Wealth and Divorce

Abstract

In the United States, wealthier couples have lower divorce risk. Wealth may stabilize marriage through its material value, especially by easing financial stress, or by providing symbolic resources, especially signaling that couples meet normative financial standards for marriage. We first show that the negative association between wealth and divorce holds net of a rich set of controls. All else being equal, having $40,000 in wealth rather than $0 is associated with as big a decline in average predicted divorce risk as having no nonmarital births versus at least one. Second, we show that the negative association between wealth and divorce risk is steepest at low positive wealth levels. Net of covariates, having $40,000 in wealth rather than $0 is associated with as big a decline in average predicted divorce risk as having $400,000 rather than $40,000. Third, we consider evidence for the symbolic perspective, which emphasizes the stabilizing role of owning visible physical assets, and the material perspective, which suggests unsecured debt heightens divorce risk. Consistent with the symbolic perspective, we find that with net worth held constant, ownership of homes and vehicles is negatively associated with divorce risk. However, more research is needed to fully adjudicate between the symbolic and material perspectives.

Introduction

Wealth inequality in the United States is high and rising (Pfeffer and Schoeni 2016). Variation in marital stability amplifies wealth inequality: wealthier couples are more likely to stay married (Eads and Tach 2016; Eads et al. forthcoming), and remaining married is associated with higher wealth (Addo and Lichter 2013; Yamokoski and Keister 2006; Zagorsky 2005). If wealth stabilizes marriages, rising wealth inequality may also further stratify American family life.

Wealth may increase marital stability through its material value, providing couples with economic resources they can use to improve their marital quality, thereby lowering their divorce risk. Alternatively, wealth may stabilize marriages through its symbolic value, allowing couples who have achieved the economic success expected of married couples to receive more social support or perceive greater value in their marriage. Or the association between wealth and divorce may not reflect a causal relationship.

In this article, we describe the association between wealth and divorce risk for Americans born in the late 1950s and early 1960s and explore the underlying mechanisms linking wealth and marital stability. We first consider that the association between wealth and marital stability may be spurious rather than causal. Although we cannot estimate the causal effect of wealth on divorce, we show that the association between wealth and marital stability remains net of a richer set of control variables than considered in prior research. Thus, we show that wealth is a distinct predictor of divorce risk, above and beyond other measures of socioeconomic resources, such as income and education.

Next, we seek to uncover the sociological processes that lead to greater marital stability for wealthier couples. Again, we cannot draw firm causal conclusions, but we evaluate whether empirical patterns are consistent with the predictions of various theoretical perspectives. We consider that wealth's material benefits may largely derive from avoiding acute financial stress and that its symbolic benefits may largely derive from clearing a wealth “marriage bar” that defines a minimum appropriate level of affluence for married couples. Both possibilities suggest the wealth–divorce association will be more negative toward the bottom of the net worth distribution than at the top. Consistent with these predictions, we find the largest differences in divorce risk across modest values of positive net worth, not between the highest wealth levels and more moderate wealth.

Finally, we continue to investigate the possible underlying mechanisms linking wealth and marital stability by seeking to determine whether the negative association between net worth and divorce risk is due to wealth's material or symbolic value. Because each perspective foregrounds particular assets or debts as especially important for marital stability, we examine how portfolio composition is associated with divorce, conditional on total wealth. Unsecured debt—not backed by collateral—may increase financial stress. Home and vehicle ownership, in addition to their material value, are visible symbolic markers of middle-class status. We find that, holding net worth constant, ownership of homes and vehicles is negatively associated with divorce risk, consistent with the symbolic perspective. However, further research is needed to fully evaluate and distinguish between the material and symbolic perspectives.

Wealth's Material Value in Reducing Divorce Risk

In this section, we describe how wealth's material value may shape couples' risk of divorce, and we denote the associated hypotheses with “M” for “material.” In the next section, we consider the possibility that wealth carries symbolic value that reduces divorce risk, and we denote the associated hypotheses with “S” for “symbolic.”

Material economic resources may improve couples' relationship quality by allowing them to face fewer contentious decisions about spending, outsource household labor, experience less crowded living quarters, and engage in valued leisure activities. By contrast, limited economic resources are associated with increased economic strain, which is associated with poorer marital quality (Conger et al. 1990). Eads and Tach (2016) theorized that assets may stabilize relationships by buffering negative economic shocks and reducing material hardship. Consistent with this idea, consumer debt is positively associated with economic pressure and negatively associated with marital quality, whereas assets tend to have the opposite associations (Dew 2007, 2009, 2011). Thus, the material perspective on wealth suggests wealthier couples are less likely to divorce because they can use their financial resources to purchase valued goods and services and achieve greater economic stability in ways that improve their relationship.

