Original research

Power outage statistics by state, and the one column that changes the ranking

Reliability for all 51 US jurisdictions on the three IEEE indices, with major event days separated out.

Updated September 2026 · Data as of EIA-861 2024 final release

Written by HyreSolar Research team Research and analysis

Audited by HyreSolar Research team Data audit and fact check

296.8 minutes off, median state SAIDI including major event days, 2024

121.3 minutes on ordinary days Median SAIDI excluding major event days

1.36 interruptions a year Median SAIFI, all 51 jurisdictions

The finding

The median US state left each customer without power for 296.8 minutes in 2024, but only 121.3 of those minutes came from ordinary days. The rest came from major event days: storms, ice and hurricanes. That gap is the most useful thing in this dataset and almost nobody publishes it. A state whose two figures sit close together has a grid that fails routinely. A state where they diverge has weather, not maintenance. Florida ranks 5 of 51 on total outage minutes and 46 of 51 once major event days are removed (rank 1 being the worst) the same grid, ranked 41 places apart depending on which column you quote.

The three indices, and what each one actually measures

Distribution reliability is reported under IEEE Standard 1366, and every utility in the United States files it to the same definitions on Form EIA-861. Three indices do the work. They are frequently quoted interchangeably in coverage, which is a mistake: two of them are independent measurements and the third is arithmetic performed on the first two.

The definitions below are the ones used on this page and in the tables underneath. Where a figure is labelled excluding major event days, it is the second of the two SAIDI columns EIA publishes, and it is the column that tells you about the grid rather than about the weather.

SAIDI: System Average Interruption Duration Index. Total customer-minutes of interruption divided by the number of customers served. Read it as: the average customer went without power for this many minutes over the year. The median jurisdiction sits at 296.8 minutes, which is a little under 4.9 hours.
SAIFI: System Average Interruption Frequency Index. Total customer interruptions divided by customers served, how many times the average customer lost power. The median is 1.36 interruptions. A sustained interruption is one lasting more than five minutes; momentary blips are counted separately and are not in these figures.
CAIDI: Customer Average Interruption Duration Index, and it is SAIDI ÷ SAIFI. It is not an independent measurement of anything. It answers one question only: when the power did go out, how long did it stay out? A high CAIDI means long restorations, not frequent ones. Hawaii has the most frequent interruptions in the country at 4.39 a year and one of the shortest average restorations at 87.2 minutes.
Major event day (MED): A day on which daily SAIDI exceeds a statistical threshold derived from the utility’s own five-year history: the 2.5 beta method, in which the threshold is the exponential of the log-mean plus 2.5 log standard deviations of daily SAIDI. It is deliberately utility-specific, so a day that is a major event for a small rural co-operative may be routine for a large urban utility.
SAIDI excluding MED: The same index with those days removed. This is the number that describes normal operation: tree contact, equipment failure, animal contact, vehicle strikes, planned work. Across the 51 jurisdictions the median is 121.3 minutes, against 296.8 with major event days included.

The 2.5 beta threshold method is set out in the IEEE Power & Energy Society Distribution Reliability Working Group’s classification paper, cited in full at the foot of this page.

The reference points a reader needs before any ranking

The reference points below frame everything that follows. Two are medians across the 51 jurisdictions, EIA publishes each state and the median across them is our arithmetic. Two are the spread between the extremes, and they are wide enough that any sentence beginning "the average American loses" is doing violence to the data. Two describe how much of the burden arrives on major event days.

296.8 minutes, median state SAIDI, major event days included HyreSolar analysis of EIA-861 2024

121.3 minutes, median state SAIDI, major event days excluded The same median, ordinary days only

97× spread between South Carolina and District of Columbia on total minutes 3,155.0 against 32.6 minutes

18.4× spread on ordinary days only, West Virginia to District of Columbia 486.1 against 26.4 minutes

58% of the median state’s outage minutes fall on major event days Median of the 51 per-state shares

31 of 51 jurisdictions where major event days are the majority of outage minutes 10 are above three-quarters

5 states with the same outage minutes and 5 different experiences

The clearest way to show what CAIDI adds is to hold SAIDI still. Every jurisdiction in the table below sits within a quarter of an hour of the median state on total outage minutes. On a ranking built from SAIDI alone they are indistinguishable. They are not remotely the same to live with.

