When the Option Meets a Real Job
The question here is empirical: when moonlighters do move into full-time self-employment, which observable conditions travel with that move? The SIPP study offers a useful answer without pretending to observe every private motive. It separates immediate movement from eventual switching and compares the financial signal from side work with the attachment and protection supplied by the primary job. The result is a more demanding picture than “successful side hustle equals inevitable exit”: weaker ties to the main job and strong side-work income are associated with entry, while health insurance can hold the option open rather than let it be exercised.
Two clocks and one imperfect window into side work
The Survey of Income and Program Participation follows households through repeated survey waves. The study uses four panels—the 1996, 2001, 2004, and 2008 panels—with interviews four months apart. Taken together, the observations run from 1996 through 2013, a period before ride-share and food-delivery platforms became common forms of supplemental work. That date range is not a footnote. The findings describe a side-work economy in which freelancing, consulting, and other extra jobs were visible, but they cannot simply be assumed to describe the composition of platform work today.
The data provide two valuable pieces of information: whether a respondent moonlights and how much money comes from that activity. The question asks whether the person earned extra money doing work outside their regular job, including freelancing, consulting, or moonlighting. It does not record the industry, business structure, duration, scalability, or purpose of the activity. A single broad measure can therefore identify a relationship between side-work income and later employment status, but it cannot directly identify who was running a product business, who was doing occasional contract work, and who was covering a shortfall in wages.
The descriptive sample gives a sense of the starting population. Among 176,175 respondents observed in the first wave of the four panels, 4,955 were moonlighters, so 2.8% of the full sample reported moonlighting. Moonlighters were not otherwise identical to non-moonlighters. Their mean monthly total earnings were $3,875, compared with $2,563 for non-moonlighters; they averaged 14.4 years of education, compared with 13.4; and they worked 42.5 hours per week in total, compared with 34.4. They were also more likely to be male and slightly younger in this sample. These differences matter because a later transition may reflect resources or selection as well as the side-work activity itself.
The raw transition gap is nevertheless substantial. During the observed survey period, 17.6% of moonlighters ever entered self-employment, compared with 6.8% of non-moonlighters. For full-time self-employment, the corresponding figures were 14.2% and 6.9%. These comparisons establish that moonlighting and later entry move together in the data. They do not establish that moonlighting caused the entry. People who choose or are able to moonlight may already differ in income, education, available time, ambition, or access to opportunities.
The study then narrows its focus to moonlighters and asks two different questions. The “next-wave entry” model estimates whether a moonlighter enters full-time self-employment in the following four-month survey wave. The “ever-switch” model asks whether the person enters full-time self-employment at any point during the panel. The second model uses people observed in the first wave who are not already self-employed, and contains 4,051 observations in the full sample; the next-wave model contains 13,550. The distinction is essential. A side venture may produce a strong signal today while the person waits several months or years to manage benefits, savings, family obligations, or the practical work of scaling.
Both models are linear probability models. In plain terms, a coefficient can be read approximately as a change in the probability of entry, measured in percentage points, holding the model’s other included variables constant. The specifications include primary-job income and hours, indicators for the share of income coming from moonlighting, employer and spousal health insurance, marital and household variables, demographic controls, and survey-year fixed effects. Those controls make the comparison more disciplined, but they do not turn the study into a randomized experiment. The full-sample R-squared values are 0.014 for the next-wave model and 0.043 for the ever-switch model, reminders that much of the variation in individual transitions remains unexplained.
The two clocks also protect against a misleading interpretation of “entry.” A four-month result describes a near-term change in employment state. An ever-switch result captures a cumulative outcome across the panel. The latter can reflect a gradual process: a person may keep the primary job while building income, wait for a suitable opening, or leave only after a benefit or household constraint changes. It is therefore possible for a variable to predict eventual movement without functioning as an immediate trigger.
The employment definitions are similarly useful and limited. In the SIPP analysis, a respondent counts as self-employed when they own a business and work at least 15 hours in it; full-time self-employment means working at least 35 hours in that business. Those thresholds create a consistent outcome for estimation. They do not tell us whether the business is stable, whether the person has stopped all other work, or whether the new arrangement is preferable. Measurement makes the transition visible; it does not finish the interpretation.
What the associations say about exercising the option
The first important result concerns attachment to the primary job. Primary-job hours are negative and statistically significant in both models. The paper reports that an additional 10 hours of work per week in the primary job is associated with a 2.7-percentage-point reduction in the probability of transitioning during the sample period. That pattern fits a necessity account: someone with fewer hours or a weaker tie to wage employment may have less to lose by moving into self-employment. But it can also fit an opportunity account. Fewer hours may leave more time for testing and building the side venture. The variable measures hours, not the person’s reason for working them, so the data cannot decide between those explanations.
Primary-job income complicates the necessity story. The study hypothesized that lower wage income would push moonlighters toward full-time self-employment by reducing the opportunity cost of leaving. Instead, the coefficient on log primary-job income is positive and statistically significant in five of six specifications. It is essentially zero for women in the next-wave model and positive, though smaller than the men’s estimate, in the women-only ever-switch model. The finding is consistent with a different constraint: higher primary income may provide savings, credit access, or other resources that make a transition affordable. It may also capture unobserved ability or opportunity. The result does not mean that higher wages cause entrepreneurship. It means that, once the model is estimated, the simple prediction that low wages alone create the strongest pressure to exit is not supported.
