Is Employer-Based Health Insurance a Barrier to Entrepreneurship?
The focus on employer-provided health insurance in the
The predominant source of health insurance in the
Concerns about disruptions in health insurance coverage could also influence the decisions of individuals who are contemplating starting new businesses. Such individuals who are currently covered by employer-sponsored health insurance would eventually lose that coverage if they leave their job. Potential entrepreneurs could face high premiums in the individual health insurance market and the possibly prohibitive health costs of being uninsured. Furthermore, changes in health plans and providers may be disruptive and costly. New entrepreneurs may also be exposed to pre-existing condition limitations and waiting periods for coverage if they have a spell of uninsured unemployment between their employer-provided coverage and their new health insurance policy. Unless they have alternative sources of health insurance coverage, such as through a spouse’s employer, this health insurance conundrum may influence their decision to start a new business. The end result is that the
Promoting entrepreneurship is viewed as a national priority by governments around the world. The interest is driven primarily by evidence that small businesses create a disproportionate share of new jobs in the economy, represent an important source of innovation, increase national productivity and alleviate poverty (see Birch 1979, Reynolds 2005, and OECD 2006 for example). The self-employed are also unique in that they create jobs for themselves, representing more than ten percent of total employment in the
Given the potential consequences, it is surprising that only a handful of studies have examined whether employer-provided health insurance limits entrepreneurship. The few studies in the literature find mixed results, with some estimating that health insurance reduces transitions into self-employment by as much as 25 percent and others finding no evidence that health insurance reduces self-employment (Holtz-Eakin et al., 1996; Madrian and Lefgren, 1998; Wellington, 2001). The lack of research on the topic contrasts sharply with a much larger literature that examines the effects of employer-provided health insurance on job mobility (Currie and Madrian, 1999). Furthermore, the previous research on health insurance and entrepreneurship generally uses data from the 1980s and early 1990s. Since then, health insurance costs have grown dramatically.
In this paper, we address the lack of current research on the topic of “entrepreneurship lock” by providing a new study of whether the
The second identification strategy that we use takes advantage of the abrupt change in health insurance coverage occurring at age 65 due to Medicare. This identification strategy is particularly well-suited suited for identifying the importance of health insurance for the near-elderly. Following the approach of Card, Dobkin, and Maestas (2006), we estimate regression discontinuity models that identify the effects of health insurance coverage on outcomes (in our case entrepreneurship) by comparing similar individuals just below the age 65 cutoff to those just above the age 65 cutoff. A discontinuous jump in business creation suggests that health insurance coverage from Medicare may be responsible. We use a novel procedure for identifying a person's age in months from matched CPS data. To our knowledge, this is the first study using this procedure and the first study to use the discontinuous change in health insurance qualification at age 65 to test the "entrepreneurship lock" hypothesis. The results from this new identification strategy and the difference-in-difference approach using recent data shed light on the question of whether employer-based health insurance restricts entrepreneurship.
2. Previous Literature
The few studies that examine the relationship between entrepreneurship and an individual’s health insurance coverage status find mixed results. Holtz-Eakin et al. (1996) considered the effect of health insurance coverage status on transitions from employment to self-employment using the 1984-86 Survey of Income and Program Participation (SIPP) and the 1982-84 waves of the Panel Study of Income Dynamics (PSID). While their estimates were quantitatively large (a lack of health insurance portability stemming from employer-sponsored insurance reduced the probability of transition from employment to self-employment by 9 to 15 percent in the SIPP population), they were statistically insignificant. Therefore, the authors could not confirm that health insurance impeded transitions to self-employment. Madrian and Lefgren (1998) also examine this issue using the SIPP and find that by using additional waves of SIPP data (1984-93), estimates of the effect of health insurance coverage status on transitions to self-employment attain statistical significance. In addition to using the difference-in-difference methodology used by Holtz-Eakin et al., they also use the passage of continuation of coverage mandates to identify the effect of health insurance coverage status on self-employment transitions. Their estimates imply that a lack of health insurance portability accounts for a 25 percent reduction in transitions to self-employment. In other work using a similar estimation methodology,
Another potential source of variation in the health insurance market for the self-employed comes from the tax treatment of health insurance. The tax subsidy to health insurance for the self employed, introduced at 25% in 1986 rose to 100% by 2003 in a number of discrete changes. Velamuri (2005) uses this variation and compares the female self-employment rate in 1984-85 to that in 1990-91 using cross-sectional self-employment rates rather than individual transitions to self-employment. She finds that women with no spousal health insurance were substantially more likely (12 to 25%) to be self-employed when tax subsides were higher compared to women who had access to spouse health insurance. However, estimates based on transitions to self-employment were statistically insignificant. Gumus and Regan (2007) present the raw percentages of workers transitioning into self employment between 1995 and 2005 and find that the transition rate has been stable over time and does not show any evidence of increasing when tax credits were increased. The relatively small estimates obtained by the studies using the variation provided by tax subsidies may be driven by the fact that many small businesses have very low levels of sales and profits in the first few years of existence (U.S. Census Bureau 1997), and thus are not eligible for or benefit only slightly from the tax credit.
DeCicca (2007) presents additional evidence on the effect of legislative changes on transitions to self employment. This study focuses on the effect of
Despite having received far more attention than the literature on entrepreneurship lock, the literature on the effects of health insurance coverage on job mobility among wage/salary workers has not reached a consensus either. Most studies in the literature use difference-in-difference methodologies to compare workers who hold employer-provided health insurance and have a high demand for this health insurance to a comparison group of workers who are believed to have a lower demand for their current health insurance plan. The comparison group has low demand for their current plan either because they do not have any insurance or because they have an alternative source of health insurance. A number of studies have found little evidence of job lock (Holtz-Eakin, 1994; Penrod, 1994; Kapur, 1998; Berger, Black, and Scott, 2004; Gilleskie and Lutz, 2002). However, other studies have found economically meaningful and significant estimates of job lock (Madrian, 1994; Cooper and Monheit, 1994; and Buchmueller and Valletta 1996; Stroupe et al. 2001; Okunade and Wunnave, 2002; Bradley et al., 2005; Sanz de Galdeano, 2006).
