Wealth and Inequality in the Stability of Romantic Relationships
The family is a key institution that transmits inequality, and racial and socioeconomic inequalities in family life have grown markedly. We use data from the 1996 to 2008 panels of the Survey of Income and Program Participation to offer a comprehensive account of how wealth relates to family stability and how that relationship varies by union type, age cohort, and both type and amount of wealth. We find that liquid and illiquid assets and secured debts are associated with a decrease in the likelihood of dissolution, and that large unsecured debts are associated with an increase. These associations do not differ significantly for married and cohabiting couples. We find evidence of both the material and the symbolic importance of wealth for stability. We also find that wealth explains a significant degree of the racial inequality in family stability.
wealth inequality, marriage, cohabitation, family stability
The Great Recession of 2008–2009 raised public awareness about inequality in American society and invigorated scholarly activity into the causes and consequences of extreme and rising wealth inequality (Piketty 2014; Pfeffer, Danziger, and Schoeni 2013). The family is a key institution that transmits inequality across generations, and racial and socioeconomic inequalities in family life have grown markedly over the past half century (McLanahan 2004; Pfeffer and Schoeni, this issue). Most research has examined how family processes reproduce income inequality (Western et al. 2012), but wealth may influence the formation and stability of family relationships in distinct ways (Keister 2000, 6–16, 225–29). In this paper, we use longitudinal data from the 1996, 2001, 2004, and 2008 panels of the Survey of Income and Program Participation (SIPP) to offer a comprehensive account of how wealth relates to family stability and how that association varies by relationship type, age cohort, and type and amount of wealth. After examining both the material and symbolic significance of wealth for relationship stability, we consider whether wealth inequalities contribute to population-level inequalities in family stability by race and by macroeconomic context.
Family structure and family stability are profoundly unequal along the lines of race and class (McLanahan and Percheski 2008; Ellwood and Jencks 2004; Kennedy and Bumpass 2008). These inequalities have grown substantially over the past half century, contributing to what Sara McLanahan (2004) and others refer to as the “diverging destinies” of children. The socioeconomic [End Page 197] gradient in marriage and child-bearing was minimal in the 1960s, but today most children of college-educated parents grow up in households characterized by stable married families and stable finances, whereas the children of less-educated parents are increasingly exposed to unstable family and economic situations. Socioeconomically disadvantaged adults are less likely to form marital relationships, are more likely to have children outside of marriage, and have less stable relationships than their more advantaged counterparts (McLanahan 2004; Tach, Mincy, and Edin 2010; Ventura and Bachrach 2000). As a result, the children of married parents spend the vast majority (84 percent) of their childhoods with both parents, whereas the children of unmarried parents can expect to spend only about half (52 percent) (Bumpass and Lu 2000). These patterns also fuel racial inequality in family life: African American couples are significantly less likely to marry or to have stable romantic relationships and more likely to have children outside marriage than whites; patterns for Latino families are more variable (Kennedy and Bum-pass 2008; McLanahan and Percheski 2008).
The growing educational and racial gradients in family formation and stability are particularly consequential because they overlap with a period of growing income and wealth inequality. Part of changing dynamics in income inequality comes from rising incomes for those in the top decile of the income distribution. Emmanuel Saez (2009) calculates that from the 1940s until the early 1980s, the top decile accounted for just over 33 percent of total income in the United States. However, from the early 1980s forward, the percentage of income going to the top 10 percent rose such that by 2007, it accounted for fully half of total income. Wealth inequality has also been driven by the top of the distribution pulling away from the rest: the top 0.1 percent owned a staggering 22 percent of total wealth in 2012 (Saez and Zucman 2014). This growth at the top was accompanied by stagnation and even decline among the lower quintiles. These trends were exacerbated by the 2008 recession, when median wealth plummeted and wealth inequality increased sharply (Wolff, this issue).1
Socioeconomic status structures the formation, progression, and dissolution of romantic relationships. Much of the work on family dynamics and economic inequality has focused on education- or income-based measures of inequality (for a review, see McLanahan and Percheski 2008). We argue that wealth is an important but understudied dimension of family relationships. Wealth is not simply a function of income or education (Hurst et al. 1998; Keister 2000). First, wealth is a stock rather than a flow. It is transmitted across generations in very tangible and unequal ways. It buys access to elite social settings such as neighborhoods, schools, and colleges (Keister 2000; Oliver and Shapiro 1995; Rauscher, this issue). It also allows families to insure against economic risks in other domains of life and may serve as a buffer against adverse effects of income volatility on consumption (Fisher et al., this issue). Further, income and wealth are not highly correlated. Although those with long-term low income may begin to look like those with low wealth, this is not necessarily the case. The very wealthy may have low earnings and support consumption with income from assets (Wolff 1995) and differences in saving and investment are large at all income levels (Brimmer 1988). Because of these important distinctions, wealth may shape the progression and stability of family relationships in distinct and consequential ways.
