Of the many surprising statistics about America’s money bail system, this one may be the most astounding: More than 60 percent of people in America’s overcrowded jails are there because they can’t afford to pay their bail amount. That works out to roughly 450,000 Americans in jail daily, and how long they stay there can vary with waiting times for trials potentially lasting months (or sometimes, years).

The American money-bail system, which has been around since 1789, has ripple effect. Some reformers argue that poor defendants might plead guilty in order to be released. Others say that there are more effective alternatives to money bail, such as using a risk score or supervising defendants before trial. Concerns about the use of money in the bail system and the bail bond industry have also raised questions about America’s pretrial system and the way it affects the lives of unconvicted people.

A bail is set based on several factors: the severity of the crime, whether a defendant is a flight risk, and whether a defendant has a criminal record or has skipped trial in the past. Previous studies have found that pretrial detentions are a predictor of convictions and guilty pleas, as well as the likelihood of future crime. But that correlation might exist because some of the factors in the way bail is set are likely related to guilt—so correlations found between bail and convictions have been hard to interpret in the sense that past studies have not been able to pinpoint whether those negative outcomes are really due to bail being assigned.

Now, two economists from Columbia University and a public defender from Maryland have co-authored a study showing the size of the impact on assigning money bail on the likelihood of a guilty plea and recidivism. The experiment uses criminal data from the arraignment system in Philadelphia from 2010 to 2015, where the assignment of defendants to bail magistrates is close to random. This randomness is ideal for what economists call a natural experiment, where defendants are exposed to different conditions (in this case, different magistrates with different predispositions to assigning bail) at random.

Chris Hansman, a PhD candidate in economics at Columbia University and a co-author of the paper, says that he and his fellow researchers found that being assigned money bail increases the probability of conviction by about 6 percentage points and also causes a 4 percentage point increase in the risk that someone would go on to commit another crime.

The implication of these findings is that the money-bail system imposes costs on the defendants and the public in the form of conviction and new crime. While this experiment only used data from Philadelphia, the authors argue that since Philadelphia’s bail system is similar to those used around the country—their results may be insightful in estimating the impact of money bail nationwide, although the causal impact of money bail is likely lower in geographical regions where defendants are wealthy (and can afford bail). Further, Hansman says that another researcher has independently confirmed their results: That paper used the same data set from Philadelphia, and concluded that pretrial detention led to a 6.6 percentage point increase in the likelihood of conviction.

The novelty of these studies is the causal link that being assigned bail (or being detained before trial) leads to a higher chance of conviction. “Many of these convictions appear to be the result of guilty pleas by detained persons,” says Ethan Frenchman, a public defender in Baltimore and co-author of the study. “Our paper further demonstrates how the American money-bail system is not only grossly unfair and irrational, but imposes far greater harms on defendants and the public than we realize.”