Skip to main content
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2014 Jun 1.
Published in final edited form as: J Health Soc Behav. 2013 May 30;54(2):166–182. doi: 10.1177/0022146513481230

Explaining the Widening Education Gap in Mortality among U.S. White Women

Jennifer Karas Montez 1, Anna Zajacova 2
PMCID: PMC3747639  NIHMSID: NIHMS502858  PMID: 23723344

Abstract

Over the last half century the gap in mortality across education levels grew in the United States, and since the mid-1980s the growth was especially pronounced among white women. The reasons for the growth among white women are unclear. We investigated three explanations—social-psychological factors, economic circumstances, and health behaviors—for the widening education gap in mortality across 1997-2006 among white women 45-84 years of age. We used data from the National Health Interview Survey Linked Mortality File (N=46,744; deaths=4,053). We found little support for social-psychological factors; however, economic circumstances and health behaviors jointly explained the growing education gap in mortality to statistical nonsignificance. Employment and smoking were the most important individual components. Increasing high school graduation rates, reducing smoking prevalence, and designing work-family policies that help women find and maintain desirable employment may reduce mortality inequalities among women.


The inequality in mortality risk across education levels in the United States is well-documented (e.g., Hummer and Lariscy 2011) and it has increased over the last half century (Crimmins and Saito 2001; Feldman et al. 1989; Lauderdale 2001; Pappas et al. 1993; Preston and Elo 1995; Rogot, Sorlie, and Johnson 1992). Since the mid-1980s, the inequality appears to have grown more among women than men (Cutler et al. 2011; Meara, Richards, and Cutler 2008; Montez et al. 2011). Among women, this recent growth has reflected declines in mortality among the higher-educated alongside increases in mortality among the lower-educated (Meara et al. 2008; Montez et al. 2011). The trends, which we describe below, have been particularly unfavorable among white women (Jemal et al. 2008; Meara et al. 2008; Montez et al. 2011).

The reasons for the widening mortality gap among white women are not fully understood. Only a few studies have investigated the reasons. Those studies largely focused on trends in specific causes of death (Meara et al. 2008; Montez and Zajacova 2013) or in health-related behaviors (Cutler et al. 2011) across education levels. They generally conclude that diverging smoking patterns played an important role. For example, two causes of death for which smoking is a major risk factor—lung cancer and chronic lower respiratory disease—explain one-quarter to one-half of the growing gap in all-cause mortality since the mid-1980s among white women 45-84 years of age (Meara et al. 2008; Montez and Zajacova 2013). While those studies are informative, they do not provide a complete explanation.

By focusing almost exclusively on education-specific trends in causes of death or health behaviors, prior studies have drawn attention to “downstream,” behavioral explanations (e.g., smoking and lung cancer mortality; obesity and diabetes mortality). Thus, they have not revealed why diverging behavioral patterns emerged nor have they addressed non-behavioral explanations such as economic policy, labor market participation, and social integration that are further “upstream” in the causal chain and that may be linked to numerous causes of death.

A complete explanation of the growing gap requires identifying the “causes of causes” (Rose 2008:128). In other words, it requires searching for broader contextual explanations that lie further upstream in the causal chain. The best insights will then be gleaned by triangulating the findings regarding the contextual explanations with findings regarding education-specific trends in causes of death. The benefits of looking for the “causes of causes” are two-fold. First, the closer we get to the root causes of the growing inequality the better chance we have of reducing it. A fundamental cause perspective impels us to search for explanations as far upstream as possible because eliminating a downstream mechanism may have limited benefit since other downstream mechanisms will likely take its place (Link 2008). Second, this approach may also be powerful in identifying policy levers to stem the divergence since such levers can be aimed at the contextual factors such as employment and educational policies.

In this study, we provide new evidence about the reasons for the widening gap in all-cause mortality risk across education levels among white women. We investigate a range of factors that are well-known mechanisms linking education and mortality but have not been examined in this context: social-psychological factors, economic circumstances, and health behaviors. Our findings address the growing concern among scholars and policy makers about the widening educational divide in longevity. For instance, the Healthy People initiative has sought to reduce health disparities within the U.S. population, including disparities by socioeconomic status (DHHS 2000). This study aims to identify mechanisms that may have the greatest leverage in achieving those goals.

BACKGROUND

Prior Research

Recent studies have found that since the mid-1980s, mortality risk among white women slightly decreased among the college-educated, remained fairly stable among women with a high school credential or some college, and increased among women with 0-11 years of education; the latter group was primarily responsible for the growing gap (Jemal et al. 2008; Miech et al. 2011; Montez et al. 2011; Montez and Zajacova 2013). For example, during 1986-1994, the age-standardized death rate of white women 45-84 years of age with 0-11 years of education (2,400 deaths per 100,000 women, or 0.024) was 3.4 times larger than the death rate for college-educated women (0.007). By 1995-1998 the death rate of the low-educated group was 4.3 times larger (0.026/0.006) and by 2003-2006 it was 4.7 times larger (0.028/0.006) than the rate for the high-educated group (Montez and Zajacova 2013). Life expectancy has also diverged. Meara and colleagues (2008) estimated life expectancy at age 25 among low- (0-12 years) and high- (13 or more years) educated adults in 1990 and 2000. Life expectancy increased by one year among high-educated white women but decreased by nearly one year among their low-educated peers.

As we mentioned above, the handful of studies that have investigated the reasons for the widening mortality gap has focused on describing trends in causes of death. The causes identified as most important in each study vary somewhat depending on the age range examined. For example, among white women 45-84 years of age, lung cancer and chronic lower respiratory disease explained one-quarter to one-half of the growing gap in all-cause mortality risk since the mid-1980s (Meara et al. 2008; Montez and Zajacova 2013). The mortality gap also increased for diabetes, cerebrovascular disease, and Alzheimer's disease among this age group (Montez and Zajacova 2013). Among younger women 25-64 years of age, deaths from accidents contributed the largest percentage to the widening gap during the mid to late 1990s (Jemal et al. 2008). In a middle-aged sample of women 40-64 years of age, accidental poisoning, chronic lower respiratory disease, and cancers of the trachea, lung, and bronchus were the main contributors between 1999 and 2007 (Miech et al. 2011).

We are aware of only one previous study that investigated some of the mechanisms that might explain the growing mortality gap. Using two U.S. population surveys, Cutler et al. (2011) examined the role of smoking and obesity in explaining the widening gap in mortality risk across education levels from the 1970s to 1990s among non-Hispanic whites 25-74 years of age. They found that smoking and obesity explained 10-40 percent of the increasing gap among women. However, they concluded that trends in the distribution of the behavioral risk factors were not the major explanation; instead, the mortality consequences of the behaviors became more severe.

Hypothesized Mechanisms

The association between education and mortality can strengthen for several reasons. First, the returns from education may increase such that specific mechanisms (e.g., smoking) become more strongly linked to education. This possibility is the focus of our study. In addition to the changing returns to education, new mechanisms linking education and mortality may emerge (for instance, the internet and the “digital divide” are one possibility). Alternatively, the content of education may change in ways that make it more relevant for health. And finally, the composition of education groups may change. Specifically, as the average level of educational attainment rose steadily during the twentieth century, individuals who do not graduate from high school today may be more negatively select than in the past. Addressing compositional changes is beyond the scope of our study, but we comment on this possibility in the discussion section.

