The Market Pay Estimate Standard, and the Numbers That Fail It

You have pulled pay figures for your role from three or four places, and now they disagree. This guide is for the job seeker who needs to decide whether the picture is solid enough to anchor a target and an ask, or whether it is a pile of numbers dressed up as a conclusion. It gives you a pass/fail grading rubric with six criteria, so you can stop second-guessing, commit to a target, and know exactly which figures to throw out.

Most pages on this topic review each source one by one and end with "use it as one input." That advice never tells you when your own compiled number is done. This standard does. It defines done, and it names the numbers that fail it.

What "trustworthy" means for a compiled pay estimate

A trustworthy market pay estimate is one that passes all six criteria at once: a real sample floor, at least two independent source types, current-or-aged recency, a single-market geography match, a content-based level match, and a clean separation of base from total comp. Miss any one and the estimate fails, because each criterion closes a different way the number lies.

The reason for pass/fail rather than a weighted score is that these failures are not additive. A stale figure is not "mostly right." A total-comp number compared against a base target is not "close." Each defect can move your anchor by more than the negotiation gap you are fighting over. So the grade is binary, and a single fail sinks the estimate until you fix it.

That 67% figure is the whole case for the independence criterion. A number that misses one time in three cannot be adopted from a single source. It has to be corroborated. Everything below turns that instinct into a rule two people can apply the same way.

A stale figure is not mostly right, and a total-comp number is not close. Each defect fails the grade outright.

The six criteria, and what each one proves

The six criteria are sample floor, source independence, recency, geography match, level match, and base-versus-total clarity. Each one exists because a specific, documented distortion lives in the gap it covers. Here is what each proves, and what it looks like when it lies.

• Sample floor. Proves the median is not an accident of two people. It lies when a confident-looking median actually rests on one or two reports. Glassdoor's floor is three or more submissions before it shows a median at all.
• Source independence. Proves the number was seen by more than one measurement method. It lies when three "sources" all re-scrape the same crowdsourced pool. Independence requires at least one payroll, survey, or government source distinct from crowdsourced.
• Recency. Proves the figure reflects the current market. It lies when a five-year-old submission sits next to fresh data with no expiry date. Un-aged staleness can understate pay by 5 to 15 percent.
• Geography match. Proves the number belongs to your market. It lies when a national figure is applied to a high-cost metro. Differentials run 40 to 60 percent between markets.
• Level match. Proves the benchmark is your job, not its name. It lies when the title matches but the scope does not. Matching a senior role to a mid-level benchmark is the primary source of benchmarking error.
• Base-versus-total clarity. Proves you are comparing like with like. It lies when a "high" number is total comp measured against a base-only target. Crowdsourced tools layer bonuses, profit-sharing, and commissions into total pay.

Why one source is never enough

No single source clears the independence criterion, because each source type has a documented direction of bias. Grading is a matter of pulling from types that fail in different directions and reconciling them, not of finding the one honest site.

The three source types bias predictably. Crowdsourced self-report rounds up and never retires stale points. Employer-posted ranges skew low and truncated. Government series lag and report broad categories that understate niche or senior roles. Put a crowdsourced figure next to a government figure and the truth is bracketed between two known errors, which is the entire point of triangulation.

The government option is stronger than most job seekers assume. The BLS Occupational Employment and Wage Statistics program covers over 830 occupations across all US states and more than 500 metro areas, free, drawn from employer payroll records rather than self-report. Its weakness is lag: it updates annually 12 to 18 months behind, and reports broad occupational categories rather than company-specific figures. As of May 2025 the BLS median hourly wage across all US occupations was $24.51, which is a useful anchor for a broad role and useless for a specialized one.

The gold standard is HR-reported survey data from providers like Culpepper, WTW, Radford, or Korn Ferry, which comes directly from employers. Most job seekers cannot buy it, so the realistic independent pair is one crowdsourced tool plus one government series. If you can reach someone who builds these benchmarks, one conversation replaces hours of reconciliation.

The range-accuracy problem, in numbers

Employer-posted ranges are the most tempting source and the most misread. They feel official, but the numbers say a posted range is a weak input that has to be discounted before it counts. Two distortions do the damage: the miss rate and the median trap.

Read the last row carefully, because it is the one that costs people money. Even when a range is accurate, over 60 percent of real pay falls below the range median. Anchoring your ask at the posted midpoint systematically overshoots what most people in that role actually receive. The mechanism is that employers pad the top of the band, so the midpoint is not the center of a bell curve - it sits above where most pay lands.

There is a quieter distortion too. The recommended high end of a range should sit 40 to 60 percent above the minimum, but an analysis of more than 12 million listings found the actual spread is only about 28 percent. That compression means the advertised top is often below the real ceiling. Meanwhile, under half of US job postings include salary information at all, and only about 10 percent note an exact pay level, so you are often reconciling a handful of truncated bands.

Grade a compiled estimate in seven steps

The procedure below takes a pile of figures to a pass/fail verdict in about half a day of focused work. Run it in order; each step assumes the previous one is done, and the last step is only meaningful once the first six have been executed.