A Wisconsin household loses power about 1.04 times a year and waits roughly 4.8 hours each time. A Tennessee household loses it 1.94 times and waits about 2.5 hours. Same annual minutes, nearly double the number of interruptions, and each one resolved in roughly half the time. Which of those is the worse service is a judgement, not a measurement, but a page that only prints SAIDI has hidden the question.

State	SAIDI (min/yr)	SAIFI (per yr)	CAIDI (min per event)	Reads as
Wisconsin	296.8	1.04	285.2	rare, long
Indiana	310.0	1.28	242.9	occasional, long
California	285.0	1.35	210.7	occasional, long
Kansas	289.8	1.50	192.8	occasional, long
Tennessee	289.3	1.94	149.2	frequent, short

Jurisdictions within 15 minutes of the median state’s 296.8 minutes, ordered by interruption frequency. HyreSolar analysis of EIA-861 2024.

CAIDI is SAIDI ÷ SAIFI, so these three columns are not independent: fix any two and the third follows. Small discrepancies between the published CAIDI and the quotient of the published SAIDI and SAIFI are rounding, because EIA rounds each index before publishing it.

Two ways to have a bad grid

The 25 states with the most outage minutes, 2024. The grey point is SAIDI excluding major event days; the navy point includes them. The length of the bar is the storm burden. HyreSolar analysis of EIA-861 2024 Reliability schedule, IEEE Standard 1366 basis. The horizontal scale is set by South Carolina; the remaining 26 jurisdictions are in the full table below.

Here is the distinction the rest of this page is built on. Two states can report identical total outage minutes and have nothing in common. The published pair of SAIDI columns separates them, and reading only the first one collapses two entirely different problems into a single number.

Storm exposure: the columns diverge

South Carolina reported 3,155.0 minutes in 2024, the highest in the country. Excluding major event days it reported 115.9 minutes, better than the median state, and 30th of 51 on ordinary-day performance. The multiple between its two columns is 27.2×.

Florida is the purer case. It is 5th of 51 on total minutes and 46th on ordinary days, at 66.7 minutes: one of the best-performing distribution systems in the United States on a day when nothing is happening to it. 96% of its outage minutes fall on major event days.

What a state in this column needs is hardening, undergrounding, vegetation management ahead of the season and mutual-aid restoration capacity. What it does not have is a maintenance problem.

Chronic weakness: the columns sit close

West Virginia reported 1,165.5 total minutes, well below South Carolina. But its ordinary-day figure is 486.1 minutes, the highest in the country, and the multiple between its columns is only 2.4×. Its customers lose power 2.83 times a year on average.

Idaho makes the same point from the other end of the total-minutes ranking: 35th of 51 on total SAIDI, 16th once major event days come out. The storms are not what is wrong with it.

What a state in this column needs is capital: reconductoring, feeder automation, pole replacement, fault indicators. Storm hardening will not move its numbers, because storms are not where its minutes are coming from.

All 51 jurisdictions sorted into the four cases

Cutting the two columns at their respective medians gives four groups. The interesting cells are the off-diagonal ones, because those are the states a single-column ranking gets wrong.

Case	What it means	Count	Jurisdictions
Weak and exposed	Above the median on both columns	21	Maine, Texas, North Carolina, Georgia, West Virginia, Virginia, Vermont, Washington, Louisiana, Mississippi, Oregon, Kentucky, Arkansas, Ohio, Michigan, New York, Hawaii, Pennsylvania, Montana, Alaska, Indiana
Storm-exposed only	Above the median on total minutes, below it on ordinary days	4	South Carolina, Florida, New Hampshire, Nebraska
Chronic but storm-spared	Below the median on total minutes, above it on ordinary days	4	Tennessee, California, New Mexico, Idaho
Solid on both	Below the median on both columns	22	Wisconsin, Kansas, Colorado, Minnesota, Oklahoma, Missouri, Iowa, Alabama, New Jersey, Nevada, Illinois, Connecticut, Wyoming, Maryland, Delaware, Utah, Rhode Island, Massachusetts, North Dakota, South Dakota, Arizona, District of Columbia

Medians used as the cut: 296.8 minutes including major event days, 121.3 excluding them. HyreSolar analysis of EIA-861 2024.

