The Small Sample Problem: When BLS Data for Your Job Becomes Unreliable

BLS OEWS does not publish a wage estimate for every occupation in every metro area. When the survey sample in a given SOC code and metro combination is too small, the Bureau either suppresses the figure entirely or flags it as having a wide relative standard error — and if you happen to land on a published number anyway, that doesn't mean the number is stable. The mechanism behind this is worth understanding before you anchor a negotiation or a relocation decision to a single percentile.

OEWS is a survey, not a census. The Bureau of Labor Statistics samples business establishments and asks them to report wages by occupation, then uses those responses to estimate employment and percentile wages for each SOC code, at the national level and within each metropolitan statistical area. National estimates draw on a huge pool of respondents — the all-occupations row alone represents 155,495,730 jobs in the May 2025 OEWS estimates (published May 2026). But once you slice that same occupation down to one metro, the number of establishments reporting a given job title can shrink to a handful. A small sample doesn't just mean a wider confidence interval BLS quietly tolerates — it means BLS may withhold the estimate, or publish one with real statistical noise baked in that the published table won't dramatize for you.

Why suppression happens at the metro level

BLS suppresses OEWS estimates for two related reasons: confidentiality and reliability. If only one or two establishments in a metro area employ a given occupation, publishing their wage data could effectively reveal what a specific employer pays specific workers — a disclosure risk the Bureau is required to avoid. Separately, even where disclosure isn't the issue, an estimate built from very few respondents carries a large relative standard error, meaning the published percentile could be far from the true metro wage for that job. Neither of these is a flaw in the data; it's the survey being honest about its own limits. The problem is that a suppressed or thin cell doesn't always look different from a robust one to a reader scanning a table.

How to recognize a thin data cell

There's no single number in a public-facing tool that tells you the underlying sample size, but there are signals worth watching for:

• A missing hourly wage next to a present annual wage, or vice versa. In the national detailed-occupation data, for example, Teaching Assistants (SOC 25-9045), Elementary School Teachers (SOC 25-2021), and Secondary School Teachers (SOC 25-2031) are published as annual-only figures — BLS doesn't estimate an hourly wage for these roles at all, a structural gap tied to how the occupation is surveyed, not a reliability failure, but a reminder that OEWS doesn't uniformly report every metric for every job.
• A narrower-than-expected gap between the 10th and 90th percentile. When a distribution looks unusually compressed relative to what you'd expect for a role with real variance — commission-heavy sales jobs, management titles, legal roles — that can be a sign the estimate is smoothed from very few data points rather than reflecting genuine pay compression.
• An occupation that appears at the national level but returns nothing, or a flat single figure, at the metro level. This is the most direct signal: if a role has meaningful national employment but the same SOC code produces no usable metro breakout, the local sample was almost certainly too thin to publish.

None of these signals substitute for BLS's own documentation, but together they should make you slow down before treating a single metro percentile as gospel.

A worked example: why national numbers are the fallback, not the norm

Consider Financial and Investment Analysts (SOC 13-2051). Nationally, OEWS reports a median annual wage of $102,740, with the 25th percentile at $79,290 and the 75th percentile at $133,340 — a wide, credible spread built from a large national sample. That national spread is trustworthy precisely because it aggregates across every metro area with employment in that SOC code, including many where the local sample alone would be too small to publish independently. If you're evaluating an offer in a small or mid-sized metro and the local percentile breakout for that occupation looks unusually narrow or is simply unavailable, the right move isn't to force a local number — it's to fall back to the national distribution and treat it as your best available anchor, explicitly noting that it isn't metro-specific.

This is also where comparing multiple markets side by side matters more than trusting one thin cell in isolation. The highest-paying jobs tool lets you rank occupations within a city or rank cities for a single occupation using published OEWS figures — and because it surfaces the underlying BLS numbers rather than smoothing them, a suppressed or missing metro cell will show up as an absence rather than a fabricated estimate. That absence is itself useful information: it tells you the local sample didn't support a reliable figure, which is a more honest answer than a confident-looking number with no data behind it.

If you want to sanity-check a specific percentile before negotiating, the salary percentile lookup shows the same OEWS percentile bands this guide draws from, and pairs well with the reasoning in Why Your Salary Negotiation Should Start with BLS Data, Not Glassdoor — a defensible number only holds up if the sample behind it does too. It's also worth reading alongside The Metro Area Trap, since a metro boundary that bundles in low-employment counties is one of the ways a thin sample gets hidden inside a seemingly normal-looking MSA figure.

None of this is professional financial or legal advice — it's a guide to reading a federal statistical product correctly. When the sample is thin, the honest move is to say so, use the national figure as a labeled fallback, and treat any single metro percentile in a rare occupation with real skepticism.

Frequently asked questions

How do I know if BLS suppressed data for my occupation and metro?

The clearest sign is a missing wage figure where a national or larger-metro equivalent exists; BLS's OEWS methodology documentation explains that estimates are withheld when publishing them would risk disclosing individual employer data or when the relative standard error is too high to be reliable.

Should I ever use the national wage figure instead of a metro figure?

Yes — when a metro-level estimate for your SOC code is unavailable or looks unstable, the national OEWS figure, such as the $102,740 median for Financial and Investment Analysts (SOC 13-2051), is a legitimate fallback as long as you label it as national rather than local.

Does a small sample mean the median is wrong?

Not necessarily wrong, but less certain — a median built from a thin sample can shift meaningfully in the next survey cycle, so it deserves less confidence than a median built from a large metro sample like the ones behind national major-group figures.

Why do some occupations only have annual wages and not hourly wages?

BLS OEWS publishes annual-only estimates for certain occupations — Teaching Assistants (SOC 25-9045), Elementary School Teachers (SOC 25-2021), and Secondary School Teachers (SOC 25-2031) among them — reflecting how those roles are structured and surveyed, not a data quality problem.

Is this guide giving me career or salary advice?

No — this is informational only, explaining how a federal data product is built and where its limits are, so you can read percentile figures correctly rather than over-trusting a single number.

Informational only, not professional or financial advice.