Noncompete Enforcement Reduces Mobility and Innovation Despite Firm Investment Claims

A noncompete clause is a restriction inside an employment contract that bars a departing worker from joining or starting a competitor, usually for a stated period and sometimes within a stated geography. Evan Starr described it as one member of a broader family of “restrictive covenants,” alongside nondisclosure agreements and nonsolicitation agreements covering clients or former coworkers.

The distinctive feature of a noncompete is that it limits where the worker can work after leaving. In Starr’s description, the scope can be narrow or extremely broad: one or two years is common, some restrictions last longer, and the geographic boundary can range from a few miles around a workplace to a state, the United States, or “the world.”

The broadest examples matter because they undermine the common assumption that noncompetes are mainly about highly paid employees with access to trade secrets. Starr pointed to Jimmy John’s workers around 2015 who, while making sandwiches, were asked to sign a two-year noncompete that would bar them from working in nearly every food establishment within three miles of any Jimmy John’s location. He also cited a temporarily employed Amazon packer making $13 an hour who had an 18-month noncompete restricting him from working on anything Amazon produced or manufactured.

Steven Davis drew the implication directly: the Jimmy John’s example shows that the issue is not only about high-end knowledge workers, technical employees, or salespeople with access to unusually valuable proprietary information. There may be know-how in making a sandwich, Davis said, but it is not the sort of know-how one ordinarily thinks of as comparable to the Coca-Cola formula or the algorithm behind an AI company. Starr allowed that Jimmy John’s lawyers might argue their sandwich-making process contains confidential or proprietary elements, but agreed that the tools and process of making sandwiches differ from highly detailed technical secrets.

The systematic evidence Starr summarized points in the same direction. Noncompetes are more common in high-skill jobs, especially where workers have access to trade secrets or clients. Among workers with trade-secret access or client access, Starr said, about one in three have a noncompete. Higher-paid workers also face them at relatively high rates. But the practice extends down the wage distribution.

Even among workers in the lowest earnings deciles, Starr said, roughly one in ten have noncompetes. At the firm level, several surveys show that about half of firms report using them, and about 30 percent of those firms use them for every worker. That is how noncompetes reach not only highly paid employees but janitors and unpaid interns.

Davis connected that pattern to a contract-writing habit he has seen in large organizations: lawyers often draft the most sweeping language favorable to their client, then insert that language broadly. A clause that might make business sense for some employees can become boilerplate for an entire workforce. Starr said that lawyers have given him a similar account. Their orientation is risk mitigation, a “belt and suspenders” approach that tries to cover even unlikely risks, such as a janitor stealing a thumb drive or an independent contractor with little relationship to the firm beyond the contract.

The classic economic defense of noncompetes assumes that a worker knowingly accepts a restriction in exchange for something else. Evan Starr laid out that version. A worker might receive a wage premium, better amenities, training, or other benefits whose expected value exceeds the future cost of limited mobility. A worker might also expect the clause not to matter: perhaps they plan to retire soon, do not expect to leave, or believe the restricted opportunities are irrelevant.

Steven Davis raised the assumption embedded in that model: what if the worker does not really understand what they are accepting?

Starr described a field experiment with Bo Cowgill and Brandon Freiberg in which real workers were randomly given contracts, some containing a noncompete. The researchers tracked how long workers spent on the noncompete page, which was page seven of the contract. About one-third skipped the page entirely. The median time spent on it was two and a half seconds.

The problem is not only inattentiveness. Starr said somewhere between 30 and 50 percent of noncompetes are deployed after the worker has already accepted the job, often on the first day. Davis called that a hold-up problem. The worker may have moved a family, turned down another job, and arrived for work before HR presents a stack of forms to sign. The worker may find themselves under a noncompete only after the practical costs of refusing have risen sharply.

Davis did not insist that this is always strategic. It may be administratively convenient to collect all paperwork on the first day, especially when much of it is required by law. But that convenience also creates an opportunity to introduce restrictions in a setting where they may not receive serious consideration. Starr added that noncompetes sometimes appear not in the employment contract itself but in employee handbooks, which the worker may not encounter until later.

Davis broadened the point beyond noncompetes. Employment contracts and handbooks can run to many pages, sometimes clear, sometimes dense with legal language. In his experience in senior academic leadership roles, most people do not read and fully digest such materials before accepting a job, and he did not suggest they should. If workers spent their lives reading every contract they were handed, they would have little time left for anything else.

