How to Run a Wage and Benefits Survey

Compensation is the price of labor, and labor markets run on information. Employers need to know the going rate to attract and keep the people they need without overpaying. Workers judge whether their pay is fair by comparing it against the market. A wage and benefits survey is how an organization replaces guesswork with evidence: it collects pay and benefits data from comparable employers and turns it into a defensible picture of what the market actually pays.

This is a guide to running one. It covers the full process, from deciding what to ask through publishing results that hold up, and it shows where Sensible Surveys removes the parts that usually go wrong.

Why run a survey at all

Every wage survey answers two questions. The first is external: are we competitive with the market we hire from and lose people to. The second is internal: are we paying our own people equitably relative to one another. The same dataset informs pay structures, merit budgets, hiring ranges, and the answers you give when an employee or a board asks why a job pays what it pays.

For a single employer, that is reason enough. But the highest-value surveys are often run by organizations that field them on behalf of others: economic development organizations, trade and professional associations, state agencies, and regional authorities. For these organizations a wage and benefits survey is not an internal tool, it is a product. Members get pay and benefits data tuned to their exact region or industry, the kind of granularity that no broad national dataset can match. The sponsoring organization gets a recurring, trusted asset that reinforces why membership is worth it.

The catch is that a survey is only as good as three things: how comparable the jobs are, how fresh the data is, and how defensible the whole exercise is. Get those wrong and you publish numbers that quietly mislead the people relying on them. The rest of this guide is organized around getting them right.

What good survey data has to deliver

Before you write a single question, it helps to know what a usable result looks like, because the design decisions you make early are what determine whether you can produce it. Every wage figure you eventually publish should carry these dimensions.

The job, matched on content rather than title. A “coordinator” at one organization does the work of a “manager” at another. Titles drift; the work is what you compare.

A clear definition of what is inside the number. Is it base pay only, base plus incentive, or total cash compensation including bonuses. Mixing these without saying so is the single most common way a survey lies to its readers.

Central tendency and spread together. A lone average hides everything interesting. You want the median and the mean, plus percentiles on either side (10th, 25th, 75th, 90th) and ideally the standard deviation, because almost every job pays across a band rather than at a single point. The spread is often more useful than the midpoint.

Level and experience splits. Entry, mid, and senior pay for the “same” job are effectively different benchmarks. A survey that blends them produces an average that describes no one.

The right market cut. Industry, geography, and organization size each move the number, sometimes substantially. The relevant market for a warehouse role is local; for a specialized professional it may be regional or national.

An effective date. All data ages from the moment it is collected. You cannot combine or compare figures without knowing as of when each one is true.

A trend. Because you make pay decisions in the future, you need a sense of how fast the market is moving so you can project to the date your decision takes effect.

Hold every source to this bar, including the survey you run yourself. The sections that follow are about clearing it.

Wages and benefits are two different surveys

It is worth separating the two halves of “wage and benefits” early, because they behave differently. Wage data is about levels: what a specific job pays. Benefits data is usually about prevalence and design: which plans an employer offers, how generous the employer contribution is, how much paid time off accrues, and how a retirement match is structured. Most experienced practitioners survey wages and benefits as separate sections or even separate surveys, partly because the questions are shaped so differently and partly because asking for too much in one instrument depresses response.

In practice that means benefits questions lean on structured formats: plan-type selections for medical, dental, vision, and retirement, contribution percentages, and paid-time-off accrual broken out by employee class. Sensible Surveys ships pre-built question patterns for exactly these, including a paid-time-off matrix with a “week counts” preset (none, one through five-plus weeks, or not applicable) and drop-in snippets for things like a retirement match matrix, so you are not reconstructing the standard benefits questions from scratch every cycle.

The comparability problem, and how to solve it

If there is one thing that separates a trustworthy survey from a misleading one, it is job comparability. This is the Achilles heel of every compensation survey ever run. If your “staff accountant” is being matched against someone else’s “bookkeeper,” the resulting wage is noise dressed up as a benchmark. The long-standing rule of thumb is a seventy percent match on the actual work performed and the knowledge, skills, and experience required, not a match on job title.

