HEALTH INSURANCE AND JOB MOBILITY:

THE EFFECTS OF PUBLIC POLICY ON JOB-LOCK

JONATHAN GRUBER and BRIGITTE G. MADRIAN*

The authors study a policy of limited insurance portability that has

been adopted by a number of states and the federal government over the
past 20 years. They find that these "continuation ofcoverage" mandates,
which grant individuals the right to continue purchasing health insur-
ance through their former employers for a specified period after leaving
their jobs, are associated with a significant increase in the job mobility
of prime age male workers. This finding suggests that "job-lock"—lack
of mobility out of jobs that offer health insurance—arises in large part
from short-run concerns over portability rather than from long-run
problems.

T he predominant feature of private

health insurance in the United States
today is its link with employment. Work-
important to ask how public policy can
alleviate job-lock.
Three possible strategies to alleviate or
place pooling of risks has several advan- eliminate job-lock are to remove the link
tages—it allows for economies of scale in between health insurance coverage and the
administration, reduces the problem of employment relationship (by, for example,
adverse selection as long as workplaces are establishing universal coverage); maintain
chosen independently of health status, and employment-based insurance but ensure
allows individuals to take advantage of the full portability of coverage across jobs; or
non-taxation of employee fringe benefits— continue the provision ofcoverage through
but it also has an important disadvan- employers, but forbid the exclusion from
tage: if wages do not perfectly offset coverage of pre-existing medical conditions
differences in the valuation of health and conditions discovered in a mandatory
insurance across differentjobs, individu- initial physical examination (two practices
als may not changejobs even when new
employment opportunities with higher
match-specific productivity arise—a phe-
nomenon known as "job-lock." Recent 'Jonathan Gruber is an Assistant Professor of Eco-
nomics at the Massachusetts Institute of Technology
research has demonstrated the empirical and Brigitte C. Madrian is an Assistant Professor of
significance of this phenomenon. For Economics at Harvard University. The authors thank
example, Madrian (1994) estimated in- John Abowd, Maria Hanratty, Olivia Mitchell, and
surance-induced reductions in mobility participants at the Cornell Seminar on Health Insur-
of approximately 25%. Given a distor- ance and the Labor Market for helpful comments,
and Aaron Yelowitz for his help in creating the SIPP
tion of this magnitude, it is obviously extracts.

Industrial and Labor Relations Review, Vol. 48, No. 1 (October 1994). ) by Cornell University.
0019-7939/94/4801 $01.00

86
 EFFECTS OF PUBLIC POLICY ON JOB-LOCK 87

that have become increasingly widespread). benefits consulting firm Foster Higgins re-
We argue, however, that implementing any ports that 57% of employers excluded pre-
of these relatively comprehensive solutions existing conditions in their health plans in
may be complex and costly. 1987 (Cotton 1991). Furthermore, almost
In this study, we examine instead a par- half of full-time workers in firms of 250
tial corrective that has already been used employees or more face a length of service
for two decades: "continuation of cover- requirement before being eligible for any
age" mandates, which require employers to health insurance coverage (Bureau of La-
continue providing health insurance cov- bor Statistics 1989). Although the length
erage to workers who leave the firm for a of service requirement is generally short, it
specified period of time. In recent re- may be important for the employee antici-
search, we have found that these mandates pating major medical expenses.^
have large effects on the retirement behav- In addition, for small firms with experi-
ior of older workers (Gruber and Madrian ence-rated insurance premiums, hiring a
1993a, 1993b). Our goal in this paper is to sick employee can entail large increases in
determine whether they have also been insurance costs. As a result, employees with
effective in alleviating job-lock. If they medical problems may face discrimination
have, they may offer a way to reduce job- when they seek a new job with coverage.
lock and still retain the advantages of em- Even if there is not explicit discrimination
ployer-provided health insurance coverage. in hiring, many insurance policies at small
We are able to study the effects of con- firms are now medically underwritten so
tinuation ofcoverage mandates because of that sick employees may not be able to
the substantial variation in their timing qualify for coverage. Furthermore, only
and generosity. Continuation ofcoverage 60% of firms with fewer than 100 employ-
laws were passed at different times and in ees even offer insurance, so a newjob with
different states from the mid-1970s through insurance coverage is far from guaranteed
the mid-1980s before being federally man- for the potentialjob changer (Gruber, forth-
dated in 1986. In addition, some laws were coming).
quite liberal, allowing 15-20 months ofcov- The theory of compensating differen-
erage, whereas under others the extent of tials predicts that in a perfectly competitive
coverage was minimal. We use data from labor market, differences in the generosity
the Survey of Income and Program Partici- of benefits across firms will be offset by
pation (SIPP) to study the effects of these differing wage levels or other aspects of
laws on job mobility and subsequent insur- compensation. As Rosen (1986) noted,
ance coverage. however, the compensating differential for
a given health insurance package will equate
Health Insurance and Job Mobility the wage reduction to the valuation of that
The primary problem with workplace
pooling of risks is that a workplace is not for which a prudent person would have sought care
necessarily chosen by workers indepen- even if no physician was actually consulted. An insur-
dently of their health status. This fact has ance company may also require all employees to
undergo medical examination, which it then uses to
led many insurers to exclude pre-existing exclude certain medical conditions on an individual
conditions from coverage for a period after basis for the life of the contract. This practice is
an individual is hired, typically between six known as medical underwriting.
months and two years.' The New York ^Pre-existing conditions exclusions and length-of-
service requirements are likely to pose a greater de-
terrent to job change as individuals get older and the
'A pre-existing condition is generally defined as any incidence of chronic medical conditions increases.
medical problem that has been treated or diagnosed Our analysis controls for age effects, however, and we
within the past six months to two years. In some cases consider only those age 54 and younger. See Gruber
it may be more broadly defined as any medical prob- and Madrian (1993a, 1993b) for a discussion of the
lem for which an individual has ever received care. It health problems of older workers and the effect of
may also be extended to include medical conditions continuation coverage on retirement.
88 INDUSTRIAL AND LABOR RELATIONS REVIEW

