As mentioned, this study uses data from the Federal Reserve Bank of New York Consumer Credit Panel/Equifax (CCP) data set. The CCP is an anonymous, nationally representative random 5 percent sample of the U.S. population with a Social Security number and a credit history. The CCP is an unbalanced panel, which means that the randomly selected panelists are added to the data set once they meet the entrance criteria and are dropped in the event that they die or no longer have sufficient information in their credit file. Entry into the CCP is limited to individuals that Equifax knows to have at least one of the following: a public record (e.g., a judgment) within the past seven years; a bankruptcy filing within the past 10 years; an open credit account; or a closed account that is still being reported. Note that a closed account can be reported for up to seven years if it did not close in good standing.a
Based on these selection criteria, it is clear that the CCP does not include all adults: As noted previously, around 8 percent of households do not have a member with a credit report and therefore cannot be included in the data set.b Furthermore, there is an apparent delay in reporting some loans for young borrowers (ages 18 to 23) in the CCP, and this analysis suggests that the proportion of young borrowers omitted from the CCP has increased since 2011. This means that, although the loans do eventually make it into the data set, a small proportion of loans are omitted each quarter. Because recent quarters appear to be disproportionately affected, estimates of aggregate student loan debt since 2011 may be somewhat conservative relative to prior estimates.
The raw data have information on each individual loan a borrower holds. However, since many borrowers have more than one loan, for this analysis loan records are aggregated to the level of the borrower. This analysis excludes deceased borrowers and those who appear in the data set for no more than one year, unless they are present in the most recent quarter. Borrowers with a nonresidential address (e.g., a post office box) and those for which relative neighborhood income is unknown are excluded from income category estimates but included in total estimates.
In addition to the restrictions based on borrower characteristics, the analysis also excludes loans with Equal Credit Opportunity Act (ECOA) codes of C (comaker), S (shared, but unknown type), T (terminated), and U (undesignated), and loans that are being paid under a wage earner plan. For comaker loans, the panelist is responsible for the loan only in the event that the maker of the loan defaults, and so the estimates include the makers (ECOA code M) but not the comakers. The shared and undesignated codes indicate that the credit bureau can identify the loan as a student loan but does not have sufficient information to categorize it further, which means the loan may be of a type that should be excluded. Terminated loans may still be existing accounts, but they are no longer associated with the panelist and should not be treated as such.
Loans with more than one borrower, referred to as cosigned loans in this report, appear on the credit report of each party to the loan. In order to avoid double counting those loans when calculating aggregate student loan debt, the value of all loans with an ECOA code of J (joint account) are halved, but for median calculations, the full value of the loan is retained. Loans with an ECOA code of M (maker) are joint loans, but the cosigning party, or comaker, becomes responsible for repayment only in the event that the primary borrower cannot make the required payments. In this study, maker loans are included, but comaker loans (ECOA code of C) are excluded. Because of this exclusion, the maker and comaker loans do not double count the same loan, so maker loans are not halved in aggregate balance calculations. Finally, the analysis excludes the few loans with a value in excess of $1,000,000.
It is worth emphasizing that this analysis does not exclude borrowers based on age or enrollment status. Ideally, the analysis of student loan debt by neighborhood income category would exclude current students because the neighborhood income of a current student’s credit bureau address may not be a good proxy for that student’s socioeconomic status, financial resources, or future prospects for debt repayment. While it is, unfortunately, not possible to identify current students in this data set with any degree of certainty, I did develop alternative neighborhood income estimates that excluded borrowers who did not appear to have begun repaying their loans — a proxy for current students and recent graduates. Other than slightly higher median balances for all of the neighborhood income categories, the alternative estimates were not qualitatively different from those presented in this paper.