Wealth and the Progression of Romantic Relationships
The institution of marriage is held in high esteem by Americans of all races and classes (Axinn and Thornton 2000; Thornton and Young-DeMarco 2001), and it is increasingly viewed as a coveted social status, or capstone, in the life course (Cherlin 2004). Today, most Americans believe that they should have not just steady employment but also some assets—money saved, a car, or even a home—before they marry (Dew and Price 2011; Edin and Kefalas 2005; Gibson-Davis, Edin, and McLanahan 2005; [End Page 198] Lloyd and South 1996). These prerequisites for marriage—the “marriage bar”—are held by rich and poor alike, but the poor are significantly less likely to be able to realize them (Gibson-Davis, Edin, and McLanahan 2005). Researchers have used the idea of the marriage bar to explain racial and socioeconomic gaps in entry into marriage. Although wealth figures strongly in qualitative narratives (Edin and Kefalas 2005), only a handful of studies have examined it as an independent component. These studies find that wealth, particularly homeownership, increases the likelihood of marriage (Lloyd and South 1996; Gibson-Davis 2009), and that racial inequalities in wealth explain a significant part of the racial gap in the decision to marry (Schneider 2011).
Daniel Schneider (2011) argues that wealth may influence the decision to marry because of its symbolic value or its use value. Wealth has use value because it can be deployed to boost material well-being by mitigating material hardship and insuring against future economic uncertainty (Fisher, this issue; Oppenheimer, Kalmijn, and Lim 1997). The symbolic value of wealth inheres in what wealth signifies to others (Lamont and Molnár 2002; Cherlin 2004). To the extent that marriage has become a status marker, displays of wealth—a big wedding, purchasing a house—signal that the couple has achieved the requisite social status deemed worthy of marriage (Veblen 1973; Zelizer 1997; Cherlin 2004). The decision to hold assets jointly or solely may also hold symbolic meaning within the relationship as a signal of a couple’s commitment, independence, or expectations about the future stability of their relationship (Addo and Sassler 2010; Kenney 2004; Treas 1993).
Compared with the accumulation of research on wealth and marriage entry, we know less about how wealth influences the stability of marriages. The family stress model predicts that economic hardships lead to feelings of economic pressure, which undermine interpersonal interactions and emotions within marriages, resulting in increased marital conflict (Conger and Elder 1994; Conger et al. 1990; Gudmunson et al. 2007). Marital conflict, in turn, is a key predictor of divorce (Amato and Rogers 1997). Economic hardship has been conceptualized in various ways—including income, poverty, and indicators of specific hardship experiences such as food or housing insecurity—and the associations with marital distress are robust to the specific indicator of hardship used (Conger, Conger, and Martin 2010).
Although wealth is moderately correlated with other indicators of socioeconomic status, it is not reducible to them, and scholars have only recently begun to untangle the unique effects of asset and debt accumulation on marital satisfaction and divorce (Dew 2011). Liquid assets (which can be converted to cash relatively easily) and illiquid assets (houses, cars, property) could promote marital stability because couples can draw on them to buffer against transitory shocks to income, which could reduce the marital strain that tends to accompany economic shocks. Asset holdings may have symbolic benefits as well: financial assets are associated with a positive future orientation, enhanced personal efficacy, and greater social participation (Sherraden 1991; Yadama and Sherraden 1996), which could have positive repercussions for interpersonal interactions and commitment within romantic relationships. Joint ownership of assets might signal particularly committed relationships (Addo and Sassler 2010; Treas 1993). Additionally, significant asset holdings might deter divorce because of the transaction costs associated with adjudicating the division of assets in divorce proceedings.
In contrast, predictions about how debt might influence marital stability are less clear. According to life-cycle theories of debt, secured debts, such as mortgages or educational loans, are investments that individuals (or couples) make to boost their long-term economic well-being.2 In the long term, the financial benefits [End Page 199] are expected to outweigh the short-term financial costs (Modigliani and Brumberg 1954). Such debts may have little short-term influence on relationships and may ultimately promote marital quality and stability in the long term. The meaning of secured debts may also have a life-cycle component, with greater secured debt at young ages being less troubling than it is at older ages.
Unsecured debts, such as consumer or credit card debt, also have ambiguous associations with relationship stability. They may be used to smooth consumption, thus averting financial hardships, and preserving relationship quality and stability. But they may also signal financial hardship or even cause it directly by diverting household income to debt repayment. Consistent with this latter hypothesis, consumer debt is associated with feelings of anxiety, economic pressure, and marital conflict (Conger and Elder 1994; Dew 2007; Drentea 2000).
Variation by Union Type
Virtually all the literature focuses on how wealth shapes decisions to start or end a marriage. Given the retreat from marriage and the concomitant growth of cohabitation over the past half century (Lundberg and Pollak 2013), whether economic forces influence the stability of cohabiting unions in the same way they influence the stability of marriages is significant. Hypotheses conflict. On one hand, ending a marriage has greater symbolic and financial costs than ending a cohabiting union. Divorce is a legal procedure that requires retaining a lawyer and undergoing court-mediated division of assets and belongings, which can be lengthy, costly, and emotionally painful. Divorce also has symbolic costs, given that partners lose their place in a legally recognized kinship system and the support that system provides. Thus, one might predict that married couples will be less likely to end their relationships in the face of low or declining levels of wealth than cohabiting couples, net of other economic characteristics.
