What Explains Divorced Women’s Poorer Health?: The Mediating Role of Health Insurance and Access to Health Care in a Rural Iowan Sample

Abstract

The economic restructuring in rural areas in recent decades has been accompanied by rising marital instability. To examine the implications of the increase in divorce for the health of rural women, we examine how marital status predicts adequacy of health insurance coverage and health care access, and whether these factors help to account for the documented association between divorce and later illness. Analyzing longitudinal data from a cohort of over 400 married and recently divorced rural Iowan women, we decompose the total effect of divorce on physical illness a decade later using structural equation modeling. Divorced women are less likely to report adequate health insurance in the years following divorce, inhibiting their access to medical care and threatening their physical health. Full-time employment acts as a buffer against insurance loss for divorced women. The growth of marital instability in rural areas has had significant ramifications for women’s health; the decline of adequate health insurance coverage following divorce explains a component of the association between divorced status and poorer long-term health outcomes.

In recent decades, families in rural areas have endured sweeping changes in local economies. Along with the loss of the majority of family farms over the second half of the twentieth century (with particularly rapid losses in the Farm Crisis of the 1980s), rural communities experienced a simultaneous erosion of manufacturing jobs, and rapid increases in poverty and unemployment (Lichter and McLaughlin 1995; MacTavish and Salamon 2004). Although some families fled rural areas as economic prospects deteriorated, many others remained, and coped. In the wake of these macroeconomic changes, rural families themselves underwent a destabilizing transformation (Tickamyer and Henderson 2004). In a departure from their historically more traditional family norms and structures, rural families have experienced a significant retreat from marriage and rise in marital instability, yielding similar family patterns in rural and urban areas by the end of the twentieth century (Albrecht and Albrecht 2004; Lichter and McLaughlin 1995; MacTavish and Salamon 2004; McLaughlin, Gardner, and Lichter 1999; McLaughlin, Lichter, and Johnston 1993; Snyder and McLaughlin 2004). Such changes are consistent with a body of literature showing that economic strain is associated with divorce (Amato 2010).

Because families exert a powerful influence on health, the rise in divorce in rural areas has potential health ramifications for rural women (Carr and Springer 2010; Koball et al. 2010). Family demographers have documented that divorced men and women have, on average, poorer physical health and higher mortality rates than their married, never-married, and widowed counterparts (Amato 2010; Carr and Springer 2010; Liu and Umberson 2008; Schoenborn, 2004; Koball et al. 2010). Although healthy women are more likely to remain married than to divorce, evidence from longitudinal data suggests that some women also undergo a decline in physical health following divorce, in both national (Hughes and Waite 2009; Liu 2012) and rural samples (Lorenz et al. 2006). While many past studies document an association between divorce and women’s poorer health outcomes, a major unresolved question concerns the mechanisms whereby divorce may cause poorer health (Carr and Springer 2010).

Two major theoretical perspectives have been proposed to elucidate this association. The crisis model suggests that the interpersonal conflict preceding divorce, the marital dissolution itself, and the ensuing life changes following divorce may cause psychological distress, in detriment to physical health (Amato 2000; Booth and Amato 1991; Williams and Umberson 2004). In contrast, the resource model argues that marriage confers financial resources, as well as social and emotional support, to married individuals (Hahn 1993; Joung et al. 1997). These resources, which may reduce one’s stress and associated illness or permit healthier lifestyles or access to medical care, are often diminished in the wake of divorce, especially for women (Holden and Smock 1991; Koball et al. 2010; Smock 1994).

While some studies find that the crisis model accounts for health differentials between married and divorced individuals (e.g. Williams and Umberson 2004), other studies suggest that women’s decreased material well-being after divorce plays a more central role in determining their health outcomes (Hahn 1993; Joung et al. 1997). However, few of these studies have traced the impact of divorce on health outcomes through the mediating mechanisms proposed by the resource model. Additionally, only a handful of studies have examined disadvantage in female-headed families in rural areas (for exceptions, see Snyder and McLaughlin 2004 and Turner 2006), and none to our knowledge examine the physical health ramifications of divorce. Because compared to women in metro areas, rural divorced women have lower levels of economic well-being, greater poverty, less access to well-paying jobs, and less access to medical care, examining divorce and health in this context may be particularly important (Brown and Lichter 2004; Jensen, McLaughlin, and Slack 2004).

The current study advances our understanding of the linkage between divorce and poor health in rural areas by examining whether policy-relevant pathways associated with health insurance and access to health care can explain a proportion of the association. Using the resource perspective, this study follows a panel of married and recently divorced women in rural Iowa over ten years with the purpose of investigating how change in marital status predicts adequacy of health insurance coverage and health care access, and whether these factors may account for a component of the association between divorce and later illness.

Literature Review

Marital Status and Health Insurance

Examining the connection between marital status and health insurance coverage is important to understand the effects of divorce on subsequent health outcomes. In rural areas as in the nation as a whole, divorced and separated women are at a substantially higher risk of being uninsured; some studies suggest that the risk for single women is more than twice that of married women (Hummer et al. 2004; Jovanovic, Lin, and Chang 2003; Meyer and Pavalko 1996). A key mechanism linking women’s marital status and health insurance coverage may be husbands’ access to employer-based coverage. Between one-quarter and one-third of all women 18 to 64 are insured as dependents on family members’ (usually husbands’) employer-based health insurance policies (Kaiser Family Foundation 2008; Meyer and Pavalko 1996; Short 1998). Due to the lower proportion of jobs offering health benefits in rural areas (Larson and Hill 2005; National Advisory Committee on Rural Health and Human Services 2008; Ziller, Coburn, and Yousefian 2006), somewhat fewer rural women may receive health insurance through their husbands’ employers, but many are nonetheless vulnerable to insurance loss upon divorce.

