Journal of Family and Economic Issues

, Volume 37, Issue 1, pp 42–57

Household Debt and Adult Depressive Symptoms in the United States

Authors

  • Lawrence M. Berger
    • Institute for Research on Poverty, School of Social WorkUniversity of Wisconsin-Madison
  • J. Michael Collins
    • Center for Financial Security, La Follette School of Public Affairs and The School of Human EcologyUniversity of Wisconsin-Madison
    • Institute for Research on Poverty, School of Social WorkUniversity of Wisconsin-Madison
Original Paper

DOI: 10.1007/s10834-015-9443-6

Cite this article as:
Berger, L.M., Collins, J. & Cuesta, L. J Fam Econ Iss (2016) 37: 42. doi:10.1007/s10834-015-9443-6

Abstract

This study used data from Waves 1 (1987–1989) and 2 (1992–1994) of the National Survey of Families and Households in the United States and a series of regression models, some of which included individual-specific fixed effects, to estimate associations of particular types and levels of debt with adult depressive symptoms. Results suggest that household debt is positively associated with greater depressive symptoms. However, this association appears to be driven by short-term (unsecured) debt; we found little evidence of associations with depressive symptoms for mid- or long-term debt. The link between short-term debt and depressive symptoms persisted with alternative estimation strategies, including defining debt in absolute and relative terms. Furthermore, this association was particularly concentrated among 51–64 year-old adults, those with a high school education or less, and those who were not stably married throughout the observation period. These findings suggest that short-term debt may have an adverse influence on psychological wellbeing, particularly for those who are less educated, approaching retirement age, or unmarried.

Keywords

DebtDepressive symptomsNational Survey of Families and Households

Introduction

Household debt increased dramatically in the United States during the last four decades. Between 1962 and 2008, the median ratio of household debt to income rose from 0.1 to 0.6 and aggregate household debt rose from about 60 % to about 120 % of aggregate household income (Dynan 2009). Much of this trend was spurred by increased homeownership financed by mortgages, but it also reflects increased accumulation of unsecured, revolving credit card debt (Durkin 2000; Xiao and Yao 2011a, 2011b). Household debt levels have declined since 2008, as credit has become less available, households have reduced spending, and debt has been forgiven through bankruptcy filings, but it remains at historically high levels and is subject to much discussion by policymakers (US Senate 2011).

Whereas the ability to borrow has clear benefits—allowing individuals and households to smooth consumption and invest in homeownership, human capital acquisition, and other large ticket items that they cannot fully pay for in the present (Dynan and Kohn 2007; Hyman 2011)—it may also result in increased financial pressure given that debt must eventually be repaid. To date, the small body of research that has specifically focused on household debt has identified adverse associations with financial and other stress (Norvilitis and MacLean 2010; Watson et al. 2014; Worthington 2006), adult physical health (Drentea and Lavrakas 2000; Keese and Schmitz 2014; Lenton and Mosley 2008), medical care (Kalousova and Burgard 2013, 2014), college completion (Dwyer et al. 2012), and marital quality (Dew 2007, 2008). Yet, findings regarding associations between debt and psychological wellbeing have not been completely consistent. Studies have found debt to be associated with poorer psychological functioning (Brown et al. 2005), increased probability of mental disorder (Jenkins et al. 2008), and higher levels of anger (Drentea and Reynolds 2012), anxiety (Drentea 2000; Drentea and Reynolds 2012), and depression (Bridges and Disney 2010; Drentea and Reynolds 2012; Gathergood 2012). At the same time, however, Dew (2007) has found debt to be associated with less depression among married couples, and Dwyer et al. (2011) have found positive associations between debt and both self-esteem and mastery.

Much remains unknown about the nature of the associations between debt and mental health. First, as noted above, results from prior studies are not conclusive. Second, social selection and reverse causality pose formidable challenges to estimating unbiased associations. With respect to social selection, other individual and household characteristics may influence both debt accumulation and psychological wellbeing such that correlations between the two are spurious. For example, factors such as unemployment or limited access to economic resources may both induce one to borrow, and also adversely influence one’s psychological wellbeing. It is also possible that psychological wellbeing influences debt accumulation as much as, or more than, debt accumulation influences psychological wellbeing. Most existing analyses have been unable to adequately adjust for these possibilities.

In addition, prior work has not fully considered potential heterogeneity in associations between debt and psychological wellbeing, even though links between debt and psychological wellbeing may differ depending on whether debt was accumulated for investment in education or homeownership versus immediate consumption, the debtor’s age or proximity to retirement, and the debtor’s income or asset levels. Particular types of debt may be differentially associated with psychological wellbeing based on common factors that determine both economic status and mental health.

This study adds to a small but growing literature on associations between debt and psychological wellbeing. We used data on roughly 8500 individuals, who were interviewed between 1987 and 1989 and again between 1992 and 1994 as part of the National Survey of Families and Households (NSFH), to examine associations of particular types and amounts of debt with adult depressive symptoms. The NSFH included a nationally representative sample of US households. It was fielded during a period characterized by rapid expansion of (particularly short-term, unsecured) debt. We estimated associations between debt and depressive symptoms through a series of standard ordinary least squares (OLS) regressions with extensive controls and OLS regressions with individual-specific fixed effects. We examined these associations in terms of whether a household had accrued any debt as well as the amount of debt it had accrued, focusing not only on total debt, but also on particular types (short-, mid-, and long-term) of debt. Additionally, we tested the robustness of our results when debt was modeled in terms of absolute level and when it was modeled as a proportion of annual income and total assets. Finally, we conducted subgroup analyses to examine whether associations of particular types and amounts of debt with adult depressive symptoms varied by age, educational attainment, and whether an individual was stably married throughout the observation period. We found that accumulation of short-term debt was associated with greater depressive symptoms. Furthermore, this association was particularly concentrated among 51–64 year-old adults, those with a high school education or less, and those who were not stably married throughout the observation period. These findings suggest that short-term debt may have an adverse influence on psychological wellbeing, particularly for those who are less educated, approaching retirement age, or unmarried. Financial professionals, educators and counselors should be aware of the coincidence of depression and high levels of short-term debt, particularly for less-advantaged adults and those approaching retirement age.

Conceptual and Theoretical Framework

Social Stress Theory (Pearlin 1989) and the Family Stress Model (Conger and Elder 1994; Conger et al. 1990)—hereafter “stress theory”— suggest that, whereas debt may help alleviate economic stress in the short-term (Dwyer et al. 2011), over the long-term, debt burdens may lead to increased economic stress and, thereby, decreased psychological wellbeing. Specifically, borrowing provides expanded opportunities to purchase goods and services that could not otherwise be purchased. As such, debt may be positively associated with wellbeing by allowing individuals and households to maintain or increase consumption as well as to make long-term investments. At the same time, however, debt burden may be inversely associated with (particularly longer-term) wellbeing both directly, because resources must be allocated to debt repayment, and indirectly, as a result of increased financial or other stress. For example, if high levels of debt require that a significant portion of income be allocated to debt repayment, then the potential benefits of borrowing may be offset by financial pressure or distress (Conger and Elder 1994; Conger et al. 1990), which may precipitate declines in psychological wellbeing.

