Eleven Hours or Two: A Major Event Day Depends on the Prior Five Years

Form EIA-861 collects a year of distribution outage data from every reporting electric utility in the United States. The 2024 row of the table built from it carries five SAIDI values for that single year. Two of them are 662.6 and 131.6. Both are minutes per customer. Both are 2024. Neither corrects the other.

Divide by sixty and the familiar headlines appear. 662.6 minutes is 11.04 hours, and the U.S. Energy Information Administration's write-up of the year says customers "experienced an average of 11 hours of electricity interruptions in 2024." 131.6 minutes is 2.19 hours, and the same write-up says interruptions "that aren't triggered by major events routinely average about two hours per year." The 11-hour year and the 2-hour year are the same year, on the same form, printed on the same line.

What separates them is one test applied a day at a time, and it is not a fixed severity. It is a number each utility computes from its own recent history.

Two Columns, One Year

The table is Table 11.1, Reliability Metrics of U.S. Distribution System, in the Electric Power Annual. Its five column groups are five questions, and the headers say so. Two spanning labels, IEEE and Any Method, stand above five subheads: All Events (With Major Event Days), Without Major Event Days and Loss of Supply Removed under the first; All Events (With Major Events) and Without Major Events under the second. Each prints SAIDI, SAIFI and CAIDI.

The printed definitions are narrow. SAIDI is the "minutes of non-momentary electric interruptions, per year, the average customer experienced," SAIFI the "number of non-momentary electric interruptions, per year, the average customer experienced," CAIDI the "average number of minutes it takes to restore non-momentary electric interruptions." Each begins by excluding something: the momentary interruption.

For 2024 the IEEE group with major event days reads 662.6 minutes, 1.531 interruptions, 432.7 minutes to restore. The group without them reads 131.6, 1.065 and 123.6. CAIDI is not an independent measurement; it is the quotient. 131.6 divided by 1.065 is 123.6, the printed value exactly, and 662.6 divided by 1.531 is 432.8 against a printed 432.7 — a tenth of a minute of rounding in the inputs, not a third measurement.

That quotient shows the year's shape. Moving from the "without" column to the "with" column adds 531.0 minutes but only 0.466 interruptions. Divide one difference by the other: the interruptions major event days contributed averaged roughly 1,140 minutes each, about nineteen hours. Major event days are not days with many more outages. They are days when the outages did not end.

The Chain

Follow one customer minute from the pole to the published table.

First, the minute has to qualify. Part B of the EIA-861 instructions defines SAIFI as how often "the average customer experiences a sustained interruption (of over 5 minutes)." Part C, which governs respondents on another method, instead states that "Momentary outages are not included in determining SAIDI and SAIFI" and asks each of them to declare where that boundary sits. Under either part, a recloser restoring in ninety seconds never enters the numerator.

Second, the minute is divided. The instructions print "SAIDI = [Sum of customer minutes interrupted over the year] / [Total number of customers served]," and define the denominator's unit by cross-reference: "A customer is defined as a metered electrical service point for which an active bill account is established at a specific location (e.g., premise). (IEEE 1366-2003 pg. 2)" The denominator is meters, not people. A household of five and a vacant studio count the same.

Third, the minute is dated, because SAIDI is computed per day as well as per year, and the daily value is what the test reads.

Fourth, the day is compared to a threshold. IEEE Std 1366: "A major event day is a day in which the daily system SAIDI exceeds a threshold value, T_MED." Above the threshold, the day's minutes stay in the "with major event days" column and drop out of the "without" column. Nothing about the minute changed. The date it landed on did.

Fifth, the threshold has to come from somewhere, and it is not a national constant. The standard's procedure reads, in order: "Collect values of daily SAIDI for five sequential years ending on the last day of the last complete reporting period." Then: "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)." Then: "Take the natural logarithm (ln) of each daily SAIDI value in the data set." Then find "a (Alpha), the average of the logarithms," and "ß (Beta), the standard deviation of the logarithms." Then equation (25): T_MED = e^(α+2.5β). Finally: "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 threshold is 2.5 log-standard deviations above a utility's own log-average day. Every utility has a different one.

What the Threshold Is Made Of

Three features of that procedure decide what the 131.6-minute column can mean.