8 of 51 jurisdictions land off the diagonal: they are ranked differently depending on which SAIDI column you use. That is 16% of the country, and it includes Florida, which moves 41 places.

The chronic grid: what is left when the weather is removed

Ordinary-day performance against the median state. The twelve worst and twelve best jurisdictions on SAIDI excluding major event days, in minutes above or below the median of 121.3. HyreSolar analysis of EIA-861 2024. The middle 27 jurisdictions are omitted from the chart for legibility and appear in the full table below.

Strip out major event days and the map redraws. The spread narrows from 97× to 18.4×, which is the first useful thing the exercise tells you: almost all of the enormous variation between American states is weather, not engineering. On an ordinary Tuesday the difference between the best and worst distribution systems in the country is a factor of 18.4, not a factor of 97.

The worst ordinary-day performers are West Virginia (486.1 min), Mississippi (279.0 min), Maine (274.0 min), Virginia (245.1 min). All four are rural states with long overhead distribution and heavy tree cover, which is the physical signature of extended radial feeders with a great deal of vegetation exposure and few switching options. The best are District of Columbia (26.4 min), Illinois (59.1 min), Rhode Island (60.8 min), South Dakota (61.5 min), dense, largely urban service territories with short feeders and, in the case of District of Columbia, substantial undergrounding.

Idaho is the case worth dwelling on, because it is invisible in every ranking built on total minutes. It sits 35th of 51 on total SAIDI, comfortably in the better half of the country, and 16th on ordinary days. A reader looking at the headline column would conclude its grid is fine. It is not; it simply had a quiet year for storms.

Where the outage minutes are

Outage minutes per customer by state, 2024. SAIDI including major event days, IEEE Standard 1366 basis. HyreSolar analysis of EIA-861 2024. Colour uses a square-root scale: SAIDI spans two orders of magnitude across the jurisdictions and a linear scale renders forty states as one indistinguishable block. Each jurisdiction is an equal square, because a geographic map would give the empty west the most ink for a per-customer rate where land area is irrelevant.

Total outage minutes have a clean geography, and it is a hurricane geography. The Atlantic and Gulf coasts from Texas round to Maine carry most of the high values, joined by the Appalachian states and the Pacific Northwest. The interior west and the dense north-east corridor carry the low ones.

Arizona is the lowest of the fifty states at 83.6 minutes; District of Columbia is lower still at 32.6, which is what an entirely urban, heavily undergrounded distribution system with 0.27 interruptions per customer per year looks like. It is not a fair comparison with a state and we include it because EIA does, not because it belongs in the same conversation.

Texas is the largest anomaly by population. It reports 1,614.3 total minutes, 3rd worst in the country, against 153.4 on ordinary days, 15th, essentially mid-table. 90% of the outage minutes a Texan customer experiences arrive on major event days. Any account of Texan grid reliability that does not separate those two things is describing the weather and calling it the grid.

Frequency and duration are different failures

Interruption frequency against average interruption length, 2024. Two states can share a SAIDI and sit at opposite corners of this chart. HyreSolar analysis of EIA-861 2024. No trend line is drawn: CAIDI is SAIDI ÷ SAIFI, so regressing one axis on the other would be regressing a variable on its own denominator.

Plotting SAIFI against CAIDI separates the country into regimes that the minute count alone conceals. Four states sit in a category of their own on restoration time, and all four are hurricane-coast states: South Carolina (1,332.3 min), Florida (833.9 min), North Carolina (831.3 min), Texas (704.2 min). Those are averages across the whole year, dragged upward by multi-day storm restorations, and they say more about the scale of the 2024 hurricane season than about routine crew response.

At the other end, Hawaii is the most interrupted service territory in the United States (4.39 interruptions per customer per year, more than triple the median of 1.36) and one of the quickest to restore, at 87.2 minutes per event. Island grids with short feeders and no interconnection fail often and recover fast. Arizona has the shortest average interruption in the country at 86.8 minutes.

