Financial Stress, Race, and Student Debt during the Great Recession

Abstract

As the onus of paying for higher education shifted from the state onto students and their families, student indebtedness grew across a wide range of households in the United States in the 2000s, especially among Black and Hispanic households. Holding student debt is a financial risk that may leave households more vulnerable to economic shocks. We study the relationship between household student loan burden and the likelihood of financial stress during the Great Recession using the unique 2007-2009 panel of the Survey of Consumer Finances. We find a robust positive relationship across four dimensions of student loan burden and holding constant household characteristics and previous financial stress. We find that Black and Hispanic households who held student loans experienced particularly high levels of financial stress relative to White households. Our results demonstrate the importance of considering the household risk incurred in the US system of financed attainment, especially during the inevitable downturns of a capitalist economy.

Over the course of the 2000s, students and families increasingly took on student loan debt to finance higher education. This growth was especially stark among Black and Hispanic households (Goldrick-Rab 2016; Houle and Addo 2019; Jackson and Reynolds 2013; Seamster and Charron-Chénier 2017). Whereas higher education was once funded largely by state and federal governments as a public good, individuals have become increasingly responsible for financing their own educations (Dwyer, McCloud, and Hodson 2012; Glater 2015; Hacker 2006; Zaloom 2019). Declining public investments occurred just as prospects for workers without a college degree worsened. Job insecurity and slow earnings growth left Americans with fewer resources to cope with instability even as they were increasingly responsible for managing risks on their own.

Defenders of the student loan system argue that concerns with student loans are overstated, citing higher earnings over the long term, increases in expensive but lucrative post-graduate degrees, and concentration of poor returns among a small subset of borrowers (Looney and Yannelis 2015). These perspectives overlook the inevitable shocks in capitalist systems which produce cyclical downturns that scar many who live through them (Redbird and Grusky 2016). American households endured significant financial upheaval over the course of the 2000s, including during the Great Recession, the worst economic crisis in the preceding 25 years (Friedline, Chen, and Morrow 2020). Student loan-holders experience severe material hardship more often than those without loans, especially during recessionary periods (Bricker and Thompson 2016; Despard et al. 2016; Faber and Rich 2018). Yet there has been surprisingly little attention to the risk of more moderate but still important financial stress for student debtors.

Without better institutional safeguards, economic downturns may raise financial stress for student loan-holders, but it is also unclear whether all student loan holders face significantly elevated risk of financial stress or whether those risks concentrate among particularly vulnerable groups. Prior research demonstrates that lower-income, Black, and Hispanic populations experience higher levels of student loan default (Jackson and Reynolds 2013; Jiménez and Glater 2020). Less is known about these dynamics during economic downturns, when vulnerabilities among broader populations may occur as well. To what degree is the insecurity associated with student loans widely held versus unequally distributed during economic recessions?

Unequal educational institutions raise the possibility of significantly unequal risks of student debt, especially by race and ethnicity. Black and Hispanic populations more often attend schools with lower returns on education. Louise Seamster and Raphaël Charron-Chénier argue that institutions encouraging high debt to attend low-reward schools represents a form of “predatory inclusion” where disadvantaged groups have been “provided with access to a good, service, or opportunity from which they have historically been excluded but under conditions that jeopardize the benefits of access” (2017:200). These conditions may be linked to the capacity to manage significant economic downturns. To the extent that the financial stress of loans is concentrated among Black and Hispanic households, the structure may contribute to systemic racism in educational opportunities.

We employ the unique data source of the 2007-2009 panel of the Survey of Consumer Finances (SCF), which captures experiences during the Great Recession. We find that households carrying student loans experienced higher odds of financial stress during the Recession across four different measures of student loan burdens, with particularly high odds of stress among non-Hispanic Black and Hispanic households relative to non-Hispanic White households (henceforth Black, Hispanic, and White for brevity). We study the Great Recession in order understand the cyclical character of economic insecurity and the specific historic conditions that continue to unfold in the inevitable crises of a capitalist economy (Redbird and Grusky 2016). Further, we contribute to knowledge on family financial stress by taking into account the economic environment as an important explanation of economic wellbeing—a gap noted by Friedline and colleagues (2020) in a review of post-Recession scholarship on financial stress. We conclude by arguing for changes to financial aid policy that at the very least better protect student loan-holding households from vulnerability to shocks.

FINANCIAL STRESS AND STUDENT DEBT

Prior research on financial stress and student debt has focused on severe difficulty and hardship. Student loan-holders experience higher likelihoods of material hardships, including trouble meeting basic needs such as food, medical care, and shelter (Despard et al. 2016). Student debtors also face higher risks of severe debt problems such as bankruptcy and foreclosure (Bricker and Thompson 2016; Maroto 2015; McCloud and Dwyer 2011). Increasing rates of college attendance were associated with increases in foreclosures across neighborhoods during the Great Recession due to the rise in financially overextended households (Faber and Rich 2018).

Focusing on the most severe hardships may only identify strain among the most vulnerable student loan-holders, understating the broader exposure to risk. In other words, studies of severe financial hardship provide only a partial view, especially because student loan-holders are on average more advantaged given that most have achieved at least some higher education. For many households, financial risk can be both more moderate and more generalized than more acute material hardships. To identify risks of economic insecurity among the entire population of student loan holders, we study moderate but still consequential financial stress. Prior work provides some evidence of broader financial stress due to student loans. Holding a student loan is negatively associated with financial wellness overall (Henager and Wilmarth 2018) and students loans are also associated with poorer health (Sweet et al. 2013; Walsemann, Gee, and Gentile 2015), lower asset accumulation (Zhan and Xiang 2018) and delayed life course transitions (Nau, Dwyer, and Hodson 2015).

We conceptualize financial stress as economic difficulty along two dimensions: expense stress and debt stress. Expense stress includes evidence of insufficient income, Specifically, trouble paying bills, which indicates that a household is struggling to make ends meet (Bricker and Thompson 2016; McCloud and Dwyer 2011); and spending more than earned, which signals that a household does not have enough income to meet their spending goals (Baek and DeVaney 2010). Expense stress parallels the more extreme hardship of being unable to meet basic housing and food needs.