package by the marginal worker who is just high cost of medical care and the substan-
indifferent between wages and benefits. In tial variation across individuals in health
a heterogeneous population, some workers status. Individual medical expenditures
will value health insurance more highly among those aged 20-54 averaged $1200 in
than the marginal worker; these "infra- 1990, with a standard deviation of $3878.*
marginal" workers will receive a surplus Large differences in the productivity of an
from their excess valuation of the benefit. employee across different firms may there-
That is, these workers will value total com- fore not be sufficient to induce mobility if
petisation at their current job at above the other firms do not provide health insur-
sum of their wages plus the cost to the ance or if the insurance excludes pre-exist-
employer of providing the benefit. As a ing conditions or is medically underwrit-
result, these workers may turn down alter- ten.
native employment opportunities in which Second, there are very large differences
their wages and total productivity would be across firms in the cost of providing health
higher but their health insurance inferior insurance, particularly by firm size. The
to that in the current job. cost of insurance is estimated to be 35%
The potential market failure in this case lower in the largest (10,000-1- employee)
arises from the lack of a complete market firms than in smaller (1-4 employee) firms
for worker/job-specific compensation pack- (Congressional Research Service 1988).
ages. If firms could accurately determine This difference will make it difficult for
worker valuation of benefits, and if they small firms to offer insurance, even if they
could offer each individual a customized know that they could attract much more
wage and benefits package, then workers productive workers by doing so."*
would move to jobs in which their produc- The welfare consequences of health in-
tivity was highest. surance-induced job-lock are unclear
Offering worker-specific compensation (Madrian 1994). Many studies have found
packages is generally not feasible, however, large increases in wages for workers who
for at least three reasons. First, such pack- switch to newjobs (Minter 1992; Bartel and
ages would violate a number of rules de- Borjas 1977; Mincer 1986; Topel 1986).
signed to guard against discrimination in These higher wages may represent a gain to
the design of benefit plans. The Internal society from the increased productivity of
Revenue Code gives favorable tax treat- an improved job match. On the other
ment to employer expenditures on health hand, it may be that somejobs offer rents to
insurance only if all workers are provided workers, and individuals tend to move into
with an equivalent benefits package. Sec- these jobs as their careers advance. Fur-
ond, even if worker-specific compensation thermore, a key consideration is the tem-
were allowed, problems with eliciting true poral nature of the job-lock. Job-lock aris-
worker valuation of these benefits would ing from chronic illnesses could have ma-
make it difficult to design. Finally, the costs jor and long-lasting consequences for work-
involved in administering such packages
might be large enough to absorb most of
the rents that the worker would earn from 'Authors' calculations using data from the 1980
their existence. National Medical Care LItilization and Expenditure
Although this market failure will exist Survey (inflated to 1990 dollars).
••One other interesting point about health insur-
for any job amenity, such as workplace ance, relative to other job amenities such as work-
safety, pensions, or parking, there are two place safety, is that it does not, by definition, need to
reasons why it may be more important for be tied to the employment relationship. This fact
health insurance than for other job ameni- allows the use of a range of policy instruments, such
as regional or national pooling of risks, to alleviate
ties. First, the variance in employees' valu- job-lock of this type. As noted in the introduction,
ation of health insurance is, in general, however, our discussion focuses on policy options
much greater than the variance in their that retain the link between health insurance and
valuation of other benefits, due to both the employment.
 EFFECTS OF PUBLIC POLICY ON JOB-LOCK 89

ers' job productivity, whereas job-lock comparable coverage on each job, and if
linked to short-term conditions such as preg- coverage is less generous in alternative
nancy may be less important from a life- employment, the problem of job-lock re-
cycle perspective. mains.
Finally, it can be argued that job-lock Perhaps the least disruptive option would
allows firms to reap the returns from gen- be to continue having individuals receive
eral human capital investments by reduc- insurance from their current employer but
ing the ability of workers to use those in- to forbid both the exclusion from coverage
vestments elsewhere. However, job-lock of pre-existing conditions and the growing
seems a particularly inefficient mechanism practice of medical underwriting. Those
for achieving this goal, relative to back- prohibitions would assure individuals of
loaded compensation devices such as pen- coverage when they changejobs. Unfortu-
sions, since presumably it is not just the nately, they might also make it infeasible
sickest workers that the firm wishes to re- for many small firms to offer insurance,
tain. since one very sick employee with suffi-
ciently high medical expenses could drive
the firm out of business. Thus, with no
Possible Solutions to Job-Lock other government action, many small firms
A solution to job-lock that suggests itself might cease to offer health insurance at all.
on first consideration of the problem is to Even if all firms were required to provide
divorce health insurance coverage from the coverage, mechanisms would be needed to
employment relationship. Pooling could reinsure small firms against these major
occur, instead, along other dimensions, shocks to their risk pools. Thus, effective
such as regionally (as proposed in Dia- implementation of this policy option may
mond 1992) or nationally (as with Cana- reqtiire much more than simple insurance
dian National Health Insurance). As un- market regulation.
derscored by the current debate about al- The key point is that the design of public
ternative national health care plans, how- policies to alleviate labor market distor-
ever, policies that change the fundamental tions without radical change to our health
nature of health insurance coverage in the care system is an important and difficult
United States, where 83% of all private issue. In considering various alternatives,
insurance is provided through the work- it is useful to refiect on our experience with
place (Gruber, forthcoming), have enor- the major policy intervention to date that
mous transition costs that may make them addresses this problem: the limited port-
infeasible. ability of health insurance provided by "con-
Another alternative is to maintain em- tinuation ofcoverage" mandates.
ployment-based insurance but to ensure
portability of coverage acrossjobs. A policy Alleviating Job-Lock Through
of full insurance portability, however, raises Continuation Benefits
a number of tricky design issues. For ex-
ample, will employees who changejobs fre- State and federal continuation ofcover-
quently be covered by their first employer, age laws require that employers sponsoring
each employer in succession, or the em- group health insurance plans offer termi-
ployer of their choice? Each of these policy nating employees and their families the
options has difficulties. Making the initial right to continue their health insurance
employer responsible for all future insur- coverage through the employer's plan for a
ance would lead to large distortions in ini- specified period of time. Although indi-
tial hiring decisions and massive adminis- viduals must pay the full average cost of
trative costs in tracking employees through their group insurance, the price may be
their job changes. Assigning responsibility well below that of a policy purchased in the
to each subsequent employer would make individual market, especially for individu-
it difficult to guarantee individuals fully als with high medical expenditures. Thus,
90 INDUSTRIAL AND LABOR RELATIONS REVIEW

Table 1. State Continuation of Coverage Laws. or widowed spouses and their families. The
first such law was implemented by Minne-
Effective Months of
Slate Date Coverage
sota in 1974. More than 20 states passed
similar laws over the next decade before
Arkansas 7/20/79 4 the federal government, as part of its 1985
California 1/1/85 3 Consolidated Omnibus Budget Reconcilia-
Colorado 7/1/86 3
tion Act (COBRA), mandated such cover-
Connecticut 10/1/75 10
1/1/87 20 age at the national level.
Georgia 7/1/86 3 The various state statutes are summa-
Illinois 1/1/84 6 rized in Table 1.^ The length ofcoverage is
8/23/85 9 generally quite short, from three to six
Iowa 7/1/87 9
Kansas 1/1/78 6 months, although ten states mandate cov-
Kentucky 7/15/80 9 erage of nine months or more. Most state
Minnesota 8/1/74 6 laws stipulate that an employee must have
3/19/83 12 been covered by an employer's insurance
6/1/87 18 for three to six months before being eli-
Missouri 9/28/85 9
Nevada 1/1/88 18 gible for continuation coverage. The state
New Hampshire 8/22/81 10 laws also apply only to firms that actually
New Mexico 7/1/83 6 purchase insurance through an insurance
New York 1/1/86 6 company;self-insuredfirms, under the 1974
North Carolina 1/1/82 3
North Dakota 7/1/83 10
Employee Retirement Income and Security
Oklahoma 1/1/76 1 Act (ERISA), are not subject to these (or
Oregon 1/1/82 6 any other) state mandates.
Rhode Island //88 18 Although similar in spirit, the state and
South Carolina 1/1/79 2 federal laws differ in a number of impor-
1/1/90 6
South Dakota 7/1/84 3 tant ways. First, the length of coverage
3/3/88 18 mandated under the federal law, 18 months,
Tennessee 1/1/81 3 equals or exceeds that mandated by all but
Texas 1/1/81 6 one state (as of January 1987, Connecticut
Utah 7/1/86 2
Vermont 5/14/86 6 law has provided for up to 20 months of
Virginia 4/17/86 3 coverage).' Second, there is no minimal
Wisconsin 5/14/80 18 length of time for which an employee must
Sources: Hewitt (1985), Thompson Publishing be covered under an employer's plan be-
Group (1992), and state statutes. fore being eligible for continuation ben-
efits. Third, the federal law applies to self-
insured firms, which are exempt from the
continuation benefits provide a way for state laws, as well as to those who purchase
workers who leave their jobs to maintain their coverage from insurers. The federal
their health insurance coverage for at least law, however, does not apply to small firms
a few months, when the alternative would employing fewer than 20 workers. Finally,
most likely be to go uninsured.
Continuation ofcoverage laws generally
apply to all separations (except those due '•Details on state laws are from Hewitt (1985) and
Thompson Publishing (1992) and have been cross-
to an employee's gross misconduct), al- checked against the actual statutes. Table 1 lists only
though in some states benefits are restricted those states with laws that apply to employees who
to those who leave their jobs involuntarily.^ terminate their employment voluntarily.
They often also provide benefits to divorced 'Eighteen months is the maximum length ofcov-
erage available following the voluntary or involuntary
termination of employment. COBRA also provides
up to 36 months ofcoverage for family members who
^We consider only states with laws that cover both would otherwise lose their insurance coverage through
voluntary and involuntary separations, since we have events such as an employee's death, divorce from the
only very noisy data on the nature of the separation. employee, or the employee's eligibility for Medicare.
 EFFECTS OF PUBLIC POLICY ON JOB-LOCK 91

