Will tax-based health insurance reforms help the self-employed stay in business?

I. INTRODUCTION

The self-employed face a tax-induced disadvantage relative to wage and salary workers when it comes to the payment of health insurance premiums. Wage and salary workers often pay their premiums out of pre-tax dollars for both income and payroll taxes but the self-employed are not able to deduct premium costs when calculating payroll taxes. This study examines whether expected claims of the health insurance deduction affect household self-employment decisions and serves as a source of empirical evidence for the ongoing policy debate over health reforms. The analysis expands the existing literature on the behavioral effects of taxation by focusing on a change in tax base, namely the deductibility of health insurance premiums, rather than tax rates.

Using a panel of tax return data, this paper investigates whether the 25% self-employment health insurance deduction implemented under TRA86 had an effect on the probability of exit from self-employment. Admittedly, the 25% deduction represented a modest savings for a self-employed filer as would recent proposals for an exemption from Self-Employment Contributions Act (SECA) taxes. For example, among those claiming the deduction in 1988, the average value was $431. For a household in the 28% tax bracket, this represents a tax savings of $121 or 7% of the premium total. (1) In comparison, full deductibility in calculating SECA taxes, as proposed in recent bills in the House and Senate, would represent about a $250 tax savings. (2)

The self-employed are an economically important group as they are most often associated with entrepreneurship. And entrepreneurs are thought to drive much of the growth and innovation in the economy. Permitting deductions of health insurance premiums in calculating the payroll tax would lower the after-tax price of health insurance premiums, increasing self-employment income and potentially reducing the variance in out-of-pocket health expenditures. The increase in income makes self-employment more attractive relative to the wage and salary sector. The reduction in health-related risks is also important from a policy perspective because although the self-employed are often assumed to be less risk averse than wage and salary workers, there is little empirical evidence on this issue and high-ability households, those most likely to generate innovation and economic growth, might have relatively lower risk tolerances.

This work contributes to a growing literature on the effects of tax policy on labor markets. Most previous analyses of the effects of tax policy have focused exclusively on tax rates or entry into self-employment. This paper contributes to the literature by moving beyond rates and assessing the impacts on the currently self-employed. Results of the analysis also provide general insight into whether aspects of the tax code outside of tax rates affect employment sector decisions. The results also offer specific guidance in terms of whether equalizing the tax treatment of health insurance premiums would enhance the survival of sole proprietorships. Prior studies of the labor market effects of health insurance have examined the effects of portability or availability of health coverage but few have specifically addressed the effects of the tax treatment of premiums.

The remainder of this document is organized as follows. Section II provides a brief history of the relative tax treatment of health insurance premiums paid by the self-employed. Related literature is summarized in Section III, and Section IV contains a description of the data and empirical methodology. Results are presented in Section V, and Section VI concludes with policy implications, limitations, and directions for future work.

II. TAX POLICY AND INCENTIVE EFFECTS

Health insurance premiums paid by employers on behalf of employees have historically received different tax treatment than premiums paid by the self-employed for themselves and their dependants. Health insurance premiums paid by employers are generally deductible in calculating both income and payroll taxes for wage workers. Other tax provisions such as flexible spending accounts allow employees to pay their share of premiums and medical expenses out of pre-tax dollars. (3) Estimates of how many firms offer their employees the option of paying their share of the premium out of pre-tax dollars, sometimes called premium conversion, are rare. Wiatrowski (1995) estimates that about one-third of employees in mid-to large-sized businesses had access to premium conversion in 1993 and Dowd et al. (2001) report that 75% of employees of large public employers had premium conversion options as of 1994. Evidence from a 2002 survey of employers by the Kaiser Family Foundation suggests that about half of employers offer flexible benefit plans that could be used for premium conversion (Gruber and Washington 2005). Quincy (2008) estimates that more than 90% of employees in firms with more than 100 employees offer flexible benefits plans but that availability falls by firm size as 50% of workers in firms with 10-24 employees and 35% of workers in firms with 2-9 employees have access to flexible benefits plans.

In contrast, self-employed health insurance premiums were not deductible when calculating income tax or payroll tax liabilities prior to 1987. A modest move toward equalization of the income tax treatment occurred in 1987 with the implementation of a self-employed health insurance deduction of 25%. The temporary 25% income tax deduction implemented for the self-employed in 1987 was expanded in 1990 to include filers who were more than 2% shareholders in an S corporation. (4)

The temporary health insurance deduction was allowed to expire in 1992 but was made retroactively available for those filing an amended return (Holtz-Eakin, Penrod, and Rosen 1996). Temporary deductions were available until 1997 when the deduction was made permanent and increased to 40%. Further incremental increases were made in 1998 (45%), 1999 (60%), and 2002 (79%). Effective in 2003, 100% of self-employed insurance premiums became deductible in calculating income tax liabilities.

However, as noted earlier, this 100% income tax deduction did not equalize the treatment of health insurance premiums paid on behalf of the self-employed and wage and salary workers. Unlike premiums paid by employers, the self-employed must still pay SECA, or payroll taxes, on income used to pay insurance premiums for themselves and their dependants. (5) Employer insurance premiums for wage and salary employees are generally paid out of pretax dollars and it is often the case that employee portions are also paid out of pre-tax dollars; thus, this income is not subject to the Federal income tax or payroll taxes (Federal Insurance Contributions Act [FICA] taxes). (6) Further compounding the issue is the fact that most households now pay more payroll taxes than income taxes (Mitrusi and Poterba 2000).

To see how health insurance tax incentives might alter household decisions regarding self-employment, consider a household that maximizes expected utility where the utility function depends both on the level and variance of household income. In this context, after-tax health insurance prices affect income levels and health insurance policies provide a safeguard against large outlays for health services, reducing the variance of income. High-ability households find self-employment attractive as a means to increase income levels, and households with lower risk tolerances find the higher cost of health insurance less attractive. If households differ in both risk attitudes and entrepreneurial ability, households whose self-employment decisions are affected at the margin might include those with high levels of ability but with lower tolerances for risk. That is, the increased cost of reducing the background risk for health expenses (outside of the income risks associated with the self-employment activity) might be enough to lead some households with high ability levels to forgo self-employment if their risk tolerances are relatively low and they have good opportunities in the wage and salary sector. This might have negative consequences for the broader economy as these high-ability households are those most likely to produce innovations and create new jobs.

It should be noted that although deductibility of self-employed health insurance premiums from the income tax reduces the after-tax price of health insurance for the self-employed it also reduces the implicit insurance of the tax system. A progressive tax system with loss offsets can generally reduce income risks because the tax rate falls as income decreases (e.g., Bruce 2002). This occurs as lower income levels move the filer into a lower tax bracket. Although this effect is possible, the modest amount of tax savings from the deduction and the small number of tax brackets make it unlikely that a large number of filers would be bumped into a lower tax bracket based on the amount of the health insurance deduction.

Differential tax treatment of self-employed and wage and salary health insurance premiums might be expected to affect self-employment rates through two avenues: (1) deter entry or (2) expedite exit. Empirical studies of the impacts on entry and exit have produced inconclusive results.

III. PREVIOUS LITERATURE

The literature reviewed in the following section can be divided into two main categories, health insurance and labor outcomes and the effectiveness of tax incentives. The current analysis builds on the previous work by focusing on the effects of the health insurance deduction and includes robustness checks with marginal tax rates for comparisons to the previous literature. It examines the exit decisions of single filers in a probit framework where information available in the current time period (t) affects decisions in the next time period (t + 1).

The literature most closely related to this study examines the relationship between health insurance and the probability of self-employment. Two recent studies have found evidence that the after-tax price of health insurance premiums affects self-employment entry decisions but the results are mixed for exit decisions with one study finding that lowering after-tax premiums reduces exit (Heim and Lurie 2009a) and one finding no effect on exit rates or the probability of being self-employed (Gumus and Regan 2009).

Several studies have addressed the effects of health insurance availability on self-employment, but do not specifically address tax incentives or survival in self-employment. One recent study focused on the self-employed finds that a New Jersey program to facilitate access to non-employer-linked health insurance coverage increased self-employment among New Jersey residents compared to residents of Pennsylvania where equivalent reforms were not undertaken (Decicca 2007). The possibility that employer-provided insurance creates job-lock and discourages transitions to self-employment is also examined by Holtz-Eakin, Penrod, and Rosen (1996). The authors use longitudinal data from the Survey of Income and Program Participation and a difference in differences approach to compare wage and salary individuals with individuals who transition into self-employment. They find no statistically significant effects of health insurance portability on transitions into self-employment.

Wellington (2001) considers the effects of health insurance coverage on the probability of self-employment. She uses a cross section of data from the 1993 Current Population Survey and finds that the availability of coverage from another source (spouse) increases the likelihood of self-employment between 2 and 5 percentage points.

