Student debt and hardship: Evidence from a large sample of low- and moderate-income households

Children and Youth Services Review 70 (2016) 8–18

Contents lists available at ScienceDirect

Children and Youth Services Review journal homepage: www.elsevier.com/locate/childyouth

Student debt and hardship: Evidence from a large sample of low- and moderate-income households Mathieu R. Despard a,⁎, Dana Perantie b,1, Samuel Taylor b,1, Michal Grinstein-Weiss b,1, Terri Friedline c,2, Ramesh Raghavan b,3 a b c

University of Michigan, USA Washington University in St. Louis, USA University of Kansas, USA

a r t i c l e

i n f o

Article history: Received 28 April 2016 Received in revised form 1 September 2016 Accepted 1 September 2016 Available online 05 September 2016 Keywords: Higher education College financing Student debt Student loans Student loan repayment Low-income students Material hardship Financial difficulty

a b s t r a c t Student debt has risen in recent years as higher education costs have shifted to students and their families, particularly those with low-to-moderate incomes (LMI). Though a college degree continues to convey higher earnings, those who finance their degrees have lower net worth and greater financial difficulties than persons without student debt. We assess the relationship between student debt and material and health care hardship among a large sample (n = 5558) of LMI tax filers, using propensity score analysis to adjust for self-selection into student debt status and loan amount and monthly payment quartiles. We find that participants with student debt have a higher likelihood of hardship. Loan amounts only partially predict hardship, and borrowers making current loan payments are at lower odds for hardship than non-payers. We also find that among those with student debt, non-payers and those without college degrees have much greater social and economic disadvantages. © 2016 Elsevier Ltd. All rights reserved.

1. Introduction Evidence consistently demonstrates that higher education is a wise investment, generating myriad economic and social returns for graduates (e.g., Dee, 2004; Carneiro, Heckman, & Vytlacil, 2010; Hérault & Zakirova, 2015; Hout, 2012; Perna, 2003). Yet as the costs of higher education in the US have grown, so too have concerns about rising student loan debt. Almost 70% of US college students borrow to finance their degree, amounting to an average debt burden of $28,950 for four-year graduates (Institute for College Access & Success, 2015). Cumulative student debt is estimated at over $1.2 trillion, exceeding credit cards as the largest form of consumer debt in the US (Chopra, 2013). The

⁎ Corresponding author. School of Social Work, 1080 S. University Avenue, Ann Arbor, MI 48109, USA. E-mail addresses: [email protected] (M.R. Despard), [email protected] (D. Perantie), [email protected] (S. Taylor), [email protected] (M. Grinstein-Weiss), [email protected] (T. Friedline), [email protected] (R. Raghavan). 1 George Warren Brown School of Social Work, One Brookings Drive, Campus Box 1196, St. Louis, MO, 63130, USA. 2 School of Social Welfare, 1545 Lilac Lane, Lawrence, KS 66045-3129, USA. 3 School of Social Work, Rutgers, The State University of New Jersey, 536 George Street New Brunswick, New Jersey 08901.

http://dx.doi.org/10.1016/j.childyouth.2016.09.001 0190-7409/© 2016 Elsevier Ltd. All rights reserved.

proportion of households with student debt rose from 9% in 1989 to 19% in 2010, and the proportion of households with student debt totaling $25,000 or greater increased by 24 percentage points (Bricker & Thompson, 2016). Recent policy proposals and actions such as free tuition at public colleges and universities, early Pell Grants (U.S. Department of Education, 2015a), and the Obama Administration's expansion of income-based student loan repayment plans reflect policymakers' concerns rising student loan debt. These historic increases in student loan borrowing and debt can be attributed to several factors. More students are attending college and staying in college for longer periods than previously (Lee, van der Klaauw, Haughwout, Brown, & Scally, 2014). Notably, college costs have outpaced inflation, prompting greater borrowing. From the 2000–2001 to 2015–2016 academic years, tuition, fees, and housing costs at four-year institutions rose 67% and 43% at private and public nonprofit universities, respectively (The College Board, 2015). Concurrently, state-level disinvestment in higher education has prompted institutions to reduce need-based forms of financial aid and shift the burden of costs to students and their families (Best & Keppo, 2014). As cost burdens have shifted, lawmakers have increased access to student loans for a larger share of students, particularly those from low- and moderate-income (LMI) households (Elliott & Friedline, 2013). Recent evidence has uncovered possible consequences of borrowing to pay for college. Households with student debt obligations fare worse

M.R. Despard et al. / Children and Youth Services Review 70 (2016) 8–18

on measures of assets and net worth compared to non-indebted counterparts (Elliott & Nam, 2013). Moreover, rising borrowing rates may be leaving borrowers with unmanageable debt. Student debt-to-income ratios have risen steadily from 12% in 1989 to 32% in 2010 (Bricker & Thompson, 2016), signaling possible repayment troubles for borrowers and graduates. Indeed, recent figures have pointed to increases in loan delinquency and default for many borrowers. Student loan delinquency rates increased from 10% in 2004 to 17% in 2012. The rate of severe delinquency or default increased from 10% for cohorts in 2005–06 to 15% among cohorts in 2007–10 (Brown, Haughwout, Lee, Scally, & van der Klaauw, 2015). Nationally, federal loan default rates began steadily rising through the mid-to-late 2000s (Hillman, 2014; U.S. Department of Education, 2011), reaching a high of 15% in 2013 before receding slightly to 12% in 2015 (U.S. Department of Education, 2015b). Certain subgroups of borrowers may be uniquely burdened by student debt and repayment obligations. Odds of student loan default are greater among Black and Hispanic borrowers, borrowers from low-income families, and borrowers who have dependents (Hillman, 2014; Lochner, Stinebrickner, & Suleymanoglu, 2013). Huelsman (2015) found significant disparities in student debt amounts between Black and White students at most 4-year institutions. Among LMI households with incomes under $30,000, Black borrowers accrued $7721 more in student debt than White borrowers (Grinstein-Weiss, Perantie, Taylor, & Raghavan, 2016). While higher education reliably predicts strong lifetime returns, the evidence reviewed above suggests that financing a degree may carry risks for some borrowers and may make it difficult for households to meet basic needs, save, and build assets while in repayment. However, little research has explored the unique associations between student loan borrowing and hardship among LMI borrowers, who are more likely to have trouble repaying loans. The purpose of this study is to examine the association between student debt and material, health, and financial hardship among LMI borrowers. We examine whether having student debt, and the amounts of student debt and monthly debt payments increase the odds of experiencing hardship – difficulty meeting needs for food, shelter, medical care and other basic needs (Beverly, 2001; Nelson, 2011; Short, 2005). Concurrent to shifting costs to students and rising student loan debt has been a push to increase college attendance among LMI students (Executive Office of the President, 2014). However, the higher education literature has paid little attention to LMI student loan borrowers as a fast-growing segment of the student loan market (The Federal Reserve Bank of New York, 2015). LMI students borrow at higher rates (Kim, 2007; Houle, 2014) and assume greater student loan (Price, 2004) and other debt (Soria, Weiner, & Lu, 2014) burdens than higher-income borrowers. Greater student loan usage and debt burden may constrain the ability of LMI borrowers to meet basic needs. To the authors' knowledge, this is the first empirical study examining the association of student debt with defined hardship measures among an LMI sample. 2. Literature and background 2.1. Student debt and financial difficulty The existing literature suggests student debt may be associated with financial difficulty. Examining data from National Student Loan Surveys from 1998 to 2003, Baum and O'Malley (2003) found that LMI borrowers had greater difficulty repaying student loans compared to other borrowers. Analyzing data from the Survey of Consumer Finances (SCF), Bricker and Thompson (2016) found that households with student debt were 4 percentage points more likely to be 60 days late on bill payments and 18% more likely to have been denied credit, or feared credit denial, than those without student debt. As student debt levels increased, the likelihood of experiencing financial difficulty also

