Health Insurance and Risk of Divorce: Does Having Your Own Insurance Matter?

Abstract

Most American adults under 65 obtain health insurance through their employers or their spouses’ employers. The absence of a universal healthcare system in the United States puts Americans at considerable risk for losing their coverage when transitioning out of jobs or marriages. Scholars have found evidence of reduced job mobility among individuals who are dependent on their employers for healthcare coverage. This paper finds similar relationships between insurance and divorce. I apply the hazard model to married individuals in the longitudinal Survey of Income Program Participation (N=17,388) and find lower divorce rates among people who are insured through their partners’ plans without alternative sources of their own. Furthermore, I find gender differences in the relationship between healthcare coverage and divorce rates: insurance dependent women have lower rates of divorce than men in similar situations. These findings draw attention to the importance of considering family processes when debating and evaluating health policies.

Love and commitment are often what couples believe secure and protect marriages from divorce. Sociologists of the family however, find that practical considerations are probably more important (Kalmijn, 1998). Married couples with high incomes are more likely to stay married (Amato, 2010; Gibson-Davis, Edin, & McLanahan, 2005). Educational attainment—an indicator of earnings potential—is also associated with greater marital stability (Amato, 2010). Education and income are consistently stronger predictors of divorce than sentiment-driven indicators. Recognizing the significance of these factors, researchers are careful to take household income, the couple’s educational attainment, wealth, and other resources into consideration as they study divorce patterns. Researchers have yet to study insurance coverage as a factor that can influence divorce. This paper examines the relationship between health insurance and divorce by asking three main research questions. (1) Is there an association between being insured by a spouse and divorce? (2) Does this association get stronger when one partner does not have an independent source of health insurance? (3) Do these associations differ by gender?

The United States is among the few and perhaps the only developed country that does not provide universal healthcare to its residents (Jost, 2003). While seniors over 64 years of age are assured coverage under Medicare, the majority of non-elderly adult men and women are left responsible for securing their own health insurance. Employment and marriage are the top two sources of health insurance for American adults; 24% of non-elderly adult women and 14% of men are covered as a dependent (Kaiser Family Foundation, 2011). In comparison, 55.3% of the adult population gains health insurance though employment (US Census, 2012).

Insurance is consequential. Being uninsured is associated with lower healthcare utilization, increased morbidity, and higher mortality (Institute of Medicine, 2004). Not having insurance creates barriers to adequate access to healthcare (Institute of Medicine, 2002). Uninsured individuals are more likely to be diagnosed with late-stage cancer—a disease that is often detected at early states during routine doctor visits (Halpern, War, Pavluck, Schrag, Bian, & Chen, 2008)—and are less likely to have an ongoing relationship with a health care provider (Holahan & Spillman, 2002). In addition, medical expenses contribute to a large portion of personal bankruptcies in the United States (Himmelstein, Thorne, Warren, & Woolhandler, 2009). Considering these known risks of being uninsured, it is not surprising that Americans treat health insurance as a valuable commodity.

Even brief periods of uninsurance can be costly. As American adults transition into and out of jobs and marriages, they risk losing health insurance coverage (Lavelle & Smock, 2012; Meyer & Pavalko, 1996). No social infrastructure guarantees continued health insurance through these transitions. The Consolidated Omnibus Budget Reconciliation Act (COBRA) allows people to purchase insurance at the lower group rate under certain conditions, but is only available for limited periods and its costs can be prohibitive, especially for recently unemployed or divorced individuals. Many cannot help but experience a gap in insurance coverage as they change jobs or go through divorce (Lavelle & Smock, 2012; Swartz & McBride, 1990). These gaps are significant. They can increase premiums or limit payouts even when individuals gain coverage, particularly for those who have on-going health care needs (van de Ven, van Vliet, Schut, & van Barneveld, 2000). Gaps can also have significant financial consequences for those who fall ill while uninsured.

Researchers who study employment and wages have established that the health insurance benefits are an important part of an employee’s compensation; a job that provides health insurance yields substantially higher total compensation than a job with the same salary but no health benefits (Woodbury, 1983). Studies also show that insurance can act as a significant motivator for people to seek and retain employment, at times even deterring workers from pursuing otherwise better opportunities (Cooper & Monheit, 1993; Madrian, 1994; Monheit & Cooper, 1994). If health insurance plays such a large role in the labor market, we can expect it also play a role in the marriage market. After all, approximately 36 million American adults under the age of 65 rely on a family member to provide their health insurance (KFF, 2011). This paper finds that married individuals who are dependent on their spouses for health insurance have lower rates of divorce. Having private health insurance is highly correlated with socio-economic characteristics such as high education, stable employment, and high income. These factors all lower the risk of divorce, but the analyses show health insurance does not follow the same pattern. This paper shows that divorce risk is lowest among people who do not have the option of employment-based coverage and are insured under their spouses’ plans. I use the longitudinal 2004 Panel of Survey of Income Program Participation (SIPP). The 2004 SIPP tracks a nationally representative sample of over 43,500 households over time with great granularity recording each member’s marital and healthcare status every month from October 2003 through December 2007. I employ the Cox proportional hazard model to estimate the relative risk of divorce between insurance-dependent and non-dependent individuals.