In this study, we use recent panel data created by matching consecutive years or months of the CPS to estimate the effect of health insurance coverage status on business creation. Existing prior research on this topic uses data from the 1980s and early 1990s, and many important changes have occurred in the health insurance market and in the labor market for entrepreneurs. In particular, health insurance costs have risen dramatically since the 1990s, particularly for small group and individual plans. The demographic composition of new entrepreneurs has also changed, with the near-elderly -- a rapid-growing segment of the
3. Conceptual Framework
Access to health insurance is a major concern among entrepreneurs. In a recent survey, health insurance costs were most frequently listed as the top critical problem faced by small businesses (NFIB, 2008). In a related survey, three-quarters of the self-employed listed cost as an important barrier to offering health insurance through their business and 78% rated the satisfaction with their premium costs as “low” (AWP, 2005). Furthermore, the burden of premium costs is disproportionately high on the smallest establishments – representing 5.7% of sales for solo practitioners compared to 2.8% for larger establishments (AWP, 2005). Only 24% of the self-employed hold employer-based coverage in their own name compared to 65% of workers in firms that employ more than 1000 workers. This gap shrinks somewhat when all sources of employer-based health insurance are considered, but it remains substantial, with half of self-employed and small business workers having employer-based coverage compared to 80% of workers in large firms (EBRI, 2007). Self-employed individuals who do not have access to employer-based coverage through a spouse may need to rely on the individual health insurance market. Premiums in the individual health insurance market can be high. Hadley and Reschovsky (2003) find that the annual premium for a healthy single 25-year-old male without children is about $1,200. For a family of two adults who are 55, have major health problems, and have two children, both with health problems, the annual premium may be over $14,000. Health plans offered by professional associations were once havens for millions of people who could not get coverage anywhere else. But as medical costs have increased, groups representing a wide range of professions have been forced to stop offering the benefit or been dropped by insurers (Girion, 2007).
In this section we provide a formal conceptual framework to describe why the market for health insurance, as it currently exists in the
We assume that all employer-sponsored group health insurance coverage is the same (health insurance is a homogenous good) and individuals either have it or they do not. Individuals have preferences over wage compensation (or the monetary return from self-employment) and employer-sponsored group health insurance.
A worker’s utility can be described by Uij = U(Wij, Hij), where Uij is the utility of worker i at firm j. Wij is the wage of worker i at firm j, and Hij is a binary indicator of employer-sponsored health insurance coverage of worker i at firm j. Let DWij denote the compensating wage differential in firms offering health insurance reflecting the fact that if individuals value health insurance, they will accept a lower wage from an employer that offers health insurance. Firms face a cost, Cij, of providing workers with health insurance. If self-employed individuals and firms could purchase insurance on a per-worker basis and this insurance was perfectly experience rated and wages were perfectly flexible, the compensating differential DWij would be equal to the cost of health insurance Ci. In this highly stylized model, health insurance would have no effect on the labor market equilibrium since self-employed individuals could purchase health insurance for the same cost as other employers. Workers pay the same compensating differential if they choose a job with insurance and as a result, they select a job or self-employment where they have the highest marginal product of labor. So, workers will switch from a job (j) with group employer-provided health insurance to self employment (s) with no group health insurance if U(Wij – DW, 1) < U(Wis,0). Self-employed workers can then choose to purchase non-group coverage for a cost of Ci in the individual market. In addition, wage earners who do not have employer-sponsored health insurance will transition to self-employment based on a simple comparison of their marginal productivity in the two sectors, and therefore should be more likely to move into self-employment than wage earners who have group health insurance.
This stylized model is not realistic in several ways. First, the self-employed face substantially higher health insurance costs than large firms. Large firms can capitalize on economies of scale, lower administrative costs, and higher bargaining power to gain lower health insurance costs compared to individuals. Second, health insurance is not a homogenous good that can seamlessly be transferred from an employer to self-employment. Despite the HIPAA protections noted above, individuals may incur disruptions in their relationships with providers and changes in policy quality as a result of purchasing new insurance under self-employment. Third, employers do not have complete flexibility to offer health insurance to some employees and not others, and to vary wages in accordance with each worker’s insurance costs. Therefore, workers with high health costs may be paying far less than the true costs of their insurance under group insurance. As a result, health insurance can lead to distortions in the employment market.
Using the framework described earlier, even if an individual was less productive in job, j, with group health insurance than when self employed (Wij <
We use data from the 1996 to 2006 Annual Demographic and Income Surveys (March) of the CPS. Each annual survey, conducted by the U.S. Census Bureau and the Bureau of Labor Statistics, is representative of the entire
The main advantage of the matched CPS is the large sample size. The matched CPS sample that we use includes more than 160,000 observations for wage and salary workers in the first survey year. The sample includes 5,100 transitions into self-employment, which is considerably larger than the other panel datasets such as SIPP and PSID. In their study of health insurance and entrepreneurship, Holtz-Eakin, Penrod, and Rosen (1996) report 700 transitions from the wage and salary sector to self-employment in their sample from SIPP and considerably less in the PSID.
Across, the 1996-2006 CPS surveys, we find that roughly 75 percent of CPS respondents in one survey can be identified in the subsequent year’s survey. The main reason that match rates are less than 100 percent is because of the movement of individuals or households out of sampled dwelling units. The CPS does not follow individuals who move out of CPS sampled dwelling units in future months. Another problem is due to false positive matches. Although unique household and person identifiers are available in the CPS to match non-moving individuals over time, false matches occur because of miscoding. We use a procedure that compares the sex, race and age of the person in each March file to remove false matches. Any changes in coding are identified as false matches. False match rates, however, are very low (roughly 3 percent) and do not vary substantially across years.