Because married women may have multiple avenues through which to acquire health insurance—through their husbands’ jobs, in addition to their own jobs or through alternative avenues—some scholars have conceptualized marriage as an insurance “safety net” for women (e.g. Meyer and Pavalko 1996). This safety net may provide extra security during transitions within, into and out of the workplace; it may allow women to have both primary and secondary insurers; or it may provide women with multiple insurance plan options from which to choose (Meyer and Pavalko 1996; Short 1998; Wood, Goesling, and Avellar 2007). Women who divorce lose this safety net, and as single women face a higher ongoing likelihood of being uninsured than they had as married women (Jovanovic et al. 2003; Meyer and Pavalko 1996).

Family transitions are often periods of increased vulnerability for family members. Just as job loss is frequently accompanied by changes in health insurance coverage (Gruber and Madrian 1997), marital dissolution also puts women at risk of insurance loss (Lavelle and Smock 2011; Zimmer 2007). This risk is particularly threatening for women, who are as likely to lose their health insurance coverage after a change in marital status as they are after a job change in the family (Short 1998). Research also suggests that divorced women who retain their insurance coverage may face a higher likelihood of underinsurance, including benefit limits, gaps in coverage, and high deductibles or copayments that hinder access to medical care (Raiz 2006). Specifically, women, individuals in single-adult households, lower income adults, and less healthy adults have been shown to have higher rates of underinsurance than their respective counterparts (Oswald et al. 2005; Schoen et al. 2008). Divorce compounds the risk of underinsurance for women in rural areas, as rural populations already face an elevated risk (Ziller et al. 2003).

Full-Time Employment and Health Insurance

Conceptualizing marriage as a health insurance “safety net” for women experiencing a transition in employment status (e.g. Meyer and Pavalko 1996) exposes an obvious vulnerability. Due to the increasing age at first marriage, the declining universality of marriage, and the frequency of divorce (Raley and Bumpass 2003; Schoen and Standish 2001; Teachman, Tedrow, and Crowder 2004), this “safety net” is becoming more and more tenuous. Reversing this conceptualization, we propose that full-time employment may serve as a “safety net” for women experiencing a transition in marital status. This is consistent with Lavelle and Smock (2011), who find that women with stable long-term employment do not experience a significant decline in the probability of insurance coverage after divorce, in contrast to their peers. Indeed, employed women are more likely to hold health insurance than unemployed women (Meyer and Pavalko 1996) and that full-time workers are more likely than part-time workers to be insured (Kaiser Family Foundation 2011).

Unfortunately, recent decades of economic restructuring have left fewer and fewer jobs in rural areas which provide workers with health insurance. The erosion of the manufacturing industry and rise in service sector jobs in rural areas has meant a loss of quality jobs and an increase in low-wage work with poor benefits (National Advisory Committee on Rural Health and Human Services 2008). Tickamyer and Henderson (2004:111) note the catch-22: “[R]ural women’s labor has increased…at the same time that their communities have suffered from economic contraction, deindustrialization, or restructuring, making their labor both more necessary and less rewarding.” Despite this, some jobs remain in rural that provide health insurance benefits, and these likely help to preserve health insurance coverage for women after marital dissolution.

Health Insurance, Access to Health Care, and Health Outcomes

The purpose of health insurance coverage is to ensure sufficient access to medical care, and to protect individuals and families from the financial burden associated with serious medical problems. Inadequate health insurance, whether uninsurance or underinsurance, thus undermines access to care (Ayanian et al. 2000; Berk, Schur, and Cantor 1995; Jovanovic et al. 2003). Schoen and colleagues (2008) found that 53% of the underinsured and 68% of the uninsured reported forgoing at least one needed medical service or treatment during 2007, roughly twice as frequently as the adequately insured (31%). The inadequately insured are also significantly more likely to delay preventative care and to skip doses or not fill a prescribed medication due to cost (Jovanovic et al. 2003; Kaiser Family Foundation 2008; Schoen et al. 2008; Weissman et al. 1991). Some evidence additionally suggests that the inadequately insured may receive poorer quality care than their insured counterparts (Kaiser Family Foundation 2008; Schoen et al. 2008). Moreover, the uninsured and the underinsured face similarly high rates of financial stress relating to medical costs (Schoen et al. 2008).

In addition to improving access to quality health care and reducing financial stress (and in part as a result of this), health insurance is also strongly correlated with positive health outcomes in the general population, and has been shown to improve the health of vulnerable subpopulations (Levy and Meltzer 2008). Rural divorced women are vulnerable along multiple dimensions: both their marital status and their geography are associated with poorer average health status and greater economic disadvantage relative to the overall population. Some evidence further suggests that the disadvantage imposed by marital status and geography may be multiplicative, with the greater economic hardship single women face relative to married women being greater in rural areas (Brown and Lichter 2004). As such, being uninsured may lead to further health deterioration.

Being uninsured compounds the other difficulties in accessing care in rural areas. Many rural areas in the U.S. suffer from inadequate and deteriorating health infrastructures, and a corresponding shortage of physicians, nurses, and other medical personnel (Morton 2004; National Advisory Committee on Rural Health and Human Services 2008; Ziller et al. 2003). As a result, rural residents in need of medical care often face problems with appointment scheduling, may be required to travel long distances to find appropriate specialists and medical services, and may lack public transportation options to reach those services (Glasgow, Morton and Johnson 2004). In this context, uninsurance and inadequate insurance coverage pose yet another barrier to accessing health care for rural divorced women.