One material benefit of net worth is the reduction of financial stress. The marginal returns to net worth for reducing financial stress are expected to be highest at low net worth levels and to decline as net worth rises. Put differently, avoiding net debt and having some asset safety net—being comfortable versus precarious—should reduce financial stress more than being very wealthy versus moderately wealthy. If the material benefits of wealth reduce divorce risk by reducing financial stress, higher net worth should reduce divorce risk most among those experiencing acute financial stress: those with low net worth.

Finally, the material perspective suggests that different types of assets and debts may have different consequences for marital stability. Imagine that two couples have the same total net worth, but the first couple holds $10,000 in unsecured debt, whereas the second couple has no unsecured debt and $10,000 less in home equity than the first couple. In theory, assets and debts are fungible. However, in practice, transaction costs may prevent converting home equity to cash to pay off debt, and the first couple may be strained by monthly bills to repay the loans. Thus, the couple holding unsecured debt is expected to experience greater financial stress and marital instability. Having a small amount of unsecured debt may not be disruptive because it may permit couples to cushion short-term costs, such as by charging car repair costs on a credit card, but more unsecured debt is expected to be associated with greater marital instability.

Wealth's Symbolic Value in Reducing Divorce Risk

A second perspective predicts that wealth promotes marital stability through the symbolic cultural resources it provides rather than its financial value. Under this perspective, prior scholarship has theorized that asset accumulation may be one way that spouses perform and meet expected marital roles (Dew 2007, 2009) and that wealth's symbolic benefits may improve couples' interactions and commitment (Eads and Tach 2016). Thus, the symbolic perspective, too, predicts a negative association between net worth and divorce risk.

Marriage has become a capstone event, to be entered into only after achieving other life course milestones (Cherlin 2004). Unmarried couples describe delaying marriage until they have achieved what they perceive as a sufficient economic standard—clearing an economic bar—which may include owning a car and home, paying off debts, or saving enough to pay for a wedding (Cherlin 2004; Edin and Kefalas 2005; Gibson-Davis et al. 2005; Smock et al. 2005).

Once married, couples whose marriages do not meet these normative standards may experience marital dissatisfaction and higher divorce risk. This logic suggests that wealth's symbolic value for marriage primarily distinguishes between couples who have and have not achieved a minimum level of affluence deemed appropriate for married couples. A couple who once met the marriage bar may divorce because they no longer meet the bar. A couple who never met the marriage bar may divorce because other aspects of their marriage become less rewarding and no longer offset the costs of not meeting the bar.

So far, the predictions of the material and symbolic perspectives on wealth are identical: both predict a negative relationship between net worth and divorce risk, and versions of each perspective suggest the steepest association at the bottom of the wealth distribution. Therefore, in what follows, we drop the “M” and “S” designators when referring to Hypotheses 1 and 2.

However, the two perspectives highlight different wealth portfolio components as particularly relevant for marital stability. The material perspective suggests that unsecured debt will increase financial stress and, therefore, marital instability, even conditional on total net worth (Hypothesis M3). By contrast, the symbolic perspective emphasizes the importance of visible assets in demonstrating that the couple has cleared the marriage bar. Visible assets, such as homes and cars, feature in unmarried couples' descriptions of desired economic thresholds to clear before marriage (Edin and Kefalas 2005; Gibson-Davis et al. 2005; Smock et al. 2005). The same logic may apply to decisions to remain married: ownership of visible assets demonstrates to the couple and others that they meet the marriage bar. Thus, the symbolic perspective predicts greater marital stability for owners of homes and vehicles, even conditional on total net worth. Put differently, wealth in the form of homeownership or vehicle ownership is likely to confer more stabilizing symbolic resources than holding the same amount of wealth in a less visible asset, such as a retirement account.

Homes and vehicles are not the only components of wealth with symbolic value; as mentioned earlier, couples considering marriage also attach meaning to paying down debts and saving for a wedding. However, homes and vehicles are two visible, symbolic, and easily measured aspects of net worth.

We caution that a lack of support for Hypothesis 2, Hypothesis M3, or Hypothesis S3 would not rule out the material or symbolic perspectives. Wealth's material and symbolic benefits for marital stability may operate through processes other than reducing acute financial stress or clearing a normative marriage bar (Hypothesis 2). Even if these processes are at work, our analyses may not capture them: unsecured debt may not heighten financial stress more than other components of net worth (Hypothesis M3), and homes and vehicles may not carry more symbolic value than other assets (Hypothesis S3). Thus, our analyses probing the potential underlying mechanisms linking wealth and divorce risk are exploratory rather than definitive.

Prior Research and Remaining Questions on the Wealth–Divorce Association

Is Wealth Associated With Lower Divorce Risk?

Using Survey of Income and Program Participation (SIPP) data, Eads and Tach (2016) found that, net of controls, a one-standard-deviation increase in net worth was associated with a 31% decline in the risk of union dissolution.¹ Their model included a parsimonious set of controls: family income; both partners' employment statuses; the household reference person's race/ethnicity, age, and educational attainment; whether there are children in the household; and whether the union is a marriage or cohabitation.

