Abstract
Background
Self‐employed workers are 10% of the US labor force, with growth projected over the next decade. Whether existing policy mechanisms are sufficient to ensure health insurance coverage for self‐employed workers, who do not have access to employer‐sponsored coverage, is unclear.
Objective
To determine whether self‐employment is associated with lack of health insurance coverage.
Data Sources
Secondary analysis of Medical Expenditure Panel Survey (MEPS) data collected 2014‐2017.
Study Design
Participants were working age (18‐64 years), employed, civilian noninstitutionalized US adults with two years of Medical Expenditure Panel Survey (MEPS) participation in 2014‐2017. We compared those who were employees vs those who were self‐employed. Key outcomes were self‐report of health insurance coverage, and of delaying needed medical care.
Data Extraction Methods
Longitudinal design among individuals who were employees during study year 1, comparing health insurance coverage among those who did vs did not transition to self‐employment in year 2.
Principal Findings
16 335 individuals, representing 121 473 345 working‐age adults, met inclusion criteria; of these, 147, representing 1 097 582 individuals, transitioned to self‐employment. In unadjusted analyses, 25.7% of those who became self‐employed were uninsured in year 2, vs 8.1% of those who remained employees (P < .0001). In adjusted models, self‐employment was associated with greater risk of being uninsured (26.1% vs 8.0%, risk difference 18.0%, 95% confidence interval [CI] 9.2% to 26.9%, P = .0001). A time‐by‐employment type product term suggests that 10.0 percentage points (95%CI 0.3 to 19.7 percentage points, P = .04) of the risk difference may be attributable to the change to self‐employment. Self‐employment was also associated with delaying needed medical care (12.0% vs 3.1%, risk difference: 8.9%, 95% CI 3.1% to 14.6%, P = .003).
Conclusions
One in four self‐employed workers lack health insurance coverage. Given the rise in self‐employment, it is imperative to identify ways to improve health care insurance access for self‐employed working‐age US adults.
Keywords: employment, health insurance, medically uninsured, self‐employment
What This Study Adds.
1. What is Known?
Self‐employment is a growing trend in the American workforce.
Many working‐age adults receive employer‐sponsored health insurance, but this is typically not available to self‐employed workers.
The extent to which those who are self‐employed are covered by health insurance is not known.
2. What this Study Adds
Nearly 1 in 4 self‐employed workers is uninsured, which is much greater than those who are employees.
Self‐employed workers report higher rates of delaying needed medical care.
As policymakers grapple with how to respond to the changing nature of work in the United States, facilitating health insurance coverage for self‐employed workers is an important policy goal.
1. INTRODUCTION
Self‐employment is increasing in the American workforce. The US Bureau of Labor Statistics reports that as of 2016, 10% of all workers are self‐employed, and that, after trending downward for the last two decades, growth in self‐employment is now expected to outpace overall growth in the labor force through at least 2026. 1 , 2 , 3 , 4 One potential explanation for this rise in self‐employment is the expansion in job opportunities afforded by app‐based services, such as Uber, Instacart, and TaskRabbit. A recent report found that this type of self‐employment work increased by 15% from 2010 to 2019, 5 with signs that it will continue to grow in the next decade. Another explanation is that employers increasingly seek to classify workers as self‐employed independent contractors rather than employees. 3
In response to this changing nature of work in the United States, some states have created regulatory reforms to addresses emerging challenges. These efforts are exemplified by the recent passage of California bill AB5, 6 which makes it more difficult to classify certain workers as independent contractors. However, regulatory changes have not yet directly addressed the possible health consequences of self‐employment. Designing policies and other interventions to do so is hampered by the lack of research on the potential health consequences of self‐employment.
A key pathway whereby self‐employment could impact health is through health insurance coverage. In the United States, employer‐sponsored health insurance has historically been the predominant source of coverage for working‐age adults. 7 The 2014 implementation of the 2010 Patient Protection and Affordable Care Act required all individuals to obtain insurance coverage, and sought to make this feasible by expanding Medicaid eligibility, establishing health insurance exchanges, and subsidizing exchange‐purchased insurance for some individuals in a means‐tested fashion. 8 This led to greater insurance coverage. 9 However, subsequent legislative and regulatory changes to the Affordable Care Act have been associated with increases in adults lacking health insurance coverage. 7 , 9
Prior research examined whether lack of employer‐sponsored health insurance is an impediment to entrepreneurism (ie, transitions to self‐employment), and whether those who seek self‐employment are healthier than those who remain employees. 10 , 11 , 12 , 13 , 14 However, these studies did not assess the impact on health insurance coverage among persons who transition to self‐employment. This is important to study, as numerous studies show that people lacking health insurance often forego necessary medical care. 15 , 16 , 17 , 18 , 19 , 20
To assess the potential impact of self‐employment on health insurance coverage, we used a nationally representative longitudinal data set with detailed, self‐reported employment and insurance coverage information. We used a longitudinal design to examine workers who switched from being employees to being self‐employed (compared with those who remained employees). We hypothesize that self‐employment would be associated with greater risk of lacking health insurance coverage, and greater risk of deferred care.