The loss of observations due to household movement raises concerns about the representativeness of the matched CPS sample. We investigate this issue further by conducting a comparison of mean values from the original cross-sectional CPS sample to means values from the matched CPS sample. As expected, we find that the matched sample has higher insurance, employment and marriage rates, and is more educated and older. The matched sample is also less likely to be a minority, live in the central city and receive public assistance. But, in all of these cases the differences are very small. For example, health insurance coverage rates are only 3 percent different and the matched sample is only one year older than the original sample (see Fairlie and London 2008 for more details). Previous research comparing matched CPS estimates on labor market transitions and outcomes to estimates from retrospective, current and SIPP panel data also does not find evidence of attrition bias (Stewart 2007 and Neumark and Kawaguchi 2004).
CPS HEALTH INSURANCE MEASURE
The CPS health insurance questions asks individuals to report all sources of health insurance coverage during the entire year prior to survey month.. However, comparisons of CPS estimates of health insurance coverage to other surveys that ask about insurance at the time of the survey reveal similar numbers. Estimates from the SIPP, MEPS and National Health Interview Survey (NHIS) indicate that roughly 40 million individuals were uninsured at the time of the survey in 1998 (CBO 2003). CPS estimates for the number of individuals with no insurance for the entire year were also roughly 40 million in that year, suggesting that the CPS overstates the number of individuals who are uninsured for an entire year. Bhandari (2004) finds similar estimates of insurance coverage rates in the CPS and point-in-time estimates from the SIPP even within several demographic groups. Estimates from the SIPP and MEPS indicate the number of people who are uninsured for an entire year is between 21 and 31 million. Thus, CPS respondents may be underreporting health insurance coverage over the previous calendar year because of recall bias or because they simply report their current coverage (see Bennefield 1996; Swartz 1986; CBO 2003; and Bhandari 2004 for further discussion). Even if the CPS estimates capture a point-in-time measure of health insurance coverage, the measure of health insurance status does not change from year to year and thus allows for an analysis of transitions in status. However, this would alter the interpretation of our results. In this article, we interpret results assuming that respondents correctly respond to the question and report all insurance held in the previous year.
5. Health Insurance Coverage and Self-Employment
Table 1 provides a descriptive profile of the variation in health insurance coverage by employment status. We find that the self-employed are nearly twice as likely to be uninsured than wage/salary workers. Roughly 20% of self employed men and women report no insurance compared to 11.8% of male wage/salary workers and 10.5% of female wage/salary workers. The uninsured rates for the self-employed are also higher than those for the other/not working population. Although this group includes the unemployed, not in the labor force and low hours workers, health insurance rates are 6.8 percentage points higher than rates for the self-employed for men and 4.5 percentage points higher for women.
Insured self-employed men are most likely to get their coverage from employment (33%), followed by dependent employer coverage (21%) and individual coverage (21%). However, insured self-employed women are most likely to get dependent employer coverage (35%), followed by individual coverage (22%) and coverage from own employment (19%). The distinction between individual coverage and own employer coverage for the self-employed is nebulous. Self-employed individuals may obtain health insurance only for themselves, but purchase it through their business, and report this coverage as employment-based insurance rather than individual insurance.
The lack of health insurance among full-time, full-year self-employed business owners is similarly high. Slightly more than 20% of full-time, self-employed men are uninsured and 19.7% of full-time, self-employed women are uninsured. These rates of uninsurance are considerably higher than for full-time, wage/salary workers.
In table 2, we use the two-year panel structure of our data to examine health insurance types and coverage in the second year for the newly self-employed, self-employment leavers, and the self-employed in both survey years. These estimates provide further evidence on the strong relationship between self-employment and not having health insurance. Individuals who are newly self-employed have very high rates of uninsurance -- 24.5 percent for men and 23.2 percent for women -- indicating that initial movement to self-employment is strongly associated with the loss of health insurance. As reported in Table 1, both wage/salary workers and those not working had substantially lower rates of uninsurance.
Although individuals who have been self-employed for at least two consecutive years have higher rates of health insurance coverage than the newly self-employed, coverage rates remain very low. Among men, 18.6% lack health insurance, and 17.4% of women are uninsured. Another interesting finding is that more than half of the male workers who leave self-employment move to jobs that have employer-provided health insurance. A large percentage of women leaving self-employment also move to jobs with employer-provided insurance. Overall, these results suggest that being uninsured is associated with movements to and from self-employment.
Four percent of all male wage/salary workers start a business each year (see Table 3). For those who have health insurance coverage from their employer, business creation rates are substantially lower at 2.9%. In contrast, 6.6% of workers who have health insurance coverage from a spouse start a business. Wage/salary workers who have no insurance coverage have a similarly high likelihood of starting a business. This result is not being driven by the unemployed or low-hours workers because only wage/salary workers with 20 or more weeks and 15 or more hours per week are included in the sample. Furthermore, when we condition on full-time, full-year work we find similar results. Business creation rates are substantially lower among wage/salary workers who have employer insurance than among wage/salary workers who have insurance coverage through a spouse or do not have insurance.
Although business entry rates are lower for women, similar patterns across health insurance coverage emerge. Entrepreneurship rates are much lower for female workers with employer insurance than for female workers with spousal coverage or no insurance. Conditioning on full-time work does not change this conclusion.
Of course, we cannot interpret these descriptive results as evidence that employer health insurance is an impediment to self-employment, since employer health insurance is correlated with job quality. Workers who have employer-provided health insurance may be less likely to move to self-employment or another job simply because they already have a job with a good compensation package. We attempt to address these concerns in the next section.
6. Estimating the Effects of Health Insurance Coverage Status on Entrepreneurship
We use two main estimation strategies to identify the effect of health insurance coverage status on entrepreneurship. First, we construct difference-in-difference models of the transition to self-employment from wage-based employment as a function of access to alternative health insurance and family health. Individuals with no alternative means of health insurance who obtain health insurance from their own jobs, and individuals who have poor family health should be less likely to become self-employed, all else equal. The second identification strategy that we use takes advantage of the abrupt change in health insurance coverage occurring at age 65 due to Medicare. We examine if the gain in health insurance at age 65 encourages individuals to become self-employed business owners. We describe each of the estimation methodologies in turn.
