Governors face every temptation to put their thumbs on the scale in the run-up to an election. Hold off on layoffs until after the vote. Lift the hiring freeze a few months early. Veto budget cuts, and race to sign a big new contract that will goose employment.

Economists have long struggled to prove the real-world effects of such machinations in the United States. Instead, they have shifted their focus to developing countries that, University of California at San Diego PhD candidate Dodge Cahan said, “are perceived as being more susceptible to these kinds of things.”

But story after story convinced Cahan that something similar must be going on here.

It’s difficult to prove. The labor market, even at the state level, is a fidgety, chaotic beast. Every day, businesses hire and fire. People graduate or retire. A single store closure or factory opening can obscure myriad moves in state and local government.

But Cahan saw a way to cut through the noise with detailed real-world data and statistical methodology that wasn’t available to earlier researchers.

In a study published in the European Economic Review, Cahan analyzed county jobs numbers during 359 gubernatorial elections from 1990 to 2015.

In the quarter before an election, state government employment was up by about seven jobs per 100,000 residents and local government employment was up by 13. Both figures are too large and too persistent to be explained away as random noise in the data. They are, Cahan writes, “consistent with manipulation.”

In the economic literature, “manipulation” is an informal term for actions motivated by something other than society’s best interests.

A few quarters after the election, government job numbers return to normal. It suggests governors were following a similar playbook to Connecticut’s Dannel Malloy. In 2014, the Democratic governor denied the state had budget problems until two weeks after he’d won reelection, at which point he announced a hiring freeze.

States elect governors in different years, so most elections come with a built-in control group: counties in states that aren’t holding elections in the same year.

After excluding outliers and counties with insufficient data, Cahan was able to analyze 1,751 counties representing about three-quarters of the U.S. population. After controlling for seasonal variation and other complicating factors, he isolated the manipulation outlined above.

To confirm his analysis, Cahan considered 430 pairs of counties on state borders, giving him the opportunity to compare areas facing elections with their across-the-border neighbors, which were in an off-year. The two counties were presumably similar, allowing Cahan to rule out long-term trends and geographic factors. The results didn’t change.

UMass Amherst economist Arindrajit Dube, who used a similar strategy to compare minimum-wage impacts, said comparing neighboring counties across state borders gives findings additional credibility, especially when attempting to show causality.

Governors typically don’t have direct power over local government hiring, but they influence it via spending and political channels. And because there are many more local government jobs than state government ones, the sector presents more opportunities for political interference.

Cahan discovered the pre-election jump in local government jobs was much higher when a county’s voters leaned toward the governor’s political party, suggesting that local officials tend to lean on the labor market to boost their party’s candidate in the election.

He also found that the hiring increase in state government was higher during close elections, and in cases where the legislature was controlled by the governor’s party.

It’s unlikely the variation is caused by seasonal poll workers, whose brief work periods and low pay mean they rarely show up in this data.

After he was reelected in 2014, Florida Gov. Rick Scott (R) axed about nine times as much spending from the budget and supplemental measures than he did when he was seeking reelection the year before, Cahan found. The casualties included programs he’d supported during his campaign.

“The problem isn’t as big as it might be in some developing countries,” Cahan said. “We don’t have a massive problem here, but it does still exist."