On the other hand, given the marriage bar standards described, married couples might be more responsive to wealth than cohabiting couples when deciding to end relationships. If marriage is a coveted social status that signifies that a couple has made it financially, perhaps married couples will be more likely to break up than cohabiting couples when adverse economic conditions cause them to fall below the marriage bar. In fact, qualitative researchers have found that couples do not hold the same standards for their cohabiting relationships that they do for marital relationships, and they tolerate adverse economic and interpersonal conditions in cohabitation they say they would not tolerate within marriage (Edin and Kefalas 2005). Indeed, research examining the relative importance of economic conditions across union types has found that socioeconomic factors are more important predictors of relationship stability for marital unions than for cohabiting unions (Tach and Edin 2013).
Racial differences are large in the formation and stability of marital relationships. African American couples are less likely to enter marriage and have less stable marriages than white couples, even taking into account differences in economic characteristics such as income and employment (Kennedy and Bumpass 2008; McLanahan and Percheski 2008). Because racial inequalities in wealth are stark (Oliver and Shapiro 1995; Conley 1999; Bucks, Kennickell, and Moore 2006; Killewald and Bryan, this issue), if wealth shapes the progression of romantic relationships via the identified mechanisms—serving a use value by buffering financial hardships or serving a symbolic value by signaling the achievement and maintenance of the marriage bar—racial wealth gaps might explain some of the racial inequality in the stability of family relationships. Schneider (2011) finds that including measures of wealth as use value and symbolic value in models of first marriage reduced the black-white gap in marriage by about 30 percent, which was more than conventional economic covariates like employment and income explained. We know less about whether differences in wealth can help explain racial gaps in the stability of romantic relationships after they form. [End Page 200]
Macroeconomic Contexts and Relationship Instability
Macroeconomic downturns—characterized by some combination of high unemployment rates, stock market volatility, falling gross domestic product, and declining housing values—have the potential to shape the economic well-being of large segments of the population and to have cascading effects on family dynamics. Under normal macroeconomic conditions, we might expect family-level economic hardship to reduce family stability by heightening economic strain, reducing marital quality, and thereby increasing divorce rates. But the effects of family-level economic hardship may be different during macroeconomic downturns, when many others are experiencing hardship as well. In particular, it may be more costly for couples to dissolve their unions (Modestino and Dennett 2013). Couples may face greater constraint in their housing and employment options. They may lack the financial where-withal to set up two households or to cover the cost of a divorce. They may also find it more difficult to divide certain assets, such as homes or stock market holdings, if the value of those assets declined.
These forces may be one reason researchers have found mixed effects of macroeconomic conditions on divorce rates (Cherlin et al. 2013; Chowdhury 2013; Harknett and Schneider 2012). Divorce rates dropped during the Great Depression (Cherlin et al. 2013), but recessionary periods after World War II were associated with greater divorce risk (South 1985). Recent estimates from work on the Great Recession that began in 2008 found that rising unemployment rates were associated with reductions in the divorce rate (Amato and Beattie 2010; Hellerstein and Morrill 2011; Schaller 2012; Harknett and Schneider 2012; Cherlin et al. 2013; but see Arkes and Shen 2014). However, evidence from the Great Recession on the effect of foreclosure rates—another indicator of macroeconomic conditions—and marital stability is mixed. Kristen Harknett and Daniel Schneider (2012) find that higher foreclosure rates reduced divorce rates, but Philip Cohen (2014) finds that they increased them.
To our knowledge, research on the macroeconomic contexts of relationship stability has focused exclusively on marital stability. Even though the empirical record on marital stability is mixed, we predict that recessions might boost the stability of cohabiting unions, in part because the financial benefits to pooling incomes may be particularly necessary during macroeconomic hardship. Individuals are more likely to double up (share living arrangements) during recessionary times (Mykyta and Macartney 2011; Bitler and Hoynes 2015; Cherlin et al. 2013); cohabitation is one form of doubling up, and many cohabiting couples cite economic reasons for moving in together (Sassler 2004).
DATA AND METHOD
We use data from multiple panels of the Survey of Income and Program Participation, starting with the 1996 panel and ending with the 2008 panel, which concludes in 2013. The SIPP is a nationally representative survey designed to provide comprehensive information about the sources of income and government program participation of individuals and households in the United States on a subannual basis. The survey is designed as a series of national panels, each lasting three to four years. Together, the panels provide almost-continuous coverage of the U.S. household population since 1996. Unlike most other longitudinal surveys, each panel draws a new nationally representative sample rather than focusing on a single cohort (for which age and period effects are confounded).
In each SIPP panel, every member of the household age fifteen or older was interviewed every four months and asked about the previous four months. All were interviewed directly if possible or by proxy response from another household member otherwise. The SIPP imputes item—and person—nonresponse in all waves (Westat 2001, chapter 4). A household roster indicates the relationship of each household member to the household head, and monthly changes in the household roster are assessed at each survey. The SIPP follows all original sample members (who are present at the first survey wave) regardless of where they move in subsequent survey waves (unless they [End Page 201] are institutionalized, in military barracks, or abroad). The SIPP also surveys new individuals who live in households with original sample members over the course of the panel; these new individuals are not followed after they stop living with an original sample member. The SIPP also includes topical modules that are a separate set of questions asked in addition to the regular core survey questions during two or three waves of each panel. One set of topical modules asks detailed questions about asset and debt types and values, which we use for this analysis.3
In this article, we construct an analytic sub-sample of families by identifying the household reference person, who we follow until the survey ends. By following just the reference person, we avoid having both parties to a single union in the data. We restrict the sample to working-age adults, ages eighteen to sixty-four. We further restrict the sample to opposite-sex couples, because there are too few same-sex couples in our sample, especially in earlier panels, for separate analyses. This results in a sample of 1,613,586 married family observations (67,460 distinct relationships) and 124,846 cohabiting family observations (8,632 distinct relationships) across all four panels of the SIPP. We observe about 8 percent of married couples, and 37 percent of cohabiting couples, ending their unions during the SIPP panels.
