How the numbers work
1. The outage measures
Utilities report two standard reliability indices each year on the U.S. Energy Information Administration's Form EIA-861 (Reliability file):
- SAIDI (System Average Interruption Duration Index): total minutes without power per customer over the year. We call it "outage time".
- SAIFI (System Average Interruption Frequency Index): number of sustained outages per customer. We call it "outages".
- Both come with major event days (all events: what customers actually lived through) and without them (a "normal year"). Major event days are days so far outside the utility's usual record, typically big storms, that the IEEE 1366 standard sets them aside.
- Typical outage length (CAIDI) = SAIDI ÷ SAIFI.
We use the IEEE 1366 figures where a utility files them and its "other standard" figures otherwise (labelled on utility pages). Momentary blips under five minutes are not counted. The data covers 2016–2025; 2025 is the latest final year, released by the EIA in October 2026.
2. Utilities
922 utilities filed reliability data for at least one of 2023–2025 and have a page. Small utilities that file the EIA's short form (Form EIA-861S) do not report reliability, so they appear on county pages without figures. A utility that operates in several states files separately for each; its page shows each state.
3. Matching utilities to counties
The EIA-861 2025 Service Territory file lists every county each utility delivers power in. It does not say how many customers the utility has in each county, so we estimate it: each utility's customers in the state (from the Sales to Ultimate Customers file, bundled plus delivery-only service, or the short form) are spread across the counties it serves in proportion to county population (Census Vintage 2025 estimates). Delivery-only utilities missing from the sales file use the customer count on their reliability filing.
A county's outage time is the customer-weighted average of the outage times of the utilities serving it (each utility's statewide figure, not a county-specific one, which the EIA does not publish). This is an estimate. It is most reliable where one utility serves almost everyone and least reliable where a county is split between a big utility and several small ones. Each utility's latest filing from 2023 to 2025 is used. We drop filings that contradict themselves: outage time including major events lower than the figure excluding them (2 filings, both Nashville Electric Service, 2024 and 2025, which report 0 minutes with major events).
Coverage: the share of a county's estimated customers whose utility filed reliability data. Counties under 50% coverage (242) show a warning, are not ranked and are not indexed by search engines. 80 counties have no filing utility at all and show only their utilities and the state record.
Special cases: Connecticut's Census estimates now use planning regions, but EIA still files utilities under the eight historical counties, so we keep those counties (without population). Five Alaska areas filed under pre-2019 names are mapped to their successor areas. U.S. territories are left out.
4. Ranks and averages
Ranks use 2025 outage time including major events: 1 = fewest minutes. A county is ranked only if utilities with 2025 filings cover at least half its estimated customers (so a county whose main utility's latest usable filing is older is shown but not ranked). 2,784 counties are ranked. State and U.S. averages are customer-weighted across every utility that filed for that year (the U.S. figure for 2025 is 4 h 23 min). They match the EIA's own state totals closely; small differences come from the EIA's handling of utilities filing under both standards.
The reading-guide labels on each page compare outage time with the U.S. average for the same year: under half is "far fewer outages than average", up to 90% "better than average", 90–110% "about average", up to twice "worse than average" and more than twice "far worse than average". They are our own bands, not an EIA scale.
5. Grid emergencies (OE-417)
Utilities, grid operators and others must report major disturbances to the Department of Energy on Form OE-417: loss of service to 50,000 or more customers, large load shedding, physical or cyber attacks and more. We read the DOE's annual summary tables for 2016–2025 and attribute each report to every state named in its "area affected". Weather-related means the event type mentions weather, storms, wind, ice, heat, fire, flood or similar. Customer counts are as filed; we ignore obviously impossible entries (more than 10 million customers) and only add a report's customers to a state total when it names a single state. Reporting criteria widened over the decade, so counts are not strictly comparable between early and late years.
6. Backup power calculator
Appliance wattages, the refrigerator duty-cycle rule (plugged-in hours ÷ 3) and the motor-start rule (about 3× running current) come from Virginia Cooperative Extension, "Estimating Appliance and Home Electronic Energy Use" (2901-9014). Our own assumptions: 85% of a power station's rated capacity is usable after losses, and a generator should run at no more than 80% of its running rating. The examples on county pages use the county's typical outage length (outage time ÷ outages).
What this cannot tell you
Utility averages hide large differences between streets: rural lines, tree cover and distance from a substation matter a lot. Utilities measure outages with different systems, and some estimate rather than record them automatically. None of this is a forecast. For a current outage use your utility's outage map; for emergency planning see Ready.gov.
Refresh
The EIA publishes a new year of EIA-861 data each autumn; we rebuild every page then. OE-417 summaries are updated by the DOE through the year.