Primary datasets
| Source | Period | Licence | SHA-256 (first 16) |
|---|---|---|---|
| U.S. Energy Information Administration, Electric Power Monthly, Table 5.6.A | June 2026 | U.S. Government work, public domain | ccb6392e962fc396 |
| U.S. EIA, 2020 Residential Energy Consumption Survey (RECS), Table CE4.6 | June 2023, revised March 2024 | U.S. Government work, public domain | 52455267579e319b |
Appliance inputs
Nameplate and standby figures are typical published manufacturer ratings for mainstream current models. Duty cycles are engineering values for the stated operating assumption. Every figure is a typical value for a typical unit, not a measurement of any specific product; ranges are given where the spread across models is wide enough to change the answer.
- FTC EnergyGuide label estimated yearly electricity use (16 CFR Part 305)
- U.S. DOE appliance efficiency standards test procedures (10 CFR Part 430)
- ENERGY STAR product specifications and qualified product lists
- U.S. EIA 2020 Residential Energy Consumption Survey, end-use tables
These are not a single downloadable dataset. Ratings and standby figures are typical published values for mainstream current models; duty cycles are engineering values for the operating assumption stated on each appliance’s own page. Where a regulated annual figure exists, the inputs are calibrated against it — the calibration table shows every one of those checks.
What each dataset actually contains
Electric Power Monthly, tables 5.6.A and 5.6.B
Published monthly by the EIA, around the 25th, covering the month two months prior. Table 5.6.A gives the average price per kilowatt-hour for each state and the District of Columbia, broken down by end-use sector — residential, commercial, industrial and transportation — for the reporting month and the same month a year earlier. Table 5.6.B gives the same figures averaged across the year to date.
These are average revenue per kilowatt-hour: total revenue collected divided by total kilowatt-hours sold. That means they include fixed monthly charges spread across the energy sold, which is why a state average is usually a little below the effective rate a low-usage household actually pays, and why they cannot see time-of-use pricing at all.
Residential Energy Consumption Survey 2020, table CE4.6
RECS is a national sample survey of housing units, run periodically since 1978; the 2020 cycle is the most recent published. Table CE4.6 gives average annual site consumption per household by end use and by fuel, for the United States and for each census region and division.
The critical property, and the one most often misread: these averages are conditional. The space-heating figure is the mean among households that heat with that fuel, not among all households. That is why the end-use columns in the published table do not sum to the total column — each is conditioned on a different subset — and it is why applying them only to the end uses a home actually has produces a breakdown that reconciles instead of one that double-counts.
This site uses the national row for the bill tool’s defaults and the census-division rows for the state pages, because a Pacific household averages 7,665 kWh a year against 13,609 in the West South Central — a national figure describes neither.
Reproducing any figure here
Every derived file is regenerated from the raw sources by one script, and the script fails rather than guesses: it reads the reporting period out of the sheet’s own title, refuses to run if the column headings have moved, and refuses to publish if any state is missing. So a figure on this site is not a number someone typed — it is the output of a transformation you can re-run.
- Download the source file from the EIA link in the table above.
- Compare its SHA-256 against the one published there. If they differ, the EIA has revised the data since this build.
- The derivation applies the formula on the method page with no further adjustment.
Why the raw files are kept
Government statistics get revised. Keeping the exact file a figure was derived from, with its checksum, means a number on this site can be reproduced years later even after the publisher has replaced the table — and it means a revision shows up as a changed checksum rather than as numbers quietly moving.
How an appliance figure is put together
Unlike the two EIA tables, the appliance inputs are not a single downloadable file, and it would be misleading to present them as one. Each appliance carries four numbers, and they come from different places:
| Input | Where it comes from | How firm it is |
|---|---|---|
| Rated watts | Typical published nameplate ratings for mainstream current models | Firm. This is printed on the appliance. |
| Duty cycle | Engineering value for the stated operating assumption, constrained by the published annual figure where one exists | The softest input, and the one the sliders exist for. |
| Hours and days | A stated typical usage pattern, written on each appliance’s page | A stated assumption, not a claim about you. |
| Standby watts | Typical published standby figures; regulated in many categories | Firm for current models, much higher for pre-2010 equipment. |
The honest summary: the rated and standby figures are close to facts, the usage pattern is an assumption you can change, and the duty cycle is a modelled value pinned to an external range wherever regulation provides one. The seven appliances that are pinned are listed here, each with its computed figure beside the published range it has to fall inside.
When these figures go out of date
The two datasets age very differently, and it is worth knowing which is which.
| Dataset | Published | How fast it ages |
|---|---|---|
| Electric Power Monthly | Monthly, around the 25th, covering two months prior | Quickly. Rates moved by several per cent nationally in the last year and by far more in some states. Refreshed here each quarter at minimum. |
| Residential Energy Consumption Survey | Every five to six years; 2020 is current | Slowly. How much a household uses for water heating changes over decades, not seasons — but the housing stock and the equipment in it do shift, so the 2020 cycle describes 2020. |
The build stamps the date it ran, and the footer of every page shows it. If that date is old and the rates look stale, they are — the fix is one command and a redeploy, and the derivation refuses to run if the EIA has changed the shape of the table rather than silently publishing the wrong column.
What these sources cannot tell you
- Your utility’s actual tariff. A state average blends every utility and every rate class in that state. It cannot see time-of-use pricing, tiered rates, demand charges or a fixed monthly charge as a separate line.
- Anything sub-annual about consumption. RECS reports annual totals. A January heating figure is not in the data; only the year is.
- Your specific equipment. Survey means describe populations. A fifteen-year-old refrigerator with a tired seal is not the average of anything.
- Anything outside the United States. Both datasets are US-only, and every figure on this site is therefore US-only.
Why not use a live API
The EIA publishes an API, and this site deliberately does not call it. Three reasons, and they all come down to the same principle: a published figure should be traceable to a fixed artefact rather than to whatever a server returned at an unknown moment.
- Reproducibility. A file with a checksum can be re-read years later. An API response cannot be, and when the publisher revises a series the old value simply ceases to exist.
- Failure mode. A build that depends on a live service fails, or worse half-fails, when that service is unavailable. A build that reads a committed file cannot.
- Visible revisions. Government statistics get revised. With a stored file, a revision shows up as a changed checksum in a diff that someone reviews. With an API, the numbers move silently between builds and nothing records that they did.
The cost of this choice is that refreshing the data is a deliberate act rather than an automatic one. That is the intended trade: figures on this site change when someone decides they should, and the change is recorded.
Reuse
The underlying government data is public domain and you may use it freely. The derived figures, the duty-cycle model and the writing on this site are ours; quote them with a link and that is fine.