|
Federal public health data is getting leaner. Data sets once published on a regular schedule are being delayed, changed, or quietly taken offline. For reporters, the work now starts with finding the data before any analysis can take place.
Quieter deletions, frozen data
In the first months of the Trump administration, thousands of web pages from the Centers for Disease Control and Prevention vanished overnight, Qin said. Today, the losses are harder to spot.
“A lot of it is just more quietly going down. It’s not going down en masse,” she said.
Some removals are one-offs. For example, mortality reports for people who died in ICE detention were recently taken down from a government website without notice, Qin said.
Other data simply stops arriving as annual data sets are delayed or discontinued after staffing cuts. Qin said that after the CDC’s reproductive health division lost most of its staff, federal data on abortion and maternal health became “just kind of frozen in time.”
Survey questions are disappearing, too. After a Trump executive order on gender, questions about gender identity and sexual orientation were removed from many federal surveys. Unless that policy is reversed, Qin said, those years of data will contain no information on sexual orientation.
“Data is accountability,” Qin said. “Stonewalling reporters when we ask for data is a form of weaponizing it.”
Data as accountability
Public data helps people make informed decisions about their health. It also gives reporters evidence to “hold powerful institutions to account,” Beckler said, adding that the absence of data can be just as meaningful as its presence.
Beckler’s investigation into C-sections for Business Insider shows what’s at stake. Academic research shows the hospital where a patient delivers is one of the strongest predictors of whether she will have a C-section. Yet, most states don’t publish C-section rates by hospital.
“Even if it becomes this really important piece of information that folks need to make an informed choice about the health of their pregnancy, [the patients] couldn’t get that data,” Beckler said.
So she asked for raw numbers of C-sections from all 50 states and Washington, D.C., then created a searchable map of low-risk C-section rates at more than 1,700 hospitals. The map showed “wildly different outcomes in hospitals that are only separated by a mile or less,” she said. Lowering C-section rates goes against a hospital’s inertia, Beckler explained, and without oversight, the rates can creep up.
Getting the data
When federal data dries up, state agencies can fill the gap. As the federal government doesn’t track C-section rates hospital by hospital, Beckler went to the state instead. Because she knew that states reported diagnosis codes to the CDC, she also knew that the raw data existed even when states didn’t publish it in the form she needed.
A public records officer is usually her point of contact, Beckler said, but is rarely the person who directly handles the data. So she asks for a call with the staff who do.
“The question I’m usually asking is, can we just discuss the underlying data, so I understand how it’s collected and collated and produced,” she said. Agencies are often open to that, she said, because “it’s not in their interest for me to analyze data incorrectly.”
When agencies worried about releasing protected health information, Beckler offered privacy protections up front and only asked for data sets large enough to avoid concerns about patient privacy. She also learned that some states steer reporters away from public records requests toward a separate data request track, which can come with strict conditions, such as academic credentials, a siloed computer, or a promise that only one person in the newsroom will see the data.
At the federal level, reporters should be ready to sue. The Guardian’s request to DHS stalled as the agency “blew every single deadline,” Qin said. The Reporters Committee for Freedom of the Press sued on the Guardian’s behalf, pro bono, and the lawsuit produced the records.
Even then, no one at the agency explained the data. A footnote in the spreadsheet DHS released said a medical claims processing agreement with Veterans Affairs had ended, leaving no miscarriage data after October 2025. Further reporting showed ICE had terminated that contract.
Beckler said she once got expedited processing from DHS, and it still took almost two years to receive anything. “It does seem like you have to sue to get really any data out,” she said, noting that groups such as the Deportation Data Project are also turning to the courts.
Qin suggested a shortcut before filing: Search CourtListener or PACER for FOIA lawsuits on your topic, and borrow the language of requests others have already sued over.
Keep people at the center
Data stories can turn technical fast. Qin said her editor keeps asking one question: “Why does this matter for your average person?”
Her answer is to trace the “trickle-down impacts” of a data set going dark. For a story about cuts at the federal education statistics office, her team asked researchers what they could no longer study. For example, one lost survey had followed a group of children from pre-K through fifth grade, which would have been the first such group surveyed since the COVID-19 pandemic.
Beckler said to look for critical gaps, then ask who is harmed. In her data center series, she used models from the Environmental Protection Agency to put a dollar figure on air quality impacts. That showed how data centers often land in communities already burdened by pollution, as in Memphis, Tennessee.
The Guardian investigation chose to open with women who had miscarriages in detention, not with a spreadsheet. “This is a story about missing data, yes, but it’s ultimately a story about the women who are experiencing these conditions,” Qin said.
How to know when data goes missing
No single system tracks deletions. Qin, who said she keeps hundreds of web monitors running, called her approach “a hodgepodge of different methods.” These are the tools the panelists named:
-
Klaxon, a free tool maintained by MuckRock and The Marshall Project, watches specific parts of a web page. It sends a Slack or email alert when something changes and saves a snapshot each time it checks. Visualping is another paid tool that works much like Klaxon.
-
The Internet Archive’s Wayback Machine lets reporters compare versions of a page to see what language changed and what was deleted. The Guardian used it to document how the Consumer Financial Protection Bureau removed 10 years of press releases.
-
Some agencies keep public GitHub pages where programmers note what they are changing, sometimes before something is deleted.
-
CourtListener and PACER show the text of FOIA requests that others have sued over, plus documents surfaced in discovery.
-
Python scripts can now be written with plain-language AI tools, Beckler said, but also suggested caution: “You really have to know how to read Python, and you have to understand how to interrogate your data and know its limitations.”
- Your own archive may be the most reliable. Download and save a data set as soon as you use it.
Sources still matter most. When a source told Beckler that the Environmental Justice Index, a federal data set she used investigating data centers, was going to disappear within 24 hours of her using it, she was able to pull and archive the data she needed before it went down.
Watch the webinar
| 