The Current Study

The current study traces the impact of divorce on health through the mediating pathways of inadequate health insurance and lack of access to health care. As such, it builds on previous analyses by Lorenz and colleagues (2006) which demonstrated that Iowan women who divorced between 1989 and 1990 had significantly more illnesses in 2001 than did a parallel sample of continuously married women. To extend their analyses, the model in Figure 1 presents hypotheses about the mediating pathways.

FIGURE 1.

The Hypothesized Pathway from Divorce to Subsequent Illness as Mediated by Adequacy of Health Insurance and Postponement of Medical Care

In Figure 1, the significant total effect of divorce on illness found by Lorenz et al. (2006) is expected to decline or become non-significant (β₄₁ = 0) when the mediating pathways are introduced. In the elaborated model, divorced women are expected to report having adequate health insurance less often than married women (β₂₁ < 0), which in turn is hypothesized to lead to more illnesses both directly (β₄₂ < 0) and indirectly through a higher likelihood of postponing medical care ((β₃₂ < 0) and (β₄₃ > 0)). The hypothesized pathways are predicted net of women’s age, education level, gross family income, initial self-rated health, and remarried status. Because full-time employment at the time of divorce may provide a safety net for some women in the event of marital dissolution, we further compare the model in Figure 1 by employment status to determine whether and to what extent full-time employment moderates the pathway from divorce to subsequent illness.

Methods

Data

The data for this study were obtained from the Iowa Midlife Transitions Project (MTP), a longitudinal (1991–2001) study of 539 rural families from an eight-county region in north central Iowa that closely mirrors the economic diversity of the rural Midwest (Lorenz et al. 2006). Originally designed to examine the effects of the Farm Crisis of the 1980s on rural families, the dataset contains information on family demographics, work experiences, financial well-being, and a wide variety of other variables.

MTP families originally participated in either the Iowa Youth and Families Project (IYFP; Conger and Elder 1994) or the Iowa Single Parent Project (ISPP; Simons 1996). The IYFP began in 1989 as a study of married-parent families with at least two children per family, one of whom was a seventh grader in 1989, and another a sibling within four years of age. The ISPP joined the study two years later, adding recently divorced mothers with at least two children, one of whom was a ninth grader in 1991, and another a sibling within four years of age. Researchers worked with public and private schools in the eight-county region to identify families meeting selection criteria. Seventy-eight percent of eligible married couples and ninety-nine percent of single mothers agreed to participate (Conger and Elder 1994; Simons 1996).

The sample for this study includes 448 women who participated at least once in 1991, 1992, or 1994 and again in 2001, the final wave of the MTP (83% retention). Women who were originally from the IYFP sample (86% retention) were marginally more likely to stay in the study than the ISPP mothers (79%). As part of an attrition analysis, we estimate that those who dropped out of the panel were slightly younger (38.9 vs. 39.9 years of age; t = 2.03), less well educated (12.7 vs. 13.5 years of school; t = 4.20), and had significantly lower gross family income ($26,631 vs. $36,134; t = 2.76) in 1991. We observed no other significant differences, including baseline self-reported health. If women who attritted from our sample are slightly more disadvantaged than the population, on average (as in most studies), then our results will likely underestimate real disparities, and our analyses less likely to detect significant findings.

Measures

Table 1 provides descriptive statistics for the variables used to measure constructs outlined in the conceptual model above, and examines differences in means by marital status in the first wave of the survey. Roughly one-fifth of the women in the study had divorced between the beginning of 1989 and the end of 1990, and the other women were married. In multivariate analyses, Divorced (1991) is coded to contrast women divorced at baseline (1) with married women (−1). To acknowledge that some of the divorced women remarried over the ten-year study period, the variable Remarried (1991–2001) contrasts the divorced women who remarried (1) with divorced women who stayed single (−1) while coding the continuously married women with a zero (0). Forty-three percent of women who were divorced at baseline remarried over the ten-year study period.

TABLE 1.

Descriptive Statistics for Married and Divorced Women

Variable	Round	Total (N=448)a		Married (N=361)		Divorced (N=87)		t
		Mean	SD	Mean	SD	Mean	SD
Physical Illness	2001	5.2	4.0	4.9	3.8	6.4	4.8	−3.0 **
Adequate Health Insurance	1992	69.6%	46.0%	73.3%	44.3%	55.2%	50.0%	3.3 **
	1994	73.7%	44.1%	77.5%	41.8%	57.7%	49.7%	3.8 ***
Postpone Medical Care	1991	37.2%	48.4%	34.2%	47.5%	49.4%	50.3%	−2.6 **
	1992	40.0%	49.0%	37.3%	48.4%	50.6%	50.3%	−2.3 *
	1994	35.9%	48.0%	32.6%	46.9%	50.6%	50.3%	−3.1 **
Age	1991	39.9	4.0	40.1	4.1	39.0	3.7	2.3 *
Education	1991	13.5	1.7	13.6	1.7	13.4	1.7	0.9
Family Income ($1,000s)	1991	36.1	30.2	40.0	31.0	20.1	19.4	5.7 ***
Self-Rated Health	1991	3.8	0.9	3.8	0.8	3.7	0.9	1.2
Full-Time Job	1991	56.7%	49.6%	53.2%	50.0%	70.9%	45.7%	−3.0 **

^aMissing data and pairwise deletion result in sample sizes ranging from 430 to 448.