9

increased. Interestingly, these findings did not persist when researchers looked at other types of consumer debt. Indeed, student loan debt appeared to be a primary source of financial burden for households under investigation. But Thompson and Bricker's results were modest in size and even weakened when isolated to households with indebted non-graduates. Furthermore, the SCF data used in this study originated from households during the Great Recession, complicating the ability to isolate the effects of student debt over other economic forces destabilizing many households at the time. Student debt is also associated with increased odds of bankruptcy for some borrowers. Using SCF data over a longer time period (1995 to 2010), Gicheva and Thompson (2015) found that as the amount of student debt increases, the likelihood of declaring bankruptcy increases, even after controlling for income, predicted earnings, and other demographic factors. The strength of the relationship between student debt and bankruptcy was greater for households with at least one borrower who did not complete their degree and decreases, but was still statistically significant, when controlling for economic condition in their models (aggregate unemployment and bankruptcy rates). In contrast to other research, student debt amount was unrelated to late bill payments or credit denials (Gicheva & Thompson, 2015). Despite wage and earnings premiums long associated with earning a college degree (Greenstone & Looney, 2012; Hershbein, Harris, & Kearney, 2014), student debt may constrain graduates' investment choices and inhibit the accumulation of assets (Gicheva & Thompson, 2015). College-educated households without student debt have seven times the net worth of similar households with student debt (Fry, 2014). Retirement savings are 52% higher for non-indebted households than indebted ones (Elliott, Grinstein-Weiss, & Nam, 2013). Because students borrow against future earnings, higher borrowing rates during college reduce the availability of discretionary income to build wealth post-college (Elliott & Lewis, 2015). This may be particularly true for recent, early-career graduates who are repaying debt while earnings are lower (Hershbein et al., 2014). Conventional life-cycle vehicles through which households accumulate assets may also be affected by student debt. College graduates with large student debt levels had significantly lower odds of purchasing a home than those without outstanding debt (Brown & Caldwell, 2013; Brown et al., 2015; Gicheva & Thompson, 2015; Shand, 2007), which may be due to reluctance to assume more debt (Houle & Berger, 2015). These divergent rates of homeownership amount to vastly different short- and long-term wealth profiles. Compared to homeowners without student debt, indebted homeowners are estimated to have $70,000 less in home equity (Hiltonsmith, 2013). 2.2. Student debt and health Research examining the independent effects of student debt on health is limited. However, a small body of evidence has demonstrated that the burden and stress associated with student debt may have adverse mental or physical health impacts. Nelson, Lust, Story, and Ehlinger (2008) found that credit card debt of at least $1000 was associated with several health risks including obesity, overeating, substance abuse, and lack of physical activity among undergraduate and graduate students attending a public university. Among a sample of university students in the United Kingdom, attitudes toward one's debt were associated with worse mental health (Cooke, Barkham, Audin, Bradley, & Davy, 2004). Walsemann, Ailshire, and Gee (2016) found that higher levels of student debt among Black young adults were associated with fewer hours of sleep, though no such relationship was found among White and Latino young adults. Evidence concerning the relationship between other forms of consumer debt and health is more robust. Household financial debt is associated with poor health and mental health outcomes (Sweet, Nandi, Adam, & McDade, 2013). Based on a meta-analysis of 65 studies,

10

M.R. Despard et al. / Children and Youth Services Review 70 (2016) 8–18

Richardson, Elliott, and Roberts (2013) found that increases in personal debt were associated with greater odds of mental health problems, substance abuse, and suicide. Similarly, Fitch, Hamilton, Bassett, and Davey (2011) found that indebtedness or increasing debt levels were associated with poorer mental health based on a systematic review of 50 studies. 2.3. Study purpose and research questions The evidence reviewed above indicates that LMI borrowers experience repayment and other financial difficulties, student debt may constrain saving and investment, and is associated with health and mental health problems. However, prior research has not specifically addressed the relationship between student debt and specific types of hardship among a sample of LMI borrowers, such as difficulty paying rent or affording needed medical care. The current study aims to assess the relationship between student debt and material, health care hardship, and financial difficulty among a large sample of LMI households. Specifically, our research questions are: 1. Does having student debt predict greater odds of experiencing material and health care hardship and financial difficulty? 2. Is there a relationship between the amount of student debt and material and health care hardship and financial difficulty among student loan borrowers? 3. Among households paying back their student loans, is there a relationship between the amount of monthly debt payments and material and health care hardship and financial difficulty? 4. Do hardship and financial difficulty experiences among borrowers differ based on repayment and college completion status? Answering these questions is important because college financing for LMI students has shifted in recent years from need-based aid (Best & Keppo, 2014) to loans (Elliott & Friedline, 2013). Consequently, increasing student debt burden may make it difficult for LMI households to pay for basic needs like housing, food, and medical care (Baum & Schwartz, 2006). This study contributes to the literature on student debt and material hardship in four ways. First, using a sample of LMI households, we focus on a population that has not been well studied, though some evidence indicates LMI households experience repayment difficulty. Second, in addition to financial difficulty, we use multiple indicators of hardship. Third, we incorporate important household-level covariates such as financial shocks, health insurance coverage, and inability to access funds in an emergency to better understand the unique relationship between student debt and hardship. Finally, our study contributes to the literature by using propensity score analysis to adjust for observed factors associated with borrowing for higher education, such as age, income and gender. This method reduces selection bias that might affect the relationship between student debt and hardship and financial difficulty. Our findings can inform public policies aimed at promoting greater access to and affordability of college education among LMI students. 3. Methods Data for this study come from the Refund to Savings (R2S) experiment. R2S is a large-scale, multiyear research project assessing whether behavioral economics techniques encourage low-income tax filers using Intuit's TurboTax Freedom Edition (TTFE) to deposit all or part of their refunds into savings accounts or U.S. Savings Bonds. As part of the Internal Revenue Service's (IRS) Free File Alliance Program, TTFE is free for use by tax filers who meet certain income or military service criteria. In 2013, the criteria for qualifying to use TurboTax Freedom Edition were: (a) household gross income less than $31,000, (b)

claiming the Earned Income Tax Credit, and/or (c) active duty military serviceperson in a household with gross income less than $57,000 (Grinstein-Weiss et al., 2015). 3.1. Sample and design The 2013 R2S sample includes 680,545 individuals who submitted their 2012 federal income tax returns using TurboTax Freedom Edition (TTFE), an online tax-preparation software program, and expected to receive a refund. From the R2S sample, 20,813 and 8251 individuals completed a household financial survey upon filing their tax returns (HFS1) and six months later (HFS2), respectively. This survey gathered detailed information on financial status, perceptions, and behaviors. Individuals were included in this study if they completed both waves of the HFS (n = 8251). Participants were excluded from the study if they were younger than 18 years of age (n = 78)4 and/or had a high school diploma or less (n = 1099), indicating they had not attended college and could not have incurred any student debt. We also excluded participants who said there were three or more adults living in the household (n = 1461) because participants might not know or accurately report student debt for other adults, such as roommates. Lastly, we excluded cases with missing data for having student debt (n = 31 or b1%) and/ or for amount of student debt (n = 24 or b1%). These exclusions produced an analytical sample of 5558 participants (see Table 1 for sample characteristics). We used an observational study design to examine the relationship between student debt and material and health care hardship and financial difficulty. Having student debt is a result of a choice to pursue postsecondary education and to finance this education with student loans; participants may systematically differ on observed characteristics related to student debt such as age and gender that might also explain outcomes. Therefore, we use propensity score analysis to reduce this source of self-selection bias and estimate the relationship between student debt and outcomes (Stuart, 2010). 3.2. Measures Measures used in this study come from administrative tax data from TTFE filers and from two waves of a household financial survey. Immediately after filing their taxes, all TTFE users anticipating refunds were invited to take a household financial survey (HFS1). The survey included a detailed assessment of participant demographic characteristics, and financial circumstances, behavior, and experiences. Six months after filing, the survey participants were invited to take another survey (HFS2), which included similar questions as HFS1 and added questions concerning financial shocks that occurred during the six months since filing. With the informed consent of survey participants concerning the release of tax information under Internal Revenue Code Section 7216, we merged administrative tax data with HFS1 and HFS2 data. 3.2.1. Dependent variables We examined the association between student debt and material and health care hardship. We defined these hardships as experiencing difficulties in meeting basic needs in the household (Beverly, 2001; Nelson, 2011; Short, 2005). Construction of hardship variables were informed by similar items used in large panel studies such as the Survey of Income and Program Participation (SIPP) and the Fragile Families and Child Well-Being Study. A dummy variable for material hardship was created and coded as ‘1’ if the participant indicated they had experienced any of the following hardships in the six months after filing taxes: unable to make a full rent or mortgage payment, skipped a bill 4 Participants were required to acknowledge they were over 18 in order to consent to participate, yet some TTFE users younger than 18 ignored this requirement and thus were excluded from the study.

M.R. Despard et al. / Children and Youth Services Review 70 (2016) 8–18 Table 1 Sample characteristics.

Sample size (n) Outcomes Material hardship Health care hardship Financial difficulty Covariates Race/ethnicity White Black Asian Hispanic Other Gender Female Male Mean age (sd) Filing status Single Head of household Married filing jointly/widow Married filing separately Number of adults in household One Two Has dependents No Yes School enrollment status Not currently enrolled Currently enrolled Employment status Full-time Part-time Not working Education Some college College degree Some graduate/professional school Graduate or professional degree Mean gross income (sd)

Full sample (%)

No student debt (%)

Student debt (%)

5558

38.84

61.16

60.33 51.48 30.60

65.85 54.40 35.24

51.62 46.87 23.29

76.06 10.65 2.59 6.93 3.78

41.49 22.24 37.76 36.29 37.32

58.51 77.76 62.24 63.71 62.68

64.69 35.31 35.91 (12.65)

35.90 44.14 41.25 (14.89)

64.10 55.86 32.35 (9.47)

61.25 21.34 16.46 0.95

39.22 33.14 44.81 39.62

60.78 66.86 55.19 60.38

46.99 53.01

41.56 36.44

58.44 63.56

64.72 35.28

40.91 35.05

59.09 64.95

69.20 30.80

45.38 23.93

54.62 76.07

51.13 18.97 29.90

37.51 41.60 39.28

62.49 58.40 60.72

39.90 33.25 11.98 14.87 $18,854 ($10,095)

47.88 35.23 28.57 30.91 $19,224 ($10,343)