2. METHODS
2.1. Data source
Data for this study were obtained from the longitudinal data files of the Medical Expenditure Panel Survey (MEPS), which is a person‐year data format. 21 MEPS is a nationally representative survey of health care costs and usage among noninstitutionalized US civilian adults. 21 Because this project used only de‐identified, publicly available data, it was deemed nonhuman subjects research by the UNC IRB. 17
2.2. Setting and participants
We used data from three longitudinal MEPS panels: MEPS panel 19 (years 2014‐2015), panel 20 (2015‐2016), and panel 21 (2016‐2017). Panel 21 was the most recent MEPS panel with full data at the time of the study. MEPS has a longitudinal design which follows individuals for five rounds of interviews over a two‐year period. 21 This means that each participant has an initial year of MEPS participation (“year 1,” corresponding to either 2014, 2015, or 2016 depending on the MEPS panel) and a follow‐up year of participation (“year 2,” corresponding to the year following their initial participation). The longitudinal data files do not include individuals with only partial participation, so follow‐up is complete within the files. The longitudinal files contain weights that can be used to produce nationally representative estimates.
The study years selected for these analyses occurred after implementation of the Patient Protection and Affordable Care Act, but prior to major changes that went into effect in 2019 (eg, the repeal of the fine for violating the individual mandate to obtain coverage). Therefore, regulations were consistent across the study period. We included all MEPS participants aged 18‐64 years (working‐age adults) at the time their MEPS participation began. Lack of employment is known to be associated with lack of health insurance, 22 so we further restricted inclusion to those who were employed at the beginning of both year 1 and year 2 of their MEPS participation; this focused the study on type of employment rather than whether an individual was employed.
2.3. Exposure: employment type
The exposure of interest was employment type (self‐employment vs being an employee). MEPS respondents who reported being currently employed or had a job to return to (eg, they were currently on leave) were considered employed. Those who were employees for at least one job and had an additional self‐employed job were categorized as employees.
Change from employee to self‐employed was determined as follows. MEPS queries respondents on employment type at multiple time points throughout their 2‐year participation. To determine study eligibility, we used responses obtained in the 1st assessment period, which covers January 1 of the year they begin study participation through their first interview round, in quarter 1 or quarter 2 of that year. To determine whether an individual changed to self‐employment or remained an employee, we used responses to questions asked during the 3rd interview round, which corresponds to the period around the beginning of their second year of participation. This means that initial employment classification corresponds to the beginning of the first year of study participation, and the next employment classification corresponds to the beginning of the second year of study participation. To understand the types of jobs participants were working in (eg, construction, manufacturing), we used US Census Bureau 2007 Industry codes. 23 , 24
2.4. Outcomes
The primary outcome was whether an individual had health insurance of any type (public or private, primary beneficiary or as a dependent), for any portion of their second year of MEPS participation. Being uninsured under this definition means lacking coverage for the entire year. We used this conservative definition of being uninsured because we thought lacking insurance for at least a full year would have particular policy relevance. Because of the longitudinal data analysis framework used (see Statistical Analysis section, below), we also included year 1 insurance information in our models, categorized in the same way.
Secondary outcomes included type of insurance in year 2, categorized as private, private and exchange‐based, and public insurance. An individual can have more than one of these coverage types during a given year and thus be included as having the specific type of coverage in more than one secondary analysis.
Last, lack of insurance may affect health by limiting access to needed health care. Therefore, we also assessed whether participants reported delaying necessary medical care.
2.5. Covariates
We considered several covariates that could confound the association between employment type and health insurance coverage. The covariates were as follows: age, gender, race/ethnicity, educational attainment (less than high school diploma, high school diploma/GED, and greater than high school diploma), family income (percentage of the federal poverty guideline for the appropriate year and household size), primary language (English, other), census region (Northeast, Midwest, South, West), self‐reported health status (excellent, very good, good, fair, and poor) and, to account for secular trends, the calendar year in which data were collected. Census region, self‐reported health status, income, and year of data collection were considered time‐varying covariates and so assessments from both year 1 and year 2 were used. Other covariates were considered time‐fixed, so only year 1 data were used.
2.6. Study design and statistical analysis
To enhance internal validity, we used a longitudinal study design that included individuals who were employed at the start of years 1 and 2 but may have changed employment type during year 2. By thus allowing individuals to serve as their own controls, this accounted for unmeasured time‐invariant covariates within participants (eg, personal preferences, prevalent illness, prior work experience, and skills).
For the statistical analysis, we conducted mixed‐effects repeated‐measures regression analyses, recording an observation for each participant in year 1 and year 2 of their MEPS participation (a “person‐year” data format). Models contained a term for employment type (self‐employed vs. employee), time (year 1 or year 2), and a time‐by‐employment type product term. The models also contained the above‐described covariates for adjustment. Models used a random‐effects term to account for repeated assessments within individuals. Owing to difficulties in interpreting results from nonlinear models, we fit linear working models. 25 , 26 There are two results from these models that are of interest. The first is the adjusted estimates of the year 2 risk of being uninsured (or other secondary outcomes). This presents the estimated risk of each outcome for those who are self‐employed and those who employees. The second is the time‐by‐employment type product term, which estimates what amount of the difference in year 2 risk of lacking health insurance, if any, may be attributable to the change to self‐employment, rather than the time‐fixed characteristics of those who change to self‐employment.
