The Impact of Medicare Part D on Self-Employment

Abstract

We explore the relationship between access to affordable health insurance and self-employment using exogenous variation from the introduction of Medicare Part D that reduced the out-of-pocket cost of prescription drugs and improved health outcomes in a difference-in-differences model using the American Community Survey. We find that our treatment group of individuals aged 65–69 were 0.5 percentage points (or 5%) more likely to be self-employed in relation to a control group aged 60–64.

Keywords

self-employment entrepreneurship Medicare Part D health insurance

Prior research has shown that the skyrocketing costs of health care made changing jobs, entering self-employment, or retirement more expensive for older individuals. The Affordable Care Act (ACA) has been promoted as a tool to decrease this “job lock” by making insurance available in the private market at lower prices, but the ACA’s impact cannot presently be estimated. To gain some insight into the relationship between access to affordable health insurance and self-employment, we investigate the effect of Medicare Part D (Part D) on self-employment rates for older Americans.

We document the relationship between access to health insurance and self-employment both in the literature and using the American Community Survey (ACS). We analyze how changes in prescription drug insurance access affected self-employment. We employ the exogenous variation in prescription drug coverage among Medicare beneficiaries following the establishment of Part D in 2006 that reduced the out-of-pocket cost of prescription drugs and improved the health of participants. We use a difference-in-differences (DD) model to show that Part D increased the proportion of Medicare eligible (65 or over) individuals who were self-employed, using slightly younger individuals as a counterfactual. Although Part D is different in scope and targeted population compared to the ACA, our estimates are similar to existing research that employs variation in access to general health insurance or reductions in the cost.

The next section of the article provides a summary of research on the relationship between health insurance and employment with a focus on self-employment at the older ages. We then provide a background discussion of Part D as well as a review of relevant research on the program. As part of this literature review, we summarize some of the basic conditions that lead to self-employment. This background and review provides the basis for our research arguments leading to the research design section that presents the data set and method. Our results are discussed in the context of both the discussion over the effects of the ACA and recent interest in modifying public policy to encourage entrepreneurship at the older ages. Our discussion provides insight into the possible effects of Medicare and the ACA on self-employment.

Literature Review

Although there is ample research regarding the effect of health insurance on wage-and-salary work, self-employment and entrepreneurship,¹ the study of the effect of prescription drug insurance on these outcomes does not, to our knowledge, exist. Prescription drugs have become a cornerstone of modern medicine, and the coverage for these treatments is often subsumed within comprehensive health insurance plans. Therefore, we first review the prior research on work and self-employment at the older ages and then how other changes in health insurance coverage generally affect work and self-employment.

Work, Aging, and Self-Employment

Increasingly, older workers are engaging in a variety of work arrangements, such as self-employment (Zissimopolous & Karoly, 2009) and phased retirement, in which older workers gradually reduce time devoted to their career employment as a transitional step toward full retirement (Scott & Chen, 2003). A significant minority of older workers also engage in part-time work after holding career jobs and before fully retiring, which Quinn (1980) has labeled “bridge jobs.”

Scholars of self-employment usually identify certain factors that either “push” or “pull” a person into self-employment, but particular push and pull factors may be more salient at the older ages (Walker & Webster, 2007). However, as Parker (2006) notes, the body of theory and research that attempts to explain self-employment at older ages is “patchy.”

Push factors are those factors that push some away from wage-and-salary work or retirement and into self-employment. A number of scholars have identified key push factors such as unemployment, recessions, and short-run market conditions (Biehl, Gurley-Calvez, & Hill, 2013; Cahill, Giandrea, & Quinn, 2008). Push factors also include job loss or the loss of health or pension benefits (Blanchflower & Oswald, 1998; Moulton & Scott, 2016). Financial incentives from Social Security and pension plans may push older workers from wage-and-salary work at certain ages (Clark & Quinn, 2005).

Pull factors pull a person into self-employment when they cause the self-employment wage to be greater than the wage from a wage-and-salary job (Lombard, 2001). For example, older people may have an advantage over younger people due to their accumulated human, financial, and social capital that can be converted into entrepreneurial resources (Singh & DeNoble, 2003). Zissimopoulos and Karoly (2009) confirmed that wealth is positively associated with self-employment. In addition, Zissimopoulos and Karoly found that openness to risk enabled the transition to self-employment. Good health, which may be facilitated by improved access to prescription drugs, as we note below, and stable family or household relationships may also facilitate entry into self-employment and entrepreneurship (Berger & Pelkowski, 2004; Singh & Verma, 2003). Finally, Cahill, Giandrea, and Quinn (2008) suggest that self-employment may provide flexibility for workers looking for a different balance of work and leisure late in life.

Health Insurance

Since Part D is an expansion of health insurance coverage, we now review the literature that studies the relationship between health insurance and self-employment. Research has focused on possible job or entrepreneurship “lock,” or the theory that workers will not leave their current employment to engage in self-employment if leaving would mean losing health insurance (Cadden, Ortiz-Walters, & Schneider, 2012). Not all studies, however, have found a significant relationship between health insurance coverage and self-employment. Holtz-Eakin, Penrod, and Rosen (1996) found no impact of employer-provided health insurance on transitions from wage-and-salary employment to self-employment using the 1984–1986 waves of the Survey of Income and Program Participation and the 1982–1984 waves of the Panel Study of Income Dynamics. The authors suggest in their discussion that employer-provided health insurance may be a signal of job quality rather than a source of job lock. Bruce, Holtz-Eakin, and Quinn (2000) found similar results using transitions between the 1992 and 1996 waves of the Health and Retirement Study (HRS).

