Say the head of a claims organization has just approved a new pay plan. For as long as anyone can remember, raises in her group arrived on a schedule and differed very little from one person to the next. The new plan replaces those steps with merit raises and wider pay bands, so that managers can pay their best people noticeably more. The case for it fits on one slide: pay should follow performance, and people will respond by performing.

Six months in, the first readout is encouraging. Output per person is up, and managers report that raise conversations have become easier to have. She also notices, with only passing interest, that resignations are running a little above last year, and she files that under the ordinary churn of a tight labor market.

A long payroll file with a built in comparison

A working paper first circulated in 2024 and revised in August 2026 follows what happens after a change like hers. Jason Chen of Auburn University and his coauthors Jakub Hajda and Joseph Kalmenovitz studied U.S. financial agencies that moved their staff from pay that rose nearly automatically to pay that depended on merit. Seven agencies made that move between 1981 and 2006, among them the SEC and the FDIC. For comparison the authors used three similar agencies whose pay rules never changed, two of which had paid for performance from the day they opened, along with agencies that had not yet made the switch. The economists followed roughly 31,000 regulators through payroll records covering 1973 to 2013 and compared the agencies that switched with those that had not.1

More regulatory output and more resignations

The reform did what its sponsors wanted, at least where that could be measured. The Federal Register began naming the staff who drafted rules in 2000, so this test covers only the three switches that fell inside that record, at the SEC in 2002, the FDIC in 2003 and the CFTC in 2006, and the 743 regulators who drafted rules there. Among them, the chance of working on a new regulatory document in a given year rose 58 to 63 percent relative to the sample average.1

It also did something that never made the planning slide. Voluntary exits from government employment rose by 40 to 54 percent, which the authors calculate as an increase of 3.1 to 4.2 percentage points on an average annual departure rate of 7.8 percent.1 About 90 percent of that increase came from people under sixty, whom the authors treat as likelier to be moving to private sector jobs than retiring.1

The plan sorted people by what they could get elsewhere and why they came

The next question was who left. Compared with similar leavers from agencies that had not made the switch, regulators who left after the reform took new jobs with starting salaries 38 percent higher and total compensation 53 percent higher.1 The departures also tracked how people had come to the agency in the first place. The authors read the job market at the moment someone joined as a sign of how attached that person was to public service, since anyone who signed on while private employers were hiring freely gave up more to be there. Regulators who entered during recessions, when private employers were hiring little, showed a reform effect on exits about twice as large as those who entered in expansions, 0.054 against 0.025.1 The authors compress the finding into one sentence: "Performance pay therefore sorts on that attachment."1

A pay plan changes how hard the people who stay work and also changes who stays, and the first readout usually reports only the first change.

The authors give two reasons for the exits. The reform raised what regulators could earn outside, partly because harder work made each person's value easier for private employers to see. It also made staying less appealing, because pay now carried more risk and rested on evaluations that some people did not trust. The evidence on trust is the softer part: exits rose most where few colleagues did similar work and ratings were presumably noisier, while a 2008 employee survey gave a weaker signal.1 As I read it, the plan raised the price of staying, and the people best placed to decline that price were the ones with somewhere else to go.

What the first readout cannot show

The paper does not measure ability directly, so it cannot settle whether the people who left were the strongest or the weakest, though the authors suspect they were among those contributing most to the agencies' technical capacity. Output in the study is counted in regulatory documents, which tally volume and say nothing about quality, and the authors leave open whether the agencies ended up more or less effective overall.1 They also point out that a regulator's outside option is usually the industry being regulated, a conflict of interest that does not arise for corporate executives.1 I would not carry the 40 to 54 percent figure into a company. The authors do report that a separate 2004 pay reform for federal senior executives raised exits by roughly 35 percent of the average rate, which they take as evidence that the pattern reaches beyond financial regulators.1

That makes the first readout a half report. The mechanism also differs from the one in an earlier piece about a retailer that paid apprentices to show up, where the payment changed what attendance meant to the people who stayed. Here the change shows up in who is still on the roster.

Before a plan like this goes live, I would write down the people whose departure would cost the most and ask two things about each: what they could earn elsewhere, and what keeps them here today. The answers show which part of the roster the plan is about to test. When the first readout arrives, I would ask for the exit list on the same page as the output chart, with tenure and level beside each name, before anyone presents the improvement as settled.

If your last pay change produced a good first readout, do you know who stopped appearing on the roster afterward, and what each of them would have told you about why they were still there the year before?