For hypothesis testing that compared the adjusted year 2 absolute risk of each outcome in those who changed from being employees to being self‐employed with those who remained employees, we used the Stata command margins, which gives point estimates, 95% confidence intervals from delta method standard errors, and p‐values from F statistics. For hypothesis testing regarding the time‐by‐employment type product term, we used P‐values from the coefficient of that term in the linear mixed model. For analyses of delayed medical care, we fit models both with and without adjustment for health insurance, to examine what role, if any, insurance coverage played in delaying needed medical care.
We conducted several sensitivity analyses. First, as our main analyses focused on persons who were employees in year 1, then did or did not change to self‐employment in year 2, we conducted the same outcome assessment among persons who were self‐employed at baseline, then did or did not change to being employees, to ensure that our results were not sensitive to choice of cohort construction. Second, to ensure the results were not sensitive to the choice of the functional form of the income variable, we fit models where income was treated as a 5‐knot restricted cubic spline. Third, to avoid issues with endogeneity of year 2 self‐reported health status, we fit models without this variable. Next, given the small percentage of individuals who transition from being employees to being self‐employed in any given year, we also fit cross‐sectional models examining difference in insurance coverage based on prevalent year 1 employment type (ie, comparing those who were self‐employed in year 1 to those who were employees in year 1). In addition, we conducted analyses that additionally adjusted for the number of hours worked per week. Finally, we conducted analyses that used only year 1 income, treating it as time‐fixed covariate, as year 2 income may be influenced by the transition to self‐employment.
The statistical software packages used for this project were SAS version 9.4 (SAS Institute) and Stata MP 16.0. Unadjusted comparisons used t tests for continuous variables and chi‐squared tests for categorical variables. All analyses used MEPS sampling weights to obtain nationally representative estimates, and accounted for the complex sampling design, and its implications for standard error estimation, using the svy commands in Stata. A two‐sided P‐value of <.05 indicated statistical significance.
3. RESULTS
There were 19 617 working‐age adults included in the three MEPS panels, representing 145 468 170 individuals. Of these, 1892 (representing 15 158 841 individuals; weighted percentage: 11.6%) were self‐employed at baseline and thus excluded from the main analysis cohort (Table S1). Of the remaining individuals, 1390 were either no longer in scope (ie, not eligible for further MEPS participation) or not working at follow‐up. The remaining 16 335 individuals, representing 121 473 345 working‐age adults, met inclusion criteria for the main analysis cohort. Of these, 147, representing 1 097 582 individuals, transitioned to self‐employment in year 2. Those who transitioned to self‐employment were of similar age, gender, race/ethnicity, and educational attainment, and had similar family income, compared with those who remained employees (Table 1). Private insurance coverage was the largest source of health insurance coverage at baseline in this group (84.0%). Of those who transitioned to self‐employment, 16.6% were uninsured during year 1, and of those who did not transition to self‐employment, 9.2% were uninsured during year 1. Industries that were more common among those who transitioned to self‐employment, compared with those who remained employees, included construction and professional and business services (Table S2).
TABLE 1.
Characteristics in year 1 of MEPS participants who were employees in year 1
| Overall | Employee in Year 2 | Self‐employed in Year 2 | |
|---|---|---|---|
| N | 16 335 | 16 188 | 147 |
| Weighted N | 121 473 345 | 120 375 762 | 1 097 582 |
| Mean (SD) or N (weighted %) | Mean (SD) or N (weighted %) | Mean (SD) or N (weighted %) | |
|---|---|---|---|
| Age, y | 40.72 (12.45) | 40.73 (12.46) | 39.72 (11.80) |
| Female | 7959 (48.4) | 7889 (48.3) | 70 (52.2) |
| Race/ethnicity | |||
| Hispanic | 4946 (17.1) | 4896 (17.1) | 50 (18.5) |
| Non‐Hispanic White | 6589 (61.6) | 6528 (61.6) | 61 (63.9) |