Part of the difficulty in identifying the mechanisms linking education and mortality risk (or any health outcome) is that they are multifarious at any point in time, they may change over time, and they may vary across demographic subgroups. Indeed, if we consider education as a fundamental cause (Link and Phelan 1995) of mortality disparities, then a search for mechanisms must be tempered by the recognition that, “The persistence of the association over time and its generality across very different places suggests that no fixed set of intervening risk and protective factors can account for the connection” (Link et al. 2008:72). Thus, our task here is particularly challenging and we expect to find relatively small effects of the mechanisms we examine. We examine the three main groups of mechanisms through which education has been consistently linked to health and mortality (Hummer and Lariscy 2011; Ross and Wu 1995): social-psychological factors, economic circumstances, and health behaviors.

Social-Psychological Factors

Higher-educated adults tend to have more social-psychological resources, such as social ties (McPherson, Smith-Lovin, and Brashears 2006) and a sense of control (Mirowsky and Ross 2003), than less-educated adults. For instance, higher-educated adults are more likely to be married than their less-educated peers, and this disparity has grown in recent decades among whites—no such growth was observed among blacks (DiPrete and Buchmann 2006). This trend may have played a role in the widening mortality gap because marriage is the most salient social tie for many adults (Walen and Lachman 2000) and it is linked with lower mortality risk (Umberson 1992; Waite 1995). Furthermore, since the early 1970s, educational homogamy within marriage has increased, particularly at the tails of the education distribution (Schwartz and Mare 2005). The increasing homogamy may have exacerbated health disparities because better-educated adults experience higher marital quality (Cherlin 1992) and the education of both spouses may influence each other's mortality (Montez et al. 2009).

Better-educated individuals are less likely to experience marital, parental, and financial stress, traumatic events such as divorce, assault, and death of a child (Lantz et al. 2005), and depression (Miech and Shanahan 2000). To the extent that the other types of mechanisms that we examine have disproportionately increased exposure to these stressors and events among low-educated women, they may have suffered even greater levels of psychological distress.

Economic Circumstances

Education is associated with a higher likelihood of being employed, avoiding financial hardship, owning a home, and having employment-related health insurance (Ross and Wu 1995). These resources have become progressively more concentrated among higher educated groups as the U.S. labor market has bifurcated in recent decades. For example, employment rates have grown substantially more for women with higher education. From 1960 to 1990, the percentage of women 25 years of age and older employed full-time grew from just 12.0 to 14.7 percent among women without a high school credential while it nearly doubled from 19.7 to 39.0 percent among women with at least five years of college (Spain and Bianchi 1996). The returns from education for labor participation and economic well-being have grown more among women than men (Blank and Shierholz 2006; DiPrete and Buchmann 2006).

Health Behaviors

Better educated adults are more likely to exercise, not smoke, drink alcohol in moderation, and maintain a healthy body weight compared with less-educated adults (Pampel, Krueger, and Denney 2010). They have greater access to health-related information and more quickly integrate it into their life styles. For example, in 1954 when definitive studies linking cigarette smoking and cancer appeared in the media, 95.1 percent of college graduates claimed they had heard the information compared with 81.0 percent of adults with less than a high school credential (Link 2008). While at that time there was little difference in smoking prevalence by education, more educated adults adopted this information faster so that by the 1990s an educational gradient emerged (Link 2008). Smoking trends across the 1908-1967 birth cohorts reveal that while smoking declined among adults with at least a high school credential, the prevalence among adults with 0-11 years of education remained stable among men but increased among women, especially white women (Escobedo and Peddicord 1996). In addition, between 1988-1994 and 2005-2008 educational disparities in obesity increased slightly among women but decreased among men (Ogden et al. 2010).

Aims of this Study

We examine the extent to which the three categories of mechanisms described above account for the widening education gap in all-cause mortality risk during 1997-2006 among non-Hispanic white women 45-84 years of age in the United States. We focus on white women because of their unique mortality trends across education levels during the last few decades. As described above their mortality gap has grown markedly since the mid-1980s; moreover, life expectancy has declined among low-educated white women, in contrast to gains in life expectancy in the population overall. We also focus on white women because the reasons for the growing gap may differ by gender and race/ethnicity, given historical disparities in school quality, employment, immigration patterns, and family structure, for example. We examine the contribution of each category of mechanisms as well as individual mechanisms, and relate our findings to what is currently known about the growing education gap in specific causes of death.

DATA AND METHODS

Data

Data for this study are from the public-use National Health Interview Survey Linked Mortality File (NHIS-LMF) downloaded from the Minnesota Population Center (2012). The NHIS-LMF links adults in the 1986 through 2004 annual cross-sectional waves of the NHIS with death records in the National Death Index through December 31, 2006. The linkage is mainly based on a probabilistic matching algorithm, which correctly classifies the vital status of 98.5 percent of eligible survey records (NCHS 2009). In 1997, the NHIS began annually collecting data on the mechanisms of interest in this study. Thus, we use data from the 1997-2004 NHIS surveys and vital status information through 2006.

Educational Attainment and Time

Our analytic objective is to assess the extent to which the growing gap in mortality risk across education levels can be explained by the hypothesized mechanisms. This entails estimating an education-by-time interaction coefficient in our statistical models (described below) and quantifying how the size and statistical significance of the coefficient attenuate when the mechanisms are included in the models.

Educational Attainment

We dichotomize education level (0=12 years or more, 1=0-11 years). This specification is based on prior research that has found unfavorable mortality trends among white women with 0-11 years of education were the main contributor to the widening mortality gap since the mid-1980s (Montez and Zajacova 2013).1 The percent of non-Hispanic white women 45-84 years of age with 0-11 years of education was 15.7 percent during the first four survey years (1997-2000) and 12.3 percent during the latter four (2001-2004). We refer to women with 0-11 years of education as “low-educated” and other women as “high-educated.”

Time

Time indicates the year of exposure to the risk of death during the 1997-2006 follow-up. Prior studies using the NHIS-LMF have aggregated follow-up years to produce stable mortality rates for race-gender-education subgroups (Montez et al. 2011; Montez and Zajacova 2013). In preliminary analyses we estimated mortality rates across 1997-2006 using continuous and aggregated specifications of time, and then determined which specification best reflected the annual rates. The aggregated term (0=1997-2001, 1=2002-2006) smoothed the annual variation and reflected the overall trend in the annual rates much better than did the continuous term. Thus, we use the aggregated (i.e., dichotomized) specification.

Mortality

The outcome is a dichotomous indicator of whether the individual died from any cause during the 1997-2006 follow-up. Of the 7,189 low-educated women in our sample, 1,161 (16.1 percent) died. Of the 39,555 high-educated women, 2,892 (7.3 percent) died.

Mechanisms

Social-Psychological Factors

We assess two components of social-psychological circumstances with strong ties to mortality: marriage and psychological distress. We include legal marital status (1=married, 0=unmarried) and spouse's education level. Other measures of social ties are not available in the NHIS. Spouse's education is defined as a three-level ordinal variable indicating 0-11 years of education (0), a high school credential or some college (1), or a bachelor's degree or higher (2) and used as a continuous predictor in the analyses. In preliminary analyses, we also examined a categorical measure of spouse's education. The findings were similar so we chose the ordinal (continuous) measure for parsimony. Since unmarried women do not have data on spousal education, we use an internal moderator approach (Mirowsky 1999) and include marital status and the product of marital status and spouse's education.

We measure nonspecific psychological distress using the K6 Scale. The scale has strong psychometric properties and provides a standardized estimate of the prevalence and severity of mental illness in community-based populations (Kessler et al. 2002). It consists of responses to six questions about how often during the last 30 days the respondent felt so sad nothing could cheer them up; nervous; restless or fidgety; hopeless; that everything was an effort; worthless. Response categories range from 0 (none) to 4 (all the time). For each respondent, we replaced missing values with the mean of their provided values and then summed the six responses for a total score between 0 and 24.