Three aspects of debt are particularly important to understand its association with depressive symptoms: (1) agency in borrowing (the degree to which one’s economic choice set is constrained), (2) magnitude (the amount of debt, often relative to income or assets), and (3), cost (total charges and fees incurred during the full period over which debt is repaid). Cost is closely linked to the default risk of each type of debt. Most notably, unsecured (short-term) debt is more expensive than secured (long-term) debt since the lender has no collateral and default rates are relatively high. Cost and magnitude are also linked. Long-term debt, such as a home or education loan, often involves a large principal amount, but is amortized such that some of the initial principal is repaid each month. These loans also have lower annual interest rates and are frequently viewed as an investment by the debtor. In contrast, short-term unsecured debt costs considerably more than long-term debt, particularly if not paid off quickly and balances revolve such that little of the initial balance is paid off.

Existing studies have predominantly focused either on total debt or on a specific type of debt (home, education, auto, or unsecured). Few have simultaneously considered the full range of types and amounts of debt that households may accumulate. This is problematic for several reasons. First, different types of debt have different associated costs (interest rates and fees), which affect the amount that can be used for consumption or investment. Notably, (typically low-cost and amortized) long-term debt is generally used for asset or human capital investment, which may be positively associated with psychological wellbeing. In contrast, (higher cost and revolving) unsecured debt is more often used for immediate consumption and may be negatively associated with psychological wellbeing, particularly over time or if incurred in a context of limited agency. Second, there is likely to be social selection into particular types of debt based on factors such as socioeconomic status (SES) and associated financial literacy (ability to fully comprehend the implications of loan terms). This may reflect borrower decisions vis-a-vis types of debt to pursue as well as a household’s ability to qualify for particular types of loans (Lusardi and Tufano 2009). Third, specific types of debt may be fungible (e.g., home equity loans may be taken to repay unsecured debt; larger education loans may substitute for unsecured debt in order to cover living expenses during schooling) and individuals may substitute or move between various types of debt for the same purposes (consumption, investment) given differences in costs (interest rates and fees). Thus, it is important to account for the full range of types and amounts of debt affecting a household at any point in time.

Individual and household characteristics may influence one’s ability to qualify for credit, one’s propensity to borrow, and the types of debt one is likely qualify for and/or utilize. Such factors may include SES, life stage, and marital status. In regards to SES, stress theory predicts that households with limited resources are more vulnerable to potentially stress-inducing factors—such as debt burden—than those with greater access to resources (Pearlin 1989). In addition, debt incurred in response to traumatic events is thought to be particularly stress-inducing (McCloud and Dwyer 2011; Sullivan et al. 2000). Disadvantaged households are especially likely to incur debt under such circumstances. Indeed, they tend to incur debt with less agency, of greater relative (though not necessarily absolute) magnitude, and at a higher cost. Most low-SES households lack the liquid assets necessary to support consumption at the federal poverty level for 3 months without income (McKernan and Ratcliffe 2009). Research further suggests that having insufficient funds to meet basic needs encourages borrowing, even at high-cost (Shah et al. 2012). That disadvantaged households tend to use debt to meet basic needs, whereas affluent households do so as a convenience or investment strategy (Dwyer et al. 2011, 2012; Sullivan et al. 2000), implies that, on average, wealthier households exercise greater agency in borrowing. This is also consistent with studies that have shown correlations between self-efficacy and overall levels of wealth (e.g., Lown et al. 2014).

The influence of debt on psychological wellbeing may vary by life stage as well. Lifecycle theory (Attanasio and Weber 2010; Modigliani 1986; Modigliani and Brumberg 1954; Shefrin and Thaler 1988) suggests that individuals and families borrow to meet consumption needs at particular life stages and save (or repay debt) at others. For example, education debt tends to accumulate in early-adulthood and home loans in mid-adulthood. These large but low cost loans are typically paid off over a relatively long period, but are not expected to persist into older-adulthood; in reality, however, they often do (Mann 2011; Federal Reserve Bank of New York 2012). By contrast, unsecured debt may occur throughout the life course, sometimes in a context of limited agency, and it may be most problematic if held in mid- to late-adulthood when income streams attenuate (Yilmazer and Devaney 2005; Xiao and Yao 2011b). As such, the influence of debt on psychological wellbeing may vary by the life course stage at which an individual experiences debt. Specifically, there are likely to be stronger associations between debt and psychological wellbeing among older working-age adults than among younger ones. Yet, despite considerable research that has linked debt accumulation and repayment to life stages (Azicorbe et al. 2003; Baek and Hong 2004; Brown and Taylor 2008; Xiao and Yao 2011b), little has focused on how associations between debt and wellbeing may vary depending on the life stage at which they are experienced, nor by the type of debt held during a given life stage.

Family structure may also influence debt levels. Whereas married households and households with children have greater average amounts of debt (Baek and Hong 2004; Yilmazer and Devaney 2005) than other households, cohabiting and single-parent households are disproportionately likely to experience debt-related hardship and delinquency (Xiao and Yao 2011a).These hardships may exacerbate adverse outcomes associated with social disadvantage, including compromised psychological wellbeing.

Literature Review

Only a handful of studies have examined associations between debt and various forms of psychological wellbeing. Whereas most have utilized longitudinal data, few have employed rigorous methods for dealing with social selection and reverse causality. Of the 11 studies of which we are aware, 6 used UK data, 4 used US data, and 1 used German data. Two studies used cross-sectional data. Drentea (2000) used a sample of adults in Ohio from 1997 and found links between credit card debt and anxiety, with stress playing a major mediating role in these associations. Furthermore, she found higher levels of anxiety among younger adults, which she concluded reflected higher levels of stress about credit card debt for these individuals relative to their older counterparts. Jenkins et al. (2008) analyzed UK data and found a positive association between the number of debts held in the past year and the probability of a mental disorder. However, the possibility that these findings were driven by social selection and/or reverse causality cannot be ruled out given that both studies relied on point-in-time data.

The nine additional studies of which we are aware all utilized longitudinal data. Yet, there is considerable variation in the methods employed to address social selection and reverse causality. Despite having longitudinal data, several studies did not analyze repeated measures of the predictors and outcomes. For example, Drentea and Reynolds (2012) used two waves of data on a sample of US adults in the Miami area. They found that having any debt was associated with higher levels of depression, anxiety, and anger. However, both debt and the outcome variables were measured at the same point in time, the follow-up interview, with only the control variables measured at baseline. Thus, the causal direction of association between debt and psychological functioning could not be determined. Reading and Reynolds (2001) used two waves of data, measured 6 months apart. They found that worrying about debt at baseline was associated with postnatal depression at follow-up. However, their models did not account for earlier depression status or changes in debt between observation points.

Using two waves of data from the NSFH (the data source used in the current study), Dew (2007) examined associations of short-term (unsecured or consumer) debt with marital conflict using a sample of consistently married individuals and structural equation modeling. Most relevant to this study, he found that, controlling for assets and a composite measure of “economic pressure,” constructed from measures of frequency of worrying about paying bills and satisfaction with finances, short-term debt was associated with fewer later depressive symptoms. However, it is important to note that the analyses estimated only between-individual differences in short-term debt levels at baseline (Wave 1) and between-individual levels of depressive symptoms at follow-up (Wave 2). Changes in individuals’ own debt levels were not used to estimate changes in depressive symptoms over time. Notably, the study focused only on short-term debt rather than all types of debt that an individual may have held.

