Method · last tested 19 October 2026
How we estimate a home’s energy use
A home’s electricity use depends on the weather it sits in: how hot it gets, how much sun falls on the roof, and how humid the air is. This page sets out exactly how we turn bills and weather into the figures on a home’s record, the standards behind each step, and how well it did when we tested it on 18,496 homes across the United States.
The short version
- For a home with its own bills, we fit the bills to that home’s weather, month by month, including sunlight and humidity.
- For a home without bills, we give a range from similar homes in a similar climate, never a single number.
- Tested on the federal Residential Energy Consumption Survey, the range estimate was typically within 27% of what a home really used, against 42% for the national average and 37% for a temperature-only estimate. Its range held the real figure for 80% of homes, as designed.
- The sunlight and humidity terms need monthly readings to test, and that test is still to come. We say so wherever they are used.
Level 1: a home with its own bills
Each billing period becomes an average day. The home’s own bills are fitted to:
| CDD(Tb) | Cooling degree days per day above the home’s own balance temperature Tb: the outdoor temperature at which this home starts to need cooling. We test every whole degree and keep the one that fits the bills best, rather than assuming 65°F. Daily mean temperature is (high + low) ÷ 2 at the nearest weather station. |
| S | Sunlight on cooling days: the solar energy falling on flat ground (kWh per square meter, the load on the roof), counted only on days warm enough to need cooling, so winter sun is not credited to the air conditioner. |
| L | Humidity load: enthalpy degree days, the heat and moisture in the outdoor air above that of the indoor air at the owner’s thermostat setting. We assume 50% humidity indoors and state that assumption on every result. |
| HDD(Th) | Heating degree days below the home’s heating balance temperature, used where the home heats with electricity. |
| b0 | Base use: what does not change with the weather, such as the refrigerator, lighting, water heater and pool pump. |
When a term is kept
A sunlight or humidity term stays in a home’s model only if it pushes use up (its b is above zero) andit improves the fit when each bill is left out in turn and predicted from the others. If not, the record says the effect is not detectable in that home’s bills. Because sunny days are usually hot days, the two overlap; we report how much, so a strong sun term is never read as proof on its own. A home needs at least three bills from months it was lived in, and no model has more than four terms for fewer than two years of bills.
How good a fit is
These are the two measures ASHRAE Guideline 14 uses to judge a model fitted to measured energy use (n bills, p terms). Each home’s record shows both. We follow the Guideline’s family of models; we do not claim certification under it.
Level 2: a home without bills
Without bills, we cannot know how a particular home is built or used, so we give a range drawn from similar homes:
The x terms are cooled, heated and total floor area; cooling and heating degree days at 65°F; floor area × degree days; electric heat × heating degree days; pool; people in the home; electric water heater; electric heat; type of air conditioning; the decade the home was built; the type of home; and its climate zone. The β values are fitted to the survey homes using the survey’s own weights, so the result represents the country’s homes rather than only the ones that answered. On a home’s record, the weather drivers (degree days, summer sun on the roof, humidity load) are shown beside the range.
How it tested
The test used the U.S. Energy Information Administration’s Residential Energy Consumption Survey for 2020: 18,496homes, whose annual electricity came from their utilities’ billing records. The homes were split into ten groups and each group was predicted from the other nine, so no home was predicted by a model that had seen it. “Median error” is the typical gap between the estimate and what the home really used, as a share of what it used.
| Climate zone | Homes | Average kWh a year | Median error: average | Temperature only | Our method | Variation explained | Range held real use |
|---|---|---|---|---|---|---|---|
| All U.S. homes | 18,496 | — | 41.7% | 36.9% | 26.7% | 0.53 | 79.8% |
| Hot-humid | 2,545 | 13,781 | 36% | 31.5% | 24.1% | 0.49 | 79.7% |
| Mixed-humid | 5,579 | 11,556 | 38.3% | 34.6% | 25.3% | 0.49 | 79.8% |
| Hot-dry | 1,577 | 8,637 | 44.3% | 40.8% | 26.8% | 0.56 | 80.1% |
| Mixed-dry | 142 | 9,133 | 35.6% | 42.1% | 25% | 0.55 | 78.9% |
| Marine | 911 | 8,410 | 49.1% | 41.7% | 27.3% | 0.53 | 79.9% |
| Cold | 7,116 | 9,015 | 46.7% | 40.2% | 28.8% | 0.46 | 79.9% |
| Very cold | 572 | 10,190 | 45.5% | 35.6% | 30.3% | 0.42 | 80.2% |
| Florida (within hot-humid) | 655 | — | 36.2% | 29.2% | 23.5% | 0.56 | 80.2% |
Climate zones are the U.S. Department of Energy’s Building America zones, as the survey records them. The 54 subarctic homes are counted in “All U.S. homes” but too few to show alone. The range is built to hold 80% of homes.
Why temperature alone is not enough
The survey also splits each home’s use into end uses with an engineering model calibrated to the bills. We compared the cooling our temperature terms account for with that modelled cooling. They agree most closely in dry climates and least in humid and coastal ones, where moisture in the air and sun on the roof do more of the work. That is why Level 1 carries a humidity term and a sunlight term.
| Climate zone | Homes with cooling | Agreement (correlation) | Temperature-based share of modelled cooling |
|---|---|---|---|
| Hot-dry | 1,404 | 0.75 | 98% |
| Mixed-dry | 128 | 0.58 | 108% |
| Mixed-humid | 5,262 | 0.55 | 74% |
| Hot-humid | 2,317 | 0.54 | 87% |
| Cold | 6,250 | 0.50 | 81% |
| Very cold | 287 | 0.50 | 70% |
| Marine | 512 | 0.41 | 64% |
What is not tested yet
- Sunlight and humidity in Level 1. The survey gives each home’s state and climate zone, not its location, and one total for the year, not months. Testing these terms needs monthly readings from real homes. We are seeking measured data, and owners’ own bills will add to it.
- Homes outside the United States. The weather inputs are global: NASA’s POWER data covers the whole planet, and the Global Historical Climatology Network holds stations worldwide. So the method can run anywhere a home has bills. Its accuracy outside the United States has not been tested, and we will not claim it until it has.
Data and standards
- U.S. Energy Information Administration, Residential Energy Consumption Survey 2020, public microdata version 7.
- NASA Langley Research Center, POWER project: daily solar energy, humidity and temperature.
- NOAA, Global Historical Climatology Network Daily: station highs and lows.
- ASHRAE Guideline 14, measurement of energy, demand and water savings: change-point models and the CV(RMSE) and NMBE measures.
- CalTRACK billing methods: the open, temperature-only method Level 1 builds on, including its search for each home’s balance temperature.
- ISO 52016-1:2017, energy needs for heating and cooling, which counts solar gains and the latent (humidity) load. Level 1 approximates this balance from bills; it does not claim compliance, which needs details of a home’s walls, windows and insulation that we do not hold.
- Enthalpy-based cooling degree days for humid climates: Shin and Do, Energy and Buildings (2015); Krese, Prek and Butala, Energy and Buildings (2011).
- Princeton Scorekeeping Method (PRISM, 1986), the first widely used method for fitting a home’s bills to its weather.
Questions or corrections about this method: tell us. When the method or its test changes, this page changes with it and says when.