Methodology
How every number on this site is produced
This page exists so that any figure we publish can be checked, argued with, or shown to be wrong. If a number here cannot be traced back through these steps, it should not be on the site.
The effective hourly rate
Every method is ranked by one computed figure: what an hour of your life is worth doing it, after money spent and after time spent that nobody pays for. It is never entered by hand and there is no editorial override, which is what makes the ordering meaningful rather than a matter of opinion.
The calculation is deliberately unflattering. Where a judgement call exists we take the more conservative reading, because a number that overstates what a method pays does real damage to someone deciding how to spend their next six months.
Which unpaid time is counted
This is where most published hourly figures quietly inflate. A rate computed only on billable hours is not the rate anyone experiences. We count all of the following:
- Setup before the first paid job: account creation, verification, equipment purchase and learning.
- Finding work — applications, pitching, bidding, listing, and the time between accepted jobs.
- Admin: invoicing, chasing payment, bookkeeping, and tax preparation attributable to this income.
- Unpaid waiting — time on-platform but not earning, which is substantial in delivery and rideshare work.
- Rework: revisions, returns, disputes and refunds.
Which expenses are subtracted
Gross earnings are the second common inflation route. We subtract platform and payment-processor fees, materials and stock, fuel and vehicle depreciation, software subscriptions, insurance, and any licence or permit the method requires. Where a cost varies by geography we use the median for the market rather than the cheapest case.
Why every figure is a median
A mean is the wrong statistic for income distributions that are heavily skewed, and income from independent work is about as skewed as data gets. A small number of very high earners pulls an average far above what a typical participant experiences. Quoting that average is technically true and practically a lie.
The scale of the distortion is not subtle. US side-hustle earnings average $885 a month against a median of $200. French micro-entrepreneurs average €773 a month against a median under €350. In both cases the average is more than four times the middle of the distribution.
We therefore publish the median, and we publish the 25th and 75th percentiles beside it so that the spread stays visible. A publishing rule enforces this: a mean appearing in a published field is treated as a defect, not a stylistic choice.
How failure rates are derived
A failure rate needs a denominator — the number of people who genuinely attempted the thing, including everyone who quietly earned nothing and never wrote about it. Online communities cannot supply this, because the people who failed are the people who stopped posting.
Tax records can. When someone operates a business, they file, and the resulting statistics describe the whole population rather than a self-selected sample. US census and IRS data report how many people ran a business of a given type and how their receipts and net income were distributed. The share sitting in the lowest bands is the failure rate.
This is public, free and authoritative data that this field does not use, for a straightforward reason: the results are damaging to any business model that depends on the reader believing the method will work.
What a pay figure leaves out
Federal wage data covers people who are EMPLOYED in an occupation. Everyone who trained for it and never got hired is absent from the median, and so is anyone doing the same work self-employed. Every median on this site is therefore a survivor's figure, and the pages say so.
Where a licensing or certifying body publishes a standing count of everyone who holds the credential, we set that count beside the number of people federal records show employed in the work, so the size of the gap is visible. Both sides have to be the same kind of quantity for that division to mean anything: a standing stock of holders against a standing stock of employed people. Where the only count available is an annual flow — people newly credentialed in one year — we publish the count and refuse the ratio, because dividing a flow into a stock produces something that reads like a hiring rate and is not one.
The gap that division produces is a ceiling, never an unemployment rate. It contains everyone who moved into management, teaching, sales or inspection, everyone who retired without surrendering the credential, everyone working part time, agency or self-employed in a way the employment survey does not reach, and everyone holding it as a second string to another job. The real number of trained people shut out of the work is lower than the gap, and we do not know how much lower. Most occupations have no verified holder count at all; those pages show none rather than an estimate.
The payback baseline
A credential page reports how many months of extra pay it takes to cover what the credential costs. The premium is measured against the employment-weighted median of every occupation on this site — the midpoint American worker — because we do not know what any particular reader earns now. If you already earn more than that, the payback is longer than shown; if you earn less, it is shorter.
The figure ignores the months of training, during which most people earn less or nothing, and it ignores interest on anything borrowed. It is also conditional on being hired into the work at all — every payback number on this site assumes you get the job, which is the assumption the holder counts above exist to test.
The evidence ladder
Every number carries a tier, shown on the page rather than buried in a footnote. The tier tells you how much weight the figure can bear.
| Tier | What it means |
|---|---|
| 1 Population | Federal tax or census aggregates covering everyone who filed. The strongest evidence available and the basis of every failure rate. |
| 2 Platform | Official payout schedules, published rate cards and platform transparency reports. |
| 3 Verified log | Practitioner earnings submitted with redacted proof — a tax form, a dated payout screenshot, a bank line. |
| 4 Self-reported | Survey or community input. Displayed, weighted down, and never the sole basis for a published figure. |
| 5 Anecdote | Not published as a number at all. |
Where this method is weak
Industry classification codes are coarser than a method. Several genuinely different ways of earning can sit under one code, which means they share a federal denominator. Where that happens we say so on the page and label the figure as shared across a group rather than presenting it as specific to one method.
Nonemployer statistics exclude the very smallest receipts, so the true share of people earning almost nothing is likely higher than we report, not lower. Gig income paid on a W-2 does not appear in sole-proprietor data at all. And tax data lags: the most recent complete year is typically two years behind, so a method that collapsed last year may still look viable in the federal figures.
None of these limits are reasons to use worse data. They are reasons to state the confidence level on every record, which we do.
If you think a number is wrong
Tell us, and include the method, the figure you are disputing and the source you believe is correct. Corrections that hold up are applied and the change is noted on the record. A site that publishes a failure rate for other people's work should be able to take a correction about its own.