The Permanent Record / Dossier · Credit Files & Scoring
Two People, One File
You assume your credit file is about you. It is about whoever the bureau's matching software decided you were, one incoming record at a time.
A credit bureau does not check identity when a new account arrives. It looks at the name, Social Security number, birth date and address attached to the record and files it with the consumer who fits best, even when the fit is not perfect. Sometimes the best fit is a stranger. In 2013 a jury in Portland heard that Equifax had merged an Oregon woman's file with another woman's who had the same name and a similar Social Security number, and that it took two years and a lawsuit to pull them apart.
Join the discussion →
The Fair Credit Reporting Act asks one thing of a credit bureau before anything else. Whenever it prepares a consumer report, section 1681e(b) says, it shall follow reasonable procedures to assure maximum possible accuracy of the information concerning the individual about whom the report relates.
Read the last clause again. The statute does not only require the information to be accurate. It requires it to be about the right person. That second requirement turns out to be the harder one, because a credit bureau never actually meets the people whose files it keeps. It meets records, and every record arrives carrying a description of someone.
A File Is a Guess That Kept Being Right
Each month, lenders and collectors send the nationwide bureaus updates on millions of accounts. Each update carries what the industry calls header information: a name, an address, a Social Security number, sometimes a date of birth. The bureau's job is to decide which consumer that header describes, and to post the account there.
The Consumer Financial Protection Bureau described the process in a December 2012 white paper on how the system works. At least one bureau organizes its data like a traditional filing cabinet, one file per consumer, with files told apart by matching logic using name, address, SSN and date of birth. Another assigns each consumer a personal identification number and links trade lines, inquiries, addresses and public records to it in a relational database, assembling the report in real time when someone asks for it. In both designs, the CFPB wrote, an algorithm assigns an incoming record to the file or PIN that represents the best match, and in some cases it does so even when all of the identifiers do not match up perfectly, or when only a limited number of identifiers came with the record.
The question the system asks is not "is this the same person?" It is "which file is this most like?" Those are different questions, and the gap between them has a name.
The CFPB also admitted what it did not know. It had no data, it said, on the relative accuracy of the flat-file and the PIN-based designs. The rules that decide whose account is whose are proprietary to each bureau.
Why Names Are a Weak Key
Matching would be easy if every record carried a clean, unique identifier. Few do. The same CFPB paper laid out why. By 2000 census figures there were more than 2.3 million Americans named Smith, 1.8 million named Johnson, a million named Davis, 850,000 named Garcia, and 600,000 named Lee. Relatives living at the same address can share a first and last name and differ only by a middle name that neither uses on applications. People change surnames at marriage and divorce. A woman named Elizabeth may apply as Elizabeth on one form and Betty, Beth or Liz on the next. And lenders differ in which identifiers they collect at all, so the header that describes you to one creditor may not look much like the header another creditor sends.
That leaves the Social Security number doing a great deal of work, and it is a number typed by people. A transposed digit, a misread form, a collector's file built from a skip-trace: each yields a record whose header looks nearly like yours and nearly like someone else's. A best-fit algorithm has to decide anyway. If it leans toward attaching, it risks a merge. If it leans toward keeping records apart, it risks the opposite failure, which the CFPB also listed: fragments of one person's history left stranded in a separate file until some unifying detail, a shared address or phone number, convinces the system they belong together.
What a Mixed File Looks Like
The CFPB's list of the kinds of inaccuracy that appear in credit files begins with this one. Its words: inclusion of accounts or records in a credit file that do not belong to the consumer, commonly called a mixed file. The second item on the list is the mirror image, a record belonging to you that was placed in someone else's file instead. The paper notes that the cause need not be the bureau's algorithm alone. A data error upstream, a wrong digit from a furnisher, can make a header match the wrong person, and the algorithm then does exactly what it was built to do.
From the consumer's side, a mixed file is hard to see coming. Accounts you never opened appear. So do addresses in states you have never lived in, and inquiries from lenders you never applied to. If the other person has a worse history than yours, your score falls with nothing you did to explain it. And because the merged file is the one the bureau sells, a lender who pulls your report may also be reading someone else's identity, account numbers included.
Julie Miller's Two Years
In October 2011 a woman named Julie Miller sued Equifax Information Services in federal court in Portland, Oregon. The case went to a jury in July 2013, and the court's later opinion summarized what was not in dispute at trial: Equifax had merged Miller's credit file with the file of a different person who had the same name and a similar Social Security number, who lived in a different state, and who had a negative credit record significantly different from hers.
Miller learned of the merger, reported it to Equifax, and began what the court called a two-year saga to fix it. She disputed repeatedly. Equifax reported back to her, repeatedly, that it had investigated. The file was not corrected. According to the opinion, Equifax did not take steps to correct the information until Miller filed her lawsuit, and at trial it argued to the jury that taking corrective steps only after a civil action is filed complied with its policy.