Although not generally focused on the self-employed, there is a large body of literature that addresses the labor market effects of health insurance for various groups including those near retirement and married women. There is a general consensus that individuals with more portable insurance options retire earlier (Rust and Phelan 1997; Karoly and Rogowski 1994; Rogowski and Karoly 2000; Blau and Gilleskie 2001; Madrian 1994a) and married women with health insurance through their spouses are less likely to work outside the home (Buchmueller and Valletta 1999; Olson 1998; Wellington and Cobb-Clark 2000). However, the evidence on the effects of health insurance portability on job choice is inconclusive and the existing research is split among three outcomes: (1) gaining health insurance is negatively associated with job change and vice versa (Cooper and Mon-heit 1993; Madrian 1994b; Gruber and Madrian 2004; Stroupe, Kenney, and Kniesner 2001; Bansak and Raphael 2008); (2) there is no significant effect of insurance availability (Holtz-Eakin 1994; Holtz-Eakin, Penrod, and Rosen 1996; Kapur 1998); and (3) the effect is contingent on specification or sub-group (Buch-mueller and Valletta 1996; Berger, Black, and Scott 2004; Gilleskie and Lutz 2002).

The second strain of literature takes a broader perspective, and addresses the general effectiveness of tax incentives. Several studies focus specifically on the effectiveness of tax incentives to encourage health insurance coverage. While not the focus of this paper, the topic is worth exploring briefly because it provides insight into whether or not tax incentives are expected to generate behavioral responses. Examining small and large firms alike, Gruber and Lettau (2004) found that firms are quite responsive to health insurance tax incentives. Small firms (less than 100 employees) were most responsive in their decisions to offer insurance as an employee benefit. Large firms (more than 1,000 employees) were more responsive in levels of spending on employee health premiums.

The decision to offer health insurance is explored by Gruber and Poterba (1994) who find that a 1% increase in the price of insurance reduces the probability that a single, self-employed person will purchase health insurance by nearly 2 percentage points. Heim and Lurie (2009b) also find that an increase in the price of insurance leads to a lower coverage take-up rate with an estimated elasticity of--0.71. Selden (2009) finds that increased deductibility of premiums led to more private health insurance coverage, but most of the subsidy represents a transfer to persons who would have purchased insurance in the absence of the subsidy.

The conclusion that firms, including the self-employed, are quite responsive to health insurance tax incentives is echoed in studies of taxable income (Wu 2005) and the responsiveness of the self-employed to tax incentives targeted at savings (Power and Rider 2002). The self-employed are more likely to contribute to tax-deferred retirement savings plans and contribute more as the after-tax price declines. In addition, Bruce (2002) and Gurley-Calvez and Bruce (2008) provide evidence that taxes have a significant effect on the duration of entrepreneurial activities, but the authors only consider tax rates.

In summary, the literature provides a general guideline for thinking about the issue of health insurance tax incentives and self-employment. First, the evidence on whether health insurance availability affects decisions to become self-employed is mixed. Second, the self-employed are generally responsive to tax incentives, particularly in their decisions to purchase health insurance, suggesting that tax incentives might indeed be expected to generate behavioral effects.

IV. DATA AND METHODOLOGY

A tangible definition of self-employment is necessary in order to empirically examine its responsiveness to tax policy. In the empirical analysis below, the presence of a Schedule C on a filer's tax return is used to identify the self-employed. As policies to increase the income and payroll tax deductibility of health insurance premiums are generally targeted at Schedule C filers, this is the most relevant group for the behavioral analysis that follows. Tax return data are currently the only data source that includes both self-employment status and information regarding the use of the health insurance deduction. Much of the previous literature has relied on self-reported survey responses to identify the self-employed, but a growing number of studies have used tax return data, identifying the self-employed by the presence of a Schedule C (e.g., Holtz-Eakin, Joulfaian, and Rosen 1994; Power and Rider 2002).

The data for this project are drawn from the University of Michigan Tax Research Database. This panel of data was constructed by the Office of Tax Policy Research (OTPR) at the University of Michigan from the public-use tax return data released by the Internal Revenue Service (IRS) Statistics of Income (SOI) Division and represents the most recent publicly available panel of tax return data. The data contain nearly 300,000 tax returns representing a random sample of all filers. Approximately 6,000 households are present in all 12 yr of the data.

Although the panel contains data for 1979-1990, a health insurance deduction for the self-employed was only available beginning in tax year 1987. Figure 1 presents the percent of Schedule C filers reporting a health insurance deduction for years 1988-1990. Overall, the percent of Schedule C filers claiming the deduction increased by 4 percentage points from just under 11% in 1988 to almost 15% in 1990. The average dollar value of deductions claimed also increased from $431 to $550 (Figure 2). As these values reflect at most 25% of the health insurance premiums paid, a conservative estimate for average annual premiums is $l,724-$2,200 as the deduction could not exceed net profits.

Figures 1 and 2 provide information about use of the health insurance deduction and dollar amounts claimed by marital status. In 1988 married filers were somewhat more likely to claim the deduction (10.84%) than single filers (10.22%) but the difference was not statistically significant at conventional levels. The growth in use from 1988 to 1990 for single filers outpaced that of married filers so that by the end of the 3-yr period, single filers were slightly more likely to report a deduction (14.64% vs. 14.58%) but again the difference was not statistically significant.

Unsurprisingly, single filers had smaller deductions on average (Figure 2). In 1990 the average dollar value of the deduction was 55% higher for married filers ($628.38) than for single filers ($345.23). This is likely because married filers have more dependants and are generally older and therefore pay higher premiums. Both groups of filers experienced an increase in the average dollar value reported between 1988 and 1990 (31.41% for married filers and 19.36% for single filers). The increase for married filers primarily took place from 1988 to 1989 while single filers experienced higher growth in the average value of deductions from 1989 to 1990.

Exit from self-employment is defined as being self-employed in one year (filing a Schedule C) and not the following year, but does not necessarily imply failure. Thus, only those with data for both years who are self-employed in the first year (at risk for exiting) are included in the sample. This simple indicator avoids more complicated issues such as degree of self-employment (e.g., percentage of total income derived from self-employment) or whether a venture is risky enough to be considered entrepreneurial (e.g., exhibits substantial losses). The analysis is conducted on pooled 2-yr transitions for years 1988-1990. (7) Table 1 contains suggestive evidence that health insurance deductions might decrease exit rates among Schedule C filers. The first entry in Table 1 indicates that 6.75% of Schedule C filers claiming a deduction in 1988 had exited self-employment by 1989. The exit rate for those not claiming the deduction in 1988 was significantly higher at 20.11%. The pattern is similar across years and filing status. A discrete choice framework is used to analyze the effects of health insurance deductibility on exits from self-employment as in Bruce and Gurley (2005). (8) The estimated equation is of the form:

TABLE 1

Exit Rates for Schedule C Filers By Previous Year's Health Insurance
Deduction Status

Exit                 All Schedule    Married      Single
Rates                  C Filers      Filers       Filers

1989   Deduction            6.75%        6.74%      6.78%
                         (17/252)     (13/193)     (4/59)
       No Deduction      20.1 1 %       18.15%     25.88%
                      (407/2,024)  (274/1.510)  (133/514)
       Overall             18.63%       16.85%     23.91%
                      (424/2.276)  (287/1.703)  (137/573)

1990   Deduction            8.62%        6.16%     15.19%
                         (25/290)     (13/211)    (12/79)
       No Deduction        20.49%       17.18%     29.59%
                      (410/2,001)  (252/1.467)  (158/534)
       Overall             18.99%       15.79%     27.73%
                      (435/2,291)  (265/1,678)  (170/613)

Notes: Differences in exit rales by deduction status are
statistically significant at the 1% level for each filing
status group (all Schedule C, married, and single filers)

[FIGURE 1 OMITTED]

[FIGURE 2 OMITTED]

(1) [D.sub.(i, t+1)] = [beta]' [X sub.(i, t)] +[[tau]I.sub.(i, AVG)] +[[gamma]H.sub.(i, t)] +[[micro].sub.i] + [V.sub.(i, t+1)]

where [D.sub.(i, t+1)] is a binary variable, which takes a value of 1 if an individual is self-employed as of time t and not self-employed as of time t + 1 and zero if the household remains self-employed in both t and t + 1. Thus, only those self-employed in the first time period are "at risk" of exit and included in the analysis. [X.sub.(i, t)] is a vector containing a constant and a set of exogenous control variables as of time t.

