A new higher education accountability system is now in effect, tying eligibility for federal loans to postsecondary programs’ labor market outcomes. The accountability measure will compare the earnings of each program’s graduates, produced by matching a cohort of college graduates to their earnings reported in Internal Revenue Service tax records, against a particular threshold, calculated using the U.S. Census Bureau’s American Community Survey (ACS) of a representative sample of households (and the individuals in them) across the United States. The threshold for undergraduate programs is the earnings of the typical working adult in the same state, aged 25 to 34, with only a high school diploma; for graduate programs, the median earnings in the same age range for workers with a bachelor’s degree (overall or in the same field of study).
Over the past year, the Education Department (ED) has published six different versions of these earnings thresholds across three policy initiatives. First, as part of the rulemaking on the accountability standards, the Department published a dataset in December 2025 (the Program Performance Data: 2026 [PPD:26]) to illustrate the likely impact of the thresholds. Second, it has published three versions of these thresholds (in March, June, and October 2026) as data updates to the College Scorecard. Finally, ED also produced earnings thresholds to flag lower-earning colleges (those where the median undergraduate completer earns less than a typical high school-only graduate in the state) in the online FAFSA portal, first in a December 2025 file and later updated in March 2026.
Each one of these six Department-published earnings threshold calculations uses a different methodology. Five of the six releases either use an incorrect methodology, rely on an ACS data file with an important error, or both. Some of the errors have also propagated into external analyses and data tools, including previous PEER Center analyses. They can be consequential – for example, resulting in as many as 196 programs mislabeled in the Program Performance Data as passing the new earnings test, and 74 institutions mis-flagged as “lower earnings” on the FAFSA portal. In other cases, such as the June College Scorecard data release, if users had applied the published thresholds to published program earnings in the manner prescribed by the regulations, they would have found 199 undergraduate programs at 178 institutions to be below the earnings standard that were actually above the threshold as properly calculated.
In the interest of ensuring that future earnings thresholds are produced correctly, and giving accurate information to the field about the likely impacts of the new accountability regime, this short brief provides an overview of how the earnings thresholds should be constructed, while highlighting key data sources and methodological choices that vary across different threshold releases from the Department. We also illustrate how ED’s different earnings threshold releases either under- or over-estimate the true thresholds. Finally, we make available the correct earnings thresholds, along with the programs that produce these thresholds, to allow analysts to more accurately model the impact of the new accountability system and to replicate or further examine our work.
Estimating Earnings Thresholds Using the American Community Survey
So what has caused ED’s estimates to change across releases? There are five data source- and methodology-related factors that account for the differences across the six earnings threshold estimates released by ED. We summarize each of these factors below, and Figure 1 shows the specific data and methodology used in each of the six releases, noting where it differs with the preferred methodology for calculating thresholds.
(1) Who Is Included As Working?
An especially consequential error ED makes in calculating earnings thresholds lies in an improper definition of which ACS sample members are working. Under the One Big Beautiful Bill Act and associated regulations, the threshold earnings should be calculated for working adults aged 25-34 who are not enrolled in college “during the year of the associated measured earnings.” This is meant to mirror the construction of the median earnings of program graduates using IRS data, which is based on the subset of non-enrolled graduates reporting positive earnings during a given calendar year.
However, in some releases (December 2025’s low-earnings flags; PPD:26, which was published during the accountability rulemaking; and the June 2026 College Scorecard update), the Department defines working adults as those with positive earnings in the previous 12 months and who are also employed in the week prior to responding to the survey. The additional requirement of being employed in the week prior to the interview results in the earnings thresholds being estimated based on a group of workers who are more attached to the labor market (since it excludes those who had earnings earlier in the year, but who were not employed in the previous week). That leads to a notably higher earnings threshold, and a greater number of failing programs. Oddly, this error was present in both December 2025 releases (the FAFSA indicator and the PPD:26 data), and corrected in a March release of College Scorecard data – but was then reintroduced in the June College Scorecard release, and corrected again in an October College Scorecard release.
(2) Which Types of Earnings Are Counted?
Another choice that explains the inconsistent ED earnings threshold estimates is the type of income from the ACS that gets counted toward the totals that ED uses. In all versions of its earnings threshold releases except one, ED defines earnings as the combined amount of wages, salaries, and self-employment income. That matches the Department’s calculations of program or institutional earnings, which include self-employment income alongside wages and salaries. However, based on the Department’s documentation and our replication analyses, in one release – its December 2025 release of earnings thresholds that defined which institutions were flagged by the “lower earnings indicator” for FAFSA filers – ED used only earnings from wages and salaries, excluding self-employment income. This has the effect of lowering the earnings threshold (and reducing the number of programs/institutions that are below the threshold), since self-employment income is incorrectly omitted when computing the earnings of high school graduates in the sample.
