, studies of socioeconomic differences in parental
on average, have more time and money to invest in support aimed at college-going have not examined
their children (Hastings and Schneider 2021). This trends in these activities over time. Second, most
increasing correlation over time between socio- studies of parental investments over time focus
economic (i.e., education and income) advantage on investments of time and money broadly, with-
and two-parent family structure suggests income- out an examination of specific education-focused
based parental investment gaps will increase. activities in which parents are engaging. Finally,
Some of this may operate through income inequal- few studies that document gap changes consider
ity (Western, Percheski, and Bloome 2008). How- potential explanations for changes over time.
ever, trends in family structure may independently In this study, I examine trends by family
affect investments. Two-parent families have income in financial preparation, conversations,
more time to spend with their children and poten- and college-admissions test-taking aimed at foster-
tially more money to invest, net of income through ing high school students’ college attainment. I
economies of scale. The second trend is growing focus on low-, median-, and high-income families
differences in median maternal age (McLanahan (defined as families at the 10th, 50th, and 90th
2004). Older mothers, on average, have more income percentile ranks, respectively, of the
years of education, which may allow them to income distribution of families with an adolescent
invest in their children in ways that are more in 10th or 12th grade). My research questions are
developmentally advantageous (Kalil, Ryan, and as follows: (1) What are the trends by family
Corey 2012). Finally, widening differences by income in parents’ engagement in activities that
income in number of children may also affect support their adolescents’ college attendance?
parental investments of time and money such Have income-based gaps grown? and (2) To
that gaps may widen. what extent do changes over time in college-going
These explanations lend themselves to different activity gaps reflect changes in the income gap
hypotheses about which gaps might widen over and characteristics of low-, median-, and high-
time. If parents perceive increased competition income families?
and pressure for students to obtain elite college
degrees, I would anticipate widening gaps across
all three domains: financial preparation, conversa- DATA
tions, and college-admissions test-taking. However,
if the relevant mechanisms are rising income I used data from two National Center for Educa-
inequality and diverging family structures, I would tion Statistics (NCES) surveys of high school stu-
expect widening gaps in financial preparation and dents: the National Education Longitudinal Study
test-taking but not necessarily in conversations. of 1988 (hereafter referred to as the high school
Even as the gaps widen, it is possible that all class of 1992) and the Education Longitudinal
families may increase their college-going efforts Study of 2002 (hereafter referred to as the high
such that rates of parental involvement rise for school class of 2004).1 These are nationally repre-
all income groups. In particular, this may occur sentative longitudinal surveys of high school stu-
because a ‘‘College for All’’ ethos has become dents. Both studies followed students through their
widespread (Carnevale 2008; Goyette 2008; Rose- secondary and postsecondary years and were
nbaum 2001). Investment theory posits that intended to measure similar constructs. Parents/
parents will invest in the development of their guardians, teachers, and other school officials
children’s human capital because they anticipate were also surveyed at various data collection
future returns or benefits to them and their chil- waves. I used student- and parent/guardian-
dren (Becker 1993; Guryan et al. 2008). As the reported survey items and demographic data
monetary and nonmonetary returns to a college from waves conducted when students were in
degree grow (Autor 2014; Baum et al. 2013), high school.
66 Sociology of Education 96(1)

The survey administration processes for these 2007). I generated 20 imputed data sets using Sta-
two studies were similar but not identical. The ta’s mi impute command. To estimate the ana-
class of 1992 was first surveyed when they were lytic models, I used Stata’s mi estimate com-
8th graders (spring 1988); they were also surveyed mand prefix to combine parameter estimates
twice in high school (10th grade in spring 1990 using Rubin’s rules across the imputed data sets.5
and 12th grade in spring 1992). For brevity, I refer
to the second high school data collection for both
studies as 12th grade; however, although it was the
Measures
modal grade, not all students were in 12th grade.2
Parents of this cohort were surveyed twice: when This study focuses on three domains of college-
students were in 8th grade and 12th grade. The going activities: financial preparation for postsec-
class of 2004 was first surveyed when they were ondary education, conversations about the process,
in 10th grade (spring 2002) and again in 12th and college-entrance exam preparation and taking.
grade (spring 2004). Parents were surveyed when I selected measures that were worded nearly iden-
students were in 10th grade. tically across the two data sets to avoid confound-
Critical for this study, both cohorts of students ing changes in wording with changes in parental
were surveyed at the same points in their high involvement over time. Table A.1 in the online
school careers. From each survey, I constructed Supplemental Material provides information about
samples of students who were enrolled in 10th the source of information (student- or parent-
grade and who responded to both waves of high reported), modal cohort grade level at the time
school data collection. I further limited my ana- of reporting, question text, and response categories
lytic sample to students whose parents responded for each outcome variable and the key predictor of
to at least one survey.3 This resulted in a sample interest, family income.
of 16,210 students in the class of 1992 and a sam-
ple of 12,530 students in the class of 2004.4 The Financial preparation. This domain com-
fact that parents were surveyed at different points prises three measures from parent survey items.
in students’ high school careers was not ideal; I The first is whether parents had set up a college
address this concern in describing the parent- investment fund to financially prepare for their
reported measures. teenager’s education after high school. This item
I conducted multiple imputation using chained is ideal in that it was worded identically in the
equations to address item-missing data on the fam- two surveys and directly related to college-going,
ily income and family characteristics variables yet what constitutes ‘‘college investment funds’’
(White, Royston, and Wood 2011). Approximately changed with the introduction of 529 College Sav-
15 percent of weighted cases were missing family ings Plans in 1996. As a robustness check, I used
income information in the class of 1992. The a measure of whether parents had made invest-
weighted share of missing values on the family ments in stocks or real estate to prepare financially
characteristics variables ranged from 6.4 percent for their teen’s education after high school.
(mother’s age) to 12.8 percent (number of sib- Results are directionally consistent across the
lings). The class of 2004 had no missing values two measures.
on income because these data were imputed by The third measure is how much money parents
NCES. Approximately 8 percent of weighted had set aside for their teenager’s future educa-
cases were missing the number of siblings in the tional needs. Parents were asked to select a dollar
home, and 9 percent of weighted cases were miss- value range from seven or eight categories, rang-
ing mother’s age. The imputation model has all the ing from ‘‘none’’ to ‘‘more than $30,000’’
variables described in the next section as well as (1992) and ‘‘none’’ to ‘‘more than 50,000’’
auxiliary variables hypothesized to be related to (2004). Except for the first and last categories, I
the imputed variables: student race/ethnicity, gen- recoded each category to its midpoint value. I
der, school type (public or private), math achieve- maintained the first category as zero, and for the
ment quartile, parents’ highest education level, topmost category, which was bounded only at
and student year of birth. I include the dependent the minimum, I multiplied the value by 1.4
variables in the imputation model, but I exclude (Card and DiNardo 2002; Mouw and Kalleberg
cases with missing dependent variables from my 2010; Weeden et al. 2007).6 I inflated values to
analytic models (Allison 2001; von Hippel 2021 dollars using the Consumer Price Index-All
Mbekeani 67