52.12 64.77 71.43 69.09 $18,619 ($9929)

or made a late payment, or could not afford the type or amount of food desired. If the participant did not indicate experiencing any of these events, material hardship was coded ‘0’. Each material hardship was also used as a dependent variable in analyses. A dummy variable for health care hardship was created and coded as ‘1’ if the participant indicated they could not afford to see a doctor or go to a hospital for medical care, to see a dentist, or to fill a prescription in the six months after filing taxes. If the participant did not indicate experiencing any of these events, health care hardship was coded ‘0’. Each health care hardship was also used as a dependent variable in analyses. We also examined the relationship between student debt and financial difficulty. A dummy variable for financial difficulty was created and coded as ‘1’ if the participant indicated they experienced a bank overdraft, or had a credit card declined for being over the credit limit in the six months after filing taxes. If the participant did not indicate experiencing any of these events, financial difficulty was coded ‘0’. Each financial difficulty was also used as a dependent variable in analyses. Material hardship, health care hardship, and financial difficulty were all measured with data from HFS2, the second observational wave conducted six months after filing taxes. We drew a conceptual distinction between hardship and difficulty. With material or health care hardship, a basic need may go unfulfilled. With financial difficulty, such as a bank overdraft, an individual

11

experiences adversity in managing resources which may be stressful and may render meeting basic needs more difficult, but is not directly indicative of lack of need fulfillment. 3.2.2. Independent variables The key predictor variables of interest were the presence and amounts of student debt and monthly debt payments measured at the household level. Responses for student debt were coded as ‘1’ if the participant indicated anyone in their household currently had an education or school loan(s) and ‘0’ otherwise. For debt amount, participants were asked “About how much does your household owe on education or school loans?” Missing responses to debt amount (3%) were replaced with responses to a follow-up question that allowed participants to estimate the debt amount by selecting a range (e.g., $2000 to $5000). The replacement value was the mid-point of the selected range (e.g., $3500 for the $2000 to $5000 response option). For the highest range (“more than $50,000”), the replacement value was $200,000, the midpoint between $50,000 and the 99th percentile. For monthly debt payments, participants were asked, “In a typical month, how much does your household pay in student loan debt?” Missing responses were replaced with midpoint categorical responses (e.g., $150 for a response of “$100 to $200”). For both student debt amounts and monthly student debt payment amounts, data were winsorized, recoding extreme outliers to the 99th percentile value as a form of data censoring. 3.2.3. Covariates and conditioning variables Several additional demographic and financial variables from HFS1 and HFS2 served as conditioning variables for propensity score analysis and/or as covariates in multivariate models: age, gender, race/ethnicity, educational attainment, current student status (enrolled or not), employment status, number of adults in the household, amount of unsecured debt (excluding student debt), amount of liquid financial assets (reported amount in checking accounts, savings accounts, and cash), number of financial shocks experienced in the six months after filing taxes, ability to access $2000 in an emergency (yes/no), health insurance status (insured or uninsured), and careful budgeting habits (measured on a five-point likert scale from “not at all like me” to “very much like me”). Both unsecured debt and liquid financial assets were winsorized to censor extreme outliers. In addition, filing status, gross income, and whether the participant claimed any dependents when they filed their taxes were derived from TTFE administrative data. All covariates and conditioning variables were measured at baseline (HFS1), except for financial shocks, which were measured retrospectively at six-month follow-up (HFS2), i.e., participants were asked whether they or another member of the household experienced any of the following in the six months since filing their taxes: a period of unemployment, an emergency room visit or hospitalization, major car repair, and legal fees or expenses. 3.3. Data analysis We used propensity score analysis to estimate the probability – conditional on observed covariates – that a participant received treatment (Rosenbaum & Rubin, 1983), which we defined as having student debt. Propensity score analysis reduces selection bias in estimating outcomes and increases treatment-control group balance, though this method does not address unobserved differences between treatment and control group participants that may affect outcomes. We used an inverse probability of treatment weighting procedure (IPTW) (Hirano & Imbens, 2001) to construct average treatment effect (ATE) estimates. With IPTW, a participant's weight is the inverse of the probability of receiving the treatment she or he actually received (Austin, 2011). Use of IPTW has been found to produce unbiased estimates of proportional differences in binary outcomes in observational studies (Austin, 2010, 2013). An ATE estimator is appropriate to assess

12

M.R. Despard et al. / Children and Youth Services Review 70 (2016) 8–18

the effect of a treatment offered to a population (Winship & Morgan, 1999). In the current study context, student debt is a treatment offered to the population of LMI tax filers. We used an average treatment effect (ATE) estimator, which is defined as:

ATE ¼ Eðŷ1 ǀw ¼ 1Þ−Eðŷ0 ǀw ¼ 0Þ

where ŷ1 and ŷ0 denote outcomes for the treated (participants with student debt) and comparison (participants without student debt) groups, respectively, and w as the binary treatment condition. Propensity score weights to estimate ATE were calculated using the formula 1=eðxi Þ for a treated participant (with student debt) and ½1=ð1−eðxi ÞÞ for a comparison participant (without student debt) and were used in subsequent logistic regression models. To implement IPTW, we assessed sample balance between treated and comparison group participants on demographic variables that may explain self-selection into student debt status and debt payment amounts. Conditioning variables were selected if there were statistically significant bivariate differences indicating sample imbalance. For student debt status, statistically significant imbalance was found for age, gender, race/ethnicity, current student status, number of adults in the household, educational attainment, and any dependents in the household. We ran a logistic regression with these conditioning variables to estimate propensity scores concerning the probability of having student debt. Thus, a participant without student debt who has an estimated high probability to have student debt is weighted relatively more than a participant without student debt who has an estimated low probability to have student debt. Boxplots demonstrated sufficient overlap of propensity scores between participants with and without student debt. No statistically significant differences remained between participants with and without student debt on conditioning variables following the propensity score adjustment using the ATE estimator. We also examined total student debt and monthly payment amounts as predictors of hardship. We implemented IPTW with multinomial logistic regression to estimate generalized propensity scores using 1=eðxi Þ as sampling weights (Guo & Fraser, 2010; Imbens, 2000) which predicted the probability of having student debt or a monthly debt payment in a particular quartile. Variables for which there were statistically significant differences predicting quartile membership were used as conditioning variables in the selection models. Boxplots demonstrated sufficient overlap of propensity scores across debt amount and monthly payment quartile groups. In addition, to answer our fourth research question, we examined differences in the predicted probabilities of material hardship, health care hardship, and financial difficulty by debt payment and graduation status, comparing participants who were making loan payments and those who were not making payments, and comparing participants who completed or did not complete college and were not currently in school. Because material and health care hardship and financial difficulty were binary outcomes, IPTW logistic regression models were used to estimate the association of student debt and debt amounts with hardship and financial difficulty outcomes, controlling for age, gender, race/ethnicity, educational attainment, employment status, dependents, income, unsecured debt, liquid financial assets, financial shocks, ability to access $2000 in an emergency, health insurance status, and careful budgeting habits. These covariates were included in models to control for factors in addition to student debt that might explain outcomes, including demographic characteristics of participants and household financial circumstances. The set of covariates used in this response equation were different than the set of conditioning variables used in the selection model that produced propensity scores (Freedman & Berk, 2008).

4. Results 4.1. Sample description A majority of the sample was female (65%), White (76%), employed (70%), had single tax filing status (61%), and claimed no tax dependents (65%). Average age and income were 36 years and nearly $19,000, respectively. Well over half of the sample (61%) reported having a student loan. Of the 31% of participants currently enrolled in post-secondary education, most (76%) had student loans. Compared to other race and ethnicity groups, Black participants were more likely to have student debt, and women were more likely than men to have student debt. Participants with student debt were also younger and more highly educated than participants without student debt (see Table 1 for sample description). Hardship and financial difficulty were very common in the sample. Most participants said they had experienced one or more instances of material (60%) and health care (51%) hardship in the six months after filing taxes. Nearly a third (31%) said they experienced a financial difficulty. Participants with student debt were more likely than those without student debt to experience each of these outcomes. 4.2. Student debt and financial characteristics The mean amount of household student debt among borrowers in the sample (n = 3399) was $37,896 (SD = $39,376) and the median amount was $25,000. Among participants currently making student loan payments (n = 1822), the average household monthly payment was $226 (SD = $205). Participants had an average of $6673 (SD = 12,287) in unsecured, non-education debt and $5901 (SD = $8604) in liquid financial assets. Participants with student debt had $1199 more in unsecured debt t(5556) = 3.54, p b 0.001 and $1957 less in liquid financial assets t(5556) = 8.31, p b 0.001 than participants without student debt. 4.3. Effects of student debt on hardships Table 2 displays results regarding the association between having student debt and material and health care hardship and financial difficulty. Participants with student debt had 51%, 19%, and 27% greater odds of experiencing material hardship (p b 0.001), health care hardship (p b 0.05), and financial difficulty (p b 0.01), respectively, compared to participants with no student debt. Certain demographic variables were also associated with outcomes. An increase in one year in age was associated with greater odds for health care hardship (p b 0.01). Women had 59%, 31%, and 31% greater odds than men of experiencing material hardship, health care hardship, and financial difficulty. Participants with dependents had 88% and 67% greater odds of material hardship, and financial difficulty (both p b 0.001), respectively, compared to participants without dependents. Several household financial circumstances were also associated with outcomes. Every additional financial shock – such as a major car repair – was associated with 66%, 74%, and 36% greater odds of material hardship, health care hardship, and financial difficulty (all p b 0.001). An increase of $1000 in unsecured debt (not including student debt) was also associated with greater odds for these three outcomes. Other household financial characteristics were associated with lesser odds of hardship and financial difficulty, including liquid financial assets, health insurance, ability to come up with $2000 in an emergency, and careful budgeting habits. Higher levels of educational attainment were associated with lesser odds of material hardship, but not health care hardship or financial difficulty. Table 3 displays results concerning the association of having student debt with odds of experiencing specific events. Participants with student debt had greater odds of skipping housing payments, bill payments, medical care, dental care, and prescription medications, as well

