Why Power Outage Maps Show Different Numbers: A Research Guide
Customer units, geographic precision, timestamps, missing coverage, and repeated observations explain why outage maps and historical charts can disagree.
Two outage maps can show different totals without either containing an arithmetic error. They may cover different utilities, use different observation times, or measure different geographic units. A serious comparison first establishes what each number represents. Only then can it distinguish a real inconsistency from a difference in scope.
Key findings
- An affected customer account is not automatically one affected person.
- Missing observations and confirmed zero-valued reports must remain distinct.
- Repeated snapshots should not be added as though each represents a new set of customers.
Identify the unit and the observation time
Oak Ridge National Laboratory’s historical EAGLE-I description presents county-level observations at 15-minute intervals and explicitly discusses changing coverage and data-quality assessment. Its published coverage figure describes that historical dataset, not this website or every present-day feed. The important research principle is to examine a dataset’s own geography, time resolution, and coverage documentation before comparing its totals. [1]
A utility account, meter, incident, and person are different units. One incident can affect many meters, while a building account can represent more than one occupant. If a source reports 500 affected accounts, reporting 500 physical failures or 500 affected people would require information the account count does not supply.
Missing is a data state, not a zero
Our site selects one authoritative count source per utility and separates delayed observations from fresh totals. An inventory entry establishes that a utility exists or participates in reporting; it does not establish a complete public observation for every county. That is why a provider can appear in the directory while its current status remains unknown.
An illustrative comparison makes the consequence clear. Map A observes two providers and reports 300 accounts. Map B observes only one and reports 250. The difference could reflect coverage rather than conflicting measurements. A valid reconciliation would compare provider-level records at aligned times before attempting to identify an error. These values are hypothetical.
A time series does not count unique customers
ORNL’s historical-data portal distinguishes downloadable historical data from near-real-time situational awareness and documents the dataset and its attribution. A research workflow must preserve which product was used. A live county feed and a historical export can have different update schedules and processing histories, even when associated with the same broader research program. [2]
Suppose a synthetic series reports 100 accounts out at 1 p.m., 100 at 1:15 p.m., and 40 at 1:30 p.m. Adding those values gives 240 account-observations, not 240 unique interrupted customers. Multiplying a count by its applicable time interval can estimate customer-time under explicit assumptions; unique customers require information about identity and overlap.
Separate peak, duration, and restoration
EIA’s 2018 explanation of distribution reliability treats interruption frequency and duration as distinct measures and notes that service experience differs among distribution systems. A live affected-account count is therefore not a substitute for an annual reliability statistic. Peak impact, average interruption duration, and an individual restoration estimate answer different questions. [3]
Our calendar history reports observed peaks, sampled days, and days with positive reports. It does not add repeated polling values into an outage-event count. A falling total also cannot prove that a particular household was restored: other customers could have become disconnected between observations. A defensible interpretation preserves that uncertainty rather than equating net change with identified restoration.
A compact method for reconciling two maps
Write down the unit, provider coverage, geographic boundary, source reporting time, last fetch time, and treatment of stale or future planned records for each map. Compare the overlapping providers first. If the values still differ, examine the underlying records and whether the source changed between observations.
This procedure turns a vague disagreement into a set of testable questions. It also shows when a comparison cannot be completed because one publisher lacks adequate metadata. The absence of enough evidence is itself a result; it is not a reason to invent a city total, assume complete coverage, or fill a missing interval with zero.
Sources & methodology
This methodological article combines published dataset descriptions with the documented behavior of this site. Numerical examples are synthetic and illustrate accounting principles; they are not measured outages. Historical coverage claims are not transferred to this platform’s current feed coverage.
- Recorded county electricity outages, 2014–2022: dataset descriptionBrelsford and colleagues / Oak Ridge National Laboratory · 2024 publication
- EAGLE-I historical-outage dataset portalOak Ridge National Laboratory · Portal metadata reviewed October 23, 2026
- Frequency and duration of distribution interruptionsU.S. Energy Information Administration · April 5, 2018
Explore current outage information
The historical figures above are separate from current reported outages. Available coverage and reporting timestamps appear on these pages.