Suppressed cells: why some wages are blank

Some tables on this site have gaps: a state row with employment but no wages, a metro missing from an occupation’s list, a 90th percentile with nothing in it. These are not errors in the site’s pipeline. They are cells BLS chose not to publish, and the reasons are worth knowing.

Three markers

The OEWS files use three symbols where a number would otherwise be. A single asterisk means a wage estimate is not available. A double asterisk means an employment estimate is not available. A hash sign marks a wage at or above the top code, which is a different case covered in top-coded wages. The first two are suppression.

Why an estimate is withheld

Confidentiality. BLS is bound by law not to publish anything that would identify an individual establishment’s data. Where an occupation in an area is dominated by one or two employers, publishing its wage would reveal what those employers pay, so the cell is withheld. This is why a specialized occupation in a small metro is often blank even when the workers clearly exist.

Reliability. Estimates with too large a relative standard error, or based on too few responding establishments, fail BLS’s publication standards and are withheld. A wage can fail while employment passes, which produces rows with a headcount and no pay.

Annual-only and hourly-only occupations. For occupations that do not work a standard 2,080-hour year, BLS publishes either an annual wage or an hourly wage, and the other row carries asterisks. Teachers, pilots and athletes have only annual figures; actors and musicians have only hourly ones. This is a reporting convention, not suppression, but it looks the same in the file; see annual versus hourly figures.

What this site does with a blank

Nothing. The family rule is that suppressed stays suppressed. A withheld median is shown as not published, with the marker, and the page says why where it can. The site does not interpolate a missing state figure from neighboring states, does not average metros to fill a state, and does not carry an older release’s value forward. A page that would consist mostly of blanks is not built at all; the publish gates require a published median and employment, and for occupation-by-state pages also the 10th and 90th percentiles, so the thin cases stay as rows in a larger table rather than becoming pages.

Reading around a gap

If a state’s median is blank for an occupation, the metro rows for that state may still be published, and the national row always is. The occupation page’s state table shows which states have a figure. If employment is published but wages are not, the employment is still informative: it says the job exists there in measurable numbers.

Be careful with rankings. A list of the ten highest-paying states for an occupation is a list of the ten highest among states with a published figure. A state whose wage is withheld could belong anywhere in it. The state pages note how many occupations are published for the state so the denominator is visible.

The scale of it

Suppression is concentrated where sample sizes are small: nonmetropolitan areas, small metros and small occupations. Nationally, almost every detailed occupation has a full row. At the state level most do, and at the metro level the share of blank cells rises quickly with the specificity of the occupation. The site’s data documentation records the counts for each release.

A blank cell is information. It says the question has an answer BLS could not publish reliably or safely, which is a more honest statement than a number invented to fill the space.