| Non‐Hispanic Black | 2938 (11.9) | 2922 (12.0) | 16 (6.5) |
| Non‐Hispanic Asian | 1394 (6.4) | 1380 (6.4) | 14 (8.4) |
| Non‐Hispanic other or multi | 468 (2.9) | 462 (2.9) | 6 (2.7) |
| Educational attainment | |||
| <HS diploma | 2417 (8.6) | 2397 (8.6) | 20 (5.8) |
| HS diploma or GED | 4522 (25.8) | 4485 (25.8) | 37 (24.4) |
| >HS diploma | 9271 (65.6) | 9182 (65.6) | 89 (69.8) |
| Family income as percentage of poverty level | 471.28 (347.14) | 471.66 (346.93) | 428.99 (368.86) |
| Private insurance | 12 343 (84.0) | 12 255 (84.2) | 88 (69.7) |
| Exchange‐based insurance | 522 (2.8) | 516 (2.8) | 6 (6.4) |
| Medicare | 74 (0.4) | 73 (0.4) | 1 (1.1) |
| Medicaid or other public insurance | 2022 (8.9) | 1996 (8.9) | 26 (17.6) |
| Uninsured | 2379 (9.2) | 2341 (9.2) | 38 (16.6) |
| English as primary language | 13 436 (91.9) | 13 318 (91.9) | 118 (91.6) |
| Census region | |||
| Northeast | 2415 (17.3) | 2399 (17.3) | 16 (11.9) |
| Midwest | 3263 (21.9) | 3240 (22.0) | 23 (18.5) |
| South | 6077 (37.4) | 6019 (37.3) | 58 (41.4) |
| West | 4580 (23.4) | 4530 (23.3) | 50 (28.2) |
| Self‐reported health status | |||
| Excellent | 4745 (30.3) | 4701 (30.3) | 44 (29.9) |
| Very Good | 5537 (36.9) | 5484 (36.9) | 53 (36.0) |
| Good | 4386 (24.8) | 4351 (24.8) | 35 (22.9) |
| Fair | 1453 (7.0) | 1443 (6.9) | 10 (8.2) |
| Poor | 199 (1.0) | 195 (1.0) | 4 (3.0) |
| Delay care | 470 (3.1) | 460 (3.1) | 10 (8.8) |
| First year of MEPS participation | |||
| 2014 | 5391 (32.5) | 5344 (32.4) | 47 (35.0) |
| 2015 | 5710 (33.7) | 5653 (33.6) | 57 (36.1) |
| 2016 | 5234 (33.9) | 5191 (33.9) | 43 (29.0) |
Insurance categories do not sum to 100% as individuals can have more than one status in a given year; however, the uninsured category represents lacking health insurance the entire year. Descriptive statistics are presented here to transparently summarize the sample analyzed. For cells with small sizes (<20), the Agency for Healthcare Research and Quality provides guidelines for the precision of the estimates presented, and note that owing to a high potential for imprecision, these estimates should not necessarily be interpreted as nationally representative. For further details see: https://meps.ahrq.gov/survey_comp/ic_precision_guidelines.shtml.
Abbreviations: GED, General Educational Development; MEPS, Medical Expenditure Panel Survey.
In unadjusted analyses, year 2 income levels were similar among those who became self‐employed and those who remained employees (Table 2). However, 25.7% of those who transitioned to self‐employment were uninsured in year 2, compared with 8.1% of those who remained employees (P < .0001). In models adjusted for age, gender, race/ethnicity, education, income, primary language, census region, self‐reported health status, and year of survey, self‐employment was associated with significantly greater risk of being uninsured in year 2 (26.1% vs 8.0%, risk difference 18.0%, 95% confidence interval [CI] 9.2% to 26.9%, P = .0001) (Table 3, full model results in Table S3). The time‐by‐employment type product term suggests that 10.0 percentage points (95%CI 0.3 to 19.7 percentage points, P = .04) of the risk difference may be attributable to the change to self‐employment. Sensitivity analyses that modeled income as a restricted cubic spline, analyses that treated income as fixed, that additionally adjusted for hours worked, and that did not include self‐reported health status, yielded results similar to those described here (Table S4). Adjusted models also show that compared with those who remained employees, those who transitioned to self‐employment were less likely to be privately insured in year 2 (58.0% vs 85.7%, risk difference: −27.7%, 95% CI –37.3% to −18.1%, P < .0001; time‐by‐employment type product term: −12.9 percentage points, 95%CI −21.6 to −4.2 percentage points, P = .004). Those who changed to self‐employment were more likely to have exchange‐based coverage (10.5% vs 2.7%, risk difference: 7.8%, 95%CI 0.2% to 15.5%, P = .04, time‐by‐employment type product term: 4.4 percentage points, 95% CI 0.4 to 8.3 percentage points, P = .03). They were also more likely to have public insurance coverage in year 2 (16.9% vs 9.0%, risk difference: 7.9%, 95% CI 0.2% to 15.5, P = .04), but this was not a significant change from year 1 public insurance coverage rates (time‐by‐employment type product term: −0.1 percentage points, 95%CI −9.3 to 8.0 percentage points, P = .88).
TABLE 2.