To identify the effect of health insurance coverage status on entrepreneurship, we compare the rate of entrepreneurial entry for an experimental group that potentially faces a disruption in health insurance coverage to the rate of entry for a control group that does not face a disruption. In addition, we also use the fact that groups with a high demand for their current health insurance policy should be less likely to leave their jobs to start a business. The literature has used several different variables to proxy for high demand including number and health status of family members (Holtz-Eakin et al., 1996, Madrian 1998). We focus on a few of these measures that are available in the CPS and best capture potential demand for health insurance and care. The measures of potential health care demand that we include are the following: (i) having a family member in bad health, (ii) number of family members in bad health, and (iii) lacking an alternative source of health insurance coverage through a spouse’s employer plan. Individuals who are in poor health or who have a family member in poor health are likely to have a high demand for their current employer-provided health insurance policy since they may face high premiums in the individual health insurance market or a discontinuity in their treatment if they change insurance plans. Workers who have only a single source of employer-provided health insurance are likely to have a higher demand for this health insurance compared to workers who have access to an alternative source of health insurance from a spouse’s employer-provided health insurance plan.
While there is considerable flexibility in the choice of experimental and control groups in a difference-in-difference estimator, the comparability of the two groups is important to obtain a consistent estimator. The key assumption, which is likely to hold only if the groups are comparable, is that the effect of any exogenous influences is the same on the control and the experimental groups (Meyer 1995). We use two main classifications of experimental and control groups. First, we define individuals who hold employer-provided health insurance as the experimental group and individuals who do not hold employer-provided health insurance as the control group. By definition, individuals who hold health insurance are more likely to be deterred from moving to self employment because of their current health insurance status than individuals who do not hold health insurance. Empirically, we estimate the following probit model:
6.1 Prob(yi) = Φ(β0 + β1Hi + β2Di + δ3HiDi + γ'Xi),
where Hi denotes whether an individual holds employer-provided health insurance, Di is potential health care demand, and Xi is a vector of demographic controls. We estimate separate models for men and for women. The sample consists of wage and salaried workers in the baseline year (t). The dependent variable, yi, equals 1 if the worker moves to self-employment in the following year (t+1). We estimate several versions of this model with the measures of potential health care demand discussed above. The coefficient on the interaction between health insurance and potential health care demand, β3, captures the difference-in-difference estimate of the entrepreneurship-lock. A negative coefficient is consistent with the notion that current employer-provided health insurance is a disincentive to entrepreneurship, and suggests that those individuals who would face a disruption in their health insurance and have a high demand for health care are relatively less likely to move into self-employment. Note that we cannot simply interpret β1 as the estimate of the effect of employer-provided health insurance on entrepreneurship because having own employer-provided health insurance may be correlated with high quality jobs and therefore this estimate would be biased.
A potential problem with this classification of experimental and control groups is that individuals who hold employer-provided health insurance differ from those who do not. Insurance holders have higher wages, longer tenure, and more education than non-holders. In additional specifications, we restrict the sample to individuals who hold employer-provided health insurance to improve the comparability of the experimental and control groups. We define the control group as individuals who have access to alternative health insurance from a spouse’s employer. We do not require that the individual is covered by the spouse’s plan, only that the spouse has own employer-provided health insurance, since individuals can usually obtain coverage from a spouse’s employer even if they are not currently covered by the policy. The experimental group is defined as individuals who do not have access to spouse employer-provided health insurance. Individuals who do not have access to an alternative plan should be more likely to be deterred from moving to self-employment because of health insurance. Workers without spousal coverage face a potential disruption in health insurance coverage when moving from wage/salary work to self-employment, whereas workers with spousal coverage potentially do not face a potential disruption in health insurance. Individuals in these two groups are relatively similar across several dimensions such as wages, education, and tenure, suggesting that individuals with own and spousal employer-provided health insure form a more comparable control group for individuals with only employer-provided health insurance.
We estimate the following probit model on the sample of individuals who hold employer-provided health insurance.
6.1 Prob(yi) = Φ(β0 + β1NSi + β2Di + δ3NSiDi + γ'Xi),
where NSi denotes that an individual does not have a spouse who holds an employer-provided health insurance plan. The sample now only consists of wage and salaried workers in the baseline year (t) who hold employer-provided health insurance. The dependent variable equals 1 if the worker moves to self-employment in the following year (t+1). We estimate this model with the remaining measures of potential health insurance demand. The coefficient on the interaction between no spouse health insurance and high health care demand, β3, captures the difference-in-difference estimate of the entrepreneurship-lock. A negative coefficient suggests that those individuals who would face a disruption in their health insurance and have a high demand for health care are relatively less likely to move into self-employment.
To obtain even further compatibility between the experimental and control groups, we estimate several robustness checks in which the sample is restricted based on spousal characteristics. We impose two restrictions to the sample of individuals who hold employer-provided health insurance: (1) only married couples and (2) only spouses who are both working at least 30 hours per week.
Table 4A reports the results from estimating equation (6.1) for men using the full sample. Columns 1 – 3 present three different measures of high health care demand, no spouse health insurance, anyone in the family in bad health, and number of family members in bad health. The estimates from the models in Table 4A show that whites and immigrants are more likely to become self-employed. Workers with relatively more education, with higher family incomes and home-owners are also more likely to start businesses. In general, these results are consistent with findings from the previous literature and the notion that workers with more resources are the most likely to be able to start a business.
The direct effect of own employer provided health insurance is large – workers who have such health insurance are between 2.5 and 3.9 percentage points less likely to move to self employment relative to a baseline transition rate of 4 percent. Interestingly, high wage workers, who are likely to be employed in high quality jobs, are more likely to leave their current job to start a business. The opposing effect of wage and health insurance on transitions is intriguing since we would expect unobserved job quality to bias both of these estimates downwards. However, we cannot place much weight on the direct effect of health insurance since it could be contaminated by unobserved job quality, and so we rely on the interaction of the high demand variables with employer health insurance (e.g. β3) to obtain an estimate of the effect of health insurance on entrepreneurship.