BACKGROUND

In his economic models of marriage, Gary Becker (1974) posits that positive gains to marriage and negative consequences of divorce motivate two individuals to stay married. His theory argues that couples with greater economic resources have lower risks of marital dissolution as these resources will increase the gains that individuals would derive from their unions. Higher levels of income can protect married couples from financial stressors (Sawhill, 1975). Economic assets, such as homeownership, can also have stabilizing effects on marriage (Becker, Landes, & Michael, 1977; Levinger, 1979; South & Spitze, 1986). By the same token, couples with higher levels of educational attainment are less likely to divorce (Martin, 2006).

The decision to divorce is not always unilateral. Becker and his coauthors (1977) explain that couples will remain married if the combined gains of staying married exceed the combined benefits of separating. If person A wants to divorce while their partner B does not, the latter can ‘compensate’ the former to make it worthwhile for both parties. If the combined net benefit of the union is not enough to satisfy both partners, however, the marriage will result in dissolution.

This study examines health insurance as a potential gain to a marriage. Because health insurance pertains to the household, it is perhaps a prominent feature of the combined gains of marriage. A family health insurance plan that covers both spouses can be considered a shared good that benefits from economies of scale. It is less costly than two people being insured independently. A person who is not covered by their spouse must either purchase an individual plan at a higher price or gain access to a group plan, usually through employment. In this way, securing private health insurance for a married couple is cheaper than two individuals obtaining their own and, therefore, it can be considered an economic ‘gain’ of marriage.

While health insurance has not been explicitly studied as an economic gain of marriage, it has been examined in the context of employment and job mobility (Cooper & Monheit, 1993; Madrian, 1994; Monheit & Cooper, 1994). Studies find that jobs that offer health insurance plans have lower turnover rates (Cooper & Monheit, 1993; Madrian, 1994; Monheit & Cooper, 1994). Likewise, being married to someone who can provide a health insurance policy may increase the gains of the marriage, leading to lower rates of divorce. My first hypothesis is:

• Hypothesis 1: Married individuals who are insured through their spouses’ health plans have lower rates of divorce

The second hypothesis examines the risk of losing health insurance coverage as a potential negative consequence of divorce. Does the association between insurance coverage and divorce get stronger when one partner does not have an independent source of health insurance? Researchers of employer-specific insurance plans and job-mobility find that people who are expected to be worse off if they lose current coverage—for example, employees with greater medical expenditures and those without spouses who can provide coverage during transitions—have lower rates of job-mobility (Buchmueller & Valletta, 1996; Cooper & Monheit, 1993; Gruber & Madrian, 2002; Madrian, 1994).

The economic consequence of divorce can also be influenced by whether those in a marriage have access to group coverage independently. The cost of divorce for people who can switch over to their employers’ plans on divorce will not be as high as those whose only sources of insurance coverage is from their spouses. Not having a comparable source of health insurance outside the marriage increases the negative consequences of divorce that may lead to lower divorce rates. I use employer-sponsored health insurance as a comparable alternative to spouse-provided health insurance plans. There are alternatives, but they are not as relevant. Public means-tested cash transfers, for example, do not appear to lower the cost of divorce. (Hoffman & Duncan, 1995). Likewise, needs-based public health insurance, such as Medicaid, would not be a comparable substitute to a private, family insurance plan. Needs-based public insurance is often available only at very low levels of income and is sometimes considered to be of lower quality than private insurance plans (Quesnel-Vallee, 2004).

• Hypothesis 2: Not having an employment-based source of health insurance coverage outside the marriage further lowers the risk of divorce for people enrolled in their spouses’ plans.

Gender, Health Insurance, and Marriage

The story of divorce and health insurance is further complicated by the issue of gender. Historically, the division of labor in a US household has fallen along gendered lines (Cherlin, 1995; Greenstein, 1995; Kalmijn, Loeve, & Manting, 2007; Nock, 1995; Presser, 2000; Sayer, England, Allison, & Kangas, 2011). While spousal roles and duties have become more flexible and negotiable over time, the traditional male breadwinner model of the family persists (Greenstein, 1995). Women still do most housework and childrearing (Cherlin, 1995; Nock, 1995) and marriages are more stable when the husband earns more than his wife (Kalmijn et al., 2007). The responsibility of financial contribution still primarily falls onto the man. A husband’s unemployment is strongly related to divorce whereas a wife’s is not (Sayer et al., 2011). By the same token, Teachman (2010) found that health-related work limitations among men but not among women were associated with higher rates of marital disruption. The gains of a marriage increase when the gendered expectations of their husbands resuming the role of the primary breadwinner are met. For wives, spouses’ incomes have a positive relationship with their self-reported levels of marital commitment (Nock, 1995).