[H sub.(i, t)] contains information regarding the health insurance deduction at time t. In some of the specifications estimated below it is a dummy variable taking a value of 1 if the health insurance deduction was used. In other models it represents the dollar value of the tax savings from the health insurance deduction (deduction amount times the tax rate) as actual tax savings depend both on the dollar value of the tax deduction and the tax bracket of the filer. The time t deduction amount is used to mitigate concerns of simultaneity of the health deduction and exit decisions and endogeneity caused because a household's employment sector affects its ability to claim the deduction. Using a lagged value of health insurance deductions is appropriate for exit decisions as the currently self-employed are likely to base future self-employment decisions on past experiences. That is, claims of the deduction from this period are used to proxy for expected claims in the next period when the exit decision is made. An analysis of entry decisions would be more complex as it is difficult to estimate the expected benefit of the deduction for new entrants.

[I.sub.(i, AVG)] represents household-level controls for income. To mitigate endogeneity problems as income levels are likely to affect employment sector, lagged income levels are used. Further, the analysis sample consists of the self-employed, and self-employment income can vary substantially over time. To account for the potentially uneven self-employment income stream, average income from time t and t--1 is used. Further, income is divided into three categories, self-employment income, wage income, and non-labor income as income from different sources could have opposing effects. At first glance it might seem unnecessary to control for wage income but about half of single filers and more than two-thirds of married filers with a Schedule C also report wage income. This issue of partial self-employment is addressed in a robustness check described below.

The error term in this equation includes an individual-specific time-invariant random effect ([[mu].sub.i]) to capture unobserved individual hetergeneity, and an independently and identically distributed residual component ([V.sub.(i, t+1)]) with zero mean and finite variance. The ability to account for unobserved individual heterogeneity is particularly valuable given that the data set does not contain much in the way of demographic information. Education, race, and gender are among the missing variables for which there does not appear to be even a rough proxy. However, the panel of tax returns provides more in terms of additional control variables than might be immediately apparent. Proxies are included in the estimation for age, household size, number of children, liquidity constraints, and risk attitudes. Specifically, Xi, t includes the number of children living away from home as a rough proxy for age (9) and total number of exemptions to proxy for household size and capture the number of children not living away from home.

The presence of financial constraints limiting self-employment is an often-cited argument for government intervention. For filers who itemize their deductions, the presence of a mortgage interest deduction is used as a source of information regarding liquidity constraints, as in Bruce and Holtz-Eakin (2001). While those with housing equity might be more likely to have access to sources of loanable funds, it might also be the case that the presence of a mortgage limits the household's ability to obtain additional financing.

Risk attitudes are also thought to be important in self-employment decisions. As in Gurley-Calvez and Bruce (2008), the tax balance due is included in an attempt to capture a household's risk attitude as it seems plausible that more risk-averse households would be more likely to over-withhold their taxes, thereby receiving a refund from the IRS. Relatively risk-loving filers might prefer to under-withhold such that the money is available for alternative uses. In the case of the self-employed who make estimated tax payments, the choice to under-or over-withhold would be based on the expected self-employment income. (10) Finally, aggregations of state identifiers in the tax panel are used to control for region of residence and a dummy variable is used for households where the state of residence is omitted for confidentiality reasons. Regional dummy variables are included to capture economic conditions that are particular to a region that are not captured in the time effect. These conditions might include trends in unemployment, which would affect job opportunities outside of self-employment. The missing region identifier is, in practice, an indicator for very high-income households. The implicit assumption is that those without state identifiers (i.e., very high-income households) are likely to behave similarly despite where they are located in the country. State dummy variables are explored as a robustness check.

The sample is limited to single filers for the empirical analysis in-line with most of the existing empirical literature, which follows the custom of limiting the sample to male heads of household who are of prime working age (25-54). In the case of a single filer, the household self-employment decision is equivalent to the individual decision. For married filers, spousal employment status and earnings cannot be separated in the present data. In addition, married filers potentially have access to health insurance through a spouse. Filers claiming an age 65 exemption are excluded from the estimation sample as they are generally eligible for Medicare. Filers are included if they are present in two consecutive years from 1988 to 1990 (i.e., have data for both time t and time t+1).

A. Robustness Check--Addressing Potential Selection Bias

Based on the means of the regression variables (Appendix Table A1), claimants and non-claimants differ in several areas. Claimants are less likely to exit (12% vs. 28% for non-claimants), have lower non-labor income, live in states with lower rates of uninsured persons, have more self-employment income, higher balances due, and are substantially more likely to have claimed a general business credit in the preceding 10 years. (11) The baseline analysis uses a one-period lag of health insurance deduction claims to proxy for future expected claims but does not address the endogeneity problem that there still might remain unobserved factors that are correlated with health insurance deductions that also affect exit rates. This problem is addressed using instrumental variables (TV) methods where the probability of claiming the health insurance deduction (or the tax savings) is predicted in the first stage and the effects of the deduction on exit rates is predicted in the second stage. The difficulty in implementing this approach is finding suitable instruments that are highly correlated with claims of the health insurance deduction but not with the unobservables that affect exit rates.

TABLE A1

Summary Statistics for Regression Variables

                    Single Filers       Claimants     Non-Claimants

                   Mean     Std.       Mean     Std.          Mean

Exit *              0.26     0.44       0.12   0.32           0.28

Health insurance    0.12     0.32       1.00   0.00            N/A
deduction
(dummy)

Tax savings         5.87    21.97      50.48  43.67            N/A
(dollars)

Wage income        47.09   766.24       4.49  11.01          52.70
($1,000)

Self-employment     4.03   382.36      17.33  18.24           2.28
income ($1,000)

Non-labor income   60.79  1770.63       2.26  10.00          68.50
($1,000)

                                 Full-Time      Not Full-Time

                    Std.       Mean     Std.   Mean     Std.

Exit *               0.45       0.11   0.31   0.36       0.48

Health insurance      N/A       0.20   0.40   0.06       0.24
deduction
(dummy)

Tax savings           N/A      10.07  27.00   3.16      17.49
(dollars)

Wage income        815.00       0.73   2.18  22.64      38.03
($1,000)

Self-employment    406.69      11.87  28.68   2.27      11.64
income ($1,000)

Non-labor income  1883.57       3.12  19.47   5.30      53.59
($1,000)

                  Married  Filers

                    Mean      Std.

Exit *               0.16    0.37

Health insurance     0.12    0.32
deduction
(dummy)

Tax savings         12.74   70.21
(dollars)

Wage income         30.91   70.26
($1,000)

Self-employment      8.81   25.61
income ($1,000)

Non-labor income     7.12   53.74
($1,000)

                      Single      Claimants   Non-Claimants  Full-Time
                      Filers

                    Mean   Std.  Mean    Std. Mean    Std.  Mean   Std.

West                 0.29  0.45   0.33  0.47  0.28    0.45  0.30   0.46

Midwest              0.20  0.40   0.17  0.37  0.21    0.41  0.19   0.39

Northeast            0.21  0.41   0.31  0.46  0.20    0.40  0.23   0.42

Other region         0.00  0.05   0.00  0.00  0.00    0.05  0.00   0.05

Missing region       0.01  0.12   0.00  0.00  0.02    0.12  0.01   0.09

Mortgage interest    0.25  0.43   0.36  0.48  0.24    0.43  0.21   0.41
deduction

Kids away            0.02  0.16   0.02  0.19  0.02    0.15  0.00   0.07
(number)

Total exemptions     1.37  0.89   1.27  0.69  1.38    0.91  1.23   0.70
(number)

Balance due          0.46  4.88   1.76  4.30  0.29    4.93  1.15   4.51
($100)

State percent       14.00  4.67  13.53  4.54 14.06    4.68 14.30   4.62
uninsured

General business     0.23  0.42   0.34  0.48  0.21    0.41  0.27   0.44
credit

Sample size         1,186          138        1.048          469

                        Not Full-Time     Married Filers

                       Mean       Std.    Mean      Std.

West                   0.28       0.45     0.23    0.42

Midwest                0.21       0.41     0.25    0.43

Northeast              0.20       0.40     0.17    0.37

Other region           0.00       0.05     0.00    0.02

Missing region         0.02       0.13     0.02    0.14

Mortgage interest      0.28       0.45     0.53    0.50
deduction

Kids away              0.03       0.20     0.01    0.09
(number)

Total exemptions       1.45       0.98     3.30    1.29
(number)

Balance due            0.02       5.08     0.76    8.49
($100)

State percent         13.80       4.69    13.72    4.69
uninsured

General business       0.21       0.40     0.48    0.50
credit

Sample size             711               3,381

Nate: Income variables represent averages over time t and t--1,
Unless otherwise noted, entries represent the percent of tilers.
Variables denoted with an asterisk are as of time r-h 1, all other
variables are as of time t except General Business Credit, which
takes a value of I if a filer claimed the general business credit,
or related credit from 1979 to 1988.
Data Source: OTPR Panel for years 1988-1990.

The first instrument is the percent of uninsured persons in the filer's state of residence. (12) It is expected that the percent of uninsured at the state level would be positively correlated with health insurance deduction claims. That is, in a state with relatively high rates of uninsured persons, the filer is less likely to have insurance from another source (e.g., employer, public insurance) and be more likely to purchase their own insurance and claim the deduction.