Debt stress includes evidence of carrying high-interest unsecured debts, including holding a credit card balance or a payday loan (Rutherford and Fox 2010). Rather than using credit cards out of convenience, holding high interest balances from month to month reveals that a household is short on cash. Even more harmful are payday loans, or small dollar, high interest loans meant to be repaid with a debtor’s next paycheck, a particularly pernicious case of predatory inclusion (Charron-Chénier 2020). These loans are known to trap debtors into cycles of lending—80% of payday loans are renewed or rolled over within two weeks of repayment of the previous loan (Burke et al. 2014). Subprime loans are usually a last resort for those without cash and excluded from less risky types of credit products and thus represent an important indicator of financial fragility (Lusardi, Schneider, and Tufano 2011). Debt stresses and may lead to more severe financial difficulties that manifest in bankruptcy and foreclosure.

FINANCIAL STRESS, STUDENT DEBT, AND INEQUALITIES IN INSECURITY

Financial vulnerability is often latent and difficult to capture (Dwyer 2018). An intense and widespread shock, such as the Great Recession, offers analytical leverage to evaluate the association of student debt with vulnerability to adverse events. We ask whether student debt raises financial stress broadly or whether any association of financial stress concentrates among the most vulnerable populations. To evaluate these associations, we analyze the role of both socioeconomic resources and race/ethnicity in the association between student debt and economic stress. First, we compare the relationships between four alternative measures of student loan burden and financial stress, identifying any global increase in risk associated with student debt holding. Second, we consider whether these relationships vary by race and ethnicity by examining interaction effects. Third, we evaluate whether and to what degree socioeconomic resources mediate any unequal associations between student debt, race/ethnicity, and financial stress.

Financial Stress and Student Debt Burden

We analyze four different dimensions of student loan holding to grasp the role of socioeconomic resources in whether and how student loans are associated with financial stress. Each captures a distinct element of student loan burden, and by comparing across the measures we draw inferences about whether and how much student loans contribute to stress above and beyond socioeconomic resources. First, we analyze a binary indicator of whether a household holds student loans, the most commonly available measure of student loan-holding. Many other associated outcomes appear to fall mainly along the divide between debtors and non-debtors (Despard et al. 2016; Nau et al. 2015). Second, total debt owed captures burden and is particularly important in the US system where there is very little opportunity to cancel any debt. Dwyer and colleagues (2012) find that there are diminishing returns as loan amounts increase for the likelihood of completing college. Pyne and Grodsky (2019) argue that high levels of loans are particularly likely among disadvantaged students pursuing graduate school, a population that may be more likely to experience financial stress. Third, monthly payments capture the impact of student loans on a household’s balance sheet and can be adjusted through the various relief options offered by the federal loan system. When faced with a crisis like the Recession, the total amount owed may be less important than the payments that are due (Lusardi et al. 2011). Finally, the burden of payments relative to income may be most associated with stress, rather than the absolute value of monthly payments. Those with the highest payments may simply have the highest paying jobs, but those whose payments represent a larger portion of their household spending every month may find those payments increase their likelihood of financial stress (Chapman and Dearden 2017).

To the extent that student debt raises risks broadly, all four measures will be associated with financial stress. To the extent that student debt raises risks mainly among those with fewer socioeconomic resources, the association should decline and then disappear with increasing precision relative to current socioeconomic resources from total debt owed to monthly student debt payments, and monthly student debt payment to income ratio. In our first hypothesis, we expect that the association will decline but still hold, indicating broadly spread risk that is worst for those with fewest socioeconomic resources:

H1. Households holding student debt at the beginning of the Great Recession will be more likely to experience financial stress in 2009 than households without student debt. Those with more debt and more burdensome payments will also be more likely to experience financial stress in 2009.

Financial Stress and Student Debt Burden by Race and Ethnicity

The financial stress associated with student loan burden may be worse for Black and Hispanic households compared to White households because of considerable racial inequalities in the returns to college and the ability to pay back student loans. Research has shown significant disadvantages for Black young adults pursuing higher education (Jiménez and Glater 2020; Seamster and Charron-Chénier 2017). Black students are more likely to take on student loans, and on average, borrow more than their White peers (Addo, Houle, and Simon 2016; Jackson and Reynolds 2013). Black students are also more likely to drop out before obtaining a credential, take longer to repay their debts, and face higher rates of default on their student loans (Houle and Addo 2019; Jackson and Reynolds 2013; Scott-Clayton and Li 2016). Studies have also shown labor market discrimination against Black graduates, leading us to expect unequal returns to similar credentials (Gaddis 2015).These disparities between Black and White students are well documented and have only increased over the course of the 2000s.

Less is known about borrowing and repayment among Hispanic populations. Most recent evidence suggests that while Hispanic students are less likely to borrow than Black or White students, those who do face similar troubles to Black students, including dropping out and repayment woes (Chan et al. 2019; Jiménez and Glater 2020). Both Black and Hispanic students are more likely to attend high-cost, low-reward for-profit institutions than White students (Cottom 2017). High levels of student lending to Black and Hispanic students is part of the process of predatory inclusion given that credentials from schools more often attended by these populations produce lower returns relative to schools more often attended by White students (Seamster and Charron-Chénier 2017).

In our second hypothesis we expect that financial stress will be higher for Black and Hispanic households relative to White households. We are particularly interested in examining the dynamics of each of the student loan measures and whether their effects differ by race. We expect that any differential effects will appear principally for the measures of student loan burden that are unadjusted for current resources, including loan-holding, amount of debt, and monthly payments. We expect fewer racial differences for the payment-to-income ratio given that racial differences may be driven by differential repayment resources:

H2. Black and Hispanic households holding student debt at the beginning of the Great Recession will experience higher levels of subsequent financial stress than White households holding student debt especially for the measures of overall burden unadjusted for current resources.