The same relationship applies to health insurance as well. If traditional gender roles put pressure on husbands to contribute financially via economic activity in the labor market, it is also likely that men are expected to provide the family with health insurance. Thus, I expect to see gender differences in the divorce rates associated with insurance dependence in my analysis. My third hypothesis is:

• Hypothesis 3: Women who are insured through their spouses have lower rates of divorce than men who are insured through their spouses

The fourth hypothesis examines whether the association between divorce and having an alternative source of insurance is stronger for women than it is for men. With men taking on the primary responsibility of the household income, it is not surprising the economic consequences of divorce are less favorable for women than men (Burkhauser, Duncan, Hauser, & Berntsen, 1991). Burkhauser and Duncan (1989) found that divorce or separation was the single most financially detrimental event that could happen to non-elderly women. Divorce’s consequences for health insurance are equally striking (Bernstein, Cohen, Brett, & Bush, 2008; Lavelle & Smock, 2012). Lavelle and Smock (2012) find that, divorce leads to an eight percentage-point decline in women’s private health insurance coverage, net of changes in employment, economic resources, and other factors. This decline was even sharper for women insured as a dependent. A corollary of these findings is that wives’ economic independence through labor force participation may lower barriers against divorce (Duncan & Hoffman, 1985; Lavelle & Smock, 2012). Duncan and Hoffman (1985) found that women’s human capital investments have some modest effects on mitigating the decline in their economic situation following a divorce. Similarly, Lavelle and Smock (2012) also found that just-divorced women who were insured through their own plans were largely protected from the risk of losing private coverage.

• Hypothesis 4: Not having an alternative source of health insurance outside the marriage lowers divorce risk for women more so than for men

METHOD

Data

I use the 2004 panel of the Survey of Income and Program Participation (SIPP) to examine the relationship between health insurance status and divorce. The SIPP is a nationally representative series of longitudinal panels whose survey duration ranges from 2.5 to 4 years. The first SIPP panel was sampled in the early 1980s and a new panel was re-sampled from the non-institutionalized population in the U.S. every one to four years. The 2004 SIPP panel is a longitudinal dataset that follows its respondents for about four years from October 2003 until December 2007. The strength of this dataset lies in its large size, its wide range of household insurance and demographic variables, and its longitudinal structure. The SIPP survey is divided into core questionnaires and topical modules. The core questionnaire collects data for the same variables every month throughout the study period. The SIPP administers a topical module containing a different theme (i.e. marital history) once every four months. The SIPP then creates longitudinal panels checking for data inconsistencies and missing values. The Census Bureau accounts for missing values by logically deriving from other available information when possible. They rely on several imputation techniques to fill in the remaining missing data. Thus, if a respondent completed a questionnaire or a module, all variables had valid values. Most of the variables that I use in my analyses are from the core questionnaires. Since these variables are recorded every month, I can get relatively close estimates of when changes occurred. Life events such as divorce and changes to insurance rarely happen multiple times within a single month.

The 2004 SIPP collected data for 45,540 individuals. The hypotheses in this paper pertain to the risk of divorce. Therefore, I only include married individuals in the analytic sample. I dropped 14,239 individuals (33% of total initial 2004 SIPP dataset) who were not married during the study period. Furthermore, I only include married individuals whose marriage duration is known. The sample includes people who were already married when the SIPP started to collect data in October 2003 and people who became married during the study period before it ended in December 2007. For those who were already married in October 2003, I derive their marriage durations from their marital history topical module that the SIPP administered in May 2004. I exclude 1,354 (3% of initial 2004 SIPP dataset) married respondents who did not complete the marital history topical module from the sample. Lastly, I limit my analysis to non-elderly adults. Those over 65 are almost universally insured through Medicare and divorce may affect their insurance status differently than the rest of the adult population. I exclude 10,559 persons (24% of initial data) over the age of 64 or under the age of 18 from the sample.

My analysis sample consists of 17,388 individuals who were married at some point during the study period between October 2003 and December 2007. They collectively experienced over 500 divorces. I weight the observations using sampling weights to compensate for SIPP’s different selection probabilities into the 2004 panel across subpopulations.

Variables

Marital Status

The 17,388 individuals in my sample, weighted to represent approximately 46 million US residents, were married at some point during the study period. A change in a respondent’s marital status reflects a marriage or divorce/separation sometime during the previous 30 days. Only the respondents who are married are at risk of divorce or separation and I limit hazard calculations to the currently married at any point in time. The incremental hazards of separation in response to all covariates are similar to the hazards of divorce. Sensitivity tests result in similar hazard patterns for all three events: divorce only, separation only, and divorce or separation. The analyses in this paper examine the hazard of either divorce or separation.

Health Insurance Status

I combine two SIPP variables asking health insurance type and coverage source to create a nominal scale to differentiate individuals who are covered by their own plan from those covered under someone else's, government sponsored need based insurance (Medicare or Medicaid), or none at all. The SIPP specifically asks if the primary subscriber is the respondent’s spouse at only one point in time during the survey period. More than 95% of married people on June 2005 who are insured on someone else’s plan are insured on their spouses’ plan. I assume that this extremely high rate of spousal coverage among married individuals who are not primary subscribers is approximately constant throughout the study period. I lag insurance status by one month to associate the insurance status prior to the divorce, as this marital event is often accompanied by a simultaneous change in health care status. I have conducted sensitivity test using a two, three, and six month lag time between insurance status and marital disruption. There is little difference in the coefficients and the hypotheses tests up to a lag time of three months. At six months, the coefficients in the hazard models become smaller in size, but even with a lag time that is one-eighth of the total survey period the tests of hypotheses yield the same results.