employees of religious organizations and the coverage. At the federal level, full cost
the federal government were originally ex- is defined specifically as 102% of the aver-
empt from COBRA, although federal em- age employer cost of providing coverage.
ployees have subsequently been included The coverage must be identical to that pro-
(beginning in 1990). When the specific vided to similarly situated active employ-
details of the state and federal statutes are ees, including the option to continue en-
at odds, firm provision of continuation ben- rollment in supplemental insurance plans
efits is governed by the law that provides for (such as for vision or dental care) if these
more generous coverage. are available. Although the continuation
An important feature of continuation cost is substantially more than an individual
coverage from our perspective is the inter- generally had to pay as an employee, it may
action between continuation coverage from still be much more attractive than purchas-
an old job and insurance coverage from a ing individual insurance, due to the econo-
new job. Before 1990, once individuals mies of scale in administering group insur-
who were continuing their health insur- ance and the reduced potential for adverse
ance benefits from their old jobs obtained selection with large employee groups.
new employment that offered health insur- In Massachusetts, the average costof fam-
ance, the former employers could drop ily health insurance coverage per employee
their continuation benefits, regardless of in 1989 was $3882.^ When inflated by the
whether they were covered by the new medical care component of the Consumer
policy. This is an important provision, Price Index, this cost is equivalent to $5047
because moving to a firm that offers insur- in 1993 dollars. In contrast, a New England
ance does not guarantee coverage if the commercial insurance company is offering
firm excludes pre-existing conditions from a family policy for a 40-year-old man with a
its policy, or if it has a length of service wife and two children, with a 1—year exclu-
requirement. After 1990, individuals were sion of pre-existing conditions, for $7000.
allowed to continue their COBRA coverage Thus, an individual can realize non-trivial
even if they started a newjob with health financial savings by purchasing continua-
insurance, as long as they continued to pay tion benefits rather than individual cover-
the required premiums. This change im- age.
proved the portability features of COBRA, Furthermore, for less healthy persons or
since it could now be used to bridge peri- those with high family medical expenses,
ods of non-coverage due to unemployment, individual coverage will be even less attrac-
length of service requirements, or pre-ex- tive, for two reasons. First, individual cov-
isting conditions exclusions.* erage is often medically underwritten, so
The effective dates of the state laws are that it may be much more expensive than
listed in Table I. The federal coverage the figure given above, or not available at
mandated under COBRA was phased in. all. Second, individual coverage is typically
Beginning in July 1986, firms had to offer much less generous than group coverage.
continuation benefits at the start of their For example, the individual policy cited
next plan year. For workers who were pro- above excludes pre-existing conditions for
vided health insurance under union con- one year, whereas such conditions would
tracts, such benefits did not have to be be covered from the start for someone con-
offered until the next contract negotiation tinuing a group policy. Table 2 compares
after January 1987. the health insurance benefits of individuals
Both the state and federal laws stipulate covered under group and nongroup poli-
that the employee must pay the full cost of cies in 1977. In every category, those cov-
ered under nongroup policies receive more
^Whether this was much of a change in practice is
unclear, since it is not obvious how the old employer
could verify that the employee had obtained new ^Authors' calculation using unpublished data from
employment that offered health insurance coverage. the Health Insurance Association of America.
92 INDUSTRIAL AND LABOR RELATIONS REVIEW

Table 2. Group and Nongroup Health Table 3. Sources of Health

Insurance Benefits. Insurance Coverage.
Fraction of Job Job
Individuals with All Stayers Leavers
Specified Benefits
Employer-Provided
Non- Any 66.1% 72.0% 45.1%
Croup group Own name 57.5 64.2 33.9
Description of Benefits Plans Plans Union 3.5 3.6 3.1
Other Group Policy 1.2 1.0 1.9
A. Primary Benefits Non-Group Policy 6.1 6.2 5.6
Major medical coverage 86.9% 39.1% GHAMPUS 2.5 2.1 4.0
Hospital room and board 98.4 91.4 Medicaid 1.1 0.6 2.9
Surgery 97.6 91.6
Physician office visit 87.9 40.4 Authors' calculations using data from the 1987
B. Other Benefits National Medical Expenditure Survey.
Ambulance 89.0 54.0
Outpatient diagnostic services 95.9 66.0
Prescribed medicines 87.3 30.3
Mental health 92.2 66.0 age is in their own name. Individuals who
C.Generosity of Benefits
leave jobs are much less likely to have had
(Conditional on Having Benefit) insurance coverage on those jobs. This
Major medical deductible < $100 94.3 61.6 pattern could be construed as evidence of
Full semi-private room charge 77.8 38.2 job-lock; on the other hand, it may just be
80-100% of UCR surgical charge 70.6 60.0 that individuals are more likely to leave
80-100% of LICR physician charge 91.8 81.3 "bad" jobs that do not offer health insur-
Source: Farley (1986), Tables 45-58. ance.
An important indication of the potential
effectiveness of continuation mandates is
limited benefits. Relative to those with the extent to which eligible persons avail
nongroup coverage, those with group poli- themselves of these benefits. Using data
cies are more than twice as likely to receive from a company that administers COBRA
major medical coverage or coverage for claims, Elynn (1992) estimated that 19.3%
physician office visits and prescription of individuals who were terminated, were
drugs, and more than 50% more likely to laid off, or quit from firms that had health
receive ambulance, mental health, and insurance chose to take up continuation
outpatient diagnostic service coverage. benefits."* As Klerman and Rahman (1992)
Furthermore, nongroup policies generally noted, this estimate represents a lower
feature both higher deductibles and higher bound on the take-up rate ofbenefits among
copayments. the eligible population, since many of these
Thus, whereas for a healthy person con- persons will move to newjobs with health
tinuation coverage may provide only mod- insurance or are covered by their spouses;
est savings, for persons with high expected these are individuals we would not expect
medical expenditures the difference in costs to take up continuation benefits. Data
may be quite sizable. Since this population from the 1987 National Medical Expendi-
is exactly the one for which job-lock is ture Survey (NMES) suggest that about two-
potentially the largest problem, continua- thirds of those who leave jobs with health
tion mandates may have some effect in insurance move to jobs in which they also
alleviating this distortion. receive health insurance." Klerman and
Table 3 shows the sources of health in-
surance coverage among 20-54-year-old '"This figure excludes retirees. Take-up among
men from the 1987 National Medical Ex- retirees is likely to be substantially higher than that
among individuals still working (see Gruber and
penditure Survey. Sixty-six percent of per- Madrian 1993b).
sons in this age group have employer-pro- "Authors' calculation. This figure actually repre-
vided health insurance; most of that cover- sents an upper bound on the fraction who move to
 EFFECTS OF PUBLIC POLICY ON JOB-LOCK 93