Urban Consumers. The resulting values range The second variable is whether students had
from zero to $81,117 (1992) and zero to taken the SAT or ACT college-admissions tests.
$105,436 (2004). Many of the programs and policies that increased
That parents were surveyed at different points access to ACT and SAT test-taking among more
in students’ high school careers is not ideal, but recent cohorts, such as statewide or district-wide
it is not a significant threat to the validity of testing, were not widespread when these cohorts
results. Research on parental rates of saving by were in high school.9 Therefore, students’
grade level is limited; however, I posit that parents college-admissions test-taking can reasonably be
of older students have, on average, saved more.7 considered within the umbrella of family col-
The financial preparation measures are correlated lege-going activities for these cohorts.
with grade, but given that the later grade is in
the earlier data set (12th grade for the class of Family income. Family income is the key pre-
1992 and 10th grade for the class of 2004), esti- dictor of interest, and I primarily operationalize it
mates of increased preparation or widening gaps as family income percentile rank. Family income
are likely conservative. was reported by a parent or guardian. Parents
were asked to select the relevant range for the
Conversations. This domain comprises three total family income in the prior calendar year
student-reported measures. Two questions asked (1991 for the class of 1992 and 2001 for the class
students to report how frequently they discussed of 2004). Parents selected from 15 (1992) and 13
with their parents going to college (10th grade) (2004) categories. From these, I generated a mea-
and college-entrance exam preparation (12th sure of family income percentile rank (Reardon
grade). Response options for these items were 2011b), which I describe in the next section.10
‘‘never,’’ ‘‘sometimes,’’ or ‘‘often.’’ I dichoto- For some analyses, I used family income (not
mized these, with 1 indicating often and 0 indicat- income rank) to estimate income inequality among
ing never or sometimes, because the relevant mar- families in each cohort. Following Reardon
gin along which gaps seemed likely to grow was (2011b), I used the Current Population Survey
the intensive rather than the extensive. Addition- (CPS) to obtain measures of household income,
ally, this approach resulted in less loss of informa- rather than using the NCES datasets.11 Using
tion given that ‘‘never’’ was the least selected CPS data from 1992 and 2002 (Flood et al.
option. 2020), I identified families with adolescents
In 10th grade, students were asked how far in (who were 17 or 18 in 1992 and 15 or 16 in
school they think their mother (or stepmother/ 2002) and ranked them, using the relevant
female guardian) wanted them to go. Students weights, to generate the income distribution for
were presented a series of mutually exclusive households with adolescents.
options, ranging from less than high school gradu- I matched household income from the CPS,
ation to more than a college degree. From these, I using its rank, to the estimated income percentile
created a binary measure indicating whether their ranks in my data sets. This provided me with an
mother wanted them to complete a bachelor’s income value for each income bin in my NCES
degree or more. Although this is not a direct mea- data sets. I used a nearest neighbor matching strat-
sure of conversations, I consider it a proxy for egy when there was not an exact match between an
whether parents communicated their aspirations income rank value in my sample and in the CPS
to their adolescent. data.12 I also identified incomes at the 10th,
50th, and 90th percentile ranks of the income dis-
Test-taking. I constructed two measures from tribution for each cohort.
items in the 12th-grade student survey. I created I inflated income values to 2021 dollars using
a binary variable for students who indicated they the Consumer Price Index.13
had engaged in or planned to engage in non-
school-based college-entrance exam preparation. Demographic measures. I used measures of
This encompassed taking a private course, work- family structure, number of siblings, maternal age,
ing with a private tutor, or studying with commer- and parental education. I categorized families into
cially available test-preparation materials (e.g., four types: two biological/adoptive parents, one
books, computer programs).8 biological parent and one stepparent/partner,
68 Sociology of Education 96(1)

single biological/adoptive parent, and other. The from the fitted parameters of Equation 1. Second,
measure of siblings counts those in the home I computed gaps (e.g., 90/10) and associated stan-
and was top-coded at seven in both surveys. For dard errors for each cohort. Using a fully inter-
the class of 2004, there was a continuous measure acted model, I conducted t tests to assess formally
of maternal age; however, for the class of 1992, whether income-rank-based parental investment
there were four categories for maternal age at gaps changed between cohorts. The presentation
the time of the teen’s birth. I categorized the con- of results focuses on the 90/10 gaps; 90/50, 50/
tinuous age variable in the later cohort to match 10, and 75/25 gaps are presented in the online Sup-
the categories in the earlier cohort. The four cate- plemental Material (Table B.1). The 75/25 gap
gories were less than 20 years old, 20 to 24, 25 to changes were attenuated relative to 90/10 gap
29, and 30 or older. The parental education vari- changes, indicating that differences across cohorts
able had four categories of parents’ highest educa- were largely driven by changes among high- and
tion level: less than a high school diploma, high low-income families rather than families at less
school diploma, some college, and bachelor’s extreme positions in the income distribution.
degree or higher graduate degree. Table 1 lists Aligned with research documenting differences
the demographic categories. within income by gender or race (e.g., Conwell
2021; Quadlin and Conwell 2021), I examined
heterogeneity in 90/10 gap changes by race,
METHODS math achievement, and gender. I stratified the
sample across each dimension and estimated 90/
I would ideally have a continuous measure of fam- 10 gap changes within subgroups, but I did not for-
ily income and fit a polynomial regression model mally test differences across groups. Results by
predicting each college-going activity measure as race and math achievement are presented in the
a function of a family’s percentile rank. However, online Supplemental Material. I found no notable
because income was measured in discrete catego- differences by gender.14
ries, I used Reardon’s (2011b) method for estimat- I examined potential explanations for widening
ing a continuous income distribution from a set of income-based gaps in college-going activities
ordered income categories. The first step was to using regression analyses and estimating gaps
identify the proportion of the population in each under three counterfactual scenarios.
income bin. The next step determined the percen- First, I fitted a within-1992-cohort regression
tile ranks associated with the lower and upper model predicting a given college-going measure,
bounds and midpoint of each binned income Y, as a function of household income with
category. higher-order terms and family characteristics.
I then estimated the following regression The model was off the following form:
equation:
Yic 5 b0c 1 b1c HHinci 1 b2c HHinc2i
Y 5 b0 1 b1 Q 1 b2 Q2 1 b3 Q3 1 e; ð1Þ
1 b3c HHinc3i 1 gc FamilyStructurei
ð2Þ
where Y is a college-going activity and Q indi- 1 dc Siblingsi 1 zc MaternalAgei
cates the percentile rank associated with the mid- 1 hc ParentEdi 1 eic ;
point of the family’s income category. I used
visual inspection and statistical tests to determine where b1c , .., and b3c are coefficients on income
the lowest degree polynomial that reasonably fit for the 1992 cohort, gc is the vector of coefficients
the data. In most cases, this was degree three, so, on family structure, dc is the coefficient on number
for simplicity, I used a cubic function for all of siblings, zc represents the vector of coefficients
measures. on the maternal age categories, and hc is the vec-
To examine income-based trends and gaps in tor of coefficients for parent education categories.
college-going activities, I made two types of com- Using the coefficients from this model, I estimated
parisons. First, I tested differences between two counterfactual scenarios: changing income
cohorts in average levels of parental investments distribution and changing family characteristics.
at the 10th, 50th, and 90th percentile ranks of Under the first, I predict the 90/10 gap in a college
the family income distribution. These estimates activity if the income distribution had changed to
and associated standard errors were generated that of the 2004 cohort and the distribution of
 Table 1. Estimated Demographic and Family Characteristics, 10th, 50th, and 90th Income Percentile Ranks: High School Classes of 1992 and 2004.