While subsequent studies have found significant associations between health insurance and self-employment, the findings have not been uniform. For those studies that found a relationship, an important factor is the source and nature of the health insurance coverage. Insurance can come from one’s employer or the employer of a spouse, and insurance can cover the individual only as an employee or also as a retiree. Given our interest in the effect of Part D, we highlight findings that are specific to older workers.

For example, Fairlie, Kapur, and Gates (2011) looked at the effect of Medicare eligibility at age 65 on changes in new business ownership using 10 years of data, 1996–2006, from the March Supplement to the Current Population Survey (CPS). Controlling for retirement, partial retirement, Social Security, and pension eligibility in the month individuals turn 65, they found that business ownership rates increased for those persons just under age 65 to just over age 65. In contrast, business ownership rates did not change over (other) ages in the range of 55–75.

In the case of married individuals, health insurance coverage through a spouse’s current employer may free a worker from job lock (Lombard, 2001; Wellington, 2001). Wellington (2001) used the 1993 CPS to estimate the impact of having health insurance through one’s spouse and found that the probability of business ownership increased by upwards of 3–4%. Fairlie et al. (2011) also found evidence to support that job lock decreased for those with coverage through their spouse.

Finally, many employers provided health insurance coverage after retirement, which often filled a gap prior to Medicare eligibility or even for services not covered by Medicare, although this coverage is declining. Zissimopolous and Karoly (2007) analyzed self-employment entry using the HRS 1992–2000 waves by including the different sources of coverage, such as health insurance from a current employer but no retiree coverage, health insurance from a current employer and access to retiree coverage, and no current employment coverage but access to retiree health insurance. In addition, these authors controlled for important gender distinctions in work and retirement. For example, men—and to some extent women—without employer-provided coverage had the highest likelihood of self-employment, while self-employment was less likely for those with employer coverage. Self-employment was more likely for those who had retiree health coverage as compared to those who only had coverage as an active worker.

Taking advantage of the fragmented nature of American health insurance, other studies used natural experiments to study changes in self-employment. In what is seen as a precursor to the ACA, the Commonwealth of Massachusetts enacted health-care reform, which took effect in 2006. Heim and Lurie (2014a) analyzed a panel of tax returns for the years 1999–2010. Their results modestly associate the reform with an increase in the number of older taxpayers who are earning the majority of income from self-employment.

These same authors also examined the effects of state-level insurance reforms such as community rating and guaranteed issue regulations in the individual market on self-employment (Heim & Lurie, 2014b). Using tax returns for 1987–2000, they found no statistically significant effect of the reforms on the propensity to be self-employed overall, although they did show evidence of an increase in self-employment among older taxpayers.

Taking advantage of legislation that increased the deductibility of health insurance premiums for the self-employed, Heim and Lurie (2010) again studied tax returns filed over 1999–2004. They found an increase in the probability of being self-employed of 1.5 percentage points, and an increase in the probability that a taxpayer would be primarily or exclusively self-employed by 1.1 and 0.35 percentage points, respectively. Velamuri (2012) also assessed the effects of these tax changes and found that the incidence of self-employment among single women rose between 10% and 13%. Gumus and Regan (2014) used CPS data from 1996 to 2007 and found that the increasingly generous tax deductions largely did not affect decisions to enter self-employment but had limited effects on exits.

The research that we summarize finds, with some variation, a connection between health insurance coverage and the ability to shift from wage-and-salary work to self-employment. Because our analysis is focused on the natural experiment posed by Part D, the next section provides background on the program.

Medicare Part D

The Medicare Prescription Drug, Improvement, and Modernization Act of 2003 established Part D, a new outpatient prescription drug coverage benefit. Part D began enrollment in January 2006 and marked the largest expansion to Medicare since the program’s inception. Prior to Part D, most Medicare beneficiaries received drug coverage through Medicaid, an employer, or a privately purchased plan. When Part D was implemented, many of these individuals substituted their existing coverage with a Part D plan or, in the case of Medicaid recipients, were automatically enrolled in one (Levy & Weir, 2009; Safran et al., 2010). The Part D enrollment rates—approximately 60%—were highest among the one quarter of beneficiaries who lacked prescription coverage before 2006 (Neuman et al., 2007; Safran et al., 2010). Part D now provides coverage to 70% of all Medicare beneficiaries (Hoadley, Summer, Hargrave, Cubanski, & Neuman, 2014).