Time‐varying characteristics in year 2 of MEPS participation
| Overall | Employee in Year 2 | Self‐Employed in Year 2 | |
|---|---|---|---|
| N | 16 335 | 16 188 | 147 |
| Weighted N | 12 147 335 | 120 375 762 | 1 097 582 |
| Mean (SD) or N (weighted %) | Mean (SD) or N (weighted %) | Mean (SD) or N (weighted %) | |
|---|---|---|---|
| Family Income as percentage of poverty level | 485.74 (347.55) | 485.91 (346.72) | 466.58 (429.42) |
| Private insurance | 12 650 (85.4) | 12 574 (85.6) | 76 (58.4) |
| Exchange‐based insurance | 539 (2.7) | 526 (2.7) | 13 (10.7) |
| Medicare | 190 (1.2) | 188 (1.2) | 2 (1.4) |
| Medicaid or other public insurance | 2114 (9.2) | 2085 (9.1) | 29 (16.9) |
| Uninsured | 2091 (8.2) | 2045 (8.1) | 46 (25.7) |
| Census region | |||
| Northeast | 2402 (17.1) | 2385 (17.2) | 17 (12.0) |
| Midwest | 3236 (21.8) | 3214 (21.9) | 22 (18.1) |
| South | 6092 (37.7) | 6034 (37.7) | 58 (41.6) |
| West | 4573 (23.3) | 4523 (23.3) | 50 (28.3) |
| Self‐reported health status | |||
| Excellent | 4281 (27.2) | 4235 (27.2) | 46 (31.6) |
| Very Good | 5831 (38.5) | 5771 (38.5) | 60 (40.6) |
| Good | 4849 (27.5) | 4818 (27.5) | 31 (22.9) |
| Fair | 1170 (5.9) | 1160 (5.9) | 10 (5.0) |
| Poor | 153 (0.9) | 153 (0.9) | 0 (0.0) |
| Delay care | 457 (3.2) | 442 (3.1) | 15 (11.9) |
Insurance categories do not sum to 100% as individuals can have more than one status in a given year; however, the uninsured category represents lacking health insurance the entire year. Descriptive statistics are presented here to transparently summarize the sample analyzed. For cells with small sizes (<20), the Agency for Healthcare Research and Quality provides guidelines for the precision of the estimates presented, and note that owing to a high potential for imprecision, these estimates should not necessarily be interpreted as nationally representative. For further details see: https://meps.ahrq.gov/survey_comp/ic_precision_guidelines.shtml.
TABLE 3.
Insurance coverage and type of coverage
| Unadjusted Year 1 prevalence, % | Unadjusted Year 2 prevalence, % | Adjusted Year 2 prevalence, % | Year 2 risk difference, % | Lower 95% CI, % | Upper 95% CI, % | Year 2 risk ratio | Lower 95% CI | Upper 95% CI | P * | |
|---|---|---|---|---|---|---|---|---|---|---|
| Uninsured | ||||||||||
| Employee | 9.2 | 8.1 | 8.0 | |||||||
| Self‐employed | 16.6 | 25.7 | 26.1 | 18.0 | 9.2 | 26.9 | 3.25 | 2.30 | 4.59 | .0001 |
| Private insurance | ||||||||||
| Employee | 84.2 | 85.6 | 85.7 | |||||||
| Self‐employed | 69.7 | 58.4 | 58.0 | −27.7 | −37.3 | −18.1 | 0.68 | 0.57 | 0.80 | <.0001 |
| Exchange‐based insurance | ||||||||||
| Employee | 2.8 | 2.7 | 2.7 | |||||||
| Self‐employed | 6.4 | 10.7 | 10.5 | 7.8 | 0.2 | 15.5 | 3.91 | 1.89 | 8.09 | .04 |
| Medicaid or other public insurance | ||||||||||
| Employee | 8.9 | 9.1 | 9.0 | |||||||
| Self‐employed | 17.6 | 16.9 | 16.9 | 7.9 | 0.2 | 15.5 | 1.88 | 1.19 | 2.97 | .04 |
Analyses adjusted for age, gender, race/ethnicity, education, income, primary language, census region, self‐reported health status, and year of survey.
Abbreviation: CI, confidence interval.
P‐values from predictive margins.
In sensitivity analyses, among those self‐employed in year 1, remaining self‐employed in year 2 was similarly associated with greater risk of being uninsured (21.2% vs 9.6%, risk difference: 11.7%, 95% CI 5.8% to 17.5%, P = .0001) (demographic characteristics of this cohort are presented in Table S5). In a cross‐sectional cohort using prevalent employment type, comparing individuals who were self‐employed vs employees in year 1, results were also similar (Table S4).
In models examining delay of needed medical care, adjusted for the same covariates as above except for health insurance coverage, individuals who changed to self‐employment were more likely to delay needed medical care in year 2 (12.0% vs 3.1%, risk difference: 8.9%, 95% CI 3.1% to 14.6%, P = .003) compared with individuals who remained employees. The risk difference was slightly smaller, but still significantly different, when additionally adjusting for insurance coverage (11.5% vs 3.2%, risk difference: 8.4% 95%CI 2.7% to 14.2%, P = .004). This suggests that self‐employed people may experience “underinsurance” that is a barrier to accessing needed health care—even when they have health insurance coverage (Table 4).
TABLE 4.