In column 1, the interaction of employer health insurance and no spouse health insurance is negative and statistically significant. The magnitude of the estimated effect is 2 percentage points which is quite large relative to a base entrepreneurship transition rate of 4 percent suggesting that the lack of spouse health insurance is a disincentive to entrepreneurship for those who rely on their own employer policy. For the other measures of potential demand for health insurance in columns 2 and 3, the results are not as clear. The coefficients on the interactions between own employer health insurance and anyone with bad health and own employer health insurance and the number of family members with bad health are both positive, but statistically insignificant.
The results for women in table 4B are somewhat similar. Employer provided health insurance has a large negative direct effect on transitions. It appears that higher wage women are also less likely to move to self-employment -- the effects of wage and health insurance are similar for women, unlike for men. Similar to the results for men, the coefficient on the interaction between own employer health insurance and no spouse employer insurance is negative and statistically significant. The coefficient estimate is also large implying an effect of 1.75 percentage points. Using the alternative measures for potential demand, we do not find negative coefficients on the interaction terms.
The estimates reported in Tables 4A and 4B provide some support for the hypothesis that employer-based health insurance limits entrepreneurship, but the evidence is not consistent across different measures of health demand. As noted above, a potential problem with this approach is that we are defining the experimental group as individuals who have their own employer health insurance and the control group as individuals who do not have their own employer health insurance. Although these estimates may be suggestive, the comparability of these two groups is in question. To address these concerns, we restrict the sample to only workers who have their own employer health insurance and estimate equation (6.2).
Table 5A reports the results for men. We report the main effects and interactions between not having a spouse with employer health insurance and the two remaining health demand measures in Columns 1-2. The experimental group is defined as individuals who do not have spouses with employer health insurance and the control group is defined as individuals who have spouses with employer health insurance. The coefficient on the interaction between no spouse health insurance and anyone in bad health in the family is large, negative and statistically significant. The coefficient estimates on the number of family members in bad health is also negative and statistically significant. These estimates show that men with poor family health and no spouse health insurance are significantly less likely to give up their employer plan and move to self employment.
The results are similar for women (Table 5.B). Female workers in families with poor health and do not have spouses with health insurance are less likely to start businesses. For both measures of poor family health the coefficients are large, negative and statistically significant.
To further improve the comparability of the experimental and control groups we impose additional restrictions on the sample. First, we limit the sample to only include married couples. A concern with the previous specification is that the experimental group includes both married and unmarried men whereas as the control group only includes married men because this group has a spouse with employer health insurance. Second, we limit the sample to individuals with full-time, full-year working spouses. In columns 3 and 4 of Tables 5A and 5B, we report results from the restricted sample of full-time working couples. We find that the interaction between the health measures and spousal health insurance strengthens in magnitude slightly and continues to be statistically significant. However, for women, the interaction term becomes somewhat smaller in magnitude and statistically insignificant. Results are similar for the alternative samples.
REGRESSION DISCONTINUITY ESTIMATES
We take a new approach to examining the question of whether health insurance discourages entrepreneurship by examining the discontinuity created at age 65 through the qualification for Medicare. In the month that individuals turn 65 they automatically qualify for Medicare, changing their access to health insurance coverage. Attaining Medicare eligibility should immediately reduce the value an individual places on employer-sponsored health insurance. In particular, would-be-entrepreneurs no longer have to be concerned about losing basic employer-sponsored health insurance coverage after that date. Although business creation rates are likely to vary substantially by age, we can isolate the effects of the "Medicare notch" by comparing business creation rates just before the age 65 birth month and just after the age 65 birth month. This approach addresses additional concerns over the potential influence of unobservables, such as individual health status, on the results.
To take this approach we use matched monthly data from the CPS. By matching consecutive months of the CPS we can identify when a person changes ages. However, the CPS only interviews households for 4 consecutive months, which limits us to identifying up to two months before the birth month, the birth month, and 2 months after the birth month. We cannot identify the birth month of individuals whose birth month does not fall in the four month interview window. This approach has not been used in the previous literature to estimate "entrepreneurship lock". Few data sets contain a large enough sample size as well as information needed to identify exact birth month.
We first examine business ownership rates by age in years. Figure 1 reports estimates for men. There is a clear jump at age 65 in business ownership rates. We find that the percentage of individuals owning a business increases from 26.5 percent at age 64 to 28.7 percent at age 65. Although we do not focus on women, we also find a large increase in business ownership rates at age 65. We do not focus on women in this analysis because the use of data from the 1996 to 2006 CPS implies that individuals who reach age 65 in the sample were born in the 1930s. Given the dramatic changes in labor force participation among women between the 1950s and 1980s and evidence that these changes were driven by differences in the behavior of women born after World War II (Lichter and Costanszo, 1987; McEwen, Orrenius and Wynne, 2005) this age group has too low of a labor force participation rate. We find that only around 30 percent of women ages 55-75 are employed. We thus focus on men.
The estimates by age in years indicate an increase in self-employment at age 65, but there is a lot of variation by age. Because of this variation we focus the analysis around the month of the 65th birthday when individuals become eligible for Medicare. To compare individuals on either side of their birth month we limit the sample to individuals whose birth month falls in the four consecutive month interview period. We create three groups: the two months before a birth month (just under age 65), the month in which the age changes (almost age 65), and the two months following an age change (just over age 65). The "almost age 65" category is created because of the ambiguity over whether the individual's birthday is in the same month as the survey month or if it falls in the month after the survey month. The survey date is typically in the second week of the month.