The establishment of Part D may alter the incentives for a job change or retirement by weakening the link between traditional employment and prescription coverage. Prior to Part D, about one third of Medicare beneficiaries were enrolled in prescription coverage provided by their employer (Levy & Weir, 2009). The value of this particular benefit to employed beneficiaries increased considerably in the decades prior to Part D due to significant changes in the prescription drug market. From 1994 to 2004, the number of prescriptions filled per capita rose from 8 to 12 and almost all of those aged 65 or older were taking at least one prescription medication. Over that same period, prescription prices were increasing by 8.3% annually, over triple the inflation rate (Hargrave, Hoadley, Summer, Cubanski, & Neuman, 2010). By 2005, prior to Part D, Medicare beneficiaries paid nearly US$32.4 billion in out-of-pocket prescription expenses.² This amounted to an average annual expenditure of US$875 for beneficiaries although there was considerable heterogeneity in these expenses. At the time, out-of-pocket prescription expenses were highest among those without drug coverage. Sambamoorthi, Shea, and Crystal (2003) found that almost 16% of people without prescription drug coverage spent more than 10% of their household income on prescription medications compared to only 3.5% of those with coverage. Safran et al. (2005) estimate that 14% of those with no coverage had monthly out-of-pocket prescription costs that exceeded US$300, more than double the share of those with coverage at the time. The financial implications associated with the loss of their employer-provided prescription drug benefit may have contributed to lower job mobility or job lock among employed Medicare beneficiaries prior to Part D. Once established, these older workers were free to leave their employer without compromising their access to prescription coverage.

Part D may have contributed to higher rates of self-employment either through the program’s impact on household budgets or, potentially more importantly, on individual health outcomes. A number of studies show that Part D significantly reduced the out-of-pocket prescription expenditures of program enrollees even while increasing their prescription utilization (Engelhardt & Gruber, 2011; Kaestner & Khan, 2012). For example, Ketcham and Simon (2008) estimate that Part D was responsible for a decline of almost 22% in out-of-pocket expenditures among all beneficiaries. Yin et al. (2008) estimated a slightly smaller reduction in out-of-pocket prescription expenditures of almost 13% while increasing utilization by nearly 6%. By reducing the cost of prescription medication, Part D has reduced cost-related prescription nonadherence, which likely accounts for some of the health improvements linked to the program (Diebold, 2016; Kennedy, Maciejewski, Liu, & Blodgett, 2011). Afendulis, He, Zaslavsky, and Chernew (2011) show that Part D reduced the likelihood of Medicare beneficiaries being hospitalized for congestive heart failure by 6.4%, diabetes by 15.5%, and asthma by 12.1%. A related study by Kaestner, Long, and Alexander (2014) indicates that Part D reduced hospital admissions for congestive heart failure by 18%, coronary artery disease by 13%, and chronic obstructive pulmonary disorder by 32%. Most recently, Diebold (2016) shows that Part D improved the self-reported health status of newly covered beneficiaries and reduced the likelihood that this group is diagnosed with high blood pressure. Poor health status and chronic health conditions are associated with work impairments that result in a higher likelihood of labor force exits and a reduced likelihood of reentry (Blau & Gilleskie, 2001; Mutchler, Burr, Massagli, & Pienta, 1999; Sammartino, 1987; Zissimopoulos & Karoly, 2007). By improving beneficiaries’ access to prescription medications—most of which treat chronic, high-risk conditions (Hargrave et al., 2010)—Part D can help beneficiaries better manage their health and reduce the likelihood of a negative health shock, which may make it easier for older workers to leave their employer and transition into self-employment before they retire or reenter the labor market after retirement.

Research Design

Model

Part D served to exogenously reduce individuals’ out-of-pocket prescription drug costs and improve their health. This variation is helpful as some prior work acknowledged that it is difficult to disentangle the impact of health insurance from job quality or spousal labor force status. We investigate the impact of Part D on an individual’s decision to be self-employed using a DD research design. Equation 1 outlines a standard DD model.

{Self-Employed}_{i t} = α + β_{1} 1 ({Age}_{i t} \geq 65) + β_{2} 1 ({Year}_{t} > 2006) + β_{3} 1 ({Age}_{i t} \geq 65) \times 1 ({Year}_{t} > 2006) + δ X_{i t} + ∊_{i t} .

A DD includes two dimensions of treatment, where the interaction of the two represents the treatment state. A traditional DD estimates the differences in the outcome for a treated and control group before and after the treatment occurs, with the difference in the two differences estimating the treatment effect. Our treatment group is designated by Medicare eligibility and the time component is whether the observation is observed before or after Part D was in full effect. We classify those aged 65 or older as treated and those younger than 65 as the control group. This slightly younger group serves as the counterfactual for any change in self-employment for the treatment group over the time period observed.

The base model includes three indicator variables, the first equals one when the individual is 65 or over and the coefficient (β₁) estimates the pretreatment difference in the proportion self-employed in the treatment group (older sample) compared to the control group (younger sample). The second equals one when the sample year is past Part D’s introduction (after 2006) and the coefficient (β₂) estimates the change in self-employment over time for the control group that is used as a counterfactual for the treatment group. Lastly, the coefficient on the interaction of these two indicators (β₃) estimates the difference in self-employment for the treatment group between the pre- and post-Part D time periods, controlling for any difference for the control group over the same time period.