M.R. Despard et al. / Children and Youth Services Review 70 (2016) 8–18 Table 2 Student debt, material and health care hardship and financial difficulty: results of IPTW logistic regression (n = 5388). Material hardship

Has student debt Age Gender (male) Female Race/ethnicity (White) Black Asian Latino Other Education (some college) College degree Some graduate school Graduate/professional degree Employment (not working) Part-time Full-time Has dependent(s) Gross income/$1k Liquid financial assets/$1k Unsecured debt/$1k Number of shocks Can access $2k in emergency Has health insurance Careful budgeting habit Constant

Health care hardship

Financial difficulty

SE

OR

SE

OR

SE

1.51⁎⁎⁎ 1.00

0.12 0.00

1.19⁎ 1.01⁎⁎

0.09 0.00

1.27⁎⁎ 0.99⁎⁎⁎

0.10 0.00

1.59⁎⁎⁎

0.13

1.31⁎⁎

0.10

1.31⁎⁎

0.11

1.18 0.60 1.00 0.95

0.17 0.16 0.16 0.20

0.79 0.49⁎⁎ 0.88 0.94

0.11 0.10 0.13 0.19

1.21 0.63 1.08 1.02

0.14 0.15 0.15 0.17

0.80⁎ 0.73⁎ 0.71⁎⁎

0.07 0.10 0.09

0.95 0.98 0.91

0.08 0.13 0.12

0.85 0.87 0.96

0.07 0.12 0.13

0.95 0.92 1.88⁎⁎⁎ 1.00 0.95⁎⁎⁎ 1.01⁎ 1.66⁎⁎⁎ 0.28⁎⁎⁎ 0.64⁎⁎⁎ 0.89⁎⁎ 2.64⁎⁎⁎

0.12 0.10 0.18 0.00 0.01 0.00 0.08 0.02 0.06 0.03 0.63

0.85 0.90 1.17 1.00 0.98⁎⁎⁎ 1.01⁎⁎ 1.74⁎⁎⁎ 0.38⁎⁎⁎ 0.27⁎⁎⁎

0.10 0.09 0.10 0.00 0.01 0.00 0.08 0.03 0.02 0.03 0.47

1.25 1.24⁎ 1.67⁎⁎⁎

0.15 0.13 0.15 0.00 0.01 0.00 0.06 0.04 0.07 0.02 0.32

0.97 2.07⁎⁎

1.42

Table 4 Student debt amount and material and health hardship (n = 3318). Material hardship Variable

OR

1.00 0.95⁎⁎⁎ 1.01⁎⁎⁎ 1.36⁎⁎⁎ 0.45⁎⁎⁎ 0.82⁎ 0.72⁎⁎⁎

13

Note: Reference group(s) in parentheses. ⁎ p b 0.05. ⁎⁎ p b 0.01. ⁎⁎⁎ p b 0.001.

as experiencing food insecurity and overdrawing bank accounts. For example, participants with student debt had 60% greater odds of skipping a bill payment (p b 0.001) compared to participants without student debt. Table 4 presents results concerning the relationship between different amounts of student debt and material and health care hardship and financial difficulty (n = 3318). Compared to borrowers in the lowest debt amount quartile ($1 to $10,500), borrowers in the next highest quartile ($10,501–$25,000) had no greater odds of hardship and financial difficulty. However, borrowers in the third ($25,001–$50,000) and fourth ($51,000–$200,000) quartiles had greater odds of health care hardship (both p b 0.05). Age, gender, having dependents, unsecured debt, and financial shocks were associated with greater odds of material hardship, while higher levels of education, liquid assets, ability to access $2000 in an emergency, health insurance, and careful budgeting habits Table 3 Student debt and specific material and health care hardships and financial difficulties: results of IPTW logistic regression (n = 5392). Material hardship

OR

SE

Skipped housing payment Skipped bills Food insecurity

1.56⁎⁎⁎ 1.60⁎⁎⁎ 1.22⁎⁎

0.16 0.13 0.09

Health care hardship Skipped necessary medical care Skipped necessary dental care Skipped prescription medications

1.26⁎⁎ 1.20⁎ 1.24⁎

0.10 0.09 0.11

Financial difficulty Overdrew account Credit card declined

1.28⁎⁎

Health care hardship

Financial difficulty

OR

SE

OR

SE

OR

SE

0.98 1.18 1.30 1.01⁎

0.13 0.15 0.19 0.01

0.98 1.31⁎ 1.53⁎⁎ 1.01⁎⁎

0.12 0.16 0.21 0.01

0.88 1.26 1.23 1.01

0.11 0.16 0.17 0.00

1.56⁎⁎⁎

0.16

1.35⁎⁎

0.13

1.22

0.13

1.21 1.06 0.89 1.05

0.20 0.33 0.15 0.24

1.01 0.65 0.83 0.85

0.14 0.17 0.14 0.20

1.50⁎⁎ 1.22 1.47⁎ 1.09

0.19 0.34 0.23 0.23

0.72⁎⁎ 0.73⁎ 0.46⁎⁎⁎

0.09 0.11 0.07

0.88 0.83 0.55⁎⁎⁎

0.10 0.12 0.09

0.82 0.84 0.75

0.09 0.13 0.12

1.05 1.13 2.04⁎⁎⁎ 0.99 0.96⁎⁎ 1.01⁎⁎ 1.65⁎⁎⁎ 0.30⁎⁎⁎ 0.64⁎⁎⁎ 0.83⁎⁎⁎ 3.51⁎⁎⁎

0.15 0.14 0.25 0.01 0.01 0.01 0.10 0.03 0.08 0.04 1.04

0.71⁎ 0.96 1.05 0.99 0.98⁎ 1.01⁎⁎ 1.66⁎⁎⁎ 0.45⁎⁎⁎ 0.27⁎⁎⁎

0.10 0.11 0.11 0.01 0.01 0.00 0.09 0.04 0.03 0.04 0.53

1.24 1.20 1.65⁎⁎⁎ 1.00 0.98 1.01⁎⁎ 1.26⁎⁎⁎ 0.39⁎⁎⁎ 0.81⁎ 0.74⁎⁎⁎

0.17 0.14 0.18 0.01 0.01 0.00 0.06 0.04 0.09 0.03 0.24

a

Student debt amount, quartile Q2 ($10,501–$25,000) Q3 ($25,001–$50,000) Q4 ($50,001–$200,000) Age Gender (male) Female Race/ethnicity (White) Black Asian Latino Other Education (some college) College Some graduate school Graduate/professional degree Employment (not working) Part-time Full-time Has dependents Gross income/$1k Liquid financial assets/$1k Unsecured debt/$1k Number of shocks Can access $2k in emergency Has health insurance Careful budgeting habit Constant

0.99 2.00⁎⁎

0.90

Note: Estimates based on IPTW multinomial logistic regression; sample only includes participants with student debt. ⁎ p b 0.05. ⁎⁎ p b 0.01. ⁎⁎⁎ p b 0.001. a Reference group is Quartile 1 ($1–$10,500).

were associated with lesser odds of material hardship. To a lesser extent, certain of these variables were also associated with health care hardship and financial difficulty. In addition, Black and Latino participants had greater odds for financial difficulty compared to White participants. Table 5 displays results concerning the association of varying amounts of student debt with odds of experiencing specific events. Compared to participants in the first debt quartile, participants in the third and fourth quartiles had greater odds of skipping bills, medical care, and dental care. Fourth quartile participants also had greater odds of skipping prescription medications. Table 6 displays results concerning the association of monthly debt payment amounts with outcomes. Compared to participants in the first quartile for monthly payments ($1 to $88), there were no statistically significant differences in odds for material and health care hardship and financial difficulty among participants in the second ($89 to $153), third ($154 to $300), and fourth ($301 to $900) quartiles. Similar to results reported in Tables 2 and 4 for student debt and debt amounts, respectively, other variables like gender, having dependents, financial shocks, liquid financial assets, and having health insurance were associated with odds for material and health care hardship and financial difficulty.

4.4. Differences between payers and non-payers

Note: Model covariates are same as reported in Table 2. ⁎ p b 0.05. ⁎⁎ p b 0.01. ⁎⁎⁎ p b 0.001.