Dwyer et al. (2011) used longitudinal data from the 1997 National Longitudinal Survey of Youth in the United States and found positive associations of both credit card and education debt with both self-esteem and mastery among young adults after adjusting for earlier measures of these outcomes. The authors posited that the direction of these associations suggested that there may be psychological benefits to borrowing for young adults who accumulate debt as a form of human capital investment. Furthermore, these associations were concentrated among lower-SES and younger young adults, implying that debt may have a more positive influence for less advantaged young adults who are investing in their future than for their more advantaged counterparts, as well as that such investments may have a more positive influence on psychological wellbeing when made at an earlier rather than later period of young adulthood.

The most rigorous studies to date have employed empirical strategies that have taken full advantage of longitudinal data to estimate within-individual associations between changes in debt and changes in psychological wellbeing over time. Three prior studies exemplify this strategy. Keese and Schmitz (2014) used German panel data and both lagged dependent variable and fixed-effects regressions and found inverse associations of consumer and housing debt with both physical health satisfaction and overall mental health. Lenton and Mosley (2008) used UK data and a simultaneous-equation generalized probit model to estimate relationships between debt burden (size) and structure (high, medium, or low interest rate) with physical and psychological health. They found evidence that relationships between these factors were bi-directional, and also that repayment structure mattered, such that higher interest (unsecured) debt was more strongly inversely linked with health than was lower interest debt. Bridges and Disney (2010) used data from a longitudinal household survey in the UK and fixed-effects regressions to estimate associations between debt problems (difficulty repaying debt) and the probability of depression. They found that individuals who experienced the onset of debt problems also exhibited an increased probability of depression.

Two additional studies used instrumental variables methods to leverage variation in debt accumulation which was determined only by factors that did not directly influence psychological wellbeing (i.e., was exogenously determined). This strategy has the potential to account for social selection and reverse causality when estimating associations between debt and psychological wellbeing and thereby produce unbiased estimates that justify causal interpretation. Brown et al. (2005), for example, used British Household Panel Survey data to estimate effects using a variety of instrumental variables. They found that outstanding non-mortgage debt was associated with poorer overall psychological wellbeing across multiple model specifications; this was not true for mortgage debt. However, their instrument for predicting debt was whether an individual had a credit card. This is unlikely to have been a valid instrument as it is not exogenous to debt accumulation. Finally, Gathergood (2012) used UK panel data and variation in local housing prices to isolate the exogenous component of what he called “problem mortgage debt.” He found such debt to be associated with decreased general psychological wellbeing and anxiety-related illness. However, the “problem debt” measure assessed difficulty paying housing costs and whether consumer credit repayment was a “heavy burden,” both of which may reflect (perceived) material hardship or limited income rather than actual debt burden.

The current study extends prior literature in three ways. First, we utilized a large, nationally representative US dataset in which various types and amounts of debt as well as depressive symptoms were measured at two time points. These data were collected during a period in which unsecured debt expanded rapidly among US households. They allowed us to estimate associations of changes in debt and changes in depressive symptoms across an approximately 6-year interval. Although this time period was necessitated by the NSFH data, which were collected between 1987 and 1989 (Wave 1) and 1992 and 1994 (Wave 2), it is a relevant period over which to study these associations. For example, personal bankruptcy in the United States remains in credit bureau and other records for 7–10 years. It also takes debtors several years to recover from a default and return to asset and debt levels experienced before a negative shock (Jagtiani and Li 2013). Overall, 6 years is thus an appropriate time frame over which to consider substantial changes in debt and their influences on psychological wellbeing. It is also notable that the 1987–1994 period pre-dated bankruptcy reform in the United States, which shifted borrowing behavior in the 2000s (Bird et al. 1999). It also pre-dated the rise of subprime mortgage lending (Chomsisengphet and Pennington-Cross 2006). Avoiding these confounding periods is a benefit of using the NSFH.

Second, we exploited the longitudinal nature of the NSFH to estimate a series of OLS models with individual-specific fixed effects. This strategy adjusted for time-invariant unobserved factors with persistent effects. It was therefore a more rigorous identification strategy than has previously been employed with US data.

Third, we included in our models the three major categories of debt held by households: short-term (unsecured), mid-term (non-mortgage bank, installment, auto, and personal), and long-term (home mortgage and education) debt. We also modeled debt in three ways: as a dichotomous indicator of whether a household had debt and as continuous measures of absolute and relative (to income and assets) amounts of debt. For each measure, we estimated models focusing on total debt as well as models focusing on particular types of debt. Finally, we tested whether associations between debt and depressive symptoms varied by age, which is a proxy for life stage, by educational attainment, which is an important indicator of SES, and by marital stability which, in addition to being a measure of SES, is related to the ability of households to have stable or multiple incomes (and the focus of at least one prior study).

Our analyses tested the hypothesis that debt burden would be positively associated with adult depressive symptoms over an approximately 6-year time period. In light of prior theory and evidence, we expected that short-term debt, which tends to be of high cost and is often incurred for immediate consumption and with limited agency, would be associated with increased depressive symptoms. Our a priori expectations for medium- and long-term debt were ambiguous. Furthermore, we expected the magnitude of (particularly short-term) debt to be positively associated with depressive symptoms. Although we assumed consumption and economic pressure are the primary mechanisms through which these associations would operate, we estimated only the direct effects of debt on depressive symptoms, rather than examining potential mediators. We took this approach for two reasons. First, we focused on estimating the full association of debt with depressive symptoms, rather than parsing out the portion of this association which was explained through each of these mechanisms. That is, we were ultimately interested in whether there is likely to be a causal link between debt and depressive symptoms. Second, our data did not include high quality measures of consumption or economic stress, making adequate measurement of these mediators problematic.

Data and Measures

Sample

Our analyses used data from the first two waves of the NSFH, a longitudinal, nationally representative household survey that started with a sample of 13,007 respondents in Wave 1 (1987–1989), and re-interviewed 10,005 of these respondents in Wave 2 (1992–1994). Households were randomly selected from 1700 selection units that resulted from selecting 17 enumeration districts within each of the 100 primary sampling units; the interview response rate was 74 % (for further detail, see Sweet et al. 1988; Sweet and Bumpass 1996). The primary respondent was a person living in the household who was randomly selected using a Kish grid; this person was asked to provide detailed information about his or her family of origin; current family structure, composition, and living arrangements; fertility, education and employment history; earnings and income; health and mental health; and household assets and debt.

Missing Data

We utilized multiple imputation techniques to impute values for all variables with missing data for the full initial NSFH sample (N = 13,007). Specifically, we imputed 25 complete datasets using Stata’s MI program (StataCorp 2013). We then limited our sample to observations of individuals who were between 21 and 65 years of age in both survey waves and who did not own their own business in either wave. We excluded individuals younger than 21 and older than 65 to focus on working-age adults. We excluded business owners in order to ensure that our debt measures reflected personal, rather than business, debt. These criteria resulted in the exclusion of just over a third of the initial sample. Our final analysis consisted of 16,964–17,059 individual-wave observations (two observations per respondent) of 8457–8516 individuals per dataset, across the 25 imputed datasets.