District of Columbia has the fewest interruptions of any jurisdiction at 0.27 per customer per year. Among the fifty states the lowest is South Dakota at 0.81.

The storm share, ranked

Expressed as a share rather than a multiple, the storm burden is easier to compare across states of very different reliability. The figure below is the proportion of each jurisdiction's total outage minutes that fall on major event days. It is our arithmetic on EIA's two published columns: (SAIDI − SAIDI excluding MED) ÷ SAIDI.

The median jurisdiction is at 58%, meaning that for a typical American state, most of the time spent without power in 2024 arrived on a handful of days. 31 of 51 jurisdictions are above half. 10 are above three-quarters.

This is the single most under-reported fact about US grid reliability. Utility reliability targets, regulatory performance mechanisms and most published rankings all use the excluding-MED figure, because that is the figure a utility can be held accountable for. Customer experience is the other one. Both are correct and they are answers to different questions.

South Carolina96%

115.9 min on ordinary days, 3,155.0 in total

Florida96%

66.7 min on ordinary days, 1,487.6 in total

North Carolina91%

143.9 min on ordinary days, 1,518.1 in total

Texas90%

153.4 min on ordinary days, 1,614.3 in total

Nebraska90%

69.8 min on ordinary days, 697.8 in total

New Hampshire89%

84.2 min on ordinary days, 735.5 in total

Maine84%

274.0 min on ordinary days, 1,748.6 in total

Georgia83%

211.4 min on ordinary days, 1,208.1 in total

Oregon79%

124.1 min on ordinary days, 579.3 in total

Washington76%

157.3 min on ordinary days, 643.5 in total

Virginia75%

245.1 min on ordinary days, 962.1 in total

Vermont74%

231.8 min on ordinary days, 895.8 in total

Ohio74%

133.3 min on ordinary days, 511.0 in total

Kentucky73%

145.7 min on ordinary days, 537.7 in total

New York70%

142.8 min on ordinary days, 477.3 in total

The fifteen jurisdictions with the highest share of outage minutes falling on major event days, 2024. HyreSolar analysis of EIA-861.

All 51 jurisdictions, 2024

Sorted by total outage minutes, worst first. The two rank columns are the point of the table: where they diverge, the state is being described differently by the two measurements.