Socioeconomic Mediators by Race and Ethnicity

In our third hypothesis we examine whether socioeconomic indicators mediate any racially unequal deleterious associations between student debt on financial stress over the Recession. We expect that education of the household head, household income, and household assets, may each be important mediators. Education may be important if the association is driven by larger numbers of Black and Hispanic debt holders who do not finish their degrees. Income might help identify any associations resulting from unequal economic returns to degrees by race. Assets captures racial inequalities in stock of resources that households have to help weather the financial shock from the Great Recession. We expect racially disparate effects will remain beyond these indicators as they cannot capture all the potential mechanisms of structural racism.

H3. Socioeconomic mediators including education, income, and wealth will reduce but not eliminate the differential effects by race of student debt burdens at the beginning of the Great Recession and subsequent financial stress.

DATA AND METHODS

We analyze the unique 2007-2009 panel of the Survey of Consumer Finances to examine the relationship between student loans and financial stress. The SCF is a series of nationally representative household-level cross-sectional surveys conducted every three years beginning in 1983 (The Federal Reserve Board of Governors 2018). The Federal Reserve took advantage of the timing of the 2007 wave right before the start of the Great Recession, and re-interviewed respondents in 2009 to understand how the crisis affected households. The re-interview rate was nearly 89%, resulting in a sample of 3,862 households, with nonresponse not strongly correlated with observable characteristics (Kennickell 2010). Our analytic sample includes all households in the 2009 recession survey. We analyze recession experiences with measures from the 2009 survey, based on pre-recession circumstances collected on the same sample from the 2007 survey.

Household Unit of Analysis

The SCF treats households as the unit of analysis for most economic variables because many financial matters involve pooled resources and demands. The sociodemographic variables refer to the head of the household or to the survey respondent, who is either the head or spouse/partner of the head. The SCF designates the man in a different-sex couple or the eldest person in a same-sex couple as the head of household. The 2007-2009 SCF reports data on the education attainment only of the head and spouse, lacking educational experiences of others. We include all households in the sample, because even heads and spouses without any higher education may hold a Parent Plus student loan, have co-signed on a child’s loan, or may live with a student debtor.¹

Measures

Financial stress.

Our outcome is a binary variable for financial stress in 2009, coded as one if a household had expense or debt stress on one or more of five indicators: behind in any loan or mortgage payment², spending exceeded income in the last year (unless spending was high due to a large purchase like a home or car)³, had a revolving credit card balance after last payment, had a revolving balance on a store charge account, or took out a payday loan in the last year. This outcome captures generalized financial stress associated with both debt and expense distress. A household is financially stressed if they experienced at least one of those circumstances.⁴

Student loan burden.

Our main predictor is student loan burden in 2007, just before onset of the Great Recession. We analyze four different measures of student loan holdings and payments: 1) a dummy variable indicating whether the household had any student loans, 2) the amount of money owed on student loans, 3) the monthly payments the household owed on their student loans, and 4) the percent of a household’s monthly income devoted to student loan payments. Both student debt amount and monthly payments are skewed with many zeros and low levels of debt. We prefer the non-transformed measures given the demands of interpretation in the nonlinear categorical modeling context (Long and Mustillo 2018; Mize 2019). Further, the assumption of normality does not apply to independent variables in logistic regression, and natural log transformations may obscure non-linear relationships (Friedline, Masa, and Chowa 2015).

Race and ethnicity.

We construct race/ethnic groups by combining responses to a question on racial identification with responses to a question on Hispanic ethnic origin. The publicly available SCF race variable combines the smaller samples of Asian, Native American, Pacific Islander and other groups into the “other race” category. Respondents can select multiple races and Hispanic as a race category, but the public data only releases the first answer selected. We coded a respondent Hispanic if they identified as such on either the ethnicity or the race question. Then, we follow federal agency practices combining the race and ethnic responses for other groups, while recognizing that these may differ from individual self-identities. (Bhutta et al. 2020). Because the SCF does not include the race and ethnicity of every household member, we must rely on the racial identification of the survey respondent (Lindamood, Hanna, and Bi 2007). Though limiting, this assumption likely holds in the majority of cases given that only 8.4% of married couples in the US had spouses with different race or ethnicities in 2010 (Wang 2012).

Controls.

We control for the main factors that drive financial well-being to assess the extent to which these factors protect against the risks of carrying student debt. We control for age of the head of household because we expect different exposures to financial risk given stage in the life course. We control for household income, divided into quintiles using probability weights so a household’s income quintile is reflective of where they fall in the population distribution. We estimate quintiles to capture the non-linear associations of student loans based on prior findings that middle-income groups carry the most loans relative to more and less advantaged populations (Houle 2014). Results are not sensitive to alternative measures of income. We control for education of the head of household (no higher education, some college, Associate degree, Bachelor’s degree, and advanced degree), and household size (1-13). In some models we include asset quintiles as a control, preferring assets to net worth because the latter includes levels of student debt. We choose quintiles to account for the extreme skewness of wealth, and to capture non-linear associations between wealth and financial stress. Findings are consistent with alternative measures. Finally, we control for 2007 financial stress to account for households that were already experiencing trouble even before the Recession hit. Appendix Table 1 presents descriptive statistics on all variables included in any model.

Limitations.

The structure of the SCF data limits our analysis in some ways. We observe only the beginning of the Great Recession because of the single follow up in 2009. The household unit of analysis limits our capacity to analyze some of the micro-dynamics of loan holding. We are unable to determine which household member carries student loans, which limits the conclusions we can draw. Not being able to identify the credentials and incomes associated with student debts makes it impossible to fully determine the degree to which inequalities stem from degree non-completion, institutional characteristics, or returns to credentials. Thus, we use cautious language when discussing our mediation findings as more suggestive of mechanisms than determinative. A further issue is the increase in enrollment in post-secondary institutions during the recession (Barr and Turner 2013). It is possible that some borrow student loans in response to financial stress rather than student loans causing financial stress. We limit the influence of reverse causality as much as possible in our modeling strategy as we discuss next. As our analysis is based on observational data, it is necessarily not causal in nature.