While not required for DD, we show that observables are relatively similar for these two groups, both before and after the Part D expansion and include a vector of controls (X_it ) described in the Data section to control for any differences. In most specifications, we use ages from 60 to 69, but because there may be concerns that those at the ends of the age range are dissimilar, we also provide estimates shrinking the range to include only 62–64 as the control group and 66–67 as the treatment group. We also provide estimates that drop age 65 since these individuals may have had little time to respond to the changing incentives and/or increased health.

Part D was implemented in 2006, but given that enrollment was open the entire year and our data set was collected throughout the year, we along with others³ exclude data from 2006 and consider 2007 the first full treatment year.⁴ In our base specification, we include only 2 years and designate 2005 as the pre-Part D time period and 2007 as the post. The most important assumption of DD is that the trends do not differ for the treatment and control groups prior to treatment—otherwise the control group’s difference cannot be used as a proper counterfactual for the treatment group. We show that there was no difference in the treatment and control group’s trends prior to the treatment, between 2001 and 2005. We also provide estimates that use data from 2001 to 2013 controlling for any potential trend differences using Equation 2. Equation 2 is identical to Equation 1 aside from including a linear year trend for the control and treatment groups rather than just the post 2006 indictor.

{Self-Employed}_{i t} = α + β_{1} 1 ({Age}_{i t} \geq 65) + γ_{1} {Year}_{t} + γ_{2} 1 ({Age}_{i t} \geq 65) \times {Year}_{t} + β_{2} 1 ({Year}_{t} > 2006) + β_{3} 1 ({Age}_{i t} \geq 65) \times 1 ({Year}_{t} > 2006) + δ X_{i t} + ∊_{i t} .

Data

The ACS (Ruggles, Genadek, Goeken, Grover, & Sobek, 2015) is a nationally representative sample that includes variables previously collected in the long-form decennial census, but on an annual basis. We retain U.S. citizens who did not have their citizenship, class of worker, or gender imputed by the Census Bureau. We classify individuals as self-employed if they reported that their “class of worker” was self-employed. We convert education to five indicator variables (no high school [omitted category], high school, some college, college, and graduate school). We create a female indicator and an indicator equal to 1 if the respondent reported that they are married (spouse present or absent). We simplified the race variable included in our summary statistics to “nonwhite” but include all possible race indicators in our regressions. We also include state of residence fixed effects to control for any potential time-invariant difference in the favorability of self-employment at the state level.

Results

Table 1 reports the summary statistics for the full sample in column 1 and stratified by year (2005 and 2007) and treatment status (age) in columns 2–5. Approximately 12% of the younger sample and 10% of the older sample are self-employed. The means for female, non-White, and married are very similar across the samples. Because educational attainment has been increasing over time, there are expectedly small differences in the educational outcomes between the younger and older samples, but including education as a control does not affect the estimates.

Table 1.

Summary Statistics.

	Full Sample	Pre-Part D		Post-Part D
Year	2001–2013	2005		2007
Age	60–69	<65	≥65	<65	≥65
Variables	(1)	(2)	(3)	(4)	(5)
Self-employed	0.106 (0.308)	0.123 (0.328)	0.098 (0.298)	0.120 (0.325)	0.102 (0.302)
Female	0.527 (0.499)	0.525 (0.499)	0.532 (0.499)	0.525 (0.499)	0.530 (0.499)
Non-White	0.140 (0.347)	0.130 (0.337)	0.132 (0.339)	0.136 (0.343)	0.135 (0.342)
Married	0.693 (0.461)	0.717 (0.450)	0.700 (0.459)	0.709 (0.454)	0.692 (0.461)
Education (no high school omitted)
High school	0.406 (0.491)	0.422 (0.494)	0.447 (0.497)	0.400 (0.490)	0.438 (0.496)
Some college	0.200 (0.400)	0.191 (0.393)	0.167 (0.373)	0.201 (0.400)	0.175 (0.380)
College	0.149 (0.356)	0.139 (0.346)	0.118 (0.323)	0.156 (0.363)	0.125 (0.331)
Graduate school	0.127 (0.333)	0.126 (0.332)	0.099 (0.299)	0.136 (0.343)	0.110 (0.313)
n	3,078,117	140,583	112,324	157,887	120,669

Source. U.S. Census American Community Survey 2001–2013 for those aged 60–69.

Note. Summary statistics are reported for the full sample and stratified by treatment status (age above or below 65) and pre- and post-Part D.

Table 2 includes the interaction (treatment effect) coefficient for several variants of Equations 1 and 2. It is important to note that while we control for time and treatment group in many different ways, the estimated effect of Part D is to increase self-employment roughly 0.5–0.6 percentage points, or 5% in relation to the base.⁵ We present the estimates from the simplest version of the DD (Equation 1) in column 1, as it includes one pre- and one posttime period (2005 and 2007), ages 60–69, and only the indicator variables for “age 65 or over” (treatment group), year over 2006 (2007, treatment time period), and the interaction of these variables. Column 2 is the same as 1 but adds the control variables listed in the Data section. Column 3 is the same as 2 but uses age fixed effects instead of a single age 65 or over to control for the pre-Part D difference between the treatment and control groups. Column 4 reverts to using the single age treatment indicator but omits age 65. Given that this is the first age for which the treatment occurs, the inclusion of these observations may introduce noise, as these individuals may have not had access to the treatment for long enough to respond. As expected, the effect is slightly larger in this model. Column 5 limits the sample to those aged 62–64 for the control group and 66–67 for the treatment group to limit differences between the two groups and the effects are similar. Column 6 uses Equation 2 and expands the years considered to 2001 through 2013 (still omitting 2006) and includes a linear year trend for the control and treatment groups.⁶ The effect size is slightly smaller. Column 7 mimics column 6 but includes age fixed effects. Our estimates are also robust to including federal and state unemployment rates (not presented in paper as the coefficients are unchanged) to control for labor market changes over time.

