The Fundamental Flaw of Earnings Premium

Insights from the Gainful Employment lawsuit on the fundamental problems with Earnings Premium that is embedded in the Senate's institutional accountability

Was this forwarded to you by a friend? Sign up, and get your own copy of the news that matters sent to your inbox every week. Sign up for the On EdTech newsletter. Interested in additional analysis? Try with our 30-day free trial and Upgrade to the On EdTech+ newsletter.

By this past weekend, it was obvious that for institutional accountability, the US Congress had abandoned the House Risk-Sharing approach and was going for broke with the Senate’s Gainful Employment for All approach, which they are now calling the Do No Harm approach. As of today, this bill passed the Senate and is going back to the House for final vote before potentially being signed into law by July 4th.

The only changes to the metrics behind the Senate’s Do No Harm approach since last week’s negotiations were to:

  • Remove non-completers from the earnings cohort; and

  • Change the earnings timeframe to measure graduate programs to 4 years after completion.

In this post I’d like to highlight the major opportunity and major flaw of the Senate bill.

Disclosure: I have provided a declaration in the lawsuit against the FVT & GE regulations, analyzing the data provided. I am also providing visualizations and data analysis of the new proposals to clients.

Major Opportunity

The key strength of the Senate Do No Harm accountability approach is that it is realistic to implement. In effect, it is the Earnings Premium half of the Gainful Employment regulations that were finalized in 2023, but with two key differences.

  • This Earnings Premium applies to all degree types except for Undergraduate Certificates, but the Senate keeps Gainful Employment on the books, and those regulations include these certificate types. Hence the more accurate moniker Gainful Employment for All.

  • This Earnings Premium approach fixes some of the issues from Gainful Employment that schools have complained about, such as comparing to all students without the degree in question rather than just for working adults. And it sets more realistic time frames.

Major Flaw

The major flaw is that the Senate Do No Harm accountability measure actually builds on a flawed metric from Gainful Employment while keeping both versions in place. We’re more than doubling down on Earnings Premium.

The concept in question for this accountability is that federal financial aid should not go to programs that leave graduates no better off financially than if they hadn’t paid for and completed that degree or certificate. That’s not a hard concept to get behind.

The problem is one of aggregation that leads to the moniker being just plain wrong. This accountability metric Does Unintended Harm™. To explain further, I will draw upon my declaration in the lawsuit challenging the Gainful Employment regulations.

And note that I am not arguing the legality of these metrics - for that, you can read the Gainful Employment lawsuit. Rather, I am making an argument about effective policy to meet stated goals.

Individual vs. Mythical Median

An ideal metric to back up this concept would be to measure an individual’s earnings with and without the credential in question. What would Julia from McAllen, Texas be earning with a high school degree vs. with an undergraduate certificate in Diagnostic Imaging? The data on individuals are not available, which leads to the need to aggregate data from groups of people. To keep this simple, just look at undergraduate certificates and degrees.

What both Gainful Employment (mostly for undergraduate certificates) and Senate Do No Harm (for undergraduate degrees) do for the comparison group (i.e., the Earnings Threshold) is to use Census Bureau data for adults aged 25 - 34 people with a high school but no college degree, aggregated at the state level without any reference to demographics (gender, race, sub-state geographic location, etc). Further, this is a weighted sample of data.

The earnings metric takes all of a program cohort’s individual’s earnings measured per person and aggregated to a program cohort. Demographics are embedded into that cohort. The only aggregation is measuring the median earnings of that cohort.

The Earnings Premium simply compares the median earnings with the Earnings Threshold to see if it is above $0. What you now have is a college or university being responsible for a comparison of a group of specific individuals to a mythical median aggregation that ignores all demographics beyond age and state.

GE Analysis - Gender

Now I’ll quote from my declaration in the GE lawsuit to show impact of this aggregation flaw [emphasis added].

