How the tool works

Updated November 6, 2026

The data

• Prices and usage: U.S. Energy Information Administration, Form EIA-861M, monthly residential revenue, sales and customers by state, through August 2026.
• Weather: NOAA nClimDiv state cooling and heating degree days. For recent months we use faster Climate Prediction Center estimates, calibrated per state to nClimDiv. These months are marked preliminary. Weather data runs through October 2026.

Splitting the bill

With prices r and kWh k for the two bills, the change is divided like this:

• Rate: (r2 minus r1) times the average of k1 and k2.
• Usage: (k2 minus k1) times the average of r1 and r2.
• Weather: k1 times (predicted usage in month 2 divided by predicted usage in month 1, minus 1), times the average price.
• Everything else: usage minus weather.

These add up exactly to the total change. If you leave out kWh, we estimate it from your state's average price, and the rate line then reflects your state and not your plan.

The weather model

For each state we fit monthly kWh per customer against cooling degree days, heating degree days and a slow time trend, with 90% weight on the current month and 10% on the previous month to reflect billing cycles.

How we tested it

We fit each state's model on earlier years and tested on 2023 onward. Results from that statewide test:

• Median R² on unseen data: 0.92.
• Year-over-year usage change: typical change of 4.7%, with 2.9% left unexplained, and the correct direction in 87% of cases.
• Month-to-month usage change: typical change of 15.1%, with 4.9% left unexplained, and the correct direction in 95% of cases.
• Injected-load test: when we added known usage to an average-home simulation, the typical error was $4, and $11 at the 90th percentile.

What we have not proven

These tests use statewide averages, so they show the method works for an average home in a state. We have not tested against real household bills, and we have not independently confirmed that ranges hold 80% of the time. Local weather can also differ from the state average: the gap we measured was 1.3 points at the median and 5.1 at the 90th percentile.

Treat the output as a place to start looking and not as proof of a billing error. The tool cannot see tiered plans, time-of-use pricing, fixed fees or your billing dates.

States left out of weather

We do not show a weather estimate for Alaska and Hawaii, since they are not in the same weather series, or for Maine and DC, since the fit was too weak.

EIA data is public domain, and we acknowledge the U.S. Energy Information Administration and NOAA as sources. Neither agency endorses this site.

Common questions

How accurate is the tool?

On statewide data it had not seen, the median state model scored an R² of 0.92. Testing with statewide data doesn't prove accuracy for one household, because we haven't tested it against real household bills.

Is my data stored?

No. The tool runs in your browser and your bill amounts aren't sent to our server.

Why is the weather shown as a range?

The weather share is an estimate, so we show a range based on how far the state's model missed in testing, plus an allowance for preliminary data where needed.

How often is the data updated?

We update it by hand when new federal data is published. The current data has prices through August 2026 and weather through October 2026.

Why do you say everything else rather than usage?

It is what remains after price and estimated weather. It includes real changes like new appliances, but also billing days, fees and our own estimation error.