In earlier research, Galligan and Bahr (1978) and Ross and Sawhill (1975) also found a negative association between wealth and marital disruption. More recently, Dew (2009, 2011) found that marital dissolution risk rose with consumer debt and fell with assets, whereas Sanchez and Gager (2000) found that divorce risk was not significantly associated with capital assets, capital debt, or consumer debt. Because Eads and Tach (2016) provided the most comprehensive evaluation of the association between wealth and marital stability, considering both net worth and asset and debt components and seeking to distinguish between the symbolic and material perspectives on wealth, we focus on how our analyses differ from theirs.

Our first contribution is to include a richer set of control variables to provide a more rigorous test that the wealth–divorce association is robust to adjusting for confounding factors (Hypothesis 1). We use National Longitudinal Survey of Youth 1979 cohort (NLSY79) data. Relative to the SIPP, the NLSY79 has the key advantage of including detailed measures of respondents' social origins and family demographic characteristics. Although some controls we add may be endogenous to wealth, both the material and symbolic perspectives predict a direct effect of current net worth on marital stability, above and beyond how prior net worth may have shaped characteristics such as marriage timing or fertility. We reiterate that we cannot estimate the causal effect of wealth on divorce, but our inclusion of additional control variables provides stronger evidence that the association between wealth and divorce is not entirely spurious.

Is the Wealth–Divorce Association Due to Wealth's Material Benefits?

Like us, Eads and Tach (2016) considered financial stress to be one possible manifestation of the material perspective. They found that being in the fourth quartile of unsecured debt holdings was associated with a statistically significantly higher risk of union dissolution than being in the first quartile. Furthermore, they showed that material hardship, measured as reported difficulty meeting essential household expenses, explained approximately 30% of this disparity. The NLSY79 does not include measures of respondent-assessed material hardship, so we cannot replicate that portion of their analysis.

However, we extend Eads and Tach's (2016) analysis of unsecured debt by conditioning on total net worth (Hypothesis M3). This allows us to test whether there is anything distinctive about unsecured debt's relationship to marital stability, as opposed to merely increasing divorce risk as any other decrease in total net worth would. For example, recalling our two hypothetical couples from above, finding that the two couples, who have the same total net worth, have the same divorce risk, despite their different portfolio compositions and differences in unsecured debt, would not provide equally strong evidence for the financial stress perspective as showing that differences in unsecured debt are related to divorce risk with total wealth held constant.²

Although Eads and Tach (2016) estimated a model in which they specified secured debt, unsecured debt, liquid assets, and illiquid assets with quartiles, they did not formally test whether the associations were consistent with linearity. Thus, our tests of the shape of the association between net worth and divorce (Hypothesis 2) are a new way to evaluate whether empirical patterns are consistent with the expectations of financial stress (and the economic marriage bar). Furthermore, describing the shape of the wealth–divorce association sheds empirical light on the economic stratification of family life, clarifying which wealth positions are associated with particularly distinctive divorce risks.

Is the Wealth–Divorce Association Due to Wealth's Symbolic Benefits?

Eads and Tach (2016) considered that wealth's symbolic benefits may result from clearing a marriage bar. However, as noted earlier, they did not test the implication that the wealth–divorce association is more negative at lower wealth levels, so our analyses testing for this nonlinearity are new (Hypothesis 2).

Like us, Eads and Tach (2016) argued that the symbolic perspective suggests that asset ownership affects union stability. They found that binary indicators for having any secured debt, any liquid assets, and any illiquid assets were each negatively associated with the risk of union dissolution, whereas having any unsecured debt had the opposite association. We argue that the interpretation of this pattern is not straightforward. Those who own a given asset, unless they owe more on the asset than it is worth, hold more value in the asset than nonowners; thus, an association between binary measures of ownership and marital stability does not reveal whether ownership has a larger role in marital stability than increases in asset value conditional on ownership. In our models seeking to evaluate whether asset ownership has symbolic value that reduces divorce risk (Hypothesis S3), we include both indicators of asset ownership and measures of the value of the asset, distinguishing the role of ownership from that of financial value. Furthermore, we control for net worth, isolating the predictive power of portfolio composition net of total wealth. Controlling for net worth allows us to assess whether asset ownership is associated with divorce risk separate from the fact that owners of an asset tend to be wealthier overall.

Our analyses also highlight that not all assets and debts have equal symbolic value. We focus on two examples of visible assets identified as symbolically meaningful in prior research: vehicles and homes (Edin and Kefalas 2005; Gibson-Davis et al. 2005; Townsend 2002). Prior research has found that homeownership is negatively associated with divorce (Cooke 2006; Ono 1998; South 2001) but has not evaluated whether this association merely reflects the overall negative association between net worth and divorce risk.

Last, Eads and Tach (2016) found homeownership to be associated with a lower risk of union dissolution only when partners jointly own the home, consistent with the possibility that joint investments facilitate relationship stability by symbolizing or encouraging commitment. In subsequent research, Eads et al. (forthcoming) found that couples who held a greater share of their assets and debts jointly had lower divorce risk. Because NLSY79 measures wealth only at the couple level, we cannot replicate this analysis.