Measures

Our outcome of interest was adult depressive symptoms which were measured by the Center for Epidemiologic Studies Depression Scale (CES-D) partial scale (Radolff 1977). The CES-D partial scale is a truncated version of the full CES-D, a self-reported measure of depressive symptoms that was designed for research in the general population. At each NSFH interview, respondents were asked to report the frequency with which they experienced a series of 12 depressive symptoms. Individuals were asked to report how many days in the last week they experienced each symptom on a 0- to 7-point scale. Following the standard protocol described in Radolff (1977), individuals were assigned 0 points for each symptom that they did not experience during the past week, 1 point for each symptom experienced for 1–2 days, 2 points for each symptom experienced for 3–4 days, and 3 points for each symptom experienced for 5–7 days. An individual’s overall score (0–36 points) for the scale was then constructed by summing the scores on the individual items.

As noted above, prior literature has been inconsistent with regard to how best to model household debt. As such, we modeled debt in several ways, which allowed us to test the robustness of our findings to multiple specifications of our key predictor. The first specification consisted of a simple dichotomous variable indicating (1 = yes) that a household had any debt, defined to include credit card debt, installment loans, bank loans, loans from friends, (overdue) bills owed for more than 2 months, vehicle debt, home improvement loans, education debt, and mortgage debt. The second measure consisted of the logarithm of total household debt, assessed by summing the amounts of each of these types of debt.

In addition, we estimated associations of specific types of debt with adult depressive symptoms. Here, we considered three types of debt: short-term, mid-term, and long-term. For each, we constructed both a dichotomous measure of whether the household had any debt of that type and also a continuous measure of the (logarithm of the) total amount of that type of debt. Short-term debt included credit card debt and overdue bills (bills owed for more than 2 months). The variable was not sensitive to the inclusion of unpaid bills. However, because juggling due dates is a common strategy for households to extend liquidity, we determined it was important to include overdue bills as a form of short-term debt (the costs of which are a late fee or other penalty). Mid-term debt consisted of installment loans, home improvement loans, vehicle loans, loans from friends, and other bank loans. Long-term debt consisted of education and mortgage debt. Finally, in order to test the robustness of our results using absolute measures of amounts of debt to those achieved when relative debt measures were employed, we also constructed a debt-to-annual income ratio and a debt-to-assets ratio, for total debt as well as for each type of debt (short-, mid- and long-term).

Each of the debt measures was self-reported by the respondent. Specifically, respondents were asked whether they had each type of debt and, if they responded affirmatively, how much they owed. The accuracy of self-reported debt data is a serious concern. Evidence suggests that borrower and lender reports are extremely similar for all forms of debt except credit card debt, for which borrowers report considerably less than lenders. This under-reporting likely reflects that individuals are often uniformed about their debt (Brown et al. 2011). If such under-reporting in our data was systematic in ways that are associated with depressive symptoms, it will have biased our estimates. It is less worrisome for our general conclusions if respondents’ self reports of their debt placed them in the same relative position in the debt distribution (despite potential underestimation of their absolute amount of debt) as would be the case if actual (administrative) debt data were used. There are no publicly available data linking credit reports to self-reported debt at the micro level. Thus, all existing studies of debt and wellbeing must rely on self-report data. However, Brown et al. 2011 compared aggregate debt levels based on self-reported and administrative data and found that time trends in consumer debt are similar (though at different absolute levels) across data sources, and that age and regional patterns are also similar. At the same time, they also found that single-person households tend to report debt more accurately than larger households (which may reflect better information among the former). Although we cannot rule out that our estimates were biased by systematic underreporting, we were further encouraged that Brown et al. found no evidence that the standard correction—universally multiplying credit card debt by a common factor (constant), which assumes that underreporting is consistent across groups and over time, is inappropriate in empirical analyses. Unfortunately, potential under-reporting of (particularly short-term) debt was a necessary limitation of our analyses.

In our standard OLS (but not fixed-effects) regressions, we included as control variables time-invariant baseline measures of whether the respondent was male, the respondent’s race/ethnicity (black, Hispanic, and other race/ethnicity, with white as the reference category), and indicators for the highest level of educational attainment by the respondent’s most educated parent (less than a high school education and high school degree or GED, with greater than a high school degree as the reference category). Ideally, we would have allowed the respondent’s parents’ educational attainment to vary over time; unfortunately, however, this information was collected only in Wave 1 (and not in Wave 2) of the NSFH. Given that respondents in our sample were 21–65 years old at the first (and subsequent) observation, it is unlikely that a large share of their parents increased their educational attainment over the observation period. Thus, this limitation is unlikely to have had much of an influence on our results. With the possible exception of the respondent’s parents’ educational attainment, these variables are independent of (could not have been jointly determined with) debt and psychological wellbeing. In other words, whereas each of the covariates may have influenced subsequent debt accumulation, psychological wellbeing, or any associations between the two, neither debt, psychological wellbeing, nor any associations between the two could have influenced these (fixed) characteristics. Furthermore, these baseline characteristics could not have been determined by the same factors or processes that determined debt or psychological wellbeing. As discussed below, because these characteristics are time-invariant (fixed) they cannot be directly modeled in fixed-effects regressions, which estimate average associations between changes in predictors and changes in outcomes for individuals. Although parameters are not directly estimated for these measures, the estimates produced by fixed-effects regression are adjusted for these (and all other observed or unobserved) time-invariant individual characteristics.

We included the following time-varying socioeconomic factors as controls in the standard OLS and fixed-effects regressions: respondent age; respondent marital status (indicators for currently married and never married, with currently single as the reference category); measures of the proportion of household members age 6–17, 18–24, 25–44, 45–65, and greater than 65 (with the proportion of household members under age 6 as the reference category); respondent education (less than a high school education and high school degree or GED, with greater than a high school degree as the reference category); the logarithm of total household income; whether the respondent was working at the time of the interview; total household assets (the sum of the self-reported value of all real estate holdings, vehicles, investments, and savings); and whether the respondent reported being in excellent, good, or fair health (versus poor health). It is important to note that, with the exception of age, these factors may be jointly determined with debt and psychological wellbeing. Again, as discussed further below, by estimating associations between changes in the predictors and changes in the outcome, the fixed-effects regressions functioned to reduce potential bias that may result from such jointly determined—or endogenous—relationships.

Finally, we included an indicator for wave of observation (a wave fixed effect) in all of the models. All dollar amounts were converted to constant, year 2010 dollars.

Empirical Strategy

The primary analytic challenge in estimating unbiased associations between debt and depressive symptoms was reducing bias due to social selection and ruling out reverse causality. With regard to social selection, we were particularly concerned that individual and household characteristics that we could not observe in our data were correlated with both debt accumulation and depressive symptoms, such that they would obscure the true relation between them. We were also concerned that greater depressive symptoms may have driven debt accumulation, rather than vice versa.