This combination is revealing. A person can have a relatively weak attachment to the primary job while still needing enough resources to carry the risk of full-time self-employment. Fewer hours can lower the cost of leaving, while higher income or accumulated resources can raise the ability to absorb a bad month. “Necessity” and “opportunity” are not opposite boxes into which every respondent must fit. The same transition can involve a job that is becoming less useful and a side venture that is becoming more viable.
The clearest evidence for the option-test interpretation comes from the composition of income. The study uses indicators for whether moonlighting supplies at least 10%, 25%, or 50% of total income. The lower thresholds are weaker and less consistently significant. The 50% indicator is positive and significant in both horizons. In the next-wave model, its coefficient is 0.0175 for the full sample and 0.0273 for men, roughly 1.75 and 2.73 percentage points. In the ever-switch model, the full-sample coefficient is 0.0830, which the paper interprets as about an eight-percentage-point higher probability of eventual transition; the estimates are 0.0838 for men and 0.0664 for women.
The pattern persists when the threshold indicators are replaced with a continuous measure of the moonlighting share of income. It also persists across alternative definitions of the self-employment outcome and under Logit and Probit models that report average marginal effects. That consistency makes the income-share relationship a useful empirical marker. A side activity that already supplies a substantial portion of a person’s income may have generated enough demand, confidence, or financial capacity to support a larger commitment.
Yet the result still does not prove that a business has been validated. Income share can rise because side-work earnings are high, because primary-job earnings are low, or because the worker is already depending heavily on unstable work. The SIPP does not show repeat customers, margins, duration, scalability, or the worker’s expectations. It cannot distinguish a recurring venture from an unusually favorable period of contract work. The association is therefore consistent with the option-test mechanism, but it is not a direct observation of someone updating a formal business forecast.
The strongest institutional friction is employer-provided health insurance. In the next-wave model, the coefficient is −0.0085 for the full sample and −0.0131 for men; in the ever-switch model, it is −0.0366 for the full sample and −0.0580 for men. The paper interprets these negative associations as job lock. Leaving the primary job can mean losing coverage, and the value of that benefit can make a promising side venture insufficiently attractive to justify the switch. In the long-run full-sample estimate, the magnitude is about a 3.7-percentage-point lower probability of transition, consistent with the abstract’s summary.
The gender pattern adds a boundary rather than a separate conclusion. The health-insurance coefficient is smaller and not statistically significant in the women-only specifications, while the negative association is clearer for men. The study does not establish why. A non-significant estimate does not prove that employer coverage has no effect for women; it means that this analysis does not estimate a sufficiently clear effect in that subgroup. The same caution applies to the positive 50% moonlighting-share relationship, which remains meaningful for women over the longer horizon even though the short-run women-only estimate is not statistically significant.
The proposed household substitute does not appear in the data as expected. If a spouse has employer-provided health insurance, the moonlighter might be able to leave their own job without losing household coverage. But the spousal-health-insurance coefficient is small, statistically insignificant, and inconsistently signed across the specifications. The evidence therefore supports a barrier associated with one’s own employer coverage, not a measurable conclusion that spousal coverage cancels that barrier. The two forms of insurance may differ in accessibility, household bargaining, eligibility, or the broader value of the job, but those mechanisms are not separately identified here.
Other household measures are similarly mixed. Children are negative in most specifications but generally insignificant, with a small marginal effect for women in the next-wave model. The sole-earner indicator and its interaction with marriage are not consistently significant. Marriage is positive and significant in the full and men-only next-wave estimates, and spousal earnings are positive but only marginally significant there; neither pattern remains clearly significant in the ever-switch model. This provides some support for the idea that households can share risk, but the support depends on the time horizon. It is not evidence that marriage itself causes entrepreneurship, nor that family responsibilities universally suppress it.
The broad interpretation is consequently conditional. Strong side-work income is associated with a higher likelihood of entry, and weaker primary-job attachment is associated with movement in the same direction. Employer health insurance is associated with staying out of full-time self-employment, while household support appears partial rather than decisive. These relationships fit an option whose exercise depends on both the evidence generated by the venture and the protection that would be lost. They do not describe a deterministic sequence in which a particular income share forces a person to quit.
The robustness checks strengthen the shape of the result without changing its status. Logit and Probit estimates track the linear-probability results; the continuous income share retains its positive relationship; and alternative definitions of part-time, full-time, only-full-time, or incorporated self-employment preserve the main patterns. Repeated associations are more persuasive than a single coefficient, but they remain associations in observational data. A stronger conclusion would require richer measures of the venture and the person’s intent, as well as a way to address selection into moonlighting and the unmeasured resources that make a transition possible.
The panel study therefore answers a focused question: which conditions are present around realized transitions? It cannot tell us whether someone planned the timing, whether a delayed move was deliberate, or how people form a career path before they have a primary job. The next perspective begins before the employment state is fixed and asks whether combining or delaying work can itself be an intended design.