Figure 2 reports estimates of self-employment rates around the age 65 cutoff. There is a clear break at age 65. Business ownership rates increase from 24.6 percent for those just under age 65 to 28.0 percent for those just over age 65. As a check, we also report estimates of business ownership rates for the three classifications for all other ages between 55 and 75. As expected, because age is only increasing slightly from just before to just after the birth month the business ownership rates are essentially the same around the birth month cutoff. For these age changes, there is no change in eligibility for health insurance.
The simple comparison of business ownership rates around the age 65 break provides some supportive evidence. To examine the effects of the age 65 break more carefully we estimate regression discontinuity models. The regression specification is straightforward and the source of identification is unambiguous. The probability of entrepreneurship is:
6.3 yit = α + λt + g(ai) + δ1Di65a + δ2Di65o + δ3Dia + δ4Dio + β'Xi + εi,
where λt are year fixed effects, g(a) is a function of age in months, Di65a is a dummy for almost being age 65, Di65o is a dummy for being just over age 65, Dia and Dio are the almost and over age dummies for other ages, and Xi is a vector of demographic controls. This model is more flexible than most regression discontinuity models. Identification of the Medicare effect is being driven entirely by comparing just over age 65 observations to just under age 65 observations. g(a) includes age in year fixed effects instead of a smooth function to allow for a more flexible age profile for the probability of entrepreneurship. This general form allows for different pre and post age 65 trends. We also estimate models that use a less flexible quadratic function for g(a). Finally, Di65a is included for the almost age 65 month. As noted above, there is some ambiguity over which month is the birth month, and is thus treated separately.
Table 6 reports estimates from several regressions of equation (6.3). The first specification includes only the age cutoff dummy variables. The left-out category is just under age 65. The coefficient on the just over age 65 variable is positive and statistically significant. Although there is a strong positive association between entrepreneurship and age, the results are not being driven by the small increase in age from the just before period to the just after period. We are implicitly controlling for this increase in age by including dummy variables for almost at the age cutoff and just over the age cutoff. As expected, these coefficients are very small suggesting that the small change in age between these two periods for ages other than 65 when individuals qualify for Medicare does not have an effect on self-employment rates. Nevertheless, we estimate additional specifications with further controls for age and other variables to check the robustness of the results. In Specification 2, we include a quadratic function for age in months. The coefficient estimate on just over age 65 remains large, positive and statistically significant.
In Specification 3, we replace the smooth function for age in months with a much more flexible form that includes dummies for each age in years as presented in (6.3). Even allowing for a fully flexible form for the age-self-employment relationship for before and after age 65, the estimates remain similar. We find a 0.033 higher probability of owing a business each month if the person is just over age 65 than if the individual is just under age 65. Finally, we also include controls for year, race, nativity, education, marital status, region, urban status, and industry. The coefficient estimate on the just over age 65 variable remains very similar attesting to the strength of the research design. The addition of the covariates has little effect on the estimated relationship between being just over the age 65 cutoff and business ownership. For this specification, the coefficient estimate implies a 0.031 higher probability of owing a business each month if the person is just over age 65 than if the individual is just under age 65. This increase represents 13 percent of the mean probability of self-employment.
To further check the robustness of these results, we estimate several additional regressions. First, we narrow the sample to only include workers ages 60 to 70 (see Table 7 Specification 1). The coefficient estimate on the just over age 65 variable remains large, positive and statistically significant. Second, we restrict the sample to include only full-time workers (defined as working 30 or more hours per week). This restriction rules out the possibility that movement to part-time self-employment at age 65 is driving the results. As reported in Specification 2, the coefficient estimate is similar to the original one. Checking the sensitivity of results in the opposite direction, we expand the sample to include individuals who are not working 15 or more hours per week. We now include all individuals ages 55-75 even if they are not in the labor force. The probability of self-employment for this sample is much lower (11.0 percent) because of the inclusion of non-workers. Specification 3 of Table 7 reports estimates using this sample. We find a higher rate of business ownership associated with being just over the age 65 break. The point estimate implies that the business ownership rate is 0.013 higher, which represents 12 percent of the sample mean. The relative magnitude of the coefficient is similar to the coefficient estimate using the main sample of workers. Thus, the results do not appear sensitive to the treatment of non-employment and low hours work. We also estimated a regression in which hours worked was the dependent variable and found no change in hours worked around the age 65 cutoff. The coefficient estimate on the just over age 65 variable was very small and statistically insignificant.
The final robustness check involves focusing on transitions from non-business ownership to business ownership. Changes in the probability of business ownership capture the combined effects of changes in business creation and business exits around the age 65 birth date. Qualifying for health insurance coverage through Medicare may have a larger influence on self-employment entry rates than exit rates from self-employment. Specification 4 of Table 7 reports estimates of equation (6.3) using self-employment entry as the dependent variable. One problem with this approach is that the likelihood of making a transition to self-employment each month is very low, which may introduce some noise in the estimates. The transition rate to business ownership is 0.004. Similar to the main results, we find a positive coefficient estimate on the just over age 65 variable. The coefficient, however, is not statistically significant. The point estimate implies that the entrepreneurship rate is 0.001 higher, which represents 29 percent of the sample mean.
A major concern with the
Given these concerns it is surprising that only a handful of studies have examined whether employer-provided health insurance limits entrepreneurship, with the few studies in this literature finding mixed results. We address the limited research on the topic of “entrepreneurship lock” by providing a new study using panel data created by matching consecutive years or months of the CPS and two main identification strategies -- difference-in-difference and regression discontinuity models.
A first pass at the data reveals that the self-employed are much less likely to have health insurance than are wage/salary workers and even our sample of unemployed and part-time workers. Estimates from our two-year panel data from matching consecutive March CPS files also indicate that new self-employment entrants have especially low rates of health insurance coverage. We also find that business creation rates are substantially lower among wage/salary workers who have employer insurance than among wage/salary workers who have insurance coverage through a spouse or do not have insurance.