(3) How Are Earnings Inflation-Adjusted?
Each annual American Community Survey represents approximately one percent of the U.S. population. To better ensure that earnings thresholds are based on an adequate number of observations, ED plans to use pooled five-year ACS files. (Using five pooled years of data is especially necessary for producing earnings thresholds for graduate programs, since the ACS sample sizes can be much smaller when producing data both by state and field of study.) The five-year file should generally align the final ACS year with the final calendar year of the measurement period for the median earnings of program graduates. For example, the most recent published program-level earnings data include those who graduated in 2017-18 and 2018-19, with their earnings measured in the 2022 and 2023 calendar years, respectively; those data are aligned with the 2023 ACS five-year file, which pools ACS respondents in years 2019 to 2023.
As when working with other longitudinal data, earnings captured in each year of the survey should be inflation-adjusted to be expressed in common dollars (what we’ll call a survey-year inflation adjustment). For instance, earnings in each year of the 2019-2023 five-year ACS file should all be converted to a common real dollar measure, such as 2024 dollars, before estimating the median earnings of a particular group of degree holders. All of the Department’s threshold releases include this survey-year adjustment, though across releases it sometimes converts dollar values to a different year’s constant dollars.
A second type of inflation-adjustment warranted by the unique data collection in the ACS, however, is absent from all of the Department’s data releases except its most recent October 2026 College Scorecard release. To collect data from the large number of households surveyed in the ACS in each survey year, the Census Bureau collects data from respondents throughout the year, with roughly equal numbers of households sampled in every month. In whichever month a respondent is surveyed, Census asks about their income in the preceding 12 months (their “reference period”). So someone interviewed in the beginning of January 2019 will report their income between January and December of 2018; whereas someone interviewed in late December 2019 reports their income between January and December of 2019. In other words, in the limit, sample members in the same survey year can report their incomes for reference periods separated by a full 12 months.
Many analysts ignore within-survey-year inflation adjustments, in part because the Census does not publish the month in which individuals are interviewed in the ACS (making accurate individual-level adjustments impossible), and in part because in low-inflation times the adjustment may not be very quantitatively important. When inflation is rising quickly, however, as it was across the period covered by the 2019-2023 five-year ACS, it can prove consequential to account for the change in the value of the dollar between when someone earned the money and when they are asked about it. Skipping this more granular reference-period inflation-adjustment depresses the earnings thresholds – for example, by treating an individual interviewed in early January of 2019 as if their salary was captured in 2019 dollars, when in fact that individual is reporting their salary in 2018 dollars (since their reference period is January to December of 2018). This has the effect, all else equal, of reducing the number of programs or institutions projected to fall below the standard. Census publishes an aggregate adjustment factor that allows users to adjust earnings so that aggregate statistics—like the median earnings of a particular group reported in each ACS survey—can all be expressed in the constant dollars of the survey year. The best practice for estimating the earnings thresholds used in the OBBBA accountability scheme is to use this reference-year inflation-adjustment in conjunction with the survey-year adjustment.
(4) Which Data Source Is Used?
ACS data can be obtained from two different sources. Public-use extracts of the ACS data called the Public Use Microdata Sample (PUMS) files can be downloaded directly from the Census, or from IPUMS, a university-based data distribution resource that researchers have long relied upon because it harmonizes data definitions across years. For estimating the earnings thresholds used in the new accountability system, the relevant information in the two sources of data should be identical and yield equivalent results.
Unfortunately, the first IPUMS release of the 2023 five-year ACS file had an error related to the survey-year inflation-adjustment of individuals’ earnings, in which the inflation factor applied to each year’s data was off by one year. In the five-year ACS files, IPUMS usually creates earnings variables where respondents’ earnings have already been adjusted to constant dollars of the final year included. For example, the earnings for those responding in 2020 are inflated by the change in the average price level between 2020 and 2023 – which in that case was an 18 percent increase. In the erroneous files, however, respondents’ earnings for 2020 were adjusted by the change from 2021 to 2023 (12.6 percent), respondents in 2021 were adjusted by the change from 2022 to 2023, and so on – so that respondents’ real earnings were too low for those surveyed prior to 2023. Across the Department’s multiple versions of the threshold, we have found that earnings threshold releases using data released by IPUMS before March 18, 2026 for the 2023 five-year ACS file all included this error. Releases calculated from the raw (PUMS) Census file, or from IPUMS after March 18, 2026, did not. That means the Department’s PPD:26 data, its March 2026 FAFSA lower-earnings indicator data, and its March 2026 College Scorecard release all relied on the erroneous IPUMS data, and the earnings thresholds are incorrect as a result.