Economic Circumstances

Key economic factors that shape mortality risk include employment, occupation, extreme deprivation (e.g., poverty), fluid resources (e.g., income), and accumulated resources (e.g., wealth, home ownership) (Krueger and Burgard 2011). Employment status during the previous week is dichotomized as employed (full-time or part-time) or not employed (unemployed or not in the labor force). In some analyses we include an alternative specification that combines information about occupation among the employed. This specification comprises four dummy variables: not employed (omitted reference), white collar (executive, administrative, and managerial; professional specialty), skilled (technical and related support; sales; administrative support; protective services; military), and manual (private household; service; farming; precision production; operators; transportation; handlers and laborers). Extreme deprivation is measured by an indicator of whether the family income-to-poverty ratio was below 1 during the previous year. In ancillary analyses we assessed four categories of the ratio but found similar results. We measure accumulated material resources with an indicator of home ownership because the NHIS does not collect data on wealth. We also include an indicator of private health insurance coverage at interview to reflect material and non-material resources garnered from past or current employment.

Health Behaviors

The most important behavioral risk factors for mortality include smoking, obesity (reflecting poor diet and physical activity), and heavy alcohol consumption as a distant third (Mokdad et al. 2004). We include smoking as a three-level variable indicating current smokers, former smokers, or individuals who never smoked (omitted reference). We include a binary measure of body mass index to identify whether the respondent was obese (BMI ≥ 30). Alcohol consumption is a four-level variable indicating lifetime abstainers, former drinkers (consumed no alcohol in prior year), light and moderate drinkers who consumed alcohol up to two days per week during the prior year (omitted reference), or heavy drinkers who consumed alcohol three or more days per week during the prior year.

Some women were missing data on one or more mechanisms. Less than 1 percent of women were missing marital status, smoking, employment, home ownership, or health insurance; 1-5 percent were missing psychological distress, occupation, alcohol consumption, spouse's education, or obesity; 25 percent were missing poverty status. We imputed these missing values using IVEware multiple imputation software (Raghunathan, Solenberger, and Van Hoewyk 2002). Table 1 shows the resulting distribution of the mechanisms by education level among non-Hispanic white women who were 45-84 years of age in the 1997-2004 NHIS surveys. The table splits the eight NHIS surveys into two 4-year groups (1997-2000, 2001-2004) to more clearly illustrate how the distributions changed. Low-educated women were disadvantaged on all mechanisms in both time periods. For example, in 1997-2000 they were less likely to be married (47 versus 65 percent), and if they were married they more likely to have a low-educated spouse (57 versus 10 percent) compared with high-educated women. The last column shows that several disadvantages grew over time. For example, the percent of low-educated women who had never smoked decreased from 52 to 50 between the two periods, while the percent among high-educated women increased from 54 to 55, which widened the education gap from 2 to 5 percent. Growing disparities were also pronounced for spouse's education level, employment status, private health insurance, obesity, and for being a former drinker.

Table 1.

Distribution of Hypothesized Mechanisms by Education Level and Survey Year among White Women 45-84 Years, 1997-2004

1997-2000 2001-2004

Low-educatioina High-education Diff. Low-education High-education Diff. Diff in Diff.
Social-Psychological Factors
    Married (%) 47.0 65.0 −18.0 *** 48.1 64.7 −16.6*** 1.4
    Spouse's education (%)
        0-11 years 57.0 9.9 47.1 *** 54.5 7.8 46.7 *** −0.4
        High school or some college 40.3 57.1 −16.8 *** 42.3 57.1 −14.8 *** 2.0
        College 2.7 33.0 −30.3 *** 3.3 35.1 −31.8 *** −1.5
    Psychological distress (0-24) 4.0 2.3 1.7 *** 4.3 2.5 1.8 *** 0.1
Economic Circumstances
    Employed part or full time (%) 18.4 49.8 −31.4 *** 18.9 51.5 −32.6 *** −1.2
    Income below poverty (%) 20.0 6.0 14.0 *** 20.6 6.1 14.5 *** 0.5
    Home ownership (%) 74.9 86.3 −11.4 *** 74.9 87.2 −12.3 *** −0.9
    Private health insurance (%) 61.0 85.0 −24.0 *** 56.2 83.6 −27.4 *** −3.4
    Occupation among employed (%)
        White collar 7.2 40.5 −33.3 *** 8.2 40.2 −32.0 *** 1.3
        Skilled 25.0 41.2 −16.2 *** 26.2 40.7 −14.5 *** 1.7
        Manual 67.8 18.3 49.5 *** 65.6 19.1 46.5 *** −3.0
Health Behaviors
    Smoking (%)
        Current smoker 23.2 17.4 5.8 *** 23.9 16.9 7.0 *** 1.2
        Former smoker 25.0 28.6 −3.6 *** 26.6 28.3 −1.7 *** 1.9
        Never smoked 51.8 53.9 −2.1 *** 49.5 54.9 −5.4 *** −3.3
    Obese (%) 26.1 20.8 5.3 *** 31.0 23.7 7.3*** 2.0
    Alcohol consumption (%)
        Lifetime abstainer 46.6 23.9 22.7 *** 45.2 23.2 22.0 *** −0.7
        Former drinker 26.6 17.8 8.8 *** 28.2 17.3 10.9 *** 2.1
        Current light drinker 23.9 47.4 −23.5 *** 23.4 47.7 −24.3 *** −0.8
        Current heavy drinker 2.8 10.9 −8.1 *** 3.2 11.9 −8.7 *** −0.6
Age (years) 67.2 59.6 7.6 *** 66.9 59.6 7.3 *** −0.3
N 4,081 19,810 3,108 19,745
***

p < .001 (age-adjusted, two-tailed tests)

Notes: Distributions are based on weighted respondent-level records. Age reflects age at interview. “Diff” means difference.

a

Low-education indicates 0-11 years of education and high-education indicates 12 or more years.

Methods

We first built a person-year file that aged all non-Hispanic white women 25-84 years of age at interview by one year beginning with their 1997-2004 NHIS interview until their year of death or 2006 if they survived. The age limit helps ensure that most women had completed their education through a high school credential; it also accounts for the top-coding at 85 years. Next, we retained person-year records for women who contributed person-years during 1997-2006 when they were 45-84 years of age. This allows women to “age-in” and “age-out” of the sample (see Montez et al. 2011). We set the lower limit at 45 because there are few deaths below age 45 in the NHIS-LMF (just 3 percent of deaths among white women occur before age 45 (Anderson 1999)) and because ages 25-44 contributed little to the increasing gap, at least during the 1990s (Meara et al. 2008). We set the upper limit at 84 because mortality matches are not as reliable among women 85 or older (Ingram, Lochner, and Cox 2008) and because the non-institutional sample excludes nursing home residents who are predominately older white women. The final sample contained 46,744 women who contributed 293,608 person-years and 4,053 deaths.

Using the person-year file, we estimated a series of discrete-time event history models using logistic regression. All models include age (time-varying from 45 to 84 years), education, time, and an education-by-time interaction. A positive interaction indicates that the mortality gap widened. We then introduce the mechanisms to assess the extent to which they attenuate the interaction and thus may “explain” the widening mortality gap.

The models were estimated with SAS Version 9.3 using PROC SURVEYLOGISTIC. The models were weighted by the eligibility-adjusted sample weights and accounted for the complex sample design of the NHIS-LMF. The model estimates were then analyzed with PROC MIANALYZE to account for the multiple imputation procedure.