1.07

0.10 0.12

Among participants with student debt but not currently enrolled in school, we compared outcomes for those making monthly payments (n = 1411) and those not making monthly payments (n = 676). Among non-payers, data concerning reasons for not making payments (e.g., deferment, forbearance) were not available. Payers had an average

14

M.R. Despard et al. / Children and Youth Services Review 70 (2016) 8–18

Table 5 Student debt amount and specific material and health hardships (n = 3318). Q2 ($10,501–$25,000)

Q3 ($25,001–$50,000)

Q4 ($50,001–$200,000)

Material hardships

OR

OR

OR

Skipped housing payment Skipped bills Food insecurity

0.84 1.04 0.90

1.16 1.34⁎** 1.16

1.29 1.76⁎⁎⁎ 1.15

Health hardships Skipped necessary medical care Skipped necessary dental care Skipped prescription medications

0.90 0.96 1.06

1.30⁎ 1.30⁎ 1.18

1.40⁎ 1.35⁎ 1.41⁎

Financial difficulty Overdrew account Credit card declined

0.88 0.88

1.26 1.21

1.24 1.15

Note: Estimates based on IPTW multinomial logistic regression; sample only includes participants with student debt; model covariates are same as reported in Table 4; reference group is Quartile 1 ($1–$10,500). ⁎ p b 0.05. ⁎⁎ p b 0.01. ⁎⁎⁎ p b 0.001.

student loan amount of $34,266 (SD = $37,674) compared to $41,735 (SD = $41,227) among non-payers t(2085) = 4.11, p b 0.001. As seen in Table 7, payers experienced less material and health care hardship and financial difficulty than non-payers. Compared to nonpayers, payers were younger, less likely to be Black or to have dependents, more likely to be employed or have health insurance, and had higher levels of education. Payers also had several economic advantages compared to non-payers: more income, more liquid financial assets, less

Table 6 Monthly student debt payment amounts and material and health hardship (n = 1783). Material hardship Variable Payment quartilea Q2 ($89–$153) Q3 ($154–$300) Q4 ($301–$900) Age Gender (male) Female Race/ethnicity (White) Black Asian Latino Other Education (some college) College Some graduate school Graduate/professional degree Employment (not working) Part-time Full-time Has dependents Gross income/$1k Liquid financial assets/$1k Can access $2k in emergency Unsecured debt/$1k Number of shocks Has health insurance Careful budgeting habit Constant

Health care hardship

Financial difficulty

OR

SE

OR

SE

OR

SE

0.92 0.98 1.24 1.02⁎

0.16 0.17 0.23 0.01

0.88 0.86 0.93 1.00

0.14 0.13 0.18 0.01

1.15 0.96 1.17 1.00

0.19 0.15 0.23 0.01

1.79⁎⁎⁎

0.23

1.61⁎⁎⁎

0.21

1.52⁎⁎

0.21

1.69⁎ 1.16 1.06 0.80

0.40 0.45 0.24 0.27

1.12 1.00 1.09 1.22

0.23 0.40 0.24 0.36

1.50⁎ 1.12 1.79⁎⁎ 1.08

0.29 0.46 0.37 0.30

0.73 0.74 0.43⁎⁎⁎

0.12 0.16 0.08

1.03 0.72 0.70

0.17 0.15 0.14

0.94 1.08 0.76

0.15 0.24 0.14

0.94 0.95 1.80⁎⁎⁎ 0.98⁎⁎ 0.97⁎ 0.25⁎⁎⁎

0.20 0.16 0.27 0.01 0.01 0.03 0.01 0.13 0.10 0.05 1.55

0.63⁎ 0.72 1.15 0.98⁎⁎

0.13 0.12 0.17 0.01 0.01 0.05 0.01 0.12 0.04 0.06 1.73

1.13 1.00 1.99⁎⁎⁎ 1.00 0.97 0.42⁎⁎⁎ 1.02⁎⁎ 1.27⁎⁎

0.24 0.18 0.31 0.01 0.02 0.06 0.01 0.09 0.12 0.04 0.39

1.00 1.66⁎⁎⁎ 0.61⁎⁎ 0.92 3.70⁎⁎

0.98 0.42⁎⁎⁎ 1.00 1.58⁎⁎⁎ 0.23⁎⁎⁎ 1.00 4.55⁎⁎⁎

0.79 0.74⁎⁎⁎ 0.98

Note: estimates based on IPTW multinomial logistic regression; sample only includes participants with student debt. ⁎ p b 0.05. ⁎⁎ p b 0.01. ⁎⁎⁎ p b .001. a Reference group is participants in the first quartile for monthly payments ($1 to $88).

unsecured debt, and fewer financial shocks. There was no difference, however, in careful budgeting habits between the two groups. The predicted probability of material hardship – all other things being equal - was 64% for payers compared to 77% for non-payers t(2039) = 13.20, p b 0.001. The predicted probability of health care hardship was 52% for payers and 63% for non-payers t(2039) = 10.94, p b 0.001. The predicted probability of financial difficulty was 41% for payers and 53% for non-payers t(2039) = 13.77, p b 0.001. Table 7 Differences between borrowers making and not making student debt payments (n = 2087).

Outcomes Material hardship Health care hardship Financial difficulty Covariates Age Gender Male Female Race/ethnicity White Black Asian Hispanic Other Education Some college College degree Some graduate school Graduate degree Employment status Employed full-time Employed part-time Not working Has dependents Adjusted gross income Liquid financial assets Unsecured debt Number of shocks Has health insurance Careful budgeting habita

Payers % or mean (n = 1411)

Non-payers % or mean (n = 676)

Bivariate differences p

61.77 49.79 30.64

81.51 67.31 47.93

⁎⁎⁎ ⁎⁎⁎ ⁎⁎⁎

33.08

35.63

⁎⁎⁎

32.79 67.21

34.64 65.36

ns ns

74.47 10.67 2.28 8.18 4.41

71.57 18.63 1.19 5.07 3.58

ns ⁎⁎⁎

25.27 49.04 5.75 19.94

35.41 38.22 8.00 18.37

⁎⁎⁎ ⁎⁎⁎

72.48 14.54 12.98 33.62 21,320 $5094 $6650 0.94 77.60 3.55

56.00 20.59 23.41 55.92 18,514 $4266 $9418 1.25 68.05 3.46

⁎⁎⁎ ⁎⁎⁎ ⁎⁎⁎ ⁎⁎⁎ ⁎⁎⁎ ⁎⁎ ⁎⁎⁎ ⁎⁎⁎ ⁎⁎⁎

ns ⁎ ns

ns ns

ns

Note: Sample excludes participants currently enrolled in school. ⁎ p b 0.05. ⁎⁎ p b 0.01. ⁎⁎⁎ p b 0.001. a Measured on a five-point scale (1 – “not at all like me” to 5 – “very much like me”).

M.R. Despard et al. / Children and Youth Services Review 70 (2016) 8–18

4.5. Differences between college graduates and non-graduates with student debt Among participants with any student debt, those who did not complete college (n = 599) had a mean student debt of $20,399 (SD = $22,302) compared to a mean of $33,773 (SD = $29,447) among participants who graduated from college (n = 955) t(1552) = 8.82, p b 0.001. Though their student debt was higher, the predicted probability of material hardship was lower for graduates (65%) compared to non-graduates (83%) t(1522) = 20.95, p b 0.001. Similarly, the predicted probability of health care hardship was 55% and 65% for graduates and non-graduates, respectively t(1522) = 9.46, p b 0.001. The predicted probability of financial difficulty was 41% and 55% for graduates and non-graduates, respectively t(1522) = 14.74, p b 0.001. Non-graduates were more likely to be Black χ2 (1, 1546) = 7.13, p b 0.01 and have dependents χ2 (1, 1546) = 70.28, p b 0.001. 4.6. Limitations There are several limitations of our study to note. First, our findings may not be generalizable to the LMI population in the US as our sample was comprised of LMI persons who filed their taxes online. Those who file their own taxes may differ systematically from those who use paid preparers or volunteer income tax assistance sites to file their taxes, and from LMI persons who do not file federal income taxes. Second, we did not observe whether participants were from LMI households of origin – only what their income was upon filing their taxes. As a result, we are unable to determine how the relationships we observe between student debt and hardship are affected by family economic backgrounds of participants (Elliott, Destin, & Friedline, 2011; Huang, Guo, Kim, & Sherraden, 2010) compared to their current financial circumstances. This is an important distinction as borrowers from more economically advantaged families of origin may experience greater economic mobility than their disadvantaged counterparts (Carasso & McKernan, 2007; Hertz, 2006; Pew Charitable Trusts, 2013). These borrowers may also receive financial support from their families that could be used to repay their student debt or manage other expenses (Cobb-Clark & Gørgens, 2014; Hartnett, Furstenberg, Birditt, & Fingerman, 2012; Wightman, Patrick, Schoeni, & Schulenberg, 2013). As a result, relationships we observe between student debt and hardship may be temporary for some, longer-lasting for others. This limitation can be addressed by using longitudinal designs in examining the relationship between student debt and household well-being. Third, as is a general limitation with propensity score analysis, there may be unobserved differences between participants with and without student debt and among participants with varying levels of student debt that help explain variation in outcomes. Fourth, there are various measurement limitations with HFS data. Though we restrict the analytic sample to households with only one or two adults, we are unable to distinguish a participant's student debt from that of other household members. It is conceivable that another household member's student debt obligation may bear no relation to a participant's ability to meet basic needs because financial responsibilities are segregated. Also, in examining loan amount quartiles as a predictor, we did not observe the length of time participants had been making student loan payments. Original loan amounts may be associated with hardship and financial difficulty. 5. Discussion In this study, we examine the relationship between student debt and material and health care hardship and financial difficulty with a large sample of LMI households who filed their income taxes online. Hardship and financial difficulty are common in the entire sample. Over half of the sample experienced one or more hardships in the six months after filing their taxes, such as skipping a rent payment. Nearly a third overdrew a bank account or had a credit card declined for hitting the credit limit.