10th Percentile 50th Percentile 90th Percentile

1992 2004 D 1992 2004 D 1992 2004 D

Mother’s age at teen’s birth

\ 20 0.17 0.14 20.03 0.10 0.08 20.02 0.05 0.04 20.01
20–24 0.37 0.37 0.00 0.32 0.26 20.05 0.25 0.17 20.08
25–29 0.25 0.27 0.02 0.35 0.33 20.01 0.44 0.36 20.08
 30 0.21 0.22 0.01 0.24 0.32 0.09 0.25 0.43 0.18
Family type
Two married biological or adoptive parents 0.45 0.32 20.13 0.71 0.61 20.10 0.86 0.80 20.07
Stepfamily 0.16 0.15 20.01 0.14 0.18 0.04 0.10 0.15 0.05
Single parent 0.37 0.46 0.10 0.14 0.18 0.04 0.04 0.05 0.01
Other 0.02 0.07 0.05 0.01 0.04 0.03 0.00 0.01 0.01
Parent education
Less than high school 0.25 0.16 20.09 0.04 0.04 0.00 0.00 0.01 0.00
High school 0.30 0.34 0.05 0.23 0.19 20.03 0.08 0.06 20.02
Some college 0.41 0.38 20.03 0.48 0.40 20.08 0.28 0.25 20.02
Bachelor’s degree and higher 0.05 0.12 0.07 0.25 0.37 0.12 0.64 0.68 0.04
Teen’s Race
Asian (non-Hispanic) 0.03 0.04 0.01 0.04 0.04 0.00 0.04 0.03 20.01
Black (non-Hispanic) 0.25 0.24 20.01 0.11 0.12 0.01 0.04 0.05 0.00
Hispanic 0.17 0.27 0.09 0.08 0.14 0.06 0.03 0.06 0.03
White (non-Hispanic) 0.53 0.40 20.14 0.76 0.65 20.11 0.88 0.83 20.06
Other 0.02 0.05 0.04 0.01 0.05 0.04 0.00 0.04 0.04
Siblings in the home 1.54 1.52 20.02 1.49 1.45 20.04 1.45 1.37 20.07
Household income (2021$) 21,510 24,481 2,970 81,570 84,703 3,133 171,083 202,020 30,937

Note: Estimated demographic characteristics were predicted from a fully interacted multinomial logit model, with a linear function of income rank. Sample sizes, rounded to the
nearest 10, are 16,210 (class of 1992) and 12,530 (class of 2004). Household income was obtained from the Current Population Survey and was inflated using the Consumer Price
Index.

69
70 Sociology of Education 96(1)

family characteristics remained unchanged (i.e., Despite these differences, three demographic
across all four measures). Under the second sce- trends are evident across the income distribution.
nario, I substituted 2004 family characteristics First, the share of students raised in two-parent
(structure, siblings, maternal age, and parent edu- families declined, by 7 to 13 percentage points
cation) and kept income at its 1992 distribution. (pp) across income groups. Second, the proportion
In the third scenario, I address the question of of parents who reported having completed a bach-
what if income and family characteristics stayed elor’s degree increased, on average. Third, the
at their 1992 levels and only the associations average number of siblings declined.
(i.e., coefficients) changed. I estimated the same The data present a mixed picture as to whether
model presented in Equation 2 for the 2004 cohort. income-based family characteristic gaps increased
Using the coefficients, I predict the 90/10 gap in ways that could potentially exacerbate inequal-
using the 1992 distributions for income and family ity in parent involvement in the college-going pro-
characteristics. cess. On one hand, the 90/10 gap in the proportion
If gap changes between cohorts were due to of adolescents born to older mothers widened sub-
changes in income distribution, then the gap esti- stantially, from 4 pp to 21, and the 90/10 gap in the
mates from the first counterfactual scenario should proportion of adolescents raised in two-parent
approximate the estimated gaps for the 2004 cohort. families widened (41 pp to 48 pp). Differences
If gap changes were due to changes in family struc- in the number of siblings also widened, with the
ture and parent attributes, then gap estimates from 90/10 gap growing from –0.09 to –0.15. On the
the second scenario should approximate the esti- other hand, the 90/10 gap in the proportion of ado-
mated 2004 gaps. If the gap changes were due to lescents who were born to younger mothers (i.e.,
changes in the associations between outcomes and under age 20) narrowed slightly, from –12 pp to
predictors, then estimates from the third scenario –10 pp, and the 90/10 gap in the proportion of
should approximate the gaps in 2004. A fourth pos- parents with a bachelor’s degree or higher nar-
sibility is that the changes between cohorts were the rowed by 3 pp.
result of a combination of the three.
All models account for the complex surveys by
utilizing weights and the survey design variables. I A Background of Increasing Income
used panel weights associated with the 10th- and Inequality
12th-grade waves of data collection.15 I used Stata’s Income gaps between adolescents from high- and
svyset and svy prefix, using information on the low-income families widened between the two
primary sampling unit and stratum, to estimate Tay- cohorts, primarily due to high-income families
lor linearized standard errors. pulling away from their low- and median-income
peers. Table 1 displays family incomes at the
10th, 50th, and 90th percentile ranks for both
RESULTS cohorts. The 90/10 income gap widened by just
under $28,000, from $149,9600 to $177,500,
Changes in Family Characteristics between 1992 and 2004. The 90/50 income gap
There were stark differences in the estimated fam- widened by about the same amount, from
ily characteristics of children at the 10th and 90th $89,500 to $117,300, and the 50/10 gap remained
percentile ranks of the income distribution of fam- approximately the same ($60,000).
ilies with a high school student (Table 1).16
Approximately 15 percent of low-income students
were born to mothers younger than 20 years old; Changes in College-Going Activities
this contrasts with 5 percent of high-income stu- Average levels of college-going activities
dents. Among high-income adolescents, 80 per- increased between the two cohorts of students
cent lived in two-parent households, compared to (Table 2). The proportion of students reporting
45 percent of low-income adolescents. Finally, that their mother wanted them to obtain a bache-
10 percent of low-income adolescents had at least lor’s degree increased by 12 pp from a base of
one parent with a bachelor’s degree or more, 71 percent. The share of students taking college-
whereas this was the modal level of parents’ edu- admissions tests increased by 8 pp. One measure,
cation for high-income children. whether a student had participated in college-
Mbekeani 71