The first is the reference population. Step b) removes days with no interruptions from the history: "do not include days that did not have any interruptions." The log-average day is therefore not an average calendar day. It is an average day on which something failed. Two utilities are each centering their threshold on their own failure days, and those distributions have nothing in common.

The second is the direction of time. The number is backward-looking and forward-applied: "The major event day identification threshold value, T_MED, is calculated at the end of each reporting period (typically one year) for use during the next reporting period." The IEEE working group's paper puts it arithmetically: "For utilities that have six years of reliability data, the first five are used to determine TMED and that threshold is applied during the sixth year." The line that sorted 2024 was drawn from daily values ending in 2023. The hurricane days of 2024 were not in the data set that judged them.

They are in the data set that will judge the years after. A severe year enters the five-year window, raises the log-average and usually the log-standard deviation, and lifts the threshold later days must clear. The dividing line moves toward the weather. A storm that was extraordinary in 2024 can be ordinary by the arithmetic of 2029, its minutes landing inside the "without major event days" column rather than outside it. A standard whose record is frozen fails the opposite way, as with the federal design storm still pinned to a 1961 atlas; this record refreshes annually, so the line slides instead.

The third is the short-history case: "If fewer than five years of historical data are available, use all available historical data until five years of historical data are available." A utility three years into recordkeeping uses a three-year distribution in the same equation, and nothing in the published column marks which respondents were in that position.

Which Column Is the Country

Two groups are labeled "Any Method" rather than "IEEE." Schedule 3 routes respondents with two questions: "If your entity calculates system average interruption duration index (SAIDI) and system average interruption frequency index (SAIFI) and determines Major Event Days in accordance with the IEEE 1366-2003 or IEEE 1366-2012 standard, answer 'yes' to questions 1 and 2 and complete Part B." And: "If your entity calculates SAIDI and SAIFI via another method answer 'yes' to question 1 and no to question 2 then skip Part B and complete Part C."

Part B and Part C are not the same instrument. Part C asks for SAIDI and SAIFI "Including major events" and "Excluding major events" on lines 10 and 11 without naming a standard beside those phrases; Schedule 3's IEEE reference sits in the Part B routing question. Part C then asks two things Part B's reliability lines do not. Line 14: "indicate how your utility defines momentary outages, i.e., less than how many minutes." Line 13: "indicate whether your utility includes inactive accounts in its definition of customers used to determine SAIDI and SAIFI."

Read those lines against the IEEE definitions and the "Any Method" columns come into focus — though not as the non-IEEE half of the country. The table's own note defines them as a pool rather than a rival: "Any method combines data from utilities that use IEEE standard with data from utilities that do not." In the IEEE columns the momentary boundary is over five minutes and the denominator is metered service points with an active bill account. In the Part C returns folded into that pool, both the entry test and the denominator's membership are respondent-declared, and EIA collects the declaration as data rather than imposing a rule; the same note says so plainly: "For utilities not using IEEE methods, Major Events are self-determined by the reporting utility." For 2024 the pool reads 611.3 against the IEEE columns' 662.6 with major events, 126.0 against 131.6 without — a superset against the subset inside it, not two separate countries.

One more instruction shapes both pairs. For the Part B reliability lines, EIA writes: "For lines 3 through 6 complete all that you calculate; for example, if you only calculate SAIDI and SAIFI without Major Event Days included, complete the area under the subtitle 'Excluding Major Events' and do not complete the area under the subtitle 'Including Major Events'." A respondent that produces only one of the two values files only one. The "with" and "without" columns of a published row therefore need not summarize the same set of utilities; each aggregates whoever calculated that number.

The third IEEE group is a different question again. Line 4 asks for SAIDI both including and excluding major event days while "excluding events where the reliability event was initiated from loss of supply (e.g. resulted from an event on the distribution system, not from the high-voltage system)," and the table's note states the effect without the tangle: "Loss of Supply Removed excludes outages due to loss of supply from the high-voltage/bulk power system." The printed column is the including-major-event-days variant, and that removal moves 2024 from 662.6 to 646.1 — sixteen and a half minutes. Whatever took the grid down in 2024, it was overwhelmingly not the high-voltage system handing distribution a dark feeder.

A Second Rulebook, With Fixed Numbers

A federal counting rule works the opposite way, and the contrast shows what a major event day is not.