15

Though common, hardship and financial difficulty is worse for participants with student debt than for those without student debt, even after using propensity score analysis to balance the sample on many important characteristics. Having student loans is associated with difficulty meeting basic needs and managing finances. The above findings are consistent with prior research that finds an association between student debt and financial difficulty and extend the examination to look at material and health care difficulty as well (Baum & O'Malley, 2003; Bricker & Thompson, 2016; Gicheva & Thompson, 2015; Soria et al., 2014). We build on the student debt research literature by demonstrating that student debt is associated with problems meeting basic survival needs among a LMI sample (e.g., housing, food, health care), while adjusting for selection bias by using propensity score methods. Student debt may be especially difficult for LMI borrowers to manage and may adversely affect meeting basic needs. Though we find student debt in general is problematic, we find only modest evidence that higher loan amounts are associated with greater hardship and financial difficulty. Participants with more than $25,000 in debt had greater odds for health care hardship, but not material hardship or financial difficult compared to participants in the lowest debt quartile. Also, whether borrowers are able to make monthly payments matters. Participants making monthly student debt payments have lower odds of hardship and financial difficulty and are better off financially compared to participants with student debt who are not making loan payments. Though we do not observe this directly, non-payers may be borrowers who are in loan default, deferment, or forbearance. We also find that those who are repaying their student loans have a very different social and economic profile than those who are not repaying their loans: younger, less likely to be Black, less likely to be supporting dependents, more likely to be employed full-time, and more likely to have health insurance. The differences between payers and non-payers we observe are very similar to those Hillman (2014) found between borrowers in re-payment and those in default. Thus, among LMI borrowers, there are relative advantages and disadvantages that may affect repayment ability. It is possible that LMI borrowers making payments have access to resources that those unable to make payments do not, such as parents and other family members who can help with debt payments. Also, given other factors such as race and having dependents, LMI borrowers making payments may only be temporarily LMI whereas borrowers with a greater set of social and economic disadvantages to navigate may experience less social and economic mobility. In other words, our findings suggest that LMI borrowers are a heterogeneous population. Future research should explore this heterogeneity to better understand how student debt differentially affects LMI borrowers. We also find that college completion matters. College graduates in our study have average student debt similar to national averages for all graduates (Institute for College Access & Success, 2015) and that is much higher than borrowers in our study who did not complete college. Despite higher student debt, college graduates have much lower odds of hardship and financial difficulty compared to non-graduates, a finding consistent with prior research (Bricker & Thompson, 2016; Gicheva & Thompson, 2015). In all of our models, participants with at least a college degree fare better than those with some college, but no degree. Based on these findings, earning a degree appears to pay off, even though the study sample has lower income than the general population. Those unable to finish college may find it more difficult to compete in the labor market and earn higher income to pay back student debt while meeting other needs such as housing and medical care. Black borrowers were more likely than other borrowers to have not completed college, similar to a finding of Jackson and Reynolds (2013) who analyzed data from the 1996–2001 Beginning Postsecondary Student study. Regardless of student debt status and educational attainment, hardship and financial difficulty is a common experience in this sample of LMI households. Though college graduates in our sample fare better

16

M.R. Despard et al. / Children and Youth Services Review 70 (2016) 8–18

than non-graduates, most experience hardship. Like Baum and O'Malley (2003), we find that hardship is common even among those able to repay their loans. We find several other factors are associated with hardship and financial difficulty, especially gender (women are worse off), having dependents, and experiencing financial shocks. Race also emerges as a factor. Compared to White student loan borrowers, Black – as well as Latino - borrowers have greater odds of financial difficulty, all other things being equal. This finding is consistent with prior research that finds higher credit card debt among Black students (Grable & Joo, 2006; Lyons, 2004). Conversely, liquid financial assets, health insurance, being able to access $2000 in an emergency, and careful budgeting habits are associated with lower odds of hardship and financial difficulty. These findings suggest that mitigating the negative impact of student debt on the financial well-being of LMI borrowers alone is insufficient. These borrowers also need opportunities and incentives to save, health insurance coverage, and financial education and counseling. The picture that emerges from our study is that hardship and financial difficulty are a common experience in our LMI sample and are more likely if one has student debt and for higher debt amounts. Hardship and financial difficulty are greater among borrowers not making repayments or who have not completed college. Even though borrowers making payments had many advantages relative to borrowers not making payments, hardship and financial difficulty were still common.

demographic characteristics and household financial circumstances than those repaying their loans. It is highly likely that non-payers include borrowers in deferment or forbearance (which we did not measure in this study) due to hardship and financial difficulty which makes repayment difficult. This raises questions to be answered with future research: what is the extent of material and health care hardship and financial difficulty among borrowers requesting deferments and forbearances? To what degree to these hardships and financial difficulties constrain re-payment ability? How are lenders making forbearance decisions – with what data and understanding of household financial, material, and health care circumstances? To what degree are LMI households offered income-based repayment, deferment, and forbearance options? These questions should be asked relative to the type of college students attended, as attending a for-profit college or university is associated with loan default (Hillman, 2014). In this study, we find that student debt is associated with greater odds for material and health care hardship and financial difficulty among a sample of LMI individuals, but repayment amounts are unrelated to these outcomes. Also, participants currently making student loan payments and graduates with student debt have much lower odds for hardship and financial difficulty compared to non-payers and non-graduates. The relationship between student debt and hardship needs to be further examined with respect to repayment ability and college completion to inform loan repayment and college retention policies, respectively.

5.1. Implications Disclaimer The main implication of our study is that LMI students need support paying for college, as dependence on loans to finance higher education may be burdensome for this population. Targeted efforts to support LMI students financially are likely to be effective before, during, and after enrollment. Before enrollment, asset-building approaches such as child savings accounts can incentivize LMI students and households to save early for college (Elliott & Friedline, 2013; Friedline, Elliott, & Nam, 2013; Friedline & Nam, 2014). During enrollment, LMI students can benefit from expanded grant and need-based financial aid programs such as Pell Grants or federal work-study. Grant-based forms of aid not only reduce the unmet costs of college for students (Monks, 2014), but are also superior to loans on various indicators of college achievement for low-income students (Chen & DesJardins, 2010; Kim, DesJardins, & McCall, 2009). The trend of declining state funding for public universities, which has placed a greater cost burden on students and their families, should be reversed. This is especially important for public universities with small endowments that struggle to supplement federal aid. To gauge the effectiveness of public and private sources of aid in making college affordable for LMI students and to help these students make informed choices prior to enrollment, colleges and universities should use an average net price metric (Burd, 2016). Assessing cost burden is especially important to ensure that LMI students do not leave college early with student debt and no degree due to crushing credit card and student loan debt (Soria et al., 2014). Our finding that non-graduates have greater hardship and financial difficulty than graduates suggests that efforts to retain LMI students should be strengthened. After enrollment, LMI borrowers need continued access to and expanded income-based repayment options to reduce debt burden and lessen risk for hardship and financial difficulty. This is especially true for households with other disadvantages that heighten risk for hardship. Employers can also play a role by offering student loan repayment assistance as a way to help attract and retain new employees, particularly because student debt is associated with less retirement savings and delayed homeownership (Brown & Caldwell, 2013; Brown et al., 2015; Elliott et al., 2013; Gicheva & Thompson, 2015; Shand, 2007). We find that over a third of participants with student debt who are not currently in school are not making student loan payments. As noted above, this group had higher odds of hardship and had different

Statistical compilations disclosed in this document relate directly to the bona fide research of, and public policy discussions concerning savings behavior as it relates to tax compliance. Compilations are anonymous and do not disclose information containing data from fewer than 10 tax returns or reflect taxpayer-level data with the prior explicit consent from taxpayers. Compilations follow Intuit's protocols to help ensure the privacy and confidentiality of customer tax data. Conflict of interest statement The Authors declare that there are no conflicts of interest in submitting this manuscript to Children and Youth Services Review. Acknowledgements The Center for Social Development at Washington University in St. Louis gratefully acknowledges the funders who made the Refund to Savings Initiative possible: the Ford Foundation; the Annie E. Casey Foundation; Intuit, Inc.; the Intuit Financial Freedom Foundation; the Smith Richardson Foundation; and JPMorgan Chase and Company. The Refund to Savings Initiative would not exist without the commitment of Intuit and its Tax and Financial Center. We appreciate the contributions from many individuals in the Consumer Group who worked diligently on the planning and implementation of the experiment. We thank Trina Shanks, Sherri Kossoudji, Kristin Seefeldt, Natasha Pilkauskas, and anonymous reviewers for comments. Lastly, we thank the thousands of tax payers who consented to participate in the research surveys and shared their personal financial information. References Austin, P. C. (2010). The performance of different propensity-score methods for estimating differences in proportions (risk differences or absolute risk reductions) in observational studies. Statistics in Medicine, 29, 2137–2148. http://dx.doi.org/10.1002/sim. 3854. Austin, P. C. (2011). An introduction to propensity score methods for reducing the effects of confounding in observational studies. Multivariate Behavioral Research, 46, 399–424. http://dx.doi.org/10.1080/00273171.2011.568786.