Data and Methods

Data Source: NLSY79

We used data from the 1979–2018 waves of the NLSY79 (U.S. Bureau of Labor Statistics 2022). The NLSY79 first surveyed a nationally representative sample of young adults aged 14–22 in 1979 and, except for some discontinued subsamples, attempted to interview them annually until 1994 and then biennially.

The NLSY79 collected information on net worth in all waves between 1985 and 2000 except 1991, and subsequently in 2004, 2008, 2012, and 2016. We excluded observations before the first collection of wealth information in 1985. We included observations from 1991, 2002, 2006, 2010, and 2014, even though net worth was not collected in these years; as described later, we multiply imputed wealth in those years.

Defining the Risk Set

At each survey wave, we identified whether each respondent was married and therefore at risk of divorce. The NLSY79 asked respondents to report their marital status at each survey wave and the month and year of changes in marital status. Using this information, NLSY79 created variables indicating the start and end dates of respondents' marriages, with the end of a marriage defined by divorce or widowhood. We created revised marriage dissolution dates, treating a marriage as dissolved once either the reported marriage end date occurred or the respondent reported separated as their marital status and was never observed reunited with their spouse at a subsequent interview. Thus, although we refer to “divorce” for brevity throughout, we considered marriages to have ended as soon as permanent separation began, regardless of whether or when divorce occurred. We defined a respondent as currently married if the interview month was after the start date of one of their marriages and before our constructed month of the marriage's dissolution. When necessary, such as when the interview month and the month of the marital transition were the same, we broke ties using the respondent's current marital status and changes in marital status since the last interview. Once a marriage had ended, the respondent exited the risk set until they married again (if ever), at which point they reentered the risk set and analytic sample.³

Outcome Variable: Divorce

Our outcome variable is whether a respondent in the risk set divorced or permanently separated before the next survey wave in which the respondent was observed. For respondents whose marriages ended, we used wave-specific reports of marital status and interwave changes in marital status to determine whether the marriage ended in widowhood rather than separation or divorce.

Core Predictor Variables: Net Worth, Assets, and Debts

The NLSY79 collected data on whether respondents and their spouses held a variety of assets and debts and, for those they held, the value. Questions about assets and debts differed across waves but included assets such as homes; vehicles; valuable items or collections; farms, businesses, and real estate other than residential homes; financial assets, including bank accounts, investment accounts, and retirement accounts; and other debt, such as to stores, hospitals, and banks, which we refer to as unsecured debt.

The NLSY79 computed couples' total net worth, which is the sum of the values of all assets the respondent and their spouse held, less the values of all debts. We converted all financial variables to constant 2020 dollars using the Consumer Price Index and then top- and bottom-coded at the 95th and 5th percentiles of the unweighted distribution in the analytic sample before multiple imputation.

We specified net worth as a linear spline with knots at the quartiles of the pooled, weighted distribution of married couple-years in the sample: $29,937; $108,813; and $319,011. We included an additional knot at $0 because net debt may be a distinctive state compared with low positive net worth.

In the analyses examining portfolio composition, testing Hypotheses M3 and S3, we also included three indicators for whether the couple had any unsecured debt, owned a vehicle, or owned a home, as well as three linear terms for home value, vehicle value, and unsecured debt value. In a second model, we included measures of home and vehicle equity (values less debts) rather than their values.

Control Variables

We controlled for marital duration with linear and quadratic terms for the number of years between the interview year and the year the respondent's marriage began.

We used the NLSY79 household screener to categorize respondents as Hispanic, non-Hispanic Black, or non-Black and non-Hispanic. We measured nativity with an indicator for whether the respondent was born in the United States.

We controlled for the respondent's parents' highest grade completed using four categories: less than 12th grade, 12th grade, one to three years of college, or four or more years of college. We also included an indicator for whether the respondent lived with two biological parents at age 14.

We measured each spouse's current educational attainment using the same categories as for parents' education, and we included an indicator variable for whether the respondent was currently enrolled in school. For each spouse, we measured whether they were employed full-time (at least 1,500 hours) in the prior calendar year by multiplying reported weeks and hours per week worked. We controlled for family income in the prior calendar year with a linear spline with knots at the weighted quartiles ($58,894; $88,247; and $127,706).

We controlled for respondents' age at marriage using three categories: younger than 21, 21–24, and older than 24. We measured prior nonmarital cohabitation with an indicator set to 1 if the respondent ever previously reported (1) a partner on the household roster while not married, (2) having cohabited before marriage with the most recent spouse (asked in 1990–2000), or (3) nonmarital cohabitation between survey waves (asked beginning in 2002). We measured nonmarital fertility with an indicator set to 1 if the respondent's reported birthdates of any of their children born to date fell outside the respondent's marriage spells as defined in our construction of the risk set. Using three categories, we controlled for whether the current marriage was the respondent's first, second, or third or higher. We controlled for the presence of children in the household with counts of the respondent's biological children and stepchildren in the home, each divided into those under 5 and those aged 5–17 and each top-coded at four.⁴

We controlled for region with four categories: Northeast, North Central, South, or West. We included an indicator for whether the respondent lived in an urban versus rural area.