The line from the trial that people still quote came from Equifax's own witness. Margaret Leslie, a vice president in the company's technology area, testified, as the opinion records it, that this kind of file merger was a "reasonable combination" that occurs with "regularity."
The bureau's own witness called the merger a reasonable combination that happens with regularity. That is a description of a design, not an accident.
On July 26, 2013 the jury returned its verdict. It found Equifax had violated the FCRA and had done so willfully, the finding that opens the door to punitive damages. It awarded Miller $180,000 in compensatory damages for the distress of those two years, and $18,400,000 in punitive damages.
The Remittitur
Equifax moved to reduce the punitive award, and the judge agreed that it had to come down. Punitive damages are bounded by due process, and the usual measure is the ratio between punitive and compensatory damages. Eighteen point four million against $180,000 is a ratio of about 102 to 1. In January 2014 Judge Anna J. Brown reduced the punitive award to $1,620,000, a 9-to-1 ratio, the top of the single-digit range the courts treat as presumptively permissible. An amended opinion in May 2014 corrected a typographical error in the original order's statement of the jury's figure. The amended judgment entered that June: $180,000 compensatory, $1.62 million punitive, plus a separate award of about $302,000 in attorneys' fees.
Both sides appealed. In June 2015 the Ninth Circuit's mandate closed the case on the parties' joint motion to dismiss the appeals. Whatever the final number was, it was settled privately.
The arithmetic matters less than what the jury heard. Miller's lawyers did not have to prove the merger; it was undisputed. The contest was over what Equifax did about it once told, and the jury's willfulness finding was an answer to that question.
Why a Fix Does Not Always Hold
Separating a mixed file is not like correcting a typo. The bureau can delete the stranger's accounts from your file, but the conditions that caused the merge are still there. The stranger's lenders are still reporting every month, with the same header that matched you the first time. Unless the bureau changes something about how those records are matched, or suppresses them against your file specifically, the next monthly update can attach to you again.
The statute anticipates part of this. Under section 1681i(a)(5)(B), information deleted after a dispute may not be reinserted unless the furnisher certifies that it is complete and accurate, and a bureau that reinserts must notify the consumer. But that provision speaks to a deleted item being put back. It was written for a dispute about one account, not for an algorithm that keeps deciding, every month, that two people are one.
And Then You're Dead
Your credit file is not a record of you. It is the pile of records a matching program decided were probably you, and the program's standard is best fit, not proof. Most of the time the guess is right and nobody notices it was a guess. When it is wrong, the burden of noticing falls on the one person who never sees the program: you, reading a report full of addresses you never lived at. The statute says the bureau must assure maximum possible accuracy about the individual the report relates to. In Julie Miller's case the bureau's own witness told a jury that merging her with a stranger was reasonable and regular, and it stayed merged until she sued.
That's it. That's the whole thing.
Sources
- The duty to follow reasonable procedures to assure maximum possible accuracy of the information concerning the individual about whom the report relates — 15 U.S.C. § 1681e(b); the limits on reinserting deleted information and the notice requirement — 15 U.S.C. § 1681i(a)(5)(B) (Legal Information Institute, Cornell Law School)
- The flat-file and PIN-based file architectures; matching on name, address, SSN and date of birth; best-match assignment even when identifiers do not all match or few are present; fragmented files; the absence of CFPB data on the relative accuracy of the two designs; the 2000 census surname counts and the name-variation examples; and the definition of a mixed file and its mirror-image omission error — Consumer Financial Protection Bureau, "Key Dimensions and Processes in the U.S. Credit Reporting System: A review of how the nation's largest credit bureaus manage consumer data," December 2012, sections 5 and 6
- The docket of Miller v. Equifax Information Services LLC, No. 3:11-cv-01231-BR (D. Or.): filing on October 12, 2011; verdict returned July 26, 2013; Equifax's motion for reduction of punitive damages; the January 29, 2014 opinion and order; the June 13, 2014 amended judgment; the attorneys' fee award; and the Ninth Circuit mandate of June 23, 2015 dismissing the appeals on joint motion — CourtListener (RECAP archive)
- The jury's award of $180,000 in compensatory and $18,400,000 in punitive damages — Plaintiff's memorandum in opposition to the motion for reduction of punitive damages, Dkt. 93, Oct. 11, 2013; the verdict form, Dkt. 79, July 26, 2013
- The undisputed merger with a same-name, similar-SSN consumer in another state; the "reasonable combination" that occurs with "regularity" testimony; the two-year effort and correction only after suit; and the reduction from $18,400,000 to $1,620,000 at a 9-to-1 ratio — Miller v. Equifax Information Services LLC, Amended Opinion and Order, Dkt. 124, May 20, 2014
- Photo: "IBM card storage.NARA", storage of IBM record cards at the Federal Records Center, Alexandria, Virginia, November 1959, U.S. National Archives and Records Administration, public domain, via Wikimedia Commons
Comments
Loading…