“You pay your share, we’ll pay our share”: The college cost burden and the role of race,...

Economics of Education Review 33 (2013) 134–153

Contents lists available at SciVerse ScienceDirect

Economics of Education Review

journal homepage: www.e lsev ier .com/ locate /econedurev

‘‘You pay your share, we’ll pay our share’’: The college cost burdenand the role of race, income, and college assets

William Elliott a,*, Terri Friedline b,1

a University of Kansas, School of Social Welfare, 1545 Lilac Lane, 309 Twente Hall, Lawrence, KS 66044, United Statesb University of Kansas, School of Social Welfare, 1545 Lilac Lane, 307 Twente Hall, Lawrence, KS 66044, United States

A R T I C L E I N F O

Article history:

Received 4 May 2012

Received in revised form 14 September 2012

Accepted 3 October 2012

JEL classification:

125

Keywords:

Assets

College savings

College finances

College costs

Student debt

Student loans

A B S T R A C T

Changes in financial aid policies raise questions about students being asked to pay too

much for college and whether parents’ college savings for their children helps reduce the

burden on students to pay for college. Using trivariate probit analysis with predicted

probabilities, in this exploratory study we find recent changes in the financial aid system

place a higher responsibility on African American, Latino/Hispanic, and moderate-income

students to pay for college themselves. We also find when parents open a savings account,

start a state-sponsored savings plan, or open a college investment fund students are less

likely to pay for college with student contributions. Therefore, we suggest in addition to

grants and scholarships, policies that encourage accumulation of savings for college

among minority and lower income families may help reduce the college cost burden they

experience.

� 2012 Elsevier Ltd. All rights reserved.

1. Introduction

Since the late 1970s the federal government hasincreasingly attempted to solve the equal access problemcaused by high college costs through the adoption of policiesthat make college loans accessible to more students. It haslargely done this through programs such as federal ParentPLUS Loans and Stafford subsidized and unsubsidized loanprograms. For example, the Middle Income Student Assis-tance Act (1978) brought college loans to the middle class byremoving the income limit for participation in federal aidprograms (Hansen, 1983). The 1992 amendments to theHigher Education Act made unsubsidized loans available,and the Omnibus Budget Reconciliation Act (1993) includedprovisions for the Federal Direct Loan Program. More

* Corresponding author. Tel.: +1 785 864 2283; fax: +1 785 864 5277.

E-mail addresses: [email protected] (W. Elliott), [email protected]

(T. Friedline).1 Tel.: +1 785 864 4720; fax: +1 785 864 5277.

0272-7757/$ – see front matter � 2012 Elsevier Ltd. All rights reserved.

http://dx.doi.org/10.1016/j.econedurev.2012.10.001

recently, Congress raised the ceiling on the amount ofindividual federal Stafford loans students can borrowthrough the Ensuring Continued Access to Student LoansAct (2008). The Health Care and Education ReconciliationAct (2010) routed all federal loans through the Direct Loanprogram, making it easier for students and parents toborrow directly from the U.S. Department of Education.These policies mark a shift away from or diminishing of therole that society plays in financing college (largely throughscholarship/grants) toward a greater obligation beingplaced on students and their families to pay for college.

Another important shift in financial aid policy is theshift away from need-based aid toward merit-based aid(Baum & Schwartz, 1988; Woo & Choy, 2011). Need-basedaid is determined solely on the assets and income (i.e.,financial need) of the prospective student and his or herfamily. Factors such as test scores have no bearing on theaid decision. In the case of merit-based aid, of whichscholarships are the most common form, a student withlittle financial need (i.e., higher assets and income) is justas entitled to aid as are students with high levels of


mailto:[email protected]

mailto:[email protected]


W. Elliott, T. Friedline / Economics of Education Review 33 (2013) 134–153 135

financial need (i.e., lower assets and income). Test scoresare often the key factor for determining eligibility. Wooand Choy (2011) find that the proportion of under-graduates receiving merit aid rose from 6% in 1995/96to 14% in 2007/08. Further, research suggests that merit-based aid is awarded disproportionately to students fromhigher income families (Woo & Choy, 2011) and that it hasdone little to improve college enrollment rates among low-income and minority students (Marin, 2002).

The last significant shift in financial aid that we willdiscuss is the shift from spending programs to tax subsides.With the exception of increases in the maximum Pell Grantand loan subsidies, most new federal resources have beenprovided through the tax code. Middle- and upper-incomestudents benefit most from these changes because theyhave a higher marginal tax rate than lower incomefamilies; therefore, they receive larger benefits from suchprograms (Maag & Fitzpatrick, 2004). Examples of theseprograms are college investment funds such as theCoverdell Education Savings Accounts, State 529s, andEducation Savings Bonds. These programs provide anincentive for families to begin saving for college costs priorto students enrolling in college. Money invested in thesetypes of college savings vehicles grow tax-free andwithdraws made from them to pay for college is alsotax-free. While there is some evidence to suggest assetssuch as net worth and savings accounts do have a positiverelationship with both college enrollment and graduation(for a review of this research see, Elliott, Destin, & Friedline,2011), to date, there is little information about whether taxbased college asset vehicles make college more affordable.

These policy trends along with rising college costs raisethe question, ‘‘Are students as likely as or more likely thansociety to bear the responsibility of paying for college?’’ Inthis exploratory study we investigate the probability thatstudents pay for college with student, family, and/orsocietal contributions (i.e., grants/scholarships). We alsoexamine whether differences exist by race and income.Findings may have implications for whether some studentsare disproportionately burdened by the shift towardgreater contributions by students and families. Finally,we focus on how different types of college assets affectwhether students are more likely to report paying forcollege with student, family, or societal contributions.

2. Review of research

2.1. Student contributions

As discussed in the introduction, increasingly studentloans are the primary way students contribute to collegecosts. Students must take money from future savings or jobearnings to pay the balance of their loans. As such, loansrepresent a way students make financial contributions totheir education. The College Board (2009) reports that in1989/90, 27% of all undergraduates had taken out federalStafford loans at some point during their enrollment inpostsecondary education, while in 2007/08, this proportionwas 46%. However, research suggests that student loansmay not improve attendance and completion rates, at leastafter a certain point (Dynarski, 1994, 2003; Kim, 2007;

Perna, 2008; Volkwein & Szelest, 1995; Volkwein, Szelest,Cabrera, & Napierski-Prancl, 1998). For example, among3251 first-year undergraduate students who borrowed topay for college, Kim (2007) finds that every additional $1000increase from the mean loan amount for students from low-income households resulted in a 60% decrease in theprobability of graduating from college. Moreover, accordingto Dynarski (1994), 10% of students at four-year colleges anduniversities defaulted on their student loans and were morelikely to default when they had low earnings after college ordid not complete college. This is confirmed in a more recentstudy of college graduates from the 1990 NationalGraduates Survey analyzed with probit models andpredicted probabilities (Schwartz & Finnie, 2002). Whilethe percentage of students who reported substantialproblems with repaying their loans was small, those withlower current and lifetime earnings reported the greatestdifficulty and were perhaps overburdened (Schwartz &Finnie, 2002). Given this, having more students pay forcollege through loans may not be in the best interest ofstudents or society. At the very least, there may be limits tothe utility of student loans.

Another important way that students contribute totheir education is by working. Research suggests thatstudents who work in federal work-study jobs have highercollege completion rates than when they do not (DesJardins,Ahlburg, & McCall, 2002; Stampen & Cabrera, 1988). Forinstance, in a study of 20% of the University of Wisconsin’sfirst-year students from 1979, those who participated infederal work-study (either by itself or in combination withother grants, scholarships, and loans) had lower dropoutrates than students who paid for college without federalwork-study (Stampen & Cabrera, 1988). College studentsmay benefit from working in several ways, includingacquiring career-related knowledge (Perna, Cooper, & Li,2007). However, very few students work in the federalwork-study program, which limits its ability to be aneffective tool for helping to pay for college for most students.

2.2. Family contributions

Expected family contributions and parent loans areother ways students pay for college costs. Parent PLUSLoans are a common source of family contributions andtheir use has almost tripled in the last decade (Baum &Payea, 2011). In contrast to student loans that are deferreduntil students are no longer enrolled full-time, parentsoften begin repayment on the loans immediately. ParentPLUS Loans also require credit checks to determineeligibility, making them less available to families withpoor credit ratings. During the 2010/11 academic year, 35%of parents whose children attended public, four-yearcolleges and universities paid for college costs in partthrough Parent PLUS Loans, accounting for 9% of all federaland non-federal loans borrowed (Baum & Payea, 2011).The average Parent PLUS Loan amounted to approximately$12,000 (Baum & Payea, 2011). Parents may also takeeducational loans from private, non-federal institutions,such as from local banks and credit unions. These types ofcontributions account for 7% of all federal and non-federalloans borrowed (Baum & Payea, 2011).

W. Elliott, T. Friedline / Economics of Education Review 33 (2013) 134–153136

Research suggests that students’ college attendanceand graduation rates may be positively associated withfamily contributions (Bettinger, 2004; Charles, Roscigno, &Torres, 2007; Elliott et al., 2011; Hanushek, Leung, &Yilmaz, 2004; Kim, 2007). Kim (2007), for example, findsthat students who received financial contributions fromtheir parents during their first year of college graduated ata rate 9% higher than students who did not receivecontributions from their parents.

2.3. Societal contributions

In this study, societal contributions refer to grants andscholarships such as Pell Grants that do not require futurerepayment by students. Federal and private grants andscholarships used to pay for college comprise about 53% ofstudents’ total financial aid package during the 2010/11academic year, with federal grants contributing to 27% offinancial aid packages (Baum & Payea, 2011). Bettinger(2004) conducts a study examining the relationshipbetween Pell Grants and college completion using studentdata gathered by the Ohio Board of Regents. He finds thatstudents who received Pell Grants were less likely todropout of college and that every $1000 increase in theamount of Pell Grant awards was associated with a 10%decrease in the likelihood of attrition (Bettinger, 2004).Moreover, a study of students receiving Pell Grantsbetween 1999 and 2007 (526,488 students at 70 publiccolleges and flagship universities) finds that Pell Grantreceipt influenced students’ enrollment and graduationbehavior, although these grants did not meet the need ofall students (Waddell & Singell, 2011).

While grants can be very helpful, they make up onlyabout half of all undergraduate student aid (Baum & Payea,2011). Moreover, grants are increasingly offered based onmerit as opposed to financial need (Heller, 2002). Unlikeneed-based aid which is determined based on the student’sand their family’s ability to pay, merit-based aid is basedon the student’s academic performance so that colleges canattract the students they most desire. Critics argue that thisshift is likely to result in financial aid resources beingfunneled away from those most in need, reducingeducational opportunities for low-income students (Hell-er, 2004). Taken together, these trends mean students andtheir parents – particularly those from lower incomehouseholds – cannot rely solely on grant aid and must relymore frequently on loans. As loans have become moreaccessible, the proportion of federal grants to federal loansthat a particular student receives has plummeted. Forexample, the proportion of federal grants to federal loansin 1976 was about even (Archibald, 2002). However, by1985 the ratio had shifted to 27% grants and 70% loans, andby 1998 to 17% grants and 82% loans (Archibald, 2002; alsosee Heller & Rogers, 2006 for more information on how thisshift has taken place).