Alternative Source of Health Insurance

In my analysis, I consider individuals who are unemployed, contingent workers, or employed part-time at small companies with less than 100 employees to be without access to employer-sponsored health insurance plans outside the marriage. Full-time employment status and the size of the firm are two indicators that predict whether an employer offers health insurance to an employee (Kaiser Family Foundation, 2013). Throughout the study period, 97 to 99 percent of firms with over 200 workers offered health insurance to their employees, whereas only 59 to 65 percent of smaller firms offered insurance plans (Kaiser Family Foundation, 2013). Very few firms offered insurance plans to temporary workers (three to six percent). Less than a quarter of small firms offered plans to their part-time staff but almost half (47%) of large firms offered insurance to workers who were not full time employees (Kaiser Family Foundation, 2013). Lavelle and Smock (2012)’s study also find that full time workers were largely protected from insurance loss after divorce.

I confirm from the 2004 SIPP that its respondents’ employment statuses and their employers’ firm sizes are indeed good predictors of whether the respondents have access to health benefits. I test how closely these variables capture respondents’ access to employment-based health insurance by examining single, never-married individuals. Almost 80% of single, never-married respondents who I assigned as having “access to employer-sponsored health insurance” are insured as the primary subscriber. About half of the never-married people who I categorized as having access but are in fact uninsured, voluntarily elected not to enroll in their employers’ plans. The SIPP asked specific questions on whether a respondent’s employer offers insurance plans and his or her reasons for not enrolling only once in June 2005. Therefore, I could not use this variable to create a better measurement of the availability of employment-based insurance throughout the entire period.

Control Variables

I incorporate demographic and socioeconomic variables that prior research identified as determinants or predictors of divorce—education, race and ethnic origin, number of children, age, marital history, and household income (Amato, 2010; Casper & Bianchi, 2001; Cherlin, 1992). My models include gender, race/ethnicity, educational attainment, the number of children, and a polynomial term for age as controlling covariates. Many households in the SIPP report income that fluctuates significantly from month to month. Some households report their entire annual income in one month leaving the monthly income for the rest of the year at zero. To smooth the income flow, I use the average of monthly total family income to measure the family’s level of financial standing. I expect higher educational attainment, higher income, and having more children to be associated with lower rates of divorce (Amato, 2010). Marital history is another predictor of divorce or separation (Becker et al., 1977; Lehrer, 1988; Presser, 2010). I include an indicator for whether or not the current marriage is a first marriage. Higher order marriages may be more likely to end in divorce or separation (Becker et al., 1977; Lehrer, 1988). I also include a polynomial term for the length of marriage. Probability of dissolution decreases with the duration of marriage (Becker et al.,1977; Presser, 2000). Couples who would end up divorced tend to end their marriage earlier than later (Becker et al., 1977) and people make more marriage-specific investments (e.g. children, sexual compatibility) the longer they are together (Lehrer, 1988). All covariates with the exception of gender and race are time-varying.

Analytic Strategy

I use Cox's proportional hazard model (Cox, 1972) to measure the effect of health insurance on marriage. This model estimates the incremental risk of an event happening to one group relative to a reference group. A person’s hazard is a multiplicative replica of the baseline hazard based on his or her set of covariates. In this way, the model can identify characteristics that are associated with greater or lower hazards of events such as divorce (Bumpass, 1990). I use STATA13’s stcox package to estimate all hazard coefficients. I test the model’s assumption that divorce hazards of the comparison groups are proportional over time. I interact the key variables of the analysis (insurance status and access to employer-sponsored health plans) with time (reference month) and test for statistically significant time-varying effects. A joint test of significance revealed no systematic change in the association between insurance status and access to employer-sponsored health plans.

I estimate four hazard models to test each of the four hypotheses in this paper. The first model estimates the divorce hazards associated with insurance status. Model 2 adds access to employment-based insurance to model 1 and interacts it with the respondent’s insurance status. Model 3 interacts gender with insurance status to examine any gender differences in the association between insurance and divorce. The fourth model is a three-way interaction between insurance status, access to an employment-based option, and gender. I present the hazards associated with the interactions in Models 2–4 with a series of dummy variables. I report these hazards in odds ratios. Each odds ratio indicates the relative hazard of divorce or separation of a respondent with a particular characteristic relative to the reference group. All models include covariates for age, race, education, children, higher-order marriage, marriage duration, and logged average monthly income as controls. I then select the relevant coefficients from the hazard models to explicitly test the four hypotheses of this paper. I calculate p-values from one-sided t-tests adjusted for false discovery rate (Benjamini & Hochberg, 1995). These p-values are more conservative than unadjusted p-values as they take into account that the probability of falsely rejecting a null condition increases with the number of tests performed. I present the four Cox hazard models and their hypothesis tests in the main results section of this paper.

RESULTS

Descriptive Results

I estimate divorce hazards from the 17,388 individuals in the 2004 SIPP who were married at some point between October 2003 and December 2007. Table 1 shows the basic descriptive statistics of my analysis sample. The percentages and averages presented in Table 1 are weighted to represent the adult US married population under age 65. About half the sample was insured under their own names and a third were insured as dependents. Less than 20% of the analysis sample were insured by need-based government plans or were uninsured. About a third received four-year college degrees and about another third did not receive any education beyond high school. Roughly half the respondents were women.

