13.4 million residential electricity disconnections in 2024 — the federal standard that would bar cutting power in dangerous weather has been optional since 1978

What is happening?

The United States counted its residential electricity disconnections nationally for the first time in April 2026. The Energy Information Administration published the 2024 Residential Utility Disconnections Report, the first output of Form EIA-112, and it recorded 13,447,935 residential electricity disconnections for nonpayment during calendar year 2024 against 143,033,986 residential electricity accounts — a national rate of 9.4 disconnection events per 100 accounts. The form is a census of every utility on the federal electric and gas reporting frames rather than a sample, with a weighted response rate of 93.5 percent on the electricity side.

The unit has to be read carefully, because the number invites a misreading. EIA states that a single customer account may receive more than one final notice, disconnection or reconnection within one calendar year. This is a count of events, and it is not a count of households that lost power.

No federal rule governs the act being counted. Section 201(b)(1) of the Federal Power Act, 16 U.S.C. 824(b)(1), confines federal jurisdiction to interstate transmission and wholesale sales and states that the subchapter does not apply to any other sale of electric energy and that the federal commission has no jurisdiction over facilities used in local distribution. The Congressional Research Service put the consequence plainly in a report dated 2023-01-31 — current law and policy give state and local regulators essentially all authority over disconnection policies and practices.

A federal standard nevertheless exists, and it has existed since 1978-11-09. PURPA wrote federal standard number four, procedures for termination, at 16 U.S.C. 2623(b)(4), and filled it out at 16 U.S.C. 2625(g) with reasonable prior notice, notice of rights and remedies, a reasonable opportunity to dispute, express provision for elderly and handicapped consumers, and no termination during periods a state regulatory authority finds especially dangerous to health where the consumer shows inability to pay. But 16 U.S.C. 2623(a) required only that each state regulatory authority and each nonregulated utility hold a hearing within two years and determine whether to adopt the standard. The floor was drafted, handed to the states as a question, and the two-year window closed around 1980.

The result is visible in the 2024 federal data. Oklahoma recorded 30.4 disconnection events per 100 residential electricity accounts, Texas 23.6 and Florida 20.6, against 1.9 in Massachusetts and 1.2 in Wyoming — a spread of about 24 times between the highest state and the lowest.

Whose problem is this?

The body that could level the map is the one body that decided in 1978 not to. Every actor with authority over the act itself is a state actor, and no state actor has authority over any other state.

Where does this problem end?

The boundary matters here because the rule already exists in federal text and only the obligation to apply it is missing. That is a different problem from one where nobody has written the rule.

What is the state now, and what should it be?

Now

Needs a new measurementthe target state: no source opened here names a target disconnection rate, an acceptable rate, or a threshold at which anything would trigger. The United States now publishes a national disconnection rate to two decimal places with no stated intent to read it against. The 1978 standard describes a floor in words rather than in numbers, and no opened source reports how many states adopted it or what rate it was meant to produce.

How big is it?

About 6.7 million to 13.4 million residential electricity accounts, for calendar year 2024. The upper bound is the census count itself, 13,447,935 disconnection events, and it is a ceiling on distinct accounts because repeats can only lower the account figure and never raise it. The lower bound is not published by anyone. EIA holds the number of distinct accounts disconnected at least once and does not report it, so the floor here rests on an assumption of two events per affected account, chosen to show the width of the interval rather than derived from any source.

The width of that interval is a publication choice rather than a sampling problem. The census is essentially complete, and the gap between the two bounds is entirely the distance between events and households.

A second and fully source-backed anchor measures who lacks the floor at all. The seven states found in June 2023 to have no weather-based protection of any kind held 17,057,411 residential electricity accounts in 2024, about 11.9 percent of the national total. That figure mixes two reference dates, a 2023 policy status against 2024 account counts, and the mixture is stated rather than smoothed. The corresponding number for the summer-only gap cannot be produced, because no opened source publishes the list of states that lack hot-weather protection.

A separate household-denominated measure exists and answers a different question. The residential energy consumption survey found 16.19 million of 132.54 million households, 12.2 percent, received a disconnect or delivery stop notice in 2024. That survey asks about notices and never asks whether service was actually cut.

Under what conditions does it arise?

1. Retail service was placed outside federal reach by design. The Federal Power Act confines the federal commission to interstate transmission and wholesale sales and excludes local distribution facilities. The agency that regulates electricity nationally has no authority over the act in question, and that is the architecture rather than an oversight.

2. The one federal standard that speaks to the act was written as a question. PURPA required each state to hold a hearing within two years and determine whether to adopt the termination standard. A consider-and-determine statute produces 50 answers to one question, frozen at the moment each state answered, with no mechanism to ask again.

3. The protection map and the weather map have drifted apart. Heat-related deaths in the United States totaled 21,518 between 1999 and 2023, rising from 1,069 in 1999 to 2,325 in 2023, with the age-adjusted rate up 63 percent and the annual increase running at 16.8 percent from 2016 through 2023. Billion-dollar winter disasters rose from 31 in the two decades to 2004 to 73 in the two decades to 2024. Summer protection remains the minority column in every inventory.