Rahman estimate thatan additional 10% of efits may not seem a reasonable option
job leavers who do not find jobs with health financially. Thus, by examining the effects
insurance are covered by their spouse's of continuation mandates on mobility, we
health insurance. hope to be able to draw some lessons about
Based on this information, we would ex- the nature of the job-lock problem itself.
pect approximately 30% of those who leave
jobs with health insurance to be affected by Data and Regression Framework
COBRA. This figure implies COBRA take-
up rates among those for whom we would The data we use come from the 1984,
expect it to matter of 65%. It therefore 1985, 1986, and 1987 panels of the Survey
appears that a sizable fraction of eligible of Income and Program Participation
(SIPP). This nationally representative sur-
job changers avail themselves of continua- vey collected information on the economic
tion benefits, despite the high out-of-pocket and demographic characteristics of indi-
costs. viduals and their families through a series
Should we expect continuation mandates of quarterly interviews, referred to as
to alleviate job-lock? The answer depends "waves," for roughly 2-1/2 years.''' Because
on the nature of this distortion. Ifjob-lock the panels overlap, as many as three panels
arises largely due to short-run medical con- may be interviewed concurrently. Alto-
siderations, such as pregnancy, then these gether, these four panels span the period
policies may be quite effective. On the from June 1983 to April 1989.'^
other hand, if individuals are worried about
long-run coverage, then even 18 months of During each interview, respondents were
asked questions concerning up to two jobs
continuation benefits may not be suffi- held during the previous four calendar
cient.'^ Furthermore, ifjob-lock arises from months, including questions about the in-
fear of moving to jobs with length of service dustry, occupation, hours worked, and pay
requirements or pre-existing conditions of each job, as well as the start or end date
exclusions, then temporary coverage may for each job if it was held for less than the
be all that is necessary;" but if fear of being full quarter. With this information, we are
medically underwritten out of coverage is able determine whether an individual
more influential, then more permanent changed jobs or ceased to be employed
portability is required. Finally, ifjob-lock during the quarter.
primarily occurs among individuals with
Our sample is restricted to men between
very high expected medical costs, then con- the ages of 20 and 54 who were not in
tinuing to pay the cost of group health school over the course of their participa-
insurance will not be a major deterrent to tion in the SIPP and who were not self-
the use of continuation benefits, but ifjob- employed.'^ We exclude individuals from
lock represents risk aversion on the part of
the average person, then continuation ben-
jobs with health insurance, as the NMES does not '"The 1984 Panel of the SIPP, which has nine
waves, covers three years. The 1985 Panel consists of
distinguish between those who receive health insur- eight waves (32 months), and the 1986 and 1987
ance through their current employer and those who Panels both include seven waves.
receive COBRA benefits through a former employer. '^Although we would like to be able to study the 14
'^Prior evidence on this question is mixed. Madrian state continuation ofcoverage laws that were passed
(1994) found job-lock arising from both pregnancy, before 1983, the period covered by the data provides
which is a short-run expenditure, and larger families, a window around the passage of the federal law and
which gives rise to longer-run expenditures. Holtz- includes 19 state laws that took effect after June 1983.
Eakin (1994) found statistically significant evidence '^We exclude 55-64-year-olds because the effects
of job-lock only over short periods of observation, of continuation mandates on retirement behavior
however; job changes over a period of more than one may be quite different from their effects on job mobil-
year were unaffected by the provision of health insur- ity. We focus on the older group of workers in Gruber
ance. and Madrian (1993a, 1993b), finding sizable effects
"Note, however, that in this case continuation of continuation mandates on retirement. We exclude
benefits in the period after 1990 would be much more women because the process determining their job
effective. mobility may be quite different from that for men.
94 INDUSTRIAL AND LABOR RELATIONS REVIEW

Hawaii, which has mandated health insur- job by the usual hours worked per week
ance for all employees, and West Virginia, multiplied by the number of weeks in the
for which we were unable to definitively quarter for which an individual was em-
date the continuation mandate. Individu- ployed in that job. The wage rate com-
als from several other small states are also puted in this manner is highly correlated
excluded because, out of concern for con- with the hourly wage rate reported by those
fidentiality, the SIPP has grouped these who are paid hourly.
states together, making it impossible to Each individual is assigned the maxi-
assign the appropriate state laws to indi- mum months of continuation coverage
viduals in these states." The final sample available at the beginning of the wave un-
consists of 155,151 quarterly observations der either COBRA or the appropriate state
on 29,841 individuals. law. Because the federal law was phased in
We use these data to estimate a model of between July 1986 and June 1987, we in-
quarterly job turnover. Turnover is de- crease the months of coverage available
fined as changing employers, becoming self- under COBRA by 1.5 months for each
employed, or becoming unemployed.'* month over this period. By June 1987,
Overall, the quarterly turnover rate for our therefore, months of continuation cover-
sample is 10%, and 24% of each panel age equals 18 for all individuals except
changes jobs within the first year. This those who live in Connecticut, where state
figure is similar to the annual turnover rate law allows for 20 months ofcoverage.
of 22.6% that is estimated from the 1987 Because continuation coverage is only
National Medical Expenditure Survey. It available for those with employer-provided
should be noted that we are not able to health insurance, an important variable in
separate voluntary from involuntary job our analysis is whether or not an individual
changes, because the 1984 and 1985 panels is covered by such insurance. In each wave,
of the SIPP did not ask about the reason for the SIPP asks if an individual is covered by
leaving one'sjob. Although job-lock really health insurance, if this coverage comes
applies only to voluntary turnover, we do from a policy in one's own name, and, if so,
not think that this problem seriously biases if this policy is provided by an employer or
our results. Over the time period that we union. In quarters in which individuals do
consider, late 1983 to early 1989, the not change jobs, we code them as having
economy was growing and the majority of employer-provided health insurance if they
turnover taking place was voluntary.'* Fur- report being covered by health insurance
thermore, Madrian (1994) found that her provided by an employer or union. In
estimates of job-lock were not sensitive to quarters in which individuals change jobs
whether or not her job change variable and report coverage from employer-pro-
included those who changed jobs involun- vided health insurance, it is not clear from
tarily. the SIPP questions whether the employer-
To calculate the wages associated with provided health insurance was attached to
the job that an individual left, we divide the the first job, the second job, or both. We
quarterly wage and salary earnings of the code these individuals' coverage by em-
ployer-provided health insurance based on
whether or not they report having such
"These states are Alaska, Idaho, Iowa, Maine, coverage in the previous wave of the SIPP.^"
Mississippi, Montana, New Mexico, North Dakota,
South Dakota, Vermont, and Wyoming. The means for our data sample are pre-
'*A1 though our definition of turnover includes the
switch from employment to self-employment, it does
not include moving from one self-employed job to ^"This definition of coverage by employer-pro-
another, since individuals are excluded from the vided health insurance will still present problems in
sample when they are self-employed. the case of individuals who change jobs in two con-
"'Three-quarters ofjob changes in the 1987 Na- secutive quarters. Of the job changes in our sample,
tional Medical Expenditure Survey were voluntary 15% are immediately preceded by another quarter
(see Madrian 1994). with ajob change, so only 1.5% of our sample are
 EFFECTS OF PUBLIC POLICY ON JOB-LOCK 95

sented in Table 4. Each observation is a Table 4. Descriptive Statistics.