Using IPUMS might also have contributed to why the Department skipped the reference-period inflation-adjustment. In principle, data users can do both the survey-year and reference-period inflation adjustments in either PUMS or IPUMS. In the PUMS Census files, though, that two-part inflation-adjustment process is the default adjustment; in the IPUMS, only the survey-year adjustment is the default. That helps to explain why the October College Scorecard release, for which the Department appears to have used the PUMS data, included both inflation-adjustment types. This was also a source of error in a recent PEER report; in calculating the thresholds, we used a January version of the IPUMS file, rather than the updated post-March data, with the survey-year adjustment only; that issue has since been corrected in our report and data tool.
(5) Which ACS Survey Years Are Used?
The final difference across earnings threshold data releases we identified is most straightforward: which years of the ACS are used. As noted earlier, the ACS five-year survey should align the last year of the sample to the final year in which earnings are measured (e.g., program earnings measured in 2022 and 2023 should be measured against thresholds calculated from the 2019-2023 ACS). That is the case in all of the ED earnings threshold releases where it has compared program or institution earnings outcomes to the thresholds. For example, the PPD:26 data and various Scorecard releases compare program earnings measured in 2022 and 2023 to thresholds based on the 2019 to 2023 ACS. And, in the December 2025 FAFSA lower-earnings indicator release, the Department didn’t yet have 2023 earnings for college graduates, so it appears to have used the 2015-2019 and the 2016-2020 ACS five-year files, and averaged the medians across the two files (which were matched up against the median earnings of 2015 and 2016 graduates of institutions, measured in 2019 and 2020), and adjusted all dollar values to June 2025 dollars.
(That latter choice of averaging the thresholds computed from two consecutive five-year ACS surveys [rather than a single one-year file], while not necessarily incorrect, is unnecessary at best, and represents another unexplained departure from the Department’s methodology used for all other data releases. The five-year 2020 file is constructed by pooling the 2016-2020 ACS surveys, while the 2019 five-year file is based on 2015-2019 ACS surveys. Averaging the results of the medians calculated from the two five-year files effectively uses 2015 and 2020 data once, while double-counting responses from 2016 to 2019. Relative to that awkward construct, the 2020 file seems superior in that it gives equal weight to the 2019 and 2020 years, and avoids giving any weight to the 2015 survey year, which is far away from the years in which programs’ earnings are measured [i.e., 2019 and 2020].)
In the most recent (October 2026) College Scorecard update, the Department released earnings thresholds under the most recently available ACS data, the 2020-2024 five-year file. Users should be careful not to apply these thresholds to the median earnings data that are currently available; since median program or institutional earnings measured in 2024 have yet to be released. If using the already-published program or institutional median earnings measured in 2022 and 2023, users should continue to use the 2019-23 ACS-derived earnings thresholds.
Table 1: Differences in Methodology and Data Used in Department of Education Earnings Threshold Estimates
How Much Does It Matter?
As explained above, some methodological errors tend to raise the estimated earnings thresholds while others tend to lower them: Conditioning the measurement of earnings among “working” individuals in the ACS on employment at the time of survey drives up the earnings thresholds, while excluding self-employment income drives the thresholds down, for example. Several of the Department’s releases combine the decisions in ways that produce a net result somewhere in the middle.
To see how each of these mix together across the released thresholds, we built an interactive data tool. Users can click to see how each of the Department’s published thresholds compares to the ones PEER posted today across states. Users can also download our data file to see the original threshold in the Department’s PPD:26 data, as well as the updated threshold, for every program and institution. To quickly look up your state’s thresholds, check out the look-up tool at the bottom of this page.
Recommendations
For earnings-based accountability, the Department has time to work out the kinks – and with its latest, October 2026 College Scorecard release, it seems to have done so. In the interim, though, the field has looked to data released by the Department for purposes such as internal benchmarking and news coverage. Negotiators used impacted data in developing the regulations around earnings-based accountability. And students were sometimes shown inaccurate low-earnings flags on the FAFSA.
In fact, we’re not aware of any other source for institutions, students, or policymakers to pull reliable, vetted thresholds that don’t carry these design problems and errors and that are appropriate to judge program or institutional earnings outcomes against using data currently available. You can check out our report here, and an interactive data tool we built here. For researchers working with program earnings data, check out our downloadable data file to find the applicable earnings threshold for each, and our programs to produce the thresholds that will allow analysts to kick the tires themselves.
The details matter – and the Department and its contractors should ensure they use the correct methodology going forward, across all releases. As new earnings-based accountability rules take effect, the PEER Center can provide technical assistance to federal and state policymakers, researchers, and institutions, and help add transparency into the process of calculating those measures.
Look Up Your State’s Earnings Thresholds