Starr noted one possible mitigation: AI tools may help workers summarize employment contracts, identify key provisions, and ask specifically whether a noncompete is present. Davis agreed that this could help, especially if the worker can train the tool over time to flag provisions of personal concern. But that is a partial response to a deeper problem: contractual consent can be formally clear while being practically weak.

A noncompete can influence behavior even when it is unenforceable. Evan Starr referred to “bogus” noncompetes: clauses employers use despite good reasons to think courts in that state would not enforce them. Steven Davis asked whether that was the meaning, and Starr confirmed it.

Several states do not enforce noncompetes. California, Oklahoma, and North Dakota banned noncompete clauses in the late 1800s, Starr said, and Minnesota did so in 2023. Yet companies have continued to use noncompete clauses in some of those states.

The mechanism is simple. A worker may read the clause, assume it is binding, and avoid a better job or a startup opportunity rather than risk a legal fight with a former employer. Davis added another reason: some people treat agreements as a matter of honor. If they signed something, they try to abide by it, even if the law would not require them to.

This is one reason enforcement policy cannot be evaluated only by asking what a court would ultimately do. The practical effect of a clause may occur before litigation, and often without litigation. Starr later described the difficulty of addressing such cases through enforcement: many states lack a private right of action or a clear red line that makes it easy to collect damages when employers use unenforceable clauses.

Washington State provided Starr’s example of a different design. In 2020, Washington adopted a bright-line rule banning noncompetes for workers making below $100,000 a year, with the threshold tied to inflation. Starr said the law also gave workers a private right of action to sue employers for what he thought was $5,000. That combination made enforcement comparatively easy: if a worker earned below the threshold, the clause was barred, and law firms had an incentive to bring claims. Class action filings in Washington included cases on behalf of workers such as carpet installers, and Starr said he had spoken with attorneys in the state about that behavior.

The point was not that Washington’s design resolves all questions. It was that legal clarity and remedies matter. A vague reasonableness standard may leave workers deterred by clauses that would not survive challenge. A bright-line rule plus a private enforcement mechanism changes the incentives for employers and plaintiffs’ lawyers alike.

The employer’s rationale divides into two distinct stories. The first is straightforwardly anticompetitive: a noncompete can shield the employer from competing for its own workers. If a rival offers higher pay, the worker may be unable to leave. If a worker wants to start a competing business, the noncompete can block that too. From the firm’s standpoint, the clause can help retain workers longer, pay them less, and reduce product-market competition.

The second rationale is the one courts most often recognize and the one that gives the policy debate its force. Evan Starr said noncompetes can encourage firms to create valuable information, invest in workers, develop customer relationships, and share sensitive knowledge with employees without fear that the employee will immediately carry that knowledge to a rival or use it to start a competitor.

Steven Davis pressed the economic version of the argument, especially around training. Starr noted a legal distinction: in most states, courts do not uphold noncompetes simply on the basis of training investments. Davis set that aside to make the economic point. If an employer pays for training that increases a worker’s value elsewhere, and the worker can immediately leave to use that training for a rival, the employer may underinvest in training in the first place. Starr agreed with the economic structure of the claim: the worry is “subsidizing your competitors.” The same logic applies to firm-specific information and client-specific relationships.

That creates a real line-drawing problem. Davis framed it as a mixture of business motivations that sometimes align with social value and sometimes do not. On one side, a firm may have a legitimate reason to protect investments that produce value beyond the immediate employment relationship. On the other, a firm may simply want to suppress wage competition, reduce employee mobility, and prevent new rivals.

Starr accepted that framing, using the vocabulary of procompetitive and anticompetitive justifications. There are procompetitive reasons for noncompetes, he said, but also anticompetitive ones. The hard question is not whether either category exists; it is how often each dominates, in which settings, and whether law can distinguish them at reasonable cost.

The most important empirical tension concerns innovation. The procompetitive case for noncompetes says that enforceability should induce firms to invest, create valuable information, and share it with employees. If that were the only mechanism, noncompetes might increase innovation.

Evan Starr said two recent studies reach the same broad conclusion in the evidence he was summarizing. He referred to one as by “Runmuth and Rockall” and another as by Johnson, Lipsitz, and Pei, and described them as examining how state-level changes in noncompete policy affect innovation, including patents and other measures. In Starr’s account of those studies, when states enforce noncompetes, innovation tends to fall, with high-impact innovations falling even more.

The mechanisms, as Starr described them, are slower inventor mobility and reduced knowledge sharing. Innovation does not occur in a vacuum. It often comes from recombination: workers move, bring different ways of thinking, mix ideas from different settings, and generate something new. Noncompetes impede that movement and therefore the recombination process.