The second IV is a dummy variable for whether the filer claimed the general business credit or other miscellaneous credit from 1979 through 1988. General business credits include the jobs credit, alcohol used as a fuel, research and experimentation, and the low-income housing credit. Other miscellaneous credits include the residential energy credit. Claims of these credits are relatively rare for single filers, 23% claimed a credit at least once. The variable is included as an IV as claims of the general business credit are likely to indicate that the filer (or their tax preparer) is relatively knowledgeable of the tax system or actively involved in tax planning, which would increase their awareness of the health insurance deduction.

B. Robustness Check--Degree of Entrepreneurship, Years of Experience, and Other Controls

Additionally, control variables are added to the baseline specification to proxy for profitability or experience, the unobservable factors most likely to be correlated with exit rates and correlated with health deduction claims. More successful or experienced business owners are less likely to exit self-employment (Holtz-Eakin, Joulfaian, and Rosen 1994; Taylor 1999) but might also be more likely to make use of the health insurance deduction (as they have less access to employer-provided health insurance). To test for this possibility, the models are estimated, including an interaction term combining the health deduction variable with an indicator for whether the household participated in self-employment "full-time" (i.e., the household had no wage or salary income). In the sample, the full-time exit rate is 11% and the rate for not full-time is 36%. Similar differences are observed in claims for the health insurance deduction as 20% of full-time filers claim the deduction versus 6% of not full-time filers (Appendix Table Al). (13) The IV models are also estimated with the full-time control.

Figure 3 presents the percentage of Schedule C filers that had only self-employment income by marital status. As expected, single households are much more likely to meet the definition of full-time self-employment. The trend in full-time self-employment declined for both filing status groups over the time period. Rates declined from over 45% for single filers in 1979 to less than 40% in 1990. The percent of married self-employed with no wage or salary income fell from almost 24% in 1979 to almost 16% in 1990.

Years of self-employment might also serve as a proxy for profitability of the business or experience of the owner. Using tax years 1979-1988, years of experience is measured as the number of years in which a Schedule C was filed. For gaps in the data, the most recent self-employment status was used for the missing years. On average, filers in the sample had 5.1 yr of experience in self-employment. As expected, those who exited had less experience on average, 2.9 yr, while those who did not exit had 5.8 yr of experience. Those claiming the health insurance deduction in the previous period averaged just under 7 yr of experience and non-claimants had an average of 4.8 yr of experience.

The specification in Equation (1) is also estimated including a control variable for net loss, as only those with positive net income can claim the credit. Controls for income from Schedules F (farm income) and Schedule E (rents and royalties, partnership income, S corporation income) are also included to test whether results differ based on additional types of income. Results are also obtained for each cross section of data (1989 and 1990). Additionally, state dummy variables instead of region are used to test whether state policies masked in the region groupings might be driving exit decisions.

[FIGURE 3 OMITTED]

C. Robustness Check--Comparison to the Tax Rates Literature

A number of recent studies have examined the effects of marginal tax rates on entry and exit decisions (e.g., Gurley-Calvez and Bruce 2008; Bruce and Gurley 2005). As income levels are implicitly included in tax rates, the income variables in Equation (1) are replaced with expected marginal tax rates as of time t + 1. These tax rates represent the marginal tax rates households would expect to face in each of the two outcomes: self-employment and wage and salary employment. Effective rates differ across employment sectors for a variety of reasons including increased evasion opportunities in the self-employment sector and differential tax preferences such as home office deductions for the self-employed.

Of course, t + 1 income levels needed to calculate the tax rates are only observed for the actual outcome, and income must be estimated for the alternative sector. Following Bruce (2000 and 2002) and Gurley-Calvez and Bruce (2008), income for the alternative sector is predicted by running ordinary least squares (OLS) regressions of observed income for a given sector on a constant, non-labor income, and a set of household-specific control variables including proxies for age and the number of children in the tax filer's household. (14) Estimated coefficients are then used to predict incomes for tax filers in the alternative sector.

After obtaining income estimates, tax rates are calculated by running a set of 17 variables from the tax returns through the National Bureau of Economic Research's TAXSIM model (Feenberg and Coutts 1993). Two tax rates are calculated for each filer: their actual tax rate depending on their chosen sector, and their hypothetical tax rate in the alternative sector. For purposes of comparability, TAXSIM-calculated tax rates are used for both outcomes rather than making use of the actual tax rates that are provided (for the chosen outcome only, of course) in the tax return data. Note that the calculated tax rates are actual tax rates, inclusive of evasion. Federal and state income taxes are combined with payroll taxes to arrive at a single tax rate for each sector.

However, these tax rates are likely to be endogenous as a move from the self-employment sector to the wage and salary sector is likely to affect this expected tax rate because employment sector and tax rates are simultaneously determined. This potential endogeneity is addressed as in previous research (e.g., Bruce 2000. 2002; Gurley-Calvez and Bruce 2008) using an additional set of tax rates from the TAXSIM model.

First, tax rates (one for each employment sector) are calculated using time t + 1 incomes and time t tax rules. These are the best approximations of the tax rates that would have existed had the tax rules remained constant. Next, an additional set of tax rates are calculated using incomes and tax rules as of time t + 1. This second set of tax rates represents the closest approximation to actual post-transition tax rates including any policy changes that occurred between time t and t + 1. The IV is then defined as the difference between the second (using time t + 1 tax rules) and first (using time t tax rules) tax rates, and represents the part of the actual t + 1 tax rate that is caused by the change in the tax code only. Two IVs are constructed, the difference in the wage-and-salary tax rates and the difference in the entrepreneurship tax rates. These IVs are entered separately into two first-stage panel regressions, one for each potentially endogenous tax rate.

V. RESULTS

Baseline results from Equation (1) are presented in Table 2. Claiming the health insurance deduction significantly reduces the probability of exit. Using the marginal effect presented in column three, a filer who claims the deduction in the previous period is 3.1 percentage points less likely to exit. Given the overall exit probability of 26% (Appendix Table Al), this represents a 12% reduction in the probability of exit.

Results are similar when the dummy variable for claiming the deduction is replaced by the tax savings from that deduction, allowing for more variability in the health insurance deductibility variable. The probability of exit decreases as the tax savings from the health insurance deduction increases. Based on the estimated elasticity, an increase in the tax savings from the deduction of 10% would lead to a reduction in the probability of exit of 4.9%. For perspective, full deductibility (100%) of health insurance premiums from the SECA tax would represent an increase in the dollar value of the health insurance deduction of about 15%. (15) Thus, assuming an equivalent increase in tax savings, full SECA deductibility would be expected to reduce the probability of exit for self-employed filers by about 7.4%.

The null hypothesis of no effect from income sources cannot be rejected at conventional levels of significance but the signs of the coefficients are generally as expected. More income from non-labor sources is negatively associated with exit, while more wage income is positively correlated with exit. However, income from self-employment has the opposite than expected sign, more self-employment income is associated with higher exit rates, but the coefficient is statistically and economically insignificant. Filers in the West were less likely to exit as were those with mortgage interest deductions. Filers with more total exemptions were more likely to exit, possibly indicating that filers with more dependants are likely to seek out a more stable source of income.

The results in Table 3 are from an IV model, where the probability of claiming the deduction or the tax savings from the deduction is predicted in the first stage and the probability of exit is estimated in the second stage. IV probit models are estimated on pooled data from 1989 and 1990. However, not all of the standard specification tests available for linear regressions are available for the probit model and a linear probability model is also estimated to complete the specification testing. In all instances, claiming the deduction decreases the probability of exit, and higher levels of tax savings are associated with lower probabilities of exit.

TABLE 3

Instrumental Variables Probability of Exit: Dummy Variable and Tax
Savings

                                        IV Probit

                          Dummy                        Tax Savings
                          Coeff.          M.E.           Coeff.

