Employment-Based Health Insurance, Illness, and Labor Supply of Women:
Evidence from Married Women with Breast Cancer
Cathy J. Bradley, David Neumark, Heather L. Bednarek, and Zhehui Luo*
* Bradley is Associate Professor, Department of Medicine, Michigan State University, East
Lansing, MI. Phone: (517) 432-3405 Fax: (517) 432-9471 email: Cathy.Bradley@ht.msu.edu.
Neumark is Senior Fellow, Public Policy Institute of California, San Francisco, CA., Research
Associate, National Bureau of Economic Research, Research Fellow, IZA, and Visiting Scholar,
Institute of Business and Economic Research, UC-Berkeley. Phone: (415) 291-4476 Fax: (415)
291-4428 email: firstname.lastname@example.org. Bednarek is Assistant Professor, Department of Economics,
St. Louis University, St. Louis, MO. Phone: (314) 977-3847 Fax: (314) 977-1478 email:
email@example.com. Luo is Research Associate, Department of Epidemiology, Michigan State
University, East Lansing, MI. Phone: (517) 432-2227 Fax: (517) 432-9471 email:
firstname.lastname@example.org. Bradley, Neumark, and Luo’s research was supported by an NCI grant, RO1-
CA86045-01A1. The views expressed are not those of the Public Policy Institute of California.
This paper was prepared for the “Health and Gender” session sponsored by CSWEP at the 2005
1 See, e.g., Currie and Madrian (1999).
An extensive literature has examined the relationship between health insurance and the labor
market.1 Absent from this research, however, is an examination of how the availability of health
insurance and whether it is contingent on employment influences labor supply after an adverse health
shock. The health shock creates a need for convalescence, but with employment-based health
insurance an individual may need to continue working in order to maintain the insurance. If so, then
employment-based health insurance may create a tension between recovery and work.
Because many women obtain insurance through their spouse’s employment, while many
others obtain it through their own employment, the labor supply response of women to illness—and
how that response depends on the source of the health insurance—provides a natural context for
studying this question. We collected data with which to compare the effects of breast cancer on labor
supply of married women 6 months following diagnosis and the beginning of treatment, depending on
whether health insurance comes through the spouse’s employment or the woman’s own employment.
We include in our analyses a control group of women constructed from respondents to the Current
Population Survey (CPS). The control group is essential to avoid confounding the effects of cancer
with life-cycle changes as well as typical employment dynamics, and how these might vary with the
source of health insurance.
The relationship between health insurance and labor supply has been studied with regard to
job lock (Cooper and Monheit, 1993; Gruber and Madrian, 1994; Kapur, 1998) and, to a lesser extent,
married women’s labor supply (Buchmueller and Valletta, 1999; Chou and Staiger, 2001). But
studying how an unexpected adverse health shock alters labor supply when the employee is
dependent on an employer for health insurance is more difficult, because chronic diseases are less
prevalent in the working age population and few health shocks are amenable to prospective,
longitudinal studies. Cancer—a disease with rising (diagnosed) prevalence among working age
individuals due to sensitive screening techniques applied to younger people—is an exception.
Because the incidence and severity of disease is recorded in cancer registries, patients can be
identified soon after diagnosis, contacted, and followed over time. Patients are also generally healthy,
without symptoms, when they are diagnosed with early stage cancers.
Previous research found that breast cancer and its treatment have a negative effect on
women’s labor supply 6 months following diagnosis (Bradley et al., in press[a]). Women with breast
cancer, with the exception of those having in situ cancer, were about 17 percentage points less likely
to work relative to a control sample. Women with breast cancer who remained working worked
fewer hours than women in the control group. Here, we extend this work to study differences in the
labor supply response to cancer between married women who had health insurance through their
employer and those who had health insurance through their spouse.
The theoretical framework for our analysis is based on models in Becker (1964) and
Grossman (1972), which suggest that poor health decreases labor supply by diminishing tastes for
work, raising the marginal value of leisure time, and increasing time required for health maintenance.
These implications are challenged when an ill employee has insurance through an employer. Under
these circumstances, an employee may forego health maintenance and devote time towards work in
order to maintain health insurance coverage.
Women newly diagnosed with breast cancer were identified, shortly after diagnosis, from the
Metropolitan Detroit Cancer Surveillance System (MDCSS), a population-based registry that covers
over 4 million people within the Detroit Metropolitan area. MDCSS is a participant in the National
Cancer Institute’s Surveillance, Epidemiology, and End Results (SEER) program, and as such is held
to high standards of completeness.
Study eligibility criteria were age range of 30 to 64, English-speaking, and either employed
or with an employed spouse at the time of diagnosis. The lower age bound was chosen because breast
cancer occurs more frequently after age 30 and the upper age bound was chosen to select women
prior to the traditional age of retirement. We enrolled employed women or non-employed women
2 These other categories were quite small.
with an employed spouse because we were interested in how breast cancer affected working
individuals or families with at least one working individual.