At the same time, Starr said there is evidence that enforcement increases firm investment. Steven Davis restated the duality: both claims can be true. Enforcing noncompetes can raise explicit monetary investment in innovation while reducing the overall rate of innovation because the broader ecosystem depends not only on investment dollars but on the movement and exchange of ideas.

Starr agreed and added that new firms are also a source of innovation. If noncompetes reduce startup formation, the effect on innovation can be larger than the direct effect on incumbent firms’ investment decisions.

Davis pointed to Silicon Valley as an accessible example. California does not enforce noncompetes, and Silicon Valley is highly innovative, both commercially and technically. That alone is not definitive, he said, but it should give pause. Starr noted the standard rejoinder from noncompete proponents: where is “Silicon North Dakota” or “Silicon Oklahoma,” the other long-standing nonenforcement states?

Davis then returned to heterogeneity. Even if the average effects Starr described suggest noncompete enforcement reduces innovation, that does not prove the clauses are harmful in every industry, technology, or contractual setting. He asked whether there are domains where enforcement appears to work well.

Starr did not offer a specific success case. Instead, he framed it as an open research problem. Much empirical work estimates averages or conditional means. Even when researchers examine heterogeneity, they may miss individual noncompetes that are procompetitive. Davis reinforced the point: regression models often say little about the tails of the distribution unless researchers work hard to study them, and behavior at the tails can differ not only in outcome but in causal structure.

For Starr, Minnesota’s recent ban creates a valuable modern test case precisely because it reaches executives as well as lower-wage workers. As executive contracts renew without noncompetes, researchers can study whether firms need noncompetes even at the upper end, and how they change their contracting strategies when the tool is unavailable. Starr described Minnesota as the first modern natural experiment of that kind, because previous policy makers had not been as bold.

Asked why not simply ban noncompetes, Evan Starr said he does not view a ban as an unreasonable policy idea. Long-running examples already exist: California, North Dakota, and Oklahoma have not enforced noncompetes for more than a century, and Minnesota recently joined the list of states banning them. Starr also noted the argument that Silicon Valley would not have become Silicon Valley without California’s ban, though he characterized that as a continuing policy debate rather than a settled fact.

Attorneys provide another kind of ban. Starr said they are the only occupation in the United States for whom noncompete clauses are barred nationwide, and he attributed that bar to American Bar Association rules. The justification, he said, is partly about negative externalities. If a lawyer is excluded from practicing because of a noncompete, the client loses access to the attorney of choice. The rule, in Starr’s account, rests on that client-externality logic.

The externality point generalizes. Even if a noncompete is good for a firm and an executive who knowingly consents, it may still be bad for society. It may reduce competition, slow entry, limit worker mobility, or deprive customers of preferred services.

Steven Davis nonetheless resisted a categorical national ban. His objection was not that noncompetes are generally harmless. It was that the earlier discussion had already identified circumstances in which parties may have procompetitive reasons to use them. Davis said he found it hard to believe those circumstances never dominate. A business model that relies on enforceable noncompetes might simply avoid California; that does not prove the model should be impossible anywhere in the United States.

The disagreement, then, was not over whether noncompetes can be abused. Both accepted that they can. The question was whether the abuses are best addressed through a broad ban, narrower rules, or case-by-case review. Davis emphasized alternatives that target the specific interest at stake. If the concern is a customer list, a nonsolicitation agreement may do the job. If the concern is confidential information, an NDA may be more tailored.

But narrow tailoring has limits. Economists often imagine a world in which it is costless to specify every contingency in a contract. In that world, one could design a precise restriction for each risk. Real contracts are incomplete. They are costly to draft, cannot anticipate every contingency, and may not capture all the ways legitimate business interests can be harmed. In some uncommon circumstances, Davis argued, a noncompete may be the natural and efficient contractual tool—not mandated, but allowed.

Starr’s response shifted the unit of analysis. The debate, he said, often gets stuck on the efficiency of a bilateral relationship. But the market-level consequences can be different. He invoked Ronald Gilson’s 1999 idea that noncompetes can create a prisoner’s dilemma. Each firm controls only its own workers. If one firm does not use noncompetes, others can poach from it. If it does use them, others cannot. Each firm may have a private incentive to adopt noncompetes, but if every firm uses them, mobility falls, startups become harder, hiring into an industry becomes more difficult, acquisitions become a more important mode of growth, concentration rises, and consumers can be affected.

Davis accepted that as a reasonable argument for inhibiting or banning noncompetes in some settings. But he continued to look for a policy instrument that could handle variation rather than erase it.