Table 2.

Difference-in-Differences: Self-Employment.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)
(Age ≥ 65) × (Year > 2006)	0.0057*** (0.0017)	0.0058*** (0.0017)	0.0056*** (0.0017)	0.0060*** (0.0018)	0.0053** (0.0025)	0.0051*** (0.0015)	0.0051*** (0.0015)
n	531,463	531,463	531,463	479,578	265,093	2,817,983	2,817,983
Years (2006 omitted)	2005–2007	2005–2007	2005–2007	2005–2007	2005–2007	2001–2013	2001–2013
Year control	(Year > 2006)	(Year > 2006)	(Year > 2006)	(Year > 2006)	(Year > 2006)	Year and Year × (Age ≥ 65)	Year and Year × (Age ≥ 65)
Ages (60–69 unless stated)				Omit 65	62–67, omit 65
Age control	(Age ≥ 65)	(Age ≥ 65)	Age FE	(Age ≥ 65)	(Age ≥ 65)	(Age ≥ 65)	Age FE
Controls		X	X	X	X	X	X

Source. U.S. Census American Community Survey 2001–2013 for those aged 60–69.

Note. The coefficients are from a difference-in-differences model where the Medicare eligible (age 65 or over) are the treated group and the first full treatment year is 2007 (the Part D enrollment took place throughout 2006, so it is often omitted in the literature). The ages and years included and the modeling of time varies depending on the column (see bottom panel of table for details). Controls include female, married, education indicators, all race indicators, and state of residence fixed effects. Heteroskedastic-robust standard errors are included in parentheses.

*Statistical significance at 10% level. **Statistical significance at 5% level. ***Statistical significance at 1% level.

To investigate the heterogeneous impact of the treatment effect by age, we include the estimates in Table 3 that repeats the model used in Table 2, column 3, but includes all the age fixed effects (instead of the age over 65 indicator) and their interaction with the post-Part D indicator. The interacted age fixed effects estimate the difference in the treatment effect for each age in relation to the omitted age 60. As expected, the coefficients are small and not significant for the control group (ages below 65). The effect emerges primarily for those aged 67 and over, but there is a 0.66 percentage point increase at age 65 that is significant at the 10% level. The larger effect for the older population is not unexpected as prescription drug usage and out-of-pocket costs increase as people age (National Center for Health Statistics, 2015).

Table 3.

Treatment Effect by Age.

	(1)
(Year > 2006) × 61	0.0014 (0.0037)
(Year > 2006) × 62	0.0017 (0.0036)
(Year > 2006) × 63	0.0035 (0.0037)
(Year > 2006) × 64	0.0031 (0.0037)
(Year > 2006) × 65	0.0066* (0.0037)
(Year > 2006) × 66	0.0039 (0.0038)
(Year > 2006) × 67	0.0116*** (0.0037)
(Year > 2006) × 68	0.0075** (0.0037)
(Year > 2006) × 69	0.0080** (0.0037)
n	531,463
Years (2006 omitted)	2005–2007
Year control	(Year > 2006)
Age control	Age FE
Controls	X

Source. U.S. Census American Community Survey 2005 and 2007 for those aged 60–69.

Note. The coefficients are from a difference-in-differences model similar to column 3 of Table 2 except it also includes age fixed effects interacted with the post indicator to determine whether treatment varied with age. Controls includes female, married, education indicators, all race indicators, and state fixed effects. Heteroskedastic-robust standard errors are included in parentheses.

*Statistical significance at 10% level. **Statistical significance at 5% level. ***Statistical significance at 1% level.

Robustness Check—Common Trends

The similar sized coefficients in Table 2, columns 6 and 7 compared to the coefficients in columns 1–5, provide preliminary evidence that there are not differential trends for the control and treatment groups. The validity of DD estimates depends on common trends in the outcome for each group, since the control group’s trend is used as a counterfactual for the treatment group. To further test this assumption, we provide the estimates in Table 4. Column 1 includes pretreatment years 2001–2005 and regresses self-employment on a linear time trend, a treatment group indicator (age ≥ 65), and the interaction of these two variables. The coefficient on the interaction between these two variables estimates the difference in the trend for the treatment group in relation to the control group. The coefficient is extremely small (0.01 percentage points) and is not statistically significant, indicating that there is no differential trend for the treatment group compared to the control group.

Table 4.

Robustness Check for Common Trends.