M.R. Despard et al. / Children and Youth Services Review 70 (2016) 8–18 Austin, P. C. (2013). The performance of different propensity score methods for estimating marginal hazard ratios. Statistics in Medicine, 32, 2837–2849. http://dx.doi.org/10. 1002/sim.5705. Baum, S., & O'Malley, M. (2003). College on credit: How borrowers perceive their education debt. Journal of Student Financial Aid, 33, 7–19http://publications.nasfaa.org/jsfa/ vol33/iss3/1/ Baum, S., & Schwartz, S. (2006). How much debt is too much? Defining benchmarks for manageable student debt. New York: College Board. Best, K., & Keppo, J. (2014). The credits that count: How credit growth and financial aid affect college tuition and fees. Education Economics, 22, 589–613. http://dx.doi.org/ 10.1080/09645292.2012.687102. Beverly, S. G. (2001). Measures of material hardship: Rationale and recommendations. Journal of Poverty, 5(1), 23–41. http://dx.doi.org/10.1300/J134v05n01_02. Bricker, J., & Thompson, J. (2016). Does education loan debt influence household financial distress? An assessment using the 2007–2009 Survey of Consumer Finances panel. Contemporary Economic Policy (advance online publication). http://dx.doi.org/10. 1111/coep.12164. Brown, M., & Caldwell, S. (2013). Young student loan borrowers retreat from housing and auto markets. Liberty Street Economics Retrieved from http://libertystreeteconomics. newyorkfed.org/2013/04/young-student-loan-borrowers-retreat-from-housingand-auto-markets.html#.VxWtRMd6pFI Brown, M., Haughwout, A., Lee, D., Scally, J., & Van Der Klaauw, W. (2015). Student debt growth and the repayment progress of recent cohorts. American Bankruptcy Institute. Law review, Vol. 23. (pp. 331–346). Burd, S. (2016). Undermining Pell: Volume III. The news keeps getting worse for low-income students (Education Policy Program Report). Washington, DC: New Americahttps:// static.newamerica.org/attachments/12813-undermining-pell-volume-iii/ Undermining-Pell-III-3.15bba9018bb54ad48f850f6f3a62a9fc.pdf Carasso, A., & McKernan, S. -M. (2007). The balance sheets of low-income households: What we know about their assets and liabilities. Washington, DC: The Urban Institutehttp:// www.urban.org/UploadedPDF/411594_low-income_balance_sheets.pdf Carneiro, P., Heckman, J. J., & Vytlacil, E. J. (2010). Estimating marginal returns to education. Working Paper No. 16474 National Bureau of Economic Research Retrieved from http://www.nber.org/papers/w16474 Chen, R., & DesJardins, S. L. (2010). Investigating the impact of financial aid on student dropout risks: Racial and ethnic differences. Journal of Higher Education, 81(2), 179–208. Chopra, R. (2013). Student debt swells. Federal loans now top a trillion Retrieved from http://www.consumerfinance.gov/newsroom/student-debt-swells-federal-loansnow-top-a-trillion/ Cobb-Clark, D., & Gørgens, T. (2014). Parents' economic support of young adult children: Do socioeconomic circumstances matter? Journal of Population Economics, 27, 447–471. http://dx.doi.org/10.1007/s00148-013-0484-6. College Board (2015). Trends in college pricing 2015 (trends in higher education series). New York, NY: The College Board" located in New York, NY Retrieved form http:// trends.collegeboard.org/sites/default/files/trends-college-pricing-web-final-508-2. pdf. Cooke, R., Barkham, M., Audin, K., Bradley, M., & Davy, J. (2004). Student debt and its relation to student mental health. Journal of Further and Higher Education, 28, 53–66. http://dx.doi.org/10.1080/0309877032000161814. Dee, T. S. (2004). Are there civic returns to education? Journal of Public Economics, 88(9– 10), 1697–1720. http://dx.doi.org/10.1016/j.jpubeco.2003.11.002. Elliott, W., & Friedline, T. (2013). “You pay your share, we'll pay our share”: The college cost burden and the role of race, income, and college assets [special issue]. Economics of Education Review, 33, 134–153. http://dx.doi.org/10.1016/j.econedurev. 2012.10.001. Elliott, W., & Lewis, M. (2015). Student debt effects on financial well-being: Research and policy implications. Journal of Economic Surveys, 29, 614–636. http://dx.doi.org/10. 1111/joes.12124. Elliott, W., III, & Nam, I. (2013). Is student deb jeopardizing the short-term financial health of U.S. households? Federal Reserve Bank of St. Louis Review, 95, 405–424https:// research.stlouisfed.org/publications/review/13/09/Elliott.pdf Elliott, W., Destin, M., & Friedline, T. (2011). Taking stock of ten years of research on the relationship between assets and children's educational outcomes: Implications for theory, policy and intervention. Children and Youth Services Review, 33, 2312–2328. http://dx.doi.org/10.1016/j.childyouth.2011.08.001. Elliott, W., Grinstein-Weiss, M., & Nam, I. (2013). Student debt and declining retirement savings (CSD Research Brief 13–34). St. Louis, MO: Washington University, Center for Social Developmenthttp://csd.wustl.edu/Publications/Pages/displayresultitem.aspx? ID1=1186 Executive Office of the President (2014). Increasing college opportunity for low-income students: Promising models and a call to action. https://www.whitehouse.gov/sites/ default/files/docs/white_house_report_on_increasing_college_opportunity_for_lowincome_students_1-16-2014_final.pdf Federal Reserve Bank of New York (2015, April). Student loan borrowing and repayment trends, 2015. Original research presented at press conference on student loan debt Retrieved from https://www.newyorkfed.org/medialibrary/media/newsevents/ mediaadvisory/2015/Student-Loan-Press-Briefing-Presentation.pdf Fitch, C., Hamilton, S., Bassett, P., & Davey, R. (2011). The relationship between personal debt and mental health: A systematic review. Mental Health Review Journal, 16(4), 153–166. http://dx.doi.org/10.1108/13619321111202313. Freedman, D. A., & Berk, R. A. (2008). Weighting regressions by propensity scores. Evaluation Review, 32, 392–409. http://dx.doi.org/10.1177/0193841X08317586. Friedline, T., & Nam, I. (2014). Savings from ages 16 to 35: A test to inform Child Development Account policy. Poverty & Public Policy, 6, 46–70. http://dx.doi.org/10.1002/ pop4.59.