2.4. College assets

Families are increasingly incentivized, largely throughthe tax code, to start accumulating assets specifically fortheir children’s educational costs prior to them reaching

college age. Research on the relationship between assetsand college outcomes suggests assets provide studentswith three things, each of which may improve collegeattendance and completion rates (Elliott et al., 2011). First,assets help students develop educational expectations thatinclude college (Elliott, 2009; Elliott & Beverly, 2011).Second, assets offer resources that can be used to getinformation about college costs and financial aid. Studentswhose parents have greater assets may also have greaterknowledge about financial aid, grants, and scholarships –or at least may know where to go or with whom to talk inorder to get information about financial aid. Charles andcolleagues (2007), for instance, find that students havegreater knowledge about grants and loans when theirparents are saving money for college. Third, assets mayprovide students with the financial resources needed topay for college (Charles et al., 2007; Huang, Beverly, Clancy,Lassar, & Sherraden, 2011; O’Connor, Hammack, & Scott,2010). Research consistently finds that assets are signifi-cantly related to college attendance and graduation (Elliottet al., 2011), presumably because assets provide studentswith greater financial resources that can be leveraged tocover unmet need and to pay for college costs up front.While it is often assumed that the primary benefit ofowning assets is their ability to help pay for college, there islittle research that tests whether assets are predictive ofhow children pay for college.

In sum, little is known about factors that predict whichtypes of contributions students are most likely to use topay for college: (1) student contributions, (2) familycontributions, or (3) societal contributions. The researchon this topic is typically descriptive in nature and focuseson differences by race and income level. The focus on raceand income is because it is well recognized that collegeenrollment gaps exist by both race and income level (e.g.,Trusty, 2000). Further, research on financial aid indicatesthat minorities and low-income students are moresusceptible to changes in financial aid than their counter-parts (e.g., Kim, 2007). Therefore, we focus on findings byrace and income when possible.

Learning more about factors that predict whichcontributions students use will help answer questionsregarding whether or not, for example, ‘‘Are minority andlower income students as likely as or more likely to pay forcollege with student contributions than White and higherincome students?’’ If they are, it suggests that minority andlower income students might be overly burdened bypolicies that emphasize student contributions. It also mayinform us as to where interventions should be targeted. Forexample, if low-income students are more likely to usestudent and societal contributions to pay for college,maybe strategies need to be designed to increase parents’capacity to contribute. In the following section, we reviewsome of the ways students pay for college by race andincome.

2.5. Differences by race

Using data from the 1995/96 National PostsecondaryStudent Aid Study (NPSAS) conducted by the U.S. Depart-ment of Education, King (1999) finds that students take out


loans disproportionately by race. Fifty-four percent ofAfrican American students at four-year colleges anduniversities rely on loans to pay for college comparedwith 36% of White students, 30% of Asian students, and 35%of Latino/Hispanic students (King, 1999). In part, studentsfrom racial/ethnic minority groups may rely more on loansbecause they might receive fewer family contributions topay for college. Approximately 44% of White students and37% of Asian students receive an expected familycontribution of $12,500 or more; however, far fewerAfrican American and Latino/Hispanic students receive anexpected family contribution of the same amount – 20%and 26%, respectively (King, 1999). Moreover, almost onethird of African American and Latino/Hispanic students donot expect any family contributions (King, 1999). Ifdistributed as intended, grants and scholarships shouldmake up for disproportionate contributions by parents.Among students at public four-year colleges and universi-ties, 39% of White, 44% of Asian, 62% of African American,and 56% of Latino/Hispanic students receive grants (King,1999). These percentages are confirmed by reports usingmore recent data (Santiago & Cunningham, 2005). AfricanAmerican and Latino/Hispanic students are the most likelyof all racial groups to receive grants in 2003/04 (Santiago &Cunningham, 2005).

2.6. Differences by income level

Students from low- and moderate-income householdsmay rely on student contributions like work-study andloans or societal contributions more often than familycontributions when compared to their middle- and high-income counterparts (Berkner, Wei, He, Cominole, & Siegel,2005; Choy & Berker, 2003; Choy & Bobbitt, 2000).According to data from full-time dependent students fromthe 1999/00 NPSAS, 51–59% of students from low- andmoderate-income households pay with loans comparedwith 27–49% from middle- and high-income households(Choy & Berker, 2003). Most students from low- andmoderate-income households with loans have subsidizedFederal Stafford (48–56%) and Perkins loans (10–17%),while fewer percentages rely on Parent Plus Loans (2–7%;Choy & Berker, 2003). Compared to students from low- andmoderate-income households, fewer percentages of stu-dents from middle- and high-income households pay withsubsidized Federal Stafford (26–49%) and Perkins loans(<1% to 6%) and greater percentages pay with Parent PlusLoans (5–10%; Choy & Berker, 2003). Seventy to 72% ofstudents from low- and moderate-income householdsreceive grants at public, four-year colleges and universities(Choy & Berker, 2003). Comparatively, approximately 28% offull-time students from high-income households at public,four-year colleges and universities pay with grants andscholarships (Choy & Berker, 2003; Presley & Clery, 2001).

3. Research questions

In this study, we test the following research questions:(1) Are students as likely as or more likely than society tobear the responsibility of paying for college? (2) Areminority and low-income students as likely as or more

likely to be asked to carry the responsibility of paying forcollege than White and higher income students? (3) Doassets accumulated for college increase or reduce thelikelihood that students report paying for college withstudent, parent, and/or societal contributions?

4. Methods

4.1. Dataset

This study used longitudinal data from the EducationalLongitudinal Survey (ELS): 2002, a publically availabledataset made available by the National Center forEducation Statistics (NCES). The ELS: 2002 began in2002 when students were in 10th grade. Follow-up wavestook place in 2004 and 2006. Its purpose was to followstudents as they progressed through high school andtransitioned to postsecondary education or the labormarket, making it an ideal dataset to test whether earlyexperiences or resources predicted students’ later out-comes. The ELS: 2002 aimed to present a holistic picture ofstudent achievement by gathering information frommultiple sources. Students, their parents, teachers, librar-ians, and principals provided information regardingstudents’ average grades, math achievement, and educa-tional expectations, school resources and curriculum,teacher experience, student and parent work/employment,and student post-high school enrollment in college. Thedependent variables in this study came from the 2006wave and independent variables came from the 2002 and2004 waves.

4.2. Study sample

The final sample was restricted to students in the 10thgrade cohort during the 2001/02 academic year, studentswho were both in the 2002/2006 ELS samples (i.e., follow-up questionnaire status), high school graduates, studentwho applied for financial aid, and students who everattended a two-year or four-year college. In addition,American Indian and biracial students were eliminatedfrom the analysis due to small sample sizes. Further, a fewschools contained less than five students. These schoolswere removed from the analysis. After these restrictionswere applied, the full sample included 7366 students.Applying the panel weight resulted in a weighted sampleof approximately 1,652,963 students. Two subsampleswere drawn from the full sample. One is restricted tostudents who attended two-year colleges (weightedn = 505,954; non-weighted n = 2003) and the other isrestricted to students who attended four-year colleges(weighted n = 1,147,009; non-weighted n = 5363). SeeTable 1 for sample characteristics.

4.3. Student variables

All control variables with exception of dependentstatus, which was measured in 2006, were measured inthe 2002 or the 2004 wave of the ELS. Student gender was adichotomous variable. Number of siblings was a continu-ous variable. To determine students’ dependent status the

Table 1

Study characteristics by type of college.

Covariates Full percent Two-year percent Four-year percent

Student

Dependent student 40 66 29

White 64 53 67

Asian 05 20 06

Latino/Hispanic 17 22 14

African American 14 15 13

Male 44 43 44

Student attended two-year college 31 – –

Student attended four-year college 69 – –

Parent/household variables

Head has high school diploma or less 21 31 16

Head has some college 34 40 31

Head has college degree or higher 45 29 53

Low-income ($0–20,000) 11 18 09

Moderate-income ($20,001–50,000) 37 45 33

Middle-income ($50,001–100,000) 39 31 42

High-income ($100,001 or higher) 14 06 17

School variables

Private school (by 10th grade) 09 13 11

Student expects to graduate college 94 87 97

Parent expects student to graduate college 86 74 93

Low college costs very important 36 47 31

Financial aid very important 64 72 60

Asset variables

Plan to remortgage home 09 07 10

Start a savings account 41 33 44

Have student put aside earnings 23 18 25

Start state-sponsored savings 07 06 07

College investment fund 18 10 22

Invest in real estate/stocks 29 19 33

Buy U.S. savings bonds 22 16 25

Continuous variables

M SD M SD M SD

Student and parent/household variables

Number of siblings 1.42 1.104 1.50 1.200 1.40 1.086

GPA 4.52 1.285 3.85 1.260 4.84 1.14

School variables

School climate .318 .849 .367 .982 .288 .765

Number of guidance counselors 4.32 2.747 3.89 2.761 4.37 2.741

Free/reduced lunch 3.26 1.821 3.66 1.856 3.04 1.800

Source: Weighted data from the ELS: 2002.

Notes: Data imputed using the Expectation Maximization (EM) algorithm. SD = standard deviation. Full (weighted n = 1,652,963; non-weighted n = 7366);

two-year (weighted n = 505,954; non-weighted n = 2003); four-year (weighted n = 1,147,009; non-weighted n = 5363).


ELS: 2002 asked students whether they lived with theirparents or not in 2006. It is a dichotomous variable (yes/no). Student race/ethnicity included seven categories.American Indians (less than 1%) and more than one race(about 4%) were dropped from the analysis due to smallsample sizes. Hispanic and Latino were combined. Therewere four categories in the final analysis (White = 0;Asian = 1; Latino/Hispanic = 2; and African American = 3).Type of college was drawn from the highest level ofeducation ever attended variable in the ELS: 2002. For thepurposes of this study, a dichotomous variable was created(1 = two-year college; 0 = four-year college). This variablewas used to create two-year and four-year subsamples.Students’ grade point average (GPA) was a categoricalvariable that averaged grades for all coursework in 9ththrough 12th grades. There were seven categories:(0 = .00–1.00; 1 = 1.01–1.50; 2 = 1.51–2.00; 3 = 2.01–2.50;

4 = 2.51–3.00; 5 = 3.01–3.50; and 6 = 3.51–4.00). Students’college expectations were measured by asking studentshow far they expected to go in school. A dichotomousvariable was created based on their responses (1 = expectsto graduate from a four-year college; 0 = does not expect tograduate from four-year college).

Importance of college costs were measured by askingstudents how important low costs (such as tuition, books,room and board) were for choosing a school, with responseoptions including not important, somewhat important,or very important. The responses were dichotomized(1 = very important; 0 = not very important). Importance offinancial aid was measured by asking students howimportant the availability of financial aid was for choosinga school, with responses including not important, some-what important, or very important. Responses weredichotomized (1 = very important; 0 = not very important).


4.4. Parent/household variables

Parents’ education level was equivalent to mother’shighest level of education or father’s highest level ofeducation whichever was higher. Parents’ level of educa-tion was composed of eight distinct levels. The eight levelswere collapsed into three for the final analysis (0 = Highschool diploma or less; 1 = Some college; and 2 = four-yearcollege degree or higher). In the ELS:2002, householdincome was composed of 13 distinct levels. For thepurposes of this study, the levels of householdincome were combined into four levels (0 = Low-income[$0–20,000]; 1 = Moderate-income [$20,001–50,000];2 = Middle-income [$50,001–100,000]; and 3 = High-income [$100,001 or higher] due to sample size.

4.5. School variables

In the case of school climate, principals are asked todescribe their school’s climate using a Likert Scale (1 = notaccurate at all to 5 = very accurate). They are asked to ratesuch statements as ‘‘student morale is high,’’ ‘‘teachers atthis school press students to achieve academically,’’ and‘‘students are expected to do homework.’’ Higher valuesrepresent principle’s perceptions of a more academicallyoriented climate. Variable was created through principalfactor analysis by the ELS staff. It is a composite score. Thecoefficient of reliability (alpha) for the scale is .86.

Number of guidance counselors is the number of full-time guidance counselors in a particular school. Number ofguidance counselors is included because findings suggestthat student who have access to high school guidancecounselors receive information about college and helpwith college-admissions requirements; consequently, theyare more likely to enroll in college and fill out financial aidforms (Perna & Titus, 2005; Stanton-Salazar, 1997).