RESULTS

We first estimated the growth of the education gap in mortality during 1997-2006. Model 1 in Table 2 contains coefficients from a logistic regression model predicting the ln(odds) of death from age, time, education, and the education-by-time interaction. During 1997-2001 the odds of death among low-educated white women 45-84 years of age were 1.37 times greater (e0.317=1.37) than the odds among their high-educated peers. During 2002-2006, the odds among low-educated women were 1.66 times greater (e0.317+0.192). The 21 percent growth in the odds [100(e0.192-1)] was significant at p<0.05.

Table 2.

Coefficients Predicting the ln(odds) of Death from Age, Time, Education, and Mechanisms among White Women 45-84 Years, 1997-2006

Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 Model 9 Model 10
Intercept −11.094*** −10.526*** −11.445*** −9.251*** −9.224*** −9.743*** −12.125*** −11.983*** −10.450*** −10.678***
Age 0.089*** 0.084*** 0.092*** 0.073*** 0.072*** 0.072*** 0.094*** 0.097*** 0.079*** 0.081***
Timea −0.050 −0.026 −0.047 0.001 0.002 −0.008 −0.045 −0.056 0.001 −0.014
Low Educationa 0.317*** 0.222** 0.180* 0.150* 0.143 0.277*** 0.152* 0.284*** 0.039 0.244**
Low Education × Time 0.192* 0.184* 0.204* 0.170 0.165 0.170 0.155 0.163 0.139 0.144
        p-value for interaction 0.026 0.034 0.022 0.052 0.057 0.051 0.076 0.062 0.113 0.101
        percent explainedc --- 4.2% −6.2% 11.5% 14.1% 11.5% 19.3% 15.1% 27.6% 25.0%
        percent explainedd --- 5.1% −5.4% 13.9% 16.6% 13.6% 23.0% 17.6% 32.7% 28.9%
Social-Psychological
    Married −0.151**
    Married × spouse ed. −0.230***
    Psychological distress 0.069***
Economic
    Employed −0.769*** −0.817*** −0.784***
    Income below poverty 0.205** 0.205** 0.155*
    Home ownership −0.450*** −0.447*** −0.359***
    Health insurance −0.221*** −0.215*** −0.148**
    Occupation (not emp.)
        White collar −1.038*** −0.890***
        Skilled −0.735*** −0.671***
        Manual −0.568*** −0.563***
Health Behaviors
    Smoking (never)
        Current 1.072*** 1.002*** 0.978*** 0.986***
        Former 0.575*** 0.506*** 0.543*** 0.495***
    Obese 0.126** 0.075
    Alcohol (current light)
        Lifetime abstainer 0.520*** 0.442***
        Former 0.571*** 0.494***
        Current heavy 0.054 0.059
AIC 16,072,146 16,014,289 15,934,455 15,878,928 15,872,142 15,970,985 15,774,522 15,867,515 15,635,271 15,773,523
***

p < .001

**

p < .01

*

p < .05

p < .10 (two-tailed tests)

a

Time is a binary variable (0=1997-2001, 1=2002-2006).

b

Low education indicates 0-11 years of education.

c

Percent of the interaction coefficient in Model 1 explained by the mechanisms using the ln(odds) coefficients in the table.

d

Percent of the interaction coefficient in Model 1 explained by the mechanisms using y-standardized coefficients available on request (Long and Freese 2005).

We next assessed the extent to which the widening gap can be attributed to the selected mechanisms. We did this by examining the degree to which the education-by-time interaction coefficient in Model 1 (b=0.192, p=0.026) attenuated when the mechanisms were statistically accounted for in Models 2-10. For each model, we report the ln(odds) coefficient and p-value, the percent attenuation of the coefficient, and the percent attenuation of its y-standardized form (y-standardized coefficients available on request). Y-standardized coefficients may improve the comparability of logistic regression models, whose unstandardized coefficients reflect the true effects of the predictors and the unobserved heterogeneity in the model (Long and Freese 2005; Mood 2010). The y-standardized coefficients exhibited a somewhat greater degree of attenuation than did the unstandardized coefficients, but the overall findings were substantially comparable. We discuss each model using the ln(odds) coefficients to aid interpretation; we discuss the percent attenuation across models (which is our main interest) using the y-standardized coefficients (hereafter “YSC”) because they may be more appropriate for model comparisons.

Models 2 and 3 examined the two components of social-psychological circumstances. Model 2 offered little support for the marriage component—the interaction attenuated slightly to 0.184 with p=0.034 (5 percent attenuation of YSC). Model 3 gave no support for the distress component. Models 4-6 offered some support for economic circumstances. Accounting for employment status, poverty, home ownership, and health insurance in Model 4 reduced the education-by-time interaction to 0.170 with p=0.052 (14 percent attenuation of YSC). When we disaggregated employment by occupational type in Model 5, the interaction coefficient was reduced to 0.165 with p=0.057 (almost 17 percent attenuation of YSC). Among the economic circumstances, employment was by far the most important component. Indeed, comparing Model 6 (which only contains employment status) with Model 4 (which also includes income, home ownership, and health insurance) reveals that, although the association between employment and mortality partly operated through income, home ownership, and health insurance, these three mechanisms contributed little to the widening gap net of employment. The final group of mechanisms, health behaviors, also received some support. Controlling for smoking, obesity, and alcohol use in Model 7 reduced the education-by-time interaction to 0.155 with p=0.076 (23 percent attenuation of YSC). Smoking was by far the most important behavior. Controlling only for smoking reduced the interaction term to 0.163 with p=0.062 (almost 18 percent attenuation of YSC) in Model 8. It is noteworthy that the interaction was attenuated to a similar extent by smoking and by economic circumstances (in Model 5).

The results summarized so far indicated that diverging economic circumstances and health behaviors contributed to the growing mortality gap. Thus, we included both groups of mechanisms in Model 9. The interaction term (b=0.139, p=0.113) became statistically nonsignificant (33 percent attenuation of YSC). Given that employment and smoking were the most important components, we included only these mechanisms in Model 10. They explained almost as much of the widening gap (29 percent attenuation of YSC) as did all economic circumstances and behaviors combined, and the interaction term remained nonsignificant.2

While accounting for economic circumstances and health behaviors reduced the interaction coefficient to statistical nonsignificance, their contribution to the growth in the gradient (33 percent) was relatively modest in practical terms. In addition, the interaction coefficient in model 9 is not statistically different from the coefficient in model 1. As we stated above, we expected to find relatively modest effects of the mechanisms that we examined given the difficulty in explaining mortality up to 10 years after a single interview and because, according to fundamental cause theory, the contribution of any particular mechanism may vary over time. However, the results can still be used judiciously. One way to assess their validity is to contrast the mediating effects of smoking and obesity against the contribution of smoking-related and obesity-related causes of death to the widening education-mortality gap. Our finding that smoking was an important mechanism is consistent with recent research showing that lung cancer and chronic lower respiratory disease explain a large proportion (25-50 percent) of the growth in the education-mortality gap among white women 45-84 years of age (Meara et al. 2008; Montez and Zajacova 2013). Thus, even a small attenuation of the interaction coefficient in our all-cause mortality models when controlling for a mechanism reported at interview (e.g., smoking) corresponds to a substantial contribution of that mechanism when assessing it using cause of death analyses (e.g., lung cancer). In addition, our finding (confirmed in ancillary models) that obesity contributed little to the growing gap in all-cause mortality is consistent with research showing that diabetes-related mortality explained just 6 percent of the growing education-mortality gap among white women 45-84 years of age (Montez and Zajacova 2013). Smoking and obesity are also related to other causes of death; however, these comparisons provide a good first approximation of the validity of our results.