Table 1.

Descriptive Statistics for Study Measures (N = 17,388)

Population at Risk	M or %	SD
Insurance Status
Insured under own name	50.84
Insured under someone else's plan	31.64
Gov't Insurance (Medicare, Medicaid)	5.64
Uninsured	11.88
Race/ethnicity
Non-Hispanic White	73.55
African American	7.66
Hispanic	12.58
Asian	3.84
Other	2.47
Educational Attainment
Less than High School	9.43
High School Diploma or Equiv.	21.06
Associate degree or some college	37.47
Bachelors' degree	20.95
Advanced degree	11.09
Children
Not living with Children	36.85
One child	23.09
Two children	24.90
Three children	15.16
Age	41.85	10.61
Family Monthly Income	6,712.19	5,116.87
Gender: 0 = male, 1 = female	0.53

Note: Population at risk at first reference month (November 2003 if already married or first month of marriage). Values weighted to represent the US population.

Being insured under one’s own name was correlated with other measures of socioeconomic standing. Over 85% of people with their own health insurance were employed full-time and contributed over half of their households’ total incomes on average. They were also more likely to have attended college. I show in Table 2 that more married women than men were insured as a dependent. In concordance with the findings by the Kaiser Family Foundation (2011) women in my sample were more likely than men to be insured on another’s plan. About 44% of married women, compared to 16% of married men, were insured as dependents. The educational attainment and income of these men were also not too different from the married men who were insured under their own names. Men who were enrolled as a dependent still have higher earnings and were more likely to have attended college than their female counterparts. While the insurance-dependent men earned less on average, the proportion with some post-secondary education was slightly higher.

Table 2.

Descriptive Statistics of Risk Population by Insurance Status and Gender (N = 17,388)

Insurance Status	Sample Size (%)		Mean Age		Avg. Monthly Earnings (USD)		Proportion with some college education (%)
	Men (n=8,091)	Women (n=9,297)	Men (n=8,091)	Women (n=9,297)	Men (n=8,091)	Women (n=9,297)	Men (n=8,091)	Women (n=9,297)
Insured under own name	66.8	36.8	42.9	41.9a	4,838	2,902a	75.9	76.6
Insured under someone else's plan	17.0	44.6	43.9b	41.8a	3,397b	1,393a,b	77.7	72.5a,b
Gov't Insurance (Medicare, Medicaid)	4.4	6.8	45.0b	38.2a,b	737b	390a,b	44.3b	40.6b
Uninsured	11.9	11.9	38.9b	37.9b	1,873b	742a,b	42.4b	42.9b

Note: Population at risk at first reference month (November 2003 if already married or first month of marriage). Values are weighted to represent the US population.

^aDenotes statistical difference between men and women at significance level, 0.05.

^bDenotes difference from being self-insured at significance level 0.05.

While individuals who were insured under their own plans contributed proportionally more to total household income, those who were insured through their spouses also made economic contributions to the household. The sizes of these contributions however, differed by gender. Men who were enrolled on another’s insurance plan contributed 41% of the total household income, on average. Women who were enrolled on another’s insurance plan, on the other hand, contributed on average, less than 20%. Table 3 shows the differences in economic contribution to the household by insurance status and gender. These percentages represent the average monthly earnings of an individual as a proportion of the monthly total household income during the marriage. Total household income includes income from means-tested cash transfers and income from property.

Table 3.

Earnings Contribution to Total Household Income by Insurance and Gender in Percentages

Insurance Status	Men	Women	Overall
Insured under own name	63.11	41.91	54.96
Insured under someone else's plan	41.96	18.64	24.52
Gov't Insurance (Medicare, Medicaid)	26.39	13.45	18.14
Uninsured	58.97	22.47	39.58

Note: Population at risk at first reference month (November 2003 if already married or first month of marriage). N = 17,388 (men n = 8,091; women n = 9,297). Values are weighted to represent the US population. Total household income includes income from property and means-tested cash transfers.

Main Results

The following section tests and reports the results of the four main hypotheses of this paper. I test each hypothesis with a separate Cox proportionate hazard model (Models 1 to 4) in the same order that I presented in the background section.

• Hypothesis 1: Married individuals who are insured through their spouses’ health plans have lower rates of divorce

At any point in time, married individuals who were insured on their spouses’ plans were indeed significantly less likely to divorce or separate than those who were covered under their own policies. Being dependent on one’s spouse for health insurance lowered the divorce hazard by almost 70%. The odds of divorce associated with being insured by a spouse (0.321) were statistically significant at the 0.01 level. Table 4 shows the odds ratios for covariates related to insurance and family income. Logged family monthly income had a significant negative association with divorce hazard consistent with prior findings (Bumpass, Martin, & Sweet, 1991). The insurance coefficients remain significant indicating that their association with divorce could not be entirely explained by the family’s income.

• Hypothesis 2: Not having an employment-based source of health insurance coverage outside the marriage further lowers the risk of divorce for people enrolled in their spouses’ plans

Table 4.