Modeling approach.

We present three sets of models to assess our three hypotheses. We first estimate logistic regression models of the association of student loan-holding in 2007 with financial stress in 2009, net of previous financial stress to identify those who were particularly affected by the recession itself. Second, we add interaction terms between the student loan variables and race/ethnicity of the respondent to examine racial inequality in the effects of student loan holding on precarity. Finally, we pursue a model-building strategy to test mediators for inequalities among student loan holders.

The SCF data provides imputed data calculated by the staff at the Federal Reserve Board for all missing responses as well as to blur data for privacy purposes, resulting in five separate implicates with no missing values (Kennickell 2011). We follow other studies using the SCF by using repeated-imputation inference methods (Rubin 1987) to combine the imputations in all of our reported models and descriptive statistics (Lindamood et al. 2007; Shin and Hanna 2017).

We use probability weights for all the descriptive statistics to adjust for the sampling approach in the SCF, but follow the general advice to avoid weighting multivariable analyses since weights often account for factors already captured in the controls (Winship and Radbill 1994). Hanna and colleagues find that unweighted analyses in the SCF are the most conservative (Lindamood et al. 2007; Shin and Hanna 2017). In supplemental analyses, we find our results are robust to weighting.

RESULTS

Did households with a member who carried student debt before the Recession face a higher likelihood of financial stress during the Recession compared with similar households without student debt? Our findings support our expectations that student loan-holding was associated with greater financial stress during the recession. The pattern of findings reveals particularly severe racial inequalities in the insecurity associated with holding student debt.

Financial Stress and Student Debt Burden

We report our first set of findings evaluating Hypothesis 1 in the odd-numbered models in Table 1 (models 1, 3, 5, and 7). We report odds ratios, which are best interpreted as the change in the odds of financial stress in 2009 associated with a one unit increase in the independent variable. Consistent with Hypothesis 1, we find that the likelihood of financial stress in 2009 was higher for those with higher student debt burdens, across all four measures. The likelihood of financial stress in 2009 was 146% higher for those with student debt in 2007 than for those without student debt. We find similar results for the three measures of levels student loan burden. The odds of financial stress in 2009 are 1.009 times higher for every thousand dollar increase in total student loan debt in 2007. Further, each dollar increase in monthly payments is associated with a 1.002 times increase in the odds of financial stress in 2009, and each percent increase in payment as percent of income is associated with a 6% increase in odds of financial stress in 2009.⁵ Taken together, these results provide abundant support for Hypothesis 1 that holding a student loan just before the Great Recession and higher levels of student debt burden are associated with increased likelihood of financial stress, even when controlling for financial conditions pre-recession.

Table 1.

Logistic Regression Models Predicting the Odds of Financial Stress

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Has Student Loan	1.791*** (0.232)	1.742*** (0.271)
Student Loan # Black		0.844 (0.295)
Student Loan # Hispanic		1.192 (0.492)
Student Loan # Other		2.655 (1.850)
Amount of Loans			1.009** (0.004)	1.004 (0.004)
Race and Amount Owed (ref=White)
Black # Amount of Loans				1.041+ (0.023)
Hispanic # Amount of Loans				1.027 (0.018)
Other # Amount of Loans				1.029+ (0.017)
Monthly Payments					1.002** (0.000)	1.001* (0.001)
Race # Monthly Payments (ref=white)
Black # Monthly Payments						1.004 (0.003)
Hispanic # Monthly Payments						1.000 (0.001)
Other # Monthly Payments						1.001 (0.002)
Payments as Percent of Income							1.057* (0.029)	1.084* (0.043)
Race # Payments as Percent of Income (ref=white)
Black # Payments as Percent of Income								0.982 (0.081)
Hispanic # Payments as Percent of Income								0.903 (0.064)
Other # Payments as percent of Income								1.398 (0.363)
Financial Stress in 2007	8.849*** (0.727)	8.873*** (0.730)	8.980*** (0.736)	8.977*** (0.738)	8.995*** (0.737)	8.991*** (0.737)	9.675*** (0.781)	9.661*** (0.780)
Race
Black	1.309+ (0.187)	1.358+ (0.219)	1.367* (0.194)	1.225 (0.185)	1.366* (0.194)	1.285+ (0.190)	1.344* (0.185)	1.351* (0.193)
Hispanic	1.254 (0.183)	1.223 (0.194)	1.265 (0.184)	1.173 (0.178)	1.243 (0.181)	1.250 (0.190)	1.254 (0.178)	1.316+ (0.192)
Other	0.713 (0.153)	0.643+ (0.147)	0.702+ (0.150)	0.644* (0.142)	0.696+ (0.149)	0.678+ (0.150)	0.730 (0.154)	0.678+ (0.149)
Age of Head	0.982*** (0.003)	0.981*** (0.003)	0.980*** (0.003)	0.979*** (0.003)	0.980*** (0.003)	0.980*** (0.003)	0.977*** (0.003)	0.977*** (0.003)
Education of Head (ref=no higher ed)
Some College	0.878 (0.108)	0.880 (0.109)	0.907 (0.111)	0.899 (0.111)	0.907 (0.111)	0.905 (0.111)	0.888 (0.107)	0.889 (0.107)
Associate’s Degree	0.617** (0.115)	0.620* (0.116)	0.652* (0.121)	0.635* (0.119)	0.652* (0.121)	0.647* (0.120)	0.677* (0.124)	0.683* (0.125)
Bachelor’s Degree	0.618*** (0.075)	0.622*** (0.076)	0.637*** (0.077)	0.623*** (0.076)	0.634*** (0.077)	0.628*** (0.076)	0.588*** (0.065)	0.585*** (0.064)
Advanced Degree	0.532*** (0.071)	0.535*** (0.071)	0.539*** (0.072)	0.535*** (0.071)	0.541*** (0.072)	0.538*** (0.072)	0.457*** (0.054)	0.457*** (0.054)
Income Quintile (ref=1)
Quintile 2	1.107 (0.163)	1.107 (0.163)	1.107 (0.163)	1.103 (0.163)	1.103 (0.162)	1.099 (0.161)
Quintile 3	1.448* (0.212)	1.442* (0.212)	1.477** (0.216)	1.471** (0.216)	1.460** (0.213)	1.452* (0.212)
Quintile 4	2.095***(0.316)	2.082***(0.313)	2.116***(0.318)	2.086***(0.315)	2.083*** (0.312)	2.071*** (0.311)
Quintile 5	0.788+ (0.108)	0.781+ (0.108)	0.783+ (0.108)	0.775+ (0.107)	0.766+ (0.105)	0.765+ (0.105)
Household Size	1.061+ (0.033)	1.060+ (0.033)	1.060+ (0.033)	1.058+ (0.033)	1.058+ (0.033)	1.058+ (0.033)	1.036 (0.031)	1.036 (0.031)
Constant	0.612* (0.141)	0.622* (0.145)	0.686+ (0.156)	0.719 (0.164)	0.699 (0.158)	0.706 (0.161)	0.976 (0.206)	0.956 (0.203)
Observations	3857	3857	3857	3857	3857	3857	3857	3857

