Most people meet a compensation survey in one of two ways. Either someone hands you a report full of percentiles and asks what it means, or someone asks you to run one. This page is for both situations, and it stops at the concept. Running one is a separate subject with its own guide.
The mechanism is simple enough to state in a sentence. A group of employers each submit what they pay for the same set of jobs, a neutral party combines the submissions, and everyone gets back the distribution without seeing anyone else’s individual figures. That trade, my data for everyone’s data, is the whole arrangement. Everything else is detail about making the trade fair and the result usable.
Wage, salary, or compensation survey?
The three terms get used almost interchangeably, and in practice most people mean the same exercise by all of them. The shades of difference are still worth knowing before a planning meeting.
A wage survey usually signals a focus on hourly roles, where pay is quoted per hour and shift or overtime premiums matter. A salary survey leans toward salaried, annually paid positions. A compensation survey is the broadest of the three: it implies the whole picture, base pay plus variable cash and often the employer’s share of benefits. A survey that covers a plant floor and a front office is a compensation survey whichever word is on the cover, because it has to handle hourly and salaried pay in one instrument.
Who runs them, and why
A single employer runs one to answer two questions it cannot answer alone: whether its pay is competitive against the employers it loses candidates to, and whether its pay is consistent internally across people doing comparable work. Neither question has an honest answer without outside data, because you cannot see a competitor’s payroll and your own numbers cannot tell you whether they are high or low.
The other kind of sponsor is an organization running the survey for other people: a county economic development authority, a chamber, a trade association. Their motivation is different. No individual employer can survey its own competitors and expect honest answers, but a neutral convener can. That is why regional and industry programs exist at all, and why they tend to run year after year once established.
The vocabulary
Market data. Compensation figures from other employers, collected through a survey or purchased from a syndicated provider. Not the same as job postings, which advertise a range rather than report what is actually paid.
Percentile. A position in a ranked list. At the 25th percentile, 75% of the market pays more. The median, or 50th percentile, is the middle value. Percentiles are more honest than an average because a single outlier cannot drag them. If the median wage for a role is $24.00 an hour, half the reporting employers pay more and half pay less.
Benchmarking. Comparing your pay for a job against market data to see whether you are ahead, behind, or aligned. The comparison is only meaningful if the jobs genuinely match.
Job matching. Deciding that your job and someone else’s job are the same job. This is done on the content of the work, not the title. A coordinator at one organization does a manager’s job at another, and titles drift constantly.
SOC codes. Standard Occupational Classification codes, published by the Bureau of Labor Statistics. They give every occupation a stable identifier so your data stays comparable across years, regions, and sources.
Total compensation. Everything of value the employer provides: base pay, bonus, incentive, equity where it exists, and the employer’s share of benefits. A base-only comparison routinely misreads who is competitive.
Safe harbor. The practice of suppressing any result computed from too few employers, so no single participant’s data can be identified. Without it, participants either decline or shade their answers.
Effective date. The date a figure was true. Compensation data ages immediately, and figures without dates cannot be safely combined or compared.
What a compensation survey is not
Several things get called compensation data that are not the same thing, and the difference usually decides how much you can trust a number.
- Not a job-posting aggregator. The pay ranges on large job sites are advertised ranges and self-reported figures, not what employers actually pay the people already in the seat. They describe the top of a hiring funnel, not a settled market rate.
- Not a syndicated dataset you buy. National providers sell large compensation databases, which are genuine survey data but priced for large employers and cut nationally. A regional or industry survey answers a narrower question: what your actual competitors pay, in your labor market.
- Not government wage data. The Bureau of Labor Statistics publishes broad occupational wage estimates for free. They are a useful sanity check, but they are averaged across large areas and lag by design, so they rarely settle a specific pay decision.
- Not an internal pay-equity audit. An equity audit looks inward, comparing your own employees to each other for fairness. A compensation survey looks outward, comparing you to the market. Both matter, and they answer different questions.
Five questions to ask about any survey you are handed
You will more often be reading someone else’s survey than commissioning your own. These are the questions that separate a report you can act on from one that should stay in the folder.
- Who is in it? Which employers, how many, what industries, what geography. If the report does not open with this, be suspicious of everything after it.
- How were the jobs matched? Ask whether participants matched against written job descriptions or just picked a title. Title-only matching quietly averages different jobs together.
- What is inside the number? Base pay alone, base plus incentive, or total cash. A report that never says has left you no way to compare it against your own figures.
- How many employers are behind this specific cell? Not the survey overall, the individual figure you are about to quote. Small cells produce confident-looking numbers that mean very little.
- When was it true? An undated benchmark gets quoted as current long after it stopped being current, usually by someone who was not in the room when it was collected.
When to run one
Annually is the common cadence, and the reason is trend rather than level. One survey tells you where you stand today; two tell you which direction the market is moving, which is the more useful fact. Beyond the regular cadence, the usual triggers are recruiting difficulty, unexpected turnover, entering a new market, or restructuring a role badly enough that no existing benchmark fits it.
Next step
That is the concept. Actually running one is a different subject, covered in depth in how to run a wage and benefits survey, which walks through instrument design, fielding, analysis, and publication. If you want to see how this plays out in a specific sector, the compensation surveys hub breaks it down by industry.