Methodology & Data Sources

Data sources

• ORNL EAGLE-I — 15-minute county-level power outage snapshots, 2014–2025, published by Oak Ridge National Laboratory / Dept. of Energy. Licensed CC BY 4.0.
• NOAA Storm Events Database— used to attribute a county's worst outages to named storms (hurricanes, ice storms, tornadoes) where one occurred.
• US Census Bureau (ACS) — county household counts, used as one of three independent denominator sources (see below).

How outage events are reconstructed

Raw EAGLE-I data is a snapshot every 15 minutes of how many customers are without power in each county. We reconstruct discrete outage "events" by requiring a county to cross a significance threshold (scaled to county size) and then chaining together strictly contiguous 15-minute snapshots — a gap of even one interval ends the event. This method was validated against a published reference dataset for Hurricane Uri (Texas, Feb 2021) and reproduced it almost exactly.

Reliability Grade & the customer-count problem

The Reliability Grade (A–F) and "percent of customers affected" figures require knowing how many customers actually live in a county — and no single published source gets this right for every county. We reconcile three independent estimates (a customer-count file, embedded per-row totals, and Census household counts) and take the largest, which biases toward understating rather than overstating outage severity.

A small number of counties are excluded from grading entirely rather than shown with a misleading grade: counties where even the reconciled estimate is implausibly exceeded by observed outage data (a known over-attribution artifact in the source data during major storms), and counties with no usable customer-count estimate at all (mostly geographies whose boundaries changed during the 2014–2025 window, e.g. Connecticut's 2022 switch from counties to planning regions).

As an independent check, we compared our state-level severity rankings against the U.S. Energy Information Administration's own 2024 utility reliability filings (SAIDI) — the two rank almost identically (Spearman correlation 0.98 across 51 states).

Known limitations

• Data covers 2014 through the last full year (currently 2025) — not real-time. We refresh annually.
• EAGLE-I coverage is roughly 92% of US customers; some rural counties have patchier reporting.
• A small number of counties don't yet have a reliability grade for the reasons above — this is disclosed on their page rather than hidden.