Historically, organizations solved this by trading written job descriptions and judging the overlap by hand. It worked, slowly and subjectively, and it was the main reason mailed surveys produced shaky data: nobody could be sure the jobs lined up. On-site visits produced better comparability because a trained analyst could verify it in person, but visits cost a fortune and did not scale.

Sensible Surveys solves comparability structurally, by anchoring the compensation question to a shared standard. The compensation question type is built on the Bureau of Labor Statistics Standard Occupational Classification system, a public taxonomy of occupations. Respondents select their roles from a curated list of these occupation codes, then enter wage data per occupation at one or more skill levels: entry, mid, senior, director, and executive for professional roles, or apprentice, journeyman, and master for the trades. Because every respondent reports against the same occupational definitions, the wages are comparable across respondents, across organizations, and across years, without anyone hand-matching descriptions. When an industry-specific title does not exist in the standard taxonomy, an administrator adds a custom occupation that receives its own stable code, so the survey can carry the roles the national taxonomy misses while keeping everything else standardized.

Two more design features defend comparability and consistency at the question level. Matrix questions default to the four employment classes that drive most U.S. compensation reporting, union, non-union, salaried exempt, and salaried non-exempt, so respondents report pay within the correct bucket instead of lumping different classes together. And Smart Suggest reads a question as you draft it and proposes the right question type, answer options, and helper text drawn from a library of around fifty common compensation-survey patterns: wages, percentages, tenure, plan-type dropdowns, per-class matrices, and so on. If nothing matches, it still returns a usable suggestion, so the feature never leaves you stuck. The effect is that your questions come out well-formed and consistent across the whole instrument rather than depending on whoever built each one.

Finally, the compensation question can capture the context that turns a raw wage into an interpretable one: optional per-occupation fields for headcount, average years of experience, fill difficulty, shift differential, and annual turnover. That context is what lets a reader understand why a number is what it is.

Where wage data comes from, and where running your own fits

There are three broad ways to get wage data, and it is worth being honest about the tradeoffs of each.

Government statistics are free and broad. National labor statistics agencies, in the United States the Bureau of Labor Statistics, publish occupational wage estimates and maintain the occupational classification system that good surveys build on. This data is excellent for a sanity check and indispensable as a framework, but it is wide rather than deep, slow to publish, and rarely specific to your exact region, your industry niche, or the precise jobs you care about.

Purchased third-party benchmarks are the second route. Compensation consultancies and data platforms sell aggregated benchmarks, and quality and cost vary enormously. The structural weakness is freshness and fit. Most of this data is collected on an annual or quarterly cycle, which means it is already behind by the time it reaches you in fast-moving labor markets. The job catalogs can be laborious to map onto your own roles, and the participant pool often skews toward large enterprises that look nothing like a regional employer base or an association’s membership.

Running or sponsoring your own survey is the third route, and it is the subject of this guide. When your jobs, your region, or your industry are not well represented in purchasable data, or when you are an organization whose entire value proposition is serving your members, you run your own. You control the jobs included, the market definition, the timing, and the participant list, and you own the result outright. It is more work than buying a report, but it is the only way to get data that genuinely fits, and the right system removes most of the work. That system is what the rest of this guide describes.

Designing the survey in Sensible Surveys

Survey design is where most of the quality is won or lost, so the tooling matters.

Templates and a question library. You start from a reusable template, which is really a library of questions, or from a blank slate. Any customization you make for a specific survey clones into a survey-scoped copy, so experimenting on one survey never pollutes your shared library. When a custom question proves its worth, you can promote it back into the reusable library for next time. The editor lays this out as a section sidebar with an outline minimap, a central question canvas, and a live split-view preview that renders exactly what a respondent will see as you edit, so there is no guessing about how a question will land.

Twelve question types. The type system covers the full range a wage and benefits survey needs: short and long text, number, single choice, multiple choice, dropdown, four matrix variants (radio, checkbox, number, percentage, and dropdown), and the compensation panel described above. Each type carries its own respondent interface, its own validation, and its own aggregation behavior, which means the analysis at the end is correct by construction rather than something you have to clean up after.