Delaying needed medical care
| Unadjusted Year 1 prevalence, % | Unadjusted Year 2 prevalence, % |
Adjusted Year 2 prevalence, % |
Year 2 risk difference, % | Lower 95% CI, % | Upper 95% CI, % | Year 2 risk ratio | Lower 95% CI | Upper 95% CI | P * | |
|---|---|---|---|---|---|---|---|---|---|---|
| Delay medical care, without insurance adjustment | ||||||||||
| Employee | 3.1 | 3.1 | 3.1 | |||||||
| Self‐employed | 8.8 | 11.9 | 12.0 | 8.9 | 3.1 | 14.6 | 3.87 | 2.37 | 6.30 | .003 |
| Delay medical care, with insurance adjustment | ||||||||||
| Employee | 3.1 | 3.1 | 3.2 | |||||||
| Self‐employed | 8.8 | 11.9 | 11.5 | 8.4 | 2.7 | 14.2 | 3.67 | 2.22 | 6.05 | .004 |
Analyses adjusted for age, gender, race/ethnicity, education, income, primary language, census region, self‐reported health status, and year of survey. Analyses with insurance adjustment additionally adjusted for insurance type (private, Medicare, other public, or uninsured)
P‐values from predictive margins.
4. DISCUSSION
In this study of nationally representative working‐age US adults who participated in the labor market, we found that one in four newly self‐employed workers lacked health insurance and that the risk of lacking health insurance was significantly greater when participants changed from being an employee to being self‐employed. Similarly, remaining self‐employed, rather than becoming an employee, was also associated with greater risk of lacking health insurance. Being self‐employed was associated with greater risk of delaying needed medical care, pointing to potential harms from lack of insurance coverage, or underinsurance. These outcomes were not explained by differences in baseline income, demographic characteristics, or previously having health insurance.
Private coverage, the major source of health insurance for working‐age US adults, was considerably lower in those who were self‐employed. This was partially mitigated by greater use of exchange‐based coverage and of public insurance such as Medicaid, but these coverage sources did not close the gap in overall coverage rates. Though we do not know the reasons for this in this study, prior studies have suggested that reasons for less than full insurance coverage in the US population overall may include lack of Medicaid expansion in some states, and unaffordability of market‐based options in certain areas. 27 Those who transitioned to self‐employment in year 2 were less likely to have health insurance coverage even in year 1 (ie, when they were still employees), which suggests that the jobs they were leaving may not have offered health insurance benefits. Nevertheless, even taking this into account in our analyses, those who transitioned to self‐employment were even more likely to be uninsured in year 2 than they were in year 1, suggesting that changing employment type was associated with net loss of health insurance coverage.
This study is consistent with and adds to the literature on self‐employment and health services use. Prior studies found that concerns about losing health insurance coverage can contribute to “job lock” and that increasing access to nonemployer‐sponsored health insurance may promote entrepreneurialism. 10 , 11 , 12 , 13 , 14 Further, a prior study found that the ACA did help improve insurance coverage among those who are self‐employed. 28 The current study suggests a potential mechanism for these findings (exchange‐based coverage), and adds evidence that despite ACA improvements, a substantial portion of self‐employed individuals remain uninsured. Further, the current study adds evidence that transitioning to self‐employment is strongly associated with losing insurance coverage.
This study has several important implications. During the time period of this study, the ACA’s “individual mandate” to obtain health insurance was still enforceable by fine. As this provision has since been repealed, it is possible that insurance coverage for the self‐employed may decline further, as has been the case in the overall US population. 9 Another implication is that as policymakers seeking to construct regulations to address changing work relationships should consider making nonemployer‐sponsored health insurance coverage more accessible. Simply re‐defining self‐employed workers as employees may be insufficient to increase access to health insurance coverage, as the “employer mandate” to offer health insurance does not apply to small firms, or workers not classified as full‐time. 8 Since self‐employed workers may work variable hours, they may be particularly affected by rules regarding what counts as full‐time work. How best to promote insurance coverage for self‐employed workers is an important direction for future research. Another interesting topic for future research would be to examine possibility heterogeneity in the relationship between self‐employment and insurance coverage by type of job (eg, blue‐collar vs white‐collar work, or jobs in different sectors of the economy). 29
The results of this study should be interpreted in light of several limitations. First, this was an observational study, so the statistical associations observed should not be interpreted causally; omitted time‐varying factors correlated with self‐employment and insurance coverage may be driving these results. These could include motivation, job experience, or other factors that did not lead to lack of health insurance during year 1, but did during year 2. Second, only a relatively small number of individuals transitioned from being employees to self‐employment each year. This limited our ability to assess details about their insurance plan characteristics and added uncertainty to the estimates of association. However, this longitudinal “incident user” approach increases internal validity over a “prevalent user” design by helping to better account for unmeasured confounding introduced by who chooses to become self‐employed. Further, a “prevalent user” sensitivity analysis that did have a larger sample size had similar results. Third, reasons for job change were not available. Understanding whether reasons for job change are related to health insurance coverage is an important area for future study. Next, we were unable to formally assess whether those who transitioned to self‐employment and those who remained employees had parallel trends in insurance coverage prior to study year 2. If trends were not parallel, a difference‐in‐difference estimate is likely to be biased, so we recommend a cautious interpretation. Finally, defining lack of health insurance as lacking coverage for the entire year means that individuals with shorter gaps were not classified as uninsured. This conservative definition is likely to result in associations that are closer to the null than a more liberal definition. We note that even short‐term gaps in coverage can be harmful. 18 , 19 , 30 , 31 Therefore, policymakers should strive to create conditions such that those transitioning to self‐employment do not lose health insurance coverage for even a brief period. These limitations are balanced by several strengths, including the use of repeated assessments within individuals to account for time‐invariant individual characteristics like previous job training and preferences about type of work, use of a nationally representative sample, and data on a variety of potential confounding factors.