Analytic Plan

We estimated discrete-time hazard models with a logit link, modeling the log odds of a married respondent divorcing before the next survey wave in which they were observed and adjusting for the length of exposure—the difference between the wave in which the predictors were measured and the wave in which the divorce outcome was measured. Our baseline model describes the association between net worth and the hazard of divorce, controlling only for marital duration. This model describes how divorce risk is stratified by wealth position.

Next, we estimated the full model, which includes the control variables described previously. We tested Hypothesis 1 by evaluating whether the slopes on the net worth spline pieces were jointly statistically significantly different from zero. We tested Hypothesis 2 by evaluating whether the slopes on all the spline pieces were identical and therefore that the association was linear in log odds. We used a .05 significance level with two-tailed tests throughout.⁵

To describe the magnitude of the association between net worth and divorce risk, we computed predictive margins, which give the average predicted probability of divorce in the next year for respondents if they were all assigned a particular value of net worth but otherwise had their own covariates. The predictive margins do not refer to the average predicted probability of divorce for those observed to have a particular net worth; they refer to the average predicted probability of divorce for the entire sample if their wealth were set to the given value. We annualized estimates by assigning each observation an exposure period of one year when generating predicted probabilities.

To test Hypotheses M3 and S3, we added to the full model indicators for homeownership, vehicle ownership, and holding any unsecured debt, plus linear measures of home value, vehicle value, and the value of unsecured debt. We then repeated this model but replaced the measures of vehicle and home values with measures of their equity.⁶

To contextualize the magnitudes of the associations in these models, we again used predictive margins, this time examining how the predicted probability of divorce changed with different portfolio compositions while holding net worth constant.

We weighted all analyses using the NLSY79 year-specific weights and clustered standard errors at the 1979 household level.

Sample Restrictions and Missing Data

We censored individuals if they attrited before the next survey wave and censored all other respondents at the final wave in 2018. We censored marriage spells that ended in widowhood before the next survey wave. For 4% of the remaining observations, we could not determine whether they were in the risk set because their marriage start dates created by the NLSY79 were incomplete or inconsistent. We excluded these cases from the analytic sample, along with observations from currently unmarried respondents.⁷ In the remaining sample of married couples, we excluded 2% of the observations because the exposure period—the number of years between the survey wave in which the predictors were measured and the survey wave in which the respondent was next observed—exceeded two years.

Only 22 same-sex couples met the criteria for inclusion in our sample. We restricted our analysis to different-sex couples because the correlates of union dissolution may differ for different-sex and same-sex couples (e.g., Weisshaar 2014).

Our analytic sample includes 88,660 couple-year observations from 8,351 respondents and 10,286 marriages. The sample includes 4,161 divorces.

We used multiple imputation with 20 imputations for item-level missing data. Because the NLSY79 did not collect wealth information in all waves, the highest missing data rates are for net worth (36%) and the home, vehicle, and unsecured debt components (23% to 30%). In waves in which the NLSY79 collected net worth data, the missing rate is 17% for net worth and no more than 9% for the home, vehicle, and unsecured debt components. Our imputation models include respondents' most recent report of the net worth and home, vehicle, and unsecured debt variables provided that it was within the last four years and while the respondent was in the same marriage. Including this most recent report allowed the couple's prior wealth to inform their imputed wealth in the years that the NLSY79 did not collect wealth data. The online supplement shows missing data rates and the results of models that treat missing data using listwise deletion and that exclude observations from survey waves in which wealth information was not collected.

Results

As shown in Table 1, which describes our analytic sample at the couple-wave level, 4% of the married respondents observed at a given wave divorce before the next survey wave in which they participated. Mean net worth after top- and bottom-coding is $226,597, and the median is $108,813. Net worth is less than zero for 7% of the sample. Further, 76% of the sample are homeowners, vehicle ownership is near universal (97%), and 57% hold unsecured debt. Among owners, median values are $191,938 for homes, $21,081 for vehicles, and $5,580 for unsecured debt.