quarter in which a member of our sample
Standard
does any work.'^' Eleven percent of the Variable Mean Deviation
sample is non-white, and the average edu-
cational level is approximately one year of Non-white 0.113 0.316
college. Seventy-seven percent of the Years of School Completed 12.94 3.021
sample holds employer-provided health Experience 16.42 10.044
Log Hourly Wage $2.20 0.619
insurance. As noted above, the quarterly Holds Employer-Provided
turnover rate is 10%. Health Insurance 0.772 • 0.420
The model that underlies the analysis is Leave Job 0.101 0.301
presented in detail in Madrian (1994). Months ofcontinuation
Coverage 6.88 7.257
Different individuals value health insur-
ance at different levels, which leads to job- Authors' calculations using data on men aged 20-
54 frortf the 1984-87 panels of the Survey of Income
lock through the mechanism described and Program Participation. The sample consists of
above. We assume that these valuations are 155,151 quarterly observations on 29,841 individuals.
distributed randomly in the population we
are studying.^^ We can examine the effec-
tiveness of continuation benefits in allevi-
ating job-lock through a probit regression the number of months ofcontinuation cover-
of the form age available in state j at time I. The regres-
sion models the probability that an individual
(1) Pr (Leave Job) = leaves his job in a particular quarter.
+ Pj • State. + pj • Time^ + • Law.), We control for a number of characteris-
tics of individuals (education, experience,
where i indexes individuals, j indexes and race) and their jobs (log hourly wage,
states, and t indexes time. X..^ is a set of eight industry dummies, six occupation
individual demographic and jobb characteris- dummies, and a dummy for the provision
tics. State, is a set of state dummies, Time^ is a of health insurance). The year dummies
set of year and month dummies,^' and Law.^ is control for any national time trends in
mobility, and the month dummies control
consecutive job changers. Our conclusions about the for seasonal mobility patterns. The state
mobility and insurance coverage effects of the man- dummies control for any time-invariant dif-
dates are unchanged when we exclude this group ferences across states that may be corre-
from the analysis.
•^'Thus, there are multiple observations on each lated with their propensity to legislate con-
individual. Accounting for the resultant intra-per- tinuation mandates, such as differences in
sonal correlation in the error term has little effect on the underlying valuation of health insur-
the restilts. There is also a potential left-censoring ance among the population. The regressor
problem that may lead to the type of "dynamic sample- of interest measures the number of months
selection" bias noted by Diamond and Hausman
(1984). Individuals who left their jobs in response to ofcontinuation benefits for which workers
the availability of continuation benefits, butwhowere are eligible if they leave their current jobs.
still unemployed one quarter later, will be excluded If continuation ofcoverage mandates alle-
from our sample. In a bivariate setting, this exclusion viate job-lock, this regressor should have a
would impart a downward bias to our estimated ef-
fect; in a multivariate setting, the effect is unclear. positive coefficient.
We tested for the importance of this bias in our study
of the effect of continuation mandates on retirement
(Gruber and Madrian 1993a) andfotindittobesmall. Effects on Mobility
^'•^Ideally, one would account for these individual
differences in valuation and use them to help identify
any potential effects of continuation benefits. We Basic Results
discuss one attempt to do so below. The basic mobility results are presented
^'The month dummies are actually dummies for
the month in which the quarter begins. The year in Table 5. The covariates have their ex-
dummies are dummies for the year in which the panel pected effects. Well-educated and older
begins. workers are less likely than other workers to
96 INDUSTRIAL AND LABOR RELATIONS REVIEW

Table 5. Effect of Continuation Coverage The coefficient on the months of con-

on Job Turnover. tinuation coverage variable in the third
column is highly significant. In this table
Independent Variable (I) (2)
and the remaining tables, the probability
A. Coefficient Estimates derivative for this coefficient is presented
Non-White .0118 -.0087 -.0087 in the last row. It indicates that one year of
(.0142) (.0144) (.0144) continuation benefits increases the mobil-
Education -.0130 -.0072 -.0071 ity rate for individuals in our sample by 0.9
(.0019) (.0020) (.0020) percentage points. This effect represents
Experience -.0127 -.0106 -.0106 approximately 9% of the baseline quarter-
(.0005) (.0005) (.0005) to-quarter mobility rate for the sample.
Log Hourly -.2645 -.1327 -.1338 Although small in absolute magnitude, this
Wage (.0078) (.0084) (.0084)
—
effect is quite sizable relative to Madrian's
Health -.5546 >-.5542
Insurance (.0108) (.0108) (1994) 25% estimate ofjob-Iock; it implies
Months of — — .0046 that a year of continuation coverage re-
Coverage (.0011) duces job-lock by almost 40%.^'*
Log-Likelihood -47,098 -45,794 -45,785
N 155,151 155,151 155,151 Specification Checks
B. Marginal Effect — .0087
of 12 Months of [9.1%] The key identifying assumption behind
Continuation our model is that the passage of these man-
Coverage [% of dates represents an exogenous change in
Baseline Mobility the opportunity set of the worker making a
Rate] mobility decision. An alternative hypoth-
Notes: The table gives estimates from a probit esis is that the laws themselves are endog-
equation for whether or not an individual changed enous responses to the pattern of mobility
jobs in a quarter using data from the 1984-87 Survey
of Income and Program Participation. The sample is across states. For example, if more indi-
comprised of men aged 20-54. Coefficients for panel, viduals are changing (or losing) jobs, states
state, month, industry, and occupation dummies are may respond by mandating benefits that
not reported. cover individuals during this transition. To
the extent that the state mandates are re-
sponses to long-standing mobility differ-
ences across states, the state fixed effects
change jobs, and workers are much less included in the regression will control for
likely to leave jobs with high wages or health this endogeneity. To the extent that legis-
insurance than jobs with neither of those latures are responding to changes in mobil-
features. The negative coefficient on health ity, however, state fixed effects will not be
insurance in the second and third columns sufficient controls.
of Table 5 cannot be taken as causal evi-
dence ofjob-Iock, since there are a number
of other reasons why individuals may be
should note that Madrian (1994) produced
reluctant to depart from the "good jobs" three estimates ofjob-Iock corresponding to three
that provide health insurance (Madrian different populations: individuals who had a spouse
1994). Further evidence for this point is with health insurance, those with families of a certain
provided by the fact that the wage coeffi- size, and those who had a pregnant spouse. The 25%
cient in the first column is substantially estimate ofjob-Iock cited here is the estimate for
individuals who had a spouse with health insurance.
reduced when health insurance coverage is This is the right estimate to consider in the context of
added to the regression.^'' this paper, because it is based on only the availability
of health insurance, rather than (as the other two
estimates are) expected medical expenditures. Con-
(1994) also found that the wage coeffi- tinuation coverage increases the availability of health
cient fell substantially when health insurance was insurance to all individuals regardless of their family
added as a regressor. Mitchell (1982) found a similar size or other factors affecting the need for health
result for pensions. insurance.
 EFFECTS OF PUBLIC POLICY ON JOB-LOCK 97