Health insurance        -3.446 ***             -0.562
deduction (dummy)          (0.132)

Tax savings                                             -0.041 ***
                                                           (0.002)

Wage income ($ 1,000)       -0.001              0.000        0.001
                           (0.001)                         (0.001)

Self-employment income    0.006 **              0.002     0.007 **
($1,000)
                           (0.003)                         (0.003)

Non-labor income             0.001              0.000      0.002 *
($1,000)
                           (0.001)                         (0.001)

West                         0.112              0.043       -0.032
                           (0.095)                         (0.095)

Midwest                      0.023              0.009        0.057
                           (0.091)                         (0.093)

Northeast                0.247 ***              0.096     0.210 **
                           (0.090)                         (0.093)

Other region                -0.274             -0.010       -0.174
                           (0.615)                         (0.633)

Missing region           -0.915 **             -0.274   -1.370 ***
                           (0.419)                         (0.436)

Mortgage interest            0.009              0.003       -0.024
deduction
                           (0.095)                         (0.096)

Kids away                   -0.220             -0.084       -0.215
                           (0.199)                         (0.204)

Total exemptions             0.011              0.004      0.067 *
                           (0.040)                         (0.039)

Balance due ($100)           0.012              0.005        0.014
                           (0.009)                         (0.009)

Sample size                  1,180                            1180

CLR                      18.35 ***                      21.100 ***

AR                       22.37 ***         23.630 ***          C-D

LM-J                   Rejected **        Rejected **       Sargan

                                    Linear Regression

                                          Dummy        Tax Savings
                           M.E.           Coeff.            Coeff

Health insurance                            -1.588 **
deduction (dummy)
                                              (0.627)

Tax savings                 -0.015                      -0.020 ***
                                                           (0.007)

Wage income ($ 1,000)        0.001              0.000        0.001
                                              (0.001)      (0.001)

Self-employment income       0.003              0.003      0.005 *
($1,000)
                                              (0.002)      (0.003)

Non-labor income             0.001              0.001      0.001 *
($1,000)
                                              (0.001)      (0.001)

West                        -0.012              0.025       -0.036
                                              (0.063)      (0.051)

Midwest                      0.021             -0.002        0.017
                                              (0.052)      (0.055)

Northeast                    0.080              0.108        0.096
                                              (0.073)      (0.067)

Other region                -0.064             -0.111       -0.082
                                              (0.359)      (0.367)

Missing region              -0.330           -0.514 *    -0.742 **
                                              (0.271)      (0.317)

Mortgage interest           -0.090             -0.026       -0.027
deduction
                                             (0.05 1)      (0.051)

Kids away                   -0.081             -0.113       -0.121
                                              (0.124)      (0.126)

Total exemptions             0.025              0.020     0.045 **
                                              (0.022)      (0.022)

Balance due ($100)           0.005            0.008 *      0.008 *
                                              (0.005)      (0.005)

Sample size                                      1180         1180

CLR                             LM           8.805 **   11.973 ***

AR                           4.375              5.966

LM-J                      4.573 **              2.300

Note: Standard errors are in parentheses. M.E. signifies marginal
effect. The effect for a binary variable is the effect of going from
0 to 1. Entries represent coefficients, marginal effects, and
elasticities from a random effects probit estimation. All
specifications include year fixed effects.
* Statistically significant at the .10 level; ** at the .05 level:
*** at the .01 level.
Data Source: OTPR Panel for years 1988-1990.

Several tests are used to assess the validity of the instruments. Starting with the simplest of verifications, both instruments are statistically significant at the 10% level in the first-stage estimation probit estimates although only the general business credit IV is statistically significant (1% level) in the linear models (Appendix Table A2). Finlay and Magnusson (2009) developed a series of inference tests that are robust to weak instruments for the IV probit model in Stata. These tests include the Anderson-Rubin (AR) statistic, the Kleibergen-Moreira Lagrange multiplier (LM) test, the overidentification (J) test, and the conditional likelihood ratio (CLR) test. The LM and J tests are combined to account for the spurious decline in power of the LM test in some regions of the parameter space, and Stata reports whether the null for the combined test is rejected or not rejected at the 5% level. The null hypothesis for all of the tests is that the coefficient on the instrumented variable (claiming the health deduction or the tax savings) is zero. Test statistics are reported in Table 6 and indicate that the null of a zero coefficient is rejected at the 1% level (or 5% level for the LM-J statistic).

TABLE 6

Probability of Exit with Marginal Tax Rates: Dummy Variable and Tax
Savings

                         Dummy          Tax Savings

                       Coeff.      M.E.     Coeff.     Elasticity

Health insurance    -0.941 ***  -0.106
deduction (dummy)
                       (0.343)

Tax savings                               -0.017 **      -0.196
                                            (0.007)

Wage MTR            -0.494 ***  -0.089   -0.603 ***     -37.713
                       (0.139)              (0.161)

Self-employment MTR  0.353 ***   0.063    0.379 ***      24.513
                       (0.092)              (0.095)

West                     0.470   0.096        0.513       0.287
                       (0.322)              (0.387)

Midwest                  0.156   0.030        0.215       0.085
                       (0.307)              (0.375)

Northeast                0.170   0.032        0.335       0.140
                       (0.387)              (0.471)

Other region             2.706   0.822        3.235       0.016
                       (2.022)              (2.483)

Missing region       -2.467 **  -0.109    -2.921 **      -0.077
                       (1.043)              (1.244)

Mortgage interest    1.603 ***   0.426    1.875 ***       0.920
deduction
                       (0.502)              (0.572)

Kids away                0.837   0.150        0.991       0.033
                       (0.658)              (0.802)

Total exemptions         0.193   0.035      0.279 *       0.744
                       (0.127)              (0.153)

Balance due (SI00)      -0.026  -0.005       -0.029      -0.026
                       (0.022)              (0.026)

Sample size               1186                 1186

Log likelihood        -606.715             -606.423

Average wage MTR         32.25
                        (2.31)

Average                  33.25
self-employment MTR
                        (2.34)

Nate: Standard errors are in parentheses. MTR signifies marginal tax
rate. M.E. signifies marginal effect. The effect for a binary variable
is the effect of going from 0 to 1. Entries represent coefficients,
marginal effects, and elasticities from a random effects probit
estimation. All specifications include year fixed effects.
* Statistically significant at the.10 level; ** at the .05 level;
*** at the .01 level.
Data Source: OTPR Panel for years 1988-1990.
TABLE A2

First-Stage Results for the Instrumental Variables Estimation

                        IV Probit              Linear Regression

                    Dummv     Tax Savings    Dummv           Tax
                   Coeff.     Coeff.      Coeff.   Savings Coeff.

Wage                 0.000        0.010    -0.001           0.008
income ($1,000)
                   (0.000)      (0.033)   (0.000)         (0.033)

Self-employment  0.003 ***    0.317 *** 0.003 ***       0.315 ***
income ($1,000)
                   (0.001)      (0.043)   (0.001)         (0.044)

Non-labor            0.000       0.040*     0.000          0.039*
income ($1,000)
                   (0.000)      (0.022)   (0.000)         (0.022)

West             0.069 ***        2.438 0.070 ***           2.492
                   (0.025)      (1.976)   (0.025)         (1.988)

Midwest              0.036      4.255 *    -0.003           2.049
                   (0.031)      (2.497)   (0.038)         (3.014)

Northeast        0.107 ***    8.390 ***  0.073 **        6.455 **
                   (0.030)      (2.415)   (0.036)         (2.841)

Other region        -0.125       -8.583    -0.149          -9.969
                   (0.189)     (14.776)   (0.190)        (14.900)

Missing region    -0.215 *  -27.956 ***  -0.229 *     -28.786 ***
                   (0.122)      (9.598)   (0.123)         (9.681)

Mortgage             0.027        1.803     0.024           1.587
interest ded.
                   (0.024)      (1.905)   (0.024)         (1.916)

Kids away           -0.095     -8.391 *    -0.100        -8.630 *
                   (0.062)      (4.832)   (0.062)         (4.861)

Total exemptions    -0.011        0.309    -0.012           0.303
                   (0.011)      (0.867)   (0.011)         (0.872)

Balance              0.002        0.171     0.002           0.174
due ($100)
                   (0.002)      (0.187)   (0.002)         (0.188)

                        IV Probit           Linear Regression

                    Dummy   Tax Savings    Dummy     Tax Savings
                   Coeff.     Coeff.      Coeff.     Coeff.

Percent Uninsured  0.003 *     0.338 **     0.072        0.064
                   (0.002)      (0.159)   (0 024)      (0 263)

General Business  0.053 **    5.435 *** 0.105 ***    6.509 ***
Credit             (0.025)      (1.886)   (0.063)      (1.916)

Note: Standard errors are in parentheses. M.E. signifies marginal
effect. The effect for a binary variable is the effect of going from
0 to I. Entries represent coefficients, marginal effects, and
elasticities from a random effects probit estimation. All
specifications include year fixed effects.
* Statistically significant at the.10 level; ** at the .05 level;
*** at the .01 level
Data Source: OTPR Panel for years 1988-1990.

Additional tests are available for linear regression IV models (Baum, Schaffer, and Stillman 2007). The under identification LM statistic tests the null that the matrix of reduced form coefficients is under identified (has rank = K1 - 1) against the alternative hypothesis that the matrix is identified (has rank = K1). The null hypothesis is rejected at the 5% level.

The Cragg-Donald (C-D) statistic tests the null that the equation is weakly identified. The estimated C-D statistic of 4.375 is not greater than the critical value for 10% maximal IV size estimated by Stock, Wright, and Yogo (2002), indicating that the instruments might be weak. Finally, the Sargan statistic is an over identification test of all instruments. The null that all instruments are valid is not rejected at the 1% level for the health deduction dummy and at the 10% level for the tax savings specification.