The procedures for subject identification and recruitment are explained in Bradley et al. (in
press[b]). Of the contacted women, the response rate for subjects screened and determined to be
eligible was 83%, and 444 employed women were enrolled. For the analysis in this paper, we
selected from this sample married women who were employed in the period just before diagnosis
with cancer, and who were either insured through their own employer or through their spouse’s
employer. Women insured through a private policy, uninsured, or with coverage through their own
and spouse’s employer were excluded from the sample to keep the analysis as straightforward as
The interview was conducted in two parts: a set of questions on labor market participation 3
months prior to diagnosis (period 1); and a nearly identical set of questions on labor market
participation as close as possible to 6 months after diagnosis (period 2). In the retrospective portion
of the interview, breast cancer subjects recalled their labor supply (employment and weekly hours)
approximately 9 months prior to the interview.3
The retrospective data covered the period March 2001 through January 2002. We therefore
constructed a control group from respondents to the CPS who participated in the March 2001
supplement, which contains questions regarding health insurance coverage. We selected CPS March
supplement respondents in their 4th month (“month-in-sample” (MIS) 4) and in MIS 5, which
occurred 9 months after MIS 4, to be consistent with the 9-month span between the cancer subjects’
pre- and post-diagnosis information. Approximately 80% of the respondents had a matching record
from one survey to the next. We chose respondents to the CPS who resided in Midwest Primary
Metropolitan Statistical Areas (PMSAs) and Metropolitan Statistical Areas (MSAs) in the Great
3 Recall over a period of this length appears relatively reliable, although with some tendency to overstate past
hours (Duncan and Hill, 1985).
4 In some analyses breast cancer will also be specified categorically to represent in situ, local, regional, and
Lakes region, and otherwise applied the same inclusion criteria to the CPS sample as to the cancer
III. Empirical Approach
The study outcomes were: the probability of employment and weekly hours worked. These
outcomes are modeled as functions of the incidence and stage of breast cancer (BCA),4 employer-
based health insurance prior to diagnosis (EHI), control variables (X), and unobserved influences (ε).
Women who obtain health insurance through their employer may systematically differ from women
who obtain their health insurance through their spouse, in terms of both characteristics and labor
supply behavior, yet experience similar health effects of cancer and its treatment. Thus, by stratifying
the sample by health insurance source, we observe how women with the two different sources of
health insurance alter their labor supply in response to cancer.
We estimate the probability of employment (E) 6 months following diagnosis or at MIS 5
(period 2) for women employed as of period 1, using
Pr(Ei2=1| EHIi1=1,Ei1=1,BCAi,Xi,εi2). (1)
Pr(Ei2=1| EHIi1=0,Ei1=1,BCAi,Xi,εi2). (2)
We define employment status as a binary variable (Ei2) that equals one if a woman reports
that she was employed in period 2 and zero otherwise, and estimate these equations as probit models,
with the estimates translated into derivatives of the probability of working with respect to the
We assume that the same variables that affect employment also affect weekly hours worked
(H), and estimate two different models for hours—one that conditions on employment post-diagnosis,
and one that does not. We also report results from a model for the percent change in hours worked.
distant stages. These four stages are summary stages indicating progression in metastases.
Columns 1 and 2 of Table 1 compare the cancer sample to the CPS sample. Women with
breast cancer were older, less educated, more likely to have health insurance through their spouse,
and had fewer young children. Turning to labor supply, women with breast cancer were less likely to
be employed, worked fewer hours in period 2 relative to women in the CPS sample, and reduced their
hours from period 1 to period 2. When we stratify the sample by health insurance source (columns 3
and 4), women with insurance through their employer were less educated and had fewer children
under age 18, but were more likely to be working and to work more hours at period 2 relative to
women who had health insurance through a spouse, and reduced their hours by less. The key
question, though, is how the labor supply responses to cancer vary with the health insurance source.
The differences between columns 1 and 2 imply that we need to account for differences between the
cancer and control samples to answer this question.
Table 2 shows that, although at 6 months following diagnosis the effects of treatment are still
likely to be present, especially for more advanced cancers, breast cancer does not statistically
significantly influence the probability of employment of women whose health insurance is through
their employer (columns 1 and 2). In contrast, women with breast cancer and who had health
insurance through a spouse were 15 percentage points less likely to be employed relative to the
corresponding control sample (column 3). A statistically significant effect was not observed for
women with in situ or local stage cancer, but a much stronger and significant negative effect (38
percentage points) was observed for women with regional or distant cancer (column 4).