Table 2. College-Going Activities: High School Classes of 1992 and 2004.

Mean
Measures 1992 2004

A. Financial preparation
Set up a college investment fund 0.09 0.17
Made investments in stocks or real estate 0.20 0.26
Amount set aside for teen’s education after high school (2021$) 10,051 10,701
B. Conversations
Mother wants teen to obtain BA or more 0.71 0.83
Often discussed going to college with parents 0.38 0.44
Often discussed college-exam preparation with parents 0.18 0.24
C. Test-taking
Teen has engaged/plans to engage in test preparation to prepare 0.56 0.58
for the SAT/ACT (non-school based)
Teen has taken the SAT or ACT 0.59 0.67

admissions test preparation, showed only a small that the gap change estimate for opening a college
change, from 56 percent for the 1992 cohort to investment fund is larger than would have
58 percent for the 2004 cohort. occurred without the advent of 529 Plans, and
thus the gap change estimate for investments in
Financial preparation. Across all income stocks and real estate (10 pp, p \ .001) is
groups, the 2004 cohort had higher levels of a more reasonable approximation of how families’
parental financial preparation for college atten- financial preparation changed over time. Finally,
dance than did the 1992 cohort, with one excep- the gap in the amount of money families had set
tion (Table 3, Panel A). Within-income group aside for their teen’s postsecondary education wid-
changes over time are calculated by subtracting ened by $2,231 (p \ .10); this change is only mar-
the value in 1992 from the value in 2004. The pro- ginally significant.
portion of high-income parents reporting they had
opened a college investment fund increased by 17 Conversations. Conversations about college-
pp (p \ .001). Among low-income families, the 1 going increased for students from low-, median-,
pp increase in the proportion who had opened and high-income families. Panel B in Table 3
a college investment fund was not significant. presents the predicted proportion by income. The
Trends in investments in stocks or real estate percentage of students who reported their
were directionally consistent, albeit somewhat mother/female guardian wanted them to obtain
smaller in magnitude for some income groups. a bachelor’s degree or more increased by 13, 13,
Low- and high-income families’ savings for col- and 11 percentage points for low-, median-, and
lege increased by $485 and $2,716, respectively, high-income students, respectively; these are sta-
but this change was only marginally statistically tistically significant for all groups. Discussing
significant for high-income families and statisti- college-exam preparation increased by 10 pp for
cally insignificant for low-income families. The low-income students and by 5 pp for high-income
average amount saved for median-income families students.
declined by $1,070 (p \ .05). Among high- and low-income students, parent
Income-based gaps in the domain of financial college-conversation gaps narrowed between
preparation for college-going widened consider- cohorts; however, these changes are not all signif-
ably. Figure 1 displays the estimated proportion icant. Figure 2 displays the fitted trend lines and
of parents who opened a college investment fund mean values for each income bin for the propor-
by income percentile rank for each cohort. The fit- tion of students who reported often discussing
ted function for the class of 2004 is steeper than college-exam preparation with their parents. The
for the class of 1992. The 90/10 gap widened by shallower slope of the dashed line shows the nar-
16 pp (p \ .001; Table 3, Panel A). It is possible rower gap for the class of 2004 (and the shift
72
Table 3. College-Going Activities at the 10th, 50th, and 90th Income Percentile Ranks and the 90/10 Gap: High School Classes of 1992 and 2004.

10th Percentile 50th Percentile 90th Percentile 90/10 Gap

Measures 1992 2004 1992 2004 1992 2004 1992 2004 D

A. Financial preparation
College investment fund 0.03 0.04 0.06 0.11*** 0.20 0.37*** 0.17 0.33 0.16***
Investments in stocks or real 0.04 0.059* 0.14 0.19*** 0.44 0.55*** 0.39 0.49 0.10***
estate
Amount saved (2021$) 2,306 2,791 4,207 3,137* 25,650 28,366* 23,344 25,575 2,231 1
B. Conversations
Earn BA or more 0.59 0.72*** 0.69 0.82*** 0.83 0.94*** 0.25 0.22 20.03
Discussed going to college 0.33 0.41*** 0.36 0.42*** 0.47 0.52*** 0.13 0.11 20.02
Discussed college-exam 0.14 0.24*** 0.16 0.22*** 0.24 0.29** 0.09 0.05 20.05*
preparation
C. Test-taking
SAT/ACT preparation 0.54 0.52 0.53 0.53 0.64 0.69** 0.10 0.17 0.07*
Taken SAT/ACT 0.39 0.43* 0.57 0.65*** 0.80 0.88*** 0.41 0.44 0.03
1
p \ .10. *p \ .05. **p \ .01. ***p \ .001.
Mbekeani 73

Figure 1. Estimated proportion of parents, by Figure 2. Estimated proportion of students, by

income, reporting setting up a college investment income, reporting discussing college-exam prepa-
fund: class of 1992 and class of 2004. ration with their parents: class of 1992 and class
of 2004.