17

Friedline, T., Elliott, W., & Nam, I. (2013). Small-dollar children's savings accounts and children's college outcomes by race. Children and Youth Services Review, 35, 548–559. http://dx.doi.org/10.1016/j.childyouth.2012.12.007. Fry, R. (2014). Student debt and overall economic well-being (social and demographic trends). Washington, DC: Pew Research Center Retrieved from http://www. pewsocialtrends.org/2014/05/14/section-1-student-debt-and-overall-economicwell-being/ Gicheva, D., & Thompson, J. (2015). The effects of student loans on long-term household financial stability. In B. Hershbein, & K. Hollenbeck (Eds.), Student loans and the dynamics of debt. Kalamazoo, MI: W. E. Upjohn Institute. Grable, J. E., & Joo, S. H. (2006). Student racial differences in credit card debt and financial behaviors and stress. College Student Journal, 40, 400–408http://projectinnovation. biz/csj_2006.html Greenstone, M., & Looney, A. (2012). Regardless of the cost, college still matters (Economic Analysis, the Hamilton Project of the Brookings Institution). http://www. hamiltonproject.org/papers/Regardless_of_the_Cost_College_Still_Matters/ Grinstein-Weiss, M., Perantie, D. C., Russell, B. D., Comer, K., Taylor, S. H., Luo, L., Key, C., & Ariely, D. (2015). Refund to Savings 2013: Comprehensive report on a large-scale taxtime saving program (CSD Research Report 15-06). St. Louis, MO: Center for Social Development, Washington University in St. Louis http://csd.wustl.edu/Publications/ Documents/RR15-06.pdf. Grinstein-Weiss, M., Perantie, D., Taylor, R., & Raghavan, R. (2016). Racial disparities in education debt burden among low- and moderate-income households. Children and Youth Services Review, 65, 166–174. http://dx.doi.org/10.1016/j.childyouth.2016.04.010. Guo, S., & Fraser, M. W. (2010). Propensity score analysis: Statistical methods and applications. Thousand Oaks, CA: Sage Publications. Hartnett, C., Furstenberg, F., Birditt, K., & Fingerman, K. (2012). Parental support during young adulthood: Why does assistance decline with age? Journal of Family Issues, 34(7), 975–1007. http://dx.doi.org/10.1177/0192513X12454657. Hérault, N., & Zakirova, R. (2015). Returns to education: Accounting for enrolment and completion effects. Education Economics, 23(1), 84–100. http://dx.doi.org/10.1080/ 09645292.2013.805184. Hershbein, B., Harris, B. H., & Kearney, M. S. (2014). Major decisions: Graduates' earnings growth and debt repayment (Economic Analysis, the Hamilton Project of the Brookings Institution). http://www.hamiltonproject.org/assets/files/major_ decisions_graduates_earnings_growth_debt_repayment.pdf Hertz, T. (2006). Understanding mobility in America. Report for the Center for American Progresshttps://cdn.americanprogress.org/wp-content/uploads/issues/2006/04/ Hertz_MobilityAnalysis.pdf Hillman, N. W. (2014). College on credit: A multilevel analysis of student loan default. The Review of Higher Education, 37(2), 169–195. http://dx.doi.org/10.1353/rhe.2014.0011. Hiltonsmith, R. (2013). At what cost? How education debt reduces lifetime wealth. Retrieved from http://www.demos.org/sites/default/files/publications/AtWhatCostFinal_ Demos.pdf Hirano, K., & Imbens, G. W. (2001). Estimation of causal effects using propensity score weighting: An application to data on right heart catheterization. Health Services & Outcomes Research Methodology, 2, 259–278. http://dx.doi.org/10. 1023/A:1020371312283. Houle, J. N. (2014). Disparities in debt: Parents' socioeconomic resources and young adult student loan debt. Sociology of Education, 87(1), 53–69. http://dx.doi.org/10.1177/ 0038040713512213. Houle, J., & Berger, L. (2015). Is student loan debt discouraging homeownership among young adults? Social Service Review, 89(4), 589–621. http://dx.doi.org/10.1086/ 684587. Hout, M. (2012). Social and economic returns to college education in the United States. Annual Review of Sociology, 38, 379–400. http://dx.doi.org/10.1146/annurev.soc. 012809.102503. Huang, J., Guo, B., Kim, Y., & Sherraden, M. (2010). Parental income, assets, borrowing constraints and children's post-secondary education. Children and Youth Services Review, 32, 585–594. http://dx.doi.org/10.1016/j.childyouth.2009.12.005. Huelsman, M. (2015). The debt divide: The racial and class bias behind the new normal of student borrowing. New York, NY: Demos Retrieved from http://www.demos.org/ sites/default/files/publications/The%20Debt%20Divide.pdf Imbens, G. W. (2000). The role of the propensity score in estimating dose-response functions. Biometrika, 87(3), 706–710. http://dx.doi.org/10.1093/biomet/87.3.706. Institute for College Access & Success (2015). Student debt and the class of 2014 (the project on student debt). Washington, DC: Institute for College Access & Success, located in Oakland, CA Retrieved from http://ticas.org/sites/default/files/pub_files/classof2014. pdf. Jackson, B. A., & Reynolds, J. R. (2013). The price of opportunity: Race, student loan debt, and college achievement. Sociological Inquiry, 83, 335–368. http://dx.doi.org/10.1111/ soin.12012. Kim, D. (2007). The effect of loans on students' degree attainment: Differences by student and institutional characteristics. Harvard Educational Review, 77, 64–100. http://dx. doi.org/10.17763/haer.77.1.n14t69l0q8292784. Kim, J., DesJardins, S., & McCall, B. (2009). Exploring the effects of student expectations about financial aid on postsecondary choice: A focus on income and racial/ethnic differences. Research in Higher Education, 50(8), 741–774. http://dx.doi.org/10.1007/ s11162-009-9143-x. Lee, D., Van der Klaauw, W., Haughwout, A., Brown, M., & Scally, J. (2014). Measuring student debt and its performance. Federal Reserve Bank of New York Staff Reports, Vol. 668, New York: Federal Reserve Bank of New York Retrieved from http://papers.ssrn.com/ sol3/papers.cfm?abstract_id=2423377 Lochner, L., Stinebrickner, T., & Suleymanoglu, U. (2013). The importance of financial resources for student loan repayment. NBER Working Paper No. 19716 National Bureau of Economic Researchhttp://www.nber.org/papers/w19716

18

M.R. Despard et al. / Children and Youth Services Review 70 (2016) 8–18

Lyons, A. C. (2004). A profile of financially at-risk college students. Journal of Consumer Affairs, 38, 56–80. http://dx.doi.org/10.1111/j.1745-6606.2004.tb00465.x. Monks, J. (2014). The role of institutional and state aid policies in average student debt. The Annals of the American Academy of Political and Social Science, 655(1), 123–142. http://dx.doi.org/10.1177/0002716214539093. Nelson, G. (2011). Measuring poverty: The official U.S. measure and material hardship. Poverty & Public Policy, 3, 1–35. http://dx.doi.org/10.2202/1944-2858.1077. Nelson, M. C., Lust, K., Story, M., & Ehlinger, E. (2008). Credit card debt, stress and key health risk behaviors among college students. American Journal of Health Promotion, 22, 400–407. http://dx.doi.org/10.4278/ajhp.22.6.400. Perna, L. W. (2003). The private benefits of higher education: An examination of the earnings premium. Research in Higher Education, 44, 451–472. http://dx.doi.org/10.1023/ A:1024237016779. Pew Charitable Trusts (2013). Moving on up: Why do some Americans leave the bottom of the economic ladder, but not others? (Brief). Retrieved from http://www.pewtrusts. org/~/media/assets/2013/11/01/movingonuppdf.pdf. Price, D. V. (2004). Educational debt burden among student borrowers: An analysis of the Baccalaureate & Beyond Panel, 1997 follow-up. Research in Higher Education, 45, 701–737. http://dx.doi.org/10.1023/B:RIHE.0000044228.54798.4c. Richardson, T., Elliott, P., & Roberts, R. (2013). The relationship between personal unsecured debt and mental and physical health: A systematic review and meta-analysis. Clinical Psychology Review, 33, 1148–1162 http://doi.org/10.1016/j.cpr.2013.08.009. Rosenbaum, P. R., & Rubin, D. B. (1983). The central role of the propensity score in observational studies for causal effects. Biometrika, 70, 41–55. http://dx.doi.org/10.1093/ biomet/70.1.41. Shand, J. M. (2007). The impact of early-life debt on the homeownership rates of young households: An empirical investigation. Retrieved from http://www.fdic.gov/bank/ analytical/cfr/2008/jan/CFR_SS_2008Shand.pdf Short, K. S. (2005). Material and financial hardship and income-based poverty measures in the USA. Journal of Social Policy, 34, 21–38. http://dx.doi.org/10.1017/ S0047279404008244.

Soria, K. M., Weiner, B., & Lu, E. C. (2014). Financial decisions among undergraduate students from low-income and working-class social class backgrounds. Journal of Student Financial Aid, 44, 2–23. http://publications.nasfaa.org/jsfa/vol44/iss1/2 Stuart, E. A. (2010). Matching methods for causal inference: A review and a look forward. Statistical Science: A Review Journal of the Institute of Mathematical Statistics, 25(1), 1–21. http://dx.doi.org/10.1214/09-STS313. Sweet, E., Nandi, A., Adam, E. K., & McDade, T. W. (2013). The high price of debt: Household financial debt and its impact on mental and physical health. Social Science & Medicine, 91, 94–100. http://dx.doi.org/10.1016/j.socscimed.2013.05.009. U.S. Department of Education (2011). National student loan two-year default rates. https://www2.ed.gov/offices/OSFAP/defaultmanagement/defaultrates.html U.S. Department of Education (2015a). Fact sheet: Department of Education launches experiment to provide Federal Pell Grant funds to high school students taking college courses for credit. http://www.ed.gov/news/press-releases/fact-sheet-departmenteducation-launches-experiment-provide-federal-pell-grant-funds-high-schoolstudents-taking-college-courses-credit U.S. Department of Education (2015b). National student loan default rates. http://www. ifap.ed.gov/eannouncements/attachments/ 093015AttachOfficialFY20123YRCDRBriefing.pdf Walsemann, K. M., Ailshire, J. A., & Gee, G. C. (2016). Student loans and racial disparities in self-reported sleep duration: Evidence from a nationally representative sample of US young adults. Journal of Epidemiology and Community Health, 70, 42–48. http://dx.doi. org/10.1136/jech-2015-205583. Wightman, P., Patrick, M., Schoeni, R., & Schulenberg, J. (2013). Historical trends in parental financial support of young adults. Ann Arbor, MI: University of Michigan, Population Studies Center Retrieved from http://blog.narotama.ac.id/wp-content/uploads/ 2014/12/Historical-Trends-in-Parental-Financial-Support-of-Young-Adults.pdf Winship, C., & Morgan, S. L. (1999). The estimation of causal effects from observational data. Annual Review of Sociology, 659–706http://www.jstor.org/stable/223520