^*p<.05;

^**p<.01;

^***p<.001

The outcome variable, Physical Illness (2001), was constructed by summing respondent reports of symptoms and diseases. Respondents indicated (yes=1, no=0) whether during the past two years they had experienced one or more of 70 symptoms and physical illnesses, ranging from relatively minor conditions such as the common cold and sore throats to more severe diseases such as heart conditions, diabetes, and cancer. Although consideration was given to weighing illnesses according to severity, the final decision was to use simple counts because it is the conventional practice in the literature (e.g., Lin and Ensel 1989, Liang et al. 1991; Ferraro and Farmer 1996; Kubzansky, Martin and Buka 2009) and consistent with previous research using the data upon which the current study builds (Lorenz et al. 2006; Wickrama et al. 2006). Analyses not shown demonstrate that higher counts of illness correspond to the presence of more serious illnesses.

Counts of physical illness is chosen as an outcome instead of more global measures because it may be more objective relative to other survey measures of physical health status. Because respondents are asked to report on the presence or absence of each condition or illness on a list, in isolation from one another, the physical illness outcome may be less sensitive to variations in respondent psychological well-being than more subjective global measures such as self-rated health (Ferraro and Farmer 1996). Our relatively more objective measure of physical health, with its emphasis on identifiable illnesses, is also congruent with the goals of this study because it investigates whether divorce predicts poorer long-term health in part by imposing structural barriers to their ability to address specific health care issues when they arise. Women in the sample reported an average of 5.2 illnesses in the two-year period prior to the 2001 interview. The average number of illnesses divorced women reported (6.4) was significantly higher than that of married women (4.9).

The first of the two mediating variables, Adequate Health Insurance (1992;1994), indicates whether a woman held health insurance she considered adequate to meet her medical needs in 1992 and 1994. (Health insurance information was not collected on the 1991 survey.) Because “the experiences of adults with inadequate coverage mirror those of their uninsured peers” both in terms of barriers to accessing health care and financial burdens (Schoen et al. 2005:289), the zero category includes both uninsured and underinsured women, where underinsurance is self-assessed (Blewett, Ward and Beebe 2006). The inadequately insured group was split roughly equally between women without health insurance and women with coverage they deemed inadequate. Almost three-quarters of the women in the sample (73.7% in 1994) reported having adequate health insurance coverage. However, as expected, the proportion of divorced women with adequate coverage was much lower, just over half of the sample in both 1992 and 1994.

The a priori decision to combine uninsured and underinsured was further supported by other characteristics of our sample. Analyses not included in tabular form indicate that the uninsured and underinsured were roughly equal in their rate of postponing health care in 1992 (68% and 62%, respectively; see below for definition) compared with those who were adequately insured (29%). Similar results were obtained for 1994 (73% and 65% compared with 24%). Examining the longitudinal consequences of inadequate health insurance, women who were uninsured and underinsured in 1994 reported similar numbers of illnesses in 2001 (6.0 and 6.6, respectively), significantly higher than their adequately insured counterparts (4.8).

Postpone Medical Care (1991;1992;1994) indicates whether women reported postponing medical and dental care for financial reasons in 1991, 1992, and 1994. In each wave of data collection, women were presented with a list of possible financial cutbacks and asked, “During the past twelve months, has your family made any of the following adjustments because of financial need?” Women who selected “postponed medical/dental care” are coded one (1) and others are coded zero (0). In each wave of data collection, between thirty-five and forty percent of all women reported postponing medical or dental care for financial reasons, reflecting, perhaps, the economic difficulties faced by this rural sample following the farm crisis. Divorced women reported postponing care even more frequently, with about half indicating they had made this financial cutback at each point in time.

The analyses control for four key covariates, each measured in the first wave of data collection. Age (1991) gives a woman’s age in years; Education (1991) indicates years of school completed; Family Income (1991) is gross family income, measured in $1,000s; and Self-Rated Health (1991) is a five-item global scale indicating self-perceived health (1=poor; 2=fair; 3=good; 4=very good; 5=excellent). Self-rated health serves as the baseline health control for several reasons. First, the full checklist of illnesses was not administered in 1991. Second, consistent with our purpose, “Self-rated health captures the full array of illnesses a person has and possibly even symptoms of disease as yet undiagnosed but present in preclinical or prodromal stages” (Idler and Benyamini 1991: 27). Additionally, self-rated health is correlated with counts of illness both in our data (2001: −0.469; p<.001) and national data (Ferraro and Farmer 1999), and predicts medical care usage (Angel and Gronfein 1988; Gold, Franks and Erickson 1996) and a range of subsequent health outcomes including physician assessments of health and mortality (Idler and Benyamini 1997; Ferraro and Farmer 1999).) and national data (Ferraro and Farmer 1999), and predicts medical care usage (Angel and Gronfein 1988; Gold, Franks and Erickson 1996) and a range of subsequent health outcomes including physician assessments of health and mortality (Idler and Benyamini 1997; Ferraro and Farmer 1999).

In 1991, women averaged 39.9 years of age. Divorced women were, on average, slightly younger than married women. Sample respondents averaged 13.5 years of school; there was no significant difference by marital status. The average gross family income in 1991 was $36,100. The average family income among the divorced women ($20,100) was only about half the amount reported by married women, even though they were estimated to have incomes very similar to the married women before their divorce (Simons, 1996). The initial self-rated health of married (3.8) and divorced (3.7) women did not differ significantly.