Table 3 reports the change in weekly hours worked, unconditional and conditional on
employment in period 2. The unconditional estimates also reflect employment effects, whereas the
conditional estimates isolate hours effects for those working in both periods. Breast cancer is
associated with a decrease in weekly hours worked for both groups of women. However, women
diagnosed with breast cancer who had health insurance through their spouse decreased weekly hours
worked by greater than did women with health insurance through their employer (for example, as
5 It is possible that the demand for health insurance will prevent those who are employed full-time from shifting
shown in the lower panel, a 30% reduction compared to a 16% reduction in hours worked conditional
The results from our analysis of health, health insurance, and labor supply are striking. A
negative health shock, captured in a diagnosis of breast cancer, decreased labor supply to a greater
extent among women insured by their spouse’s policy than among women with health insurance
through their employer. This finding was present in all models estimated. The negative effects of
cancer were greater for women with advanced stage diseases—suggesting that even women who
required aggressive treatment6 were sensitive to employer-based health insurance. Employer-based
health insurance appears to be an incentive to remain working and to work at a greater intensity when
faced with a serious illness. The health implications of this apparent consequence of employment-
based health insurance are yet to be measured.
into part-time employment. To account for this possibility, we estimated a multinomial logit model where the
outcomes are full-time employment (working 35 or more hours per week), part-time employment (working
fewer than 35 hours per week), and non-employed. We found that women with health insurance through their
employer were less likely to shift to part-time employment. Also, in unreported regressions we estimated the
specifications included in Tables 2 and 3 controlling for husband characteristics (age, education, hours worked,
union membership); the results were qualitatively similar.
6 All women with breast cancer received treatment. The majority of the women received surgery followed by
radiation and chemotherapy. However, those with more advanced stages were more likely to have a
mastectomy instead of a lumpectomy.
Becker, Gary. Human Capital. New York: Columbia University Press for NBER, 1964.
Bradley, Cathy; Neumark, David, Bednarek, Heather and Schenk, Maryjean. “Short-term Effects of
Breast Cancer on Labor Market Attachment: Results from a Longitudinal Study.” Journal of Health
Economics, in press[a].
Bradley, Cathy; Neumark, David, Oberst, Kathleen, Brennan, Simone, Luo, Zhehui, and Schenk,
Maryjean. “Combining Registry, Primary and Secondary Data Sources to Identify the Impact of
Cancer on Labor Market Participation.” Medical Decision Making, in press[b].
Buchmueller, Thomas and Valletta, Robert. “The Effect of Health Insurance on Married Female
Labor Supply.” Journal of Human Resources, Summer 1999, 34(1), pp. 42–70.
Chou, Y.J. and Staiger, Douglas. “Health Insurance and Female Labor Supply in Taiwan.” Journal of
Health Economics, March 2002, 20(2), pp. 187-211.
Cooper, Phillip and Alan Monheit. “Does Employment-Related Health Insurance Inhibit Job
Mobility?” Inquiry, Winter 1993, 30(4), pp. 400-16.
Currie, Janet and Madrian, Brigitte. “Health, Health Insurance and the Labor Market,” in O.
Ashenfelter and D. Card, eds., Handbook of Labor Economics, Volume 3. Amsterdam: Elsevier
Science, 1999, pp. 3309-416.
Duncan, Gregory and Hill, Daniel. “An Investigation of the Extent and Consequences of
Measurement Error in Labor-Economic Survey Data.” Journal of Labor Economics, October 1985,
3(4), pp. 508-32.
Grossman, Michael. “On the Concept of Health Capital and the Demand for Health.” Journal of
Political Economy, March-April 1972, 80(2), pp. 223–55.
Gruber, Jonathan and Madrian, Brigitte. “Health Insurance and Job Mobility: The Effects of Public
Policy on Job-Lock.” Industrial and Labor Relations Review, October 1994, 48(1), pp. 86-102.
Kapur, Kanika. “The Impact of Health on Job Mobility: A Measure of Job-Lock.” Industrial and
Labor Relations Review, January 1998, 51(2), pp. 282–98.
Table 1. Descriptive statistics for breast cancer and CPS sample
In situ or local stage
Regional, distant, unknown
White, Hispanic, non-black
No high school diploma
High school diploma
Children under age 18
Employed at 2nd interview
Mean hours worked per week
Mean hours worked per week
2nd interview (employed
The sample is restricted to married women, employed in period 1, with insurance through either their own
employer or their spouse’s employer. Standard deviations of continuous variables are reported in
parentheses. Statistically significant differences between columns 1 and 2, or 3 and 4, are denoted by *
(p<.10), ** (p<.05), and *** (p<.01).
Table 2. Probability of employment 6-months following diagnosis/MIS 5 by insurance source,
Employer insurance (n=145)
Independent variables (1)
Breast cancer -0.02
In situ/local stage N/A
Spouse insurance (n=187)
Regional/unknown stage N/A N/A
Partial derivatives of probability with respect to independent variables are reported with standard errors in parentheses.
Coefficients for age, race, education, and number of children under age 18 are not shown. Statistically significant
estimates are denoted by * (p<.10), ** (p<.05), and *** (p<.01).
Table 3. Hours change models, conditional and unconditional models
employment in 2nd period
Independent variables (1)
Breast cancer -4.41*
In situ or local stage N/A
Percent change Change
Regional or unknown stage N/A N/AN/A
employment in 2nd period
In situ or local stage -3.57***
Regional or unknown stage N/A N/A N/AN/A
See notes to Table 2.