There is a natural test for this alternative Table 6. Effect of Continuation Coverage
hypothesis. Because only those workers on Job Turnover.
who have health insurance on the job are
Have No
eligible to receive continuation benefits, Health Health
workers without health insurance on the Independent Variable Insurance Insurance
job provide a control group for assessing
the effects of continuation mandates. If A. Coefficient Estimates
these mandates are simply correlated with, Non-White .0038 -.0238
or due to, exogenous changes in mobility (.0195) (.0216)
Education -.0105 -.0026
propensities, then the laws should be corre- (.0026) (.0030)
lated with the mobility of workers both with Experience -.0103 -.0103
and without health insurance on thejob. If (.0007) (.0008)
the laws are causing changes in mobility Log Hourly Wage -.1770 -.0862
patterns, however, they should only affect (.0118) (.0121)
workers with employer-provided health in- Months of Coverage .0059 .0021
surance. (.0014) (.0018)
Table 6 therefore divides the sample into Log-Likelihood -27,180 -18,469
two groups consisting of those with and N 119,700 35,451
those without employer-provided health B. Marginal Effect of 12 .0084 .0075
insurance on their currentjobs.^* For those Months of Continuation [14.3%] [3.4%]
Coverage [% of Baseline
with health insurance (column 1), the ef- Mobility Rate]
fect ofcontinuation coverage is even greater
than that estimated for the full sample, and Notes: See notes to Table 5.
it is more significant as well. The coeffi-
cient estimate implies that one year ofcon-
tinuation coverage leads to an increase in As noted earlier, tbe effect of continua-
the likelihood of changingjobs of .84 per- tion mandates will be mitigated by factors
centage points; this is 14% of the baseline sucb as tbe extent to which firms self-insure
mobility rate for this group. (since self-insured firms are exempt from
The second column of Table 6 looks at state mandates) and tbe distribution of
those without health insurance. In fact, the workers by firm size (since small firms are
mandates positively affect mobility for this exempt from the federal mandate). In
group as well. This finding suggests tbat order to appropriately measure the effect
tbe passage of the continuation regulations of continuation coverage, we need to ac-
was correlated witb other factors causing count for variation in tbese factors across
increases in mobility in tbese states during states. We do so by constructing "corrected"
the years under study. Tbe implied mobil- months of coverage variables. Tbe Appen-
ity increase for tbose without health insur- dix details tbe calculation of tbese correc-
ance, however, is not significant, and it tion factors, one for tbose who bave em-
represents only 3.4% of tbe baseline mobil- ployer-provided bealtb insurance tbat ad-
ity rate for tbis group. Furthermore, if we justs for tbe factorsjust described, and one
use tbe estimated effect of tbese mandates for all individuals tbat also adjusts for tbe
on mobility for tbose without healtb insur- likelihood of being insured.
ance to control for secular cbanges in mo- Briefly, we use unpublisbed data from
bility, tbe evidence in Table 6, consistent tbe Healtb Insurance Association of
witb tbat in Table 5, suggests that continu- America (HIAA) and tbe Current Popula-
ation mandates increase mobility by about tion Survey to estimate, by state, tbe frac-
10% (tbat is, 14% minus 3.4%). tion of workers actually offered employer-
provided bealtb insurance, tbe fraction in
that this split is imperfect due to potential
self-insured firms, and tbe fraction in small
errors in assigning insurance coverage to a given job (< 20 employee) firms. If only a state law is
for persons who changejobs in consecutive quarters. in effect, we adjust tbe montbs ofcoverage
98 INDUSTRIAL AND LABOR RELATIONS REVIEW

Table 7. Effect of Continuation Coverage a noisy correction factor that is less than
on Job Turnover (Corrected). one. T'he estimates in columns 1 and 2 of
Table 7 suggest that one year of coverage
Have
Fult Health
increases mobility by between 12% and
Independent Variable Sampte Insurance 15%."
Finally, one potential problem with this
A. Coefficient Estimates model that was discussed above is that the
Non-White -.0088 .0038 variable of interest is measured with con-
(.0144) (.0195) siderable noise. In particular, we are not
Education -.0072 -.0105 focusing on the individuals who most value
(.0020) (.0026)
Experience -.0106 -.0103
their health insurance benefits and for
(.0005) (.0007) whom the limited portability of continua-
Log Hourly Wage -.1339 -.1769 tion mandates may be the most useful. One
(.0084) (.0118) such group for whom these mandates
Health Insurance -.5542 — should be important is married workers,
(.0108) whose higher family medical expenditures
Months of Coverage .0063 .0063 may increase their valuation of health in-
(.0016) (.0017) surance. It is true that married workers are
Log-Likelihood -45,785 -27,181 more likely than single workers to have
N=Sample Size 155,151 119,700 coverage from an alternative source (namely
B. Marginal Effect of 12 .0122 .0091 their spouse), and thus potentially less likely
Months of Continuation [12.7%] [15.3%] to need continuation benefits. Madrian
Coverage [% of Baseline
Mobility Rate] (1994) estimated that 33% of married men
had coverage through a spouse's employ-
Noles: See notes to Table 5. ment in 1987. But family health care ex-
penditures for married men are three times
greater than those for single men.^* Thus,
on net, being married would appear to
to reflect the fact that the state law applies raise the costs of leaving a job with health
only to those offered insurance and only to insurance.^'
those who do not work in a firm that self- In results not reported, we repeated the
insures. If only a federal law is in effect, we regressions of Tables 5-7 for only married
adjust the months of coverage by the frac- individuals and found slightly larger, al-
tion of individuals who are offered insur- though less precisely estimated, effects. For
ance and the fraction who work in small the basic specification, there is a .0085 per-
firms. When both a state law and a federal centage point increase in mobility, which is
law are in effect, the corrected months of 9.7% of the baseline mobility rate for this
coverage account for the fact that the state subsample. This result implies that the
law will affect insured individuals working mandates had a slightly larger effect on a
in small firms that do not self-insure. When
we adjust for these factors, the average
length of continuation coverage in states "This regression is not run for those without
that offer coverage falls from 6.88 months health insurance, since their corrected months of
to 4.55 months. continuation coverage are zero by definition.
^^Authors' tabulations from the 1980 National
The results using the corrected months Medical Care Utilization and Expenditure Survey.
of coverage variables are presented in Table ^'Of course, if these higher expected expendi-
7. The findings are very similar; the magni- tures are reflected in lower wages, the mobility distor-
tudes of the coefficients are somewhat in- tion for married men will be no higher than that for
creased, as are the standard errors, although single men. The feasibility ofwage shifting by marital
status is uncertain; see Gruber (1994) for evidence in
the effects are still highly significant. This favor of the ability of employers to shift increased
result is to be expected from a procedure insurance costs to the wages of identifiable demo-
that multiplies our coefficient of interest by graphic groups within the workplace.
 EFFECTS OF PUBLIC POLICY ON JOB-LOCK 99