Family Structure and Dissolution
In each month of the SIPP, we assess family structure by identifying adults living in the same household and classifying them as household head, spouse of the household head, or unmarried partner of the household head. Households are coded as married if a spouse of the household head is listed on the household roster, or cohabiting if an unmarried partner of the household head is. In all of the SIPP panels used in this paper, participants were asked directly about the presence of an unmarried partner. We identify a marital dissolution as occurring in the month in which the household reference person’s family structure changes from married to any other household type and either a separation or a legal divorce occurred. We identify cohabitation dissolution as occurring in the month in which one of the cohabiting partners no longer lives in the household. Because the SIPP includes no direct questions about the start and end dates of nonmarital romantic relationships in these panels, we must measure the start and end of cohabitations based on the household roster.
Family Wealth and Debt Components
We create four measures of wealth that mirror John Czajka, Jonathan Jacobson, and Scott Cody’s (2003) measures using SIPP (for a detailed list of the SIPP variables comprising each measure, see table A1). First, we calculate the value of all secured debt. For most participants, this is largely their mortgage. Some have business debts secured by the value of the business. Second, we calculate the value of unsecured debt. This is largely store and credit card debt. Third, we calculate liquid assets, which includes saving and checking account balances. Fourth, we calculate illiquid assets, which include the values of participants’ car or cars and the value of their house or business.4 Information is collected on assets and debts held individually by each adult in the household, as well as some assets and debts jointly held by spouses. We add individual and jointly held assets and debts together for each person in the union to create family-level measures. The dollar value of these measures is adjusted to 2013 dollars using data from the Bureau of Labor Statistics CPI inflation calculator. Table 1 shows the mean and median values of each measure of wealth by union type and stability. We construct standard deviation measures of [End Page 202] each of the asset and debt measures for inclusion in the regression models, so that our coefficients represent how a standard deviation change in assets or debts influences relationship stability.
Race and Ethnicity
The SIPP asks directly about the race and ethnic origin of participants. We use the race and ethnicity of the reference person as our measure. We maintain four categories: non-Hispanic [End Page 203] white, non-Hispanic black, Hispanic, and non-Hispanic other race. Table 1 shows the proportion of the sample in each racial-ethnic category as well as the share of each group who experience a marital separation (given that they were married) or a cohabiting union dissolution (given that they were in a cohabiting union). Table A2 shows differences in mean and median wealth and debt accumulation among racial-ethnic groups.
The SIPP asks about years of education completed for the reference person, which we recode into a four-category measure: less than high school (fewer than twelve years of school), high school (twelve years of school), some college (thirteen to fifteen years of school), and four-year degree or higher (sixteen years of school or more). Table 1 shows the proportion of household heads with each level of education, as well as the percentages from each educational category of those who experience a marital separation or a cohabiting union dissolution.
We measure macroeconomic conditions in two ways. First we include a measure of state-level unemployment. We use monthly unemployment rates from the Bureau of Labor Statistics’ (BLS) Local Area Unemployment Statistics series at the state level. These unemployment data were merged with the SIPP data by month-year and the reference person’s geographic location.5 Paul Amato and Brett Beattie (2011) find that unemployment tends to have the strongest effect on divorce when considering unemployment rates within the year or with a year lag. However, unemployment rates rose fairly quickly during the Great Recession’s fallout; thus, following Harknett and Schneider (2012), who analyze a similar period, we separate the un employment rates into quartiles and lag the quartiles by three months. As a second measure of the macroeconomic conditions, we include a measure of national-level recession using the U.S. Business Cycle Expansions and Contractions data from the National Bureau of Economic Research. Our period of analysis includes the 2001 recession, which lasted from March 2001 until November 2001, as well as the Great Recession, which lasted from December 2007 until June 2009. We merge these data with our SIPP sample by month-year.
Additional Time-Varying Controls
Total monthly family income is measured in each month by calculating the sum of the SIPP-generated total person income measures for the reference person and his or her spouse or partner. Age is included as the reference person’s age in years. We include a dummy variable indicating whether the reference person has children living in the household. We also include dummy variables for employment that indicates whether the male partner is employed and whether the female partner is employed. In some analyses, we include a dummy variable indicator for whether the couple experienced an income shock, measured as having experienced a reduction in income in the previous month from the month before. We also include a measure of material hardship in some analyses. For this measure, we used a SIPP topical module question asked once in the 1996, 2001, and 2004 panels and twice in the 2008 panel,6 which asks respondents whether they had difficulty meeting any of their essential household expenses, such as mortgage or rent payment, utilities bills, or medical expenses at any time in the past twelve months. We created a dummy variable for whether respondents experienced hardship and applied that variable to the past twelve months of observations for each respondent.