To address concerns that workers who have employer-provided health insurance may be less likely to move to self-employment simply because they already have a job with a good compensation package and high job quality, we first estimate difference-in-difference models. Identification of "entrepreneurship lock" arises from the interaction between having employer-provided health insurance and potential demand for health care. Using this first approach, we find some evidence that employer-based health insurance limits entrepreneurship, especially for men, but the evidence is not consistent across different measures of potential demand for health care. To improve the comparability of the experimental and control groups, we limit the sample to only individuals who have employer-based health insurance. Identification then comes from the interaction between having a spouse with employer-based health insurance and potential demand for health care. For men, we find consistent evidence of a larger negative effect of health insurance demand on the entrepreneurship probability for those without spousal coverage than for those with spousal coverage. Several robustness checks that further refine the comparability between experimental and controls groups provide similar results. Our estimates suggest that entrepreneurship lock for men is just over 1 percentage point. Relative to a base transition rate of 3 percent, this is a substantial effect. However, the evidence for women much weaker and tends to be imprecisely estimated. This is not surprising given that the entrepreneurship transition rate for women between 25 and 64 is half the male rate (3% vs. 1.4%) and this makes it difficult to precisely estimate entrepreneurship lock for women.
We also take a new approach in the literature to examining the question of whether employer-based health insurance discourages entrepreneurship by examining the discontinuity created at age 65 through the qualification for Medicare. Using a novel procedure of identifying age in months from matched monthly CPS data, we compare the probability of business ownership among male workers in the months just before turning age 65 and in the months just after turning age 65. Business ownership rates increase from 24.6 percent for those just under age 65 to 28.0 percent for those just over age 65, whereas we find no change in business ownership rates from just before to just after for the remaining ages in our sample of workers ages 55-75. We estimate several regression discontinuity models to confirm these results. As expected because of the small change in actual age and the orthogonality of included controls, we find a similarly large and statistically significant increase in self-employment rates in the age 65 birth month when the worker qualifies for Medicare. These results are not sensitive to several alternative samples, dependent variables, and age functions.
Overall we find some evidence that the
Ai, C. and E.C. Norton, 2003, Interaction Terms in Logit and Probit Model," Economic
Letters, 80, pp. 123-129
AWP Research. 2005 Health Care Survey. Final Report. Prepared for the National Assocation of the Self-Employed.
Berger Mark C; Black Dan A; Scott Frank A (2004), “Is There Job-Lock?: Evidence from the Pre-HIPAA Era,” Southern Economic Journal, 70: 953-976.
Birch, David. 1979. The Job Creation Process,
Bradford, William D. 2003. "The Wealth Dynamics of Entrepreneurship for Black and White Families in the
Bradley Cathy J; Neumark David; Luo Zhehui; Bednarkek Heather L (2005), “Employment-Contingent Health Insurance, Illness, and Labor Supply of Women: Evidence from Married Women with Breast Cancer,” NBER Working Paper No. 11304.
Buchmueller Thomas C; Valletta Robert G (1996). “The effects of employer-provided health insurance on worker mobility,” Industrial and Labor Relations Review, 49: 439-55.
Bucks, Brian K., Arthur B. Kennickell, and Kevin B.Moore. 2006. "Recent Changes in
Card, David, Carlos Dobkin, and Nicole Maestas. 2005. "The Impact of Health Insurance Status on Treatment Intensity and Health Outcomes."
Cooper, P.F. and A.C. Monheit (1993), ``Does employment-related health insurance inhibit job mobility?'' Inquiry 30: 400-416.
Currie, J. and B.C. Madrian (1999). “Health, Health Insurance, and the Labor Market,” in Orley Ashenfelter and David Card, eds., Handbook of Labor Economics, Volume 3, (Elsevier Science:
DeCicca, Philip (2007) Health Insurance Availability and Entrepreneurship: Evidence from
Egerstrom, Lee (2007), “Health Insurance: The Big Bump in the Road to Entrepreneurship.” Twin Cities Daily Planet. http://www.tcdailyplanet.net/article/2007/08/28/health-insurance-big-bump-road-entrepreneurship.html#
Fairlie, Robert W., and Rebecca A. London. 2008. "The Dynamics of Health Insurance Coverage: Identifying Trigger Events for Insurance Loss and Gain," (with Rebecca London), Health Services and Outcomes Research Methodology, (forthcoming).
Fairlie, Robert W., and Alicia M. Robb. 2008. Race and Entrepreneurial Success: Black-, Asian-, and White-Owned Businesses in the
Fronstin, Paul (2007) “Sources of Health Insurance and Characteristics of the Uninsured” EBRI Issue Brief 310.
Gilleskie Donna B; Lutz Byron (2002), “The Impact of Employer-Provided Health Insurance on Dynamic Employment Transitions,” Journal of Human Resources, 37: 129-162.
Girion, Lisa (2007), “Health Insurance Options Dwindle for the Self Employed”,
Gruber, J., and J. Poterba (1994): \Tax Incentives and the Decision to Purchase Health
Insurance: Evidence from the Self-Employed," Quarterly Journal of Economics, 109(3), 701-733.
Gruber, Jonathan and Brigitte C. Madrian (2004). “Health Insurance, Labor Supply and Job Mobility: A Critical Review of the Literature,” in Catherine G. McLaughlin, ed., Health Policy and the Uninsured (Urban Institute Press:
Gumus and Regan (2007) Tax Incentives as a Solution to the Uninsured:
Evidence from the Self-Employed. IZA DP No. 2866.
Holtz-Eakin, Douglas (1994), “Health Insurance Provision and Labor Market Efficiency in the
Economic Flexibility: Is There a Tradeoff? (
Kapur, Kanika. (1998). “The Impact of Health on Job Mobility: A Measure of Job Lock,”
Industrial and Labor Relations Review 51(2): 282-297.
Karoly, Lynn and Julie Zissimopoulos (2004) Self Employment and the 50+ population. AARP Research Brief.
Keen, Judy, 2005 Health Insurance Costs May Kill New-Business Dreams. Daily News Central. http://health.dailynewscentral.com/content/view/0001533/39/
Lichter, Daniel and and Janice Costanzo. (1987). How Do Demographic Changes Affect Labor Force Participation of Women?, Monthly Labor Review, Vol. 110, No. 11. November, pp. 23-25. Available on-line at: http://www.bls.gov/opub/mlr/1987/11/rpt2full.pdf (as of August 18, 2008).
