The Government Accountability Office (GAO) took the U.S. Department of Education (ED) behind the proverbial woodshed in a new report that was extremely critical of how ED estimated the cost of income-driven repayment (IDR) programs. (Senate Republicans, which asked for the report, immediately piled on.) Between fiscal years 2011 and 2017, ED estimated that IDR plans would cost $25.1 billion. The current estimated cost is up to $52.5 billion, as shown in the figure below from the GAO report.
The latest estimate from the GAO—and the number that got front-page treatment in The Wall Street Journal—is that the federal government expects to forgive $108 billion of the estimated $352 billion of loans currently enrolled in income-driven repayment plans. Much of the forgiven loan balances are currently scheduled to be taxable (a political hot topic), but some currently unknown portion will be completely forgiven through Public Service Loan Forgiveness.
The GAO report revealed some incredible concerns with how ED estimated program costs. Alexander Holt of New America has a good summary of these concerns, calling them “gross negligence.” In addition to the baffling choices not to even account for Grad PLUS loans in IDR models until 2015 (!) and to not assume borrowers’ incomes increased at the rate of inflation (!!), ED ran very few sensitivity analyses about how different reasonable assumptions would affect program costs. As a result, the estimates have not tracked tremendously closely with reality over the last several years.
But there are several reasonable steps that could be taken to improve the accuracy of cost estimates within a reasonable period of time. They are the following:
(1) Share the current methodology and take suggestions for improvement from the research community. This idea comes from Doug Webber, a higher ed finance expert and assistant professor at Temple University:
ED could then take one of two paths to improve the models. First, they could simply collect submissions of code from the education community to see what the resulting budget estimates look like. A second—and better—way would be to convene a working group similar to the technical review panels used to improve National Center for Education Statistics surveys. This group of experts could help ED develop a set of reasonable models to estimate costs.
(2) Make available institutional-level data on income-driven repayment takeup rates and debt burdens of students enrolled in IDR plans. This would require ED to produce a new dataset from the National Student Loan Data System, which is no small feat given the rickety nature of the data system. But, as the College Scorecard shows, it is possible to compile better information on student outcomes from available data sources. ED also released information on the number of borrowers in IDR plans by state last spring, so it’s certainly possible to release better data.
(3) Make a percentage of student-level loan data available to qualified researchers. This dataset already exists—and is in fact the same dataset that ED uses in making budget projections. Yet, aside from one groundbreaking paper that looked at loan defaults over time, no independent researchers have been allowed access to the data. Researchers can use other sensitive student-level datasets compiled by ED (with the penalty for bad behavior being a class E felony!), but not student loan data. I joined over 100 researchers and organizations this fall calling for ED to make these data available to qualified researchers who already use other sensitive data sources.
These potential efforts to involve the research community to improve budget estimates are particularly important during a Presidential transition period. The election of Donald Trump may lead to a great deal of turnover within career staff members at the Department of Education—the types of people who have the skills needed to produce reasonable cost estimates. I hope that the Trump Administration works to keep top analysts in the Washington swamp, while endeavoring to work with academics to help improve the accuracy of IDR cost projections.