Our analyses are based on event history models of time to union dissolution. We use Cox [End Page 204] proportional hazards models to estimate the risk of failure, or dissolution, as a function of wealth, debt, and other family characteristics. Respondents who enter the survey period already in a marriage or cohabiting relationship are immediately in the risk set. Respondents who enter unions later during a SIPP panel enter the risk set the first time the union is reported in the survey. We measure time as months since union entry (or since the survey began for those already in a union), and participants are censored at the end of the survey period. We allow respondents to contribute multiple dissolutions and adjust for multiple relationships with robust standard errors. Thus, our unit of analysis is the relationship-month rather than the person-month.7
where h0(t) represents the baseline hazard rate at time t and Xn represents the vector of independent variables. Because we conduct a monthly survival analysis, but our key asset and debt independent variables are measured only every twelve months in the topical modules, we forward-fill the asset and debt values between topical modules. For example, if a couple responds to the topical module in wave 3, and again in wave 9, we fill in the wave 3 values for waves 4 through 8. This assumes that asset and debt values do not change between waves, but it is better than the alternative of linear interpolation, which is problematic if couples end their unions between waves of the topical modules.
We conduct three main sets of survival analyses. The first estimates the effects of total net worth, as well as detailed measures of secured and unsecured debt and liquid and illiquid assets on relationship stability. We then add interaction terms to the model to test whether the associations between wealth and union dissolution differ for married and cohabiting couples. We also test whether they differ for older or younger cohorts. Finally, we examine whether the effects of wealth and debt on relationship stability show a nonlinearity, entering separate dummy variables for quartiles of the wealth and debt distributions.
In a second set, we explore the symbolic and material meanings of wealth for relationship stability. Following Schneider (2011), we proxy the symbolic value of wealth, meaning that assets serve as a signal to others that a couple has reached the marriage bar and, thus, when marriage is appropriate, by testing whether simply holding any assets or debts affects relationship stability. We do so by including dummy variables for whether a couple holds each type of asset or debt. We then test whether joint ownership of the home is associated with relationship stability, relative to just the male partner owning the home, just the female partner owning the home, or no homeownership. We also examine the material role of wealth by considering whether having assets reduces the effect of an income shock on relationship stability, and whether self-reported financial hardships explain the associations between debt and relationship stability.
In a final set of analyses, we examine whether wealth contributes to population-level inequalities in relationship stability by race-ethnicity or macroeconomic condition. Specifically, we measure the baseline differences in relationship stability by race-ethnicity and then add in a standard set of economic controls typically used by family researchers, which includes household income, education, and employment. We then test whether adding our asset and debt measures to the model explains any more of the racial-ethnic differences in relationship stability, net of the standard set of economic controls.
Finally, we add our macroeconomic variables of state-level unemployment and national recession to the models. We test whether the associations between wealth and relationship stability vary by macroeconomic context, and whether these effects vary for married and cohabiting couples. [End Page 205]
The median couple in our sample has a net worth of $115,000, a figure that varies considerably between married and cohabiting couples and by race-ethnicity. The median married couple had about $68,000 in secured debt, $1,800 in unsecured debt, and $23,000 in liquid assets and $189,000 in illiquid assets. The median cohabiting couple, by contrast, had no secured debt, $700 in unsecured debt, $1,500 in liquid assets, and $24,000 in illiquid assets. These socioeconomic differences are also reflected in the divergent monthly household incomes and educational distributions for the two groups. Consistent with prior research, we also observe stark wealth inequalities between white and black couples, with Latino and other race couples falling in between (see table A2).
Assets, Debts, and Relationship Stability
Table 2 presents results from a Cox proportional hazards model of union dissolution. Model 1 includes family-level net worth (total assets minus total debts) and our full set of controls. A standard deviation increase in the value of a couple’s net worth decreases the risk of union dissolution by 31 percent (exp(−0.377) —1), controlling for other factors such as income, education, race, and employment. Model 2, which also adjusts for controls, tests whether components of net worth are differentially associated with relationship stability. We find that although a standard deviation increase in secured debt decreases the risk of dissolution by 12 percent, unsecured debt is not significantly associated the hazard rate. This is contrary to our predictions that secured debt would not affect short-term relationship stability and that unsecured debt would increase the hazard of dissolution. Liquid and illiquid assets are both associated with relationship stability as predicted, decreasing the risk of dissolution by 49 percent and 17 percent, respectively.8
In supplemental analyses, we examine whether assets and debts have nonlinear effects on relationship stability. To test for nonlinearity, we include asset and debt measures as quartile dummy variables rather than as continuous measures. The lowest quartile of each asset and debt measure is the reference category. We find evidence of relatively linear effects of asset holdings and secured debts on relationship stability, and the magnitude of the association increases monotonically as we move up the quartiles of the distribution (table A4). We do, however, find an interesting nonlinear association for unsecured debt, that only large amounts of unsecured debt have a significant negative influence on relationship stability. Those in the fourth quartile of unsecured debt (those holding the most unsecured debt) have an 8 percent higher risk of dissolution than those in the first quartile (those holding the least).9
Variation by Relationship Type and Age Cohort
Table 3 shows results from a set of models in which we explore how associations between wealth and relationship stability vary by relationship type and age cohort. First, we interacted the relationship-type dummy variable with each asset and debt measure (see table 3, relationship type). Models 1, 2, and 3 show that the associations of unsecured debt, secured [End Page 206]
[End Page 207]
[End Page 208]
debt, and liquid assets with the risk of relationship dissolution do not differ significantly for married versus cohabitating couples. Model 4, however, shows that illiquid assets significantly increase the risk of dissolution for cohabitating couples. We caution that this may be due to small numbers of cohabitating couples with illiquid assets. We therefore cannot reject the null hypothesis that assets and debts function similarly for married and cohabiting couples.