Hypothesis 1: Cox Regression of Divorce Hazard on Insurance Status

Model 1. Cox Regression of Divorce Hazard on Insurance Status (hazards in odds ratios)
Insurance Status
Insured under own name	(reference)	a
Insured under someone else's plan	0.32***	b
Gov't Insurance (Medicare, Medicaid)	0.59***
Uninsured	0.73*
Logged family monthly income	0.44***

Test of Hypothesis 1: Married individuals who are insured through their spouses’ health plans have lower rates of divorce
	Key Coefficient for Hypothesis Test
a	Insured under own name (reference group)	1.00
b	Insured under someone else's plan	0.32***
	Ratio of b to a	0.32***

Note: Model includes age, age-squared, race, education, children, higher-order marriage, and marriage duration as controls. Coefficients are not shown. N=17,388 (men n=8,091; women n=9,297). Values are weighted to represent the US population.

Note: P-values of one-sided t-tests are corrected adjusted for False Discovery Rate (Benjamini and Hochberg 1995).

^*p < .05.

^**p < .01.

^***p < .005.

Model 2 confirms that not having an employment-based source of insurance was associated with further declines in divorce hazards among individuals who were insured by their spouses. This model interacts insurance status with access to employer-sponsored plans as people may have had the option to enroll in their own employers’ health plans but had to forgo them in favor of their spouses’ family policies. Table 5 reports the divorce odds ratio for each insurance status group who had and who did not have an employment-based option. The divorce ratio of someone insured under their spouses’ plans with an option for an employer-sponsored plan was 0.433 relative to persons who were insured under their own names. Individuals insured under their spouses’ plans without access to employer-sponsored plans had divorce hazard ratios of 0.179. Both coefficients were significant at the alpha 0.005 level. I test whether not having access to an employer-sponsored plan significantly lowered the divorce hazards among people who were insured by their spouses in the lower panel of Table 5. Not having an insurance option outside the marriage further lowered divorce odds by 0.412 and this difference is statistically significant at the alpha 0.005 level. Logged family monthly income had a significant negative association with divorce in Model 2. However, similarly to the first model, family income did not completely moderate the relationship between insurance and divorce.

• Hypothesis 3: Women who are insured through their spouses have lower rates of divorce than men who are insured through their spouses

Table 5.

Hypothesis 2: Cox Regression of Divorce Hazard on Insurance Status and Access to Employment-based Option

Model 2. Cox Regression of Divorce Hazard on Insurance Status and Access to Employment-based Option (hazards in odds ratios)
Two-way interaction between insurance status and access to employment-based option
Insurance Status	Access to Employment-based Option
Insured under own name	Yes	(reference)
Insured under own name	No	0.59*
Insured under someone else's plan	Yes	0.43***	a
Insured under someone else's plan	No	0.18***	b
Gov't Insurance (Medicare, Medicaid)	Yes	1.03
Gov't Insurance (Medicare, Medicaid)	No	0.37***
Uninsured	Yes	1.03
Uninsured	No	0.39***
Logged family monthly income		0.41***

Test of Hypothesis 2: Not having an employment-based source of health insurance coverage outside the marriage further lowers the divorce risk of people enrolled in their spouses’ plans.
	Key Coefficients for Hypothesis Test
a	Insured under someone else's plan & has employment-based option	0.43***
b	Insured under someone else's plan & has no employment-based option	0.18***
	Ratio of b to a	0.41***

Note: Model includes age, age-squared, race, education, children, higher-order marriage, and marriage duration as controls. Coefficients are not shown. N=17,388 (men n=8,091; women n=9,297). Values are weighted to represent the US population.

Note: P-values of one-sided t-tests are corrected adjusted for False Discovery Rate (Benjamini and Hochberg 1995).

^*p < .05.

^**p < .01.

^***p < .005.

Model 3 shows that divorce hazards associated with being covered by a spouse’s health insurance did not differ by gender. Men who were insured on their spouses’ plans had divorce risks that were 0.475 that of men who were primary subscribers. Women who were insured by another had hazards of 0.373 of the same reference group (Table 6). The difference in hazard ratio between men and women were not statistically significant. Model 3 reveals another interesting gender difference. A woman who had her own source of health insurance was associated with a significantly higher hazard of divorce than a man who was the primary subscriber. This is consistent with some findings from prior research on the positive relationship between women’s financial independence and likelihood of divorce (South and Spitze, 1986; Greenstein, 1995). While a wife’s earnings can increase gains and stability to a marriage by augmenting the household income (McLanahan, 2004), empirical evidence also shows that wives who earned more than their husbands were more likely to experience divorce (Kalmijin et al., 2007). Because the divorce risks of women who were primary subscribers were so high, the differences in divorce hazards between primary and dependent insurance subscribers were greater among women (1.580 vs 0.373) than among men (1.000 vs 0.475).

• Hypothesis 4: Not having an alternative source of health insurance outside the marriage lowers divorce risk for women more so than for men

Table 6.