population that was more likely to be af- Table 8. The Effect of Continuation Mandates
fected by their presence. on Insurance Coverage.
No
Continuation Mandates Have Health
and Insurance Coverage Full Health Insur-
Independent Variable Sample Insurance ance
A necessary (but not sufficient) precon-
dition for continuation of coverage man- A. Coefficient Estimates
dates to affect job mobility is that they Non-White -.2961 -.3227 -.1392
affect insurance coverage.'" Moreover, the (.0350) (.0617) (.0500)
effects of these mandates on insurance cov- Education .1242 .0929 .1179
erage may be important for assessing their (.0046) (.0077) (.0070)
welfare implications. Ostensibly, the pur- Age .0112 .0368 -.0289
(.0096) (.0174) (.0146)
pose of these mandates was to correct a .0001 -.0002 .0005
Age^
failure in the market for private insurance (.0001) (.0002) (.0002)
for job leavers. Such market failure seems Months of Coverage .0034 .0112 -.0048
likely, given the adverse selection in the (.0027) (.0048) (.0043)
take-up of continuation coverage benefits Log-Likelihood -7,535 -2,513 -3,347
documented by Huth (1991) andLongand N 12,285 4,959 5,070
Marquis (1992). In this case, increasing B. Marginal Effect of .0145 .0411 -.0219
insurance coverage among those who would 12 Months of
have left their jobs even in the absence of Continuation
the regulations represents a clear welfare Coverage
improvement. As discussed earlier, how- Notes: This table gives estimates of the probability
ever, the welfare implications of reducing of being insured in the quarter after a job change
job-lock are less clear. Thus, the net wel- using data from the 1984-87 Survey of Income and
fare effects of the mandates may depend on Program Participation. The sample is comprised of
men aged 20—54 who change jobs or become unem-
the relation between their effect on ployed. Coefficients for panel dummies are not re-
"inframarginal" individuals, who would have ported.
left their jobs in the absence of a mandate,
and their effects on "marginal" individuals
whose mobility decision is made in response These results are presented in Table 8.
to the mandate. Since there are relatively few individuals in
some states, we run the regressions without
To model the effect of these mandates state effects; the coefficients are quite simi-
on insurance coverage, we examine the
sample of worker/quarters in which there lar, although less precisely estimated, if
was ajob change. We then create a dummy state effects are included. Once again, we
variable equal to one if an individual has use a probit specification.
insurance coverage in the quarter after his For the full sample (column 1), the coef-
job change. We regress this dummy vari- ficient on months of coverage suggests that
able on the set of covariates used earlier." continuation mandates have a positive but
insignificant effect on the probability of
being insured after changing jobs. The
'"Some evidence that these mandates did increase implied increase in insurance coverage is
coverage among those who left their jobs and re- 1.4 percentage points. In the second and
mained unemployed is presented in Klerman and third columns of Table 8, we split the
Rahman (1992). For comparability to our mobility
regressions, however, we wish to examine the effects sample, as before, into those with and with-
on the insurance coverage for any worker who leaves out health insurance. For those with health
his job. insurance, the effect of continuation cover-
"We do not include thejob characteristics (hourly age is statistically significant and implies a
wage, industry, and occupation), since it is not obvi-
ous which job (the one left or the one gained) to use. 4.1 percentage point increase in insurance
The results are similar if the characteristics of the coverage after changing jobs. For those
former job are included. without employer-provided health insur-
100 INDUSTRIAL AND LABOR RELATIONS REVIEW

ance, the likelihood of having insurance ity by about 25%. Furthermore, above and
after leaving a job falls by 2.2 percentage beyond increasing job mobility, continua-
points. Thus, the effect for those with tion mandates were successful in increas-
health insurance is strengthened if we con- ing the insurance coverage of job leavers.
sider other trends correlated with the pas- Our findings therefore imply that a sub-
sage of the mandates. Compared to the stantial portion of the job-lock problem
effect on mobility for those with health can be alleviated by mandatory limited port-
insurance (0.84 percentage points), this ability. This conclusion is supportive of
result suggests that the effect of the man- Holtz-Eakin's (1994) contention that job-
dates was largely to cover individuals who lock is a short-run problem. Presumably,
would have left their jobs even in the ab- the expansion of eligibility for COBRA cov-
sence of continuation availability. erage after 1990, allowing continued health
coverage through the previous employer
even after an individual finds employment
Conclusion
with a new firm that offers health insurance
The problem of job-lock has caught the (for which there may be length of service
public's attention, and one key measure of requirements or pre-existing conditions
the success of health care reform in the exclusions), will strengthen the effects of
United States will be its ability to mitigate COBRA. Future work to model the effects
insurance-induced mobility reductions. of this change would be very useful.
That goal, however, is only one of several Although our estimates imply that lim-
toward which reform must strive, goals that ited portability significantly reduces job-
may not be mutually compatible. If it is lock, there are a number of ways in which
possible to substantially alleviate job-lock limited portability may need to be refined
through modest policy changes, doing so to mitigate this distortion for the entire
may offer policy-makers more degrees of population. The important point from an
freedom with which to design other ele- economic perspective is that the policy
ments of the reform package. changes required to fully remove job-lock
This research suggests that continuation may be very costly relative to the changes
benefits have had some success in alleviat- necessary to greatly mitigate its effects. If
ing job-lock. We find that one year of limited portability has a substantial effect,
continuation benefits was associated with a policies that strengthen COBRA, while per-
10% increase in mobility among those with haps providing for a public reinsurance
health insurance—a moderately large ef- pool for the worst risks, may be all that are
fect if, as Madrian (1994) has estimated, necessary to greatly reduce this key labor
insurance-induced job-lock reduces mobil- market distortion.

APPENDIX

Calculating Correction Factors for the Impact of State and Federal Continuation Laws

In our basic regression specification, we assign to firms are exempt from state mandates. And third,
individuals the maximum number of months of con- those who work in firms with fewer than 20 employees
tinuation coverage mandated under either federal or will not be affected by the federal law, which does not
state law. Several factors, however, lead to less than apply to small firms, although they will be influenced
full coverage of these laws. First, those who are not by state laws that pertain to all firms.
covered by employer-provided health insurance To the extent that these factors differ across states,
should not be affected by any of these laws. Second, we would expect laws that mandate equivalen t mon ths
before the federal law took effect, employees of self- of coverage to have different effects. To account for
insured firms should not have been affected by the the less than full coverage of these laws, we compute
state laws because, under the 1974 Employee Retire- a "corrected" measure of months of coverage. As
ment Income and Security Act (ERISA), self-insured before, if there is no state or federal law in effect, we
 EFFECTS OF PUBLIC POLICY ON JOB-LOCK 101