Table 3 also shows whether the associations between wealth and relationship stability differ for older and younger age cohorts. Our results here are largely consistent with predictions from a life-cycle model of savings and debt. A one standard deviation increase in the amount of secured debt decreases the risk of dissolution by 20 percent for younger cohorts, but only by 7 percent for older cohorts. Thus, secured debt is less protective of relationship stability for older couples than for younger couples. Unsecured debt has little association with stability for any age group. Model 7 shows that liquid assets have an increased protective effect among older couples. That is, a standard deviation increase in liquid assets among older couples decreases the risk of dissolution by 61 percent (exp(–0.121–1.065)); for younger couples, liquid assets are not significantly associated with relationship stability. The association between illiquid assets and relationship stability does not differ for older and younger cohorts.
Symbolic and Material Meanings of Wealth
Scholars have argued that wealth matters for relationships because of what it symbolizes, apart from its economic value (Schneider 2011; Zelizer 1997; Cherlin 2004). Ownership of assets, such as a home or a car, independent of their value, can be a symbolic marker of success and status; researchers have found that they matter for entry into marriage (Schneider 2011; Edin and Kefalas 2005). We build on this line of research by testing whether holding any asset or debt (in contrast to assessing the effect of amounts) is associated with relationship stability. The results indicate that simply having some assets and debts, versus none, is significantly associated with the risk of dissolution, controlling for other factors (see table A5).
The sole versus joint ownership of assets may also be symbolically significant, in that couples who hold their assets—homes, cars, bank accounts—jointly report greater commitment to their relationships and higher levels of relationship satisfaction, which may be the result of greater trust and support. The results indicate that sole homeownership, whether the owner is the woman or the man in the relationship, increases the risk of dissolution by about 60 percent over not owning (see table A5).10 In contrast to the effect of sole ownership, jointly owning the home decreases the hazard rate by 49 percent over not owning. We therefore find strong support for the symbolic value of asset and debt holdings.
Wealth has potentially important material value for couples as well. They can liquidate asset holdings or draw on interest to provide extra income. Wealth is an obvious buffer against unexpected financial insecurities. To examine the material meaning of wealth for relationship stability, we ask whether the effect of income shocks on relationship stability was weaker for couples with greater asset holdings. We tested this by including an indicator for whether the couple experienced an income shock in the prior month, measured as a negative income change from the month before, and interacting this measure with liquid and illiquid assets (see table A6). A negative income shock increases the risk of dissolution, and liquid assets reduce it, though the coefficient does not reach conventional levels of statistical significance. Illiquid assets do not alter the effect of an income shock in any substantively or statistically significant way.
We also examine the possibility that debts are either markers of financial hardship or directly create financial hardship via the cost associated with debt repayment and other fees (see table A6). Consistent with prior research, [End Page 209] we find that experiences of financial hardship increase the risk of relationship dissolution significantly. We also find that the financial hardship measure explains 29 percent of the association between large amounts of unsecured debt and relationship stability.11
Wealth and Racial Inequality in Relationships
Researchers have found that socioeconomic differences, measured by income, education, and employment, explain part of the relationship stability gap, but much remains unexplained. We examine whether assets and debts explain part of the black-white gap in relationship stability. Model 1 in table 4 includes only race-ethnicity dummy variables and controls for age, children, and relationship type. This first model shows that black couples are 53 percent more likely to end their relationships than white couples. Model 2 adds in the economic measures that previous research has found decrease this gap: income, employment, and education. Indeed, in this model, the likelihood of dissolution for black couples falls to 41 percent more than for whites, leaving a significant portion of the gap unexplained. Model 3 adds our measures for assets and debts. As these results show, the likelihood of dissolution for black couples is now 29 percent higher than for whites. This suggests that assets and debts reduce the black-white relationship stability gap by about 45 percent, which is about as much as the standard set of economic controls explained. In other words, assets and debts explain a substantial portion of the black-white gap in relationship stability, rivaling that of other standard socioeconomic measures.
We find no significant differences between the relationship stability of white and Latino couples when adjusting only for age, children, and relationship type. However, in model 2, where income, employment, and education are adjusted, Latino couples actually face an 11 percent lower risk of dissolution than white couples. Latinos are 16 percent less likely than whites to end their unions in model 3 when adding in assets and debts.
Macroeconomic Conditions and Relationship Stability
The theoretical predictions for how macroeconomic conditions shape divorce are mixed: some theories suggest that adverse conditions would reduce marital stability, and others that adverse conditions would actually promote marital stability by making it more costly to divorce. The predictions for cohabitation are more clear, however, suggesting that cohabitations would be more stable in times of macroeconomic hardship.