We relied on two empirical estimation strategies to attempt to adjust for social selection. First, we estimated a series of OLS regressions with extensive controls. These models allowed us to examine the extent to which the unadjusted (bivariate) associations between debt and depressive symptoms were driven by differences in the characteristics and circumstances of debtors and non-debtors. The OLS models took the form:
$${\text{DEP}}_{it} = \beta_{0} + \rho {\text{DEBT}}_{it} + \beta X_{it} + \varepsilon_{it} \,$$
(1)
where DEPit was the depressive symptoms measure for individual i at time t; DEBTit was one or more debt measures; Xit was a vector of covariates (including a wave fixed effect); and εit was a disturbance term. We estimated three specifications of this model. The first included only the debt measure(s) and an indicator for wave of observation as predictors. We added the time-invariant baseline characteristics, including gender, race/ethnicity, and parents’ educational attainment, to the second specification, as well as a time-varying (but exogenous) measure of respondent age. In the final specification, we added the time-varying (and potentially endogenous) socioeconomic factors including, family structure and composition, educational attainment, income, work status, assets, and health status. This strategy allowed us to assess the extent to which the associations of interest varied with the inclusion of more extensive controls. The standard errors in these models were adjusted for intra-cluster correlation due to multiple observations of the same individuals.
The coefficients produced by the standard OLS models were interpreted as average differences in depressive symptoms between individuals who had (a particular type of) debt and those who did not (or between those who had greater and lesser amounts of debt in the models in which debt amount was the key predictor). Despite the inclusion of controls, however, the OLS estimates were subject to bias due to unmeasured factors associated with both debt and depressive symptoms. In addition, because debt and depressive symptoms were measured at the same point in time, these estimates did not account for the possibility of reverse causality. For these reasons, we also estimated two-period fixed-effects models, which further adjusted for time-invariant unobservable characteristics and better accounted for the possibility of reverse causality. The fixed-effects models took the form:
$${\text{ DEP}}_{it} = \alpha_{i} + \lambda_{t} + \rho DEBT_{it} + \beta X_{it} + \varepsilon_{it} \,$$
(2)
where αi was an individual-specific fixed effect, which was comprised of all time-invariant (observed or unobserved) characteristics of individual i, and \(\lambda_{t}\) was the wave fixed effect. Equation (2) was reduced to:
$${\text{ DEP}}_{it} - \overline{{{\text{DEP}}_{i} }} = \lambda_{t} - \overline{\lambda } + \rho (DEBT_{it} - \overline{{DEBT_{i} }} ) + (X_{it} - \overline{{X_{i} }} )\beta + (\varepsilon_{it} { - }\overline{{\varepsilon_{\text{i}} }} ) { }$$
(3)
in which the value of each variable at a given time point was modeled as the deviation between an individual’s actual value at that time point and the individual’s average value on that variable across both observation points. Differencing each variable from its mean across the time periods effectively removed the individual-specific fixed effect from the model. It thereby served to adjust for all observed and unobserved time-invariant characteristics of the individual and, thus, to eliminate bias due to such factors from the estimated association between debt and depressive symptoms. However, the estimate was still subject to bias due to omitted time-varying characteristics or characteristics that have time-varying associations with the outcome.

In contrast to the standard OLS regressions, the fixed-effects regression estimates reflect intra-individual change for individuals observed with and without (a particular type of) debt (or with different amounts of debt). As such, they were interpreted as the average difference in depressive symptoms when the same individual was observed with debt and without debt (or with different amounts of debt). We estimated two specifications—controlling only for wave of observation and controlling for the full set of time-varying socioeconomic factors—of this model.

Results

Descriptive Statistics

Table 1 presents descriptive statistics for the full sample and by whether the respondent had any debt. The raw data (averaged across the two time periods) revealed that those with no debt had a higher mean number of depressive symptoms than those with debt. Overall, 79 % of households had some debt, with 49, 53, and 48 % reporting short-, mid-, and long-term debt, respectively. Among households that had debt, 62 % had short-term debt, 67 % had mid-term debt, and 60 % had long-term debt. On average, households had about $42,000 of total debt, consisting of about $2000 of short-term, $6500 of mid-term, and $34,000 of long-term debt (measured in constant 2010 dollars). Among those households that had any debt, these figures were approximately $53,500, $3000, $8000, and $43,000. Long-term debt by far accounted for the largest portion of total debt in our data, followed by mid- and short-term debt. These patterns are mirrored in the Survey of Consumer Finances (SCF)—a survey widely considered as the most reliable benchmark for measures of household net worth (Kennickell and Shack-Marquez 1992). As a comparison, for example, the 1989 SCF by the Federal Reserve reported total debts of $61,000, of which $3200 was short term credit card debt. The levels are larger in the SCF than NSFH, but the ratios of short-term debt to total debt are similar (4.7 and 5.0 % respectively). The 1992 SCF reported about $65,000 in total debt and $3500 in credit card debt. The periods are not synchronized but are relatively close. Average total debt to annual household income and assets ratios were 1.7 and 1.5 (2.1 and 2.0 among those with debt).
Table 1

Descriptive statistics, full sample and by any debt status

 

Full sample

No debt

Any debt

CES-D depression (0–36 points)

9.195 (0.072)

9.993 (0.176)

8.982 (0.077)***

Debt

 Any debt

0.790

  

 Any short-term debt

0.492

 

0.624

 Any mid-term debt

0.530

 

0.671

 Any long-term debt

0.476

 

0.603

 Total debt amount

$42,257.28 (552.202)

 

$53,509.53 (661.393)

 Total short-term debt

$2,030.22 (49.753)

 

$2,570.82 (61.665)

 Total mid-term debt

$6,470.38 (110.994)

 

$8,193.34 (138.221)

 Total long-term debt

$33,756.69 (517.499)

 

$42,745.38 (628.050)

 Total debt/annual income

1.690 (1.018)

 

2.140 (1.291)

 Short-term debt/annual income

0.168 (0.102)

 

0.213 (0.129)

 Mid-term debt/annual income

0.218 (0.122)

 

0.277 (0.155)

 Long-term debt/annual income

1.303 (0.917)

 

1.651 (1.163)

 Total debt/assets

1.541 (2.033)

 

1.951 (2.573)

 Short-term debt/assets

0.326 (0.121)

 

0.413 (0.153)

 Mid-term debt/assets

0.366 (0.410)

 

0.463 (0.519)

 Long-term debt/assets

0.849 (1.745)

 

1.075 (2.209)

Baseline (exogenous and time-invariant) characteristics

 Male

0.397

0.370

0.405**

 White

0.674

0.538

0.711***

 Black

0.213

0.294

0.191***

 Hispanic

0.096

0.152

0.082***

 Other race/ethnicity

0.016

0.017

0.016

 Parents less than high school

0.409

0.571

0.366***

 Parents high school/GED

0.347

0.271

0.368***

 Parents greater than HS/GED

0.243

0.158

0.266***

Time-variant socioeconomic factors

 Age

39.442 (0.087)

41.596 (0.228)

38.869 (0.094)***

 Married

0.544

0.337

0.599***

 Never married

0.178

0.279

0.151***

 Single (divorced, separated, widowed)

0.278

0.384

0.250***

 Proportion of persons less than age 6

0.092 (0.001)

0.084 (0.003)

0.094 (0.002)**

 Proportion of persons age 6–17

0.178 (0.002)

0.162 (0.004)