Note: Shows odds ratios and standard errors in parentheses. Based on authors’ calculations of the 2007-2009 panel of the Survey of Consumer Finances, five implicates combined using Rubin’s rules.

⁺p < 0.10,

^*p < 0.05,

^**p < 0.01,

^***p < 0.001

The control variables across all four models largely conform to findings in prior studies. Supporting our goal of capturing moderate forms of financial stress that may be encountered by more advantaged populations rather than severe material hardship, income quintiles 3 and 4 consistently experience higher odds of financial stress than quintile 1, as does income quintile 2 in several models. This finding also highlights the heightened vulnerability of middle-class populations during the recession, whose risk may be less evident during good times, in contrast to lower-income groups who may experience more constant levels of vulnerability across the business cycle. At the same time, the odds of financial stress for households with Black respondents are between 31%-49% higher than households with White respondents. Black households fared worse than their White counterparts over the Recession, even when controlling for income, education, and financial stress in 2007. The results for Hispanic and “other race” versus White is insignificant, perhaps due to the diversity of these groups and smaller sample size.

We estimate predicted probabilities of financial stress for each of the student loan predictors with the controls at their means. Figure 1 reports four panels corresponding to Models 1, 3, 5, and 7 in Table 1. The figure demonstrates the positive association between the predicted probability of financial stress and student loan-holding, the amount owed, the monthly payments, and the burden of the monthly payments. Each panel illustrates the positive relationship between student loan burden and financial stress, with strong positive trends even at lower levels of holding and payments. That each measure of student debt shows a stable and consistent pattern is reassuring. Many data sources on student debt only provide information on incidence and maybe total amount owed, lacking more detailed information on payments. Our findings indicate that incidence and amount of student debt provide valid indicators of loan burden for understanding financial stress and wellbeing in the total population.

Figure 1.

Predicted probability of financial stress in 2009 by student loans in 2007. Based on models odd-numbered models in Table 1. Based on authors’ analysis of the Survey of Consumer Finances, 2007 and 2009 waves. Figures are based on the first implicate while the tables are based on all five implicates combined using Rubin’s rules. Figure results are the same for each of the five implicates.

Financial Stress and Student Debt Burden by Race/Ethnicity

We next turn to Hypothesis 2, in which we expect unequal risk of student debt for Black and Hispanic households compared to White households. In Table 1 we add an interaction of race and ethnicity with each measure of student debt burden in the even-numbered models (2, 4, 6, and 8). We focus our discussion on the predicted probabilities shown in Figures 2 and 3.⁶ In a nonlinear model, the coefficients of interaction terms cannot indicate whether an interaction effect exists, but predicted probabilities are required because the effect on the dependent variable varies across levels of independent variables (Long and Freese 2014; Mize 2019).

Figure 2.

Predicted probability of financial stress by amount of student debt interacted with race/ethnicity of the respondent. Based on even-numbered models in Table 1. Based on authors’ analysis of the Survey of Consumer Finances, 2007 and 2009 waves. Figures are based on the first implicate while the tables are based on all five implicates combined using Rubin’s rules. Figure results are the same for each of the five implicates.

Figure 3.

Predicted probability of financial stress by amount of student debt interacted with race/ethnicity of the respondent. Based on model 6 in Table 1. Based on authors’ analysis of the Survey of Consumer Finances, 2007 and 2009 waves. Figures are based on the first implicate while the tables are based on all five implicates combined using Rubin’s rules. Figure results are the same for each of the five implicates.

As expected in Hypothesis 2, we find that Black and Hispanic households holding student debt experience higher levels of subsequent financial stress, but the association is only for the measures of overall burden. The top left panel shows the results for the binary indicator of student loan holding and demonstrates that the likelihood of financial stress varies by loan status rather than by race. For each group, households of any race carrying debt have a significantly higher probability of experiencing financial stress than those without, consistent with the earlier analysis. While those probabilities vary, there are no statistically significant differences between race/ethnicity categories among those with debt. However, the association with debt-holding masks underlying race/ethnic variability based on loan amounts. The top right panel of Figure 2 shows diverging slopes by race/ethnicity as total amount owed increases. The predicted probability of financial stress among Hispanic and Black households increases much more quickly as amount owed increases compared with similar White households. The bottom panels show estimates for monthly payments and payments as a percent of income. Overlapping confidence intervals throughout demonstrate that when accounting for hardship entering the Recession, there are no unequal effects by race and payment variables. The bottom left panel shows results for the monthly payment measure, demonstrating similar slopes across race/ethnic groups, but with Black households more likely than White households to experience financial stress across all levels of payment. Holding burden equal shows no racial differences, but racial inequalities in employment and income mean that burdens are often unequal by race. The result is the racial disparities we see in financial stress associated with total loan amount.

