Salary Benchmark Sources: Audit the Data Before You Hire

Before you use a salary benchmark to set a hiring budget, check where its numbers came from, who they describe and what they measure. A survey of wages paid, a collection of advertised salary ranges and a table of total compensation can each be useful. They need distinct labels before you use them in the same hiring decision.

That sounds obvious until a plausible number arrives with three citations and a confident explanation. You follow one link, see a familiar publisher and move on. The spreadsheet keeps the figure. The uncertainty disappears.

An October 7 account published by Peec AI offers a reason to slow down. Researcher Metehan Yesilyurt says he published 50,000 invented statistics on a site called StateGlobe and later found its numbers cited elsewhere. His subjects included marketing and web technology. The experiment does not establish that salary databases contain those fabricated numbers. Its relevance here is narrower: repeated citation can give an unsupported statistic the appearance of evidence.

You can audit a compensation figure without conducting your own survey. The result should be a short record explaining why a particular range belongs in your job posting.

What does the salary benchmark actually measure?

Start by naming the observation behind the figure. “Salary data” is too broad a description to support a budget.

A job board shows what employers advertise. That can help you understand the ranges candidates encounter while searching. It does not tell you what those employers ultimately paid, how many people accepted or whether the advertised ceiling was available to a candidate with your requirements.

An employer survey can describe wages paid to existing employees. Its collection method, occupation definitions and coverage determine how closely those employees resemble the person you want to hire. An employee submission can offer another perspective if you can inspect how the publisher handles verification, data age and selection.

Then there is the pay definition. Base pay, cash compensation and equity-inclusive total compensation are different measures. Before comparing two figures, ask what each includes and how its components were valued. An impressive total-compensation figure may tell you little about your cash budget if the source does not separate the components.

Use a simple label beside every candidate benchmark:

Keep the source’s label even when it makes the comparison less tidy. Calling an advertised floor a market salary median creates a stronger claim than the observation supports.

For a founding engineer’s salary and equity package, this distinction becomes especially useful. A candidate deciding between cash and ownership needs you to explain both. A single blended figure can conceal the tradeoff you need to discuss.

Trace the number to its original collection

Follow the citation until you reach a source that explains how the information was collected. A second article repeating the same report gives you another explanation of the sample, not another sample.

Suppose an AI assistant returns a salary estimate with several links. Open them. Identify which source collected the underlying observations. If every article eventually points to the same survey, record that survey once and assess its methods. If the chain ends at another unsourced assertion, leave the figure out of the budget calculation.

The Peec account is a useful warning about that distinction, with a clear limit. Its creator describes an experiment he ran and outcomes he observed; this article has not independently audited those outcomes. You do not need to accept every traffic or citation claim to adopt a sound research habit: inspect the measurement behind a statistic before relying on its repetition.

A transparent source makes that inspection possible. The US Bureau of Labor Statistics publishes technical notes for its May 2024 Occupational Employment and Wage Statistics estimates. Those notes describe a sample of about 1.1 million establishments, with six semiannual panels collected over three years. Earlier wages are adjusted to the reference date, and the estimates use modeling.

That is a dated methodological example, not a claim that May 2024 is the latest release. The distinction matters: the reference period printed beside a number and the date you opened its webpage tell you different things.

In your research, record three dates when available: the observation period, the release date and your access date. A newly updated article can still quote an older collection. A recent collection can still describe a population that does not match your role.

If you cannot establish the collection period or pay definition, label the source as background reading. It has not yet earned a place in the numerical comparison.

Check the role, geography and employer sample

Match the benchmark to the work you need done before choosing a percentile. A familiar job title is a starting point for that match.

Write down the responsibilities first. Does your “senior engineer” own an unfamiliar service, mentor a team, handle production incidents or mainly deliver well-scoped changes? Compare those requirements with the source’s occupation and level definitions. Where the definitions are broad, preserve that limitation beside the figure.

The BLS OEWS FAQ discusses responsibility, experience, geography and employer size among influences on wage comparisons. A credible collection can therefore remain a poor match for your vacancy. A national occupational estimate does not become a startup-specific seniority band because the job titles sound similar.