The proposed moderator of the conceptual model in Figure 1 is full-time employment in the first wave of data collection. Full-Time Job (1991) indicates whether women reported working in any one job for greater than or equal to 35 hours per week in 1991. The maximum number of hours in a single job is used rather than summing total weekly hours worked across multiple jobs because receipt of benefits often depends on full-time employment status. In 1991, just over half of the women in the sample were employed full-time. This proportion was significantly greater among divorced women, of whom almost three-quarters held full-time jobs.

Analytic Strategy

The analyses that follow examine two potential mediators and a moderator of the relationship between divorce and later illness. First, we compare a sequence of structural equation models (SEMs) testing the conceptual model in Figure 1 (see Table 2). Then, we examine whether employment status moderates the pathways identified for the full group of women (Table 3).

TABLE 2.

Standardized Path Coefficients for Structural Equation Models Tracing a Pathway from Divorce to Physical Illness

	Model 1: Baseline	Model 2: Divorce -> Illness	Model 3: + Health Insurance	Model 4: + Postpone Med. Care
Measurement Model
Adequate Health Insurance
→ Adq. Health Insurance (1992)	0.66 --	0.66 --	0.67 --	0.70 --
→ Adq. Health Insurance (1994)	0.76 **	0.76 **	0.76 ***	0.72 ***
Postpone Medical Care
→ Postpone (1991)	0.71 --	0.71 --	0.71 --	0.69 --
→ Postpone (1992)	0.77 ***	0.77 ***	0.77 ***	0.74 ***
→ Postpone (1994)	0.70 ***	0.70 ***	0.70 ***	0.74 ***
Structural Model
→ Physical Illness (2001)
Postpone Medical Care				0.21 *
Adequate Health Insurance			−0.17 **	−0.04
Divorced (1991)		0.12 *	0.09	0.09
Remarried (1991–2001)		0.01	0.04	0.02
Age (1991)	0.03	0.04	0.03	0.05
Education (1991)	0.08	0.07	0.09	0.10 *
Family Income (1991)	−0.08	−0.05	−0.02	−0.01
Self-Rated Health (1991)	−0.39 ***	−0.39 ***	−0.36 ***	−0.36 ***
→ Adequate Health Insurance
Divorced (1991)			−0.15 *	−0.16 **
Remarried (1991–2001)			0.15 *	0.12 *
Age (1991)	−0.02	−0.02	−0.02	−0.02
Education (1991)	0.11	0.11	0.10	0.09
Family Income (1991)	0.24 **	0.24 **	0.20 **	0.20 **
Self-Rated Health (1991)	0.15 *	0.15 *	0.14 *	0.14 *
→ Postpone Medical Care
Adequate Health Insurance				−0.61 ***
Age (1991)	−0.06	−0.06	−0.06	−0.06
Education (1991)	−0.08	−0.08	−0.08	−0.02
Family Income (1991)	−0.20 **	−0.20 **	−0.20 **	−0.06
Self-Rated Health (1991)	−0.13 *	−0.13 *	−0.13 *	−0.04
N	448	448	448	448
R2illness	0.17	0.18	0.20	0.23
R2insurance	0.11	0.11	0.16	0.15
R2postpne	0.09	0.09	0.09	0.41
χ2	155.86	149.27	123.97	28.79
df	34	32	29	27
CFI	0.84	0.83	0.87	1.00
RMSEA	0.09	0.09	0.09	0.01

^*p<.05;

^**p<.01;

^***p<.001

TABLE 3.

Proportion of Women with Adequate Health Insurance Across Time by Marital and Work Status at Baseline (1991)

	Adequate Health Insurance (1992)		Adequate Health Insurance (1994)
	Not Employed Full-Time	Employed Full-Time	Not Employed Full-Time	Employed Full-Time
Divorced	36.0%	63.9%a	32.0%	69.5%a
Married	72.6%ab	75.1%b	75.8%ab	80.2%b

Superscripts identify statistically indistinguishible proportions (p<.10).

Each of the models in Table 2 consists of both a measurement model and a structural model. The measurement portion (top of Table 2) uses multiple indicators of each latent variable to separate the variance common across the separate indicators from the variance unique to each separate indicator (thereby controlling for measurement error and yielding stronger overall measures), while the structural portion (bottom of Table 2) estimates the strength of relationships between the latent variables and all other variables in the model after controlling for measurement error (Bollen 1989).

In our models in Table 2, there are two latent variables, Adequate Health Insurance and Postpone Medical Care. The former has two observed indicators, measured in 1992 and 1994, while the latter has three observed indicators, measured in 1991, 1992 and 1994. The standardized path coefficients linking each observed indicator to its underlying latent variable are referred to as standardized factor loadings, similar to factor loadings in traditional factor analysis. As a rough guide, factor loadings greater that 0.50 are considered adequate and all our loadings are greater than 0.65. We tested measurement models using both categorical and continuous specifications of the adequate health insurance and postpone medical care indicator variables, and found that this specification did not substantively impact results. Accordingly, the fitted models reflect the continuous specification of the measurement model.

The models are estimated using maximum likelihood with bootstrap estimates of standard errors for direct and indirect effects (1,000 iterations) as recommended by Bollen and Stine (1990). Because eight percent of the 448 sample members had missing data on some items, estimates were obtained using full-information maximum likelihood, which has been shown to have good performance in structural equation models (Enders and Bandalos 2001). All of the models correlate the six exogenous variables (Divorced (1991); Remarried (1991–2001); Age (1991); Education (1991); Family Income (1991); Self-Rated Health (1991)), and assume the residuals among endogenous variables (Adequate Health Insurance; Postpone Medical Care; Physical Illness (2001)) are uncorrelated.

