In a previous piece, I argued that ROI is the wrong lens for evaluating community college and workforce programs. The single-number obsession obscures two things that actually matter: how earnings grow over time (career ladders) and how much variation exists within a program (earnings variance).
The theory is straightforward. The data is harder. So let's look at it.
I pulled the most recent Post-Secondary Employment Outcomes (PSEO) release from the Census Bureau—2025Q4—and filtered to every short-term certificate program (less than one year) offered across Colorado's community colleges. That gives us 22 institutions and 633 program-institution combinations. PSEO provides something almost no other dataset can: earnings at the 25th, 50th, and 75th percentile at one, five, and ten years after graduation, with graduates tracked across state lines.
Here is what the data shows.
The ladder matters more than the starting rung
Take Front Range Community College, the largest community college system in the state.
Their Computer Systems Networking & Telecom certificate graduates earn a median of $43,931 in Year 1. Respectable, but not headline-grabbing. By Year 5, that climbs to $61,384. By Year 10: $62,260.
That is a career ladder—fast early growth that levels off. If you evaluated this program at Year 1 alone, you would undervalue it. If you projected Year 1 growth linearly to Year 10, you would overvalue it. Neither ROI calculation captures the actual shape.
Now compare their Electrical/Electronic Engineering Technology certificate: $46,845 at Year 1, $61,332 at Year 5, $77,093 at Year 10. Nearly identical at the 5-year mark, but a fundamentally different trajectory—this one keeps climbing.
These are programs that look identical at the 5-year mark but tell very different stories about long-term career paths. An ROI number would flatten both into the same bucket. The ladder shows you which one is still rising.
The spread tells you what a number cannot
The 50th percentile—the median—is the number everyone reports. But the distance between the 25th and 75th percentiles tells you something just as important: how predictable the outcome is.
At Emily Griffith Technical College, the Heating, Ventilation, and Air Conditioning program has a Year 5 range of roughly $44,000 to $79,000. That is a $35,000 spread. The median of $58,000 is real, but a prospective student needs to understand that the experience varies enormously. Some graduates are earning close to $80,000; others are closer to $44,000. Same credential, same school, very different outcomes.
Compare that to a program where the P25–P75 spread is tight—say $15,000. That program may have a lower median, but the outcome is more predictable. For a student weighing debt or opportunity cost, predictability has real value.
This is what earnings variance captures. When we collapse the distribution into a single ROI figure, we are telling students the average weather when they need the forecast.
What the blank spaces are telling us
Of those 633 program-institution combinations in the Colorado data, only 283 have any median earnings to report. The rest—more than half—are entirely suppressed.
This is not a data quality problem. It is a sample size problem, and it is telling us something important. The Census Bureau applies differential privacy protections that suppress cells where the graduate count is too small to release without risking individual identification. Programs with a handful of graduates over a five-year window simply cannot produce a reliable—or safe—earnings estimate.
These are real programs training real students. We just cannot see their outcomes. For the programs that serve the smallest, most specialized populations—often the students who most need good information—the data goes dark.
Fort Lewis College, the only four-year institution in Colorado's southwest, has certificate programs in the data but zero reportable earnings across every single one. Not because the programs failed, but because the cohorts are too small for the privacy math to clear.
This is a structural limitation that any honest evaluation framework has to acknowledge. If your ROI scorecard only covers the programs with enough graduates to survive privacy thresholds, you are systematically excluding the smallest programs—which are disproportionately rural, tribal-serving, and specialized workforce training. (For more on data suppression as a structural problem, see The Programs We Can't See.)
What this looks like
I built an interactive visualization of this data that shows the full picture for each school: every program as a horizontal range bar, with the median marked and whiskers stretching from the 25th to 75th percentile, three lines per program for Year 1, Year 5, and Year 10.
Select Colorado from the state dropdown, choose a school, and the programs sort by Year 5 median earnings. The patterns are immediately visible: which programs have strong ladders, which have wide spreads, and which have no data at all.
The career ladder chart traces median earnings progression from Year 1 through Year 10, with shaded bands showing the P25–P75 range at each point. The earnings growth scatter puts two dimensions together: how fast earnings grow and how much outcomes vary—so you can see which programs offer strong growth with predictable results and which carry more risk.
So what
Colorado, like most states, is making real funding and policy decisions about which workforce programs to expand, which to cut, and how to advise students. If the evaluation framework is a single ROI number, the programs that look "low-performing" may actually be the ones with the strongest growth trajectories. And the programs with the highest Year 1 earnings may be the ones that plateau fastest.
The career ladder and earnings variance framework does not replace accountability. It makes accountability honest.
A program that starts at $35,000 but reaches $65,000 by Year 10 with a tight spread is a different proposition than one that starts at $45,000 and flatlines. Both look similar in a Year 1 snapshot. Both look very different over time.
The data to see this distinction already exists. It is in the PSEO. The question is whether we use it or continue to reduce it to a single number.
Data: U.S. Census Bureau, Post-Secondary Employment Outcomes (PSEO), 2025Q4 release. Earnings in 2023 dollars (CPI-U adjusted). Short-term certificate programs (<1 year), Colorado institutions, all graduation cohorts, national geographic scope. Explore the full dataset in the interactive earnings explorer.