In this final set of results, we consider the effects of macroeconomic conditions on relationship stability and examine whether they differ for marriages and cohabitations. Models 1 and 2 in table 5 show that macroeconomic conditions do not attenuate the relationship between assets, debts, and relationship stability as we expected they might.
Model 2 of table 5 shows that couples living in states where unemployment levels are in the second are not at significantly more risk of dissolving their unions than those in states at just the first quartile (lowest unemployment). However, those in states in the third and fourth (highest unemployment) quartile face a 7 and 26 percent higher risk of union dissolution relative to those with the lowest unemployment. National-level recession increases the risk of dissolution by 54 percent. Model 2 adds the wealth and debt measures to the models. These measures do not appear to mediate the association between state-level unemployment and relationship dissolution.
Model 3 interacts the cohabitation dummy variable with the fourth quartile of unemployment dummy variable. The results indicate important differences in macroeconomic effects for married versus cohabiting couples—married couples face an increased hazard rate in poor macroeconomic conditions while cohabiting [End Page 210] couples actually face a decreased risk. That is, comparing two cohabiting couples, the couple living in a state with the highest quartile of unemployment rather than a state in the lowest quartile of unemployment is 16 percent less likely to break up (exp(0.247–0.417)). This provides support for the hypothesis that cohabiting relationships are more stable during tough macroeconomic times, and that marriages are less stable.
[End Page 211]
Our analysis provides a nuanced portrait of how wealth is related to the stability of family relationships and explores how this association varies across types of debt and types of unions. We find that both liquid and illiquid assets are associated with the stability of marital relationships. Consistent with Schneider (2011), we find evidence that these associations reflected both the material as well as the symbolic values of wealth for relationships. Evidence also suggests that liquid assets buffered against the adverse consequences of transitory shocks to income. The protective effect of liquid assets was particularly pronounced for older age cohorts, consistent with a life-cycle theory of savings. Holding any kind of asset is associated with relationship stability, consistent with an interpretation that assets hold symbolic meaning, independent of their actual amount. The joint ownership of assets also appears to have symbolic value for relationships, given that joint ownership of a home is associated with relationship stability relative to renting, but sole ownership by either partner is less stable than not owning a house.
Associations between debt and relationship stability are more mixed. Large amounts of unsecured debt are associated with a reduction in marital stability, in part because these couples reported greater financial hardship. Unsecured debt may therefore either create financial hardship directly or be a marker for it. Secured debts are associated with an increase in marital stability, however. Secured debts, like mortgages, are investments made to boost [End Page 212] long-term well-being, which may explain why they are associated with an increase rather than a decrease in stability. This may also explain why secured debt is a stronger predictor of relationship stability for younger couples than for older couples. Another possibility for the differences in the strength of the association between secured debt and dissolution for older and younger couples may be that investment or willingness to take on secured debt is a stronger signal of maturity in younger couples; older couples may have other signals to rely on (Brüderl and Kalter 2001).12 However, if unsecured debt worked solely as a signaling mechanism, illiquid assets would differ between older and younger cohorts, which we do not find.
Although theory suggests that wealth and debt shape union stability differently for married and unmarried couples, we find little evidence for this in our analysis. If marriage is protective because of its legal and institutionalized commitment mechanisms, married couples might be more likely to stay together in the face of adverse wealth conditions. If, however, couples hold marriage in high esteem, as the marriage bar theory suggests, married couples might be more likely to break up in the face of asset or debt adversity than cohabiting couples. Contrary to both of these theories, we find no significant differences between married and cohabiting couples in terms of how wealth and debt shaped the stability of their romantic unions. Several possible reasons clarify this null finding. First, it could be that both theories are at work and cancel each other out. Second, the relatively few cohabiting unions in our analysis and large standard errors around our interaction terms mean that we cannot rule out potentially meaningful differences among these two types of unions.
Consistent with prior research, we find substantial racial differences in relationship stability: black couples were 53 percent more likely to end their relationships than whites were. The conventional socioeconomic measures of income, employment, and education explained a portion of this black-white gap in relationship stability. When we include these measures in the models, the increased risk of dissolution for blacks relative to whites drops by 23 percent to 41 percent. When we add measures of wealth and debt to the models, they explain a significant additional portion of the gap: the greater relative likelihood of dissolution for blacks drops to 29 percent, a further reduction of about 29 percent. Prior research finds that wealth measures explained about 30 percent of the black-white marriage gap in marriage entry (Schneider 2011); here we find that wealth also explains a significant portion of the black-white gap in the stability of relationships after they form. This reduction may occur because wealth has a similar use value within couples of any race-ethnicity, buffering hardships or smoothing consumption. However, wealth may not serve the same symbolic value within relationships of all races and ethnicities because different racial-ethnic groups do not have the same access to assets (Brimmer 1988). For example, black Americans have more difficulty getting a mortgage or getting the same type of mortgage as white Americans (Rugh and Massey 2010). Examining the effect of different types of assets on the racial gap in relationship stability is an important area for future research. Our results suggest that the black-white wealth divide may have lasting consequences for the intergenerational reproduction of inequality via its effects on family instability, independent of other measures of socioeconomic status.