Results

SEM and the Total Effect of Divorce on Subsequent Illness

Table 2 presents standardized path coefficients for the sequence of four structural equation models tracing the proposed pathway from divorce to number of illnesses a decade later. The table is arrayed so that the four models are presented in the four columns across the top. The factor loadings for the measurement portion of each model are reported in the first rows under “Measurement Model.” These factor loadings are from 0.66 to 0.77, all within the acceptable range of values (Bollen 1989). The “Structural Model” portion of Table 2 presents coefficients linking the three endogenous variables Physical Illness (2001), Adequate Health Insurance, and Postpone Medical Care—with predictor variables. Pathways to—Physical Illness (2001), the key outcome variable, are presented first. The first and second proposed mediators, Adequate Health Insurance and Postpone Medical Care, are then presented in turn. The bottom of Table 2 reports the summary statistics for each model, including the estimated chi-square, the comparative fit index (CFI) and the root mean square error of approximation (RMSEA).

Model 1 identifies three important associations between exogenous covariates and endogenous variables. First, women with higher incomes are more likely to have adequate health insurance (0.24; p < 0.01) and less likely to postpone medical care for financial reasons (−0.20; p < 0.01). Consistent with previous research, these findings indicate that individuals with lower incomes are more likely to be uninsured or underinsured compared to those with higher incomes, and are more likely to delay needed medical care (Schoen et al. 2008; Weissman et al. 1991). Second, women with better self-rated health in the initial survey wave are more likely to have adequate health insurance (0.15; p<0.05) and less likely to postpone medical care (−0.13; p < 0.05) in intermediate waves. The causality of this association may work in both directions: women who tend to have better health insurance coverage and better access to medical care may come to have improved health; or, women with poorer health may be denied adequate health insurance by not meeting eligibility requirements. This finding is consistent with previous research, which suggests that a higher proportion of individuals with health problems are uninsured or underinsured compared to healthier individuals (Schoen et al. 2008; Ziller et al. 2006). Third, better self-rated health in the initial survey wave (1991) is strongly predictive of lower numbers of illness ten years later (−0.39; p < 0.001), thus providing evidence of continuity in health across the decade.

Adding to the baseline model, Model 2 estimates the total effect of divorce on counts of illness a decade later. Consistent with the descriptive analyses in Table 1, the path coefficient of 0.12 (p<0.05) indicates that divorced women have significantly higher levels of illness than married women 10 years after divorce occurs, even after controlling for remarriage, baseline health status and other covariates.

Effect of Divorce on Illness via Adequacy of Health Insurance and Access to Care

Model 3 adds two key pathways to Model 2: one from divorce to adequacy of health insurance (−0.15; p < 0.05) and another from adequacy of health insurance to counts of illness (0.17; p < 0.01). These estimates suggest that women who divorced between 1989 and 1990 were significantly less likely to hold adequate health insurance in the years following the divorce (1992, 1994), and that women holding inadequate health insurance report more illnesses in 2001. Adding this indirect effect to the model reduces the magnitude of the effect from divorce to illness from a significant 0.12 (p < 0.05) to a non-significant 0.09, suggesting that adequacy of health insurance partially mediates the relationship between divorce and later illness. Note that Model 3 also adds a third pathway, showing that remarriage is also significantly predictive of adequate health insurance (0.15; p<.05), perhaps suggesting that those who remarried may have entered divorce with more resources, one of which may have been health insurance. Alternatively, some women who remarried may have recovered health insurance by gaining access to the fringe benefits of their new spouses.

Model 4 represents the model in Figure 1 and further elaborates the relationship between adequacy of health insurance and subsequent health by estimating the mediating effects of postponing medical care. Women who report having inadequate health insurance are much more likely to postpone medical care for financial reasons (−0.61; p < 0.001), and those who postpone medical care report more illnesses in 2001 (0.21; p < 0.05). Adding this indirect effect to the model reduces the direct effect of adequacy of health insurance on counts of illness from −0.17 (p<.01) to −0.04 (n.s.), indicating that postponement of medical care mediates the relationship between adequacy of health insurance and illness. The key mechanism linking adequacy of health insurance and later illness may thus be access to needed health care. Model 4 also contains one unintuitive result: women with higher levels of education report higher counts of illnesses in 2001, even after controlling for other factors. Although this could be a methodological artifact of multiple regression, we also speculate that better educated individuals may be more familiar with a wider range of illnesses, and more vigilant in observing and reporting illnesses in the self-report checklist.

Because the four models in Table 2 are nested, differences in the chi-square goodness-of-fit tests can be used to compare the improvement in fit of one model over the next. Accordingly, Model 3 fits significantly better than Model 2 (Δχ² = 25.30; Δdf = 3; p < 0.001), but our hypothesized Model 4 is the strongest model, fitting clearly better than Model 3 (Δχ² = 95.18; Δdf = 2; p <0.001).

Employment as a Buffer against Insurance Loss after Divorce

Do divorced women in rural areas working full-time experience the same pathway to higher rates of illness as their unemployed and part-time-working counterparts? To answer this question, we first examine descriptive statistics on how employment status relates to health insurance coverage. Table 3 presents proportions of women in 1992 and 1994 who report adequate health insurance coverage, by their marital status and employment status in 1991. The women most likely to report being uninsured or underinsured are divorced women who were not working full-time at or shortly after their divorce. Only 36 percent and 32 percent of this group reported adequate insurance coverage in 1992 and 1994, respectively. Among women employed full-time, divorced women were again less likely to hold adequate insurance coverage compared to married women, but by a much smaller margin (1992: 63.9% vs. 75.1%; p<0.10), suggesting a partial protective effect of full-time employment on health insurance coverage. Women both married and employed full-time reported the highest rates of adequate insurance coverage (1992: 75.1%; 1994: 80.2%), although not statistically higher than those reported other married women.), suggesting a partial protective effect of full-time employment on health insurance coverage. Women both married and employed full-time reported the highest rates of adequate insurance coverage (1992: 75.1%; 1994: 80.2%), although not statistically higher than those reported other married women.