5. CONCLUSIONS
Being self‐employed is strongly associated with greater risk for lacking health insurance coverage, and in delaying needed medical care. Current policies to help ensure health insurance coverage for working‐age US adults may be insufficient to mitigate changes in the nature of labor force participation, such as self‐employment. Given the rise in self‐employment, it is important to study the impacts on health and health care utilization of self‐employment. This information can be used to design policies that better protect the health of working‐age US adults.
AUTHOR CONTRIBUTION
SAB conceived of the study and drafted the manuscript. RG, MED, and SB assisted with interpretation of the data and revised the manuscript critically for intellectual content. All authors give approval of the manuscript version to be submitted.
Supporting information
Author matrix
Tables S1‐S5
ACKNOWLEDGMENTS
Joint Acknowledgment/Disclosure Statement: All authors declare they have no conflicts to disclose. SAB has received funding from the US National Institutes of Health, and personal fees from the Aspen Institute, outside the submitted work. SB has received funding from the US National Institutes of Health, US Centers for Disease Control and Prevention, US Department of Agriculture, and Robert Wood Johnson Foundation, unrelated to the present study; and has received personal fees from Collective Health, KPMG, Research Triangle Institute, PLOS Medicine, and The New England Journal of Medicine, unrelated to the present study. MD has received funding unrelated to this project from NIH, PCORI, NC DHHS, AHRQ, HRSA, Arnold Ventures, and Society of Family Planning and has received personal fees unrelated to the present study from the American Hospital Association and Wiley.
Berkowitz SA, Gold R, Domino ME, Basu S. Health insurance coverage and self‐employment. Health Serv Res.2021;56:247–255. 10.1111/1475-6773.13598
FUNDING INFORMATIONFunding for SAB’s role on this study was provided by the National Institute of Diabetes And Digestive And Kidney Diseases of the National Institutes of Health under Award Number K23DK109200. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
REFERENCES
- 1. Hammond SFH. Self‐employment in the United States : Spotlight on Statistics: U.S. Bureau of Labor Statistics. https://www.bls.gov/spotlight/2016/self‐employment‐in‐the‐united‐states/home.htm. Accessed January 29, 2020.
- 2. Roberts ET. Small‐business options: occupational outlook for self‐employed workers : Career Outlook: U.S. Bureau of Labor Statistics. https://www.bls.gov/careeroutlook/2018/article/self‐employment.htm. Accessed January 29, 2020.
- 3. Freelanced: the rise of the contract workforce. NPR.org. https://www.npr.org/2018/01/22/578825135/rise‐of‐the‐contract‐workers‐work‐is‐different‐now. Accessed August 12, 2020.
- 4. Chen MK, Chevalier JA, Rossi PE, Oehlsen E. The value of flexible work: evidence from uber drivers. National Bureau of Economic Research. 2017. 10.3386/w23296 [DOI]
- 5. Yildirmaz A, Goldar M, Klein S.Illuminating the shadow workforce: insights into the gig workforce in businesses. Published online February 2020. https://www.adp.com/‐/media/adp/resourcehub/pdf/adpri/illuminating‐the‐shadow‐workforce‐by‐adp‐research‐institute.ashx. Accessed May 4, 2020.
- 6. Bill Text ‐ AB‐5 Worker status: employees and independent contractors. https://leginfo.legislature.ca.gov/faces/billTextClient.xhtml?bill_id=201920200AB5. Accessed January 29, 2020.
- 7. National Center for Health Statistics . Health, United States, 2018. Published November 12, 2019. https://www.cdc.gov/nchs/hus/contents2018.htm. Accessed January 29, 2020. [PubMed]
- 8. United States Senate . The Patient Protection and Affordable Care Act Detailed Summary. Published online 2009. https://www.dpc.senate.gov/healthreformbill/healthbill04.pdf. Accessed January 24, 2018.