In general, the IV results suggest a negative effect of health insurance deduction claims and tax savings on exits from self-employment. These inferences appear to be robust to a potential problem of weak instruments, and testing generally supports the validity of the IVs but sometimes at low levels of significance. The IVs perform better in the probit specifications but the weak instrument tests are only available for the linear regression models. Because the estimates are likely to be biased if the validity conditions are not met, these results are best viewed as suggestive of the direction of the effect.

The magnitudes of the effects from the IV estimates are much larger than those found in the probit model. Claiming the deduction results in a 56.2 percentage-point decrease in the probability of exit compared to a 3.1 percentage-point decrease in the probit model. Likewise the marginal effect of tax savings indicates that a $10 increase in the value of the deduction leads to a 15 percentage-point reduction in the probability of exit. According to the probit model estimated in Table 2, the same $10 increase in the value of the deduction would reduce the probability of exit by 0.38 percentage points.

TABLE 2

Probability of Exit: Dummy Variable and Tax Savings

                               Dummy          Tax Savings

                        Coeff.      M.E.     Coeff.     Elasticity

Health insurance     -1.304 ***  -0.031
deduction (dummy)       (0.439)

Tax savings                               -0.028 ***      -0.485
                                             (0.011)

Wage income ($1,000)      0.001   0.000      0.001 *       0.099
                        (0.000)              (0.000)

Self-employment           0.001   0.000        0.001       0.017
income ($1,000)
                        (0.001)              (0.001)

Non-labor                -0.002   0.000       -0.003      -0.444
income ($1,000)
                        (0.002)              (0.002)

West                  -1.000 **  -0.040    -1.206 **      -1.012
                        (0.416)              (0.474)

Midwest                  -0.212  -0.010       -0.251      -0.148
                        (0.443)              (0.527)

Northeast                -0.243  -0.011       -0.238      -0.149
                        (0.433)              (0.512)

Other region              1.407   0.252        1.832       0.014
                        (3.421)              (6.247)

Missing region           -0.596  -0.018       -0.785      -0.031
                        (1.328)              (1.465)

Mortgage              -0.816 **  -0.032    -0.913 **      -0.672
interest deduction      (0.346)              (0.391)

Kids away                 0.526   0.028        0.504       0.025
                        (0.993)              (1.252)

Total exemptions       0.319 **   0.017     0.412 **       1.645
                        (0.159)              (0.194)

Balance                   0.035   0.002        0.045       0.060
due ($100)
                        (0.028)              (0.032)

Sample size               1.186                1,186

Log likelihood         -636.830             -634.418

Note; Standard errors are in parentheses. M.b. signifies marginal
effect. The effect for a binary variable is the effect of going from
0 lo 1. Entries represent coefficients, marginal effects, and
elasticities from a random effects probil estimation. All
specifications include year fixed effects.
* Statistically significant at the.10 level; ** at the.05 level;
*** at the.01 level.
Data Source: OTPR Panel for years 1988-1990.

Results presented in Table 4 include a full-time dummy variable as well as an interaction of the full-time dummy and the health deduction variable. This specification allows for full-time status to have an independent effect on the probability of exit as well as the effect through increasing the probability of claiming the health insurance deduction. For these specifications, the null hypothesis of no effect from the health insurance deduction cannot be rejected although the coefficients are negative as in the earlier results. The third and seventh columns of Table 4 contain results for the IV models including the full-time control. In these specifications the coefficients on the health insurance deduction dummy and the tax savings variable are negative and the null of no effect is rejected at the 1% level.16 Once again, the magnitudes of the effects from the IV models are much larger than those in the probit models presented in Table 2.

TABLE 4

Probability of Exit with Full-Time Controls: Dummy Variable and Tax
Savings

                                        Dum my

                          Probit         M.E.       IV Probit   M.E.

Health insurance           -0.284       -0.052 -3.460 ***  -0.563
deduction (dummy)
                          (0.374)                 (0.154)

Tax savings
Full-time dummy        -1,381 ***       -0.254      0.060   0.023
                          (0.276)                 (0.255)

Interaction (lull-time -0.522 ***       -0.084
with health variable)
                          (0.579)

Wage income ($1,000)       -0.003       -0.001      0.000   0.000
                          (0.004)                 (0.002)

Self-employment income   -0.012 *       -0.003    0.007 *   0.003
($1,000)
                          (0.007)                 (0.003)

Non-labor income            0.001        0.000      0.001   0.000
($1,000)
                          (0.003)                 (0.001)

West                    -0.481 **       -0.090      0.110   0.042
                          (0.231)                 (0.116)

Midwest                    -0.103       -0.021      0.023   0.009
                          (0.228)                 (0.093)

Northeast                  -0.022       -0.005  0.242 ***   0.094
                          (0.225)                 (0.091)

Other region                0.435        0.114     -0.284  -0.103
                          (1.580)                 (0.618)

Missing region             -0.411       -0.068  -0.927 **  -0.276
                          (1.045)                 (0.417)

Mortgage interest       -0.516 **       -0.093      0.014   0.006
deduction
                          (0.212)                 (0.121)

Kids away                   0.151        0.032     -0.225  -0.086
                          (0.494)                 (0.199)

Total exemptions          0.128 *        0.027      0.015   0.006
                          (0.084)                 (0.044)

Balance due ($100)          0.026        0.006      0.013   0.005
                          (0.022)                 (0.009)

Sample size                 1.180

Log likelihood           -596.578

                                   Tax Savings

                          Probit         M.E.    IV Probit   M.E.

Health insurance
deduction (dummy)

Tax savings                -0.009       -0.084 -0.040 ***  -0.015
                          (0.006)                 (0.003)

Full-time dummy        -1.465 ***       -0.964     -0.202  -0.074
                          (0.290)                 (0.221)

Interaction (lull-time -0.003 ***       -0.018
with health variable)
                          (0.010)

Wage income ($1,000)       -0.003       -0.062      0.001   0.000
                          (0.004)                 (0.002)

Self-employment income   -0.013 *       -0.130   0.007 **   0.003
($1,000)
                          (0.008)                 (0.004)

Non-labor income            0.001        0.010    0.002 *   0.001
($1,000)
                          (0.003)

West                   -0.4.92 **       -0.235     -0.062  -0.023
                          (0.239)                 (0.110)

Midwest                    -0.109       -0.036      0.044   0.016
                          (0.236)                 (0.098)

Northeast                   0.002        0.001   0.021 **   0.076
                          (0.233)                 (0.098)

Other region                0.452        0.002     -0.181  -0.064
                          (1.634)                 (0.658)

Missing region             -0.489      -0.01 1 -1.217 ***  -0.295
                          (1.092)                 (0.475)

Mortgage interest       -0.528 **       -0.221     -0.052  -0.019
deduction
                          (0.218)                 (0.116)

Kids away                   0.151        0.004     -0.210  -0.077
                          (0.509)                 (0.244)

Total exemptions          0.144 *        0.326    0.074 *   0.027
                          (0.088)                 (0.041)

Balance due ($100)          0.029        0.023      0.015   0.006
                          (0.023)                 (0.010)

Sample size                 1,180

Log likelihood             -596.1

Note: Standard errors are in parentheses. Entries represent
coefficients from a random effects probit estimation and a pooled
IV probit model. All specifications include year fixed effects.

* Statistically significant at the.10 level; ** at the .05 level;
*** at the .01 level. Data Source: OTPR Panel for years 1988-1990

Results for the remaining robustness checks are presented in Table 5. For brevity, only the results for the health insurance variables are included. The negative effect of the health insurance dummy and tax savings variables remains after adding the control for years of experience, which has a negative and statistically significant effect on exits from self-employment. Results are generally similar for cross-sectional probits and specifications including state dummy variables. In addition, the probability of exit was estimated separately for filers whose average prior self-employment income (from 1979 to 1988) was equal to or more than the median and for those whose average income was below the median. The effect of the health insurance dummy and tax savings was negative for both groups but only statistically significant (at the 10% level) for the median or above group with the health insurance dummy variable. These results suggest that the effect might be the strongest for the highest earners although there are important limitations to these results including small sample sizes and incomplete data for previous earnings.

