Removing Major Event Days Takes 2024's Outage Average From 662.6 Minutes to 131.6

The U.S. Energy Information Administration publishes two averages for how long an electricity customer in the United States went without power during 2024. One is 662.6 minutes. The other is 131.6 minutes. They sit on the same row of the same table, they carry the same name, and they describe the same twelve months.

The distance between them is a set of calendar days that were identified one utility at a time and then set aside. The larger figure counts every non-momentary interruption in the year; the smaller counts the year after its Major Event Days have been taken out. Between the two lies 531.0 minutes, or 8.85 hours, which is 80.1 percent of the total.

Which days get taken out is decided by a formula built out of each utility’s own past. The line between an ordinary bad day and a major event is therefore not the same in every service territory, and it does not stay put over time.

The Chain

The IEEE method turns a year of outage minutes into two numbers through a fixed sequence. EIA’s footnote states the rule in two sentences: “For utilities using the IEEE method, a Major Event Day is any day that exceeds a daily SAIDI threshold called Tmed.” and “Tmed is a duration statistic calculated from daily SAIDI values from the past five years.” The standard itself, filed in state utility dockets, defines the day this way: “A major event day is a day in which the daily system SAIDI exceeds a threshold value, TMED.”

The standard sets the threshold out as steps a) through f). The utility collects daily SAIDI values covering five sequential years. Days without interruptions are dropped, because the standard directs that “Only those days that have a SAIDI/Day value will be used to calculate the TMED (do not include days that did not have any interruptions).” The natural logarithm of each remaining daily value is taken. Alpha is set to the average of those logarithms and beta to their standard deviation. The threshold is then e raised to the power of alpha plus 2.5 times beta. Classifying days against that threshold is a separate step, g).

The threshold then governs the following year, not the year it was built from. In the standard’s words, “Any day with daily SAIDI greater than the threshold value TMED that occurs during the subsequent reporting period is classified as a major event day.” The sentence that follows it is the part that explains why two columns exist at all: “Activities that occur on days classified as major event days should be separately analyzed and reported.” The purpose the standard gives for the whole procedure is “to allow major events to be studied separately from daily operation, and in the process, to better reveal trends in daily operation that would be hidden by the large statistical effect of major events”.

Two properties follow from the arithmetic. Because alpha is the log-average of every day on which that utility recorded an interruption, a system whose ordinary days are calm carries a low threshold, and a moderate storm can clear it. Because beta is a log-standard deviation, a system whose daily record swings widely carries a high threshold, and the same storm may fail to clear it. The threshold measures unusualness relative to one utility rather than severity in any absolute sense. A short record does not avoid this: “If fewer than five years of historical data are available, use all available historical data until five years of historical data are available.”

How a calendar day leaves the reported reliability number The chain: five years of daily SAIDI, drop days with no interruptions, take natural logarithms, compute alpha and beta, set the threshold, then classify any day above it in the following year as a major event day. How a calendar day leaves the reported reliability number Steps a) to g) of the IEEE Std 1366 2.5 beta method; the EIA footnote states only what Tmed is a-b Collect daily SAIDI values for five sequential years, for that utility alone Days with no interruptions are left out of the data set c Take the natural logarithm of each remaining daily SAIDI value d-e Alpha is the average of those logarithms; beta is their standard deviation f Threshold T-MED = e to the power of (alpha + 2.5 x beta) g In the next reporting year, any day above T-MED is a Major Event Day Its minutes leave the without-major-event-days column and stay in all events Day above T-MED Minutes counted in the all-events column only Day at or below T-MED Minutes counted in both columns Source: IEEE Std 1366 as filed with the Missouri Public Service Commission; footnotes to EIA Table 11.1.

One Table, Two Definitions of a Major Event

Table 11.1 of the Electric Power Annual, built from Form EIA-861 returns, does not hold two averages for 2024. It holds five. Three are computed from utilities reporting under IEEE Std 1366: all events, 662.6 minutes; without major event days, 131.6 minutes; and loss of supply removed, 646.1 minutes. Two more are computed under the heading “any method”: all events, 611.3 minutes, and without major events, 126.0 minutes.

The last pair covers a wider set of utilities, not a different arithmetic on the same set. EIA states it plainly: “Any method combines data from utilities that use IEEE standard with data from utilities that do not.” For the utilities that do not, the exclusion is not computed at all. “For utilities not using IEEE methods, Major Events are self-determined by the reporting utility.” An EIA summary from an earlier data year made the consequence explicit, noting that “Not all utilities are consistent in what they consider major events”.

The gap between the two all-events aggregates is not stable, and its sign is not stable either. In 2024 the any-method figure ran 51.3 minutes below the IEEE figure. In 2021 it ran 35.8 minutes below. In 2020 it ran 35.8 minutes above. Over the eleven years in the table, the any-method total came in higher than the IEEE total in two years and lower in nine. A reader comparing a state against a national average is choosing between two aggregates that differ in who is counted.