State	SAIDI	SAIDI ex-MED	Storm share	SAIFI	CAIDI	Rank, total	Rank, ordinary days
South Carolina	3,155.0	115.9	96%	2.37	1,332.3	1	30
Maine	1,748.6	274.0	84%	3.64	480.9	2	3
Texas	1,614.3	153.4	90%	2.29	704.2	3	15
North Carolina	1,518.1	143.9	91%	1.83	831.3	4	20
Florida	1,487.6	66.7	96%	1.78	833.9	5	46
Georgia	1,208.1	211.4	83%	2.10	574.3	6	8
West Virginia	1,165.5	486.1	58%	2.83	411.5	7	1
Virginia	962.1	245.1	75%	2.24	429.9	8	4
Vermont	895.8	231.8	74%	3.34	268.2	9	5
New Hampshire	735.5	84.2	89%	1.62	455.2	10	38
Nebraska	697.8	69.8	90%	1.24	560.8	11	44
Washington	643.5	157.3	76%	1.77	362.8	12	14
Louisiana	630.6	215.9	66%	2.36	267.7	13	7
Mississippi	607.3	279.0	54%	2.48	244.7	14	2
Oregon	579.3	124.1	79%	1.32	437.6	15	24
Kentucky	537.7	145.7	73%	1.78	301.6	16	19
Arkansas	527.8	201.9	62%	2.14	247.0	17	9
Ohio	511.0	133.3	74%	1.30	393.7	18	22
Michigan	481.7	161.3	67%	1.36	353.0	19	13
New York	477.3	142.8	70%	1.67	286.1	20	21
Hawaii	382.6	219.3	43%	4.39	87.2	21	6
Pennsylvania	353.1	131.5	63%	1.39	253.5	22	23
Montana	341.7	152.3	55%	1.63	209.8	23	17
Alaska	323.2	192.4	40%	2.32	139.3	24	10
Indiana	310.0	123.3	60%	1.28	242.9	25	25
Wisconsin	296.8	97.5	67%	1.04	285.2	26	35
Kansas	289.8	104.0	64%	1.50	192.8	27	33
Tennessee	289.3	169.9	41%	1.94	149.2	28	11
California	285.0	164.4	42%	1.35	210.7	29	12
New Mexico	272.6	152.0	44%	1.32	205.7	30	18
Colorado	257.0	118.9	54%	1.42	181.6	31	28
Minnesota	222.4	91.5	59%	1.26	176.3	32	36
Oklahoma	220.7	121.3	45%	1.58	139.3	33	26
Missouri	212.6	106.6	50%	1.18	180.8	34	31
Idaho	204.4	152.5	25%	1.31	155.5	35	16
Iowa	200.3	84.1	58%	1.21	165.1	36	39
Alabama	197.7	120.8	39%	1.75	113.0	37	27
New Jersey	173.3	98.8	43%	1.12	155.3	38	34
Nevada	158.9	63.8	60%	1.11	143.4	39	47
Illinois	158.0	59.1	63%	0.85	186.3	40	50
Connecticut	152.8	73.6	52%	0.88	174.3	41	42
Wyoming	137.8	116.0	16%	1.00	138.0	42	29
Maryland	118.1	75.1	36%	0.82	144.5	43	41
Delaware	117.9	66.8	43%	0.85	138.0	44	45
Utah	115.4	105.0	9%	1.01	114.5	45	32
Rhode Island	104.3	60.8	42%	0.90	116.4	46	49
Massachusetts	103.7	77.0	26%	0.85	121.8	47	40
North Dakota	102.7	87.0	15%	0.94	109.0	48	37
South Dakota	85.6	61.5	28%	0.81	105.0	49	48
Arizona	83.6	69.8	17%	0.96	86.8	50	43
District of Columbia	32.6	26.4	19%	0.27	120.2	51	51

SAIDI and CAIDI in minutes per customer per year; SAIFI in interruptions per customer per year. Rank 1 is the most outage minutes. HyreSolar analysis of EIA-861 2024 Reliability schedule.

Figures labelled HyreSolar analysis are computed by us from the EIA source files named below. EIA publishes the inputs; it does not publish these ratios or rankings.

Five ways an outage statistic gets misread

Every one of these appears regularly in published coverage of US grid reliability, including in material produced by people selling backup power.

Quoting SAIDI including major event days as "how bad the grid is"
For Florida the two columns differ by a factor of 22.3. The including-MED figure describes a hurricane season. The excluding-MED figure describes the distribution system.
Treating CAIDI as a measure of how unreliable a state is
CAIDI is SAIDI ÷ SAIFI. A state that has one very long interruption and nothing else scores badly on CAIDI and well on SAIFI. Hawaii has the country's worst SAIFI and one of its best CAIDIs.
Comparing a US state figure with a European national figure
Most European reliability figures are reported on a different basis for planned interruptions and use different major-event exclusions. The numbers are not comparable without restating both, and this dataset does not let us do that.
Reading a year of data as a trend
A single hurricane season moves a coastal state's SAIDI by an order of magnitude. South Carolina at 3,155.0 minutes in 2024 is a statement about 2024, not about South Carolina. The ordinary-day column is far more stable year to year and is the one to use for comparison.
Using the state figure to describe a household
SAIDI is a customer-weighted average across every utility filing in the state. Rural co-operative territory and dense urban territory inside the same state routinely differ by more than the gap between the best and worst states. A state number cannot tell you what your street does.

Does unreliable power drive people to solar? The data says no

Outage minutes against rooftop solar adoption, 2024. Each point is one of the 51 jurisdictions. The relationship is weak and negative. HyreSolar analysis of EIA-861 2024. Correlation coefficients computed across all 51 jurisdictions; the version excluding Hawaii and California is reported in the text.