Geography needs the same care. Decide which labor market you are hiring in: a particular city, a country or a defined set of locations. For a remote role, explain which locations candidates may work from and how you intend to set pay. “Remote” alone does not identify a single market.

Next, look at the sample’s composition. Ask how many observations contribute to the relevant slice, how many employers they represent and whether one employer supplies much of the data. A page with many advertisements may offer limited diversity if they come from the same company.

Do not assume advertisements are independent observations. The same vacancy may appear on several boards or be republished. Check the publisher’s approach to duplicate listings before treating a count as a count of distinct opportunities.

There is no universal sample threshold in this article. A larger sample from the wrong population can be less useful than a smaller, clearly described comparison. If narrowing the role and location leaves too little evidence, say so. You can broaden the search deliberately, but record which condition you relaxed and why.

Reconcile pay periods before comparing ranges

Put comparable figures on the same currency and time basis, then keep the assumptions visible. Arithmetic cannot reconcile different employment terms.

An annual salary and a monthly salary need a period conversion. An hourly rate needs an explicit hours assumption. A currency conversion needs a recorded rate and date so someone else can reproduce it. Retain the original figures alongside the converted values so a later reviewer can see what changed.

Also establish whether the amount is gross or net, and whether it describes an employee or a contractor. Converting a contractor’s invoice rate into a monthly number does not account for benefits, leave, tax treatment or other differences. Keep those observations separate until you have a defensible comparison for your decision.

Even well-documented public sources have specific pay definitions. The BLS technical notes describe OEWS wages as straight-time gross pay, with specified incentive pay included and items such as overtime, nonproduction bonuses and employer benefit costs excluded. Those figures should not be relabeled as comprehensive employer cost or total compensation.

Dates require restraint too. The BLS FAQ explains that OEWS is designed for cross-sectional estimates and discourages time-series comparisons without accounting for comparability issues. Two annual tables do not automatically establish that your hiring budget should rise by their apparent difference.

When sources disagree, first check definitions, populations and periods. Keep both results in view while you investigate. Averaging incompatible numbers can make a spreadsheet look settled while leaving the underlying decision unresolved.

Keep a decision record with the salary band

A useful audit ends with a decision another person can reconstruct. Keep the evidence, exclusions and chosen range together.

The worksheet below can live in the document where you approve the vacancy. Complete the source-related fields for each underlying collection, then record the chosen band and ownership. It is a proposed working format, not a claim about a built-in Kit record.

The most valuable sentence in the record may explain what you excluded: “We left out total-compensation submissions because the cash component was unavailable.” Or: “We widened the location comparison because the exact city slice did not support a useful comparison.” Those are hypothetical examples of decision notes, not findings about a real dataset.

Choose the range after reviewing the evidence against your budget, responsibilities and compensation policy. A market comparison informs that choice; it does not make it for you. If you select a higher or lower range, write down the reason while it is still clear.

Then check that you publish the approved range. Our guide to honest salary ranges covers that next step. The research record should help you explain the posting, including where the available evidence was incomplete.

Use Kit’s listing evidence with its limits in view

Kit’s Compensation Research can help you inspect the advertised listings behind a role comparison. Use that access to examine the observations before adopting the summary.

The role view provides source links where a usable URL is available, along with company information and posted dates in its desktop table. Filters help narrow the comparison. Location comparisons show listing and company counts, with percentiles withheld unless a row has at least five listings and three identified companies. That is a product guardrail, not a guarantee of statistical representativeness.

The calculation needs an accurate label too. By default, the role view’s percentiles describe advertised salary minima, not accepted offers or employee wages. The API also supports range midpoints and maxima. For these comparisons, annual amounts are divided by 12, hourly amounts use an assumption of 160 hours per month, and listings without a stated period are excluded.

A remote listing can also belong to a city row. Those rows overlap; adding their counts would misstate the sample. Source links and visible counts help you investigate these details, but they do not verify an employer’s claim or establish complete market coverage.

We have written about bringing compensation research into the recruiting workflow. The practical standard remains simple: preserve what the source observed, explain the comparison you made and keep the reason for your band. If you try Kit, start with the listings for a role and region you know, and judge how well the evidence matches the hire you intend to make.