Guardrails that protect the data before it ever lands. Number validation lets you set minimums and maximums, decimal precision, and currency or custom prefixes and suffixes, and the server rejects anything out of range, so a mistyped “one million dollars per hour” never enters your dataset and skews a percentile. Conditional logic shows or hides questions based on earlier answers, using a full set of comparison operators joined with AND or OR, with wildcards that handle “show this question if any row in that matrix was answered” without you enumerating every row. A visual logic graph maps every visibility rule so you can catch a circular dependency before launch rather than after a respondent gets stuck. Question-level flags let you mark a question required, make it a filter axis for the results page, keep it private, or tag it as demographic so personally identifying fields group into an admin-only table instead of cluttering public charts.

Section and question management. You can add, rename, duplicate, delete, and drag to reorder sections, expand or collapse questions individually or in bulk, and lean on a fifty-step undo and redo history while you work. A bulk edit mode lets you revise many questions’ text in a single column view, and find-and-replace works across the whole instrument, which matters when you decide late that “associate” should read “team member” everywhere.

The five-step launch wizard. Setup is a guided sequence that is hard to get wrong.

1. Basic info: title, open and close dates, the anonymity threshold (five, ten, or fifteen responses), and a link to the predecessor survey, which is what unlocks prefill and trend reporting in later cycles.
2. Template: select an existing one or start blank.
3. Customize: scope this particular survey by including or excluding sections, then confirm it is ready, a step that prevents accidentally launching a half-built instrument.
4. Invites: a structured invitee table plus a CSV upload that dedupes by email.
5. Review and launch: a summary card and pre-launch validation that enforces valid, unique email addresses and at least one invitee.

Throughout, the wizard autosaves on every change and shows who saved it last, so setup can span several stakeholders across several days and the next person can pick up exactly where the last one left off.

Fielding it without friction

The classic survey methods were mail, telephone, and on-site visits, and each forced a tradeoff. Mail was cheap but produced weak comparability and low response. Phone worked only inside small, well-acquainted professional networks. Visits produced the best data but cost so much they could not scale. For decades the comp-survey literature treated this as an unavoidable bind.

Web collection removes the bind, and Sensible Surveys is built to remove friction at every point where a respondent might drop off.

Passwordless access. A first-time respondent clicks a signed magic link and lands directly on their survey. No account creation, no password, no signup form standing between them and the first question. For situations where email delivery is unreliable, or where the respondent list is small enough to reach out personally, an invite-by-link mode skips email entirely and lets you share the link through whatever channel your organization actually uses.

Roster import and gentle follow-up. You upload an invitee roster by CSV and the system dedupes by email and creates the invitations in one operation. Chasing non-responders is a single click per row or a bulk action, and resends are throttled so a reminder campaign never turns into spam.

Autosave on every keystroke. A respondent’s work is written continuously, with a visible saved indicator, so nobody loses a two-hundred-question survey because they closed a tab. That alone changes completion rates on long instruments.

Predecessor prefill, the retention lever. When a survey is linked to last year’s, returning respondents see their prior answers already filled in and visually highlighted, so their job becomes confirming and updating rather than retyping. Returning respondents finish quickly because they are reviewing, not starting over, and quick is what gets a survey finished.

Auto-filtering by employment class. If a respondent identifies early as a union employer, the matrix questions stop showing them the salaried-exempt rows that do not apply. Less friction for the respondent and cleaner data for you, with fewer junk “not applicable” cells to filter out later.

A review step that enforces completeness. The final section is a read-only summary showing each section’s status, and any missing required answers surface in a banner that blocks submission until they are fixed. You stop chasing incomplete responses after the survey closes because the survey will not let them be incomplete.

Respondents also get a simple dashboard of their active and past surveys, and can return after submitting to view exactly what they submitted. The net effect is the data quality people associate with an in-person visit, delivered at the cost and scale of a web form.

From responses to market rates

This is where the classic analytical workflow happens: match the jobs, choose your measure, age the data to a common date, weight the sources, and arrive at a composite market rate. In Sensible Surveys most of it happens automatically once the survey closes.

Anonymity thresholds gate everything, so start there. You set a threshold of five, ten, or fifteen responses per reported cell. Below that count, a cell is either flagged as below threshold or suppressed entirely, and the suppression happens on the server, which means even an administrator’s CSV export respects the rule for any non-admin recipient. No individual respondent can be reverse-identified from a thin cell. The legal reasoning behind this is covered below, but the practical point is that the guardrail is enforced by the system rather than by everyone remembering to be careful.

The compilation pipeline. When a survey closes, the system aggregates the responses, applies the anonymity threshold, and computes percentiles, means, and standard deviations for each occupation at each level, then caches the result. That caching is why the analytics render quickly even on a survey with hundreds of respondents and thousands of data points.

The analytics. The compensation panel is the workhorse, with four views of every occupation: a percentile table (10th, 25th, median, 75th, 90th, mean, and standard deviation), range bars showing the tenth-to-ninetieth-percentile band with the quartile box inside it, distribution curves with a standard-deviation envelope, and year-over-year trends. Around it, histograms show wage distributions with the interquartile band, median, and mean overlaid, and heatmaps handle matrix questions where stacked bars would mislead. Every occupation expands inline to its entry, mid, and senior splits, each with its own range and distribution, so a reader can drill from the headline number down to the level that actually matches their job.

Slicing the data. Any question you mark as filterable becomes an analysis axis: industry, headcount band, region, employment class. Multiple values within a question combine with OR, and filters across questions combine with AND, so you can ask for, say, manufacturing employers with fifty to two hundred staff in a particular county and read just that slice. You can change which questions are filterable after launch without rerunning anything.

Data hygiene. You can mark any response as excluded from results without deleting it. The row stays in place for the audit trail but drops out of every aggregate, which is how you handle a typo outlier, a test submission, or a known bad-faith entry, cleanly and reversibly. Open-text responses are paginated and searchable so a popular free-text question does not slow the page, and they are flagged separately at export time because free text is the most common way personally identifying information leaks out of a dataset.

Export. A subset modal lets you choose which sections to include and independently toggle compensation data and verbatim text. Compensation exports as one row per occupation, level, and statistic, which drops straight into a pivot table for anyone who wants to do their own cut.

There is also a transparency control worth knowing about. With an opt-in toggle, results can include a panel listing the participating organizations. Some economic development organizations want this for credibility; many trade associations turn it off for competitive reasons. The choice is yours per survey.

Year-over-year continuity

Aging data is the part the old guides spend the most effort on, because surveys collected at different moments had to be manually trended to a common date using adjustment factors pulled from salary-increase surveys. It was tedious and error-prone. Sensible Surveys handles continuity structurally through the predecessor link, which turns year-over-year comparison into a property of the data rather than a project.

Returning respondents prefill from last year, the retention lever already described. Successor mapping handles staff turnover: when the person who answered last year has moved on, you map their replacement to them, and the replacement’s form prefills from the prior answers, so your time series stays continuous even as the people filling it out change. Trend reporting then renders automatically. The system walks the chain of predecessor surveys and shows per-occupation median wage growth, response-rate movement, and respondent-count deltas, degrading gracefully along the way: a single year hides the trend views, two years shows year-over-year movement, and three or more years shows compound annual growth. Instead of an aging exercise every cycle, you get a series that compounds in value the longer you run it.

Controlling who sees what

For an association or authority, the finished result is an asset, and you control its distribution precisely.

Per-audience release. You release to respondents, to paid subscribers, and to the public independently, each with an optional scheduled date. You can give respondents the first look as a thank-you for participating, open it to subscribers a week later, and publish a public version a month after that, and you can schedule all of it in advance and walk away. Every release is logged.

Privileged preview. Administrators and managers see compiled results the moment compilation finishes, which makes them the quality-assurance pass before anything reaches an outside audience, and a “view as” control lets them preview exactly what each audience will see.