Private school indicated the type of school attended bythe respondent in the base-year interview: (1) public, (2)Catholic school, or (3) other private. For the purposes ofthis study, a dichotomous variable was created (1 = privateor other private; 0 = public). Free/reduced lunch was thepercent of 10th graders receiving free or reduced pricelunch and was a categorical variable in the ELS: 2002(1 = 0–5%; 2 = 6–10%; 3 = 11–20%; 4 = 21–30%; 5 = 31–50%;51–75%; and 76–100%). Parents’ college expectations weremeasured by asking parents how far they thought theirchild would go in school. A dichotomous variable wascreated based on their responses (1 = expect child tograduate from a four-year college; 0 = do not expect childto graduate from four-year college).

4.6. College assets variables

Variables of interest came from questions askingparents what they were doing to financially prepare fortheir child to attend college. These variables representedthe types of assets available to students to pay for collegecosts. The following college assets were included: started asavings account; bought U.S. savings bonds; invested instock/real estate; opened a college investment fund (i.e.,mutual fund); planned to take out a home equity loan;

opened a state-sponsored savings plan; and told student toput aside money for college. All variables were dichoto-mous (1 = yes; 0 = no).

4.7. Outcome variables

Student contributions were based on three questionsthat asked students whether or not they paid for collegewith (1) student loans, (2) savings or job earnings, and (3)federal work-study grants. Responses to these questionswere combined to create two categories (1 = paid withstudent contributions; 0 = did not pay with studentcontributions). Family contributions were based on twoquestions that asked students whether or not they paid forcollege with (1) parent loans and (2) contributions fromfamily. Responses to these questions were combined tocreate two categories (1 = paid with family contributions;0 = did not pay with family contributions). Societalcontributions were based on a question that askedstudents whether or not they paid for college with grantsand scholarships. Responses to these questions werecombined to create two categories (1 = paid with societalcontributions; 0 = did not pay with societal contributions).The three outcome variables were measured in 2006.

Overall, Table 2 findings suggest that students weremore likely to report having paid for college with societalcontributions (73%, full sample) and student loans (63%,full sample) than any of the other factors considered.Work-study (18%, full sample) was the least commonlyreported method for having paid for college. Not surpris-ingly, a higher percentage of students who attended a four-year college reported using each of the different methodsfor having paid for college than students who attended atwo-year college. African Americans (68%) and moderate-income (69%) students who attend a four-year collegereport using student loans more often than any othergroup and Asian (51%) students report using studentloans less than any other group. High-income (72%) andAsian (62%) students have the highest percentage ofstudents who report paying for college with familycontributions while low-income (87%) students are mostlikely to report paying for four-year college with societalcontributions.

Table 3 provides information on the percent of studentswho reported using student, family, and societal contribu-tions. In the aggregate and in the case of the four-yearcollege sample, a higher percentage of students reportedhaving paid for college with student contributions thansocietal contributions. In regards to the two-year sample,White students were the only racial/ethnic group to have ahigher percentage of students who reported using studentcontributions when compared to the percentage ofstudents who reported using societal contributions.However, in the case of the four-year college sample, onlyAsian students did not have a higher percentage ofstudents who reported that they used student contribu-tions to pay for college when compared to students whoreported using societal contributions. Interestingly, ahigher percentage of low-income students at two-yearcolleges reported using student contributions to pay forcollege than they did societal contributions.

Table 2

Percent of students who report paying for college with student, family, and societal contributions.

Covariates Student contributions Family contributions Societal contributions

Student

loans

Work

study

Savings/job

earnings

Family

[26_TD$DIFF]contributions

Parent

[27_TD$DIFF]loans

Grants/scholarships

Full (full sample) 52 15 44 52 21 71

White 46 14 47 56 23 69

Asian 57 19 41 56 16 74

Latino/Hispanic 56 14 43 45 18 69

African American 44 18 34 39 22 76

Low-income 57 18 39 32 11 81

Moderate-income 47 18 47 43 19 75

Middle-income 44 14 46 58 26 66

High-income 58 11 39 71 23 65

Full (two-year college sample) 28 06 44 38 09 61

White 32 04 46 42 10 59

Asian 16 09 45 36 10 63



Low-income 28 03 39 50 08 53


Middle-income 29 06 45 34 10 65

High-income 20 07 37 26 06 72

Full (four-year college sample) 61 19 45 57 26 74

White 61 17 48 60 27 72

Asian 51 22 40 62 19 77



Low-income 60 25 41 37 14 87


Middle-income 63 16 45 62 31 70

High-income 44 12 39 73 25 67


Notes: Row percentages are reported. Data imputed using the Expectation Maximization (EM) algorithm. Full (weighted n = 1,652,963; non-weighted

n = 7366); two-year (weighted n = 505,954; non-weighted n = 2003); four-year (weighted n = 1,147,009; non-weighted n = 5363).

Table 3

Percent of students who report paying for college with student, family, and societal contributions.

Covariates Full Two-year Four-year

Student Family Societal Student Family Societal Student Family Societal

Full sample 72% 59% 69% 59% 42% 61% 78% 67% 73%

Race

White 75 64 68 64 46 59 78 71 71

Asian 63 64 74 52 42 62 67 70 77

Latino/Hispanic 68 50 66 56 42 57 77 58 75

African American 69 47 74 52 29 73 80 57 75

Income level

Low-income ($0–20,000) 64 36 78 56 53 53 75 41 87

Moderate-income ($20,001–50,000) 74 50 73 64 52 50 83 58 79

Middle-income ($50,001–100,000) 76 68 63 60 39 65 79 74 68

High-income ($100,001 or higher) 64 78 64 50 30 72 65 81 66


Notes: Row percentages are reported. Data imputed using the Expectation Maximization (EM) algorithm. Full (weighted n = 1,652,963; non-weighted

n = 7366); Two-year (weighted n = 505,954; non-weighted n = 2003); Four-year (weighted n = 1,147,009; non-weighted n = 5363).


5. Analysis plan

5.1. Missing data

The first step in the analysis was to account for missingdata. Missing data were assumed to be missing at random,and handled by expectation–maximization (EM) imputation(Dempster, Laird, & Rubin, 1977). This method estimates

unmeasured data and is based on iterating through twoalternating steps (i.e., the expectation and maximizationsteps). A value is calculated for the missing data based onthe observed data and its distribution in the expectationstep, and calculated based on the current updated datasetin the maximization step. These two steps are alternatednumerous times until a better model can be specified toestimate more accurate missing values (Little & Rubin, 1987).


5.2. Multivariate (trivariate) probit model

In the second step in the analysis we conducted amultivariate probit model using the ‘‘mvprobit’’ programin STATA 11.0. Preliminary analyses of the three primaryoutcomes of interest (student contributions, family con-tributions, and societal contributions) revealed that therewas a significant correlation between different pairs ofoutcomes. Therefore we concluded that analyses thatignored correlations across outcomes, such as simpleunivariate probits, might lead to bias (Cappellari & Jenkins,2003). A trivariate probit model is a generalization ofunivariate probit model. It allowed us to estimate threedichotomous dependent variables simultaneously whileexplicitly modeling the correlation in disturbance termsusing simulated maximum likelihood methods (i.e., thismethod accounts for the fact that the same child can usestudent, family, and societal contributions to pay forcollege; Cappellari & Jenkins, 2003). The coefficientestimates from the trivariate probit model accounted forunobserved correlation among the outcomes. Appendix Aprovides equations for conducting the analysis.

Because ELS: 2002 randomly selected approximately 26students within each school, we adjusted standard errorsby clustering them into the same school unit. Further, boththe descriptive and binary regression analyses wereweighted using the ELS: 2002s second follow-up baseyear panel weight. Weights are used to compensate forunequal probabilities of selection and to adjust for theeffects of nonresponse. Using weights allows a researcherto make generalizations to the national populations (formore information see, Ingels et al., 2007).

Marginal effects are typically calculated in probitmodels; however, they are difficult to compute in trivariateprobit models. Given this, predicted probabilities of apositive response for each of the three outcomes based onthe weighted trivariate probit model were calculatedinstead. The ‘mvppred’ program in STATA Version 11.0 wasused to calculate the predicted probabilities (Cappellari &Jenkins, 2003). We present the weighted mean of thepredicted probabilities for each race/ethnic subgroup ofour sample (e.g., the mean predicted probability of usingstudents’ contributions for Whites, Asians, Latinos/Hispa-nics, and African Americans) and for each income subgroupof our sample (e.g., low-income, moderate-income, mid-dle-income, and high-income).

We also calculate what we refer to as students’ collegecost burden from predicted probabilities. The college costburden is the difference between the predicted probabilitystudents report using societal contributions to pay forcollege from the predicted probability he/she uses studentcontributions.

6. Trivariate probit results

To reduce space and to make comparisons of resultsacross the three outcomes and the three samples, signs ofsignificant predictors of student, family, and societalcontributions for the full, two-year, and four-year samplesare presented in Table 4. Appendices B–D provide detailedtables of trivariate probit estimates, adjusted standard

errors, confidence intervals, and estimated correlationcoefficients for all three outcomes (i.e., student, family, andsocietal contributions) for all three samples (i.e., fullsample, two-year college sample, and four-year collegesample). In this section we report statistical significance,direction of the relationship and we interpret correlationcoefficients for each of the three samples. To conservespace, however, we only report on statistical significancefor the four-year college sample.

6.1. Full sample

The trivariate probit regression results for the fullsample are presented in Appendix B. Being a dependentstudent, being Asian compared to being white, having ahigher GPA, attending a private school in 10th grade orbefore, and being a student who expects to graduatecollege all reduce the chance that students report payingfor college with student contributions. Conversely, astudent of moderate- or middle-income compared to astudent of low-income and being a student who perceivesthat financial aid is very important increase the likelihoodthat students report paying for college with studentcontributions. In regards to parents’ savings, being astudent with a parent who started a savings account, astate-sponsored savings account, a college investmentfund, and/or invested in real estate or stocks all reduce thelikelihood that students report paying for college withstudent contributions. However, having parents who planto mortgage their home to pay for college or having parentswho tell you to put aside money for college increase theodds that students report paying for college with studentcontributions.

The only significant and positive correlation in the fullmodel is between student contributions and familycontributions; the correlation coefficient for these twooutcomes is .147 (95% CI: .095, .198) (see Appendix B). Thissuggests that these equations share the same unobser-vables in the error terms. The correlations between studentand societal contributions and family and societal con-tributions are both significant and negative (�.085 [95%CI: �.139, �.031] and �.122 [95% CI: �.172, �.072],respectively). This indicates that the expected uncondi-tional relationship between family contributions andsocietal contributions, for example, is not fully removedthrough the inclusion of the explanatory variables.Dependent status, being Asian, Latino/Hispanic, AfricanAmerican, attending a two-year college.

6.2. Two-year college sample

The trivariate probit regression results for the fullsample are presented in Appendix C. Being a dependentstudent, being in a family with a higher number of siblings,attending a school with a higher percentage of students onthe free/reduced lunch program, and being a student whoperceives that college costs of financial aid are veryimportant for choosing a college all reduce the chance thatstudents report paying for college with family contribu-tions. Conversely, being Asian compared to being white,living with a head of family with more schooling, a student

Table 4

Statistically significant predictors of student, family, and societal contributions.