4. Whole classes of provider sit outside the state rules themselves. Moratoria bind commission-regulated companies. In the 2024 federal data municipal utilities disconnected at 12.3 per 100 customers and cooperatives at 7.8, against 6.3 for investor-owned utilities, and large municipal systems reached 16.4 per 100.

5. Arrears grew faster than assistance. Residential utility arrears passed USD 21 billion by September 2024, roughly one household in six or seven was behind on an electric or gas bill, and LIHEAP reached 18 percent of income-eligible households in 2023 with under 3 percent of income-eligible households receiving cooling help in FY2023.

What has been tried?

Two of these are the whole story in miniature. The one time a near-national floor existed it was built in months and dismantled in months, and no federal instrument was created to retain any part of it. The one time Congress finally moved, it moved the measuring instrument and not the rule.

What was found?

Why is it still unsolved?

Supply disparity — the same household need meets a different rule at every state line and at every change of provider class, and no procedure exists to level it.

The first movement is jurisdictional and deliberate. In 1935 the Federal Power Act put retail service outside federal reach, and in 1978 Congress wrote the termination standard anyway and then declined to require it. What 16 U.S.C. 2623(a) asked of each state was a determination, not compliance. A statute of that shape does not produce a floor; it produces a map. The two-year window closed around 1980, the answers were never re-collected, and the map has been carried forward for 46 years while the weather it was answering has changed. Heat-related deaths rose 117 percent between 1999 and 2023 and the sharpest part of that rise came after 2016, entirely outside the window in which any state was asked the question.

The second movement is that the disparity nests inside itself. A household inside a protected state can still be outside the protection in four ways, and each of them is invisible in the national count. Municipal utilities and rural cooperatives are generally outside state commission jurisdiction, and in 2024 they disconnected at roughly twice the investor-owned rate. Delivered fuels have no disconnection to prohibit, because the supplier simply does not deliver and no form records it. Pay-in-advance lapses are excluded from the federal definition, so the households whose service ends most quietly are the ones the count is built not to see. Tenants can lose service for a nonpayment that is not theirs, under notice periods that in some states are shorter than the ones customers get. The least-covered providers cut the most, and the least-covered households are the least counted.

The third movement is the inversion that keeps the pattern stable. In May 2025 a federal budget proposal argued for ending the assistance program on the ground that states already have policies preventing disconnection. The unevenness that is the problem was offered as the reason no federal action is needed. The premise is testable and no one on either side tested it, because the one measurement that would settle it — the share of households covered by an enforceable protection at the moment the temperature crosses a threshold — is not produced by any body in the country. Meanwhile the route by which a protection actually gets adopted runs through a single death that becomes newsworthy inside a single state, as it did in Arizona in 2019. There is no route by which the same fact changes the rule in Florida, Texas or Oklahoma.

What observation would mean it is solved?

Candidates — (a) a binding national minimum takes effect, whether by making the 1978 standard mandatory or by new statute, with a stated effective date (b) the spread between the highest and lowest state disconnection rates narrows over several years (c) the disconnection rate falls in the states and provider classes that currently sit outside protection.

(a) alone is weaker than it looks. The 1978 text shows how a fully drafted floor can sit in the code for 46 years without binding anyone. A mandatory version could still leave municipal utilities, cooperatives, delivered fuels and pay-in-advance customers outside its reach, and every one of those exclusions already exists inside state rules today. A national minimum that inherits the existing carve-outs would change the map and not the holes in it.

(b) alone measures the wrong thing. Convergence can come from protected states getting worse as easily as from unprotected states getting better, and the national rate moves with arrears, weather and utility practice at the same time. The spread also depends on a unit that is not the one readers assume, because these are events per 100 accounts and not households cut off.

(c) alone counts what the instrument can see. The federal census excludes pay-in-advance lapses by definition, records no duration between a disconnection and a reconnection, and carries no demographic breakdown at any level. A rate that falls while prepay adoption rises would look like progress and might not be any. The three have to be read together, and all three have to be read against the fact that no source anywhere states what rate would be acceptable.

What is it connected to?

Fills with researchwater and telecommunications shutoffs under the same state regulatory structure, prepaid electricity markets and their growth, landlord and tenant utility law, energy burden and housing cost, and disconnection rules in other countries. Relation type and evidence grade were not confirmed in this round.