Hypothesis 3: Cox Regression of Divorce Hazard on Insurance Status and Gender

Model 3. Cox Regression of Divorce Hazard on Insurance Status and Gender (hazards in odds ratios)
Two-way interaction between insurance status and gender
Insurance Status	Gender
Insured under own name	Male	(reference)
Insured under own name	Female	1.58***
Insured under someone else's plan	Male	0.48**	a
Insured under someone else's plan	Female	0.37***	b
Gov't Insurance (Medicare, Medicaid)	Male	0.41*
Gov't Insurance (Medicare, Medicaid)	Female	0.88
Uninsured	Male	0.91
Uninsured	Female	0.89
Logged family monthly income		0.44***

Test of Hypothesis 3: Women who are insured on their spouse’s health plans have lower rates of divorce than men who are insured by their spouse
	Key Coefficients for Hypothesis Test
a	Insured under someone else's plan & male	0.48**
b	Insured under someone else's plan & female	0.37***
	Ratio of b to a	0.78

Note: Model includes age, age-squared, race, education, children, higher-order marriage, and marriage duration as controls; coefficients are not shown. N=17,388 (men n=8,091; women n=9,297). Values are weighted to represent the US population.

Note: P-values of one-sided t-tests are corrected adjusted for False Discovery Rate (Benjamini and Hochberg 1995).

^*p < .05.

^**p < .01.

^***p < .005.

The divorce hazards of spouse-insured women who did not have an employment-based source were significantly lower than their male counterparts. The divorce hazards of men who were insured by their spouses and did not have employment-based options were 0.489. The hazards for similar women were 0.179 (Table 7). The difference in divorce odds between these two groups was statistically significant at the alpha 0.05 level. The reference group for Model 4 consists of male primary subscribers who were most likely to have employer-sponsored insurance.

Table 7.

Hypothesis 4: Cox Regression of Divorce Hazard on Insurance Status, Access to Employment-based Option, and Gender

Model 4. Cox Regression of Divorce Hazard on Insurance Status, Access to Employment-based Option, and Gender (hazards in odds ratios)
Three-way interaction between insurance status, access to employment-based option, and gender
Insurance Status	Access to Employment-based Option	Gender
Insured under own name	Yes	Male	(reference)
	Yes	Female	1.70***
	No	Male	0.75
	No	Female	0.75
Insured under someone else's plan	Yes	Male	0.44*	a
	Yes	Female	0.61*	b
	No	Male	0.49	c
	No	Female	0.19***	d
Gov't Insurance (Medicare, Medicaid)	Yes	Male	0.22
	Yes	Female	2.03*
	No	Male	0.41*
	No	Female	0.49***
Uninsured	Yes	Male	1.13
	Yes	Female	1.49
	No	Male	0.52*
	No	Female	0.47**
Logged family monthly income			0.40***

Test of Hypothesis 4: Not having an alternative source of health insurance outside the marriage lowers divorce risk for women more so than for men
	Key Coefficients for Hypothesis Test
c	Insured under someone else's plan & has no employment-based option & male	0.49
d	Insured under someone else's plan & has no employment-based option & female	0.18***
	Ratio of d to c	0.37*
Not having an alternative source of health insurance outside the marriage lowers divorce risk for men
a	Insured under someone else's plan & has employment-based option & male	0.44*
c	Insured under someone else's plan & has no employment-based option & male	0.49
	Ratio of c to a	1.12
Not having an alternative source of health insurance outside the marriage lowers divorce risk for women
b	Insured under someone else's plan & has employment-based option & female	0.61*
d	Insured under someone else's plan & has no employment-based option & female	0.18***
	Ratio of d to b	0.30***

Note: Model includes age, age-squared, race, education, children, higher-order marriage, and marriage duration as controls. Coefficients are not shown. N=17,388 (men n=8,091; women n=9,297). Values are weighted to represent the US population.

Note: P-values of one-sided t-tests are corrected adjusted for False Discovery Rate (Benjamini and Hochberg 1995).

^*p < .05.

^**p < .01.

^***p < .005.

Having an outside option for insurance was not associated with higher divorce hazards among men when they were already insured by their wives (Table 7). The story is different for women. The divorce hazards of women who were insured by their spouses but also had access to employment-based plans outside their marriages were significantly higher than spouse-insured women who did not have alternative sources through employment. The odds ratio of 0.412 associated with having access to employer-sponsored insurance plans among women who were insured through their spouses was statistically significant (Table 7). These coefficients’ significance persists even when family income is included as a covariate in Model 4. Taken together, these findings indicate that the insurance that a spouse provides may act as a deterrent to marital disruption in addition to other economic predictors of divorce such as employment and income.

DISCUSSION

This paper tests two main ideas. Does being dependent on a spouse for health insurance lower the hazard of divorce? And, does this relationship between health insurance dependency and divorce differ between men and women? The results affirm that on average, people who were insured through their spouses’ health plans had lower rates of divorce. Incorporating employment status into the baseline model shows that higher levels of dependency on their spouses for health insurance coverage further led to diminished risks of divorce or separation. These results are in concordance with the findings from researchers of employer-provided health benefits and job mobility. Not having an alternative source for health care outside their current arrangement—employment and marriage—made individuals less likely to terminate their jobs and marriages.