- 9. Griffith KN, Jones DK, Bor JH, Sommers BD. Changes in health insurance coverage, access to care, and income‐based disparities among US adults, 2011–17. Health Aff Proj Hope. 2020;39(2):319‐326. 10.1377/hlthaff.2019.00904 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Bailey J. Health insurance and the supply of entrepreneurs: new evidence from the affordable care act. Small Bus Econ. 2017;49(3):627‐646. 10.1007/s11187-017-9856-8 [DOI] [Google Scholar]
- 11. Bailey J, Chorniy A. Employer‐provided health insurance and job mobility: did the Affordable Care Act Reduce Job Lock? Contemp Econ Policy. 2016;34(1):173‐183. 10.1111/coep.12119 [DOI] [Google Scholar]
- 12. Gai Y, Minniti M. Health insurance, job lock, and the supply of self‐employment. J Small Bus Manag. 2015;53(2):558‐580. 10.1111/jsbm.12169 [DOI] [Google Scholar]
- 13. Rietveld CA, van Kippersluis H, Thurik AR. Self‐employment and health: barriers or benefits? Health Econ. 2015;24(10):1302‐1313. 10.1002/hec.3087 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Gumus G, Regan TL. Self‐employment and the role of health insurance in the U.S. J Bus Ventur. 2015;30(3):357‐374. 10.1016/j.jbusvent.2014.01.001 [DOI] [Google Scholar]
- 15. Woolhandler S, Himmelstein DU. The relationship of health insurance and mortality: is lack of insurance deadly? Ann Intern Med. 2017;167(6):424‐431. 10.7326/M17-1403 [DOI] [PubMed] [Google Scholar]
- 16. Freeman JD, Kadiyala S, Bell JF, Martin DP. The causal effect of health insurance on utilization and outcomes in adults: a systematic review of US studies. Med Care. 2008;46(10):1023‐1032. [DOI] [PubMed] [Google Scholar]
- 17. Institute of Medicine (US) Committee on the Consequences of Uninsurance . Effects of Health Insurance on Health. National Academies Press; 2002. https://www.ncbi.nlm.nih.gov/books/NBK220636/. Accessed February 14, 2020. [Google Scholar]
- 18. Gold R, DeVoe J, Shah A, Chauvie S. Insurance continuity and receipt of diabetes preventive care in a network of federally qualified health centers. Med Care. 2009;47(4):431‐439. 10.1097/mlr.0b013e318190ccac [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. O'Malley JP, O'Keeffe‐Rosetti M, Lowe RA, et al. Health Care Utilization Rates After Oregon’s 2008 Medicaid Expansion: within‐group and between‐group differences over time among new, returning, and continuously insured enrollees. Med Care. 2016;54(11):984‐991. 10.1097/MLR.0000000000000600 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Huguet N, Valenzuela S, Marino M, et al. Following uninsured patients through medicaid expansion: ambulatory care use and diagnosed conditions. Ann Fam Med. 2019;17(4):336‐344. 10.1370/afm.2385 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. AHRQ . Medical expenditure panel survey home. Published June 26, 2020. https://meps.ahrq.gov/mepsweb/. Accessed June 26, 2020.
- 22. Driscoll AK, Bernstein AB. Health and access to care among employed and unemployed adults: United States, 2009–2010. NCHS Data Brief. 2012;83:1‐8. [PubMed] [Google Scholar]
- 23. The United States Census Bureau . Industry and Occupation Code Lists & Crosswalks. https://www.census.gov/topics/employment/industry‐occupation/guidance/code‐lists.html. Accessed February 14, 2020.
- 24. Medical expenditure panel survey public use file details. https://www.meps.ahrq.gov/mepsweb/data_stats/download_data_files_detail.jsp?cboPufNumber=HC‐195. Accessed February 14, 2020.
- 25. Ai C, Norton EC. Interaction terms in logit and probit models. Econ Lett. 2003;80(1):123‐129. 10.1016/S0165-1765(03)00032-6 [DOI] [Google Scholar]
- 26. Karaca‐Mandic P, Norton EC, Dowd B. Interaction terms in nonlinear models. Health Serv Res. 2012;47(1 Pt 1):255‐274. 10.1111/j.1475-6773.2011.01314.x [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Sommers BD. Health insurance coverage: what comes after the ACA? Health Aff Proj Hope. 2020;39(3):502‐508. 10.1377/hlthaff.2019.01416 [DOI] [PubMed] [Google Scholar]
- 28. Decker SL, Moriya AS, Soni A. Coverage for self‐employed and others without employer offers increased after 2014. Health Aff Proj Hope. 2018;37(8):1238‐1242. 10.1377/hlthaff.2017.1663 [DOI] [PubMed] [Google Scholar]
- 29. Agarwal SD, Goldman AL, Sommers BD. Blue‐collar workers had greatest insurance gains after ACA implementation. Health Aff Proj Hope. 2019;38(7):1140‐1144. 10.1377/hlthaff.2018.05454 [DOI] [PubMed] [Google Scholar]
- 30. Sommers BD, Gourevitch R, Maylone B, Blendon RJ, Epstein AM. Insurance churning rates for low‐income adults under health reform: lower than expected but still harmful for many. Health Aff Proj Hope. 2016;35(10):1816‐1824. 10.1377/hlthaff.2016.0455 [DOI] [PubMed] [Google Scholar]
- 31. Roberts ET, Pollack CE. Does churning in Medicaid affect health care use? Med Care. 2016;54(5):483‐489. 10.1097/MLR.0000000000000509 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Author matrix
Tables S1‐S5