TABLE 5

Probability of Exit Robustness Checks: Dummy Variable and Tax Savings

                                      Coeff.      M.E.   Elasticity

Baseline -- dummy                  -1.304 ***  -0.031
                                      (0.439)

Years of experience -- dummy        -0.498 **  -0.102
                                      (0.226)

Cross-section: 1989 -- dummy       -0.857 ***  -0.188
                                      (0.269)

Cross-section: 1990 -- dummy           -0.310  -0.088
                                      (0.194)

State controls -- dummy             -0.976 **  -0.026
                                      (0.449)

Baseline--tax savings              -0.028 ***              -0.485
                                      (0.011)

Years of experience -- tax savings   -0.008 *              -0.068
                                      (0.004)

Cross-section: 1989 -- tax savings  -0.012 **              -0.076
                                      (0.005)

Cross-section: 1990 -- tax savings   -0.005 *              -0.049
                                      (0.003)

State controls -- tax savings       -0.011 **              -0.108
                                      (0.004)

Mote: Standard errors are in parentheses. M.E. signifies marginal
effect. The effect for a binary variable is the effect of going from
0 to 1. Entries represent coefficients, marginal effects, and
elasticities from a random effects probit estimation. All
specifications include year fixed effects.

* Statistically significant at the.10 level; ** at the.05 level;
*** at the.01 level.
Data Source: OTPR Panel for years 1988-1990.

Table 6 is included for comparability with previous research regarding the effects of marginal tax rates on self-employment transitions. Consistent with Gurley-Calvez and Bruce (2008), a relative increase in the marginal tax rate on wage and salary income reduces the probability of exit, while an increase in the self-employment tax rate increases the probability of exit. As above, the presence of the health insurance premium deduction decreases the probability of exit and the effect is larger with larger values of tax savings. As is the case in Gurley-Calvez and Bruce (2008), the magnitude of the effects is implausibly large. A possible explanation for the overstated tax effects is that marginal tax rates are poor proxies for income variables but income numbers cannot be included as additional control variables as income is implicitly included in the marginal tax rates.

VI. CONCLUSIONS

In summary, the results provide consistent evidence that health insurance deductibility in the calculation of income taxes would likely reduce the probability of exit for the existing self-employed who claim the deduction. The effects on exit are negative and statistically significant in 8 of 10 specifications with a dummy variable for claiming the deduction and 9 of 10 specifications using the tax savings from the deduction. Thus, equalizing the treatment of health insurance premiums for the self-employed and wage workers by allowing full deductibility from SECA taxes (an increase in the dollar amount of the deduction of about 15%) would result in a decrease in exits for single filers. The estimated size of the effect is 7.4% from the baseline model. Although the presence of an effect is established in nearly all specifications, the magnitude of the effect varies across specifications and more research is needed, possibly using more recent data, to more convincingly establish the size of the effect. In particular, the effects from the IV models are implausibly large.

More broadly, these results provide further evidence that changing aspects of the tax code beyond tax rates can potentially create significant behavioral responses. These results are broadly consistent with the literature on the behavioral effects of taxation, suggesting that tax rates and base impact household decisions. They are also similar in spirit to recent work by Heim and Lurie (2009a) who used a panel of tax return data to examine how the change in premium prices induced by changes in the tax code affect entry and exit decisions.

From a policy perspective there are a couple of important limitations. First, the analysis is limited to single filers given the complexity of intra-household decision-making for joint filers. Thus, it is unclear whether allowing SECA deductibility for health insurance premiums would also lead married self-employed filers who claim the deduction to remain in business longer. Additionally, it is likely that not all of the single filers in the analysis are eligible to claim the deduction as they might have access to health insurance through a wage and salary employer. It is not possible to discern eligible households based on the information available in the data. Second, claims of the health insurance deduction are likely to be endogenous and correlated with other unobservable factors affecting exit. Efforts were made to mitigate these concerns through an IV approach and by including full-time and years of experience controls. Results strongly suggest a negative effect of claiming the health deduction on exit rates, even accounting for weak instruments in the IV specifications. However, further research is needed to more definitively establish the size of the effect.

Another area for future research is examining what factors affect claims of the health insurance deduction amongst eligible households. One cannot determine conclusively whether a filer in the tax data is eligible to claim the deduction, but results from the first-stage IV estimates indicate that there are several factors that affect claims of the deduction and the amount of tax savings in significant ways. Most notably, filers who claimed a general business credit over the previous 10 years were far more likely to claim the health insurance deduction and had greater tax savings from the deduction. This result suggests that information is likely an important component of deduction take-up rates because those filers (or their accountants) most familiar with the tax system are likely to be aware of the array of tax credits available.

REFERENCES

Bansak. C., and S. Raphael. "The State Children's Health Insurance Program and Job Mobility: Identifying Job Lock among Working Parents in Near-Poor Households." Industrial Labor Relations Review, 61(4), 2008, Article 7.

Baum, C. F., M. E. Schaffer, and S. Stillman. "Enhanced Routines for Instrumental Variables/GMM Estimation and Testing." Stata Journal, 7(4). 2007, 465-506.

Berger, M. C., D. A. Black, and F. A. Scott. "Is There Job-Lock?: Evidence from the Pre-HIPAA Era." Southern Economic Journal, 70(4). 2004, 953-76.

Blau, D. M., and D. B. Gilleskie. "Retiree Health Insurance and Labor Force Behavior of Older Men in the 1990s." Review of Economics and Statistics, 83(1). 2001, 64-80.

Bruce, D. "Taxes and Entrepreneurial Endurance: Evidence from the Self-Employed." National Tax Journal, 55(1), 2002, 5-24.

Bruce, D., and T. Gurley. "Taxes and Entrepreneurial Activity: An Empirical Investigation Using Longitudinal Tax Return Data." Office of Advocacy, U.S. Small Business Administration Contract No. SBAHQ-03-M-0535, 2005.

Bruce, D., and D. Holtz-Eakin. "Who Are the Entrepreneurs? Evidence from Taxpayer Data." Journal of Entrepreneurial Finance and Business Ventures, 1(1), 2001, 1-10.

Buchmueller, T. C., and R. G. Valletta. "The Effects of Employer-Provided Health Insurance on Worker Mobility." Industrial and Labor Relations Review, 49(3), 1996, 439-55.

Buchmueller, T. C., and R. G. Valletta. "The Effect of Health Insurance on Married Female Labor Supply." Journal of Human Resources, 34(1). 1999, 42-70.

Cooper, P, F., and A. C. Monheit. "Does Employment-Related Health Insurance Inhibit Job Mobility?" Inquiry, 30(4), 1993, 400-16.

Decicca, P. "Health Insurance Availability and Entrepreneur-ship: Evidence from New Jersey." Working Paper No. 1003309, 2007. Accessed 2 July 2009. Available at SSRN: http://ssrnxom/abstract=1003309.

Dowd, B., R. Feldman, M. Maciejewski, and M. V. Pauly. "The Effect of Tax-Exempt Out-of-Pocket Premiums on Health Plan Choice." National Tax Journal, 54(4). 2001, 741-56.

Feenberg, D., and E. Coutts. "An Introduction to the TAXSIM Model." Journal of Policy Analysis and Management, 12(1), 1993, 189-94.

Feldstein, M-, and A. Samwick. "Social Security Rules and Marginal Tax Rates." National Tax Journal. 45(1), 1992, 1-22.

Finlay, K., and L. M. Magnusson. "Implementing Weak Instrument Robust Tests for a General Class of Instrumental Variables Models." Stata Journal, 9(3). 2009, 398-421.

Gilleskie, D. B., and B. Lutz. "The Impact of Employer-Provided Health Insurance on Dynamic Employment Transitions." Journal of Human Resources, 37(1), 2002, 129-62.

Gruber, J., and M. Lettau. "How Elastic is the Firm's Demand for Health Insurance?" Journal of Public Economics, 88, 2004, 1273-93.

Gruber, J., and B. C. Madrian. '"Health Insurance, Labor Supply and Job Mobility: A Critical Review of the Literature," in Health Policy and the Uninsured, edited by C. G. McLaughlin. Washington, DC: Urban Institute Press, 2004, 97-178.

Gruber, J., and J. Poterba. "Tax Incentives and the Decision to Purchase Health Insurance: Evidence from the Self-Employed." The Quarterly Journal of Economics, 109(3). 1994, 701-33.

Gruber; J., and E. Washington. "Subsidies to Employee Health Insurance Premiums and the Health Insurance Market." Journal of Health Economics, 24(2), 2005, 253-76.

Gums, G., and T. Regan. "Self-Employment and the Role of Health Insurance." IZA Discussion Paper No. 3952, 2009. Accessed 2 July 2009. Available at SSRN: http://ssrn.com/abs(racl=I332593.

Gurley-Calvez, T., and D. Bruce. "Do Tax Cuts Promote Entrepreneurial Longevity?" National Tax Journal, 61(2). 2008, 225-50.

Heim, B. T., and I. Z. Lurie. "The Effect of Health Insurance Premium Subsidies on Entry into and Exit from Self-Employment." Mimeo, U.S. Department of the Treasury, 2009a.

--. "Do Increased Premium Subsidies Affect How Much Health Insurance Is Purchased? Evidence from the Self-Employed."Journal of Health Economics, 28(6), 2009b, 1197-210.