	(1)	(2)
Year trend × (Age ≥ 65)	0.0001 (0.0005)
Year trend	0.0020*** (0.0004)
(Age ≥ 65)	−0.2981 (1.0447)
2002 × (Age ≥ 65)		−0.0006 (0.0029)
2003 × (Age ≥ 65)		0.0013 (0.0028)
2004 × (Age ≥ 65)		−0.0001 (0.0028)
2005 × (Age ≥ 65)		0.0005 (0.0023)
Partial treatment year
2006 × (Age ≥ 65)		0.0030 (0.0023)
Posttreatment
2007 × (Age ≥ 65)		0.0062*** (0.0023)
2008 × (Age ≥ 65)		0.0065*** (0.0023)
2009 × (Age ≥ 65)		0.0094*** (0.0023)
2010 × (Age ≥ 65)		0.0137*** (0.0023)
2011 × (Age ≥ 65)		0.0123*** (0.0022)
2012 × (Age ≥ 65)		0.0127*** (0.0022)
2013 × (Age ≥ 65)		0.0144*** (0.0022)
n	633,116	3,078,117
Years	2001–2005	2001–2013

Source. U.S. Census American Community Survey 2001–2013 for those aged 60–69.

Note. Column 1 is a test to determine whether the trends are different for the treatment (age ≥ 65) and control (age < 65) samples prior to treatment. Column 2 includes a treatment indicator, year fixed effects (2001 omitted), the interaction of year fixed effects and the treatment indicator variable. The interacted variables test the common trends assumption by estimating the deviation between the treatment and control samples in relation to the difference at baseline for pretreatment years and also estimates the treatment effect for all posttreatment years. Heteroskedastic-robust standard errors are included in parentheses.

*Statistical significance at 10% level. **Statistical significance at 5% level. ***Statistical significance at 1% level.

To more robustly test for any pretreatment difference in the control and treatment groups’ self-employment trends, we regress self-employment on year fixed effects, a treatment group indicator, and the interaction of these variables. The treatment indicator removes any baseline difference in the proportion self-employed between the treatment and control groups. The noninteracted year fixed effects estimate differences between year t and the omitted year (2001) for the control group. The interaction of the year fixed effects and treatment indicator estimates the difference between the proportion self-employed for the treatment and the control group in year t compared to the difference in the omitted year. For instance, the first coefficient in column 2 indicates that in 2002, there was only a −0.06 percentage point difference in the proportion of the treatment group that was self-employed in relation to the control group, controlling for baseline differences. This is another way of testing if the trends differ for the treatment and control, without imposing a functional form on the trends. Note that there are no sizable or statistically significant differences prior to the treatment in 2006. Following 2006, the difference increases and is statistically significant. The coefficients from this table are plotted in the middle panel of Figure 1 along with the trends for the treatment and control groups in the top panel. Note that there is a small, but statistically insignificant jump in the partial treatment year 2006 that becomes larger and significant in 2007. There is no discernable increase in 2008, but the effect gets larger until 2010 when it appears to plateau at a roughly 1.25 percentage point increase. This estimate is larger than that found using our DD model.

Figure 1.

Self-employment over time. Source. U.S. Census American Community Survey 2001–2013 for those aged 60–69. Expenditure data from the 2007 to 2014 Annual Reports of the Boards of Trustees of the Federal Hospital Insurance and Federal Supplementary Medical Insurance Trust Funds, table II.B1. National Unemployment data from the Bureau of Labor Statistics. Note. The top panel plots the trend in self-employment for the treated (age ≥ 65) and control (age < 65) samples. The middle panel plots the coefficients from Table 4, column 2—or the difference in the deviation of the treated sample from the omitted year (2001) in relation to the control sample’s deviation. The thicker lines in the top panel correspond to the pre-Part D trends. The vertical dashed lines and gray portion in the middle panel demarcate the partial treatment year 2006. The bottom panel plots annual Part D expenditures and the national unemployment rate.

It is difficult to disentangle the impact of the Great Recession, the recovery, and the potential long-term effects of reducing out-of-pocket prescription drug expenditures and increasing health status on self-employment, but to gain some insight we plot annual Part D expenditures⁷ and the national unemployment rate in the bottom panel of Figure 1. We highlight the strong visual correlation between the Part D expenditures and self-employment differences (middle panel) that is not detectable for the unemployment rate. Exploratory regression analysis further confirms the visual findings.⁸ While it is possible that these macro-level events had an effect, it appears that Part D played a significant role in increasing self-employment among the Medicare-eligible population.

Conclusion

Using a natural experiment, we have shown that access to more affordable health insurance in the form of cheaper prescription drugs is an important factor in the self-employment decision-making process for older individuals. We show that a large expansion to health insurance affordability and access, Part D, increased the likelihood of older individuals being self-employed. Increasing access to affordable health insurance seems to be an effective tool for policy makers who want to encourage job mobility, self-employment, and entrepreneurship, especially at the older ages.

The estimates in the paper suggest a 0.5 percentage point or 5% increase in self-employment as a result of Part D. Prescription drug benefits are but one part of the whole health insurance coverage, such that we may not expect very large effects compared to an expansion in overall health insurance coverage. However, this effect is quite similar in size to having access to health insurance through a spouse (at 3–4%, Wellington, 2001) or increasing the tax deductibility of health insurance for the self-employed (at 1.1 percentage points for primarily self-employed and 0.35 percentage points for the exclusively self-employed; Heim and Lurie, 2010).