The finding that employment was the most important dimension of economic well-being raised the possibility that it was not employment per se that was important, but rather that non-employment may indicate inability to work due to poor health. We tested this alternative explanation in Table 3. First, we controlled for self-rated health at interview in Model 2a (1=poor; 0=fair, good, very good, or excellent). Self-rated health had only a small effect on the education-by-time interaction. Next, we added employment status in Model 2b. Net of baseline health, employment attenuated the interaction term from 0.183 (p=0.040) in Model 2a to 0.161 (p=0.070) in Model 2b, which is similar degree of attenuation when employment was included in Model 6 in Table 2. We repeated the analysis using an indicator of whether the respondent reported having no “physical, mental, or emotional problem that kept them from working at a job or business” at interview. We found that any divergence in simply being able to work across education levels had little impact on the growing mortality gap (comparing Model 1 to Model 3a). Moreover, even when controlling for ability to work, employment contributed to the growing gap as shown in Model 3b. These tests suggest that the contribution of employment to diverging mortality across education levels is at least partly due to the health benefits derived from employment.3

Table 3.

Coefficients Predicting the ln(odds) of Death from Age, Time, Education, and Health Status at Interview among White Women 45-84 Years, 1997-2006

Model 1 Model 2a Model 2b Model 3a Model 3b
Intercept −11.094*** −11.150*** −10.080*** −9.974*** −9.215***
Age 0.089*** 0.088*** 0.075*** 0.085*** 0.075***
Timea −0.050 −0.031 0.002 −0.001 0.022
Low Educationb 0.317*** 0.160* 0.137 0.144 0.125
Low Education × Time 0.192* 0.183* 0.161 0.188* 0.171
        p-value for interaction 0.026 0.040 0.070 0.032 0.050
        percent explainedc --- 4.7% 16.1% 2.1% 10.9%
        percent explained --- 5.5% 18.1% 3.7% 13.7%
Poor Self-Rated Health 1.458*** 1.358***
No Work Limitationse −1.142*** −1.063***
Employed −0.672*** −0.548***
AIC 16,072,146 15,784,864 15,718,380 15,671,858 15,629,039
***

p < .001

**p < .01

*

p < .05

p < .10 (two-tailed tests)

a

Time is a binary variable (0=1997-2001, 1=2002-2006).

b

Low education indicates 0-11 years of education.

c

Percent of the interaction coefficient in Model 1 explained by the mechanisms using the ln(odds) coefficients in the table.

d

Percent of the interaction coefficient in Model 1 explained by the mechanisms using y-standardized coefficients available on request (Long and Freese 2005).

e

The NHIS asked all respondents, “Does a physical, mental, or emotional problem now keep you from working at a job or business?” Respondents were classified as either “unable to work,” “able to work but limited in the amount or type of work,” or “able to work without any limitation.”

A Glance at Causes of Death

We replicated the analysis for deaths from heart disease, and from lung cancer or chronic lower respiratory disease (CLRD), to assess whether the results were consistent with etiological differences in these causes of death and thus support the validity of the mechanisms.4 The results (available on request) support the mechanisms’ validity. As expected, the growth in the gap for these causes was larger than the growth for all-cause mortality, but it was only marginally significant (p<0.10) due to small numbers of deaths. The widening gap in heart disease mortality reflected multiple mechanisms—marriage and spousal education, economic well-being, and health behaviors—while the widening gap in lung cancer and CLRD mortality mainly reflected smoking. Smoking explained over three times as much of the widening gap in lung cancer and CLRD mortality (34 percent) than it did for heart disease mortality (10 percent).

DISCUSSION

This study examined three explanations—social-psychological factors, economic circumstances, and health behaviors—for the widening education gap in mortality risk during 1997-2006 among white women 45-84 years of age. Social-psychological factors contributed little to the increasing gap; however, economic circumstances and health behaviors played an important role. Accounting for economic circumstances (employment, occupation, poverty, home ownership, health insurance) and health behaviors (smoking, obesity, alcohol consumption) together explained the growing education gap in mortality to statistical nonsignificance. In practical terms, the contribution of economic circumstances and health behaviors was relatively modest—they explained roughly one-third of the growth in the gap (although they fully explained the main effect of education). Employment status and smoking were the most important components.

The role of employment is intriguing and, to our knowledge, has not been previously examined as a potential explanation of the growing education gap in mortality. While divergence in health-related ability to work may have played some role, our results indicate that employment was, in and of itself, an important contributor. Indeed, studies have found that the better health of employed adults was mostly attributed to employment itself rather than health selection into employment (Graetz 1993). Employment provides both manifest (e.g., income) and latent benefits such as social networks and supports and a sense of purpose, it enhances self-esteem, and it offers mental and physical activity (Creed and Macintyre 2001). Access to social networks and support through employment may have become more important in recent decades with high divorce rates, smaller families, and geographic mobility disrupting other avenues of support.

Future studies should examine employment in detail to better understand its contribution. Like most cross-sectional analyses this study assumes that the mechanisms measured at survey reflect a meaningful degree of exposure to the mechanism during the life course. Thus, it assumes that differences in employment rates at the time of interview reflect differences in the propensity to work during adulthood. This is the same assumption, for example, that women who report being married (or smoking) at the time of survey have a history of being married (or smoking). Nevertheless, more research is needed to elucidate the role of employment. Employment histories could be informative, particularly among older respondents who may be retired at the time of survey.

The findings have several policy implications. One goal should be to increase the high school graduation rate (i.e., receiving a diploma, not a GED), which has stagnated for white women since the mid-1940s birth cohorts (Heckman and LaFontaine 2010). This goal is bolstered by research that finds a substantial drop in mortality risk associated with a high school credential, more so than any other year of education (Montez, Hummer, and Hayward 2012).

Social protection policies are also needed. For example, work-family policies should be (re)designed to allow women who want desirable jobs outside the home to secure them. Women disproportionately head single parent households and care for children and aging parents, all of which create major obstacles to employment outside the home. The obstacles are even higher for low-educated women who tend to be confined to low-paying jobs with little flexibility in hours, little allowance for time off to care for family members, and limited social support outside the home (Heymann 2000). Some national work-family policies do exist; however, they may not help low-educated women. For example, the Family and Medical Leave Act (FMLA) entitles eligible employees of certain employers up to 12 weeks of unpaid but protected leave per year to attend to certain medical and family-related needs (U.S. Department of Labor n.d.). The policy, however, does not pertain to private-sector employers with fewer than 50 employees where low-educated women are disproportionately employed. It may also be financially prohibitive for women who do not have an alternative source of income, yet women with the fewest financial resources and social ties may need a policy like FMLA the most.

Other hurdles that low-educated women disproportionately face in finding and maintaining employment should also be addressed (for examples, see Damaske 2011; Heymann 2000; Moen and Roehling 2005). Increasing the number of good jobs for working class adults could help low-educated women enter the labor force. Maintaining employment could be facilitated by widespread availability of affordable day care. Mandating paid parental leave, paid sick time, and flexible work schedules (where possible) could also help low-educated women maintain employment since the jobs available to them usually lack these benefits.