upward on the y-axis demonstrates the overall

increase). The 90/10 gap in discussing college- (see Tables C.1 and C.2 in the online Supplemen-
exam preparation shrank by 5 pp and is statisti- tal Material). Across the other domains, results for
cally significant (p \ .05). Gaps in the other two subgroups are generally consistent in direction and
measures of conversations also narrowed, but magnitude with the main results.
these changes are not statistically significant
(Table 3, Panel B).
Explanations for Widening Income-
Test-taking. The measures for test-taking gener-
ally conform to the pattern of rising levels of Based College-Going Activity Gaps
college-going activities across the three income I found descriptive evidence that widening 90/10
groups, with two exceptions (Table 3, Panel C). income gaps were largely due to changes in the
Students from high-income families increased income distribution and changing associations. In
their participation in test preparation, but partici- other words, results of the counterfactual scenarios
pation rates for students from low- and median- suggest that high-income families’ financial prepa-
income families remained the same or declined, ration and test-preparation activities increased rela-
although not significantly. All three groups, how- tive to low-income families due to having more
ever, increased their rates of SAT/ACT taking. money and being more likely to put money toward
The 90/10 gap changes in test-taking measures these activities. Figure 4 displays the estimated 90/
are positive and consistent with widening gaps but 10 gap in 1992, the 90/10 gap under the three coun-
statistically significant for only one measure. The terfactual scenarios (changing income distribution,
90/10 gap in SAT/ACT preparation widened by changing family characteristics, and changing asso-
approximately 7 pp (p \ .05), from 10 to 17, ciations), and the estimated 90/10 gap in 2004.
between the two cohorts. Figure 3 displays the fit- For two measures—opening a college invest-
ted function for test preparation by income rank ment fund and college-admissions test prepara-
for each cohort. The 90/10 gap change is largely tion—the counterfactual 90/10 gaps from the
driven by the increase in test preparation among changing income and changing associations sce-
students from families with incomes above the narios best approximate the estimated gaps in
80th percentile rank in the 2004 cohort. The 2004. Under the changing income scenario (e.g.,
income-based gap in test taking also widened; only the distribution of incomes changed), the
however, this change is not significant.17 counterfactual gap estimate is approximately 20
There is evidence of some heterogeneity in gap pp, and under the changing associations scenario,
changes. Specifically, widening gaps in financial the gap estimate is 27 pp. Changing family charac-
preparation were concentrated among White stu- teristics do not account for the widening gaps in
dents and students with above median math scores these two measures. Somewhat in contrast, all
74 Sociology of Education 96(1)

Figure 3. Estimated proportion of students, by Figure 4. Estimated and counterfactual 90/10

income, reporting participating in college-admis- gaps under three scenarios.
sions test preparation: class of 1992 and class of
2004.
Second, I could not observe differences in the
quality of families’ engagement in these college-
three factors appear to have contributed to the going activities. This study focused on quantity of
growth in the gap in investments in stocks and engagement rather than quality. It may be the
real estate. In all three counterfactual scenarios, case, however, that fewer high-quality interactions
for this measure, the predicted gaps are larger have the same or even greater benefits than do fre-
than the estimated 1992 gap. quent low-quality interactions. A related concern is
Rising income inequality also accounts for there may be unobservable qualitative differences in
most of the increase in the 90/10 gap in the amount parental support by income. Parents who attended
families had saved for their adolescents’ postsec- college and have college graduates in their social
ondary education (Figure 5). Unlike opening a col- networks, as high-income individuals often do,
lege investment fund and test preparation, chang- may provide more specific advice and information
ing associations do not account for the gap when supporting their adolescent’s college-going.
growth. In fact, changing associations mitigate
the growth in the gap, given that the 90/10 gap
under the changing income scenario is predicted DISCUSSION
to be larger (approximately $27,000) than the esti-
mated gap in 2004. Researchers know little about whether the socio-
economic gap in parental support for children’s
college attendance has changed over time. I inves-
tigated this question by examining income-based
LIMITATIONS trends and gaps in parental financial preparation
Two limitations are of note. First, survey data can for postsecondary education expenses, engage-
suffer from social desirability bias, namely, that ment in conversations about the college-going pro-
respondents answer questions in ways they think cess, and support for college-admissions tests for
others will view favorably. It is unclear to what two cohorts of high school students.
degree adolescents’ responses would suffer from I found widening income gaps in financial
this; however, parents’ responses were likely preparation for postsecondary education expenses
affected. For the purposes of this study, social and college-admissions test preparation; however,
desirability bias is of concern if the magnitude I also found narrowing gaps in parent-adolescent
of this bias changed over time or if it changed dif- conversations. Results for financial preparation
ferentially by income. Between-cohort changes showed evidence of heterogeneity within income
may be overestimated if social desirability bias by race and achievement; future research should
increased over time. If this bias increased differen- investigate these differences further.
tially across income groups, then gap change esti- The two domains in which I found widening
mates may be biased. gaps were those that required resources. For test
Mbekeani 75

Figure 5. Estimated and counterfactual 90/10 gap in amount saved under three scenarios.

preparation and test-taking, parents or students of competition around highly selective college
must have access to information about available enrollment (Ramey and Ramey 2010; Reeves
courses, materials, or tutors and the monetary 2017) or their observation of the increased wage
resources to afford them (in addition to paying premium paid to college-educated workers (Autor
to take the tests). Similarly, financial preparation 2014).
for college requires that families have the resour- There are several potential explanations for
ces to set aside funds for future education and why gaps in parent-child conversations may have
the knowledge of relevant financial products. narrowed. Low-income parents’ behaviors may
The widening gaps in these domains suggest have changed because of the perception of the
that the mechanical effect of rising income increased importance of attending college for
inequality, or widening gaps in family composi- social mobility and financial security. Unlike
tion, may have played a role, yet that was only high-income parents, low-income parents would
in part what I found. Rising income inequality not necessarily have the financial means to invest
accounted for some of the growth in the gaps, as monetarily; however, they may shift their behav-
did changing associations. This suggests that in iors in other ways, in service of the same ends.
addition to a mechanical effect, there may be A related explanation is that the norm of intensive
‘‘contextual effects of income inequality’’ parenting may have become more pervasive across
(Schneider et al. 2018:480). Under this mecha- socioeconomic classes (Ishizuka 2019). The nar-
nism, the pathway between income inequality rowing of the gap in conversations may also be
and parental investments is not through income due to the penetration of the ‘‘College for All’’
alone. Rather, rising income inequality may have movement over this study’s time period (Carne-
differentially changed parents’ attitudes, beliefs, vale 2008; Goyette 2008; Rosenbaum 2001).
and preferences such that high-income parents Although not directly relevant to questions of
feel pressured to invest in their adolescents to gap changes between cohorts, two overall trends
ensure access to particular educational experien- are noteworthy. First, average levels of parental
ces and postsecondary institutions (Ramey and involvement increased across all three domains.
Ramey 2010; Reardon 2011a; Reeves 2017; These trends are striking given the demographic
Schneider et al. 2018). This perceived need to shifts over this time, which may have made it
invest may stem from parents’ heightened sense more challenging for some parents to marshal
76 Sociology of Education 96(1)