Predictors Student contributions Family contributions Societal contributions

Full Two-year Four-year Full Two-year Four-year Full Two-year Four-year

Dependent student � � � � � � � �Asian � � � + +

Latino/Hispanic � �African American � � � + + +

Male + + �Student – two-year college � � � +

Number of siblings � �Head – some college + �Head – four-year college degree or higher + + +

Moderate-income ($20,001–50,000) + + + + � �Middle-income ($50,001–100,000) + + + + + + � � �High-income ($100,001 or higher) + + + � �GPA � � + + + + +

Private school (by 10th grade) � � �School climate

Number of guidance counselors � � �Free/reduced lunch � � � � + +

Student expects to graduate college � + + + + +

Parent expects student to graduate college + + � �Low college costs very important � � � �Financial aid very important + + � � � + +

Plan to remortgage home + + +

Start a savings account � � � +

Have student put aside earnings + + + + + +

Start state-sponsored savings � � � � � �College investment fund � � � + +

Invest in real estate/stocks � �Buy U.S. savings bonds + + +


Notes: See Appendices B–D for trivariate probit estimates and correlation coefficients. Full (weighted n = 1,652,963; non-weighted n = 7366); Two-year

(weighted n = 505,954; non-weighted n = 2003); Four-year (weighted n = 1,147,009; non-weighted n = 5363).


of moderate-, middle-, or high-income compared to astudent of low-income, having a higher GPA, and being astudent who expects to graduate from college increase thelikelihood that students report paying for college withfamily contributions. In regards to parents’ savings, onlyhaving parents who started a state-sponsored savingsaccount reduces the chance that a student reports payingfor college with family contributions. Both having parentswho start a saving account to pay for college and/or open acollege investment fund increase the likelihood thatstudents report paying for college with family contribu-tions.

The trivariate probit regression results for the two-yearcollege sample are presented in Appendix C. Estimatedcorrelation coefficients listed at the bottom of Appendix Care in the same direction and all are significant similar tothe full model. However, the strength of relationship ineach case is stronger in the two-year college sample than itwas in the full sample.

6.3. Four-year college sample

The trivariate probit regression results for the fullsample are presented in Appendix D. Being a dependentstudent, being a student of moderate-, middle-, or high-income compared to a student of low-income, being in aschool with a higher number of guidance counselors, andbeing a student who perceives that higher college costs are

very important for choosing a college all reduce the chancethat students report paying for college with societalcontributions. Instead, being an African American studentcompared to being a white student, having a higher GPAscore, and being a student who perceives that financial aidis very important for choosing a college all increase thelikelihood that students report paying for college withsocietal contributions. In regards to parents’ savings, notype of college assets reduce the chance of studentsreporting paying for college with societal contributions buthaving parents that start a savings account to help pay forcollege increases the chance that students report payingfor college with societal contributions.

The trivariate probit regression results for the four-yearcollege sample are presented in Appendix D. The onlysignificant positive correlation is between student con-tributions and family contributions; the correlationcoefficient for these two outcomes is .110 (95% CI: .046,.185). Unlike in the full sample and the two-year sample,the correlation between student and societal contributionsis not significant in the four-year sample. The correlationbetween family and societal contributions is significantand negative �.077 [95% CI: �.139, �.015].

In sum, in regards parents’ college savings, Table 4indicates that different types of college assets affect howstudents pay for college in different ways. Planning tomortgage a home to pay for college and telling a student toput aside earnings for college in 10th grade are positive


predictors of student contributions in all three samples.U.S. savings bonds are a positive predictor in the two-yearsample but not the four-year sample. Conversely, openingup a savings account, starting a state-sponsored savingsplan, or a college investment fund reduces the chance thatstudents pay for college with student contributions in allthree samples. Investment in real estate/stocks reducesthe chances of reporting paying for student contributionsin full and the four-year samples. With respect to familycontributions, starting a savings account, having studentsput aside earnings, opening a college invest fund, andbuying U.S. savings bonds all are related to students beingmore likely to report paying for college with familycontributions. However, whether they have a significantpositive effect varies by the type of college students areattending. Finally, having students put aside earnings isthe only college asset variable that is a significant

[(Fig._1)TD$FIG]

Fig. 1. Predicted probabilities for student, parent, and societal contributions and

attended college and applied for financial aid. Notes: Estimates adjusted for clus

(EM) algorithm. Full = students who attended either a two-year or four-year c

(weighted n = 505,954; non-weighted n = 2003); Four-year (n = 1,147,009; non-

societal contributions. Open brackets indicate the financial aid gap (societal co


predictor of societal contributions in the four-year sampleonly.

7. Predicted probabilities and students’ college costburden by race

It is difficult to determine the magnitude of differencesby race and income, key variables of interest in this study,by interpreting the coefficient estimates. Thus, we calcu-late marginal predicted probabilities for race and income.In addition we calculate students’ college cost burden.

7.1. Within groups

Fig. 1 presents predicted probabilities and the collegecost burden for racial/ethnic groups for all three outcomes.White students who attend a two-year college by 2006

the financial aid gap by race and type of college among students who have

tering within schools. Data imputed using the Expectation Maximization

ollege. Full (weighted n = 1,652,963; non-weighted n = 7366); Two-year

weighted n = 5363). *Student contributions are equal to or greater than

ntributions minus student contributions).


have a lower probability of reporting paying for collegewith family contributions than White students who attenda four-year college. Further, in the full, two-year, and four-year samples White students have a lower probability ofreporting paying for college with societal contributionsthan they do student contributions. The college costburden for White students who attend a two-year collegeis �8, and for a four-year college it is �4. Asian studentshave a higher probability of reporting that they pay forcollege with societal contributions than student contribu-tions in all three samples. Moreover, they have a collegecost burden of +14 with respect to two-year collegeattendance and +10 with respect to four-year collegeattendance. Among Latino/Hispanic students, the proba-bility of reporting that they pay for college with studentcontributions is equal or higher than the probability thatthey report paying for college with societal contributions.This is reflected in the college cost burden of zero fortwo-year colleges and �1 for four-year colleges. AfricanAmerican students have a higher probability of reportingthat they pay for college with student contributionsthan societal contributions in the four-year sample butthe opposite is true in the two-year sample. Thecollege cost burden is incentive laden +25 at two-yearcolleges for African American students. In contrast, itprovides a disincentive at four-year colleges (college costburden =�3).

7.2. Comparing groups

Among minority groups, Asian students have thehighest probability of reporting paying for college withfamily contributions regardless of whether they attendtwo-year colleges or four-year colleges. Using the collegecost burden, findings also indicate that Asian studentshave the greatest incentive to attend college overallwhen compared to all other racial/ethnic groups. Theircollege cost burden at two-year colleges is +14 and a +10at four-year colleges. Latino/Hispanic students do notseem to be given much incentive by society to attendeither two-year colleges or four-year colleges. They arethe only minority group for which the college cost burdenis zero or negative for both two-year and four-yearcollege attendance. This is also true of White students.Further, Latino/Hispanic students have the lowestprobability of reporting having paid for college withsocietal funds at a two-year college when compared to allother racial/ethnic groups. African Americans have one ofthe greatest disincentives to attend four-year colleges ofany racial/ethnic group using the college cost burden.First we find that they have the second highest collegecost burden at four-year colleges (White students �4;African American �3). Second, findings indicate that theyby far have the greatest incentive to attend two-yearcolleges of any racial/ethnic group using the college costburden. The college cost burden for African Americans attwo-year colleges is a whapping +25. Moreover, AfricanAmericans have the lowest probability of reportingpaying for college with family contributions whetherattending a two-year college or a four-year college of anyracial/ethnic group.

8. Predicted probabilities and the college cost burden byincome level

8.1. Within groups

According to the college cost burden, low-incomestudents receive societal incentive to attend both two-year and four-year colleges. The incentive is higher fortwo-year colleges (college cost burden = +20) than four-year colleges (college cost burden = +14). The probabilitythat low-income students report having used familycontributions to pay for two-year college attendance orfour-year college attendance falls considerably below .50in both cases (.32 and .42, respectively). Moderate-incomestudents have a small incentive to attend two-year colleges(college cost burden = +6). In contrast, they have adisincentive to attend four-year colleges (college costburden =�1). Middle-income students have the largestdisincentive to attend college regardless of type of collegeusing the college cost burden. The college cost burden formiddle-income students at two-year colleges is�10 and ata four-year college it is a �14. In the aggregate, high-income students receive an incentive to attend college(college cost burden in full sample = +4). When the data aredisaggregated by type of college they receive smalldisincentives to attend both two-year colleges (collegecost burden =�2) and four-year colleges (college costburden =�5). However, particularly in the case of four-year college attendance, many high-income studentsreport having received family contributions to pay forcollege.

8.2. Comparing groups

A general principle of the financial aid system is thatthe higher your income the more of a burden you shouldbear for paying for college. Findings support this inregard to family contributions and societal contributions.We find as income raises the probability that students’pay for college with family contributions also raises.Further, as income decreases the probability thatstudents pay for college with societal contributionsincreases. However, this pattern does not continue inthe case of student contributions. In the case of two-yearcolleges, only low-income students have a lower proba-bility of reporting that they pay for college with studentcontributions than high-income students (.50 vs. .57,respectively). When we consider four-year colleges, low-income students have a higher probability of reportingthat they pay for college with student contributions thando high-income students (.73 vs. .62, respectively). This ismost likely due to the low probability that low-incomestudents receive family contributions compared to high-income counterparts (.42 vs. .82, respectively). Moreover,according to the college cost burden, middle-incomestudents have the least incentive to attend collegeregardless of the type of college when compared to allother income groups.

It is also interesting to note that unlike the case ofAfrican Americans who receive very large incentivesto attend two-year colleges and disincentives to attend


four-year colleges just the opposite is true in the case ofhigh-income and middle-income students. According thecollege cost burden, middle-income students have a largerdisincentive to attend two-year colleges than they do four-year colleges (two-year college cost burden =�14 vs. four-year college cost burden =�9). Whereas, high-incomestudents have a disincentive to attend two-year colleges(college cost burden =�2) but an incentive to attend afour-year college (college cost burden = +5). What cannotbe left out from this discussion is that middle- and high-income students are almost assured of receiving familysupport if they attend four-year colleges (.75 and .82,

[(Fig._2)TD$FIG]

Fig. 2. Predicted probabilities for student, parent, and societal contributions and

attend college and applied for financial aid. Notes: Estimates adjusted for clusterin

algorithm. Full (weighted n = 1,652,963; non-weighted n = 7366); Two-year (weig

weighted n = 5363). *Student contributions are equal to or greater than socie

contributions minus student contributions).


respectively) whereas there is only about a .50 probabilitythat they report receiving family contributions to pay forcollege if they attend two-year colleges (.51 and .57,respectively).

Lastly, several income groups have a negative collegecost burden. High-income students have a negative collegecost burden with respect to two-year college attendance(college cost burden =�2), middle-income students withboth two-year college attendance (college cost burden =�14) and four-year college attendance (college costburden =�9), and moderate-income students with four-year college attendance (college cost burden =�1) (Fig. 2).

the financial aid by income level and type of college among students who

g within schools. Data imputed using the Expectation Maximization (EM)

hted n = 505,954; non-weighted n = 2003); Four-year (n = 1,147,009; non-

tal contributions. Open brackets indicate the financial aid gap (societal


9. Discussion

We find that the only statistically significant, positivecorrelation across all three samples is between studentcontributions and family contributions. We conclude fromthis that the same unmeasured variables that increase thechance of student contributions also increase the chancesof family contributions. It is as if the parents say to theirchildren, ‘‘you pay your share, we’ll pay our share.’’ If theparents say, ‘‘we won’t pay’’, it appears children are alsoless willing to pay.

