How the Grid Reliability Index is built

What the metrics measure

SAIDI. System Average Interruption Duration Index: minutes the average customer went without power over the year.

SAIFI. System Average Interruption Frequency Index: how many times the average customer lost power over the year.

Major Event Days. Days of extraordinary stress such as hurricanes, ice storms, and wildfires. EIA files each metric both including and excluding them; both are published here and neither is shown alone.

Source

U.S. Energy Information Administration, Form EIA-861 reliability file, data year 2024 — the filings report on 2024-01-01 through 2024-12-31, and that period, not the day this page was built, is what the figures describe. The file is Reliability_2024.xlsx, pinned at SHA-256 ac2805354a0a5666…. Utilities report these figures annually; the data year, not the retrieval date, is what the numbers describe.

EIA's file carries a third metric block whose exact definition could not be established from the archive. It is not published. Publishing a figure whose definition we could not establish would be worse than omitting it.

2 filed figures contradict themselves: major events are added to the ordinary year, so the inclusive figure cannot be the smaller of the two. Each metric is judged on its own pair, because a contradiction between two SAIDI values says nothing about whether the SAIFI values contradict. Nashville Electric Service (TN) filed 0 minutes including major events against 86.58 excluding them; Nashville Electric Service (TN) filed 0 interruptions including major events against 1.01 excluding them. Those cells read “filed, contradictory” in the state table rather than “not reported”, because the filer did publish a figure and the decision to withhold it was made here. The inclusive figure is withheld for each of those metrics rather than published or silently averaged in, and the excluded-events figure is kept.

How state figures are built

A state figure is the customer-weighted mean of the utilities that reported it. Weighting matters: an unweighted mean would let a cooperative serving two thousand homes count as much as a utility serving six million, and the resulting number would describe neither.

Each metric is averaged over its own filings, not over a single shared set. A utility often publishes one major-event variant and not the other, so the ordinary-year mean and the storm-inclusive mean for the same state can rest on different filers — that is true of 16 of the 51 jurisdictions here. Every state page prints the count behind each figure rather than one count for the page.

A utility that filed the form without publishing a metric is excluded from the mean rather than counted as zero. A reported 0.0 is kept, because a year without interruptions is a real result and not a missing one. A value that was filed but could not be read is excluded too, and the table shows it as “filed, unreadable” rather than “not reported”, because a defect in the source is not the same as a filer staying silent; the 2024 file contains none. Of 971 filings, 747 carried a usable figure and 224 did not.

The storm comparison uses a fixed set of utilities

Each state page states how many times worse the storm-inclusive figure is than the ordinary one. That ratio is not the quotient of the two headline averages, because those are averaged over whoever filed each metric and can rest on different utilities. It is computed instead over the filers that published both variants, so the comparison holds the customer population fixed on both sides.

The two approaches can disagree materially. The widest gap in this snapshot is Tennessee, where the paired ratio is 1.9× across the 41 filers that published both figures, against 2.0× from dividing the two published averages — the difference between saying major events account for most of the outage time there and saying they do not.

Where this is weakest

• A state is not an address. Each utility files one figure for its whole territory, so a rural feeder and an urban network are reported as a single number, and the state figure averages those territory averages again. How far any particular address sits from the published figure is not something this dataset can say.
• Thin samples. 4 jurisdictions have fewer than 3 filings behind at least one published figure — District of Columbia, Hawaii, Rhode Island, Vermont. Their pages say so; the figure describes those filings, not the state.
• One year is not a trend. A single severe storm can move a utility’s storm-inclusive figure by an order of magnitude. These are 2024 filings, not an average of years.
• Utilities classify major events themselves under an IEEE standard, and the boundary between an ordinary bad day and a major event is not identical across filers.
• The two columns are not always the same utilities. Where a filer published one variant and not the other, comparing a state’s ordinary-year figure against its storm-inclusive one compares two slightly different sets of utilities.
• Nothing here prices a battery. The index says how much outage time a territory reported. What backup is worth depends on the cost of the outage to a household, which this data does not contain.

Why a solar site publishes this

A grid-tied solar array shuts down during an outage. It is designed to, so that power is not fed onto lines that crews are working on. Panels alone therefore do nothing for resilience, and the case for adding a battery rests on how often and how long the power actually goes out where you live. The Solar Savings Index covers the other half of that decision.