What these sources do not say

• Whether any state adopted the 1978 federal standard. The termination standard is live in the United States Code. No opened source says how many state regulatory authorities adopted it, how many declined, or whether any has revisited its determination in the 46 years since. The statute that most nearly answers this problem is tracked by nobody who writes about this problem.
• What the measurement was for. Congress directed up to USD 3 million and three years of work to produce the national count, and not one opened source names a federal body empowered to act on the result. The data gap identified in January 2023 is now closed, and no source states what policymaking followed.
• What an acceptable rate would be. Not the statistical agency, not the research service, not the policy inventory, not the pending bill, not the assistance directors. There is no target, no threshold and no trigger anywhere in the opened record.
• How long anyone was without power. The federal report counts 13.4 million disconnections and 11.4 million reconnections and publishes no elapsed time between them. A two-hour lapse and a six-month one are the same row, and the residual of about 2.0 million between the two counts is not explained.
• How large the pay-in-advance population is. The federal form excludes those lapses from the definition of a disconnection, and where utilities reported them anyway the values were imputed. No opened source estimates how many residential accounts are prepay or how often they lapse, and prepay is concentrated in exactly the high-rate states.
• The disconnection rate by race, income, age or disability. The federal census carries no demographic breakdown at any level, and the household survey that carries all of them asks only whether a notice was received. The most cited disparity in this field is a disparity in warnings, not in outcomes.
• What share of households the protections actually cover. Moratoria bind commission-regulated companies and not municipal utilities, cooperatives or delivered fuels. Nobody states what share of households that excludes, even now that the ownership split is computable from the federal data.
• What happens after reconnection. Reconnection fees, deposits and the arrears that survive a moratorium are named as the reason a moratorium defers rather than forgives, and no opened source quantifies the cost of getting back on.

See the evidence

One primary federal dataset was read directly and it carries most of this document. The EIA-112 report was opened as a PDF and every national and state figure here is computed from its own tables, with the counts route cross-checked against its published monthly rates at 9.41 against 9.40. The three sections of the United States Code were read directly and are the only primary legal texts here, and the bill text of H.R. 2486 was read in full specifically to check whether it contains a disconnection prohibition; it does not, and a third-party headline claiming otherwise is not supported by the text. Everything else is secondary. Where sources disagree the disagreement is left visible rather than resolved. The count of protecting states is 41 and 20 in an inventory that publishes its method, 40 and 21 in a legislative finding that does not, and 43 and 25 on a later dashboard, and none of the three reconciles with the others. The older working figure of about 1 percent of households disconnected annually cannot be squared with 9.4 events per 100 accounts, and no opened source attempts it. The move from nearly 3 million reported disconnections in 2022 to 13.4 million in the 2024 census is a change in measurement rather than a change in conditions, and it is recorded here as such. Two figures were sought in this round and could not be reproduced from anything opened, so neither appears anywhere above — a single settled count of states barring disconnection during extreme weather, and annual disconnection rates broken out by race of householder. The utility-ownership rates rest on a single independent analysis, since the federal report publishes no ownership breakdown. The Arizona and Florida summer contrast is arithmetic on the federal tables and is not a causal estimate, and the one causal estimate cited here is scoped to pandemic-era moratoria and pandemic mortality and must not be read onto heat. This is a Path A output (research-based definition), so observation_refs is empty and provenance_mode: press-derived.

People affected

Estimated range 6,723,968–13,447,935 As of 2024

Sensitivity The width of this interval is not statistical uncertainty. The underlying collection is a census rather than a sample, essentially complete at 93.5 percent weighted response with the remainder imputed, so the distance between the two bounds is entirely the distance between disconnection events and distinct accounts, which is a publication choice rather than a sampling problem. Neither bound is a count of households. What the figure fails to count runs in one direction only, and every omission is stated by the sources themselves: pay-in-advance lapses are excluded from the federal definition of a disconnection and their size is unknown, delivered fuels such as heating oil and propane have no disconnection to prohibit and are recorded nowhere, tenants who lose service for a nonpayment that is not theirs are counted by nobody at national scale, and the 1.7 million residential natural gas disconnections of the same year cannot be added because EIA warns of significant overlap between the two customer populations. The limit in the opposite direction is that the count captures only households whose service actually ended, while a much larger group avoids that outcome by going without. In the same period 17.55 million households reported leaving the home at an unhealthy temperature and 32.89 million reported reducing or forgoing food or medicine to pay energy costs, and none of them appear in any disconnection count. Duration is also absent, so a lapse of two hours and a lapse of six months are the same row. A separate and fully source-backed exposure figure sits alongside this interval and answers a different question: the seven states found in June 2023 to have no weather-based disconnection protection of any kind held 17,057,411 residential electricity accounts in 2024, about 11.9 percent of the national total, a figure that mixes a 2023 policy status with 2024 account counts.

Regional breakdown State-level disconnection figures do exist in the federal report and are used in the dossier, but they are event counts on the same basis as the national total, not counts of distinct accounts. Converting them would require the same repeat frequency that EIA does not publish, and there is no reason to assume that frequency is uniform across states, since the states with the highest rates are also the states with the least state-level reporting and the greatest exposure to pay-in-advance service, which the federal definition excludes. A regional breakdown built on a divisor assumed constant across states would carry a precision the sources do not support.