Continued policy efforts to reduce smoking could also stem the growing gap. However, these efforts must go beyond tobacco control policies oriented at changing behavior (Graham et al. 2006). Qualitative research has shown that, despite knowing the health risks, socioeconomically disadvantaged women state that they smoke to relieve stress from the daily hassles of poverty, single parenting, and conflict-ridden relationships; because they are lonely and feel hopeless; because it provides a rare opportunity to socialize and feel a sense of belonging; and because they often have nothing interesting to do outside the home such as employment or affordable recreational activities (Stewart et al. 2011). These reasons also resonate with our findings that some of these distal correlates of smoking, like employment, played an important role in the growing mortality inequality. As others have argued, a comprehensive policy framework—addressing education, poverty, and housing, for example—is needed to improve the circumstances faced by disadvantaged women (Stewart et al. 2011) because “social policy is tobacco control policy” (Graham et al. 2006:pii11).

In addition to the three categories of well-established mechanisms we examined in this study, other factors may be important. The childhood environment, for instance, has received increasing attention as indelibly shaping adult health (e.g., Duncan, Ziol-Guest, and Kalil 2010; Hayward and Gorman 2004; Kuh and Ben-Shlomo 2004) and other outcomes (e.g., Haas, Glymour, and Berkman 2011; Palloni et al. 2009). Childhood factors might have played a role in the growing mortality gap if they have polarized over time as some research suggests (McLanahan 2004). The NHIS unfortunately does not routinely collect information about childhood conditions. Future research should assess the role of childhood contexts using other surveys that contain measures such as parental socioeconomic status and childhood health.5

Another potentially important factor is compositional changes. As average education levels increased over time, low-educated women may have become a more negatively select group. While compositional changes likely occurred to some degree, empirical tests of their role in explaining diverging trends in well-being and mortality across education levels have generated little support (Blau 1998; Cutler et al. 2011; Martikainen, Blomgren, and Valkonen 2007; Meara et al. 2008). One indication that compositional changes are not the main explanation, at least for the increasing mortality among low-educated women since the mid-1980s, is that mortality among low-educated white men continued to decline (Montez et al. 2011). To glean some insights, in ancillary analyses we controlled for the national high school graduation rate when each respondent was 17 years of age (U.S. Department of Education 1993). Controlling for the rate only slightly narrowed the widening gap (the education-by-time interaction decreased from 0.192 (p=0.026) in the baseline model to 0.185 (p=0.033), suggesting that compositional changes were not the driving force behind the growth in the mortality gap. This test should be interpreted cautiously because other contemporaneous trends that influenced mortality could confound it.

A few limitations of the study should be noted. First, data on mechanisms was collected only at the time of survey. Thus, we did not have information on long-term exposures such as work or marital histories, which may have hindered our ability to detect their full contribution. Also, we did not have information on other potentially important mechanisms such as parity, social networks, sense of control, and childhood environments. Our findings should not be generalized to other demographic groups because their mortality trends and underlying mechanisms may differ. Future studies should use other data to assess trends among women 85 and older, who comprise an increasing share of deaths. While our study period is just 10 years, it reflects almost 50 years of birth cohorts (1913-1961). The diverging trends that we find likely reflect period and cohort effects (see Masters, Hummer, and Powers 2012 for recent evidence of cohort effects in the widening gradient) which we cannot partition with our data. Last, our effect sizes were fairly small so our findings should be interpreted cautiously and validated using other datasets. In particular, the role of employment should be validated using data that contain detailed information on work histories.

Conclusions

The Healthy People initiative aims to improve the U.S. population's health and eliminate health disparities (DHHS 2000). However, disparities in the length of life among white women in the United States continued to grow during the late 1990s and through mid-2000s. This trend does not imply that such initiatives were ineffective—only that the structural forces working against them were stronger. One force appears to be the obstacles that low-educated women face in securing (desirable) employment. Our findings indicate that increasing high school graduation rates and redesigning work-family policies may improve longevity and reduce disparities among U.S. women. These solutions reflect the philosophy of Geoffrey Rose (2008:133) who argued that, “The decisions which most affect the health of the nation are not taken in government departments of health but in those of the environment, employment, education, social security, and (especially) the Treasury.”

Footnotes

1

In preliminary analyses we included more detailed categories of education and similarly found that the widening gap was statistically significant only among the 0-11 year category, using a high school credential as the reference group.

2

In ancillary analyses, we replicated the models in Table 2 to assess whether categorizing GED recipients as low-educated changed the results. We found that the higher risk of death among low-educated women was similar (b=0.317 for low-educated in Model 1 in Table 2 versus b=0.313 in the ancillary models) as was the education-by-time interaction coefficient (b=0.192 in Model 1 in Table 2 versus b=0.208 in the ancillary models). The mechanisms attenuated the interaction coefficient in a similar pattern for both specifications.

3

In ancillary analyses we stratified the models by age at interview to glean additional insights; however, the number of deaths in some age groups was too small to produce robust estimates. Note that while the percentage of women who were employed declined with age, many women remained employed after the traditional retirement age. For example, at age 70, 17 percent of high-educated women and 11 percent of low-educated women were employed.

4

We combined lung cancer and CLRD to increase power and because they share smoking as a major risk factor. Due partly to small numbers of deaths, other leading causes did not exhibit a statistically significant increase in the education-mortality gap.

5

Other studies have used adult height as a proxy for certain aspects of childhood such as nutrition and pathogen exposure (e.g, Bozzoli, Deaton, and Quintana-Domeque 2009; Dowd, Zajacova, and Aiello 2009; Fogel 2004). In ancillary models, we analyzed adult height as a proxy for childhood conditions but it did not attenuate the widening gap in mortality. However, height may be a poor proxy in high-income regions and for recent cohorts who experienced more favorable epidemiological contexts.

Contributor Information

Jennifer Karas Montez, Harvard University.

Anna Zajacova, University of Wyoming.