Davis suggested an analogy to antitrust law’s rule-of-reason approach. In complex settings where it is hard to legislate every contingency in advance, courts can apply broad principles and build case law over time. The central question becomes whether a practice is, on balance, procompetitive or anticompetitive.

Evan Starr answered that this is essentially what noncompete law has done for centuries, though usually not under antitrust law. The first noncompete case, he said, dates to 1414. The seminal English case came in 1711, establishing a reasonableness criterion. For a noncompete to be reasonable, it must protect some legitimate firm interest, such as trade secrets, clients, or other valuable assets, and it must not unduly harm the worker or society. That test crossed the Atlantic, and most U.S. states adopted it, apart from the states that ban noncompetes.

Steven Davis was struck that a 1711 rule sounded so modern: effectively a rule-of-reason framework.

The problem is that “reasonable” does not mean the same thing everywhere. Starr gave several examples of state variation. Florida, which he described as the most vigorous noncompete-enforcing state, had a provision under which a worker challenging a noncompete could not argue that the clause unduly harmed the worker. Starr contrasted that with other states where worker harm is part of the analysis. Some states allow enforcement even when the worker was fired without cause; others do not. Some require additional “consideration” when a noncompete is imposed mid-employment; in others, continued employment is enough.

Davis argued that antitrust authorities have historically paid much more attention to product markets than labor markets, which he called a mistake. Recent years have brought some efforts to correct that imbalance. He nevertheless pressed whether more aggressive antitrust enforcement could serve as an alternative to detailed rule-writing.

The central obstacle is that antitrust law is usually concerned with market-level harm, while noncompetes are signed one worker at a time. A single noncompete typically will not move market wages or prices. That makes it harder to fit into the antitrust framework than a merger or wage-fixing agreement.

Starr cited a Federal Trade Commission case involving glass-container manufacturers in a concentrated industry where multiple firms used noncompetes. There, the FTC alleged market-level harm and reached an agreement with the manufacturers. That kind of case fits the market frame. But the individual noncompete remains harder for antitrust law to reach.

He also described an FTC case against Prudential Security. The company used a noncompete with a $100,000 liquidated-damages provision for violation. A Michigan court had previously found the noncompete unenforceable, but Prudential kept using it. The FTC pursued the company and got it to remove the clause.

Davis took from this that antitrust tools have potential but have been underdeveloped and underused for noncompetes. Starr agreed, with the caveat that legal structure matters: the FTC, as he understood it, lacks the power to bring class actions in the way private plaintiffs might in state court, and many states do not make damages easy to collect.

Noncompete law in the United States remains highly state-specific. Most states inherited some version of the reasonableness test, but they differ sharply in what they count as reasonable, which workers they protect, what remedies they provide, and whether they draw bright lines by income or occupation.

That variation is now a major source of evidence. Evan Starr said that when he began studying noncompetes around 2012 or 2013, the research was coming mainly from management journals and had not yet reached the economics mainstream. He was on the academic job market in 2014 trying to convince economists that noncompetes were an important subject. Six years later, he said, the issue was on the presidential agenda of almost every Democratic presidential candidate, and the Obama administration had found the Jimmy John’s and Amazon examples as new evidence was coming out.

Steven Davis said noncompetes remain understudied relative to their importance. Compared with the minimum wage, the economics literature on noncompetes is still tiny, even though Davis suggested the area would benefit greatly from more economic insight.

Starr described the origin of his research agenda as a basic measurement problem. Working with Charlie Brown, J.J. Prescott, and Norm Bishara, he wanted to know how common noncompetes were, what was actually in employment contracts, how the contracting process worked, and how it mattered for firms, workers, and society. He was able to add questions to surveys, and since then the growing policy variation across states has created more opportunities for study.

The policy options Starr and Davis considered each match a different failure. Low-wage workers signing restrictions they do not read and cannot challenge point toward bright-line bans and private enforcement. Market-wide suppression of mobility, entry, and knowledge recombination points toward broader bans or antitrust-style intervention. Legitimate protection of firm investments in information, clients, and training points toward narrower covenants or case-by-case reasonableness review.

For states, the practical choice is therefore not a single abstract position on noncompetes, but a design problem: whether to use bright-line bans for vulnerable workers, enforceability limits for executives and technical workers, and remedies that deter employers from using bogus clauses.

None of those instruments fully resolves the tension Davis kept returning to: abuses and average harms do not prove that every noncompete is socially wasteful, while legitimate bilateral uses do not answer Starr’s market-level concern that widespread adoption can reduce mobility, startup formation, and innovation.