Dropping the noncompete lifted recruiters' earnings 12% to 17% in a 14,000-offer randomized trial

Every contract in the experiment carried a nondisclosure agreement, including the ones with no noncompete ^[6]. That control is what gives the secrets result its meaning. The comparison is a noncompete plus an NDA against an NDA alone, and the researchers found no evidence that the added clause reduced leakage of sensitive information ^[7]. A firm relying on a noncompete with no NDA behind it is outside the test.

Measuring leakage at all is the unusual part. "It's difficult to test how noncompetes affect secret sharing outside of an experiment like this because secrets are, generally, secret and thus hard to observe," Starr said ^[14]. The team partnered with two finance firms hiring thousands of freelance recruiters on short-term contracts ^[16]. They then tracked how each worker reviewed the contract, whether they accepted, and how they answered later job opportunities that could breach the noncompete or the NDA ^[17]. "By occupying the role of the employer and randomizing contract conditions, we can finally observe how workers respond to these provisions in real time," Starr said ^[15].

Employers have long argued that workers would not give up future job options unless they were paid for it ^[20]. That argument assumes the worker reads the clause. Many in this experiment skimmed past the noncompete in seconds, and a substantial share did not know about it until a firm contacted them after employment ^[9].

The result on unenforceable clauses arrives in the middle of a policy fight. A nationwide ban proposed by the Federal Trade Commission was halted in court, and states continue to advance their own restrictions ^[18]. "In other words, the law doesn't matter," Starr said. "The presence of the clause alone changes behavior." ^[11]

The authors, writing in The Quarterly Journal of Economics ^[1], hedge their own conclusion: "At least in our setting, noncompetes reduced mobility and earnings without reducing the spread of firm secrets," they wrote ^[13]. The hedge is warranted. The workers were short-term recruiters, and whether a salaried engineer holding years of proprietary work responds the same way is beyond what this design can test. The published summary also does not report the baseline rate of moves between competitors behind the 36% to 52% gain ^[3].

For roles like these, I think the trade-secret case for the noncompete no longer holds up as an empirical argument. Applying that verdict to every job goes beyond what a study at two firms can show ^[2].