With regard to the full and two-year samples, the ideathat there is a type of meta-message communicated byparents and children, ‘‘you pay your share, we’ll pay ourshare’’ is strengthened by our findings on student andparent expectations. When students expect to graduatefrom college, we find that they are more likely to reportthat their parents contribute to paying for college.Similarly, when parents expect their child to graduatefrom college, students are more likely to report contribut-ing to paying for college. It appears that one way thatpositive expectations may work is by signaling to the otherthat it is safe to contribute, ‘‘you can trust me.’’ Forexample, for parents to invest in college, they mustaccurately predict that their child will complete college inorder to receive a return on their investment. Positivestudent expectations may provide parents with muchneeded confidence that the student will graduate. Studentexpectations remain an important predictor of familycontributions in the four-year sample, but parent expec-tations do not. Carrying our line of reasoning forward, thissuggests that among four-year college goers it remainsimportant for parents that students provide them with atype of insurance that it is safe to invest. However, itappears it might be less important that parents signal tostudents that it is safe to invest. Perhaps, when studentsattend a four-year college the meta message is, ‘‘If youdon’t pay, we won’t pay.’’ However, given that the outcomevariables are measured at the same time, an alternativeexplanation cannot be ruled out. That is, students may firstmake efforts to help pay for school then parents act.

The correlation between student and societal contribu-tions and family and societal contributions are bothsignificant and negative in the full and two-year samples.As such, we suggest that both sets of relationships might beinterpreted as substitutes for one another. For example, ifthe unmeasured effects raise family contributions, theyalso reduce societal contributions in all three samples.Since the fitted model accounts for the financial status ofthe parents, this effect may be interpreted as a substitutioneffect. The student and societal correlational relationshipmight also be interpreted as a substitution effect in boththe full and two-year sample. Since the student andsocietal correlation is not significant in the four-yearsample it cannot be interpreted as a substitute effect.However, caution is needed and results should beinterpreted as preliminary. More evidence is neededbefore any conclusion can be drawn about an actualsubstitution effect.

Further, in this study we ask, ‘‘Are students as likely asor more likely to bear the responsibility of paying for

college than society?’’ The answer seems to be yes,particularly in the case of four-year college attendance.The college cost burden among White, Latino/Hispanic,and African American students who attend a four-yearcollege is negative in each case. Similarly, it is negativeamong middle- and moderate-income students. Thisindicates in each of these cases, students have a higherprobability of reporting paying for college with studentcontributions than societal contributions. Sallie Mae(2011) reports that the percentage of student contribu-tions (i.e., student borrowing and student income andsavings) is slightly lower than societal contributions (26%vs. 33%, respectively). The most important reason for thisdifference might be that Sallie Mae (2011) uses descriptivedata; they do not attempt to predict which students usedifferent types of contributions while controlling for avariety of factors. The college cost burden may beexacerbated by the fact that in each of the cases wherethere is a negative gap there is also a higher probabilitythat students report paying for college with studentcontributions than they do family contributions. Moreover,the bulk of student contributions are in the form of studentloans that can have long-term negative effects (e.g.,American Student Assistance, 2010).

The second research question we examined was, ‘‘Areminority and low-income students as likely as or morelikely to be asked to carry the responsibility of paying forcollege than White and higher income students?’’ In thecase of Asian students, the answer is a resounding ‘‘no’’regardless of the type of college. This might be because ofthe shift to more merit-based aid. Research suggests thatAsian students have the highest test scores of any racial/ethnic group (Kao & Thompson, 2003). In general, it can besaid that Asian students have a higher incentive to attendcollege than other groups. Moreover, of any minoritygroup, they have the highest probability of paying forcollege with family support. This is in line with King’s(1999) finding which indicates that Asian students aremore likely than either African American or Latino/Hispanic students to have an expected family contributionof more than $12,500 per year.

Findings are mixed in the case of African Americans.With respect to two-year college attendance, clearlyAfrican Americans receive a far greater incentive to attendwith a college cost burden of +25. An explanation for this isAfrican American students often come from low-incomefamilies with little assets (King, 1999). As a result, theyoften are not expected to make any financial contributiontoward paying for college. For example, King (1999) findsthat 36% of African American students are not expected topay anything toward college costs. Furthermore, King(1999) suggests that a reason why African Americanstudents use grants and scholarships at higher percentagesthan other racial/ethnic groups is because a high percent-age of African American students are independent studentswith dependents. In turn, they have a lower income profilemaking them more likely to be eligible for grants andscholarships. However, this explanation is less convincingin the case of four-year colleges. With respect to four-yearcolleges, the college cost burden of African American isnegative, which means African American students are

2 Sallie Mae (2011) defines low-income as <$35,000 and high income

as $100,000 or more.


more likely to report paying with student contributionsthan they are with societal contributions. Moreover, theburden of African American students is nearly equal to thatof White students. When the two-year and four-yearfindings are considered together, it provides an explana-tion for why research might indicate that African Americanstudents are overrepresented at two-year universitieswhere there is less chance that students continue on for afour-year degree (Louie, 2007).

Alternatively, one might suggest that while college costburden is upside down for African American students, theprobability of using societal contributions to pay forcollege is roughly equal for two-year and four-year collegeattendance (73 vs. 76, respectively). However, this does nottake into consideration that African American students arethe only group for which the probability of paying forcollege with societal goods at a two-year college and four-year college is about equal. For all other racial/ethnicgroups the probability of paying for college with societalcontributions is noticeably higher with respect to four-yearcolleges. Moreover, this line of reasoning does not get at theproblem of students’ college cost burden because it onlylooks at societal contributions without considering studentcontributions. The financial aid incentive structure withrespect to African American and White students is evenmore distorted when we consider family contributions.African American students have the lowest probability ofreporting using family support to pay for college than anyother racial/ethnic group. The family contributions disparitymay be explained with the family composition argumentarticulated by King (1999) earlier in this paragraph.

While white students are more likely to report payingfor college with student contributions than they aresocietal contributions (i.e., have a larger negative collegecost burden) when compared to Latino/Hispanic students,the college cost burden still appears to be unfavorable, ifnot inequitable for Latino/Hispanic students. An explana-tion for why the gap for Latino/Hispanic students mightnot be as large as it is for African American students, forexample, is that Latino/Hispanic students are more averseto borrowing to pay for college than African Americanstudents (Cunningham & Santiago, 2008). Consistent withCunningham’s and Santiago’s (2008) findings, our descrip-tive data also suggest that Latino/Hispanic students areless likely to pay for college with student loans than allother groups but Asian students. Their aversion toborrowing may work to reduce the college cost burdenthey face since college loans are the primary ways studentscontribute. Despite this, the college cost burden is likelyworsened by the fact that Latino/Hispanic students have alower probability of reporting paying for college withfamily contributions than White students.

In the case of low-income students, there appears to belittle evidence from this study that low-income studentsare being asked to bear more of the burden of paying forcollege when compared to other income groups based ontheir college cost burden. However, there are largedisparities in family contributions when compared toother income groups, particularly high-income students.Given this, it is important to note that we are unable toascertain whether the amount of grants and scholarships is

sufficient to make-up for low family contributions amonglow-income students. That is, while the basic pattern ofhow students pay for college is one of creating equality ofopportunity, it may not be in sufficient amounts to actuallyprovide equality (e.g. ACSFA, 2002, 2006, 2010).

Moderate-income and middle-income groups appear tohave the most regressive college cost burden of any incomegroups, especially when four-year colleges are considered.While both moderate-income and middle-income stu-dents are discouraged to attend four-year collegesaccording to the college cost burden, high-income studentsare encouraged. This provides evidence that lower incomestudents, with the exception of the lowest income bracket,are being forced to bear more of the responsibility forpaying for college when compared to high-income stu-dents. Like in the case of race, this problem is onlyexacerbated when family contributions are considered.Further, when family contributions are considered, it mightbe argued that the financial system least favors moderate-income students because the probability that they receivefamily contributions is far less than that of middle-incomestudents. In line with this, Sallie Mae (2011) reports thatamong high-income students, 43% of the cost of college ispaid through family income and savings with an additional8% being paid through family loans. That means that overhalf of the cost of college for high-income students is paidfor through family contributions. In contrast, only about25% of college costs are paid for by family contributionsamong low-income students (Sallie Mae, 2011).2

We also examined whether college assets increase orreduce the likelihood that students report paying forcollege with student, family, and/or societal contributions.We find that different types of college assets affect howstudents pay for college in different ways. For example,planning to mortgage a home to pay for college and tellinga student to put aside earnings for college in 10th grade arepositive predictors of student contributions in all threesamples. It might be that planning to mortgage a home andtelling a student to put aside earnings signal to studentsthat parents do not have enough money put aside to pay forcollege and students will have to contribute if they want togo to college. It appears that this might be interpretedpositively, at least among students who apply for financialaid and who attend college. These students might interpretthese types of assets as meaning that even though theirparents cannot afford to pay for college, their parents see itas a worthwhile investment. While they might beinterpreted as positive and encourage students to progresstoward college and contribute financially, they do notprovide actual resources for paying for college. Thus theyare positively related with student contributions.

U.S. savings bonds are also a positive predictor ofstudents’ reporting paying for college with studentcontributions but only in the two-year sample. It mightbe that in some cases college-savings bonds are given tostudents to save. These students might see them as theirown money and part of the contribution they make toward


their education, thus they have a positive relationship withstudent contributions. Alternatively, it might be thatsavings bonds also provide a positive signal to studentsthat their parents expect them to attend college but as asavings mechanism provide very little in actual savings. Incontrast, starting a savings account, a state-sponsoredsavings plan, a college investment fund or investing in realestate/stocks may provide students with actual resourcesfor covering the cost of college which drive down the needfor student contributions. Therefore, we find a negativerelationship between these types of college assets andstudent contributions.

With respect to family contributions, on the one hand,putting aside earnings and U.S. savings bonds aresignificant in the two-year sample but not the four-yearsample. On the other hand, starting a savings account andcollege investment funds are significant in the four-yearsample but not in the two-year sample. Non-significantresults for both telling students to put aside earnings andbuying U.S. savings bonds in the four-year sample may bedue to the cost of four-year colleges compared to two-yearcolleges. Since both types of assets may result in far lessactual accumulation of assets, they may hold less sway onwhether or not students report paying for college withfamily contributions at more costly four-year colleges thanthey do at less expensive two-year colleges.

State-sponsored savings plans in the parent’s name arenegative predictors of family contributions. There might beseveral reasons for the negative relationship. A reasonmight be that state-sponsored savings plan managers do apoor job of informing families about how state plans can beused to finance college. For example, Sallie Mae (2011)finds 76% of families who opened a state-sponsoredsavings plan report that savings plan companies are‘‘neither’’, ‘‘fairly unhelpful’’ or ‘‘very unhelpful’’ inproviding information about financing college. The lackof information about the utility of state plans to helpfinance college, coupled with negative media coverageabout the potential of these plans to reduce the amount ofneed-based aid available, might have a negative effect onwhether students report paying for college with familycontributions. However, the small sample size of parentswith state-sponsored savings plans for their childrenmight make findings on state-savings plans misleading. Asa result, findings on state-sponsored savings plans shouldbe interpreted as tentative, and future research with largersamples sizes and more complete information should beexamined before conclusions are drawn.

Having students put aside earnings is the only collegeasset variable that is a significant predictor of societalcontributions. It has a positive relationship in the four-yearsample. This might be for the same reasons why it has apositive relationship in regards to student contributions.Clearly, more research is needed that attempts tounderstand why different types of college assets havedifferent effects.

10. Limitations

There are several notable limitations that should beconsidered when interpreting the study results. First,

while each school was supposed to include 26 randomlyselected students, there was considerable variation in thenumber of students whose data were collected throughoutthe 2004 and 2006 waves, which reduces the representa-tiveness of the population. Second, missing data variedacross the different items contained in the surveys, andmany of the later items in the student questionnaire werenot missing at random. Steps were taken to counter thispotential threat by imputing data to replace missingdata. Nevertheless, estimates may contain a degree ofmissing data bias. A third limitation is the inability toexamine contribution amounts. For example, whether theamount of family contributions is higher for high-incomestudents than the amount of societal contributions is forlow-income students. A fourth limitation is the use ofstudent reports. For example, it might be that somestudents do not see some things as family contributionsthat are. For instance, living at home provides studentswith considerable resources to help make college afford-able (such as, not having to pay room and board). However,because students may not see that as money to pay fortuition costs or books they may not report it as a familycontribution.