0.183 (0.002)***

 Proportion of persons age 18–24

0.064 (0.001)

0.069 (0.003)

0.063 (0.001)*

 Proportion of persons age 25–44

0.397 (0.003)

0.319 (0.006)

0.418 (0.003)***

 Proportion of persons age 45–65

0.248 (0.003)

0.327 (0.007)

0.227 (0.003)***

 Proportion of persons greater than 65

0.020 (0.001)

0.035 (0.002)

0.016 (0.001)***

 Less than high school

0.193

0.358

0.149***

 High school/GED

0.383

0.389

0.381

 Greater than HS/GED

0.425

0.253

0.470***

 Logarithm of household’s total income

10.615 (0.010)

10.072 (0.029)

10.759 (0.011)***

 Currently working

0.701

0.491

0.757***

 Total assets ($10,000s)

13.137 (0.164)

7.042 (0.279)

14.760 (0.194)***

 Excellent, good or fair health status

0.946

0.913

0.955***

 Wave observed

0.501

0.555

0.486***

 Proportion of sample

 

0.790

0.210

16,964–17,059 individual-wave observations of 8457–8516 individuals per dataset, across 25 imputed datasets. Means (and standard errors) or proportions presented. Statistical significance of bivariate tests for mean difference between those with any debt and those with no debt: * p < 0.05, ** p < 0.01, *** p < 0.001

Our raw data also revealed considerable differences in the characteristics of those who had debt and those who did not. Those who had debt were younger, more likely to be male, and less likely to be black or Hispanic than those who did not. The former had more highly educated parents, were more highly educated themselves, and were more likely to be married and working. On average, they also had greater income and assets and were in better health than those without debt. In short, the raw data indicated that adults who had debt tended to be more socioeconomically advantaged than those who did not. These differences may, in part, explain why having debt was associated with fewer depressive symptoms in the raw data. They highlight the importance of adjusting for individual and household characteristics when estimating associations between debt and depressive symptoms.

In addition, we examined descriptive statistics by specific types of debt (results not shown). We found that those with short-term debt had greater depressive symptoms than those without short-term debt, although this difference fell short of statistical significance. In contrast, those with mid- and long-term debt had significantly fewer depressive symptoms than those without such debt. We also found that men were more likely than women to have mid- and long-term debt, but less likely to have short-term debt. For the most part, patterns for the other covariates were similar across all three debt measures, with the exception that those with greater assets were more likely to have mid-, and particularly long-term debt than those with fewer assets. This makes sense given that those with greater assets are considerably more likely to own a home and to have taken on a mortgage for its purchase. These results suggest that examining associations between particular types of debt and depressive symptoms separately is warranted.

Regression Results

Table 2 shows results from our primary standard OLS (Models 0, 1, and 2) and fixed-effects (Models 3 and 4) regression models. Panel A presents results from models in which we used a dichotomous indicator for having any debt as the key predictor. Consistent with the bivariate results, the estimate from Model 0, which adjusted only for wave of observation, revealed that having any debt was associated with fewer depressive symptoms. This association remained negative, but became non-significant in Model 1, in which we controlled for age, sex, and race/ethnicity, and the highest educational attainment of the respondent’s parents. However, in Model 2, in which we further controlled for family structure and composition, education, income, work status, assets, and health status, this association became large and statistically significant. Thus, after controlling for these time-varying socioeconomic factors we found debt to be associated with greater depressive symptoms. This likely reflects that those who have debt tend to be more socioeconomically advantaged than those with no debt and that socioeconomic advantage is inversely associated with depressive symptoms.
Table 2

OLS and fixed effects regression results

 

(0)

(1)

(2)

(3)

(4)

Panel A: Any debt

 Any debt

−1.014 (0.199)***

−0.216 (0.203)

0.743 (0.196)***

0.559 (0.291)+

0.916 (0.293)**

Panel B: Amount of total debt

 Logarithm of total debt

−0.374 (0.030)***

−0.295 (0.031)***

0.037 (0.034)

0.024 (0.052)

0.138 (0.053)*

Panel C: Any short-, mid-, and long-term debt

 Any short-term debt

0.622 (0.154)***

0.528 (0.150)***

0.779 (0.145)***

0.640 (0.213)**

0.699 (0.207)**

 Any mid-term debt

−0.289 (0.163)+,a

−0.062 (0.163)a

0.289 (0.161)+,a

0.113 (0.219)

0.350 (0.217)

 Any long-term debt

−1.976 (0.153)***,a,b

−1.101 (0.158)***,a,b

−0.325 (0.160)*,a,b

−0.186 (0.230)a

0.272 (0.231)

Panel D: Any short-, mid-, and long-term debt

 Logarithm of short-term debt

0.193 (0.046)***

0.170 (0.045)***

0.229 (0.043)***

0.192 (0.065)**

0.215 (0.063)***

 Logarithm of mid-term debt

−0.130 (0.036)***,a

−0.062 (0.036)+,a

0.033 (0.036)a

0.010 (0.048)a

0.071 (0.048)c

 Logarithm of long-term debt

−0.351 (0.024)a,b

−0.210 (0.026)***,a,b

−0.079 (0.027)**,a,b

−0.054 (0.040)a

0.032 (0.041)a

Baseline (exogenous) characteristics

No

Yes

Yes

Yes

Yes

Socioeconomic factors

No

No

Yes

No

Yes

Fixed effects

No

No

No

Yes

Yes

16,964–17,059 individual-wave observations of 8457–8516 individuals per dataset, across 25 imputed datasets. OLS coefficients (and standard errors) presented. Standard errors from the OLS models were adjusted for intracluster correlation due to multiple observations of each individual. Covariates are listed in Table 1

+p < 0.10; * p < 0.05, ** p < 0.01, *** p < 0.001

aDiffers from short-term debt at p < 0.05

bDiffers from mid-term debt at p < 0.05

cDiffers from short-term debt at p < 0.10

The Model 2 estimate was confirmed by the fixed-effects results, both with and without controls. The estimate from Model 4, the most restrictive specification, indicates that having any debt was, on average, associated with a 0.92 point increase in depressive symptoms. This implies that having debt was associated with having approximately 9 % more depressive symptoms (given a mean of 10 depressive symptoms among those with no debt), which we interpret as constituting a modestly large magnitude of effect. Among the covariates (results not shown), we found that being married, higher income, working, and in good physical health were inversely associated with depressive symptoms, as would be predicted.

Results for total amount debt, as well as for each type of debt (any and amount), are presented in Panels B through D in Table 2. Consistent with the results for having any debt (Panel A), we found that the logarithm of total debt (Panel B) was negatively associated with depressive symptoms in Model 0 (which controlled only for wave of observation) and Model 1 (which controlled for wave of observation and time-invariant baseline characteristics). In Model 2, which also controlled for time-variant socioeconomic factors, total debt was positively associated with depressive symptoms, although this estimate (for amount of debt) failed to attain statistical significance, unlike the estimate for having any debt. Both of the fixed-effects regression estimates (Models 3 and 4) were positive, and the Model 4 estimate was statistically significant. The coefficient from Model 4 indicates that a 10 % increase in debt was roughly associated with a 1.4 point (about 14 %) increase in depressive symptoms.

