Government statistics have a way of escaping their paperwork. A figure leaves a federal agency’s data file, gets summarized in a press release, condensed into a news story, screenshotted into a social post, and quoted in a meeting — and by the time it reaches an audience, the definitions, caveats, and reference periods that made it meaningful have often been dropped. The number survives; the context does not. That is why anyone who cites government figures, whether a journalist, a student, a policy analyst, a local official, or a curious resident comparing their county to a neighboring one, should build the habit of returning to the primary source.
A primary source in this context is not simply “a website ending in .gov.” It is the specific publication — the table, file, or dataset — issued by the agency that actually collects, compiles, or administers the information, together with the documentation that explains how it was produced. Secondary retellings can be accurate and useful, but they are always one step removed from the record, and each step introduces opportunities for drift.
What Actually Counts as a Primary Source for Government Figures
Primary sources come in tiers, and knowing which tier you are looking at changes how much weight the number can bear. At the most direct level are the agencies themselves: the Census Bureau publishes population and demographic estimates along with margins of error and detailed methodology; the Bureau of Labor Statistics releases employment and price data with explicit definitions of who is counted and who is not; the FBI publishes crime data through the Uniform Crime Reporting program and its successor, the National Incident-Based Reporting System, accessible via the Crime Data Explorer; the Centers for Medicare & Medicaid Services publishes Medicare Advantage and Part D plan information; USAspending.gov records federal awards and spending; the Internal Revenue Service releases filings from tax-exempt organizations; and individual state agencies — treasurers’ unclaimed property programs, comptrollers, health departments — publish their own administrative records.
These are primary not because they are official in tone, but because the publishing agency is the one that received the reports, ran the program, or processed the filings. That distinction matters because the collector knows exactly which definitions apply. An arrest is not an offense, an offense is not an incident, an obligation is not an outlay, a plan’s premium in one county is not its premium statewide, and a filing is not an audit. Primary sources generally state which of these a given figure represents. Secondary summaries frequently do not.
A second tier includes agency-produced summaries, dashboards, and annual reports built on the agency’s own data. These can be treated as primary when they link back to the underlying tables and clearly identify the vintage, though summary dashboards sometimes lag behind the underlying files or apply their own rounding. Below that sit independent compilations, think-tank analyses, news write-ups, and encyclopedic entries — all potentially valuable, but all secondary. Their reliability depends on how faithfully they carried the definitions forward, and readers usually cannot tell without checking the original.
Practically speaking, a citation is only as strong as its weakest link. Naming the agency is not enough if the specific table, year, and geography are unstated. Primary means traceable: a reader should be able to follow your citation to the exact publication and see the number for themselves, in the form the agency released it.
How Secondary Retellings Drift From the Original Data
Drift is rarely dramatic. It accumulates through small, ordinary editorial decisions. A reporter shortens a definition; an infographic drops the footnote about suppressed values; a blog post compares two years that used different methodologies; a slide deck rounds an estimate and then treats it as exact. None of these are sabotage, and any one of them may be harmless — but stacked together, they can turn a carefully bounded estimate into a confident-sounding claim its source never supported.
Several recurring failure modes are worth watching for. Revisions and restatements are routine: agencies update prior-period data when late submissions arrive or processing improves, so a figure quoted from an old article may no longer match the current file. Vintage mismatches cause similar trouble, as when one-year estimates are compared to five-year estimates, or when a ranking blends data from different release cycles. Definitional shifts can be even more consequential; the FBI’s transition from the summary Uniform Crime Reporting system to the incident-based NIBRS changed what is collected and how participation is reported, which affects comparability across places and years. Coverage gaps matter too: not every agency reports, not every record includes every field, and categories such as unknown race or unknown location exist in the data and must be acknowledged rather than silently folded away. Compilations that index US government data by state, county, and city can make it much faster to locate the right table and the right agency link, but they are a starting point, not a substitute for the originating file.
Crime statistics deserve particular care. Arrest counts reflect law-enforcement activity and reporting practices as much as underlying events, and offense data reflect what was reported to police. Reporting coverage varies by agency and year, definitions vary by program, and socioeconomic conditions and policing patterns influence what enters the system. None of that supports claims that any demographic group is inherently more or less inclined toward crime; arrest records document police contact, and responsible analysis says so plainly. The same restraint applies elsewhere: a state unclaimed property database listing a name is not proof of entitlement, and a Medicare plan’s presence in one county in one year says nothing about its availability or cost in the next.
A Practical Checklist for Verifying a Government Figure
Verification is mostly a matter of asking the same handful of questions every time. Start with identity: who published this, what is the publication or table called, and when was it released? Then check the reference period — the year or range the data describe, which is often not the year of publication. Next, read the methodology and footnotes before quoting the number; they typically disclose sampling error, suppression rules, reporting thresholds, and definitional changes that determine whether the figure can be compared to anything else.
Geography and denominators come next. County boundaries, metropolitan definitions, school districts, and ZIP code approximations do not always line up with one another, and a tallied count without a population base or a rate can mislead when places of very different sizes are placed side by side. Confirm the unit as well: counts, rates, estimates, and projections are different objects, and estimates carry uncertainty that should be stated rather than hidden. Where categories are incomplete — unknown race, unreported incidents, unclassified spending — note them instead of assuming they are negligible.
Finally, keep a citation trail and re-check the source at the moment you publish rather than relying on a saved excerpt. For time-sensitive programs, go to the operational source: plan availability and premiums under Medicare Advantage and Part D vary by county and by year, so readers should confirm details in the official CMS Plan Finder during open enrollment rather than trusting a quoted figure. For unclaimed property, search the relevant state program directly and free of charge. For federal awards, distinguish obligations from outlays and check whether the amount reflects an initial award, a modification, or a total. For crime figures, note the program, the participation level, and whether you are citing arrests, offenses, or incidents.
This discipline costs a few extra minutes and pays for itself immediately. When someone challenges a statistic, the difference between “a study said so” and “here is the agency’s table, this is the year, this is the definition, and here is the footnote” is the difference between an argument about credibility and a conversation about the data. Government figures are public precisely so they can be inspected at the source. Citing them well is simply honoring that arrangement — treating the number not as a rhetorical prop but as a documented claim with a paper trail anyone can follow.
Busan robotics engineer roaming Casablanca’s medinas with a mirrorless camera. Mina explains swarm drones, North African street art, and K-beauty chemistry—all in crisp, bilingual prose. She bakes Moroccan-style hotteok to break language barriers.