Figure 3 breaks out the results for total amount owed separately for Black-White and Hispanic-White to better demonstrate racial gaps. The Black/White results show unequal probabilities of stress starting at $20,000 in loans. Hispanic and White households have similar odds of financial stress at lower debt amounts, but by about $80,000 owed, Hispanic households face a higher chance of stress than White households.

Taken together, our findings support Hypothesis 2 and suggests that differences in the effects of debt levels, rather than payments or just having debts, drives the association with financial stress for Black and Hispanic households beyond what is observed in the entire population. Black households especially, are more burdened by higher debt than similar White households, even when holding other important socio-economic variables constant. We observe similar patterns for Hispanic households, although the results are not as robust across all measures of indebtedness.

Black, Hispanic, and White households holding any debt face similar likelihoods of financial stress, but this masks the variability shown when looking at amounts. Could this be driven by differences in the high debt burdens—for instance, some with high student debt have prestigious and lucrative careers, whereas others with high debt burdens may owe for overpriced or predatory undergraduate degrees? Because our data are at the household level, we are unable to be sure which household member is indebted, and for what credential money was borrowed. However, household level resources capture some of the potential mechanisms in the unequal effects of debt burden on financial stress, which we probe next.

Socioeconomic Mediators of Financial Stress and Student Debt Burden by Race and Ethnicity

We evaluate Hypothesis 3 using a model-building strategy to study whether disparities in household socioeconomic resources act as mechanisms that contribute to producing the differential associations between loan holding and financial stress by race/ethnicity. We examine models predicting loan amounts since that is where we observe the most significant racial inequality. Table 2 reports a model-building approach for total loan amount interacted with race and ethnicity. In order to compare effects across models, we show fully standardized coefficients (denoted as Y*) as recommended by Long and Freese (2014), best understood as the expected standard deviation change in the dependent variable for every standard deviation increase in the independent variable. The analysis shows first, evidence of socioeconomic mediation and second, a large persistent racial interaction effect net of mediators.

Table 2.

Logistic Regression Models Predicting the Odds of Financial Stress, by Race/Ethnicity

	(9)		(10)		(11)		(12)		(13)		(14)
	OR	Y*	OR	Y*	OR	Y*	OR	Y*	OR	Y*	OR	Y*
Amount of Loans	1.011** (0.004)	0.041	1.005 (0.003)	0.019	1.001 (0.003)	0.004	1.004 (0.004)	0.015	1.004 (0.004)	0.015	1.004 (0.004)	0.012
Hardship 2007	11.287*** (0.884)	0.366	10.056*** (0.801)	0.324	10.063*** (0.803)	0.321	9.575*** (0.775)	0.304	8.977*** (0.738)	0.286	8.500*** (0.716)	0.273
Race of Respondent (ref=white)
Black			1.583*** (0.216)	0.038	1.445* (0.209)	0.030	1.201 (0.176)	0.014	1.225 (0.185)	0.015	1.271 (0.197)	0.018
Hispanic			1.507** (0.211)	0.033	1.426* (0.207)	0.028	1.170 (0.173)	0.012	1.173 (0.178)	0.012	1.235 (0.190)	0.016
Other			0.654* (0.137)	−0.023	0.597* (0.129)	−0.028	0.677+ (0.148)	−0.021	0.644* (0.142)	−0.022	0.614* (0.137)	−0.025
Head Age			0.975*** (0.003)	−0.111	0.975*** (0.003)	−0.110	0 977*** (0.003)	−0.099	0.979*** (0.003)	−0.085	0.980*** (0.003)	−0.080
Household Size			1.019 (0.030)	0.007	1.018 (0.030)	0.007	1.034 (0.031)	0.013	1.058+ (0.033)	0.021	1.068* (0.034)	0.024
Race and Amount Owed (ref=white)
Black # Amount Owed					1.036+ (0.022)	0.039	1.045* (0.023)	0.046	1.041+ (0.023)	0.041	1.042+ (0.023)	0.041
Hispanic # Amount Owed					1.023 (0.017)	0.024	1.029 (0.018)	0.029	1.027 (0.018)	0.027	1.027 (0.018)	0.026
Other # Amount Owed					1.028+ (0.016)	0.020	1.027 (0.017)	0.018	1.029+ (0.017)	0.018	1.030+ (0.016)	0.019
Education of Head (ref= no highier ed)
Some College							0.881 (0.106)	−0.012	0.899 (0.111)	−0.010	0.927 (0.115)	−0.007
Associate’s Degree							0.655* (0.120)	−0.025	0.635* (0.119)	−0.010	0.650* (0.123)	−0.024
Bachelor’s Degree							0.563*** (0.063)	−0.065	0.623*** (0.076)	−0.052	0.671** (0.085)	−0.043
Advanced Degree							0.436*** (0.052)	−0.092	0.535*** (0.071)	−0.068	0.605*** (0.085)	−0.053
Income Quintiles (ref=1)
Quintile 2									1.103 (0.163)	0.009	1.018 (0.150)	0.002
Quintile 3									1.471** (0.216)	0.036	1.309+ (0.197)	0.025
Quintile 4									2.086*** (0.315)	0.070	1.927*** (0.312)	0.062
Quintile 5									0.775+ (0.107)	−0.032	0.844 (0.134)	−0.021
Assets Quintiles (ref=1)
Quintile 2											2.070*** (0.322)	0.065
Quintile 3											1.757** (0.300)	0.049
Quintile 4											1.492* (0.255)	0.037
Quintile 5											1.025 (0.191)	0.003
Constant	0.224*** (0.013)		0.768 (0.153)		0.785 (0.156)		0.995 (0.210)		0.719 (0.164)		0.514** (0.121)
Observations	3857		3857		3857		3857		3857		3857

Note: Shows odds ratios and standard errors in parentheses. Fully standardized coefficients are denoted as Y*, and are interpretable as the expected change in the dependent variable based on a one standard deviation change in the independent variable and are the appropriate way to compare attenuation across models. Based on authors’ calculations of the 2007-2009 panel of the Survey of Consumer Finances, five implicates combined using Rubin’s rules.