Turning to specific types of debt (Panels C and D), we found that the association between debt and depressive symptoms was predominantly driven by short-term debt. Short-term debt was associated with greater depressive symptoms in all of the models. The Model 4 estimates indicate that having any short-term debt was associated with 8 % more depressive symptoms and that a 10 % increase in short-term debt was associated with roughly a 24 % increase in depressive symptoms, given an average of 9.1 depressive symptoms for those with no short-term debt (see the note to Table 2). In contrast, the Model 4 estimates for mid- and long-term debt were small and non-significant.

Table 3 presents results from the OLS (Model 2) and fixed-effects (Model 4) regressions with the full set of controls, in which we tested the robustness of our findings to two specifications of relative debt levels: debt-to-annual household income and debt-to-total household assets. Again, we present results for total debt and for the specific types of debt. We found no evidence of association between either total debt-to-income or total debt-to-assets and depressive symptoms. Turning to types of debt, the OLS and fixed-effect results using debt-to-annual income ratios were consistent with those using the absolute debt measures: short-term debt was associated with increased depressive symptoms and the short-term debt estimate also significantly differed from the estimates for mid- and long-term debt in the fixed-effects models. Whereas the OLS results using the debt-to-assets measures were also consistent with those using the absolute debt measures, the fixed-effects results suggest no association of any of the debt-to-assets ratios with depressive symptoms.
Table 3

Summary of robustness checks for ols and fixed effects regression results: debt as a proportion of income and assets

 

(2)

(4)

Panel A: Amount of total debt/annual income

 Total debt/annual income

0.000 (0.004)

−0.002 (0.003)

Panel B: Amount of short-, mid-, and long-term debt/annual income

 Short-term debt/annual income

0.023 (0.007)**

0.019 (0.010)+

 Mid-term debt/annual income

0.032 (0.038)

0.002 (0.042)a

 Long-term debt/annual income

−0.004 (0.002)+,a

−0.005 (0.004)a

Panel C: Amount of total debt/assets

 Total debt/assets

0.009 (0.007)

−0.001 (0.011)

Panel D: Amount of short-, mid-, and long-term debt/assets

 Short-term debt/assets

0.015 (0.003)***

−0.027 (0.043)

 Mid-term debt/assets

0.037 (0.037)

0.038 (0.038)

 Long-term debt/assets

−0.009 (0.014)

−0.007 (0.019)

16,964–17,059 individual-wave observations of 8457–8516 individuals per dataset, across 25 imputed datasets. OLS coefficients (and standard errors) presented. Standard errors from the OLS models were adjusted for intracluster correlation due to multiple observations of each individual. Models adjust for all of the covariates listed in Table 1

+p < 0.10; * p < 0.05, ** p < 0.01, *** p < 0.001

aDiffers from short-term debt at p < 0.05

The results presented in Table 4 focus on subgroup differences in associations between debt and depressive symptoms by age, education, and whether the respondent was stably married throughout the observation period. Specifically, we estimated separate regression models for age 21–30, 31–50, and 51–64; those with a high school education or less and those with greater than a high school education; and those who were and were not consistently married through the observation period. Results from the fixed-effects regressions indicate that associations between short-term debt and greater depressive symptoms were most heavily concentrated among the older (those age 51–64) and less educated (those with a high school degree or less) groups of adults in our sample, as well as among those who were not consistently married throughout the observation period.
Table 4

Fixed effects regression results: subgroup analyses by age, education attainment, and consistently married for amount of debt by type

 

Age

Education

Consistently married

 

21–30

31–50

51–64

≤HS

>HS

No

Yes

 

(4)

(4)

(4)

(4)

(4)

(4)

(4)

Logarithm of short-term debt

0.198 (0.192)

0.097 (0.089)

0.393 (0.192)*

0.281 (0.093)**

0.144 (0.079)+

0.299 (0.088)**

0.109 (0.084)

Logarithm of mid-term debt

0.202 (0.149)

0.077 (0.065)

0.061 (0.138)

0.052 (0.067)a

0.080 (0.062)

−0.029 (0.071)a

0.110 (0.060)+

Logarithm of long-term debt

−0.034 (0.142)

0.019 (0.063)

0.180 (0.137)

0.028 (0.062)a

0.056 (0.057)

−0.049 (0.062)a

0.018 (0.055)

Fixed effects

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Proportion of sample

0.251

0.559

0.189

0.575

0.425

0.452

0.548

16,964–17,059 individual-wave observations of 8457–8516 individuals per dataset, across 25 imputed datasets. OLS coefficients (and standard errors) presented. Standard errors from the OLS models were adjusted for intracluster correlation due to multiple observations of each individual. Models adjust for all of the covariates listed in Table 1

+p < 0.10; * p < 0.05, ** p < 0.01, *** p < 0.001

aDiffers from short-term debt at p < 0.05

Finally, because 10 % of the respondents in our sample reported having no depressive symptoms, we also estimated a series of Tobit models with random effects as an additional robustness check. The Tobit model, which is a censored regression model, was used to estimate the linear relationship between debt and depressive symptoms while adjusting for the fact that depressive symptoms may have been left censored, as indicated by the relatively large proportion of zeros, or even right censored if actual depressive symptoms may have extended beyond the maximum value of the measure. Results (not shown, available upon request) were consistent with those of our primary models.

Discussion and Conclusions

Several limitations should be considered when interpreting the results of this study. First, the fixed-effects regressions are based on the assumption that associations of types and levels of debt with depressive symptoms are time invariant. As such, our estimates may have been biased by omitted time-varying factors or by time-invariant factors that have time-varying influences on debt or depressive symptoms. This may be a particularly important concern given the length of time between observations. Second, specific types of debt are, to some extent fungible. For example, individuals or households may use home equity loans to pay off short-term debt (thus, effectively converting short-term debt to mid- or long-term debt). Individuals may also boost current (short-term) consumption by taking student loans of a larger size than is needed to cover direct educational expenses. Our analyses were unable to account for these possibilities. Rather, we observed only the amounts and types of debt owed at two given points in time. Third, there may be considerable heterogeneity in associations between debt and depressive symptoms by factors beyond age, education, and marital stability, including other measures of SES, gender, and race/ethnicity; our analyses were silent in this regard. Fourth, it is important to note that depressive symptoms, as measured by this application of the CES-D, do not constitute clinical depression. As such, our results cannot be interpreted as estimates of associations between debt and clinical depression, and it cannot be assumed that debt has a similar association with diagnosable depression. Finally, although our fixed-effects regressions account for within-individual changes in debt and estimate associations between changes in debt and changes in depressive symptoms, reverse causality continues to be a concern. Specifically, because we observe individuals at only two distinct points in time, roughly 6 years apart, we cannot determine the order in which debt levels and depressive symptoms changed (which preceded which) during the intermittent time period. To gain further insight into the potential direction of these associations, we estimated cross-lagged path (structural equation) models in which we simultaneously regressed Wave 2 debt on Wave 1 depressive symptoms and Wave 2 depressive symptoms on Wave 1 debt, while also accounting for the correlation between debt and depressive symptoms within each wave. Results (not shown) revealed that, in all cases, associations between Wave 1 debt and Wave 2 depressive symptoms were substantially larger than associations between Wave 1 depressive symptoms and Wave 2 debt. This strongly suggests that the (causal) direction of association runs from debt to depressive symptoms rather than vice versa.

