The data that isn't there.

The question

Most data stories start with what's in the dataset. This one starts with what's missing. When the power goes out, people stop running speed tests, phones stop streaming, and home routers go dark. The absence of data becomes a signal of its own.

I live in Houston, so Beryl wasn't abstract for me. The question I wanted to answer: using only public data, can you see when an outage hit and how quickly connectivity came back?

What I looked at

Public sources, each measuring something different:

• Active probing from IODA, Georgia Tech's Internet Outage Detection and Analysis project. Every ten minutes it pings blocks of internet addresses and counts how many answer.
• Routing data (BGP), also from IODA. This is the internet's address book: which networks say they can be reached.
• Speed test counts from Ookla's open data, which comes in quarterly tiles across Texas, published separately for fixed broadband and mobile.

To keep myself honest, every comparison lines 2024 up against the same weeks or quarters of 2023, a year with no hurricane. July is peak travel season in Houston, so a dip on its own doesn't prove much.

The internet noticed

On a normal July day, about 76,000 blocks of Texas addresses answer IODA's pings. Then Beryl came ashore near Matagorda Bay around 4 a.m. on July 8.

By midmorning, about one in ten of those blocks had gone quiet, and it stayed that way for a day and a half. Recovery was slow and steady: back to 95% by Wednesday afternoon, 98% by Friday night, and 99% the following Monday, a full week after landfall. The same days in 2023 barely moved.

The address book didn't flinch

Here's the part that made this story click for me. The routing data hardly budged the whole week. The networks were still telling the internet they existed, even while the homes and businesses behind them were dark.

That's the data that isn't there. If you only watched routing, Beryl would look like a quiet week. You only see the outage when you knock on the door and nobody answers.

The speed tests went up?

This is where I expected the clearest signal, and got a plot twist instead. In the 11 counties around Houston, people ran 12% more fixed broadband speed tests in Q3 2024 than in Q2. That sounds like Beryl: the power comes back and everyone checks whether their internet survived.

Then I lined it up against the rest of Texas and the year before. The rest of the state swung by about 9 points between the two years too, and Houston always gains on the rest of Texas from Q2 to Q3. Once you account for both, Houston's extra growth in 2024 is less than one point.

So I split the Houston numbers by speed. If Beryl had left connections slower for weeks afterward, you'd expect more tests landing in the slow tiles.

The opposite happened. Tests in the slowest tiles (under 25 Mbps) fell 11% from Q2 to Q3 in 2024, faster than the 3% drop a year earlier. All the growth was at the fast end, where tests in 500+ Mbps tiles jumped 57%. That looks like the usual story of networks getting faster.

A week-long outage simply gets lost in a three-month total. Quarterly tiles are great for comparing places and the wrong tool for timing a storm.

Phones picked up the slack

All of that was fixed broadband: home and office connections. Ookla publishes mobile tiles separately, and when the power is out, the phone is often the only thing still online. If people moved to cellular at scale, you'd expect more mobile tests and slower mobile speeds around Houston.

That's exactly what shows up, and it survived the same quarterly blur that hid everything else. Every big Texas metro got faster on mobile from Q2 to Q3 2024. The Houston area went backwards, down 6%, after gaining 11% over the same months in 2023. Harris County on its own fell 7%. Mobile tests in the Houston area also rose 21%, though tests rose in other metros too, so I'd lean on the speeds more than the counts.

One honest caveat: this data can't tell congestion apart from damage. More people leaning on cellular slows things down, and so do cell sites knocked out by the storm. Both probably happened. Either way, the cellular network around Houston was under strain.

What I took from it

Match the resolution to the question. Ten-minute data showed exactly when Beryl hit and how long recovery took. Quarterly fixed broadband data couldn't tell you a hurricane happened at all, but quarterly mobile data could, because the shift to phones was big enough to survive the averaging.

Absence is a signal, but only if you're measuring often enough to notice it. And "the network is up" and "people are online" are two very different statements. Routing said everything was fine. The pings said one in ten weren't home. And the phones were working overtime.

Sources

• IODA (Internet Outage Detection and Analysis), Georgia Tech. Active probing and BGP signals for Texas, July 2023 and July 2024, pulled from the IODA API. Dashboard: ioda.inetintel.cc.gatech.edu
• Speedtest® by Ookla® Global Fixed and Mobile Network Performance Maps, fixed and mobile tiles, 2023 and 2024. Based on analysis by Ookla of Speedtest Intelligence® data. CC BY-NC-SA 4.0. github.com/teamookla/ookla-open-data and the AWS Registry of Open Data
• County boundaries: US Census Bureau cartographic boundary files, via plotly/datasets
• Code: every script and chart for this story is in data-stories/beryl