the resources to support their adolescents. Across The larger economic picture also presents con-
the income spectrum, parents were likely aware cerns for low-income students’ college attainment.
of the growing monetary and nonmonetary returns As postsecondary education costs have risen and
to a college degree (Autor 2014; Baum et al. income inequality has deepened, low-income fam-
2013); consequently, they provided increased sup- ilies increasingly face resource constraints when it
port to their adolescents to achieve higher comes to higher education. Even as these families
education. do more to support college attendance, the effect
However, even with these increases and shrink- of their efforts may be dampened given the larger
ing gaps in parent-child conversations, absolute structural inequalities. These structural factors
differences between adolescents from high- and could amplify the widening gaps in college-going
low-income families remained large. To some activities, leading to increased stratification in col-
extent, these students and families may be ‘‘run- lege attainment.
ning in place’’—with absolute increases but still This study contributes to research seeking to
relatively behind students from more advantaged understand whether and to what degree socioeco-
families (Bastedo and Jaquette 2011). Over twice nomic gaps in parental involvement aimed at facil-
as many high-income students (88 percent) as itating college attainment have changed and what
low-income students (43 percent) had taken a may explain these changes. Widening gaps, during
college-admissions test in 2004. Moreover, low- a period of rising income inequality and widening
income parents were not able to keep pace with disparities in other features of children’s lives, are
high-income parents in financial preparation for of concern because they have the potential to
college. High-income parents had saved approxi- exacerbate educational achievement and attain-
mately 10 times as much for their adolescent’s ment gaps by socioeconomic status. Understand-
education after high school as low-income parents ing the nature, causes, and consequences of
($27,000 vs. $2,900). changes in parental involvement by income is
One key question is how do these changes in important for identifying policy solutions aimed
parents’ and students’ engagement in college-going at reducing these gaps.
activities relate to income-based trends in college
enrollment and completion? As noted, studies
have documented growing gaps in enrollment ACKNOWLEDGMENTS
over a similar period to the present study, with I am grateful to Daniel Koretz, Sasha Killewald, Richard
some stabilization of college enrollment gaps in Murnane, Bridget Long, Anne Lamb, Marcus Waldman,
more recent years (Bailey and Dynarski 2011; Bas- Simone Fried, Odis Johnson, Jr., John Diamond, three
tedo and Jaquette 2011; Chetty et al. 2014; Ziol- anonymous reviewers, and participants at the 2019 East-
Guest and Lee 2016). These gap changes occurred ern Sociological Society and 2019 Society for Research
on Educational Effectiveness annual research conferen-
even as college enrollment rates increased, on aver-
ces for helpful comments and suggestions on earlier ver-
age. Thus, one possibility is that the relationships
sions of this article.
between these activities and college enrollment
are mechanical. This would be consistent with the
observed increase in overall rates of enrollment as FUNDING
well as larger increases for adolescents from
The author disclosed receipt of the following financial
higher-income families. A second possibility, also
support for the research, authorship, and/or publication
consistent with observed trends, is that a subset of of this article: The research reported here was supported
activities had a greater effect, or a greater effect by the Harvard Graduate School of Education’s Dean’s
for particular groups, leading gaps to widen. Hold- Summer Fellowship.
ing constant enrollment, there are substantial differ-
ences in the colleges attended by low- and high-
income adolescents (Bastedo and Jaquette 2011). RESEARCH ETHICS
Adolescents from low-income families are much The National Education Longitudinal Study of 1988 and
more likely to attend two-year colleges or less the Education Longitudinal Study of 2002 are anony-
selective colleges, with lower average benefits in mized data sets publicly available from the National
terms of four-year degree completion and labor Center for Education Statistics. The views expressed in
market earnings. this article are those of the author and do not reflect those
Mbekeani 77