Further, caution is needed in interpreting results fromthe state-sponsored savings variable and the homeremortgage variable. Only a small percentage of parentshave started a state sponsored savings program (8%) andonly 9% of parents expect to remortgage their home topay for college Small sample size can lead to inaccuratefindings.

11. Implications

Research has consistently shown that studentexpectations are an important predictor of students’educational outcomes (Cook et al., 1996; Marjoribanks,1984; Mau, 1995; Mau & Bikos, 2000; Mickelson, 1990).However, little research examines the role of studentexpectations on family contributions for financingcollege. Our findings indicate that students who havepositive college expectations are more likely to haveparents who help pay for their education. How mightthis work? We speculate that programs that helpincrease student expectations may not only improvestudent educational outcomes, but that they might alsohelp students signal to parents that they can trust theirchild to complete college.

In regards to race and income, Asian students appear tobe the best equipped to take advantage of the currentfinancial aid system and its emphasis on merit-based aid.With regard to African American students, they are beingincentivized to attend two-year colleges over four-yearcolleges. Research shows that students who attend two-year colleges are less likely to complete a degree (McIntosh& Rouse, 2009) and less likely to go on to a four-yearcollege (Long & Kurlaender, 2008). Given this, we suggestproviding both more and higher dollar grants and scholar-ships at four-year colleges so that African Americanstudents have to rely on paying for college with studentcontributions less. We suggest higher amounts because theprobability of using grants and scholarships is about equal

ppendix A

The following equations represent the trivariate probitstimates modeled in this study, where i equals the nthubject and m equals the nth variable. These equations weresed to calculate joint results for three outcomes and accountr correlations between the errors. In other words, equations

alculate whether or not students use each type ofontribution (e.g., student, family, and societal contributions)t the same point in time while accounting for correlationsetween the errors of the three models.

.1 Student contributions�i;m¼ b00Xi0 þ b01Xi dependent student status þ b02Xi race

þb03Xi gender � � �b023Xi savings bonds þ em;

where m = 1(race), . . ., M(savings bonds)

Student contributionsi m ¼ 1 if studentcontributions�i m >0 and 0 otherwise

ei m, where m = race, . . ., savings bonds areerror terms that have multivariate normaldistributions, each with a mean of zeroand variance/covariance matrix V, with valuesof 1 on the diagonal and correlations pj k = pk j

as off-diagonal elements.

.2 Family contributions�i m

¼ b00Xi 0 þ b01Xi dependent student status þ b02Xi race

þb03Xi gender � � �b023Xi savings bonds þ ei m;

where m = race, . . ., savings bonds

Family contributionsi m = 1 if familycontributions�i m >0 and 0 otherwise

.3 Societal contributions�i m ¼ b00Xi 0

þb01Xi dependent student status þ b02Xi race

þb03Xi gender � � �b023Xi savings bonds þ em;

where m = race, . . ., savings bonds

Societal contributionsi m = 1 if societalcontributions�i m >0 and 0 otherwise

In the case of trivariate probit models in which there arehree error terms (M = 3) each distributed as multivariate


at both two-year and four-year college so it is notnecessary the availability that is at issue. However, AfricanAmerican students are far more likely to pay for collegewith student contributions at four-year colleges than theyare at two-year colleges which suggest that higher dollarscholarships and grants are needed. In contrast, Latino/Hispanic students receive no incentive in the case of two-year colleges and a disincentive in the case of four-yearcolleges. Therefore, we suggest that there is a need tocreate more grants and scholarships that target Latino/Hispanic students. With respect to income, while thelowest income students have garnered most of the mediaand research attention for good reasons, our findingsindicate that financial aid policies must pay closerattention to the opportunities they provide for moder-ate-income students to attend four-year colleges inparticular.

Another implication of this study is that parents’ collegeassets may serve as a means for reducing studentcontributions. However, among lower income and minor-ity families, with the exception of Asian students, there is agreat need to find ways to help more families contributetoward paying for college. Family contributions are amajor source of financing college for higher income, White,and Asian families and potentially a huge source ofinequality in the financial aid system. A problem thatlower income, African American and Latino/Hispanicfamilies face is that they have very little money to saveor to use for college after they pay all of their otherexpenses. So, while we find that college assets can help toreduce the burden of paying for college on students andincrease family contributions, lower income families areless likely to benefit from existing college savingsinstruments because of their low marginal tax rate (Maag& Fitzpatrick, 2004).3 These instruments are largelydesigned as tax subsidies. Examples of these instrumentsare state-sponsored savings plans, college investmentfunds, and education bonds.

Given this, lower income and minority families mayneed access to specially designed accounts called ChildSavings Accounts (CSAs), sometimes referred to as ChildDevelopment Accounts (CDAs). CSAs have been proposedas a potentially novel and promising mechanisms forhelping students and their families finance college(Boshara, 2003; Goldberg & Cohen, 2000; Sherraden,1991). An example of a CSA policy is the America Savingfor Personal Investment, Retirement, and Education(ASPIRE) Act. ASPIRE would create ‘‘KIDS Accounts,’’ or asavings account for every newborn, with an initial $500deposit, along with opportunities for financial education.4

Students living in households with incomes below thenational median would be eligible for an additionalcontribution of up to $500 at birth and a savings incentiveof $500 per year in matching funds for amounts saved in

3 The marginal tax rate is the rate on the last dollar of income earned.

This is different from the average tax rate, which is the total tax paid as a

percentage of total income earned.4 At this writing, the ASPIRE Act remains on the Congressional agenda

(http://www.newamerica.net/publications/policy/aspire_act_bill_sum-

mary).

accounts. When account holders turn 18, they would bepermitted to make tax-free withdrawals for costs associ-ated with post-secondary education, first-time homepurchase, and retirement security.

Acknowledgment

We would like to thank Dr. Paul Johnson from theCenter for Research, Methods, and Data Analysis (CRMDA)at the University of Kansas for his help in estimating thetrivariate probit model.

ormal, there are eight joint probabilities corresponding toight possible combinations of affirmative (student con-ributionsi m = 1) and negative (student contributionsi m = 0)utcomes. The joint probabilities are expressed in Eq. (A.4),sing the example where all outcomes are affirmative (i.e.,tudent contributionsi m = 1; family contributionsi m = 1; andocietal contributionsi m = 1):

A

esufoccab

A

A

A

tnetouss

http://www.newamerica.net/publications/policy/aspire_act_bill_summary

http://www.newamerica.net/publications/policy/aspire_act_bill_summary

A.4 PrðStudent contributions ½y1� ¼ 1; Family contributions ½y2� ¼ 1; Societal contributions ½y3� ¼ 1Þ¼ Prðe1 ¼ b01Xi dependent student status; e2 ¼ b02Xi race; e3 � b03Xi gender . . . e23 � b023Xi savings bondsÞ¼ Prðe23 ¼ b023Xi savings bonds j e22 <b022Xi real estate=stocks; e21 <b021Xi college investment fund; . . . ; e1 <b01Xi dependent student statusÞ� Prðe22 <b022Xi real estate=stocks j e21 <b021Xi college investment fund; . . . ; e1 <b01Xi dependent student statusÞ� Prðe1 <b01Xi dependent student statusÞ

A.5 The Cholesky decomposition of the variance/covariance matrix for the errors is expressed as follows:

E(ee0) = V = Cee0C, where

e1 = C1 1e1

e2 = C2 1e1 + C2 2e2

e3 = C3 1e1 + C3 3e3; and so forth until

e23 = C23 1e1 + C23 23e23

The trivariate normal probabilities of the three affirmative outcomes (i.e., student contributionsi m = 1; familycontributionsi m = 1; societal contributionsi m = 1) can then be expressed as:

A.6 Prðe1 ¼ b01Xi dependent student status; e2 ¼ b02Xi race; e3 ¼ b03Xi gender . . . b23 ¼ b023Xi savings bondsÞ¼ Pr½e23 � ðb023Xi savings bonds � C23 23e23 � C23 1e1Þ=C23 j e22 � ðb022Xi real estate=stocks

� C2222e22 � C22 1e1Þ=C22 j . . . e1 � b01Xi dependent student status=C1 1�� Pr½e22 � ðb022Xi real estate=stocks � C22 22e22 � C22 1e1Þ=C22 . . . e2 � ðb02Xi race � C2 1e1=C2 1Þ=C2 2je1 �b01Xi dependent student status=C1 1 � Pr½e1 ¼ b01Xi dependent student status=C1 1�

The standard normal covariates, 0, that appear in Eq. (A.5) are uncorrelated with each other.

Appendix B

Trivariate probit estimates of the probability a student pays for college with student, family, or societal contributions


b SE 95% CI b SE 95% CI b SE 95% CI

Dependent student S.228**** .044 S.313 S.143 S.115*** .044 S.201 S.028 S.186**** .044 .192 .480White (reference) 0 0 0Asian S.314**** .066 S.443 S.185 .189*** .062 .067 .310 .090 .064 S.128 .020Latino/Hispanic S.117* .062 S.238 .004 .016 .057 S.095 .128 .024 .061 S.358 S.151African American S.163** .070 S.300 S.025 S.064 .063 S.188 .060 .336**** .074 S.003 .065Male .133** .040 .055 .211 S.022 .036 S.093 .048 S.054 .038 S.194 .046Student – two-year college S.521*** .050 S.619 S.424 S.359**** .049 S.456 .262 S.254**** .053 S.172 .088Number of siblings .019 .018 S.016 .054 S.065**** .016 S.097 .033 .031* .017 S.318 S.019Head – high school or less (reference) 0 0 0Head – some college .051 .055 S.057 .159 .091* .052 S.011 .194 S.074 .061 S.660 S.341Head – four-year college degree or higher S.041 .059 S.156 .075 .211**** .056 .100 .322 S.042 .066 S.697 S.301Low-income ($0–20,000) (reference) 0 0 0Moderate-income ($20,001–50,000) .229**** .065 .101 .356 .207*** .065 .079 .335 S.169** .076 .187 .262Middle-income ($50,001–100,000) .228*** .073 .085 .372 .451**** .072 .310 .592 S.501**** .081 S.056 .182High-income ($100,001 or higher) S.109 .088 S.282 .063 .521**** .087 .351 .691 S.499**** .101 S.078 .035GPA S.045* .019 S.082 S.008 .048*** .018 .012 .083 .224**** .019 S.036 S.006Private school (by 10th grade) S.296**** .062 S.418 S.175 .008 .058 S.106 .123 .063 .061 .011 .066School climate S.020 .025 S.068 .028 S.008 .023 S.053 .037 S.022 .029 .016 .351Number of guidance counselors S.009 .009 S.027 .009 .002 .008 S.013 .018 S.021*** .008 S.242 .021Free/reduced lunch S.038** .015 S.067 S.008 S.028** .013 S.053 S.003 .039*** .014 S.184 S.005Student expects to graduate college S.070 .081 S.230 .089 .246*** .083 .082 .410 .184** .085 .268 .448Parent expects student to graduate college .139** .062 .018 .260 S.044 .060 S.162 .074 S.111* .067 S.233 .055Low college costs very important .010 .048 S.084 .103 S.079* .044 S.165 .007 S.094** .046 S.132 .064Financial aid very important .181**** .044 .093 .268 S.266**** .045 S.354 S.178 .358**** .046 S.042 .159Plan to remortgage home .241*** .073 .098 .384 .091 .071 .048 .230 S.089 .073 S.043 .271Start a savings account S.158*** .054 S.263 S.053 .035 .050 S.063 .132 S.034 .050 S.079 .147Have student put aside earnings .221**** .057 .109 .332 .112** .051 .011 .213 .058 .051 S.147 .053Start state-sponsored savings S.309**** .077 S.459 S.159 S.244** .076 S.393 S.095 .114 .080 S.149 .053College investment fund S.157*** .058 S.270 S.043 .144** .061 .024 .264 .034 .057 S.638 S.009Invest in real estate/stocks S.148*** .053 S.253 S.044 .045 .050 S.054 .144 S.047 .051 .192 .480Buy U.S. savings bonds .089 .054 S.017 .194 .125** .052 .024 .227 S.048 .052 S.128 .020Constant .990**** .152 .692 1.288 �.153 .154 �.454 .148 �.324* .160 �.358 �.151

Correlation coefficientsrho21 (Student – Family) .147**** .026 .095 .198rho31 (Student – Societal) �.085**** .028 �.139 �.031rho32 (Family – Societal) �.122**** .026 �.172 �.072Draws = 100; Log pseudolikelihood =�2,825,320.9; Wald x2 = 1935.54***; weighted n = 1,652,963; unweighted n = 7366


Notes: Estimates adjusted for clustering within schools. Data imputed using the Expectation Maximization (EM) algorithm.

