Bringing big data to the fight against benefits fraud

21 February 2015 - 10:09 By New York Times
subscribe Just R20 for the first month. Support independent journalism by subscribing to our digital news package.
Subscribe now
File photo.
File photo.
Image: Gallo Images/Thinkstock

A few years ago, the New York City Human Resources Administration decided to try a new way to root out fraud among people receiving government benefits.

Data detectives began running benefit recipients through a computerized pattern-recognition system.

They discovered that the behavior of a small percentage of people stood out. The anomalies in themselves didn’t constitute fraud, but they pointed the agency’s data scientists in potentially fruitful directions.

One of those outliers, for instance, was Parvawattie Raghunandan, a Bronx resident whose family had received more than $50,000 in health benefits over a decade. Her case was unusual because most families of similar size and income typically received multiple benefits - like health coverage, food stamps and cash assistance - but Raghunandan had applied only for Medicaid for herself and her three children, agency officials said.

So the data scientists followed up by searching state records on business ownership and car registration for more information about her family’s situation. They also tapped into a national database on property ownership from LexisNexis Risk Solutions, an information and analytics division of Reed Elsevier.

Investigators subsequently concluded that the family had under reported its assets, among them: an electrical contracting business, owned by Raghunandan’s husband, where she had claimed to work for a low wage; three residential properties in New York and one out of state; and joint bank accounts with more than $100,000, according to agency officials.

The case culminated this month in criminal charges - grand larceny and making false statements to a public office - being brought against Raghunandan by the Bronx district attorney. She has pleaded not guilty.

Agency officials say that this kind of multisource data analysis has helped them uncover more benefit abuse with less effort. Last year, agency staff members completed nearly 30,000 investigations and identified about $46.5 million in fraud compared with nearly 48,000 investigations and about $29 million in fraud in 2009, before the agency began systematic data analysis of recipients.

“The data-mining process is extremely important,” Steven Banks, the agency’s commissioner, told me recently. “It allows us to zero in on likely fraud so we don’t divert resources to finding what otherwise might be a needle in a haystack.”

But Todd A. Spodek, a lawyer representing Raghunandan, who is originally from Guyana, said, “I think there is a fundamental flaw with relying on data analytics to determine criminal culpability. New immigrants are often seduced by enrollment agents to sign up for benefits without understanding fully the process.”

Business intelligence companies like IBM, SAS and LexisNexis have long provided predictive computer modeling techniques to financial services companies seeking to inhibit fraud. But now some state and local government agencies are turning to these services. Some agencies use the software to integrate and analyze their own files on benefit applicants; others are augmenting their records with commercial data - such as lists of luxury car purchases. They are all arming themselves with data-mining software in an effort to keep up with the increasingly complicated nature of benefits fraud.

Last month, for instance, state officials in North Carolina announced the discovery of a new kind of unemployment scheme there. With the help of SAS data integration and analytics software, the state’s Division of Employment Security identified more than 100 fictitious employers who had reported wages and paid taxes to the state for imaginary employees - and then filed 672 fraudulent unemployment claims.

“Fraud has been around for many years, but the nature of the threats that governments are facing is changing,” says Shaun Barry, the principal solutions architect of SAS’ security intelligence practice. “Fraudsters are getting more organized and sophisticated, using advanced analytical techniques and taking advantage of the lack of communication between government agencies.”

Barry says the company’s government business had been experiencing “explosive growth” - with 22 agencies in 14 states now using SAS software to mitigate fraud, waste and abuse.

Some agencies are turning to commercial data-mining services because they are frustrated by the lack of integration between state and local government records across the country.

States, for instance, typically maintain their own records on births, marriages and deaths among their residents. But it can be difficult for a local fraud investigator to determine whether a person has falsely applied for benefits under the name of a dead person in another state. So some agencies use services like LexisNexis, which integrates public data nationwide, to examine applications.

“Because of our identity information, we know more than the government entities,” says Monty Faidley, director of the government division of LexisNexis Risk Solutions. “We know where virtually every individual over 18 is.”

Other state agencies use software from IBM to help their investigators identify patterns that could indicate benefit abuse. Agencies that pay for child care services, for instance, may use analytics engines to identify implausibilities - like a mother who claims to have enrolled her children in different day care centers, even though the centers are too far apart from one another and her workplace for her to drive there round trip on a daily basis. The company also offers a program that can check for close relationships between government employees and the people to whom their agencies award contracts.

“Anytime you can correlate a person, location and time, you can identify schemes,” says Deepak Mohapatra, a senior consultant in government at IBM.

As agencies embrace commercial data-mining practices, however, they may also end up using disparate details about people in ways that citizens might not expect or trust. Civil liberties advocates say there is a real risk that erroneous information or discriminatory algorithms could unfairly keep people from getting needed benefits.

“Nobody supports benefit fraud. But lots of questions are raised when governments wade into the Wild West world of commercial data,” says Jay Stanley, senior policy analyst for the American Civil Liberties Union’s project on speech, privacy and technology. “Are the steps being taken to fight that fraud fair, accurate, and are they going to have side effects?”

At New York’s Human Resources Administration, data scientists say they take measures to ensure that computer correlations do not inadvertently lead to false accusations.

“LexisNexis will tell us if a client has registered a Mercedes or an Escalade, or if they have a condo in Dade County or Miami Beach,” says Morgan Neuwirth, an agency data scientist. But because data obtained from private vendors may be wrong or out-of-date, investigators check those details against primary sources like property deeds or state car registries. “We are careful,” Neuwirth says, “about verifying and validating.”

subscribe Just R20 for the first month. Support independent journalism by subscribing to our digital news package.
Subscribe now