Two of these steps carry more weight than the rest. Defining the role by content is where most estimates are lost before they begin, and dating every figure is where a clean-looking pile silently rots. Give those two the most time.

Building the resume that matches the role you just defined by content, and tailoring it to each posting so your claimed scope lines up with the level you are benchmarking, is exactly the friction Refolk removes - it writes from your own history and scores how well you actually fit each posting, which keeps your target and your applications pointed at the same level.

Content match and geography, done precisely

Match on content, not title, and pin one market before you compare anything. These two criteria - level match and geography match - are where a pile of figures that look consistent quietly turns out to be measuring different jobs in different places.

For level match, aim for a benchmark job description whose content matches 80 percent or more of yours. Not a perfect match, and never a title match. Match by responsibilities, scope, skills, decision-making authority, and reporting relationships. The dossier is blunt that a senior-to-mid mismatch is the primary source of benchmarking error, so if the title says "analyst" but the scope says "leads a team," grade against the scope.

For geography, the number that should scare you is the differential: the same role in San Francisco can carry a 50th-percentile rate 40 to 60 percent above an equivalent secondary Midwest market. A national figure applied to a high-cost metro is not conservative, it is wrong by up to 60 percent. Filter every source to one metro before you reconcile.

How this goes wrong: eight false positives

The most valuable part of a standard is the list of things that pass a lazy look and fail a real one. Every entry below is a false positive: a number that looks trustworthy and is not. Check each one explicitly before you grade an estimate as passing.

• Median built on a thin sample. A confident-looking median that rests on one or two reports. Check the three-or-more-submission floor and the confidence badge before trusting it.
• Base-versus-total confusion. A "high" number that is total comp compared against a base-only target. Separate the figures, because tools layer bonuses and commissions into total pay.
• Range treated as a bell centered on the median. Aiming at the midpoint when over 60 percent of pay sits below the range median. Anchor to reported medians, not posted-range midpoints.
• Stale point used as current. A five-year-old submission with no expiry sitting next to fresh data. Date-stamp and discard or age it; understatement can reach 5 to 15 percent.
• Title match instead of content match. Same title, different scope. Enforce the 80 percent content-match rule, since a senior-to-mid mismatch is the primary benchmarking error.
• Wrong geography. A national number applied to a high-cost metro. Filter to a single market; differentials run 40 to 60 percent.
• Fake independence. Three "sources" that all re-scrape the same crowdsourced pool. Require at least one payroll, survey, or government source distinct from crowdsourced.
• Trusting a very wide posted range. Reading a band like $50,000 to $250,000 as informative. Discount any range far outside the 40 to 60 percent norm as non-informative.

Aging, and why the calendar downgrades your data

Aging is the practice of moving a figure forward to a common effective date because surveys are published on a lag. It matters because staleness is a silent downgrade, not a rounding error, and it usually runs in the direction that costs you money.

Data collection closes months before publication, and organizations often use a survey for a full year after it appears, so published data may lag the analysis date by 6 to 18 months. Government series lag 12 to 18 months by design. Left un-aged in a fast-moving market, that gap can understate current compensation by 5 to 15 percent - frequently larger than the negotiation gap a job seeker is arguing over.

The working rules conflict slightly across sources, and the standard resolves the conflict conservatively: age figures between roughly 9 and 18 months forward to your reference date, and replace anything more than a year old rather than aging it, since one source holds that data over a year old should be swapped for a newer source entirely. When in doubt, replace rather than stretch.

Source name | Source type (crowd/employer/gov/survey) | Base or Total | Sample size / confidence | Effective date | Metro | Level | Aged figure at reference date
Example row:
Glassdoor | crowdsourced | Base | 12 reports, high confidence | 8 months ago | Denver metro | Senior | aged +4% to today
Government series | gov OEWS | Base | full occupation | 14 months ago | Denver metro | (broad) | aged +6% to today

Keep the estimate current, and know where the data is scarce

A graded estimate is a snapshot, not a permanent verdict. Re-grade it when any source crosses its recency threshold, when you change target market or level, or when a new government release lands. The BLS OEWS cycle is the clearest calendar to track: it publishes annually, and knowing your latest release date tells you exactly when your government anchor has gone stale.

Where the standard is thinnest is outside the United States, and it is worth saying plainly. Reliable, methodologically aged, content-matched benchmarks are produced by a scarce profession that is heavily concentrated in one country.

In Refolk's index of professional profiles there are 4,648 US professionals with compensation-analyst or manager titles, against 277 in Canada and 108 in the UK - roughly a 43x US-to-UK gap. The practical consequence: outside the US, fewer practitioners produce aged, content-matched benchmarks, so non-US readers lean harder on lagging government series and should weight the recency criterion even more carefully.

number: 4,648
label: US professionals with compensation-analyst or manager titles in Refolk's index
note: Against 277 in Canada and 108 in the UK - the people who build the aged, content-matched benchmarks you are trying to grade.