The results further demonstrate that the relationship between insurance dependency and marital stability differed by gender. The gendered relationship between health insurance and divorce mirrors the dynamics of income and marital stability. Risk of divorce rises along with wives’ contribution to the family income in excess of their husbands’ with the most stable marriages being those where the husband is the primary earner (Heckert, Nowak, & Snyder, 1998; Kalmijn, Loeve, & Manting, 2007; Ono, 1998). Likewise, I find divorce rates are the highest among women who had access to health insurance independent of their husbands. While women’s employment may have transitioned from being a marital destabilizer to a stabilizer in recent decades (Oppenheimer, 1994; Sayer et al., 2011), it appears that securing the family health insurance still remains within the male domain.

There are several limitations to this study. I recognize that marital decisions may not always be unilateral and often result from joint decision-making between the two spouses. A wife may be motivated to stay together in consideration for the husband’s lack of health insurance. Whatever mechanisms at play, my results show the different divorce outcomes based on an individual’s insurance situation. While the monthly health insurance and marital status measurements in the SIPP are strengths in determining a relationship between the two, I also note that couples often obtain legal divorce decrees months after they make their decisions. The health insurance situation of the two individuals involved in the failing marriage may have changed since beginning divorce proceedings. Research has shown increases in married women’s labor force participation in the periods prior to divorce (Gray, 1995). Similarly, the insurance-dependent partner may be motivated to secure other sources for health insurance in anticipation of the change in marital status. Applications for divorce specifically address the issue of healthcare coverage and divorcing individuals are fully aware of the termination of benefits through their soon-to-be former spouse. If this is the case, the effects on divorce rates associated with insurance-dependence may be an upper-bound more relevant to couples who were not able to secure independent coverage the month immediately prior to the finalization of the divorce. My sensitivity analyses with lagged insurance statuses of two, three, and six months indeed show that the longer the lag, the coefficient decrease in magnitude. Nevertheless, hypotheses tests still yield the same conclusion; insurance-dependence lower risk of marital disruption and that this association is more pronounced for women than for men.

The health status of an individual may change the association between insurance and divorce. Having poor health may increase the dependency on a spouse’s health insurance plan as having continuous coverage become more important and poor health can also negatively impact employment prospects. The SIPP, unfortunately, does not have good measures of health to adequately capture the respondents’ need for health insurance. In a sensitivity analysis, I used self-rated health rated on a five-point scale both as a control and as a moderator for the relationship between insurance and divorce. This variable in the 2004 SIPP was problematic as less than 3% of the analytic sample indicated that they were not in good health and the SIPP only recorded the variable twice during the four-year data collection period. Self-rated health in this case, was neither a predictor of divorce nor an influential moderator.

Lastly, the SIPP’s short study period of 48 months can also be another limitation of the data. The study can only prospectively observe the risk of divorce of a couple only for a four-year window of their marriage. It cannot account for the entire history of all couples’ marriages by tracking them from their wedding till its dissolution through divorce or death. The divorce and separation hazards are estimated from this longitudinal data’s four-year study period and are extrapolated throughout couples’ marital life courses.

Despite these limitations, this paper contributes to the broader literature examining the determinants of marital stability. It distinguishes health insurance’s influence on divorce from other measures of socioeconomic status such as education, income, and employment. If private insurance coverage was simply an artifact of these characteristics, we would expect that those with stable employer-based coverage would have the lowest rates of divorce. On the contrary, the analyses demonstrate that spouse-insured persons who have less economic options are the ones who are the least likely to divorce. While the traditional economic resources contribute to lower divorce rates by making the marriage more attractive, it is the aversion to the risk of losing health insurance that deters people away from divorce. It is a subtle but an important distinction especially when studying health policies that aim to guarantee health coverage to more people. The incentive to stay in marriages for health insurance may be stronger for those with lower prospects of securing and maintaining private health insurance through employment. The negative association between insurance dependence and divorce that we see in this paper may likely to be stronger among people with low socioeconomic status; these people may rely more on their spouses to protect themselves from the risk of losing health coverage.

This paper also underscores the gendered patterns in marriage and economic dependence. The reduction in divorce rates associated with insurance dependence is stronger among women than for men. It is consistent with research showing that American marriages are still governed by gendered social norms; men are often expected to resume the responsibility of financially providing for the family through labor force participation outside the household.

This paper draws attention to the strong connection between health care and marriage. Acquiring health insurance in the United States is largely dependent on work and marriage—two things that are valued in American society (Kaiser Family Foundation, 2011). The system rewards adults who seek and maintain good employment or who remain married to partners who can provide spousal coverage. When affordable and dependable health care is not guaranteed, people are incentivized to conform to the social behaviors that American policies promote. Whether or not sociologist and health researchers agree that access to good health care should be so intricately tied to family values, they cannot neglect the inevitable influences that they have on each other. Divorce is only one of many family processes that could be affected by health care policies in the United States. The findings in this study call attention to the importance of taking into consideration family dynamics when developing and evaluating health care policies. Marriage, childbirth, divorce, remarriage, and transitions to adulthood are all significant life events that could be shaped by policies. Understanding the relationship between health policies and family processes is crucial to working toward a more effective health care system that improves the overall health and happiness of the population.