Notwithstanding these limitations, the results from this study suggest that, after adjusting for background characteristics, debt accumulation is associated with greater depressive symptoms among US adults. This finding was robust to both standard OLS and fixed-effects specifications of our regression models. Furthermore, our analyses of specific types of debt revealed that, rather than this association being driven by overall amount of debt, it was much more closely aligned with the type of debt incurred. Specifically, we found that the association between debt and depressive symptoms was primarily driven by short-term debt, whether modeled dichotomously or by amount, and whether considered in absolute or relative terms. On the whole, short-term debt was significantly associated with depressive symptoms in the majority of our models, and the magnitude of this association was moderately large. By contrast, mid-term and long-term debt never significantly differed from zero in the fixed-effects models. In addition, the estimate for short-term debt significantly differed from that for long-term debt in most models, and significantly differed from that of mid-term debt in several; the estimates for mid-term and long-term debt were rarely significantly different from one another.

Our findings with regard to short-term debt provide nuance to those of Dew (2007) who, using the same data source, found short-term debt at Wave 1 to be associated with fewer depressive symptoms at Wave 2 (the study did not examine mid- or long-term debt). The difference between results from that study and our results reflects that our estimates were adjusted for initial levels of depressive symptoms. Accounting for initial differences between debtors and non-debtors addresses selection into borrowing, as does focusing on changes in debt and changes in depressive symptoms over time. To better compare our results to Dew, we also estimated associations between debt and depressive symptoms separately for stably married individuals and other individuals. These results showed that the association between debt and depressive symptoms was considerably stronger among individuals who were not consistently married. The association was smaller and non-significant, but positive, when the model was estimated using only the subsample of consistently married families. This further suggests that the difference between our estimates and Dew’s (2007) likely reflects differences in model specification more so than differences in sample definition.

On the whole, our results suggest that short-term or unsecured debt is associated with increased depressive symptoms, whereas longer-term debt does not exhibit such an association. Short-term debt is often used for immediate consumption and entails higher interest rates and fees than long-term debt; short-term debt may also be taken on with relatively less agency than longer-term types of debt. For example, people may perceive mortgages as a lifetime asset, or education loans as an investment in human capital. Both would be incurred with considerable agency. These sorts of investments may not have the same associations with depressive symptoms as short-term borrowing for ongoing expenses. That is, individuals may experience optimism with regard to debt that will “pay off” in the future through increased earnings or wealth acquisition (in the form of home equity), but experience psychological burden with regard to debt that does not contribute to asset or skill accumulation.

This interpretation is consistent with Dwyer et al. (2011) interpretation of their finding that educational debt was positively associated with self-esteem and mastery among young adults, which they argued is likely to reflect that such debt has positive psychological influences because it is accrued for human capital investment. At the same time, we found an association between short-term debt and greater depressive symptoms, whereas they found an association between credit card debt and higher levels of self-esteem and mastery. They also interpreted this finding as reflecting positive influences associated with human capital investment. Two factors should be considered when weighing our results and theirs. First, the two studies measure psychological wellbeing in very different domains. Second, in our subgroup analyses, we found the association between short-term debt and depressive symptoms to be most concentrated among adults age 51–64. Indeed, this association was not significant among younger adults, who are the focus of their study. Thus, the two sets of findings should not be interpreted as inconsistent.

Few prior studies have investigated how mid-term debt, which we defined to include installment loans, home improvement loans, vehicle loans, loans from friends, and other (non-mortgage or education) bank loans, may influence psychological wellbeing. Such debt is often characterized by relatively low interest rates and amortized principal payments. It may also be used to for the purchase of necessity items or items that are consumed over a relatively long time period, such as a car or home improvements. As such, it may share some characteristics with long-term debt and may therefore be similarly associated with psychological functioning or wellbeing. At the same time, such debt is not generally accrued as a result of human capital investment or the purchase of an appreciable asset. In this way, it may differ considerably from long-term debt. Future research should explore these possibilities and uncover how mid-term debt may influence individual and household wellbeing.

Although we found no links between long-term debt and depressive symptoms, it is important to note that our data were collected long before the recessions of 2001 or 2008, the housing boom and bust, or the significant increases in student debt that marked the last decade. It is quite possible that mortgage debt, especially in areas with declining home values, is perceived differently now than in the past, which may suggest a differential pattern of association with depressive symptoms or other forms of psychological strain (see Talbot et al. 2014 for a discussion). Likewise, given shifts in lending for student loans and the experiences of recent cohorts of young adults (see Friedline et al. 2014), the relative agency of investments in human capital may also be eroding. Future research should study associations of mortgage and education debt with individual and household functioning in the last decade, given the extensive shifts in financial and housing markets.

Our results highlight the importance of considering specific types and amounts of debt, rather than overall debt, when modeling associations between debt and psychological wellbeing. Prior studies (e.g., Bridges and Disney 2010; Brown et al. 2005; Dew 2007; Gathergood 2012) have largely failed to simultaneously consider the full range of specific types of debt that households may hold. As such, previous findings may have confounded multiple and potentially counterbalancing influences of different types of debt. Future research should further explore the influences of specific types and amounts of debt on a range of domains of functioning and wellbeing in order to provide a more nuanced understanding of the nature of these relationships. Future research should also further investigate our finding that the association of short-term debt with depressive symptoms is particularly concentrated among 51–64 year old adults, those with a high school education or less, and those who are not stably married, which suggests that short-term debt may have an adverse influence on psychological wellbeing, particularly for those who are less educated, approaching retirement age, and either single or in unstable relationships. This may be of concern given that these populations groups may be least likely to experience substantial future income growth. If this finding is confirmed, it may have implications for targeting financial capacity building interventions toward these populations.

Our results also have implications for the financial industry and financial professionals. Financial planners, educators, counselors and advisers may not be fully considering how mental health issues and credit issues coincide. Professionals should recognize signals of depressive symptoms and, especially, associations between mental health and short-term debt. There remains the potential that lenders offering high-cost short-term debt will target disadvantaged populations by offering “easy” solutions to short-term cash flow problems, which serve to exacerbate both indebtedness and the client’s mental health problems. Debt contracts offered to vulnerable borrowers might benefit from provisions that require affirmative disclosures of terms and conditions, mandatory financial counseling, or even the right to rescission within a specified time frame. Likewise, mental health professionals may benefit from taking a closer look at their clients’ borrowing behaviors and overall financial status. Understanding how depression and borrowing may be reciprocally related remains an important issue for practitioners to consider.

Finally, our results have implications for public policy. The Great Recession and associated housing crisis prompted considerable regulatory reforms in federal laws governing mortgages and consumer (largely credit card) debt, as well as the creation of the Consumer Financial Protection Bureau. To the extent that such policies serve to limit the accrual of short-term debt, they may also function as protective factors vis-à-vis depressive symptoms that may increase with such debt. Future research should examine whether these recent policy initiatives actually affect short-term debt accrual and, in turn, whether reductions in short-term debt are associated with reductions in depressive symptoms.

Copyright information

© Springer Science+Business Media New York 2015