The protective effect of full-time employment status on divorced women’s health insurance coverage is confirmed by conducting a multiple group analysis of Model 4 in which women who are employed full-time are compared with those who are not (Bollen 1989). Although the analyses are based on unstandardized data, we report the standardized coefficients for ease of interpretation. Focusing specifically on the pathway from divorce to adequacy of health insurance, the path coefficient is a significant −0.27 for women not employed full-time in 1991, whereas the corresponding coefficient for women employed full-time in 1991 is not statistically significant and estimated at −0.13. When constraining the two paths to be equal, the increase in chi-square is significant (Δχ² = 4.1, Δdf =1, p < 0.05). This comparison suggests that full-time employment has a protective effect on divorced women’s health insurance status. As a consequence, divorced women employed full-time at or soon after divorce may be less likely to postpone necessary medical care and, on average, report better physical health outcomes than other divorced women. This is consistent with descriptive statistics (not shown in tabular form) which show that divorced women working full-time in 1991 report an average of 5.9 illnesses in 2001, compared with the 7.5 illnesses reported by divorced women not working full-time.

Discussion

Previous studies suggest that material resources, broadly speaking, are associated with health differentials between married and divorced women. Extending this line of research, this article specifically identifies health insurance coverage as an important policy-relevant resource connecting divorce with poorer long-term health outcomes in a sample of rural women. Expanding upon research from multiple disciplines, this study is the first to link the pathways from divorce to illness via lower rates of health insurance coverage—in any sample—and to investigate the causal model as a whole within an SEM framework using a longitudinal dataset. Although SEM analyses cannot prove a causal relationship, associations observed in the data are consistent with a causal explanation of divorce’s effect on health insurance and increased illness. We trace the effects of divorced women’s poorer health insurance coverage longitudinally, seeing that the absence or inadequacy of this important financial resource inhibits access to medical care, in turn eroding physical health. Our analyses indicate that divorced women are less likely to have adequate health insurance, decreasing their likelihood of making needed medical and dental visits. Divorced women’s lack of adequate health insurance proves to have a detrimental association with long-term health.

The findings lend support to the resource model in explaining health differentials between married and divorced individuals. Health insurance coverage, often jeopardized in the event of divorce, is important as both a financial and a health resource. The importance of health insurance in the resource model is also consistent with recent research showing that the gap in physical health between married and previously married individuals has been increasing over the past 30 years (Liu and Umberson 2008).

That said, it remains that neither the resource model nor its close competitor, the crisis model, adequately explains the growing health differential nationally or in rural areas. From the crisis perspective divorce rates have increased dramatically over this time period, causing this life-course transition to become “more normative and less stigmatized” (Liu and Umberson 2008:240). Thus the accompanying stress and associated health problems would be expected to decline over time rather than increase. However, the material resources model is not wholly adequate either. Because women’s labor force participation has dramatically increased over this time period, it is unlikely the material hardship of divorced individuals has grown over recent decades. Rather, Smock (1993) shows that the costs of material disruption appear to be stable over time.

Findings from the current study shed light on a possible explanation for the declining health of divorced individuals relative to married individuals over the past thirty years. During this time period, there has been a rapid increase in health care costs relative to most other categories of household expenditures, which suggests that risks posed by being un- or underinsured may have increased, disproportionately affecting divorced individuals (Altman and Levitt 2002). Future research should probe this possibility.

The current study also provides support for the notion that full-time employment may act as a health insurance safety net in times of marital instability. Full-time employment appears to protect health insurance coverage after divorce. Divorced women who do not work full-time are least likely to be adequately insured, while divorced women working full-time hold adequate health insurance at a much higher rate. This finding is aligned with previous research that indicates that the decline in women’s broader economic well-being following marital dissolution is much less acute for women employed full-time (Bianchi, Subaiya, and Kahn 1999).

Geography, generalizability and future replications

A recent review of linkages between families and health noted that the strongest emerging work in this area moves beyond the question, “Does family structure affect health?” and addresses instead, “Under what conditions, for which outcomes, for whom, and through which pathways does family structure, context and process affect health?” (Carr and Springer 2010:743). This study begins to meet this challenge, and advances the literature to examine the mechanisms through which divorce impacts health in a rural Iowan sample.

A fruitful direction for future research will be to assess heterogeneity in the marital disruption-health relationship and in potential intermediary mechanisms. The relatively small and geographically homogenous sample limited our capacity to do so in the current study, beyond considering women’s employment status as a potential buffer against insurance loss. Recent studies have looked at how changes in health after marital disruption differ by women’s age and birth cohort (Liu 2012) and by marital quality (Williams and Umberson 2004; Hawkins and Booth 2005), but none to our knowledge examine differences by social context.

Certainly, the health advantages of marriage have been documented across a variety of populations and contexts, including samples in North America, Europe and Asia (Manzoli et al. 2007; Hughes and Waite 2009) and for Whites, African Americans and Hispanics in the U.S. (Schoenborn 2004). Nevertheless, differences in the marital disruption-health relationship by social context seem likely in light of the large and growing body of evidence that the characteristics of places in which people live and work are important determinants of health status (Taylor, Repetti and Seeman 1997; Diez Roux and Mair 2010). For example, women who live in places and times with a large stock of quality, affordable housing, a tight labor market, strong community social ties, outdoor spaces for recreation, and an adequate supply of mental and physical health care services might experience minimal health declines after divorce compared to their counterparts in less healthy environments. Learning more about such specific contextual modifiers would further our understanding of the intermediary mechanisms that link marital disruption and health.