Although it is too early to investigate the impact of the ACA, if our results or even their general direction can be generalized from reductions in the cost of prescription drug coverage for older individuals to reductions in the cost of health insurance or new access through the ACA for the general population, then they indicate that the ACA may result in reduced job lock, allowing more individuals to enter and remain in self-employment. As Baby Boomers continue to age, a reduction in job lock may mean that older Americans will work longer due to improved health, perhaps in part-time jobs such as bridge jobs, in self-employment, or in other positions that may be better suited to their personal situations.

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

References

Afendulis

C. C.

Zaslavsky

Chernew

(2011). The impact of Medicare Part D on hospitalization rates. Health Services Research, 46, 1022–1038.

Berger

M. C.

Pelkowski

J. M.

(2004). Health and labor force transitions. Quarterly Journal of Business and Economics, 42, 113–138.

Biehl

A. M.

Gurley-Calvez

Hill

(2013, 4). Self-employment of older Americans: Do recessions matter? Small Business Economics. doi:10.1007/s11187-013-9479-7. Retrieved from https://link-springer-com.web.bisu.edu.cn/article/10.1007/s11187-013-9479-7

Blanchflower

Oswald

A. J.

(1998). What makes an entrepreneur? Journal of Labor Economics, 16, 26–60.

Blau

D. M.

Gilleskie

D. B.

(2001). The effect of health on employment transitions of older men. In Polachek

(Ed.), Worker wellbeing in a changing labor market. Research in labor economics (Vol. 20, pp. 35–65). Amsterdam, the Netherlands: JAI Press.

Bruce

Holtz-Eakin

Quinn

(2000). Self-employment and labor market transitions at older ages, Boston College Department of Economics (Working Paper 490). Retrieved from http://fmwww.bc.edu/EC-P/wp490.pdf

Cadden

D. T.

Ortiz-Walters

Schneider

G. P.

(2012). Employer-based health insurance: A barrier to small business formation. Proceedings of the Academy of Entrepreneurship, 18. Retrieved October 8, 2014, from http://alliedacademies.org/public/proceedings/Proceedings31/AE_Proceedings_18(2)_Fall_2012.pdf#page=9

Cahill

K. E.

Giandrea

M. D.

Quinn

J. F.

(2008, 2). A micro-level analysis of recent increases in labor force participation among older workers (Working Paper 2008-8). Chestnut Hill, MA: Center for Retirement Research, Boston College.

Clark

R. A.

Quinn

J. F.

(2005). Effects of pensions on labor markets and retirement. In Gale

Shoven

Warshawsky

(Eds.), The evolving pension system: Trends, effects, and proposals for reform (pp. 77–102). Washington, DC: Brookings Institution.

10.

Diebold

(2016). The effects of medicare part D on health outcomes of newly covered medicare beneficiaries. The Journals of Gerontology Series B: Psychological Sciences and Social Sciences. doi:10.1093/geronb/gbw030

11.

Engelhardt

G. V.

Gruber

(2011). Medicare Part D and the financial protection of the elderly. American Economic Journal: Economic Policy, 3, 77–102.

12.

Fairlie

R. W.

Kapur

Gates

(2011). Does employer-based health insurance limit entrepreneurship? Journal of Health Economics, 30, 146–162.

13.

Gumus

Regan

T. L.

(2014). Self-employment and the role of health insurance in the U.S. Journal of Business Venturing. Retrieved from https://www-sciencedirect-com.web.bisu.edu.cn/science/article/pii/S0883902614000123

14.

Hargrave

Hoadley

Summer

Cubanski

Neuman

(2010). Medicare Part D data spotlight: Coverage of top brand-name and specialty drugs. Kaiser Family Foundation. Retrieved November 4, 2014, from http://kff.org/medicare/issue-brief/coverage-of-top-brand-name-and-specialty/

15.

Heim

B. T.

Lurie

I. Z.

(2010). The effect of self-employed health insurance subsidies on self-employment. Journal of public Economics, 94, 995–1007.

16.

Heim

B. T.

Lurie

I. Z.

(2014a, 4 3). Does health reform affect self-employment? Evidence from Massachusetts. Small Business Economics, 35–49. doi:10.1007/s11187-014-9572-6

17.

Heim

B. T.

Lurie

I. Z.

(2014b). Did reform of the non-group health insurance market affect the decision to be self-employed? Evidence from state reforms in the 1990s. Health Economics, 23, 841–860. doi:10.1002/hec.2960

18.

Hoadley

Summer

Hargrave

Cubanski

Neuman

(2014, 8). Medicare Part D in its ninth year: The 2014 marketplace and key trends, 2006–2014 (Kaiser Family Foundation Report). Retrieved November 4, 2014, from http://files.kff.org/attachment/medicare-part-d-in-its-ninth-year-the-2014-marketplace-and-key-trends-2006-2014-report

19.

Holtz-Eakin

Penrod

J. R.

Rosen

H. S.

(1996). Health insurance and the supply of entrepreneurs. Journal of Public Economics, 62, 209–235.

20.

Kaestner

Khan

(2012). Medicare Part D and its effect on the use of prescription drugs and use of other health care services of the elderly. Journal of Policy Analysis and Management, 31, 253–279.