The Robust Nonlinear Association Between Wealth and Divorce

Table 2 shows the logit coefficients for the net worth terms from the baseline and full models.⁸ As shown in the bottom row of Table 2, for both models, the net worth spline terms are jointly statistically significantly associated with divorce (i.e., we can reject the null hypothesis that the slopes on the wealth spline terms are all zero). In both models, net worth is negatively and statistically significantly associated with divorce for low positive values of net worth ($0–$29,937). For the middle 50% of the wealth distribution ($29,937–$319,011), net worth remains negatively associated with divorce, but the associations are statistically significant only in the baseline model. In both models, the association between wealth and divorce is not statistically significant for either the very bottom (below $0) or the top quartile (above $319,011) of the wealth distribution. Overall, the results support Hypothesis 1 that net worth is negatively associated with divorce risk, although perhaps not across the entire distribution. Given the rich set of controls in our full model, our results provide more rigorous support for the claim that wealth is a distinct predictor of divorce, although we still cannot draw firm causal conclusions.⁹,¹⁰

In both models, the negative association between wealth and the log odds of divorce is most pronounced for positive values of net worth below the 25th percentile of the distribution (between $0 and $29,937) and then attenuates as net worth rises. As shown in the second-to-bottom row of Table 2, for both models, we can reject the null hypothesis that the association between wealth and the log odds of divorce is linear (i.e., that the wealth spline coefficients are all equal). On the log odds scale, these results support Hypothesis 2, that the association between net worth and divorce is more negative at lower (positive) values of net worth. The association between net worth and divorce risk remains statistically significant and nonlinear when sibling fixed effects are added to the full model, although the pattern of coefficients across the net worth distribution is somewhat different (see the online supplement).

In both models in Table 2, increases in net worth among net debtors (i.e., declines in net debt) are positively and not statistically significantly associated with divorce risk. Only 7% of the sample are net debtors (Table 1), and the associations are imprecisely estimated. Therefore, we cannot draw firm conclusions about how, if at all, divorce risk changes across net debt values. However, divorce risk may not increase at higher net debt levels if access to credit allows couples to meet unexpected costs and smooth consumption.

Next, we calculated predictive margins, which allow us to assess whether the wealth–divorce association is nonlinear in predicted probabilities as well as log odds. To visualize stratification in divorce risk by net worth, Figure 1 shows the predictive margins and 95% confidence band for the baseline model across values of net worth between –$20,000 and $400,000 in increments of $20,000.¹¹ For context, the overall annualized hazard of divorce, generated from a hazard model with no covariates, is 2.6% (see the online supplement).¹² The average predicted probability of divorce in the next year is 5.1% when net worth is set to $0 but falls by approximately 25%, to 3.8%, when net worth is just $20,000 higher. Divorce risk continues to fall across moderate levels of net worth but is approximately constant at 1.8% when wealth is at least $300,000. These patterns highlight the substantial differences in marital stability between couples with low net worth compared with at least moderate wealth and smaller differences across the top of the wealth distribution.

Figure 2 spotlights these differences, showing the differences in predictive margins between $0 net worth—when the average predicted probability of divorce is highest—and $20,000 increments of net worth from $20,000 to $400,000 for both the baseline model and the full model. These differences are similar to average marginal effects (“effect” does not imply causality here) of net worth on divorce, except that the “margin” is a change in net worth from $0 to a specified alternative value rather than an instantaneous rate of change. As shown in Figure 1, the predictive margin for $20,000 in net debt has a very wide confidence interval. We therefore do not show it in Figure 2.

In the full model, compared with having $0 net worth, the average predicted probability of divorce in the next year is 0.4 percentage points lower when net worth is $20,000, 0.7 percentage points lower when it is $40,000, and 1.4 percentage points lower when it is $400,000. These disparities are 56% to 66% smaller than in the baseline model. Although control variables substantially reduce the wealth–divorce association, net worth remains meaningfully negatively associated with divorce risk net of controls, supporting Hypothesis 1.¹³

The nonlinearity in the wealth–divorce association also remains net of controls, supporting Hypothesis 2. All else constant, the average decline in the predicted probability of divorce in the next year resulting from increasing net worth from $0 to $40,000 (0.7 percentage points) is the same as the average decline resulting from increasing net worth from $40,000 to $400,000.

To contextualize the magnitude of wealth's association with divorce, Table 3 shows the average marginal effects (for categorical variables) and average change in predictive margins across illustrative values (for quantitative variables) for all other predictors in the full model, analogous to the results for net worth in Figure 2. The average decline in the predicted probability of divorce in the next year associated with having $400,000 in net worth rather than $0 (1.4 percentage points, from Figure 2) is similar to that of getting married after age 24 rather than before age 21 (1.5 percentage points) or of being in a first rather than a third marriage (1.4 percentage points). The average decrease in the predicted probability of divorce associated with having $40,000 in net worth rather than $0 (0.7 percentage points, from Figure 2) is similar to that of being in a first rather than second marriage (0.5 percentage points) or of having no nonmarital births rather than at least one (0.7 percentage points).

Although our focus is on net worth, family income also shows a negative association with divorce concentrated toward the bottom of the distribution. Having a family income of $60,000 rather than $20,000 is associated with an average decline of 1.1 percentage points in the predicted probability of divorce in the next year, but an income of $100,000 rather than $60,000 is associated with an average additional decline of only 0.4 percentage points.

