Ten occupations in our data have enough salary observations on both sides to compare remote postings against onsite ones. Every single premium came out large. Security engineering, plus 119 percent. Product management, plus 99. Software engineering, plus 87 on a sample of 409 remote against 1,877 onsite. Numbers like that would make a lovely headline, and we drafted one.
Then we tried to break it, because that is the house rule.
| Occupation | Remote n | Onsite n | Apparent premium |
|---|---|---|---|
| Security engineer | 42 | 372 | +119% |
| Product manager | 102 | 873 | +99% |
| Account executive | 165 | 659 | +94% |
| Software engineer | 409 | 1,877 | +87% |
| Marketing manager | 43 | 884 | +93% |
| Management consultant | 42 | 742 | +81% |
| Product designer | 32 | 413 | +77% |
| Project manager | 72 | 1,326 | +74% |
| Machine learning engineer | 51 | 525 | +72% |
| Data scientist | 55 | 491 | +61% |
The break attempt
The problem hiding in that table is where each column comes from. Our remote observations arrive mostly through remote-first boards, which skew senior, tech-heavy, and venture-funded. The onsite pool arrives mostly through general boards carrying everything from federal agencies to regional firms. Comparing the two medians mostly tells you which kinds of companies use which kinds of boards, and only secondly what working from home pays.
The clean test would compare remote and onsite postings inside one source, same board, same employer mix. We ran it. It cannot be done with our current data: the general boards barely flag remote at all. Adzuna gave us 11 remote software postings against 1,095 onsite, and single digits for every other occupation we tried. A comparison that thin proves nothing in either direction, so we are publishing the failure instead of the headline.
What survives scrutiny
Three things, more modest than the table. Remote-first employers do post high salaries; whatever the cause, those jobs exist and are real money. The official anchor gives scale: the US all-worker median for software engineering sits near 133,000 dollars (BLS OEWS (the Bureau of Labor Statistics wage survey)), and remote-board postings cluster well above it, so the population posting remotely is simply not the median population. And the direction of the bias is knowable even where its size is not, which means any remote-pay figure you read, including ours, is an upper bound until someone shows you a same-source comparison.
How to read any remote-pay claim, including this one
The composition trap has a three-question test, and it works on every remote-salary article ever published. First, are the remote and onsite numbers drawn from the same source, or from a remote board compared against the general market? If the article does not say, assume the worst, because same-source data is rare and authors who have it brag about it. Second, is seniority controlled in any way, even crudely, since remote-first hiring skews senior and a seniority gap masquerades perfectly as a location premium. Third, are the sample sizes published next to the percentages? Our own table above includes an n of 42 producing a 119 percent headline, which is exactly the kind of number that evaporates when the sample doubles.
Run those three questions against the remote-pay statistics you have seen this year and most will fail all three. Ours fails the first two and passes the third, which is why this piece exists.
What a defensible premium would probably look like
Bounded speculation, labeled as such: studies with employer-level controls in adjacent literatures, and the few same-company disclosures that exist, tend to land location-flexible pay differences in the range of 0 to 20 percent, not 60 to 120. Remote work reprices geography and widens the buyer pool, and both effects are real, but nothing in labor economics suggests the same worker doing the same job doubles in value by leaving the building. When our within-source test becomes possible, that is the range we expect it to confirm, and we will publish whatever it says either way.
Where this leaves you
If you are negotiating a remote offer, use occupation-level anchors rather than remote-market headlines: the blended bands on our salary pages start from official statistics and declare their posting bias, and FairElephant will weigh a specific number against your location and remote rates. If you are choosing remote work expecting an automatic 87 percent raise, expect instead a wider set of employers competing for you, which is worth plenty and is not the same thing. We will rerun the within-source test as general boards improve their remote flags, and this page will change when the evidence does.
Apparent premiums: posted salary medians, remote-flagged versus not, minimum 30 observations per side, PivotHop July 2026 run. Within-source test: Adzuna-only split, reported counts above. Official anchor: BLS OEWS May 2024, Standard Occupational Classification (SOC) 15-1252 family. This piece supersedes any earlier internal use of the raw premium figures.