Five averages on one row of EIA Table 11.1 The 2024 row holds three IEEE columns and two any-method columns. The IEEE columns use a computed threshold; the any-method columns also contain utilities that determine major events themselves. One row of EIA Table 11.1 holds five different averages SAIDI in minutes per customer, United States total, data year 2024 IEEE All events with major events 662.6 IEEE Without major event days 131.6 IEEE Loss of supply removed 646.1 Any method All events with major events 611.3 Any method Without major events 126.0 Major event = a day above T-MED, a threshold computed from that utility’s own five years Combines IEEE reporters with utilities that decide for themselves The any-method group is the wider one: it holds the IEEE reporters plus utilities that do not use IEEE methods. For utilities not using IEEE methods, major events are self-determined by the reporting utility. Source: U.S. Energy Information Administration, Electric Power Annual, Table 11.1 (Form EIA-861).

Eleven Years Side by Side

Set the two IEEE series side by side for 2014 through 2024 and they behave like measurements of different quantities. The upper series runs 236.2, 209.0, 268.4, 505.9, 349.2, 295.5, 456.1, 475.8, 333.0, 366.6 and 662.6 minutes. The lower series runs 114.2, 117.0, 119.8, 117.0, 121.4, 122.2, 116.0, 125.7, 131.1, 123.9 and 131.6.

The upper series spans 453.6 minutes between its lowest year and its highest, a ratio of 3.17 to one. The lower series spans 17.4 minutes, a ratio of 1.15 to one. The full eleven-year swing in the excluded-events number is smaller than the 123.6-minute CAIDI the same table reports for 2024 on the days that were not classified out; the all-events CAIDI for that year is 432.7 minutes. Measured against its own mean, the upper series varies by 34.2 percent and the lower by 4.6 percent, a factor of 7.4.

Averaged across the eleven years, the all-events figure is 378.0 minutes and the without-major-event-days figure is 121.8. The gap is 67.8 percent of the larger figure, which is a comparison of eleven annual national averages rather than a customer-weighted share of the minutes recorded. EIA describes the same pattern in words: “Interruptions attributed to major events averaged nearly nine hours in 2024, compared with an average of nearly four hours per year in 2014 through 2023.” and “Service interruptions that aren’t triggered by major events routinely average about two hours per year.” A separate EIA summary put it as a trend, observing that “Since 2013, the average duration of electricity interruptions each year has remained consistently around two hours after excluding major events.” The 2024 frequency figures follow the same shape: SAIFI of 1.531 interruptions with major event days and 1.065 without, against a CAIDI of 432.7 minutes and 123.6 minutes respectively.

Two U.S. outage averages for the same eleven years The all-events series swings between about 209 and 663 minutes a year. The series excluding major event days stays between about 114 and 132 minutes across the same eleven years. Two U.S. outage averages for the same eleven years SAIDI in minutes per customer per year, utilities reporting under IEEE standards With major event days Without major event days 0 100 200 300 400 500 600 700 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 662.6 505.9 209.0 114.2 131.6 Across the eleven years the lower series spans 17.4 minutes; the upper series spans 453.6. Source: U.S. Energy Information Administration, Electric Power Annual, Table 11.1 (Form EIA-861).

The Same Split, State by State

Table 11.4 breaks 2024 down by state, and the two columns diverge in proportion to how much of the year arrived at once. South Carolina recorded 3,155 minutes with major event days, the longest in the country at 52.6 hours, and 115.9 minutes without them, meaning 96.3 percent of its outage minutes were classified out. North Carolina recorded 1,518.1 and 143.9, a 90.5 percent split. Texas recorded 1,614.3 and 153.4, also 90.5 percent. Georgia recorded 1,208.1 and 211.4, or 82.5 percent.

Florida is the case that shows what the two columns are actually measuring. Its all-events figure is 1,487.6 minutes, or 24.8 hours. Its without-major-event-days figure is 66.7 minutes, lower than Arizona’s 69.8, even though Florida’s all-events figure for the year was 17.8 times Arizona’s 83.6 minutes. Read one column and Florida had a catastrophic year. Both figures come from the same filing.

States without a landfalling hurricane show the split compressing rather than disappearing. South Dakota recorded 85.6 minutes and 61.5, a 28.2 percent exclusion. Massachusetts recorded 103.7 and 77.0, or 25.7 percent. Arizona recorded 83.6 and 69.8, or 16.5 percent. Hawaii sits in between at 382.6 and 219.3, a 42.7 percent exclusion, in a state where EIA attributes frequent interruptions to weather, volcanic activity, unplanned outages at oil-fired plants and interconnection of new capacity.

Minutes Removed Are Not Interruptions Removed

The exclusion does not take the same share out of every index. In 2024 the national figures dropped from 662.6 minutes to 131.6, removing 80.1 percent of the duration. Over the same step the frequency index dropped from 1.531 interruptions to 1.065, removing 0.466 interruptions, or 30.4 percent of the count.