Portfolio Composition and Assessing the Symbolic and Material Perspectives

Next, we assessed whether asset and debt types foregrounded by the material and symbolic perspectives are associated with divorce risk, holding constant total net worth. Table 4 shows the logit coefficients from these analyses. As shown in the first column, consistent with the symbolic perspective (Hypothesis S3), homeownership and vehicle ownership are negatively and statistically significantly associated with divorce holding net worth and the other asset/debt measures constant. The value of the home or vehicle is not significantly associated with divorce risk.¹⁴

The results do not support Hypothesis M3: neither ownership of unsecured debt nor its amount is statistically significantly associated with divorce risk, nor are the terms jointly statistically significant.

We repeated the previous model but replaced home and vehicle values with equity (value less debts). If homes and vehicles reduce divorce risk purely through their symbolic value, we would not expect equity to predict divorce, conditional on total wealth. By contrast, according to the material perspective, equity may reduce financial stress because it can be borrowed against, or higher equity may indicate less burdensome debt payments. The results show that vehicle equity, in addition to home and vehicle ownership, is negatively and statistically significantly associated with divorce risk. This finding raises the possibility that vehicles have material as well as symbolic value in decreasing divorce risk.

To contextualize the magnitudes of the associations in Table 4, we generated predicted probabilities of divorce in the next year for an illustrative scenario: $50,000 in net worth and varied portfolio composition. Both owners and nonowners of homes and unsecured debt are common at this level of net worth, making a range of portfolio compositions realistic. We compared the average predicted probability of divorce under three scenarios: when a couple does not own their home, when they own a home worth $75,000, and when they own a home worth $150,000. In each case, we set couples' net worth to $50,000 and otherwise used couples' observed covariates. This analysis allows us to compare the change in divorce risk associated with two equal-sized changes in home value, only one of which includes a change in ownership. We conducted analogous analyses for vehicle value (at $0; $10,000; and $20,000) and unsecured debt (at $0; $5,000; and $10,000). Because homes, vehicles, and unsecured debt have different ranges of plausible values, we considered a different range of values for each. Thus, the coefficients should be compared only within asset/debt types, not between them.

Panel A of Table 5 shows the results based on the first column of Table 4, using home and vehicle values. The decline in the average predicted probability of divorce associated with owning a home worth $75,000 rather than not owning (0.7 percentage points) is statistically significantly different from the decline associated with owning a home worth $150,000 rather than $75,000 (less than 0.1 percentage points). The same pattern holds for vehicles: the decline in the average predicted probability of divorce associated with owning a vehicle worth $10,000 compared with not owning a vehicle (0.7 percentage points) is statistically significantly different from the decline associated with owning a vehicle worth $20,000 rather than $10,000 (less than 0.1 percentage points). By contrast, the increase in the average predicted probability of divorce associated with having $5,000 in unsecured debt versus $0 is not statistically significantly different from the increase associated with having $10,000 versus $5,000, nor is either of the individual effects statistically significant.

Panel B of Table 5 shows the results based on the second column of Table 4, using home and vehicle equity. Because home equity values are often much lower than home values, here we estimated the change in the predicted probability of divorce associated with owning a home with $20,000 in equity versus not owning a home and with owning a home with $40,000 in equity versus $20,000 in equity. The results for homes and unsecured debt are similar to those in panel A. For vehicles, having $20,000 in equity rather than $10,000 is associated with a statistically significant decline in the predicted probability of divorce, just as vehicle equity was a significant predictor of divorce in the logit model in Table 4. However, ownership remains more important: the decline in the average predicted probability of divorce associated with holding $10,000 in vehicle equity compared with not owning a vehicle (0.8 percentage points) is statistically significantly larger than the decline associated with holding $20,000 rather than $10,000 in vehicle equity (0.1 percentage points).

These results confirm that for homes and vehicles, asset ownership is associated with larger declines in divorce risk than equivalent changes in asset value or equity conditional on ownership, consistent with Hypothesis S3. The results in Tables 4 and 5 do not provide evidence for Hypothesis M3 that unsecured debt increases divorce risk, holding net worth constant. Of course, this does not mean unsecured debt is unimportant; it merely means that we cannot rule out the possibility that its association with divorce risk is well captured by its contribution to overall net worth.¹⁵

Conclusion

Divorce risk varies substantially by wealth, especially across modest positive values of net worth. When we adjusted only for marital duration, the annualized average predicted probability of divorce fell from 5.1% at $0 net worth to 3.8% at $20,000 and to 1.8% at $300,000 or above. Descriptively, these patterns illuminate how marital stability is patterned by social class.

We sought to distinguish among three alternative explanations for the association between wealth and marital stability: (1) material benefits of wealth that increase marital stability, including by reducing financial stress; (2) symbolic benefits of wealth that stabilize marriage through the signal it provides of status and success, including that the couple exceeds a minimum economic bar considered appropriate for married couples; and (3) confounding factors that lead to a spurious association between wealth and divorce risk.