REFERENCES

  1. Anderson Robert N. United States Life Tables, 1997. National Vital Statistics Report. 28. Vol. 47. National Center for Health Statistics; Hyattsville, MD: 1999. [PubMed] [Google Scholar]
  2. Blank Rebecca M., Shierholz Heidi. Exploring Gender Differences in Employment and Wage Trends among Less-Skilled Workers. 2006 NBER Working Paper 12494. [Google Scholar]
  3. Blau Francine D. Trends in the Well-Being of American Women, 1970-1995. Journal of Economic Literature. 1998;36:112–65. [Google Scholar]
  4. Bozzoli Carlos, Deaton Angus, Quintana-Domeque Climent. Adult Height and Childhood Disease. Demography. 2009;46:647–69. doi: 10.1353/dem.0.0079. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Cherlin Andrew J. Marriage, Divorce, Remarriage. Harvard University Press; 1992. [Google Scholar]
  6. Creed Peter A., Macintyre Sean R. The Relative Effects of Deprivation of the Latent and Manifest Benefits of Employment on the Well-Being of Unemployed People. Journal of Occupational Health Psychology. 2001;6:324–31. [PubMed] [Google Scholar]
  7. Crimmins Eileen M., Saito Yasuhiko. Trends in Healthy Life Expectancy in the United States, 1970-1990: Gender, Racial, and Educational Differences. Social Science & Medicine. 2001;52:1629–41. doi: 10.1016/s0277-9536(00)00273-2. [DOI] [PubMed] [Google Scholar]
  8. Cutler David M., Lange Fabian, Meara Ellen, Richards-Shubik Seth, Ruhme Christopher J. Rising Educational Gradients in Mortality: The Role of Behavioral Risk Factors. Journal of Health Economics. 2011;30:1174–87. doi: 10.1016/j.jhealeco.2011.06.009. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Damaske Sarah. For the Family? How Class and Gender Shape Women's Work. Oxford University Press; New York, NY: 2011. [Google Scholar]
  10. DHHS . Department of Health and Human Services. Healthy People 2010: Understanding and Improving Health. U.S. Government Printing Office; Washington, D.C.: 2000. [Google Scholar]
  11. DiPrete Thomas A., Buchmann Claudia. Gender-Specific Trends in the Value of Education and the Emerging Gender Gap in College Completion. Demography. 2006;43:1–24. doi: 10.1353/dem.2006.0003. [DOI] [PubMed] [Google Scholar]
  12. Dowd Jennifer Beam, Zajacova Anna, Aiello Allison. Early Origins of Health Disparities: Burden of Infection, Health, and Socioeconomic Status in U.S. Children. Social Science & Medicine. 2009;68:699–707. doi: 10.1016/j.socscimed.2008.12.010. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Duncan Greg J., Ziol-Guest Kathleen M., Kalil Ariel. Early-Childhood Poverty and Adult Attainment, Behavior, and Health. Child Development. 2010;81:306–25. doi: 10.1111/j.1467-8624.2009.01396.x. [DOI] [PubMed] [Google Scholar]
  14. Escobedo Luis G., Peddicord John P. Smoking Prevalence in U.S. Birth Cohorts: The Influence of Gender and Education. American Journal of Public Health. 1996;86:231–36. doi: 10.2105/ajph.86.2.231. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Feldman Jacob J., Makuc Diane M., Kleinman Joel C., Cornoni-Huntley Joan. National Trends in Educational Differentials in Mortality. American Journal of Epidemiology. 1989;129:919–33. doi: 10.1093/oxfordjournals.aje.a115225. [DOI] [PubMed] [Google Scholar]
  16. Fogel Robert William. The Escape from Hunger and Premature Death, 1700-2100: Europe, America, and the Third World. Cambridge University Press; 2004. [Google Scholar]
  17. Graetz Brian. Health Consequences of Employment and Unemployment: Longitudinal Evidence for Young Men and Women. Social Science & Medicine. 1993;36:715–24. doi: 10.1016/0277-9536(93)90032-y. [DOI] [PubMed] [Google Scholar]
  18. Graham Hilary, Hazel M Inskip Brian Francis, Harman Juliet. Pathways of Disadvantage and Smoking Careers: Evidence and Policy Implications. Journal of Epidemiology and Community Health. 2006;60:ii7–12. doi: 10.1136/jech.2005.045583. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Haas Steven A., Maria Glymour M, Berkman Lisa F. Childhood Health and Labor Market Inequality over the Life Course. Journal of Health and Social Behavior. 2011;52:298–313. doi: 10.1177/0022146511410431. [DOI] [PubMed] [Google Scholar]
  20. Hayward Mark D., Gorman Bridget K. The Long Arm of Childhood: The Influence of Early-Life Social Conditions on Men's Mortality. Demography. 2004;41:87–107. doi: 10.1353/dem.2004.0005. [DOI] [PubMed] [Google Scholar]
  21. Heckman James J., LaFontaine Paul A. The American High School Graduation Rate: Trends and Levels. The Review of Economics and Statistics. 2010;92:244–62. doi: 10.1162/rest.2010.12366. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Heymann Jody. The Widening Gap: Why America's Working Families Are in Jeopardy and What Can Be Done About It. Basic Books; New York: 2000. [Google Scholar]
  23. Hummer Robert A., Lariscy Joseph T. Educational Attainment and Adult Mortality. In: Rogers RG, Crimmins EM, editors. International Handbook of Adult Mortality. Springer Publishers; 2011. pp. 241–61. [Google Scholar]
  24. Ingram Deborah D., Lochner Kimberly A., Cox Christine S. Mortality Experience of the 1986-2000 National Health Interview Survey Linked Mortality Files Participants. 147. Vol. 2. National Center for Health Statistics. Vital Health Stat; 2008. [PubMed] [Google Scholar]
  25. Jemal Ahmedin, Ward Elizabeth, Anderson Robert N., Murray Taylor, Thun Michael J. Widening of Socioeconomic Inequalities in U.S. Death Rates, 1993-2001. PLoS ONE 3(5):e2181. 2008;3:1–8. doi: 10.1371/journal.pone.0002181. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Kessler Ron C., Andrews G, Cople LJ, Hiripi E, Mroczek DK, Normand S-LT, Walters EE, Zaslavsky AM. Short Screening Scales to Monitor Population Prevalences and Trends in Non-Specific Psychological Distress. Psychological Medicine. 2002;32:959–76. doi: 10.1017/s0033291702006074. [DOI] [PubMed] [Google Scholar]
  27. Krueger Patrick M., Burgard Sarah A. Work, Occupation, Income, and Mortality. In: Rogers RG, Crimmins EM, editors. International Handbook of Adult Mortality. Springer Publishers; 2011. pp. 263–88. [Google Scholar]
  28. Kuh Diana, Ben-Shlomo Yoav. In: A Life Course Approach to Chronic Disease Epidemiology. Second Edition Kuh D, Ben-Shlomo Y, editors. Oxford University Press Inc; New York: 2004. [Google Scholar]
  29. Lantz Paula M., House James S., Mero Richard P., Williams David R. Stress, Life Events, and Socioeconomic Disparities in Health: Results from the American's Changing Lives Study. Journal of Health and Social Behavior. 2005;46:274–88. doi: 10.1177/002214650504600305. [DOI] [PubMed] [Google Scholar]
  30. Lauderdale Diane S. Education and Survival: Birth Cohort, Period, and Age Effects. Demography. 2001;38:551–61. doi: 10.1353/dem.2001.0035. [DOI] [PubMed] [Google Scholar]
  31. Link Bruce G. Epidemiological Sociology and the Social Shaping of Population Health. Journal of Health and Social Behavior. 2008;49:367–84. doi: 10.1177/002214650804900401. [DOI] [PubMed] [Google Scholar]
  32. Link Bruce G., Phelan Jo C. Social Conditions as Fundamental Causes of Disease. Journal of Health and Social Behavior. 1995;(extra issue):80–94. [PubMed] [Google Scholar]
  33. Link Bruce G., Phelan Jo C., Miech Richard, Leckman Westin Emily. The Resources That Matter: Fundamental Social Causes of Health Disparities and the Challenge of Intelligence. Journal of Health and Social Behavior. 2008;49:72–91. doi: 10.1177/002214650804900106. [DOI] [PubMed] [Google Scholar]
  34. Scott Long, J., Freese Jeremy. Regression Models for Categorical Outcomes Using Stata. Second Edition. Stata Press; College Station, TX: 2005. [Google Scholar]