Holtz-Eakin, D. "Health Insurance Provision and Labor Market Efficiency in the United States and Germany," in Social Protection Versus Economic Flexibility: Is There a Tradeoff?, edited by R. M. Blank. Chicago, IL: University of Chicago Press, 1994, 157-87.

Holtz-Eakin, D., D. Joulfaian, and H. S. Rosen. "Sticking It Out: Entrepreneurial Survival and Liquidity Constraints." Journal of Political Economy, 102(1). 1994. 53-75.

Holtz-Eakin, D., J. R. Penrod, and H. S. Rosen. "Health Insurance and the Supply of Entrepreneurs." Journal of Public Economics, 62, 1996, 209-35.

Kapur. K. "The Impact of Health on Job Mobility: A Measure of Job Lock." Industrial and Labor Relations Review, 51(2), 1998, 282-97.

Karoly, L. A., and J. A. Rogowski. "The Effect of Access to Post-Retirement Health Insurance on the Decision to Retire Early." Industrial and Labor Relations Review, 48(1), 1994, 103-23.

Madrian, B. C. "The Effect of Health Insurance on Retirement." Brookings Papers on Economic Activity, 1994 (1), 1994a, 181-232.

. "Employment-Based Health Insurance and Job Mobility: Is There Evidence of Job Lock?" Quarterly Journal of Economics, 109( 1), 1994b. 27-54.

Mitrusi, A., and J. Poterba, "The Distribution of Payroll and Income Tax Burdens, 1979-1999." National Tax Journal, 53. 2000, 765-94.

Olson, C. A. "A Comparison of Parametric and Semipara-metric Estimates of the Effect of Spousal Health Insurance Coverage on Weekly Hours Worked by Wives." Journal of Applied Econometrics, 13(5). 1998, 543-65.

Power, L., and M. Rider. "The Effect of Tax-Based Savings Incentives on the Self-Employed." Journal of Public Economics, 85. 2002, 33-52.

Quincy, L. "Using Section 125 Premium-Only Plans to Expand Health Coverage." Mathematica Policy Research Issue Brief, 2008. Accessed 2 July 2009. Available at http://www.mathematica-mpr.com/publications/PDFs/section125brief.pdf

Rogowski, J. A., and L. A. Karoly. "Health Insurance and Retirement Behavior: Evidence from the Health and Retirement Survey." Journal of Health Economics, 19(4), 2000, 529-39.

Rust, J., and C. Phelan. "How Social Security and Medicare Affect Retirement Behavior in a World of Incomplete Markets." Econometrics, 65(4), 1997, 781-831.

Selden, T. M. "The Impact of Increased Tax Subsidies on the Insurance Coverage of Self-Employed Families." Journal of Human Resources, 44(1), 2009, 115-39.

Stock, J. H., J. H. Wright, and M. Yogo. "A Survey of Weak Instruments and Weak Identification in Generalized Method of Moments." Journal of Business and Economic Statistics. 20(4), 2002, 518-29.

Stroupe, K. T., E. D. Kenney, and T. J. Kniesner. "Chronic Illness and Health Insurance-Related Job Lock." Journal of Policy Analysis and Management, 20(3), 2001, 525-44.

Taylor, M. P. "Survival of the Fittest? An Analysis of Self-Employment Duration in Britain." The Economic Journal, 109 (March), 1999, C140-55.

Wellington, A. "Health Insurance Coverage and Entrepreneurship." Contemporary Economic Policy, 19(4), 2001, 465-78.

Wellington, A. J., and D. A. Cobb-Clark. "The Labor-Supply Effects of Universal Health Coverage: What Can We Learn from Individuals with Spousal Coverage?" in Worker Well-Being: Research in Labor Economics, Vol. 19, edited by S. W. Polachek. Amsterdam: Elsevier Science, 2000.

Wiatrowski, W. J. "Who Really Has Access to Employer-provided Health Benefits?" Monthly Labor Review, 118(6), 1995, 36-44.

Wu, S. "The Tax Effect on Taxable Income from Privately Held Businesses." Southern Economic Journal, 71(4), 2005, 891-912.

ABBREVIATIONS

FICA: Federal Insurance Contributions Act

IRS: Internal Revenue Service

IV: Instrumental Variables

OTPR: Office of Tax Policy Research

SECA: Self-Employment Contributions Act

SOI: Statistics of Income

TRA86: Tax Reform Act of 1986

doi: 10.1111/j.1465-7287.2010.00202.x

* I thank the Center for Business and Economic Research at the University of Tennessee and the Center for Policy Research at Syracuse University for assistance in procuring the data for this project. This research was supported by a grant from the U.S. Small Business Administration (Contract No. SBAHQ-04-M-0536--Graduate Student Research). In addition, the dissertation from which this research grew was funded in part through a dissertation fellowship from the Ewing Marion Kauffman Foundation. The contents of this article are solely the responsibility of the author. Additional valuable feedback was provided by Donald Bruce, Joe Johnson, Chad Moutray, seminar participants at West Virginia University, and the University of Tennessee, and three anonymous referees.

(1.) A deduction valued at $431 represents at most a quarter of average premiums paid (this is a lower bound as deductions are only allowed in the presence of profits). Thus, 28% of $431 gives a tax savings of $ 121. Dividing this savings by the estimated premium amount ($1724) results in a deduction of 7% of the premium paid.

(2.) Efforts include House (H.R.3660 referred to the House Committee on Ways and Means 9/25/2007) and Senate (S.2239 referred to the Committee on Finance 10/25/2007) bills to allow the self-employed to deduct health insurance premium costs from the self-employment (SECA) or payroll tax. Bills introduced in previous sessions of Congress include a Senate bill titled the Equity for Our Nation's Self Employed Act of 2005 (S. 663 referred to Senate Committee on Finance 3/17/2005), an earlier House bill titled the Self-Employed Health Care Affordability Act of 2003 (H.R. 1873, referred to House Ways and Means Committee 4/30/2003), and the companion Senate bill titled the Equity for Our Nation's Self-Employed Act of 2004 (S. 2433. referred to Senate Finance Committee 5/18/2004).

(3.) Flexible spending accounts allow employees to pay out-of-pocket health expenses out of pre-tax dollars. These plans, also called cafeteria plans or Section 125 plans, were provided for in IRS Code Section 125 beginning in 1978. These accounts can be used for premium conversion (paying the employee portion of the health insurance premium out of pre-tax dollars), medical flexible spending accounts, to cover other medical expenses such as co-pays and services not covered by an employee's health insurance plan (e.g., dental services), and other benefits including dependant expenses and group life insurance coverage.

(4.) See Internal Revenue Code 16 U.S.C. Sec.162 (1) (5).

(5.) As the payroll tax contributions define future benefits, adding a SECA exemption is potentially detrimental for those with low income levels. For instance, Feldstein and Samwick (1992) found that the effective Social Security tax rate was negative for those with low earnings in 1990. Deducting income used to pay insurance premiums from SECA calculations would reduce the amount of payroll taxes paid and the value of future benefits. For households with a negative effective tax rate, the gain from avoiding current taxes is more than offset by the loss of future benefits.

(6.) FICA and SECA taxes serve as the primary funding source for the Medicare and Social Security programs.

(7.) Note that this encompasses exits that occurred in 1989 and 1990. Data for 1987 are excluded from the main sample as the health deductibility data for this year were deemed to be less reliable. Information for 1987 is only available for returns that were filed in a later year and re-assigned back to the 1987 filing year. Specifications including 1987 produced qualitatively similar results. Note that it is not possible for a filer to have multiple exits in the data.

(8.) The above equation is estimated via a random effects probit. The lack of a longer panel of data (following the introduction of the health insurance deduction) precludes the use of more sophisticated duration analysis techniques to examine the effects of the health insurance deduction on spells of entrepreneurship.

(9.) The estimations exclude filers claiming an "Age 65" deduction.

(10.) There may be a bias toward larger balances due for the newly self-employed unfamiliar with the requirement for estimated tax payments but this is not expected to have a large impact in the current analysis as it is focused on exits from self-employment amongst the already self-employed.

(11.) All differences are statistically significant at the 1% level.

(12.) The source for percent of uninsured persons by state is: U.S. Census Bureau, Current Population Survey, 1988-2006 Annual Social and Economic Supplements.

(13.) Differences in exit rates and health insurance deduction claims are significant at the 1% level.

(14.) Full results from the income-prediction regressions are available upon request from the author.

(15.) Assuming that the taxpayer is below the SECA income cap, $106,800 as of 2009.

(16.) Note that the interaction term is not included in the IV models as the effects of full-time status on claims of the health insurance deduction should be captured in the first stage. In addition, the interaction term represents an additional endogenous variable and additional IVs would be needed to run the full set of specification tests.

Gurley-Calvez: West Virginia University, Bureau of Business and Economic Research, Morgantown, WV. Phone [phone], Fax [phone], E-mail [e-mail]