Drawing on the material and symbolic perspectives on wealth's benefits for marital stability, we articulated four empirical predictions. First, both perspectives suggest a negative association between wealth and divorce risk, net of controls (Hypothesis 1). After controlling for a richer set of variables than Eads and Tach (2016), we confirmed their finding that, all else being equal, higher net worth is associated with greater marital stability. Net of controls, we found that having $40,000 in net worth rather than $0 is associated with, on average, a decrease of 0.7 percentage points in the annualized risk of divorce—comparable in magnitude to the decreases associated with being in a first rather than second marriage or having no prior nonmarital births versus at least one. Thus, as a correlate of divorce, net worth is on par with other correlates that have received far more attention.

Second, versions of the material and symbolic perspectives suggest that wealth should have diminishing marginal returns for marital stability. At the bottom of the wealth distribution, small increases in net worth may be valuable for reducing financial stress (a material benefit) or clearing a threshold for the minimum expected level of affluence for married couples (a symbolic benefit) (Hypothesis 2). Consistent with this prediction, we found that all else being equal, the average annualized predicted probability of divorce fell about as much when net worth was $40,000 rather than $0 as when it was $400,000 rather than $40,000.

Of course, we could not entirely rule out selection. Future research is needed to estimate the causal effect of wealth on divorce. This sort of examination might include, for example, exploiting exogenous wealth changes due to lottery winnings (e.g., for wealth's effect on labor supply, see Cesarini et al. 2017) or housing booms (e.g., for housing wealth's effect on offspring college outcomes, see Lovenheim and Reynolds 2013). Further, our analyses provide only a partial portrait of how wealth may affect marital stability because we controlled for characteristics that may be endogenous to prior wealth, such as income and fertility. We encourage future research that takes a broad array of approaches to collectively provide greater evidence on the effect of wealth on divorce.

Our final set of analyses sought to distinguish between the material and symbolic perspectives on wealth's benefits for marital stability. Here, we relied on the perspectives' different predictions about which asset or debt types are likely to be associated with divorce, above and beyond total net worth, and whether ownership or the dollar value is expected to be more consequential. We improved on prior research by isolating the predictive power of each asset or debt type net of overall wealth and not conflating ownership and value of the focal asset or debt. Consistent with the symbolic perspective, we found ownership of both homes and vehicles to be negatively and statistically significantly associated with divorce risk, holding constant total net worth and all other controls (Hypothesis S3). Using an illustrative case, we further showed that the same increase in asset value or equity is associated with greater declines in average divorce risk when it is paired with moving to ownership rather than an increase in value or equity conditional on ownership. This finding suggests that ownership of homes and vehicles is associated with greater marital stability in ways explained by neither their consequences for total net worth nor their implications for home or vehicle value or equity. Conditional on total net worth, we found that neither ownership of unsecured debt nor its amount is significantly associated with divorce risk, contrary to our prediction based on the material perspective (Hypothesis M3).

However, we urge caution in interpreting these results. The material and symbolic perspectives are broad, so versions of each may be consistent with various empirical patterns. Empirical estimation is challenging given the collinearity of different asset and debt types and limitations in which portfolio components can be disaggregated in the NLSY79. All else being equal, unsecured debt may be associated with divorce risk in the population, although the association is not statistically significant in our sample. Even if no such population-level association exists, this does not rule out other manifestations of the material perspective. Our analyses are exploratory rather than a definitive adjudication among spurious associations and the material and symbolic perspectives on wealth's association with marital stability.

Matters are even more complicated when we consider other possible mechanisms linking wealth and divorce risk. Wealth may affect divorce risk by shaping the expected costs and benefits of divorce as spouses divide assets (Dew 2009). Prior research has considered how divorce laws, including unilateral divorce and the division of marital property, shape in-marriage behavior and divorce risk (Clark 1999; Stevenson 2007; Voena 2015; Zang 2020). This line of research could be expanded to analyze how these laws moderate the association between wealth and divorce. We encourage future research that develops additional empirical predictions to distinguish among theoretical perspectives and future qualitative research that explores how couples understand the role of financial circumstances in their marital satisfaction and marital stability.

In an era of high and rising wealth inequality (Pfeffer and Schoeni 2016), we found that wealthier couples experience lower divorce risk, net of a host of controls. We further found that reductions in divorce risk are concentrated over modest positive values of net worth and are much less pronounced above the median of the wealth distribution. Last, ownership of visible, symbolic assets—homes and vehicles—is negatively associated with divorce risk, even conditional on total net worth. Our findings support calls to consider wealth as a distinct indicator of economic stratification, and we encourage future research to recognize wealth as a substantial correlate of divorce.

Acknowledgments

We are grateful for the helpful comments we received from three anonymous Demography reviewers. We also benefited from helpful comments at conference and seminar presentations. Killewald is grateful for support received from the Harvard Radcliffe Institute Fellowship Program. The NLSY79 survey is sponsored and directed by the U.S. Bureau of Labor Statistics and managed by the Center for Human Resource Research at The Ohio State University. Interviews are conducted by the National Opinion Research Center at the University of Chicago.

Notes

References