This is a page about the grid, and it belongs to the grid. But one intuition comes up often enough to be worth testing against the same 51 rows, because it is stated as fact in a great deal of marketing: that households in places where the power fails buy solar to escape it.

Across all 51 jurisdictions the correlation between total outage minutes and the share of households with net-metered rooftop solar is Pearson -0.205, Spearman -0.244. Removing the two adoption outliers, Hawaii and California, moves it to -0.255 across the remaining 49. The relationship is weak and it points the wrong way: American states with worse reliability have, if anything, less rooftop solar, not more.

The mechanism is not mysterious once you look at which states are which. The worst reliability sits in the rural south-east and Appalachia, where retail electricity is cheap, incentives are thin and net metering is limited. The best sits in dense urban territories and the interior west. Reliability is confounded with almost everything else that drives adoption, and it loses.

The honest version of the claim is narrower and it is about batteries rather than panels: a solar system without storage does not run during an outage at all, because grid-tied inverters disconnect for line-worker safety. Backup is a storage product. Battery attachment across the installed US fleet is 5.9%, and where it is high it tracks export tariff design rather than outage minutes, the battery analysis takes that apart. For adoption drivers generally, solar statistics by state is the page with the rankings.

Methodology

Source and extraction

US Energy Information Administration, Form EIA-861, 2024 final release, Reliability schedule, "State Totals" sheet, downloaded on 2 September 2026. For each state the sheet carries the number of customers covered, then SAIDI, SAIFI and CAIDI including major event days, then the same three indices excluding them.

We read the sheet by script and keep the state code, SAIDI, SAIFI, CAIDI and SAIDI excluding major event days. Nothing is transcribed by hand and nothing on this page is typed as a literal, every figure in every sentence above is a template expression reading the same JSON the tables read, so a number in the prose cannot drift from the same number in a table.

EIA aggregates utility filings to a state total before publishing this sheet. The aggregation is customer-weighted, which is the correct weighting for an average customer experience and the wrong one for describing any particular utility.

What we derive, and the arithmetic

Storm share, (SAIDI − SAIDI excluding MED) ÷ SAIDI, expressed as a percentage. EIA publishes both columns and does not publish the difference or the ratio.

Multiple, SAIDI ÷ SAIDI excluding MED. Reported to one decimal in the dumbbell chart above.

Ordinary-day rank: the 51 jurisdictions ranked on SAIDI excluding MED, worst first. The dataset ships a rank for total SAIDI; the second rank is ours, and the difference between the two is the finding this page is built around.

Medians: the median across the 51 jurisdictions, unweighted. A jurisdiction is one observation regardless of population, which is the right treatment for "the median state" and the wrong one for "the median American". We have not computed the latter, because the customer counts on this schedule cover only the utilities that filed reliability data and do not reconcile with the residential account totals used elsewhere in this research library.

Quadrants, each jurisdiction is placed above or below the median on each of the two SAIDI columns independently. The cut is the median, not a judgement about what constitutes acceptable reliability.

Why the major event day threshold is utility-specific

Under IEEE Standard 1366 a major event day is not defined by wind speed, by a declared emergency or by any external event. It is defined statistically, per utility, from that utility's own last five years of daily SAIDI: take the natural logarithm of each day's SAIDI, compute the mean and standard deviation, and set the threshold at the exponential of the mean plus 2.5 standard deviations. Days above it are major event days.

Two consequences follow, and both matter for reading the table. First, a day of moderate weather can be a major event day for a utility that is normally very reliable, while the same day is routine for one that is not: the threshold rewards consistency, not absolute performance. Second, because the threshold is set from five years of history, a utility in a run of severe seasons progressively raises its own bar.

The method is deliberate and it is defensible: its purpose is to let ordinary-day trends be seen at all, which a handful of catastrophic days would otherwise swamp. But it is a statistical construct, not a description of the weather, and it should not be read as one.

Limitations

One year. This is the 2024 data year. Storm-driven SAIDI is enormously volatile between years for coastal states, so the total-minutes ranking here should be read as a 2024 ranking and nothing more. The ordinary-day column is far more stable and is the one we would compare across years.