* p< .10.

** p< .05.

*** p< .01.

**** p< .001.


Appendix C

Trivariate probit estimates of probability of paying for college with student, family, and societal contributions for students whoattended a two-year college



Dependent student S.190*** .070 S.327 S.052 .040 .078 S.112 .192 S.173** .077 S.324 S.023Asian S.318*** .122 S.558 S.078 .184 .133 S.078 .445 .154 .144 S.127 .436Latino/Hispanic S.188* .098 S.380 .004 .051 .102 S.149 .251 S.111 .102 S.311 .089African American S.354*** .114 S.578 S.130 S.218** .107 S.429 S.008 .469**** .117 .239 .699Male .262**** .072 .122 .403 .043 .066 S.087 .173 S.101 .069 S.236 .034Number of siblings S.008 .028 S.062 .047 S.069** .031 S.129 S.009 .044 .030 S.014 .102Head – some college S.060 .077 S.210 .090 .131 .082 S.030 .293 S.158* .091 S.336 .020Head – four-year college degree or higher S.047 .095 S.232 .139 .238** .095 .053 .423 S.113 .101 S.310 .084Moderate-income ($20,001–50,000) .118 .100 S.078 .313 .087 .105 S.120 .294 S.042 .112 S.262 .179Middle-income ($50,001–100,000) .252** .111 .033 .470 .328*** .118 .097 .559 S.357*** .116 S.586 S.129High-income ($100,001 or higher) .132 .166 S.195 .458 .418** .177 .071 .765 S.153 .199 S.542 .237GPA S.005 .028 S.060 .049 .020 .030 S.039 .079 .177**** .030 .118 .237Private school (by 10th grade) S.386*** .138 S.658 S.115 .045 .149 S.247 .337 .204 .134 S.058 .467School climate S.020 .037 S.092 .053 .003 .035 S.065 .071 S.031 .045 S.119 .056Number of guidance counselors .000 .015 S.029 .029 S.004 .014 S.031 .024 S.032** .015 S.061 S.004Free/reduced lunch S.073*** .026 S.123 S.023 S.002 .022 S.045 .041 .087**** .025 .039 .135Student expects to graduate college .059 .100 S.137 .256 .251** .109 .038 .464 .235** .104 .030 .439Parent expects student to graduate college .149* .083 S.014 .313 S.115 .083 S.278 .049 S.173** .088 S.344 S.001Low college costs very important .073 .081 S.085 .232 S.049 .069 S.185 .086 S.023 .078 S.176 .129Financial aid very important .051 .085 S.114 .217 S.258*** .084 S.422 S.093 .299**** .080 .141 .456Plan to remortgage home .284** .141 .008 .559 .164 .134 S.098 .425 S.042 .141 S.319 .236Start a savings account S.235** .102 S.435 S.035 S.133 .089 S.308 .041 S.005 .090 S.182 .172Have student put aside earnings .177* .104 S.027 .382 .202** .102 .002 .402 S.135 .099 S.329 .059Start state-sponsored savings S.290* .157 S.598 .018 S.292* .155 S.596 .012 .050 .156 S.255 .356College investment fund S.206* .118 S.436 .025 .057 .127 S.192 .306 S.023 .130 S.278 .231Invest in real estate/stocks S.082 .101 S.279 .116 S.082 .095 S.268 .105 .010 .108 S.201 .222Buy U.S. savings bonds .198* .106 S.010 .407 .286*** .104 .081 .490 .026 .105 S.181 .232Constant .454** .209 .044 .865 �.441* .243 �.916 .035 �.585** .226 �1.029 �.142

Correlation coefficientsrho21 (Student – Family) .235**** .041 .153 .314rho31 (Student – Societal) �.316**** .040 �.393 �.235rho32 (Family – Societal) �.201**** .042 �.283 �.116Draws = 100; Log pseudolikelihood =�944,651.44; Wald x2 = 446.85***; weighted n = 505,954; unweighted n = 2003.



* p< .10.

** p< .05.

*** p< .01.

**** p< .001.

Appendix D

Trivariate probit estimates: probability of paying for college with student, family, and societal contributions for students whoattended a four-year college.



Dependent student S.293**** .056 S.402 S.184 S.179**** .051 S.279 S.079 S.178*** .054 S.284 S.073Asian S.309**** .078 S.461 S.156 .188*** .071 .049 .327 .051 .076 S.097 .199Latino/Hispanic S.074 .079 S.229 .081 S.030 .071 S.169 .110 .106 .080 S.051 .263African American S.037 .090 S.214 .140 .020 .083 S.144 .183 .249*** .090 .072 .425Male .063 .047 S.029 .155 S.054 .044 S.140 .032 S.022 .046 S.111 .068Number of siblings .037 .023 S.009 .082 S.065*** .020 S.104 S.026 .025 .022 S.018 .067Head–some college .119 .077 S.032 .271 .072 .068 S.062 .206 S.008 .086 S.177 .162Head–four-year college degree or higher S.009 .074 S.154 .136 .174** .069 .037 .310 .005 .084 S.161 .171Moderate-income ($20,001–50,000) .285*** .097 .094 .475 .303**** .079 .149 .457 S.274*** .098 S.467 S.081Middle-income ($50,001–100,000) .195* .102 S.004 .394 .558**** .088 .385 .730 S.635**** .106 S.844 S.427High-income ($100,001 or higher) S.130 .117 S.359 .099 .607**** .103 .405 .808 S.675**** .123 S.917 S.433GPA S.064** .025 S.113 S.016 .056** .024 .008 .104 .251**** .024 .204 .299Private school (by 10th grade) S.267**** .067 S.398 S.136 S.011 .064 S.137 .115 .020 .066 S.109 .149School climate S.025 .032 S.089 .038 S.022 .031 S.083 .038 .007 .034 S.059 .073Number of guidance counselors S.013 .011 S.034 .008 .004 .011 S.017 .025 S.016* .009 S.034 .002Free/reduced lunch S.022 .018 S.056 .013 S.038** .016 S.069 S.006 .017 .015 S.013 .047Student expects to graduate college S.447*** .145 S.731 S.162 .271** .131 .014 .528 .090 .142 S.189 .369Parent expects student to graduate college .144 .098 S.047 .335 .027 .091 S.151 .204 .008 .096 S.180 .196Low college costs very important S.039 .060 S.157 .078 S.097* .054 S.203 .009 S.125** .059 S.240 S.009Financial aid very important .240**** .051 .140 .339 S.262**** .053 S.366 S.157 .383**** .055 .276 .490


Appendix D. (Continued )



Plan to remortgage home .231*** .083 .067 .394 .061 .085 S.106 .228 S.097 .083 S.261 .066Start a savings account S.129* .066 S.259 .001 .103* .061 S.018 .223 S.043 .063 S.167 .081Have student put aside earnings .267**** .068 .133 .400 .085 .063 S.038 .208 .134** .060 .016 .253Start state-sponsored savings S.330**** .087 S.501 S.159 S.232** .092 S.412 S.052 .146 .093 S.035 .328College investment fund S.134* .066 S.263 S.004 .168** .070 .030 .305 .047 .063 S.076 .170Invest in real estate/stocks S.175*** .063 S.298 S.052 .079 .061 S.040 .198 S.074 .060 S.191 .042Buy U.S. savings bonds .042 .063 S.082 .167 .067 .063 S.057 .190 S.086 .062 S.207 .036Constant 1.360**** .217 .935 1.784 �.300 .214 �.721 .120 �.363* .220 �.794 .069

Correlation coefficientsrho21 (Student – Family) .110**** .033 .046 .185rho31 (Student – Societal) .045 .034 �.022 .112rho32 (Family – Societal) �.077** .032 �.139 �.015Draws = 100; Log pseudolikelihood =�1,847,640; Wald x2 = 1091.42****; weighted n = 1,147,009; unweighted n = 5363



* p< .10.

** p< .05.

*** p< .01.

**** p< .001.


References

ACSFA. (2002). Empty promises: The myth of college access in America.Washington, DC: Advisory Committee on Student Financial Assistance.

ACSFA. (2006). Mortgaging our future: How financial barriers to college under-cut America’s global competitiveness. Washington, DC: Advisory Commit-tee on Student Financial Assistance.

ACSFA. (2010). The rising price of inequality: How inadequate grant aid limitscollege access and persistence. Washington, DC: Advisory Committee onStudent Financial Assistance. Retrieved from http://chronicle.com/items/biz/pdf/acsfa_rpi.pdf.

American Student Assistance. (2010). Approaching the tipping point: Theimplications of student loan debt and the need for education debt manage-ment. Washington, DC: American Student Assistance.

Archibald, R. B. (2002). Redesigning the financial aid system. Baltimore, MD:The Johns Hopkins University Press.

Baum, S., & Payea, K. (2011). Trends in student aid 2011. New York, NY: CollegeBoard.

Baum, S., & Schwartz, S. (1988). Merit aid to college students. Economicsof Education Review, 7(1), 127–134 http://dx.doi.org/10.1016/0272-7757(88)90077-5.

Berkner, L., Wei, C. C., He, S. L., Cominole, M., & Siegel, P. (2005). 2003–04National Postsecondary Student Aid Study (NPSAS:04): Undergraduatefinancial aid estimates for 2003–04 by type of institution (NCES 2005-163). Washington, DC: US Department of Education National Center forEducation Statistics.

Bettinger, E. (2004). How financial aid affects persistence. In C. Hoxby (Ed.),College choices: The economics of where to go, when to go, and how to payfor it. Chicago, IL: University of Chicago Press.

Boshara, R. (2003). Federal policy and asset building. Social DevelopmentIssues, 25(1&2), 130–141.

Cappellari, L., & Jenkins, S. P. (2003). Multivariate probit regression usingsimulated maximum likelihood. The Stata Journal, 3(3), 278–294.

Charles, C., Roscigno, V. J., & Torres, K. (2007). Racial inequality and collegeattendance: The mediating role of parental investments. Social ScienceResearch, 36, 329–352 http://dx.doi.org/10.1016/j.ssresearch.2006.02.004.

Choy, S. P., & Berker, A. M. (2003). How families of low- and middle-incomeundergraduates pay for college: Full-time dependent students in 1999–2000(NCES 2003-162). Washington, DC: US Department of Education NationalCenter for Education Statistics.

Choy, S. P., & Bobbitt, L. (2000). Low-income students: Who they are and howthey pay for education (NCES 2000-169). Washington, DC: US Departmentof Education National Center for Education Statistics.