assign no months of coverage to an individual. If a are constant over time. This assumption clearly is
state law is operative but the federal law has not yet incorrect for self-insurance, which grew dramatically
taken effect, the corrected months of coverage equals during the 1980s. As long as the pattern of growth of
self-insurance was uncorrelated with the passage of
Months \ r- J Iracuon \ y traction in \ - |
( of State 1 • I ( Offered I . ll- Self-insured I I
Coverage/ L\Insurance/ \ Firm / J
these laws, however, it will not bias our results. Re-
cent research suggests that mandates play little role
in the firm's decision to self-insure.
If the federal law is in effect but there is no state The fraction of workers offered employer-pro-
law, the corrected months of coverage equals vided health insurance and the fraction of insured
workers in firms with fewer than 20 employees are
• Months \ w- y Fraction v y Fraction \ - | measured directly, by state, from the CPS. The frac-
I of Federal 1 . I I Offered 1 . II- Insured in 1 tion in self-insured firms is estimated from the HIAA
VCoverage/ L\Insurance/ \ <20 Firm / J ' data. In using the HIAA data, we have tried two
Finally, if both a state law and the federal law are in different approaches. The first is to calculate the
place, the corrected months of coverage equals average of these quantities by state. This strategy,
however, is hindered by very small state cell sizes in
/Months \ r- y Fraction \ y Fraction \ - i the HIAA data. Our second approach, therefore, is to
I of Federal j . I I Offered I . H- Insured in I I + use the HIAA data to run a regression predicting the
Vcoverage/ L \ Insurance/ \ <20 Firm / J rate of self-insurance as a function of firm size, indus-
try, and census division. We then use these estimated
Months \ r y Fraction \ y Fraction in \ coefficients to impute values of these quantities for
( of State I ' l l Offered 1 . II- Self-insured 1 .
Coverage/ L \ Insurance/ \ Firm /
each individual in the May CPS. Our correction
factors are state-wide averages of these imputed val-
Fraction \ - i
ues.
( Insured in 1 I
<20Firm / J '
We imputed values in this way for all male workers
between 20 and 54 years old. In our empirical work,
we use the corrected months of coverage computed as
We measure the various components of the cor- outlined above when the sample includes all workers.
rected months of coverage using data from the Health When the sample is restricted to only those with
Insurance Association of America (HIAA) 1989 em- health insurance, we use a similarly constructed cor-
ployer survey and the May 1988 Current Population rected months of coverage that does not adjust for the
Survey (CPS) pension supplement. In using data fraction of workers in the state with employer-pro-
from this late date, we are assuming that these factors vided health insurance.

REFERENCES

Bartel, Ann, and George Borjas. 1977. "Middle-Age Services, National Center for Health Services Re-
Job Mobility: Its Determinants and Consequences." search and Health Care Technology Assessment.
In Seymour Wolfbein, ed.. Men in the Pre-Retirement Flynn, Patrice. 1992. "Employment-Based Health
Years. Philadelphia: Temple University Press. Insurance: Coverage Under COBRA Continuation
Congressional Research Service. 1988. Costs and Rules." In U.S. Department of Labor, Pension and
Effects of Extending Health Insurance Coverage. Wash- Welfare Benefits Administration, Health Benefits and
ington, D.C: Library of Congress. the Workforce. Washington, D.C: GPO.
Cotton, Paul. 1991. "Pre-existing Conditions'Hold Gruber, Jonathan. Forthcoming. "State Mandated
Americans Hostage' to Employers and Insurance." Benefits and Employer-Provided Health Insurance."
Journal of the American Medical Association, Vol. 265 Journal of Public Economics, forthcoming.
(May), pp. 2451-53. . 1994. "The Efficiency of a Group-Specific
Diamond, Peter A. 1992. "A Proposal for Universal Mandated Benefit: Evidence from Health Insur-
Coverage Through Compulsory Group Health In- ance Benefits for Maternity." American Economic
surance." Mimeo, Massachusetts Institute of Tech- Review, Vol. 84 (June), pp. 622-41.
nology, Department of Economics. Gruber, Jonathan, and Brigitte C. Madrian. 1993a.
Diamond, Peter A., and Jerry A. Hausman. 1984. "Health Insurance Availability and Early Retire-
"The Retirement and Unemployment Behavior of ment." NBER Working Paper No. 4469. Cam-
Older Men." In HenryJ. Aaron and Gary Burtless, bridge, Mass.: National Bureau of Economic Re-
eds.. Retirement and Economic Behavior. Washington, search.
D.C: Brookings Institution. . 1993b. "Health Insurance and Early Retire-
Earley, Pamela J. 1986. Private Health Insurance in the ment: Evidence from the Availability of Continua-
United States. National Health Care Expenditures tion Coverage." NBER Working Paper No. 4595.
Study Data Preview 23, DHHS Publication No. (PHS) Cambridge, Mass.: National Bureau of Economic
86—3406. U.S. Department of Health and Human Research.
102 INDUSTRIAL AND LABOR RELATIONS REVIEW

Hewitt Associates. 1985. Continuation of Group Medi- Changes." In R. G. Ehrenberg, ed.. Research in
cal Coverage—A Study ofState Laws. Lincolnshire, 111.: Labor Economics, Vol. 8, part A. Greenwich, Conn.:
Hewitt Associates. JAI Press.
Holtz-Eakin, Douglas. 1994. "Health Insurance Pro- Minter, Caroline. 1992. "Tests ofJob-Matching Mod-
vision and Labor Market Efficiency in the United els Using Turnover and Wage Growth." Mimeo,
States and Germany." In Rebecca Blank, ed.. Social Massachusetts Institute of Technology.
Protection Versus EconomicFlexibility: Is There aTradeoff? Mitchell, Olivia S. 1982. "Fringe Benefits and Labor
Chicago: University of Chicago Press. Mobility." Journal of Human Resources, Vol. 17
Huth, Stephen A. 1992. "COBRA Costs up. Elections (Spring), pp. 286-98.
Down: Survey." Employee Benefit Plan Review, Sic'p- Rosen, Sherwin. 1986. "The Theory of Equalizing
tember, pp. 12-34. Differences." In Orley Ashenfelter and Richard
Klerman, Jacob Alex, and Omar Rahman. 1992. Layard, eds.. Handbook of Labor Economics, Vol. 1.
"Employment Change and Continuation of Health Amsterdam: North-Holland.
Insurance Coverage." In U.S. Department of Labor, Thompson Publishing Group. 1992. Employer's Hand-
Pension and Welfare Benefits Administration, Health book: Mandated Health Benefits—The COBRA Guide.
Benefits and the Workforce. Washington, D . C : GPO. Salisbury, Md.
Madrian, Brigitte C. 1994. "Employment-Based Topel, Robert. 1986. "Job Mobility, Search, and
Health Insurance and Job Mobility: Is There Evi- Earnings Growth: A Reinterpretation of Human
dence of Job-Lock?" Quarterly Journal of Economics, Capital Earnings Functions." In R. G. Ehrenberg,
Vol. 109 (February), pp. 27-51. ed.,ResearchinLaborEconomics,Vo\.8,partA. Green-
. 1993. "The Labor Market Effects of Employ- wich, Conn.: JAI Press.
ment-Based Health Insurance." Doctoral diss., U.S. Departmentof Labor, Bureau of Labor Statistics.
Massachusetts Institute of Technology. 1989. Employee Benefits in Medium and Large Firms.
Mincer, Jacob. 1986. "Wage Changes in Job Washington, D.C: GPO.