There’s a “Guest Viewpoint” in today’s Eugene Register-Guard discussing Measure 42 which has a couple of points that I have to take issue with.
But the premise has a problem: the credit information itself is unreliable. How can unreliable information fairly be the predictor of anything?
A 2002 study by the Consumer Federation of America estimated that tens of millions of Americans are unfairly penalized for incorrect information in their credit reports. More recently, a 2004 study by the U.S. Public Interest Research Group found that one in four credit reports contained errors serious enough to cause consumers to be denied credit, housing, or even a job.
The answer to the question “how can unreliable information fairly be the predictor of anything” comes in two parts:
First, perfectly reliable information isn’t normally found in the real world. If we want to use information, we have to accept this lack of perfection, provide a means to offer corrective action when imperfection is discovered, and move one with our lives.
Second, information doesn’t have to be perfect to have meaning, in aggregate. Cleaner data is certainly more useful, and can serve as the foundation for better models…but messy data can still be meaningful as long as the users of that data allow for the extra noise in the design of their model.
If concerns about credit data being wrong is the real issue here, the answer is to seek legislative and regulatory remedies to encourage the credit bureaus and their subscribers to be more dilligent about accuracy, rather than in banning one use of the data.
After all, banning credit scoring won’t do anything to fix the alleged errors that potentially cause consumers to pay higher interest rates or have loans denied.
I’m not entirely sure about the accuracy of the PIRG study cited, which brings us to my other issue with the column-writer:
The narrator simply states that a person’s credit score demonstrates their likelihood of making an insurance claim. This assertion has never been proven; moreover, as long as the underlying data is suspect, it can never be proven.
The information is there and available to anyone caring enough to do a simple Google search. For example, see these studies by by Epic Consulting (report, appendices) and the University of Texas (report).