Not every utility files. The reliability schedule covers utilities that report distribution reliability, which is most but not all of the country. Coverage varies between states, and a state where the filers skew towards large investor-owned utilities will look different from one where small co-operatives are well represented.

Sustained interruptions only. Momentary interruptions (under five minutes) are excluded from all three indices. A household on a feeder that blinks daily but rarely fails for long will see excellent numbers and a great deal of reset clocks.

Transmission and generation are not in here. These are distribution indices. A load-shed event driven by generation shortfall may or may not appear, depending on how the distributing utility recorded it, and the schedule gives us no way to separate the two.

No attribution. EIA reports the totals and the excluding-MED figures. It does not name the days, the storms or the causes. We therefore do not attribute any state's gap to a particular event anywhere on this page, however strongly the calendar suggests one.

State averages hide everything below them. Reliability varies more within a large state than between many states. Nothing here describes a specific address.

Reproducing this

The input is a public EIA workbook, the extraction is a short script, and the arithmetic for every derived column is stated above. If a figure here disagrees with one you have computed from the same file, we would like to know: corrections are made on this page with a dated note rather than quietly.

One thing this page deliberately does not do

2024 was a severe Atlantic hurricane season, and Hurricane Helene made landfall in the Florida Big Bend on 26 September 2024 before driving inland across Georgia and the Carolinas, producing what utilities described as the largest outage event several of those states had seen in decades. It is entirely plausible that this is most of what the storm gap in South Carolina, Georgia, North Carolina and Florida represents.

We do not say so in the analysis, because the dataset cannot support it. EIA-861 gives us a total and a figure excluding major event days. It does not tell us which days were excluded, how many there were, or what happened on them. Attributing a state's gap to a named storm would be an inference dressed as a measurement, and this library does not publish those.

Questions

Which state has the most power outages?

On total outage minutes in 2024, South Carolina, at 3,155.0 minutes per customer, but that is a storm figure. Excluding major event days it falls to 115.9 minutes, better than the median state. On frequency, the answer is different again: Hawaii customers lose power 4.39 times a year, more than anywhere else. On ordinary-day duration the worst is West Virginia at 486.1 minutes. Three defensible answers to the same question, which is why the column matters more than the ranking.

What is the difference between SAIDI and SAIFI?

SAIDI measures how many minutes the average customer spent without power over the year. SAIFI measures how many separate times they lost it. A state can be bad on one and good on the other. Dividing SAIDI by SAIFI gives CAIDI, the average length of a single interruption. The median US jurisdiction in 2024 recorded 296.8 minutes across 1.36 interruptions.

What are major event days and why are they excluded?

A major event day is a day on which a utility's daily SAIDI exceeds a threshold computed from its own five-year history under IEEE Standard 1366: the exponential of the log-mean plus 2.5 log standard deviations of daily SAIDI. They are excluded so that ordinary-day trends can be seen at all; a single hurricane would otherwise dominate a decade of data. Both figures are published and both are correct. The excluding-MED figure describes the distribution system, the including-MED figure describes what customers actually lived through.

How many hours of power outage does the average American experience?

The median US jurisdiction recorded 296.8 minutes per customer in 2024, or about 4.9 hours, with 121.3 minutes of that on ordinary days. We report the median state rather than a national per-person average deliberately: the reliability schedule's customer counts cover only the utilities that filed it, so a population-weighted national figure derived from them would not reconcile with any other customer total in this research library.

Is grid reliability in the United States getting worse?

This page covers one data year and cannot answer that. What it can show is where any deterioration would have to be looked for: the ordinary-day column, which reflects the state of the distribution system, rather than the total, which is dominated by whether a given year had a bad storm season. Across the 51 jurisdictions in 2024, 58% of the median state's outage minutes fell on major event days, so year-to-year movement in the headline figure is mostly weather.

Do power outages make people buy solar panels?

Not according to this data. Across all 51 jurisdictions the correlation between outage minutes and household solar adoption is Pearson -0.205, weak, and negative. States with worse reliability have slightly less rooftop solar, not more. It is also worth knowing that a grid-tied solar system without a battery does not operate during an outage: the inverter disconnects for line-worker safety. Outage protection is a storage purchase, not a solar one.