Under 10 CFR 205.351, every electric utility or other entity engaged in the generation, transmission or distribution of electric energy for delivery or sale to the public "shall report promptly, through the DOE Emergency Operations Center, by telephone, the occurrence of any event such as described in paragraphs (a) through (d) of this section." The criteria are quantities fixed in the rule, each with its own condition attached. Load shedding counts at "over 100 megawatts (MW) of firm customer load," but only "for reasons of maintaining the continuity of the bulk electric power supply system." Equipment failures must involve the loss of firm system loads "for a period in excess of 15 minutes," and the threshold then turns on size: "over 300 MW" for entities "with a previous year recorded peak load of over 3000 MW," and for everyone else losses that "total over 200 MW or 50 percent of the system load being supplied immediately prior to the incident, whichever is less." An interruption counts at "3 hours or longer to over 50,000 customers, or more than 50 percent of the total customers being served immediately prior to the interruption, whichever is less." Voltage reductions count "of 3 percent or greater," again "for purposes of maintaining the continuity of the bulk electric power supply system."

No criterion in that list consults five years of the entity's own daily history. Fifty thousand customers means the same in Maine and in Texas. But the list is not purely absolute either. The equipment-failure threshold sorts entities by "a previous year recorded peak load," and two criteria offer a proportional alternative — 50 percent of the load being supplied, or of the customers being served, "whichever is less." The IEEE threshold asks whether a day was unusual against five years of this utility's record; the federal rule asks what was lost on the night itself. Section 205.351 sets its criteria without reference to any reliability index, and a night reported to the Emergency Operations Center is not thereby a major event day in Table 11.1.

Where This Doesn't Apply

The segmentation is not a trick, and the 131.6-minute column is not a cover-up. The standard states its purpose: "Its purpose 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." That column is an instrument aimed at maintenance, vegetation management, equipment age and switching practice. It was never an account of what customers lived through.

So the columns cannot substitute for each other in either direction. If the question is how much power people lost, the 2024 answer is 662.6 minutes, and state figures in the same source run far higher — EIA reports customers in South Carolina at "nearly 53 hours." If the question is whether routine distribution performance improved, the answer is the 131.6 column, and only against years whose thresholds came from their own five-year windows.

A major event day is also not a severity classification. The label comes from a test on one utility's own history, so the same regional storm can be a major event day for the utility with a calm record and an ordinary day for the neighbor whose last five years were rough. Counting major event days across utilities counts history as much as weather — the pattern of a class that moves while the physical thing does not, as when a dam's hazard class changes without its condition changing.

The frame is narrow in other ways. Schedule 3 covers distribution only: "a distribution circuit is any circuit with a voltage of 35kV or below that serves end use customers directly or through step-down transformers or substations." Generation and bulk transmission are counted elsewhere. And the denominator's customer is a meter with an active bill account, so nothing in SAIDI scales with household size or medical dependence on power.

States draw their own pair of lines on top. The Michigan Public Service Commission publishes distribution metrics "calculated in two ways," one labeled "All Weather" and covering the entire year, the other labeled "Excluding Major Event Days," which "excludes events when the distribution system is experiencing major impacts." A state's excluded set is a separate decision from the one Form EIA-861 records.

Reading the Table

There is no single correct SAIDI for 2024, and the table does not pretend otherwise; it prints five and labels each. The failure mode is in the sentence "the average U.S. customer lost X hours of power in 2024," which requires choosing a column and rarely says which.

Three questions resolve the choice. Did customers lose service? Use the columns with major events. Is routine performance changing? Use the columns without them, remembering that each year's exclusion boundary came from that utility's previous five years. Did a specific night trip a federal obligation? Neither column answers that; 10 CFR 205.351 does, with fixed megawatts, customer counts and hours. The dispatch timestamp that decides what an ambulance response time measures is the same kind of choice as the threshold deciding which day counts as ordinary. The number is not wrong. It answers the question its definition asked.

Primary sources: U.S. Energy Information Administration, Electric Power Annual Table 11.1; U.S. Energy Information Administration, Form EIA-861 Instructions; U.S. Energy Information Administration, Today in Energy; Office of the Federal Register, 10 CFR 205.351; Missouri Public Service Commission, IEEE Std 1366 filing; IEEE Distribution Reliability Working Group, Classification of Major Event Days; Michigan Public Service Commission, Distribution System Reliability Metrics.