of the Harvard Graduate School of Education or Brown 8. In line with Grodsky (2010), I created a measure of
University. All errors, omissions, and conclusions are ‘‘shadow’’ test preparation that included only pri-
my own. vate tutoring or private courses. The trend and gap
change results (available upon request) were
like those presented for my more expansive test-
ORCID ID preparation measure. The measure itself, however,
did not conform to expectation due to a nonmono-
Preeya Pandya Mbekeani https://orcid.org/0000- tonic relationship with family income rank for
0003-1465-0226 below-median-income families.
9. The exception being that Colorado and Illinois
implemented statewide ACT testing beginning
SUPPLEMENTAL MATERIAL with the high school class of 2002. By way of com-
Supplemental material for this article is available online. parison, in the 2016–17 school year, 25 states
required the SAT or ACT (Gewertz 2017).
10. Reardon (2011b) found that gap estimates do not
NOTES vary systematically with the number of categories
in which income is reported.
1. Ideally, I would have also used the most recent 11. In cases where the lower and upper bounds identify
NCES survey of high school students, the High the ranks of interest (10/50/90, e.g., if the lower
School Longitudinal Study of 2009 (HSLS:09); bound of the second to last category is 0.9), they
however, only one measure was directly comparable can be matched to the income values of the catego-
to the measures used in this study. Four measures ries. However, in most cases, the lower and upper
used in this study were not in HSLS:09. Two meas- bound ranks were not those of interest. To identify
ures were either worded differently or had different incomes at other ranks, one must assume a distribu-
response options, raising questions about compara- tion of income within a category and interpolate to
bility. Finally, the timing of when students were sur- identify the income associated with the rank of
veyed (9th grade and 11th grade rather than 10th interest. This would likely introduce measurement
grade and 12th grade) confounded changes in trends error.
with the timing of the survey, a particular concern 12. The precision of the match of the percentile ranks
for questions related to test-taking. was high (at three significant digits), alleviating
2. A small number of students from each cohort were concerns of measurement error due to imprecise
retained or graduated early. matches.
3. I constructed and assigned family income and parent 13. Relevant code is available from the author upon
education ranks prior to dropping students with request.
missing parent questionnaire data. 14. I thank an anonymous reviewer for suggesting these
4. Sample sizes were rounded to the nearest 10 per analyses.
NCES reporting requirements. 15. This was f2f1pnwt for the class of 1992 and f1pnlwt
5. Results using unimputed data were directionally for the class of 2004.
consistent and quite similar in magnitude. In some 16. I estimated demographic characteristics at the 10th,
cases, results using unimputed data showed larger 50th, and 90th percentile ranks of the income distri-
or statistically significant 90/10 gap changes than bution using a multinomial logit model. The qua-
the results from imputed data, which were more dratic and cubic terms were not significant, so I
conservative estimates of gap changes. opted for a parsimonious, linear function of income
6. I followed the practice used in studies of wage percentile rank ðQ Þ. Estimates from the three spec-
inequality of adjusting top-coded categories. Stan- ifications are similar.
dard practice in the cited studies is to replace the 17. In the related literature (e.g., Bassok et al. 2016;
top category with 1.4 times the top-code value. I Kalil et al. 2016; Reardon 2011a), changes in
also assessed alternative procedures, including income-based gaps are compared to changes in
assigning the lower bound value, using alternative parent-education rank gaps. In supplementary analy-
multipliers (1.2 and 1.6), and applying the 1992 ses, presented in Table D.1 in the online Supplemen-
adjusted top-code in both data sets. Results using tal Material, I found that 90/10 parent-education rank
these alternative approaches, available upon gaps in the domain of financial preparation also wid-
request, are directionally consistent and similar in ened but by less. Gaps between highly educated and
magnitude to those using 1.4. less educated parents shrank in the domain of conver-
7. Understanding parental rates of saving across stu- sations. For test-taking, 90/10 gap changes are incon-
dents’ age/grade is an important area for future sistent, with the gap in test preparation widening and
research. the gap in test-taking narrowing.
78 Sociology of Education 96(1)

REFERENCES Cooper, Marianne, and Allison J. Pugh. 2020. ‘‘Families

across the Income Spectrum: A Decade in Review.’’
Allison, Paul D. 2001. Missing Data. Thousand Oaks, Journal of Marriage and Family 82(1):272–99.
CA: SAGE Publications. Duncan, Greg J., and Richard J. Murnane, ed. 2011.
Autor, David H. 2014. ‘‘Skills, Education, and the Rise Whither Opportunity? Rising Inequality, Schools,
of Earnings Inequality among the ‘Other 99 Per- and Children’s Life Chances. New York: Russell
cent.’’’ Science 344(6186):843–51. Sage Foundation.
Bailey, Martha J., and Susan M. Dynarski. 2011. Flaster, Allyson. 2018. ‘‘Kids, College, and Capital:
‘‘Inequality in Postsecondary Education.’’ Pp. Parental Financial Support and College Choice.’’
117–31 in Whither Opportunity? Rising Inequality, Research in Higher Education 59(8):979–1020.
Schools, and Children’s Life Chances, edited by G. Flood, Sarah, Miriam King, Renae Rodgers, Steven Rug-
J. Duncan and R. J. Murnane. New York: Russell gles, and J. Robert Warren. 2020. ‘‘Integrated Public
Sage Foundation. Use Microdata Series, Current Population Survey:
Bandelj, Nina, and Angelina Grigoryeva. 2021. ‘‘Invest- Version 9.0 [Dataset].’’ doi:10.18128/D030.V9.0
ment, Saving, and Borrowing for Children: Trends Gewertz, Catherine. 2017. ‘‘Amid Subtly Shifting Test
by Wealth, Race, and Ethnicity, 1998–2016.’’ RSF: Landscape, Support Slips for Common-Core Tests.’’
The Russell Sage Foundation Journal of the Social Education Week, February 15, 1–8.
Sciences 7(3):50–77. Goyette, Kimberly A. 2008. ‘‘College for Some to Col-
Bassok, Daphna, Jenna E. Finch, RaeHyuck Lee, Sean F. lege for All: Social Background, Occupational
Reardon, and Jane Waldfogel. 2016. ‘‘Socioeconomic Expectations, and Educational Expectations over
Gaps in Early Childhood Experiences: 1998 to 2010.’’ Time.’’ Social Science Research 37(2):461–84.
AERA Open 2(3). doi:10.1177/23328584166539. Grodsky, Eric. 2010. ‘‘Learning in the Shadows and in
Bastedo, Michael N., and Ozan Jaquette. 2011. ‘‘Run- the Light of Day: A Commentary on Shadow Educa-
ning in Place: Low-Income Students and the Dynam- tion, American Style: Test Preparation, the SAT and
ics of Higher Education Stratification.’’ Educational College Enrollment.’’ Social Forces 89(2):475–81.
Evaluation and Policy Analysis 33(3):318–39. Guryan, Jonathan, Erik Hurst, and Melissa Kearney.
Baum, Sandy, Charles Kurose, and Jennifer Ma. 2013. 2008. ‘‘Parental Education and Parental Time with
How College Shapes Lives: Understanding the Children.’’ Journal of Economic Perspectives
Issues. New York: College Board. 22(3):23–46.
Becker, Gary Stanley. 1993. Human Capital: A Theoret- Hastings, Orestes P., and Daniel Schneider. 2021. ‘‘Fam-
ical and Empirical Analysis with Special Reference ily Structure and Inequalities in Parents’ Financial
to Education. 3rd ed. Chicago, IL: University of Chi- Investments in Children.’’ Journal of Marriage and
cago Press. Family 83(3):717–36.
Buchmann, Claudia, Dennis J. Condron, and Vincent J. Hillman, Nicholas, Melanie J. Gast, and Casey George-
Roscigno. 2010. ‘‘Shadow Education, American Jackson. 2015. ‘‘When to Begin? Socioeconomic
Style: Test Preparation, the SAT and College Enroll- and Racial/Ethnic Differences in Financial Plan-
ment.’’ Social Forces 89(2):435–62. ning.’’ Teachers College Record 117(8):28.
Card, David, and John E. DiNardo. 2002. ‘‘Skill-Biased von Hippel, Paul T. 2007. ‘‘Regression with Missing Ys:
Technological Change and Rising Wage Inequality: An Improved Strategy for Analyzing Multiply
Some Problems and Puzzles.’’ Journal of Labor Eco- Imputed Data.’’ Sociological Methodology 37(1):
nomics 20(4):733–83. 83–117.
Carnevale, Anthony P. 2008. ‘‘College For All?’’ Ishizuka, Patrick. 2019. ‘‘Social Class, Gender, and Con-
Change: The Magazine of Higher Learning 40(1): temporary Parenting Standards in the United States:
22–31. Evidence from a National Survey Experiment.’’
Chetty, Raj, John N. Friedman, Emmanuel Saez, Nicho- Social Forces 98(1):31–58.
las Turner, and Danny Yagan. 2017. ‘‘Mobility Kalil, Ariel, Rebecca Ryan, and Michael Corey. 2012.
Report Cards: The Role of Colleges in Intergenera- ‘‘Diverging Destinies: Maternal Education and the
tional Mobility.’’ National Bureau of Economic Developmental Gradient in Time with Children.’’
Research, Working Paper w23618. Demography 49(4):1361–83.
Chetty, Raj, Nathaniel Hendren, Patrick Kline, Emma- Kalil, Ariel, Kathleen M. Ziol-Guest, Rebecca M. Ryan,
nuel Saez, and Nicholas Turner. 2014. ‘‘Is the United and Anna J. Markowitz. 2016. ‘‘Changes in Income-
States Still a Land of Opportunity? Recent Trends in Based Gaps in Parent Activities with Young Children
Intergenerational Mobility.’’ American Economic from 1988 to 2012.’’ AERA Open 2(3). doi:10.1177/
Review: Papers & Proceedings 104(5):141–47. 2332858416653732.
Conwell, Jordan A. 2021. ‘‘Diverging Disparities: Race, Kornrich, Sabino. 2016. ‘‘Inequalities in Parental Spend-
Parental Income, and Children’s Math Scores, ing on Young Children: 1972 to 2010.’’ AERA Open
1960-2009.’’ Sociology of Education 94(2):124–42. 2(2). doi:10.1177/2332858416644180
Mbekeani 79