21.

Kaestner

Long

Alexander

G. C.

(2014). Effects of prescription drug insurance on hospitalization and mortality: Evidence from Medicare Part D (NBER Working Paper No. 19948). Cambridge, MA: National Bureau of Economic Research.

22.

Kennedy

J. J.

Maciejewski

Liu

Blodgett

(2011). Cost-related nonadherence in the Medicare program: The impact of Part D. Medical Care, 49, 522–526.

23.

Ketcham

J. D.

Simon

(2008, 9). Medicare Part D’s effects on elderly drug costs and utilization (NBER Working Paper No. 14326). National Bureau of Economic Research. Retrieved November 4, 2014, from http://www.nber.org/papers/w14326

24.

Levy

Weir

D. R.

(2009). Take-up of Medicare Part D: Results from the health and retirement study. Journals of Gerontology Series B: Psychological Sciences & Social Sciences, 65B, 492–501.

25.

Lombard

K. V.

(2001). Female self-employment and demand for flexible, nonstandard work schedules. Economic Inquiry, 29, 214–317.

26.

Mahmoudi

Jensen

G. A.

(2014). Has Medicare Part D reduced racial/ethnic disparities in prescription drug use and spending? Health Services Research, 49, 502–525.

27.

Moulton

J. G.

Scott

J. C.

(2016). Opportunity or necessity: Disaggregating self-employment and entry at the older ages. Social Forces, 94, 1539–1566.

28.

Mutchler

J. E.

Burr

J. A.

Massagli

M. P.

Pienta

(1999). Work transitions and health in later life. The Journals of Gerontology Series B: Psychological Sciences and Social Sciences, 54, S252–S261.

29.

National Center for Health Statistics. (2015). Health, United States, 2015. Washington: U.S. Government Printing Office. Retrieved from http://www.cdc.gov/nchs/data/hus/hus15.pdf#079

30.

Neuman

Strollo

M. K.

Guterman

Rogers

W. H.

Rodday

A. M.

Safran

D. G.

(2007). Medicare prescription drug benefit progress report: Findings from a 2006 national survey of seniors. Health Affairs (Project Hope), 26, w630–x643.

31.

Parker

S. C.

(2006). Entrepreneurship, self-employment and the labour market. In Casson

Yeung

Basu

Wadeson

(Eds.), The Oxford handbook of entrepreneurship (pp. 435–460). Oxford, England: Oxford University Press.

32.

Quinn

J. F.

(1980). Labor force participation patterns of older self-employed workers. Social Security Bulletin, 43, 17–28.

33.

Ruggles

Genadek

Goeken

Grover

Sobek

(2015). Integrated public use microdata series: Version 6.0 [Machine-readable database]. Minneapolis: University of Minnesota.

34.

Safran

D. G.

Neuman

Schoen

Kitchman

M. K.

Wilson

I. B.

Cooper

… Rogers

W. H.

(2005). Prescription drug coverage and seniors: Findings from a 2003 national survey. Health Affairs. doi:10.1377/hlthaff.w5.152.

35.

Safran

D. G.

Strollo

M. K.

Guteman

Rogers

W. H.

Neuman

(2010). Prescription coverage, use and spending before and after Part D implementation: A national longitudinal panel study. JGIM: Journal of General Internal Medicine, 25, 10–17.

36.

Sambamoorthi

Shea

Crystal

(2003). Total and out-of-pocket expenditures for prescription drugs among older persons. The Gerontologist, 43, 345–359.

37.

Sammartino

F. J.

(1987). Effect of health on retirement. Social Security Bulletin, 50, 31.

38.

Scott

J. C.

Chen

Y.-P.

(2003). Gradual retirement: An additional option in work and retirement. North American Actuarial Journal, 7, 62–74.

39.

Singh

DeNoble

(2003). Early retirees as the next generation of entrepreneurs. Entrepreneurship Theory & Practice, 27, 207–226.

40.

Singh

Verma

(2003). Work history and later-life labor force participation: Evidence from a large telecommunications firm. Industrial and Labor Relations Review, 56, 699–715.

41.

Velamuri

(2012). Taxes, health insurance, and women’s self-employment. Contemporary Economic Policy, 30, 162–177. doi:10.1111/j.1465-7287.2011.00256.x

42.

Walker

E. A.

Webster

B. J.

(2007). Gender, age and self-employment: Some things change, some stay the same. Women in Management Review, 22, 122–135.

43.

Wellington

A. J.

(2001). Health insurance coverage and entrepreneurship. Contemporary Economic Policy, 19, 465–478.

44.

Yin

Basu

Zhang

J. X.

Rabbani

Melitzer

D. O.

Alexander

G. C.

(2008). The effect of the Medicare Part D prescription benefit on drug utilization and expenditures. Annals of Internal Medicine, 148, 169–177.

45.

Zissimopoulos

Karoly

L. A.

(2007). Transitions to self-employment at older ages: The role of wealth, health, health insurance, and other factors. Labour Economics, 14, 269–295.

46.

Zissimopoulos

Karoly

L. A.

(2009). Labor force dynamics at older ages: Movements into self-employers by workers and nonworkers. Research on Aging, 31, 89–111.