Finally, we examine the role of macroeconomic conditions on the stability of family relationships. We find that high levels of state unemployment (in the top quartile) were associated with an increased risk of dissolution, relative to periods of low unemployment. We also examine whether these associations differ for cohabiting unions and predict that cohabiting unions may be more stable during hard times because cohabitation allows for pooling limited resources. We find support for this theory in that cohabiters’ risk of dissolution is significantly lower in states with high unemployment than in states with low unemployment.
Our analysis has several limitations that readers should keep in mind when interpreting [End Page 213] the findings. First, our use of the SIPP data precludes our ability to look at longer-term trajectories of marital instability and wealth accumulation over the entire life course, as each SIPP panel lasts only three to four years. Second, our analysis has focused on wealth as a key predictor of relationship stability, but relationship instability is also an important potential cause of declining assets or growing indebtedness; indeed, divorce is one of the key antecedents of bankruptcy (Sullivan, Warren, and Westbrook 1999). Isolating the causal effect of relationship instability on changes in wealth is tricky: a host of unobserved factors could cause both relationship instability and financial hardship (Fisher and Lyons 2006). This is clearly an important question for future research to disentangle because it can help provide more precise estimates of the role of family instability in producing wealth inequality (see Killewald and Bryan, this issue).
Taken together, our results highlight the important yet understudied role of wealth on the stability of family relationships. Much of the research has focused on how wealth explains gaps in marriage entry; here, we find that wealth plays an important role in shaping marital stability as well. Debt and assets are significantly associated with the stability of both marital and cohabiting relationships. The importance of debt and assets remains net of the standard set of socioeconomic controls of education, employment, and income, and the magnitude of wealth effects is often comparable to the magnitude of these measures. Furthermore, assets and debts appear to have not only material value for relationships, buffering against income shocks and either creating or ameliorating financial hardships, but also symbolic value. Ownership of assets in and of themselves can be a marker of status, and joint ownership can signal relationship commitment. Because family instability has adverse consequences for children (McLanahan, Tach, and Schneider 2013), our results suggest that family instability may be one important mechanism through which the inter-generational transmission of wealth inequality operates. As a result, policy interventions that reduce wealth inequality may also reduce inequalities in children’s exposure to family instability. [End Page 214]
Alicia Eads is a graduate student in sociology at Cornell University.
Laura Tach is assistant professor of policy analysis and management at Cornell University.
[End Page 215]
[End Page 216]
[End Page 217]
[End Page 218]
[End Page 219]
[End Page 220]
The authors thank Fabian Pfeffer, Bob Schoeni, and participants at the Russell Sage Foundation conference on wealth inequality for feedback on an earlier version of this manuscript.
2. During the recent housing boom, this relationship between secured debt and long-term economic well-being may have been more tenuous as homeowners increasingly cashed in home equity to finance consumption. We thank a reviewer for pointing this out. This became more prevalent beginning in 1999 (Brown et al. 2010). We confirm that relationships between secured debt and the likelihood of relationship dissolution are not significantly different for respondents in the 1996 and 2008 SIPP panels.
3. The assets and debt topical modules were asked in the following waves: 3, 6, 9, and 12 for the 1996 panel; 3, 6, and 9 for the 2001 panel; 3 and 6 for the 2004 panel; and waves 4, 7, and 10 for the 2008 panel.
5. For the 1996 and 2001 panels, the SIPP combines two sets of states. Those living in North Dakota and South Dakota were coded identically, as were those living in Vermont and Maine. We averaged the unemployment data for these sets of states and applied the average to respondents living in these areas.
6. The financial hardship question was asked in the following waves: wave 8 for the 1996 panel, wave 8 for the 2001 panel, wave 5 for the 2004 panel, and waves 6 and 9 for the 2008 panel.
7. To address censoring, we perform our analyses using two subsamples, which have relationship duration information, and thus, for these analyses, time is measured since the start of the relationship for everyone in the subsamples.
8. We reestimated models 1 and 2 above on two subsamples to ensure that our results were not driven by the left censoring in our full sample. Table A3 shows results that count time since the start of marriage for the subsample of respondents who completed the marital history topical module. It also presents results for the subsample of respondents, married or cohabiting, who entered a relationship during the survey period, for whom we observe the beginning of the relationship during a SIPP panel. The results for these subsamples do not differ substantively from the full sample results.
9. We also separated the value of a couple’s mortgage and home value from these values, given that home equity makes up the largest share of most Americans’ investment portfolios and their mortgages are the greatest contribution to their levels of debt (Wolff, this issue; Killewald and Bryan, this issue). We found that the asset-debt associations described are not simply a house effect: they hold for assets and secured debts other than homes as well. We also test whether being underwater on a mortgage—owing more than the house is worth— affects relationship stability, possibly by increasing the costs of dissolution. However, we do not find significant effects.
10. About 2 percent of our sample lives in a household in which the female partner is the sole owner of the house. About 2 percent live in a household in which the male partner is the sole owner.
11. The models in the second panel of table A6 are estimated on a subset of observations because respondents’ answers to the financial hardship question applied to only twelve months of the three-to four-year survey period, thus many observations could not be used for this analysis.
12. We thank a reviewer for pointing this out.