⁺p < 0.10,

^*p < 0.05,

^**p < 0.01,

^***p < 0.001

First, we find support for Hypothesis 3 that socioeconomic resources mediate the interaction between race and student debt on financial stress, as indicated by reduced standardized coefficients (Y*) with added mediators. Model 9 shows the positive association between financial stress in 2009 and the amount of student debt a household owed in 2007, net of 2007 financial stress. Model 10 adds age of household head, the size of the primary economic unit, and race of the respondent. Model 11 adds the interaction between amount of debt and race. Each addition reduces the standardized coefficient for student debt, indicating a mediating effect. In Model 11, for instance, a one standard deviation increase in the interaction term between Black and holding student debt is associated with a .039 standard deviation increase in financial stress. Model 12 adds education of household head, and increases the standardized coefficients for amount owed and the interactions between race and amount owed, meaning that once education of household head is considered, loan amounts are more important for predicting financial stress. Model 13 includes income quintiles, which reduces both the standardized coefficients for the interaction terms and statistical significance. As expected in Hypothesis 3, income helps explain the unequal returns to student debt on financial stress by race. The standardized coefficient for the interaction term between Black and amount of student debt decreases from 0.046 to 0.041 after income is accounted for. Adding in a measure of assets does not affect the coefficient size of the interaction terms, although it decreases the main effect of debt holding.

Taken together, these results indicate that differential effects between Black and White households of total student debt levels are driven by unequal returns to education and income in protectiveness from financial stress during the Great Recession. Against expectations, education of head increases the standardized coefficient for loan amount. We take this as suggestive evidence supporting theories that loan amounts capture some of the variation in educational institutions that is associated both with total costs and the returns to education. Assets are also less influential mediators than we expected. In supplemental analyses we find a stronger mediating effect for assets when we exclude 2007 financial stress in the model. Thus, it appears that wealth operates mainly through pre-existing racial disparities in insecurity.

Second, we find support for our expectation in Hypothesis 3 that the racial interaction in the relationship between student debt and financial stress will remain highly significant net of socioeconomic mediators. Figure 4 shows the predicted probability of financial stress in 2009 for Black and White households over amount of debt, holding all covariates at their means, across Models 11-14. Results for Hispanic households are very similar to those for Black households, so we exclude for parsimony’s sake. Each of these models includes the interactions, which allows the slopes to vary by race. We find the relationship for predicted probabilities remains largely unchanged across models even in the fully saturated model with socioeconomic mediators. While the standardized coefficients showed some attenuation in a non-linear model (as discussed earlier), “it is not possible to determine the nature of an interaction effect on the predicted probabilities in logit/probit models based on the coefficients alone” (Mize 2019:94). The overall disparity by race remains highly significant. The implication is that socioeconomic mediators—at least at the household level available in the SCF—are insufficient to capture the racial disparities in student debt. In the most saturated model, having higher amounts of student loan debt has little effect on the predicted probability of financial stress for households with a White respondent: when socioeconomic resources are held at their means, higher levels of debt are not associated with higher levels of financial stress for White households. Among Black households, in contrast, each model shows a consistent positive relationship between more student debt and financial stress. We suspect one reason for the relatively modest effect of socioeconomic mediators is that 2007 financial stress already carries so much of those selection dynamics. The net effect with that prior vulnerability controlled thus appears to be even more driven by the differential effects of race beyond socioeconomic mediators that we interpret as likely reflecting structural racism.

Figure 4.

Predicted probability of financial stress by amount of student debt interacted with race/ethnicity of the respondent. Based on models 11-14 in Table 2. Based on authors’ analysis of the Survey of Consumer Finances, 2007 and 2009 waves. Figures are based on the first implicate while the tables are based on all five implicates combined using Rubin’s rules. Figure results are the same for each of the five implicates.

The racial disparity in returns to debt holding on the odds of financial stress during hard times is stark. The findings are consistent with research that argues that higher overall levels of student debt for Black and Hispanic students raise financial vulnerabilities in ways that may be overlooked with other measures of student debt burden. The persistence of that interaction even when controlling for household income and wealth suggests more research is needed to fully understand the mechanisms behind the unequal effects of student debt burdens on Black and Hispanic households compared to White households.

CONCLUSION

In sum, our analysis demonstrates a strong association between household student debt holding at the beginning of the Great Recession and financial stress during the crisis, even controlling for socioeconomic position and financial stress entering the recession. We find that the risks of student debt are not equal amongst households but differ for race-ethnic groups for total loan levels. Elevated odds of financial stress start at lower levels of total loan amounts and increase at a steeper slope for households with Black and Hispanic respondents compared to White respondents. These findings contribute to the growing consensus that student debt is disproportionately risky for households of color (Jackson and Reynolds 2013; Seamster and Charron-Chénier 2017). Black-White disparities in debt only increase across the life course as Black populations carry debt later in life, putting them at financial risk for a longer period of time (Houle and Addo 2019).

Our findings relative to alternative student loan burdens also provide valuable methodological insights in this area of research. Among all households, an indicator of loan-holding will capture student loan stress in the absence of more detailed measures. However, that measure may mask important race-ethnic disparities in total levels of debt held. Given that data sources are more likely to include indicators of debt-holding than levels, payments, and debt-to-income ratios, studies of other varied outcomes should consider the possible implications of alternative measures on the mechanisms expected to influence findings. Our results that there were fewer differences on the effects of payments and payments-to-income ratio supports the expectations derived from prior research that institutional differences that affect total loan amounts and wage returns to college are significant drivers of differences in the effect of student debt on financial stress (Pyne and Grodsky 2019; Seamster and Charron-Chénier 2017). The similarity in the payments to income ratio across race-ethnic groups provides some basis for cautious optimism that racial and ethnic disparities in student debt—including the associated financial stress during economic downturns—could be reduced if the institutional conditions that drive disparities in monthly burdens and incomes were reduced through anti-racist policies. At the same time, there remain significant racial disparities in wealth, family backgrounds, and social networks that also need to be addressed in addition to the student loan disparities (Addo et al. 2016; Houle and Addo 2019; McCabe and Jackson 2016). The overall loan burden is simultaneously the least likely to be adjusted with any forbearance from the federal loan system and most likely to diverge by race and ethnicity.

