BLAW Searches for Bias in Civil Asset Forfeiture Cases

When police officers or other law enforcement personnel take property, money or other assets from people suspected, but not convicted, of a crime, it’s known as a civil asset forfeiture (or CAF).

This practice is controversial for numerous reasons, not the least of which is that the person who loses the property need not be found guilty of a crime in order for law enforcement to seize it. Probable cause — essentially the suspicion that the person may have committed a crime — is usually enough.

Guilty or not, affected people have to sue to get their assets back, and, in the process, must prove they didn’t commit a crime or that they were otherwise wrongly implicated. And, legally speaking, the dispute is between the law enforcement agency and the property itself.

Legal scholars and social activists have, for years, asserted that law enforcement uses CAF procedures in a biased manner as a way of targeting and harassing people of a particular race or class. Most studies conducted on the question to date have relied on data provided by the U.S. Department of Justice. Yet, a large percentage of CAF actions occur at the state and local level.

A team of legal researchers and data scientists at Bloomberg has sought to use another source of data – court case dockets – to determine whether trends in CAF proceedings at the federal, state and local court levels might indicate a pattern of bias. The answer so far appears to be no, but there’s more work to be done on the question.

In a paper presented at the 2017 Data for Good Exchange at Bloomberg’s Global HQ in New York last week, Wayne Krug, a senior software engineer on the Bloomberg Law (BLAW) machine learning team explained the questions posed by this project.

“Can we see any indication of bias within the docket data, and, can we specifically see a targeting of any disadvantaged group,” Krug said in his remarks. “Or increased rates of enforcement in border states? Or whether targets are chosen specifically for the amount of assets that can be obtained?”

The team gathered its data in two ways. First, it conducted a full-text search of Bloomberg Law’s dockets repository for civil cases brought by law enforcement naming specific amounts of U.S. currency as the defendant. “It’s a fairly reliable way of finding CAF cases,” he said.

Next, they extracted metadata from the docket files looking for the amount of cash involved, the cause of action, and other underlying facts about the case.

Krug said one key piece of data that would usually hint at bias was missing: the race or national origin of the person from whom the property had been seized. “Court dockets rarely contain demographic data, so it was difficult to get any insight into some of the racial, ethnic and national origin biases that might apply.”

In all, the team examined CAF cases in all U.S. federal and state court systems, plus an additional 922 local jurisdictions – a total of about 117,000 cases. Of those, about 72,000 had sufficient metadata to be useful.

What they found is upward trends over time, both in the volume of CAF cases per year, and in the amount of cash seized. They also found that states with larger populations tend to have more CAF cases, and that the volume tends to track closely with population. Three states were outliers and had a higher volume of cases relative to their population: Nevada, Oklahoma and Arizona. Krug said one possible reason is that these three states each have laws that allow police agencies to keep 100 percent of the proceeds from forfeitures.

However, they found no link between the size of a state’s population and the amount of money seized in that state. “The fact that the dollar amounts correlate relatively poorly with population compared to case volume may indicate some bias in the practice of CAF,” the team wrote in the paper. “However, it could also be explained by actual spikes in drug crime that may run counter to the population.”

“We are seeing an increase in the number,” of CAF cases, Krug said. “But we’re not able to find conclusive evidence of bias in the docket data.” There are, he said, some “fishy trends” worth investigating further, including one seen in some states that saw an increase in seizures of smaller dollar amounts from single individuals.

In future studies, the team hopes to “close the gap” by adding more sources of demographic data, and additional information drawn from court filings, including pleadings and judicial opinions. “We want to try and get more information about the defendants in the underlying cases, and what the results of those cases were.”