Government Data Sources for Journalists: Where to Start

Why do government datasets belong in your toolkit?

Government agencies publish datasets, and Data.gov lists them in a catalog ^[1]. That catalog does not control the quality of the data it lists ^[1]. So a dataset is a starting point for reporting, not a finished fact.

IRE reported from its NICAR25 conference that many government datasets can be downloaded for free as Excel or CSV files. They can be localized to any city, county or state in the United States ^[2].

For local numbers, the Census Bureau describes the American Community Survey (ACS) as the only source of local estimates for most of the 40-plus topics it covers ^[3].

Numbers are not the story, though. Speakers at NICAR25 put it this way: your data story is not about data, it is about people ^[2]. Use a dataset to find and test a story. Then go talk to the people in it.

This guide covers three starting points: Data.gov, Census Bureau data, and the Bureau of Labor Statistics (BLS) API.

What is Data.gov, and what is it not?

The U.S. General Services Administration launched Data.gov on May 21, 2009. It worked with the Office of Management and Budget and other agency partners ^[1].

Data.gov is a metadata catalog. It displays information about datasets that individual agencies publish and maintain ^[1]. It points you to data. It does not hold the data.

Data.gov says it does not host or store the underlying data. The files live with the publishing agency ^[1].

It also says it does not control the quality, accuracy or availability of the data itself ^[1]. That puts the vetting on you. Check each dataset with the agency that publishes it ^[1].

Federal agencies must, by statute, create and maintain comprehensive metadata inventories. A central federal catalog run by GSA harvests them ^[1].

What if a link is broken or the data looks wrong?

Data.gov directs you to the dataset point of contact for a broken link, incorrect metadata or a data-quality problem. You will find that contact in the sidebar of the dataset page ^[1].

Start with that contact when a number looks off. Write down the name and the date you asked, in case you need to show your work later ^[1].

What about state and local data?

Many states, counties and cities keep their own open datasets and open data websites. The OPEN Government Data Act requirement applies to federal agencies ^[1].

Non-federal datasets reached through Data.gov have different terms of service and licenses than federal data. Read the terms for each one before you republish anything ^[1].

Where do you find Census Bureau data?

For the 2025 ACS 1-year release, the Census Bureau says data products on data.census.gov, the API, summary files and other tools become public by 10 a.m. ET on Oct. 29 ^[3]. Those are the main places to look.

Video from U.S. Census Bureau (YouTube). Open at youtube.com. Content from youtube.com loads only if you click.

The Census Bureau describes the ACS as the only source of local estimates for most of the 40-plus topics it covers ^[3]. For local estimates on those topics, start with the ACS.

The Census Bureau states that all ACS data are estimates ^[4].

Those estimates come from a sample of the population in the United States and Puerto Rico. They do not come from the whole population ^[4].

How do you read ACS estimates without getting burned?

Keep three habits. Read the margin of error. In general, the Census Bureau advises using ACS for characteristics such as percents and rates rather than population totals ^[4]. Compare like with like.

Read the margin of error

The Census Bureau publishes a margin of error (MOE) for every ACS estimate. You can use it to judge how reliable a figure is ^[4].

Before you write that a number rose or fell, look up the MOE for both figures. Then say "about" or "an estimated" in your copy ^[4].

Use the right kind of number

In general, the Census Bureau advises using the ACS for population characteristics rather than estimates of population totals. Characteristics include percents, means, medians and rates ^[4].

So "the share of renters in your county" is a better fit for the ACS than "how many renters live in your county." ^[4]

Compare like with like

The Census Bureau sets clear rules for comparisons:

• Compare 1-year estimates only with 1-year estimates. Compare 5-year estimates only with 5-year estimates ^[4].
• Do not compare overlapping 5-year datasets. Its example: 2005-2009 against 2006-2010 ^[4].
• Do not compare the experimental 2020 ACS 1-year estimates with any other data ^[4].

Check whether your place is covered

Not every place gets a 1-year estimate. For the 2025 ACS release, the publication threshold is a population of 65,000 or more for 1-year estimates. For the 1-year Supplemental Estimates, it is 20,000 or more ^[3].

Check those thresholds before you plan a story around 1-year numbers for a small town ^[3].

What other agency datasets can you use?

Census is only one source. Federal agencies publish and maintain their own datasets, and Data.gov catalogs information about them ^[1].

The BLS runs a public data API. Version 2.0 requires registration and lets users access more data more frequently ^[5].

Version 1.0 is more limited. It does not require registration and is open for public use ^[5].

For a quick look, version 1.0 may be enough. If you need more data more often, version 2.0 offers more access, and it requires you to register ^[5].

Do not forget state and city portals. Check the terms and licenses on each one ^[1].

How do the Census and BLS APIs work?

Both agencies run APIs, and each has its own access rules ^[6]. ^[5]

The Census Data API requires an API key on all data queries ^[6]. A single query can include up to 50 variables ^[6].

So request a key before you start. Then keep your variable list at 50 or fewer, or split it across several queries ^[6].

For BLS data, decide which version fits your needs. Version 2.0 asks you to register and gives you more data more frequently. Version 1.0 does not require registration but is more limited ^[5].

Video from U.S. Census Bureau (YouTube). Open at youtube.com. Content from youtube.com loads only if you click.

Save every query you run, so you can show how you got each number ^[6].

Step by step: from story question to a verified number

1. Write your question in one sentence. Name the place, the group and the time period.
2. Search Data.gov to see which agencies publish data on your topic ^[1].
3. Get the files from the publishing agency, since Data.gov does not host them ^[1].
4. Note the dataset point of contact in the sidebar. You will need that contact later ^[1].
5. For ACS data, confirm whether you have a 1-year or 5-year estimate, and record the margin of error ^[4].
6. Compare only matching datasets. Do not compare overlapping 5-year periods, as the Census Bureau advises ^[4].
7. Contact the dataset point of contact. Ask whether you are reading the table correctly ^[1].
8. Find real people the number describes. Their stories carry the piece ^[2].

When you contact an agency, you may need to explain who you are. If it helps, see how a press credential can document your work. An NPA credential is not government-issued. It does not give you access or legal rights.

How do embargoes and release schedules affect your deadline?

Some releases come with an embargo. The Census Bureau announced one for the 2025 ACS 1-year estimates. Embargo subscribers get access from Tuesday, Oct. 27 at 10 a.m. ET until Thursday, Oct. 29 at 12:01 a.m. ET. ^[3]

Public availability follows by 10 a.m. ET on Oct. 29 ^[3].

The Census Bureau states that wire and distribution services are prohibited from distributing embargoed news stories and data files before the public release date and time ^[3].

Read each embargo notice in full and follow its terms. Plan your reporting so your verification is done before the release window opens ^[3].

If you work alone, you can join the National Press Association and apply for membership. Membership does not give you legal rights or special access to data.

What should you check before you publish a data claim?

Run this checklist on every data-driven sentence:

• Did you get the data from the publishing agency, not just the catalog listing ^[1]?
• Did you check the dataset with the agency's point of contact ^[1]?
• Is the number an estimate? If so, did you report that and check the margin of error ^[4]?
• Are you comparing 1-year to 1-year, or 5-year to 5-year, with no overlapping periods ^[4]?
• Does your place meet the publication threshold for the product you used ^[3]?
• If the data is non-federal, did you read the license and terms ^[1]?
• Did you talk to at least one person the number describes ^[2]?

If a step fails, say so in your story or hold the claim. Check your state's law and your publisher's rules before you republish data from a state or local portal ^[1].