Our findings suggest at least three promising directions for future research. First, it would also be valuable to know how financial stress unfolds over time and whether and how households recover. Which households overcome misfortune? Which households fall into deeper difficulties? The long-term trajectory of these episodes may be particularly important since student debt cannot be discharged in bankruptcy. At the same time, the various deferment and forbearance programs associated with student debt may make trouble associated with student debt more short-term and manageable over time. Second, it will also be valuable to compare the risks of student debt with other forms of debt, and for intersecting risks among those who amass multiple liabilities relative to their resources. The findings here raise questions about the relationship between financial challenges and the accrual of high-cost debts such as credit cards and payday loans. Finally, we hope that our analysis will inform continuing efforts to understand the experience of financial stress among student loan holders during a range of shocks. The recession is a case of a widespread crisis, but everyday adverse events such as cuts in pay, job loss, family dissolution, and medical disasters can also trigger hardships and financial problems (Despard et al. 2018; Hacker 2006; McCloud and Dwyer 2011; Western et al. 2012). Future work should examine whether our findings during an economy-wide crisis hold when households face these more personal crises. It will be particularly important in future research on all of these dynamics to study inequalities by race and ethnicity and within racial groups by class (McCabe and Jackson 2016; Seamster and Charron-Chénier 2017).

Our findings that household student debt holding is associated with elevated financial stress during the Great Recession highlights the damaging societal consequences of disinvestment in public goods in the United States that has shifted risk from the state to the populace. Both state and federal governments once provided much financial support for institutions of higher education, subsidizing the costs of tuition for students (Eaton et al. 2016). It is only since these funding sources have dried up that student loans became as common and large as they are today. There are also more intangible results of a system that includes significant risk. Students aware of the risks of loans may be debt-averse to such a degree that they may not even apply for financial aid or chose low-cost two-year colleges without considering a four-year path (Ovink 2016).

Fortunately, there is new energy around policy reform in the student loan system, and growing appreciation that any reforms need to be explicitly anti-racist to avoid reinforcing existing racial inequalities. Allowing student loans to be discharged in bankruptcy is one commonly suggested change that would help a subset of borrowers. Income based repayment programs were expanded after the Great Recession in part to reduce the risk of loans. However, the promises of income-based repayment plans often came up short, with more interest paid in total and a high tax burden on any forgiven balances, as well as the risk of getting kicked off a plan if a household misses yearly paperwork deadlines (Goldrick-Rab 2016). Deferment and zero interest accrual as well as emergency grants for students were include in the CARES Act, one of the early relief packages in the COVID-19 crisis (US Department of Education 2020). Yet, the CARES act provisions only deferred payment rather than canceled payments permanently or reduced debt levels. As important as federal reform initiatives are, additional innovative reform actions by state and public university coalitions have also focused on increasing state funding for higher education overall and grant aid for low-income students in particular (Eaton et al. 2016).

Student loans are not the only source of financial risk in family portfolios, which also include housing, retirement, and saving for the next inevitable financial crisis. The disinvestment in the public sector has also reduced job opportunities a for Black populations especially, making it potentially even harder for these populations to handle increasing loan burdens (Wilson, Roscigno, and Huffman 2015). Thus, while there are important steps to be taken to control costs in higher education and reform financial aid to be more equitable and effective, there are broader issues of rising financial risk, slowing income growth, and spreading precarity in American society that may be just as important in shaping college affordability (Dwyer 2018). What is needed is a broad-based defense of public goods and social insurance that protects from all the risks associated with capitalist downturns and ecological crises. The COVID-19 crisis highlights just how much we all depend on our public commitments, and just how serious the risks we face when those bonds are torn. We hope that conversations about student loans can be leveraged to broader conversations in part by encouraging a recognition that shared risks require shared solutions.

APPENDIX A.

Table A1.

Descriptive Statistics.

	HHs Without Student Loans		HHs Carrying Student Loans
Variables	Mean	SD	Mean	SD
Financial Stress in 2009	48.85%	0.50	71.17%	0.50
Behind Loan	15.56%	0.38	29.07%	0.38
Revolving CC Balance	36.65%	0.49	52.56%	0.49
Revolving Store Balance	14.52%	0.36	21.80%	0.36
Spent More than Income	1.65%	0.13	2.26%	0.13
Took out Payday Loan	3.79%	0.20	5.77%	0.20
Student Loan Burden in 2007
Amount of Loans (thousands)	$22.02	28.41	$22.02	28.41
Monthly Payments	$141.91	201.44	$141.91	201.44
Payments as % of Income	2.60%	3.72	2.60%	3.72
Household Characteristics
Size of Household	2.39	1.40	2.79	1.40
Age of Household Head	51.54	16.80	38.44	16.80
Education of Household Head
No Higher Education	49.04%	0.50	23.81%	0.50
Some College	17.91%	0.39	22.04%	0.39
Associate’s Degree	5.29%	0.24	10.87%	0.24
Bachelor’s Degree	16.30%	0.39	29.47%	0.39
Advanced Degree	11.45%	0.32	13.80%	0.32
Race/Ethnicity of Respondent
Non-Hispanic White	71.74%	0.46	65.31%	0.46
Non-Hispanic Black	11.80%	0.34	19.62%	0.34
Hispanic	12.13%	0.33	12.10%	0.33
Non-Hispanic other	4.33%	0.20	2.97%	0.20
Total Income in 2007	$80207.9	249887.55	$78761.2	74345.594
Total Assets in 2007	$597671	2177985.6	$307056	406959.22
Observations	3,232 (84%)		625 (16%)

Note: The means and standard deviations are weighted using population weights. Based on authors’ analyses of the Survey of Consumer Finances, 2009 follow up to 2007 wave, five implicates combined using Rubin’s rules.