Kornrich, Sabino, and Frank Furstenberg. 2013. ‘‘Invest- New Evidence and Possible Explanations’’ Online
ing in Children: Changes in Parental Spending on Appendix to Chapter 5 in Whither Opportunity? Ris-
Children, 1972–2007.’’ Demography 50(1):1–23. ing Inequality, Schools, and Children’s Life Chan-
Long, Mark C. 2008. ‘‘College Quality and Early Adult ces, edited by G. J. Duncan and R. J. Murnane.
Outcomes.’’ Economics of Education Review 27(5): New York: Russell Sage Foundation.
588–602. Reeves, Richard V. 2017. Dream Hoarders: How the
Lucas, Samuel R. 2001. ‘‘Effectively Maintained American Upper Middle Class Is Leaving Everyone
Inequality: Education Transitions, Track Mobility, Else in the Dust, Why That Is a Problem, and What
and Social Background Effects.’’ American Journal to Do about It. Washington, DC: Brookings Institu-
of Sociology 106(6):1642–90. tion Press.
Ma, Jennifer, and Matea Pender. 2021. Trends in College Rosenbaum, James. 2001. Beyond College for All:
Pricing and Student Aid 2021. New York: College Career Paths for the Forgotten Half. New York:
Board. Russell Sage Foundation.
McLanahan, Sara. 2004. ‘‘Diverging Destinies: How Schneider, Daniel, Orestes P. Hastings, and Joe LaB-
Children Are Faring under the Second Demographic riola. 2018. ‘‘Income Inequality and Class Divides
Transition.’’ Demography 41(4):607–27. in Parental Investments.’’ American Sociological
Mouw, Ted, and Arne L. Kalleberg. 2010. ‘‘Occupations Review 83(3):475–507.
and the Structure of Wage Inequality in the United Weeden, Kim A., Young-Mi Kim, Matthew Di Carlo,
States, 1980s to 2000s.’’ American Sociological and David B. Grusky. 2007. ‘‘Social Class and Earn-
Review 75(3):402–31. ings Inequality.’’ American Behavioral Scientist
Piketty, Thomas, and Emmanuel Saez. 2003. ‘‘Income 50(5):702–36.
Inequality in the United States, 1913-1998.’’ The Western, Bruce, Christine Percheski, and Deirdre
Quarterly Journal of Economics 118(1):1–39. Bloome. 2008. ‘‘Inequality among American Fami-
Piketty, Thomas, and Emmanuel Saez. 2020. ‘‘Tables lies with Children, 1975 to 2005.’’ American Socio-
and Figures Updated to 2018.’’ http://elsa.berkeley logical Review 73(6):903–20.
.edu/~saez/. White, Ian R., Patrick Royston, and Angela M. Wood.
Quadlin, Natasha, and Jordan A. Conwell. 2021. ‘‘Race, 2011. ‘‘Multiple Imputation Using Chained Equa-
Gender, and Parental College Savings: Assessing tions: Issues and Guidance for Practice.’’ Statistics
Economic and Academic Factors.’’ Sociology of in Medicine 30:377–99.
Education 94(1):20–42. Ziol-Guest, Kathleen M., and Kenneth T. H. Lee. 2016.
Quadlin, Natasha, and Brian Powell. 2022. Who Should ‘‘Parent Income–Based Gaps in Schooling: Cross-
Pay? Higher Education, Responsbility, and the Pub- Cohort Trends in the NLSYs and the PSID.’’
lic. New York: Russell Sage Foundation. AERA Open 2(2). doi:10.1177/2332858416645834
Ramey, Garey, and Valerie A. Ramey. 2010. ‘‘The Rug
Rat Race.’’ Brookings Papers on Economic Activity
2010(1):129–76. Author Biography
Reardon, Sean F. 2011a. ‘‘The Widening Academic
Achievement Gap between the Rich and the Poor: Preeya Pandya Mbekeani is a postdoctoral research
New Evidence and Possible Explanations.’’ Pp. associate at the Annenberg Institute for School Reform
91–115 in Whither Opportunity? Rising Inequality at Brown University. Her research examines the transi-
Schools, and Children’s Life Chances, edited by G.
tion from high school to college, with a focus on inequal-
J. Duncan and R. J. Murnane. New York: Russell ity as both cause and consequence in this transition. Her
Sage Foundation. work also evaluates the impact of policies aimed at
Reardon, Sean F. 2011b. ‘‘The Widening Academic
improving postsecondary access and completion.
Achievement Gap between the Rich and the Poor: