Why salary websites disagree with OEWS

Search for a job title’s pay and several numbers appear, often thousands of dollars apart, one of them from BLS. The disagreement is not a sign that one source is wrong. It is what happens when different methods measure different things and call them all “salary.” This page lists the differences that account for most of the gap, using the occupation tables here as the reference point.

Who supplies the number

OEWS asks employers. BLS and the state workforce agencies draw a sample of establishments from unemployment insurance records and ask each one how many people it employs in each occupation and what it pays them; the Handbook of Methods states that exact wage rates are received for federal, postal and most state, local and private-sector records, with a small share reported in wage intervals. The sample over a three-year cycle covers about 1.1 million establishments and roughly 57 percent of United States employment.

Most salary websites ask workers, or read job postings. A figure built from visitor submissions reflects whoever chose to submit, and a figure built from postings reflects what employers advertise for open positions, which is not the same population as everyone currently in the job. Neither is a probability sample of the occupation, so neither carries a sampling error that can be stated.

What counts as pay

OEWS wages are straight-time gross pay. The Handbook of Methods lists what is in: the base rate, cost-of-living and guaranteed pay, hazard pay, incentive pay such as commissions and production bonuses, and tips. It also lists what is out: overtime pay, back pay, jury duty pay, severance, shift differentials, nonproduction bonuses, tuition reimbursement and the employer’s cost of benefits. A website reporting “total compensation” or “base plus bonus” is adding items BLS deliberately leaves out, and for jobs where bonuses and equity are large the difference can exceed the base wage itself. See how OEWS wages are measured.

How jobs are grouped

OEWS classifies every job into the Standard Occupational Classification by the work performed, not by title, education or pay grade. A website groups by the titles people type, so “software engineer” on a website may blend what the SOC splits across several codes, and a company-specific title such as “member of technical staff” may have no counterpart at all. Find your SOC code and how BLS defines occupations cover the mapping. The first thing to check when two numbers disagree is whether they describe the same set of jobs.

Which workers

OEWS covers wage and salary workers in nonfarm establishments. It omits the self-employed, owners and partners in unincorporated firms, household workers and unpaid family workers, and it does not survey military occupations. For occupations with heavy self-employment, a website that includes freelancers or practice owners is describing a different population.

Which period and which place

Every OEWS estimate carries a reference month, May of the release year, and is built from six panels collected over three years with earlier panels’ wages adjusted forward. It is a stable picture of a period, not a reading of this month’s offers. A posting-based figure can move with the market in weeks. Place matters too: the OEWS national median blends every state and metro, and a website’s figure may be weighted toward wherever its users are. The state tables and metro tables exist so that comparisons can be made at the same geography.

Which statistic

OEWS publishes a mean and five percentiles for each cell. Many websites publish a single “average” without saying whether it is a mean, a median or a midpoint of a range, and the choice moves the figure, as mean versus median shows. A range built from the OEWS 10th and 90th percentiles is wide for most occupations; a single number inside it is not more precise, only less informative.

Using both

A salary website can show something OEWS cannot: a named employer, a recent posting, or a total package. OEWS can show something a website cannot: a consistently defined, sampled, error-measured distribution for a defined occupation and area, comparable across the whole country. On this site every figure is the BLS figure, printed with its reference month and its suppression and top-code markers, and nothing from any other source is mixed in.