36. A simple review of the data shows significant impact of student demographics on the EP metric. To pick one example, consider gender. The Earnings Threshold (“ET”) is based on state-aggregated data from the [Census Bureau American Community Survey] ACS, based on median earnings for residents with a high school degree but no college, between 25-34 years of age. A simple inclusion of gender within that same data source that the Department used shows the wide variation in earnings thresholds if gender is included. The chart on the next page shows the results for female-only (red) and male-only (blue) median earnings, using a weighted median analysis of the ACS data.

38. A basic view of ACS data in the chart above shows that gender does have a significant impact on EP. For example, in Texas, the Department’s chosen Earnings Threshold is $25,899; yet ACS data show women with a high school education in Texas made $20,012 and men with a high school education made $29,932. Averaged across all states, the ACS five-year data survey shows that women make $20,020, and men make $29,911, fully 49% more than women.

Further in the declaration, I added a chart showing a simple view of how the gender problem shows up in which programs would fail the Earnings Premium metric.

40. [snip] By looking at a simple plot of the completion share of females in undergraduate certificate programs versus the calculated EP for each program using the Department’s definitions and 2022 PPD data file, the data are not random in distribution (as they would have to be if the Department’s regression analysis had produced an accurate result). In fact, there are strong clusters on the left of the chart (programs predominantly with male completers) and on the right (programs predominantly with female completers). The chart also shows the predominantly male programs mostly passing EP (above the red line) while predominantly female programs failing EP (below the red line). The distribution is not random but instead clustered by gender, indicating that the 16% female disadvantage in fact reflects reality.

GE Analysis - Geographic Location

Now let’s look at the impacts from the state-level aggregation for Earnings Threshold, again drawing from my declaration [emphasis added].

47. To better understand the specifics of this issue, consider Paul Mitchell the School Louisville (“Paul Mitchell”), with its cosmetology program showing median three-year non-enrollment earnings of $22,391. The median [Census Bureau Integrated Public Use Microdata Series] IPUMS high school equivalent earnings for that area (metropolitan area: city center) is $20,221. Based on the Department’s approach, Paul Mitchell’s program fails the EP metric since Kentucky’s overall Earnings Threshold is $24,397. But if the Department used within-state MSAs, Paul Mitchell’s program would have an Earnings Premium of $2,270 (an 11% increase over the ET).

48. Consider one more example: the Douglas J. Aveda Institute in East Lansing, MI (“Aveda”), with its cosmetology program showing median three-year non-enrollment earnings of $19,704. The median IPUMS high school equivalent earnings for that area (metropolitan area: city center) is $17,000. Based on the Department’s approach, Aveda’s program fails the EP metric since Michigan’s overall Earnings Threshold is $24,438. But if the Department used within-state MSAs, Aveda’s program would have an Earnings Premium of $2,704 (a 16% increase over the ET).

In Sum

Let’s go back to the hypothetical Julia in McAllen. She is disadvantaged by being in a program that is predominantly female and in a poor part of Texas. The Earnings Threshold is an aggregation across all of Texas (including Dallas, Austin, and many other higher-income areas) and combining male and female. Her program will likely fail, and Julia will have no access to federal financial aid. All due to a flawed metric.

With the Senate accountability section of the bill, all undergraduate degree programs are about to face this same problem.

Given the current anti-DEI mood in the country, I should note that I am not arguing that different demographic groups must have similar outcomes. I am arguing, however, that setting policy based on a fundamentally flawed metric will result in a lot of unintended harm.

Well, Senate and Senate staffers. If one goal was to show nonprofit institutions how difficult it is to manage poorly-designed metrics, and if another goal was to give me more data analysis to work on to find the unintended harms, then job well done. I fully acknowledge that there will be some positive benefits by discouraging majors that should not be enrolling as many students given the current and future job market, but if your goal was to create effective policy to discourage programs that don’t benefit graduates while doing no harm, then we have a problem.

The main On EdTech newsletter is free to share in part or in whole. All we ask is attribution.

Thanks for being a subscriber.