Can I use these figures to estimate my own outage risk?

Only loosely. These are state totals, customer-weighted across every utility that filed. Reliability inside a single large state varies by more than the gap between many states: a dense urban feeder and a long rural radial line are different products. Your own utility publishes its own SAIDI and SAIFI, usually in its annual reliability report to the state commission, and that is the figure to use.

How current is this data and when will it change?

It is the 2024 data year from EIA's final release of Form EIA-861, retrieved 2 September 2026. EIA publishes final data for a year in roughly October of the following year. We rebuild this page from the new workbook against the same URL when the next release lands, and every figure above regenerates from the source file rather than being edited.

Written and audited by

HyreSolar Research

Primary-source research, data analysis and fact checking

We are a research desk, not a sales floor. We read the statute, the tariff, the code section, the federal filing or the manufacturer data sheet ourselves, and we publish the figure with the document it came from and the date we retrieved it. Where a number cannot be traced to a primary source, we publish the shorter page and say what we could not verify. That rule has cost us whole sections, and it is the reason the rest can be trusted.

160: primary sources read and cited
220: figures with a retrieval date
115: federal and state government sources
66: researched pages published

How this desk works

Primary sources only. Statutes from the legislature’s own publishing system, federal data from the agency that collects it, code text from the adopted edition, manufacturer claims from the data sheet. We do not cite an article that cites a source; we go and read the source.
Every figure carries its provenance. A named document and the date we retrieved it, so you can check it and so you know how old it is. Retrieval dates are not decoration: an EIA rate from May is a different fact from an EIA rate from August.
We publish what we could not verify. Every research page carries a section naming the things we tried to establish and could not, and why. A paywalled standard, a state website that refused the request, a manufacturer that publishes no figure at all.
We separate measurement from modelling from our own reasoning, and label which is which on the page. A laboratory measurement, an assumption inside a modelling tool and our own inference are three different kinds of claim and they are never presented as one.
We do not sell solar, and we take no payment for placement, ranking or a favourable mention. Nobody buys a position on this site.

Data as of EIA-861 2024 final release. Authorship on this site is organisational: the analysis belongs to the desk rather than to a named individual, and we do not publish credentials we do not hold. Our editorial policy sets out how we source, date and correct what we publish.

Sources & retrieval dates

US EIA, Form EIA-861, Reliability (2024 final release) — State-total SAIDI, SAIFI and CAIDI on the IEEE Standard 1366 basis, published both including and excluding major event days. The source for every reliability figure on this page. Retrieved 2 September 2026.
IEEE Power & Energy Society, Distribution Reliability Working Group, Classification of Major Event Days — The 2.5 beta threshold method adopted into IEEE Standard 1366, and the definition of a major event day used throughout this page. Retrieved 2 September 2026.
NOAA National Hurricane Center, Tropical Cyclone Report, Hurricane Helene (AL092024) — Landfall date and location for the storm named in the closing note. Cited as context only; no figure on this page is attributed to it. Retrieved 2 September 2026.
US EIA, Form EIA-861 — Net Metering (annual files, 2014–2024) — Utility-level net-metered installations, capacity and PV-paired batteries by state and customer sector. Retrieved 2 September 2026.
US EIA, Form EIA-861 — Sales to Ultimate Customers (annual files, 2014–2024) — Residential revenue, sales and customer counts by utility and state. The basis for price, bill and consumption figures. Retrieved 2 September 2026.
US EIA, Average Price of Electricity to Ultimate Customers by End-Use Sector — EIA’s published price series, used to validate our derivation. Agreement across all 357 overlapping state-years is within 0.005¢/kWh. Retrieved 2 September 2026.

Reliability is a storage question, not a solar one

If keeping the lights on through an outage is what you are actually buying, the size of the battery matters far more than the size of the array. The calculator prices both.

Price a system All calculators

HyreSolar is an independent analysis and matching service. We are not an installer, lender or utility. When a reader asks to be introduced, installers may pay us a referral fee. That fee never buys ranking, scores or placement in research. Our editorial policy sets out the rules.