It is also worth considering how the more aggregate geographic context—rural and urban residence—affects the marital disruption-health relationship. Future studies should replicate the model in Figure 1 using urban or nationally representative datasets to determine whether the mechanisms identified in this study are operational in other geographic contexts. One hypothesis is that marital disruption might have larger impacts on rural women’s health and use of health care because rural populations have lower rates of health insurance coverage, poorer access to medical care, and poorer overall health compared to urban populations (Ziller et al. 2003). Additionally, divorced women in rural areas are perhaps even more disadvantaged than their urban counterparts, as evidenced by their poorer job prospects and higher rates of poverty (Brown and Lichter 2004). On the other hand, because family patterns of marriage and divorce in rural areas have largely converged with urban family patterns, the mechanisms identified in this study may reflect those in the U.S. population as a whole (Lichter and Brown 2011; MacTavish and Salamon 2004). This hypothesis seems plausible when considering that the disparity in rates of insurance coverage by marital status even more pronounced in urban areas, where the risk of being uninsured is twice as high for single individuals as for married ones (Hummer et al. 2004).

Replications should try, where possible, to refine measures of insurance coverage and to distinguish gradients of uninsurance and underinsurance. One advantage of the MTP data is that it allows us to consider the adequacy of a woman’s insurance coverage, rather than looking strictly at the presence or absence of insurance coverage. Still there is room for improvement. Lower-quality health insurance plans often have high deductibles, high copays, and low payout limits.¹ In 2007, nearly two-thirds (62.1%) of personal bankruptcies in the United States resulted from the high cost of medical problems; surprisingly, three-quarters of those medical debtors had health insurance at the onset of their illness or condition (Himmelstein et al. 2007). This striking figure demonstrates the lack of protection afforded by poor quality insurance coverage. Studies that examine the effects of “any insurance coverage” will likely underestimate the detrimental effects of divorce on illness as mediated by health insurance coverage (Schoen et al. 2008). Divorced women able to avoid insurance loss by switching onto a new plan (e.g., through their own employer, the private market, or a public insurance program) may experience a decline in quality of health insurance coverage, a decline which would be missed entirely if only uninsurance status is tabulated.

Replications should also examine the robustness of study results to alternative measures of health outcomes. This study focuses on counts of physical illness because it tests a model of structural detriments to post-divorce health (lack of adequate health insurance, limited access to medical care). We speculate that relatively more subjective health outcomes such as global measures of self-rated health may prove to have weaker ties with structural detriments of health following divorce (but perhaps stronger ties with social-psychological detriments), and as such, replications using these measures may fail to detect any contribution of health insurance and access to health care to divorced women’s poorer health. Future research may also examine whether the structural pathways found here are replicated using measures of mortality, functional limitations, and comprehensive indices of health that combine multiple dimensions (see e.g., Boardman 2004). Advancing our understanding of the pathways linking marital disruption and each of these health outcomes is an important next step in this literature, which has laid the groundwork by documenting the basic marital health advantage across a wide range of health outcomes (Hahn 1993; Johnson, Backlund, Sorlie and Love 2000; Schoenborn 2004; Lorenz et al. 2006; Manzoli et al. 2007; Liu and Umberson 2008; Hughes and Waite 2009; Liu 2012).

Policy implications

Considering the high instability of U.S. family structures, this study exposes the consequences of a major fault in the current health insurance system. The U.S. health care system currently offers little protection against insurance loss for individuals who divorce. One federal law—the Consolidated Omnibus Budget Reconciliation Act of 1985, commonly referred to as COBRA—grants most newly divorced individuals the right to retain dependent coverage through their ex-spouses’ employers for up to 36 months after divorce (U.S. Department of Labor 2011). Electing COBRA continuation coverage, however, is often prohibitively expensive, limiting its effectiveness. In 2010, standard COBRA premiums averaged $429 monthly for individual coverage (Kaiser/HRET 2010).

However, there is one uplifting aspect of this troubling story. Unlike many other factors contributing to the poorer health of divorced women—the stress of marital conflict and the divorce process and the loss of social support and shared economic resources—the decline in access to quality health insurance is one that is more directly policy amenable. Given the far greater instability in both family life and employment than in decades past, one might question the wisdom of continuing to tie health insurance coverage to marital and employment roles (Montez, Angel, and Angel 2009).

Certainly, the best way in which to provision health insurance coverage has been a point of significant debate in national politics in recent years. The Affordable Care Act passed in March 2010 under the Obama administration has the potential to mitigate the risk of insurance loss following divorce, thereby helping to protect the health of divorced women in rural communities and elsewhere. The law has provisions expected to expand the availability of insurance to women through their own jobs, to potentially make insurance more affordable for women on the private market, and to expand eligibility for Medicaid and other need-based public insurance programs. However, the major provisions of this new law are not scheduled to go into effect until 2014, so for several years data will remain unavailable to examine empirically how the health insurance reforms contained therein will change the health insurance and health prospects of marital separators.

Regardless of how health insurance and health care reforms unfold over the coming years (and even decades), policymakers should be aware that a system that induces a de facto linkage between marital status and insurance coverage may have unintentional and deleterious consequences.