  35. Martikainen Pekka, Blomgren Jenni, Valkonen Tapani. Change in the Total and Independent Effects of Education and Occupational Social Class on Mortality: Analyses of All Finnish Men and Women in the Period 1971-2000. Journal of Epidemiology and Community Health. 2007;61:499–505. doi: 10.1136/jech.2006.049940. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Masters Ryan K., Hummer Robert A., Powers Daniel A. Educational Differences in U.S. Adult Mortality: A Cohort Perspective. American Sociological Review. 2012;77:548–72. doi: 10.1177/0003122412451019. [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. McLanahan Sara. Diverging Destinies: How Children Are Faring under the Second Demographic Transition. Demography. 2004;41:607–27. doi: 10.1353/dem.2004.0033. [DOI] [PubMed] [Google Scholar]
  38. McPherson Miller, Smith-Lovin Lynn, Brashears Matthew E. Social Isolation in America: Changes in Core Discussion Networks over Two Decades. American Sociological Review. 2006;71:353–75. [Google Scholar]
  39. Meara Ellen R., Richards Seth, Cutler David M. The Gap Gets Bigger: Changes in Mortality and Life Expectancy, by Education, 1981-2000. Health Affairs. 2008;27:350–60. doi: 10.1377/hlthaff.27.2.350. [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Miech Richard Allen, Shanahan Michael J. Socioeconomic Status and Depression over the Life Course. Journal of Health and Social Behavior. 2000;41:162–76. [Google Scholar]
  41. Miech Richard, Fred Pampel Jinyoung Kim, Rogers Richard. The Enduring Association of Education and Mortality: The Role of Emergent and Growing Disparities. American Sociological Review. 2011;76:913–34. doi: 10.1177/0003122411411276. [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Minnesota Population Center . Minnesota Population Center and State Health Access Data Assistance Center, Integrated Health Interview Series: Version 5.0. University of Minnesota; Minneapolis: 2012. http://www.ihis.us. [Google Scholar]
  43. Mirowsky John. Analyzing Associations between Mental Health and Social Circumstances. In: Aneshensel C, Phelan J, editors. Handbook of the Sociology of Mental Health. Kluwer Academic/Plenum; New York: 1999. pp. 105–26. [Google Scholar]
  44. Mirowsky John, Ross Catherine E. Education, Social Status, and Health. Aldine de Gruyter; New York: 2003. [Google Scholar]
  45. Moen Phyllis, Roehling Patricia. The Career Mystique: Cracks in the American Dream. Rowman & Littlefield Publishers, Inc; Lanham, MD: 2005. [Google Scholar]
  46. Mokdad Ali H., Marks James S., Stroup Donna F., Gerberding Julie L. Actual Causes of Death in the United States, 2000. JAMA. 2004;291:1238–45. doi: 10.1001/jama.291.10.1238. [DOI] [PubMed] [Google Scholar]
  47. Montez Jennifer Karas, Hayward Mark D., Brown Dustin C., Hummer. Robert A. Why Is the Educational Gradient in Mortality Steeper for Men? Journal of Gerontology: Social Sciences. 2009;64:625–34. doi: 10.1093/geronb/gbp013. [DOI] [PMC free article] [PubMed] [Google Scholar]
  48. Montez Jennifer Karas, Hummer Robert A., Hayward Mark D. Educational Attainment and Adult Mortality in the United States: A Systematic Assessment of Functional Form. Demography. 2012;49:315–36. doi: 10.1007/s13524-011-0082-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  49. Montez Jennifer Karas, Hummer Robert A., Hayward Mark D., Woo Hyeyoung, Rogers Richard G. Trends in the Educational Gradient of U.S. Adult Mortality from 1986 through 2006 by Race, Gender, and Age Group. Research on Aging. 2011;33:145–71. doi: 10.1177/0164027510392388. [DOI] [PMC free article] [PubMed] [Google Scholar]
  50. Montez Jennifer Karas, Zajacova Anna. Trends in Mortality Risk by Education Level and Cause of Death among U.S. White Women from 1986 to 2006. American Journal of Public Health e-View Ahead of Print. 2013 doi: 10.2105/AJPH.2012.301128. (doi: 10.2105/AJPH.2012.301128) [DOI] [PMC free article] [PubMed] [Google Scholar]
  51. Mood Carina. Logistic Regression: Why We Cannot Do What We Think We Can Do, and What We Can Do About It. European Sociological Review. 2010;26:67–82. [Google Scholar]
  52. NCHS . The National Health Interview Survey (1986-2004) Linked Mortality Files, Mortality Follow-up through 2006: Matching Methodology. Hyattsville, MD: 2009. [Google Scholar]
  53. Ogden Cynthia L., Lamb Molly M., Carroll Margaret D., Flegal Katherine M. Obesity and Socioeconomic Status in Adults: United States 1988-1994 and 2005-2008. NCHS data brief no. 50. National Center for Health Statistics; Hyattsville, MD: 2010. [PubMed] [Google Scholar]
  54. Palloni Alberto, Milesi Carolina, White Robert G, Turner Alyn. Early Childhood Health, Reproduction of Economic Inequalities and the Persistence of Health and Mortality Differentials. Social Science & Medicine. 2009;68:1574–82. doi: 10.1016/j.socscimed.2009.02.009. [DOI] [PMC free article] [PubMed] [Google Scholar]
  55. Pampel Fred C., Krueger Patrick M., Denney Justin T. Socioeconomic Disparities in Health Behaviors. American Review of Sociology. 2010;36:349–70. doi: 10.1146/annurev.soc.012809.102529. [DOI] [PMC free article] [PubMed] [Google Scholar]
  56. Pappas Gregory, Susan Queen Wilbur Hadden, Fisher Gail. The Increasing Disparity in Mortality between Socioeconomic Groups in the United States, 1960 and 1986. New England Journal of Medicine. 1993;329:103–09. doi: 10.1056/NEJM199307083290207. [DOI] [PubMed] [Google Scholar]
  57. Preston Samuel H., Elo Irma T. Are Educational Differentials in Adult Mortality Increasing in the United States? Journal of Aging and Health. 1995;7:476–96. doi: 10.1177/089826439500700402. [DOI] [PubMed] [Google Scholar]
  58. Raghunathan TE, Solenberger Peter W., Van Hoewyk John. Iveware: Imputation and Variance Estimation Software Users Guide. University of Michigan; Ann Arbor, MI: 2002. [Google Scholar]
  59. Rogot Eugene, Sorlie Paul D., Johnson Norman J. Life Expectancy by Employment Status, Income, and Education in the National Longitudinal Mortality Study. Public Health Reports. 1992;107:457–61. [PMC free article] [PubMed] [Google Scholar]
  60. Rose Geoffrey. Rose's Strategy of Preventive Medicine. Oxford University Press, Inc; New York: 2008. [Google Scholar]
  61. Ross Catherine E., Wu Chia-Ling. The Links between Education and Health. American Sociological Review. 1995;60:719–45. [Google Scholar]
  62. Schwartz Christine R., Mare Robert D. Trends in Educational Assortative Marriage from 1940 to 2003. Demography. 2005;42:621–46. doi: 10.1353/dem.2005.0036. [DOI] [PubMed] [Google Scholar]
  63. Spain Daphne, Bianchi Suzanne M. Balancing Act: Motherhood, Marriage, and Employment among American Women. Russell Sage Foundation; New York: 1996. [Google Scholar]
  64. Stewart Miriam J., Lorraine Greaves Kaysi Eastlick Kushner, Letourneau Nicole L., Spitzer Denise L., Boscoe Madeline. Where There Is Smoke, There Is Stress: Low-Income Women Identify Support Needs and Preferences for Smoking Reduction. Health Care for Women International. 2011;32:359–83. doi: 10.1080/07399332.2010.530724. [DOI] [PubMed] [Google Scholar]
  65. U.S. Department of Education 120 Years of American Education: A Statistical Portrait. National Center for Education Statistics. 1993 [Google Scholar]
  66. U.S. Department of Labor [March 1, 2012];The Family and Medical Leave Act. n.d. ( http://www.dol.gov/whd/fmla/)
  67. Umberson Debra. Gender, Marital Status and the Social Control of Health Behavior. Social Science & Medicine. 1992;34:907–17. doi: 10.1016/0277-9536(92)90259-s. [DOI] [PubMed] [Google Scholar]
  68. Waite Linda J. Does Marriage Matter? Demography. 1995;32:483–507. [PubMed] [Google Scholar]
  69. Walen Heather R., Lachman Margie E. Social Support and Strain from Partner, Family, and Friends: Costs and Benefits for Men and Women in Adulthood. Journal of Social and Personal Relationships. 2000;17:5–30. [Google Scholar]

RESOURCES