College Board. (2009). Trends in student aid 2009. New York, NY: CollegeBoard.

Cook, T. D., Church, M. B., Ajanaku, S., Shadish, W. R. J., Kim, J., & Ran, et al.(1996). The development of occupational aspirations and expectationsamong inner-city boys. Child Development, 67, 3368–3385 http://dx.doi.org/10.2307/1131783.

Cunningham, A. F., & Santiago, D. A. (2008). Student aversion to borrowing:Who borrows and who doesn’t. Washington, DC: Institute for HigherEducation Policy and Excelencia in Education.

Dempster, A. P., Laird, N. M., & Rubin, D. B. (1977). Maximum likelihood fromincomplete data via the EM alogorithm. Journal of the Royal StatisticalSoceity, Series B: Methodological, 39(1), 1–38.

DesJardins, S. L., Ahlburg, D. A., & McCall, B. P. (2002). Simulating thelongitudinal effects of changes in financial aid on student departurefrom college. The Journal of Human Resources, 37(3), 653–679 http://dx.doi.org/10.2307/3069685.

Dynarski, M. (1994). Who defaults on student loans? Findings fromthe National Postsecondary Student Aid Study. Economics of EducationReview, 13(1), 55–68.

Dynarski, S. (2003). Loans, liquidity and school decisions. Cambridge, MA:National Bureau of Economic Research.

Elliott, W. (2009). Children’s college aspirations and expectations: Thepotential role of college development accounts (CDAs). Children andYouth Services Review, 31(2), 274–283.

Elliott, W., & Beverly, S. (2011). Staying on course: The effects of assets andsavings on the college progress of young adults. American Journal ofEducation, 117(3), 343–374.

Elliott, W., Destin, M., & Friedline, T. (2011). Taking stock of ten years ofresearch on the relationship between assets and children’s educationaloutcomes: Implications for theory, policy and intervention. Children andYouth Services Review, 33(11), 2312–2328.

Ensuring Continued Access to Student Loans Act, 110 U.S.C. § PL 110-227(2008).

Goldberg, F., & Cohen, J. (2000, September). The universal piggy bank: Design-ing and implementing a system of savings accounts for children. Paperpresented at the Inclusion in Asset Building: Research and Policy Sym-posium, Center for Social Development, Washington University in St.Louis, MO. Retrieved from http://csd.wustl.edu/Publications/Documents/60.TheUniversalPiggyBank.pdf

Hansen, L. W. (1983). Impact of student financial aid on access. TheAcademy of Political Science, 35(2), 84–96 http://dx.doi.org/10.2307/3700892.

Hanushek, E. A., Leung, C. K. Y., & Yilmaz, K. (2004). Borrowing constraints,college aid and intergenerational mobility (NBER Working Paper No.10711). Cambridge, MA: National Bureau of Economic Research.

Health Care and Education Reconciliation Act, 111 U.S.C. § PL 111-152(2010).

Heller, D. E. (2002). State merit scholarship programs: An introduction. In D.E. Heller & P. Marin (Eds.), Who should we help? The negative socialconsequences of merit scholarships (pp. 15–23). Cambridge, MA: The CivilRights Project at Harvard University.

Heller, D. E. (2004). State merit scholarship programs: An introduction.In D. E. Heller & P. Marin (Eds.), State merit scholarship program andracial inequality. Cambridge, MA: The Civil Rights Project at HarvardUniversity.

Heller, D. E., & Rogers, K. R. (2006). Shifting the burden: Public and privatefinancing of higher education in the United States and implications forEurope. Tertiary Education and Management, 12(2), 91–117 http://dx.doi.org/10.1080/13583883.2006.9967162.

Higher Education Act, 102 U.S.C. § PL 102-325 (1992).Huang, J., Beverly, S., Clancy, M., Lassar, T., & Sherraden, M. (2011).

Early enrollment in a statewide child development account program

http://chronicle.com/items/biz/pdf/acsfa_rpi.pdf

http://chronicle.com/items/biz/pdf/acsfa_rpi.pdf

http://dx.doi.org/10.1016/0272-7757(88)90077-5

http://dx.doi.org/10.1016/0272-7757(88)90077-5

http://dx.doi.org/10.1016/j.ssresearch.2006. 02.004

http://dx.doi.org/10.1016/j.ssresearch.2006. 02.004

http://dx.doi.org/10.2307/1131783

http://dx.doi.org/10.2307/1131783

http://dx.doi.org/10.2307/3069685

http://dx.doi.org/10.2307/3069685

http://csd.wustl.edu/Publications/Documents/60.TheUniversalPiggyBank.pdf

http://csd.wustl.edu/Publications/Documents/60.TheUniversalPiggyBank.pdf

http://dx.doi.org/10.2307/3700892

http://dx.doi.org/10.2307/3700892

http://dx.doi.org/10.1080/13583883.2006.9967162

http://dx.doi.org/10.1080/13583883.2006.9967162


(CSD Working Paper 11-23). St. Louis, MO Washington University, Centerfor Social Development.

Ingels, S. J., Pratt, D. J., Wilson, D., Burns, L. J., Currivan, D., Rogers, J. E., et al.(2007). Education longitudinal study of 2002 (ELS: 2002) base-year tosecond follow-up data file documentation (including field test report).Washington, DC: National Center for Education Statistics.

Kao, G., & Thompson, J. S. (2003). Racial and ethnic stratification in educa-tional achievement and attainment. Annual Review of Sociology, 29,417–442 http://dx.doi.org/10.1146/annurev.soc.29.010202.100019.

Kim, D. (2007). The effect of loans on students’ degree attainment: Differ-ences by student and institutional characteristics. Harvard EducationalReview, 77(1), 64–100.

King, J. E. (1999). Money matters: The impact of race and gender on howstudents pay for college. Washington, DC: American Council on Education.

Little, R. J. A., & Rubin, D. B. (1987). Statistical analysis with missing data. NY:Wiley.

Long, B. T., & Kurlaender, M. (2008). Do community colleges provide a viablepathway to a baccalaureate degree? Educational Evaluation and PolicyAnalysis, 31(1), 30–53 http://dx.doi.org/10.3102/0162373708327756.

Louie, V. (2007). Who makes the transition to college? Why we should care,what we know, and what we need to do. Teachers College Record, 109(10),2222–2251.

Maag, E. M., & Fitzpatrick, K. (2004). Federal financial aid for higher education:Programs and prospects. Washington, DC: The Urban Institute. Retrievedfrom http://www.urban.org/url.cfm?ID=410996.

Marin, P. (2002). Merit scholarships and the outlook for equal opportunity inhigher education. In D. E. Heller & P. Marin (Eds.), Who should we help?The negative social consequences of merit scholarships. Cambridge, MA:The Civil Rights Project at Harvard University.

Marjoribanks, K. (1984). Ethnicity, family environment and adolescents’aspirations: A follow-up study. Journal of Educational Research, 77(3),166–171.

Mau, W. C. (1995). Educational planning and academic achievement ofmiddle school students: A racial and cultural comparison. Journal ofCounseling & Development, 73(5), 518–526.

Mau, W. C., & Bikos, L. H. (2000). Educational and vocational aspirations ofminority and female students: A longitudinal study. Journal of Counseling& Development, 78(2), 186–194.

McIntosh, M. F., & Rouse, C. E. (2009). The other college: Retention andcompletion rates among two-year college students. Washington, DC: Cen-ter for American Progress.

Mickelson, R. A. (1990). The attitude–achievement paradox among blackadolescents. Sociology of Education, 63(1), 44–61 http://dx.doi.org/10.2307/2112896.

Middle Income Student Assistance Act, 95 U.S.C. § PL 95-566 (1978).O’Connor, N., Hammack, F. M., & Scott, M. A. (2010). Social capital, financial

knowledge, and Hispanic student college choices. Research in HigherEducation, 51, 195–219.

Omnibus Budget Reconciliation Act, 103 U.S.C. § PL 103-66 (1993).Perna, L. W. (2008). Understanding high school students’ willingness

to borrow to pay college prices. Research in Higher Education, 49,589–606.

Perna, L., Cooper, M., & Li, C. (2007). Improving educational opportunitiesfor student who work. In St. John, E. P. (Ed.). Confronting educationalinequality: Reframing, building understandings, and making change.Readings on equal education (Vol. 22, pp. 109–110). New York, NY:AMS Press.

Perna, L. W., & Titus, M. A. (2005). The relationship between parentalinvolvement as social capital and college enrollment: An examinationof racial/ethnic group differences. Journal of Higher Education, 76,486–518.

Presley, J. B., & Clery, S. B. (2001). Middle income undergraduates: Where theyenroll and how they pay for their education. Washington, DC: US Depart-ment of Education National Center for Education Statistics.

Sallie Mae and Gallup. (2011). How America pays for college 2011. Newark,DE: Sallie Mae. Retrieved from https://www1.SallieMae.com/about/news_info/research/how_america_pays_2011/

Santiago, D. A., & Cunningham, A. F. (2005). How Latino students pay forcollege: Patterns of financial aid in 2003–04. Washington, DC: Excelenciain Education, Institute for Higher Education Policy, and USA Funds.

Schwartz, S., & Finnie, R. (2002). Student loans in Canada: An analysisof borrowing and repayment. Economics of Education Review, 21(5),497–512 http://dx.doi.org/10.1016/S0272-7757(01)00041-3.

Sherraden, M. (1991). Assets and the poor. Armonk, New York: ME Sharpe Inc.Stampen, J. O., & Cabrera, A. F. (1988). The targeting and packaging of student

aid and its effect on attrition. Economics of Education Review, 7(1), 29–46http://dx.doi.org/10.1016/0272-7757(88)90070-2.

Stanton-Salazar, R. D. (1997). A social capital framework for understandingthe socialization of racial minority children and youth. Harvard Educa-tional Review, 67(1), 1–40.

Trusty, J. (2000). High educational expectations and low achievement:Stability of educational goals across adolescence. Journal of EducationalResearch, 93, 356–395.

Volkwein, J. F., & Szelest, B. P. (1995). Individual and campus characteristicsassociated with student loan default. Research in Higher Education, 36(1),41–72 http://dx.doi.org/10.1007/BF02207766.

Volkwein, J. F., Szelest, B. P., Cabrera, A. F., & Napierski-Prancl, M. (1998).Factors associated with student loan default among different racial andethnic groups. The Journal of Higher Education, 69(2), 206–237 http://dx.doi.org/10.2307/2649206.

Waddell, G., & Singell, L., Jr. (2011). Do no-loan policies change the matricu-lation patterns of low-income students? Economics of Education Review,30(2), 203–214 http://dx.doi.org/10.1016/j.econedurev.2010.10.004.

Woo, J., & Choy, H. S. P. (2011). Merit aid for undergraduates: Trends from1995–96 to 2007–08 (NCES 2012-160). US Department of EducationNational Center for Education Statistics.

http://dx.doi.org/10.1146/annurev.soc.29.010202.100019

http://dx.doi.org/10.3102/0162373708327756

http://www.urban.org/url.cfm?ID=410996

http://dx.doi.org/10.2307/2112896

http://dx.doi.org/10.2307/2112896

https://www1.salliemae.com/about/news_info/research/how_america_pays_2011/

https://www1.salliemae.com/about/news_info/research/how_america_pays_2011/

http://dx.doi.org/10.1016/S0272-7757(01)00041-3

http://dx.doi.org/10.1016/0272-7757(88)90070-2

http://dx.doi.org/10.1007/BF02207766

http://dx.doi.org/10.2307/2649206

http://dx.doi.org/10.2307/2649206


Date post:	27-Jan-2017
Category:	Documents
Upload:	terri
View:	213 times
Download:	0 times