Divide the 531.0 removed minutes by the 0.466 removed interruptions and the implied length of an interruption on an excluded day is about 1,139 minutes, close to 19.0 hours. The published CAIDI for the days that remained is 123.6 minutes. The interruptions taken out were therefore roughly 9.2 times longer, one at a time, than the ones left in. That is a derived comparison rather than a published one, and it assumes the two columns cover the same customer base, which the table does not state.

The distinction matters for what the two columns describe. The number that stays flat near two hours a year is a statement about how often an ordinary fault occurs and how quickly crews clear it. The number that swings between 209.0 and 662.6 minutes is mostly a statement about how long restoration takes when a large fraction of a distribution system is damaged at once. The first is a maintenance and operations measurement. The second is a restoration-capacity measurement. The table reports both because they move independently.

Where This Doesn’t Apply

The threshold logic describes only utilities reporting under IEEE Std 1366-2003 or IEEE Std 1366-2012. For the rest, EIA records that major events are self-determined, so the any-method columns mix a computed boundary with a declared one.

SAIDI is a system average weighted by customers, not a description of any household. A territory where a few customers lost service for a week and one where most customers lost service briefly can report similar minutes. The index also covers distribution systems: outages originating on the bulk power system are separated in the column EIA labels loss of supply removed, which “excludes outages due to loss of supply from the high-voltage/bulk power system.”

Short interruptions fall outside the count entirely. EIA notes that “For utilities that report SAIDI metrics using Institute of Electrical and Electronics Engineers (IEEE) standards, non-momentary interruptions are those lasting longer than five minutes.” A system with frequent brief dropouts and few long ones can post a low SAIDI while feeling unreliable to the people on it.

The figures also arrive late and in stages. EIA collects them on Form EIA-861, where the most recent survey year is posted as an early release file while the year before it carries final figures, so two adjacent years on the same page are not at the same stage of review. The Electric Power Annual carrying data for 2024 was released on October 16, 2025, about nine and a half months after that data year closed.

Finally, the method does not claim that one column is the true one. Its stated purpose is to let major events be studied apart from daily operation so that trends in daily operation are not hidden. Reading the 131.6-minute figure as the time customers spent without power in 2024 inverts that purpose; reading the 662.6-minute figure as routine performance does the same in the other direction. The two are answers to two questions, and the boundary between them is a threshold each utility computes from its own five preceding years.

Comments

Post a Comment

Popular posts from this blog

The Repair Takes Hours: Clearing a Boil Water Notice Runs on Laboratory Time

A boil water notice usually arrives with a cause attached: a broken main under an intersection, a pump that lost power, a pressure zone that fell below the level the operator treats as safe. The cause can be repaired within a day. The notice can remain in force well after the repair is finished. That gap is not slow paperwork. Issuing the notice and withdrawing it run on two different mechanisms: one has a federal deadline written into it, and the other is a sequence of physical and biological steps, several of which cannot run in parallel and one of which is a waiting period inside a warm cabinet. The distance between them is what a household experiences as an advisory that will not lift. Two clocks start at the same moment One is fixed by federal rule. The other is a sequence of physical and biological steps. NOTICE CLOCK — bounded at 24 hours CLEARANCE CLOCK — set by steps, not a deadline Hour 0 The system learns of the situation No later than 24 hours Tier 1 public no...
Read more

Twenty-Three Days Before an Outbreak Is Visible, and Longer Before a Recall Notice

A recall notice arrives as a single event: a brand, a lot code, a date range, a list of states. Inside the machinery that produces it, the notice is one of the last steps. Before a product can be named in public, several separate systems have to finish running, in order. A person's immune system, a clinical laboratory, a courier, a state public health laboratory, a national genomic database, a round of interviews, and finally a paper trail held by firms with no direct relationship to one another. The Centers for Disease Control and Prevention publishes an illustration of the first half of that sequence. In it, a person eats a contaminated food on Day 1, and the outbreak is identified on Day 23. That interval is not a measure of effort. It is a sum of waiting periods belonging to different institutions, none of them working to a shared deadline. What the First Nine Days Consume In CDC's published timeline, nothing in the first nine days is decided by a food company or by...
Read more

A Landslide Dam Forms in Seconds and Fails on Its Own Schedule

A Flood With No Storm Attached Strong ground shaking is over in well under a minute. The flood it sets up can arrive a week later, or two months later, in clear weather, with nothing in the rainfall record to account for it. The object standing between the two events is a pile of rock and soil that fell into a river channel during the shaking and stayed there. Geologists call it a landslide dam , or a barrier lake when the emphasis falls on the water rather than the debris. The U.S. Geological Survey studied the general case in The Formation and Failure of Natural Dams , Open-File Report 87-392, which compiled landslide dams worldwide and tabulated how long they lasted. Its central finding is uncomfortable: these structures do not fail on any schedule that can be pinned down, and among those that do fail, most fail well after the response has wound down. The failure mode is not the landslide. The landslide is the visible event, the one that gets counted. The failure mode is the i...
Read more