Background Checks and Credit Scores Don’t Work. Why Are They Still Society’s Gold Standard for Trust?
One year ago, an Uber driver in Kalamazoo went on a deadly shooting spree because, as he later explained to police, his Uber app “literally took over his mind and body.” As the tragic details trickled out into the press, a natural question emerged: how in the world did this guy become an Uber driver? The entire point of background checks is to keep serial killers from driving us to work (among other things), so why didn’t Uber’s mandatory background check flag this murderer?
A less gory event in Austin dealt with similar issues. There, Uber and Lyft packed their bags and moved out of the city altogether after a fight with city officials about whether or not their drivers should have to undergo a more stringent, third-party background check.
In both of these cases, people wanted more from their background checks: more safety, more security, more comprehensiveness. But not once in either of these situations was a very basic question asked: are background checks really that effective?
A flawed trust mechanism
Background checks are currently the gold standard for trust in our society, even though many of their flaws are widely known. They can be frighteningly inaccurate or incomplete. A 2015 report by the Government Accountability Office found that 25% of the records in 20 states failed to note whether arrested people had ever been convicted or not — a pretty key piece of information when you’re trying to determine someone’s criminal record. Though the FBI created a task force back in 2009 to fix this, said task force doesn’t currently have any plans to improve the situation. Incomplete records like these don’t just lead to criminals slipping through loopholes, by the way; they lead to qualified candidates being unfairly rejected from jobs when their background checks give the misleading impression that they’re ex-convicts, instead of former arrestees who were exonerated.
And let’s be honest: many of those qualified candidates are minorities. That’s because the systematic biases in our criminal justice system obviously show up in background checks. The Department of Justice recently released a report decrying the lack of diversity in law enforcement and citing background checks as part of the problem, since they have “been shown to have an unwarranted disproportionate impact on underrepresented populations.”
Why the big picture is so difficult to see
A youthful bout with marijuana shouldn’t disqualify a responsible adult from gainful employment, but unfortunately, that’s exactly what happens when we blindly accept the results of background checks instead of taking a “holistic” view, as the Department of Justice report recommends.
Of course, some types of drug usage are a big deal, but background checks don’t even do a great job revealing those instances. See, just because an individual has not been arrested for a crime does not mean much about their propensity for that crime. According to the NAACP, for example, about 16 million people in the US report illicit drug use, but the number of people arrested for drug violations has hovered around 1.5 million for the past few years. So if you’re relying on background checks (or any sort of arrest record) to find out if someone’s used drugs, you’re only getting a very small part of the picture.
Why is it so hard for people to take a “holistic” view? Because the way background checks are used encourages superficial interpretation of a simple pass/fail structure, rather than richer contextual understanding of why the alleged crime occurred and what reforming actions the subject of the background check may have taken since then. And going deeper to get this contextual understanding is difficult — because, quite frankly, background checks are simply too confusing. The information presented in a background check is nearly impossible for most typical users of the tool to interpret — the California Penal Code for 2016 alone contains over 3,000 pages and 30,000 sections. And each state has its own version of a criminal code, stuffed to the gills with idiosyncratic definitions and subtleties. Unless someone wants to pour millions into training employers to properly understand the context behind a background check report and to deeply interpret background check results, the problems with the tool will persist.
Credit scores: not much better
Think you can just pair a credit score with a background check to get a really comprehensive picture of an individual? Think again. Our revered credit scoring system has equally profound weaknesses. For example, one out of every ten American citizens has no credit score at all, more than two out of ten have “thin credit files” which make their score less reliable, and the scores’ unintended correlation with race and age has been widely studied (we know that African Americans on average have less than half the credit scores of white Americans, for instance). Paradoxically, many users of credit reports have no idea what the target class for a credit scoring model actually is! We have anecdotally asked multiple executives in the financial services industry to tell us precisely what a credit score predicts, and not one of them has been able to do it. Still, our society continues to use credit scores for a plethora of very important situations — including renting an apartment, getting a job, and taking out a decent loan. Credit scores were never developed or tested for these situations, but thanks to lots and lots of marketing by the “big three” credit unions, they’ve become our gold standard.
In spite of these obvious flaws in our legacy trust mechanisms, society continues to cling even tighter to them when faced with new use cases such as ride-sharing, home-sharing and peer-to-peer lending. These new business models can seem like the new Wild West — sprawling, confusing, unregulated — so it makes sense that people adhere to what they perceive as the best ways to interpret whether or not an individual is trustworthy. However, the time has come to question whether we should continue to use these mechanisms for life-altering decisions such as hiring, housing, lending, and insurance. Instead of wondering whether we need stricter background checks, we should ask different questions: What does failing a background check or getting a particular credit score really predict or tell us about a person? And shouldn’t we know the answer before we continue to anoint these tools the gold standard for trust?
We already know that our old-school trust mechanisms are discriminatory and allow prejudice to advance. We know they are often incomplete and inaccurate. Surely we can do better than this. In spite of the concerns about hidden biases in big data models, we are convinced that rigorous, cutting-edge computer science in the hands of ethical people can be part of the solution. It’s time to put our modern tools to use — like large scale data and cutting-edge machine learning — to create trust mechanisms that are accurate, ethical, and inclusive, instead of ones that perpetuate inequality, misinformation, and, ironically, distrust.
Originally published at /pulse/background-checks-credit-scores-dont-work-why-still-gold-savi-baveja?trk=pulse_spock-articles on January 20, 2017.
