A freelance photo editor we’ll call the freelancer for this piece—self-employed, no W-2, income that swings from a slow $1,800 month to a $7,000 month depending on which clients pay on time—ran her situation through four different Medicaid pre-screeners over the course of a week. She wasn’t shopping for the answer she liked best. She was trying to get a second opinion after the first result surprised her. By the fourth tool, she had four different eligibility conclusions and no clearer sense of where she actually stood. This is a useful case study precisely because nothing about her situation was exotic. Irregular self-employment income is common, and it happens to be the exact input that exposes how differently these tools are built underneath a nearly identical-looking form.
How the freelancer’s income was entered into each tool
The four screeners she used included a state Medicaid agency’s own pre-screening page, a national nonprofit’s benefits-navigator tool, a health insurance marketplace calculator, and a county human services department’s standalone eligibility estimator. All four asked for “income,” but they asked for it in four different shapes.
- One asked for current monthly income, with a note to enter “what you expect to earn this month.”
- One asked for total income for the current calendar year so far, divided by the number of months elapsed.
- One asked for an annual estimate, framed the way marketplace applications frame it—project what you’ll earn for the full year, similar to how you’d estimate it on a subsidy application.
- One asked for average income over the last three months, with a small text box for “self-employed applicants” that gave no further instruction.
She answered each one honestly, using the numbers she had in front of her. The problem is that “honestly” doesn’t mean “consistently” when the underlying question is being asked four different ways. A snapshot of this month, a year-to-date average, a forward-looking annual projection, and a trailing three-month average are four different numbers even when they’re describing the same person’s finances. For someone with a steady salary, these methods converge—divide the same annual figure by twelve and you get roughly the same monthly number every time. For someone whose income moves by a factor of three or four between months, they don’t converge at all. They diverge in exactly the direction that determines whether an income line sits above or below a program threshold.
The numbers she actually entered
Without reproducing her exact figures, the shape of the problem was this: her three most recent months included one very strong month and two weak ones, so a trailing-average tool skewed high. Her year-to-date total, divided evenly, sat closer to the middle because it absorbed a slow stretch earlier in the year. Her forward-looking annual projection depended entirely on how optimistic she felt about upcoming contracts, which is a judgment call, not a fact. And her single-month snapshot happened to fall on one of her better weeks, which pushed that tool’s estimate to the top of the range. None of these entries were wrong. Each was a faithful answer to the specific question asked. The tools simply asked different questions while displaying an identical-looking label: “Income.”
Why variable monthly income breaks simple annualized formulas
Most eligibility screeners are built around one core operation: take an income figure, multiply or divide it into an annual or monthly equivalent, and compare that number against a published threshold for household size. This works cleanly for a salaried employee, because a salary is designed to be smooth—the same amount lands every pay period, and any of the four methods above will produce the same answer.
Self-employment income is not smooth by design. It’s lumpy because of invoicing cycles, seasonal demand, client payment delays, and the fact that business expenses get netted out unevenly from month to month. A designer might do six weeks of unpaid production work and then get paid for all of it at once. A landscaper might earn almost nothing in January and triple that in June. None of this is unusual for self-employed people, but it is genuinely hard to compress into a single “monthly income” number without deciding, implicitly, which period counts as representative.
The formal rules that programs like Medicaid actually use for counting self-employment income are more specific than “type in a number.” They generally involve looking at income over a defined period, applying particular treatment to business expenses, and in some cases allowing for reasonable projections when past income doesn’t predict the future well. Screeners rarely implement that level of detail. They’re built to be fast and low-friction, which means they tend to pick one simplifying assumption and apply it universally, whether or not it fits the person typing in the number.
That’s the structural reason variable income breaks these tools: the screener has to choose a reference period, and for a self-employed applicant, the choice of reference period is not a minor implementation detail—it’s the whole ballgame.
Comparing how each tool guessed at averaging the income
Looking at the four tools side by side, the differences weren’t cosmetic. Each one had made a distinct, largely invisible design decision about how to handle non-salaried income, and none of them disclosed that decision to the user in a way that would let her know it was happening.
The current-month snapshot tool treated whatever she entered as a stable monthly figure and multiplied it by twelve to estimate annual income. This method assumes the current month is representative of every month. For a salaried worker, that’s a safe assumption. For a freelancer, it means the tool’s output is only as good as the timing of when she happens to fill out the form.
The year-to-date average tool was less sensitive to any single month, but it quietly weighted the earlier, slower part of her year more heavily just because there was more of it to average against. It produced a more stable-looking number, but “stable” isn’t the same as “accurate”—it just smoothed over the same volatility rather than accounting for it.
The forward-looking annual projection tool put the estimation burden on her. It asked her to guess her own annual total, which is a reasonable approach in principle—it mirrors how some programs actually want self-employed applicants to think about the question—but it offered no guidance on how to build that estimate from partial-year data, so two equally honest freelancers with identical finances could produce different projections depending on how cautious or optimistic they felt that day.
The trailing three-month average tool was the most exposed to short-term noise. Three months is a small enough sample that one unusually good or bad month can swing the average substantially, and for someone whose income cycle runs longer than three months, this window simply doesn’t capture the pattern.
None of these four approaches is “the wrong one” in isolation. Each corresponds to a real method that some part of the benefits system uses somewhere, for some purpose. The trouble is that a single applicant, entering the same real financial life into four boxes labeled the same way, has no way to know which method is running underneath the label—and the tools don’t tell her.
Why the same underlying data produced different conclusions
It’s worth being precise about what actually happened here, because it’s easy to conflate “the tools disagreed” with “the tools are broken.” They weren’t broken. Each one performed its designed calculation correctly on the input it was given. The disagreement came from upstream—from the fact that “monthly income” is not a well-defined quantity for someone whose income doesn’t arrive monthly in any consistent sense. Ask four different, valid questions about the same irregular income stream, and four different, valid-looking answers are exactly what you should expect.
What this case suggests about screeners and non-traditional earners
The lesson here isn’t that any particular screener did a bad job, or that one of the four “got it right” and the others got it wrong. It’s that pre-screeners, as a category, are built around an implicit assumption of income regularity, and self-employed or gig-based earners sit outside that assumption more often than the tools acknowledge. The same issue tends to show up for seasonal workers, tipped employees, people paid by commission, and anyone whose hours fluctuate week to week—variable income is the common thread, not self-employment specifically.
A few things follow from this that are worth keeping in mind if you’re in a similar position:
- A screener result is a rough estimate of an estimate. It’s not just approximating your eligibility—it’s approximating your income first, using a method it likely doesn’t disclose, and then approximating eligibility on top