Paid subscriber access. External paid viewers receive access scoped to the specific surveys you grant them and never see anything else in your account, which is what makes it practical to monetize the data without exposing the rest of your operation. Grants can be created, revoked, and reactivated, and each action is logged.

Per-respondent access control. When you need finer control, you can require explicit per-respondent permission to view results, granted individually or in bulk, with notifications gated so that no respondent is told their results are ready before the results are actually live.

When someone reaches a results page without access, the message is tailored to who they are rather than a bare error, which keeps respondents, subscribers, and internal viewers from confusion about why they cannot see something yet.

Staying defensible

Two legal areas matter for wage surveys, and both have shifted since most comp-survey guidance was written, so this is worth getting current on.

Antitrust. Sharing wage information among competitors has always carried antitrust risk, because a survey can shade into signaling or coordination if it is run carelessly. For roughly three decades, federal enforcement agencies offered an informal “safety zone” that effectively blessed surveys meeting certain conditions: administration by a neutral third party, data at least three months old, at least five participants with no single one dominating, and aggregated results that could not identify any participant. In 2023 that formal safe harbor was withdrawn, and the agencies signaled that they intend to scrutinize information exchanges more closely rather than less, including exchanges run through intermediaries and algorithms.

The important point is what the withdrawal did and did not change. It removed the guarantee, not the logic. Aggregating data, anonymizing it behind response thresholds, using historical rather than current figures, and running the survey through a neutral administrator are still precisely the practices that keep a survey on the right side of the line. They are now best practice and risk management rather than a checklist that buys automatic immunity, which means doing them well matters more than it used to, not less.

This is where the choice of platform actually helps your posture. Sensible Surveys is itself a neutral third-party administrator. Its anonymity thresholds are enforced on the server, so sub-threshold cells never reach anyone who should not see them, including through exports. Results are aggregated by design rather than by discipline. And every meaningful action in the system is logged with who did it and when, so if you ever need to show how the data was handled, the record exists. None of this is legal advice, and you should run your specific survey design past your own counsel, but the architecture is built to support a defensible posture under the current, stricter environment rather than a withdrawn one.

Using survey data as evidence. If your wage data ever appears in litigation, in a prevailing-wage dispute, a pay-equity claim, or expert testimony, it has to clear the reliability bar for expert evidence. That bar tightened at the end of 2023, when the federal rule governing expert testimony was amended to emphasize that the party offering an expert must show, by a preponderance of the evidence, that a reliable methodology was reliably applied to the facts, and that judges are to act as genuine gatekeepers rather than passing questions of reliability through to the jury. The practical implications for a survey are the same in spirit but higher in stakes than before: documented methodology, known sample sizes, a stated measure of dispersion or rate of error, and a clear effective date are what make data defensible. Standard deviations, percentile distributions, an audit trail, and a documented compilation process are exactly those artifacts, and in Sensible Surveys they are produced by the system rather than reconstructed from a spreadsheet months after the fact.

A quick-start checklist

To run a wage and benefits survey on Sensible Surveys:

1. Define the purpose and the market cut you need, by region, industry, and organization size.
2. Choose your benchmark occupations from the standard occupational taxonomy, adding custom codes for any industry-specific roles.
3. Build from a template, using Smart Suggest and the snippets library, and set number validation and conditional logic as you go.
4. Decide your wage and benefits sections, keeping benefits questions structured around plans, contributions, and time off.
5. Set your anonymity threshold, and if this is a recurring survey, link the predecessor.
6. Load your invitee roster by CSV, and choose email invites or invite-by-link.
7. Field it, using throttled resends to chase non-responders while autosave and prefill carry the respondent experience.
8. Close the survey, let the compile run, and use the privileged preview to quality-check results before anyone outside sees them.
9. Slice by industry, size, region, and class, exclude any outliers, and read the percentile and trend views.
10. Release per audience on the schedule you want, and next cycle, map successors so prefill and trend reporting compound the value.

The work that once required a third-party consultant and a season of manual analysis now runs end to end in one place, with comparability, anonymity, and year-over-year continuity built in rather than bolted on. That is what Sensible Surveys is for.