McEwen, Helen, Pia Orrenius and Mark Wynne (2005). Opting Out of Work: What’s Beehind the Decline in Labor Force Participation? Southwest Economy, Issue 6, November, December. Available on-line at: http://www.dallasfed.org/research/swe/2005/swe0506a.html (as of August 18, 2008).
Madrian, Brigitte C. (1994b). “Employment-Based Health Insurance and Job Mobility: Is There Evidence of Job Lock?,” Quarterly Journal of Economics 109(1): 27-54.
Madrian, Brigitte C. and Lars John Lefgren (1998). “The Effect of Health Insurance on
Transitions to Self Employment,” unpublished paper (
Madrian (2006) The
National Federation of Independent Business. 2008. “Small Business Problems and Priorities.” http://www.nfib.com
OECD. 2005. SME and Entrepreneurship Outlook - 2005 Edition, Organisation for Economic Co-operation and Development Press.
Okunade Albert A; Wunnava Phanindra V (2002), “Availability of Health Insurance and Gender Differences in ‘Job-Lock’ Behavior: Evidence from NLSY,” Journal of Forensic Economics, 15: 195-204.
Parker, Simon C. 2004. The Economics of Self-Employment and Entrepreneurship.
Penrod, John R. (1994). “Empirical Essays in the Economics of Labor and Health.” Ph.D.Dissertation,
Perry, C.W. and H.S. Rosen, 2004, The Self-Employed are Less Likely than Wage Earners to Have Health Insurance than Wage-Earners. So What?" Public Policy and the Economics of Entrepreneurship. D. Holtz-Eakin and H. Rosen, ed.,
Reynolds, Paul. 2006. Entrepreneurship in the
Royalty, Anne and Jean Abraham, 2006 Health Insurance and Labor Market Outcomes: Joint Decision-Making within Households”, Journal of Public Economics, vol. 90 (8-9):1561-1577.
Sanz de Galdeano Anna (2006), “Job-Lock and Public Policy: Evidence from
D; Kniesner Thomas J (2001), “Chronic Illness and Health Insurance-Related Job Lock,” Journal of Policy Analysis and Management, 20: 525-44.
Velamuri (2005) Taxes, Health Insurance and Women's
Zissimpolous and Karoly, Transitions to Self- Employment at Older Ages RAND Working Paper WR-135 2003
 The 1996 Health Insurance Portability and Accountability Act (HIPAA) mandates that pre-existing condition limitations and waiting periods cannot be imposed on individuals who had continuous prior health insurance coverage, but it does not apply to individuals who do not have continuous prior coverage.
 Business owners are found to have higher saving rates and accumulate more wealth than wage/salary workers (Bradford 2003). Self-employed business owners hold nearly 40 percent of total
 In related work, Gruber and Poterba (1994) and Gumus and Regan (2007) use the variation provided by the tax subsidy to examine insurance coverage among the self-employed.
 Gruber and Madrian (1999) and Madrian (2006) provide detailed reviews of the existing literature.
 Prior to matching years we remove the supplemental samples to the 2001 to 2006 ADFs, which are generally not reinterviewed in the following March.
 Age in the second survey year is allowed to be in the range from -1 to +3 from the first survey year.
 The CPS asks separate questions about employer-provided (own and dependent), privately purchased, military, Medicaid, Medicare, Indian Health Service, and other sources of health insurance
 Full-time workers work 35 or more hours per week and 40 or more weeks a year.
 Over half of the uninsured newly self-employed were insured before becoming self-employed, and for these workers the move to self-employment resulted in a loss of health insurance.
 Bad or poor health is defined by individuals reporting that their health is "fair" or "poor" instead of "good," "very good," or "excellent." Anyone with bad health and the number with bad health in the family do not include the respondent. Spousal coverage is measured by using household, family and spouse identifiers for matching spouses, and information from each individual's employer health insurance coverage.
 Less direct measures of potential health demand include pregnancy and the total number of family members. Pregnancy is associated with additional labor supply changes, and the total number of family members is less precise that the number of family members in bad health.
 We also use spouse's employer-provided health insurance coverage or non-coverage to identify the control and experimental groups, respectively, as discussed below.
 In Xi, we include control variables for the individual’s job in the baseline year, family, individual demographics, residence, and survey year.
 The marginal effects for interaction terms in a probit model may be biased (Ai and Norton, 2003). Results in the paper are very similar using a linear probability model. In addition, we have calculated predictions of the marginal effects and their distribution and found a similar pattern of results, although these are somewhat more cumbersome to report.
 In our data, insurance holders are paid $25 per hour and 35% have college degrees compared to non-holders who are paid $18 per hour and 20% have college degrees. Among insurance holders, those who have spouse health insurance are very similar to those who do not have it – both groups are paid $25 per hour. Thirty-seven percent of those with both own and spouse insurance have college degrees compared to 34% of those with only own health insurance. Among those who have employer-provided health insurance, individuals with spouse health insurance have similar demographic characteristics, such as age and race, compared to those who do not have spouse health insurance. In contrast, employer-provided health insurance holders are older and more likely to be white compared to non-holders.
 We do not have information on whether the individual was offered health insurance and turned it down.
 Individuals who have both employer-provided health insurance and access to spouse health insurance may still have a preference for their own employer policy, and as a result, prefer to stay in their current job. This would result in an under-estimate of the effect of health insurance on self-employment transitions.
 We have also estimated the models with a measure of family health that includes the individual’s own health. Results using this measure are quite similar to the results reported in the paper.
 See Parker (2004) and Fairlie and Robb (2008) for recent reviews of the literature on the determinants of business ownership.
 One limitation of the basic monthly CPS data is that we have no information on health insurance coverage. We thus cannot distinguish individuals by demand for health insurance or care.