RB-2026-005  Power Bill Shock: Residential Electricity Bills by State
NumeralQ, https://www.numeralq.com/research/electricity-bills-by-state
Author: Joshua Kurz. Data frozen 17 November 2026. License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Credit "NumeralQ".

These are AVERAGES PER RESIDENTIAL ELECTRICITY ACCOUNT from utility filings (EIA Form 861M), not any one household's
bill. Latest period = Sep 2025-Aug 2026. Prior period = Sep 2023-Aug 2024 (two years earlier). Four-year comparison =
Sep 2021-Aug 2022. Jan-Aug 2026 are EIA preliminary figures and will be revised. Dollars are nominal unless named "real".
"US" rows are the sum of the 50 states and DC.

----------------------------------------------------------------------------------------------------------
state-monthly-residential.csv  (one row per state, DC and US per month, January 2019 to August 2026)
----------------------------------------------------------------------------------------------------------
state                     Two-letter code (DC included; US = sum of 50 states and DC).
year, month               Calendar month.
eia_status                EIA's flag for the month: Final or Preliminary.
revenue_thousand_usd      Residential revenue from electricity sales, thousand dollars (EIA-861M).
sales_mwh                 Residential sales, megawatt-hours (EIA-861M).
customers                 Residential customer accounts (EIA-861M).
price_c_kwh               revenue / sales, cents per kWh = revenue_thousand_usd / sales_mwh x 100.
kwh_per_customer          sales_mwh x 1000 / customers (kWh in that month).
bill_per_customer         revenue_thousand_usd x 1000 / customers ($ in that month).

----------------------------------------------------------------------------------------------------------
state-price-vs-bill.csv  (52 rows: 50 states, DC and US)
----------------------------------------------------------------------------------------------------------
For each 12-month period (suffix _prior, _latest, _4yr_prior), with revenue and sales summed over the 12 months and
customers averaged over the 12 months:
price_c_kwh_*                         Price = total revenue / total kWh x 100 (cents per kWh, sales-weighted).
kwh_per_month_*                       Usage = total kWh / average customers / 12.
bill_per_month_*                      Bill = total revenue / average customers / 12 (EIA's own method). Price x usage = bill.
annual_bill_prior, annual_bill_latest Bill per month x 12 ($ per year).

price_change_pct, usage_change_pct, bill_change_pct   Latest vs prior, percent.
bill_change_usd_per_year              annual_bill_latest - annual_bill_prior.
bill_change_from_price_usd_per_year   Midpoint split: price change x average of the two periods' usage ($ per year).
bill_change_from_usage_usd_per_year   Midpoint split: usage change x average of the two periods' price ($ per year).
                                      The two parts add up to bill_change_usd_per_year.
price_change_4yr_pct, bill_change_4yr_pct   Latest vs Sep 2021-Aug 2022, percent.
real_bill_change_pct                  (1 + bill change) / (1 + CPI-U all-items change) - 1, percent. CPI change from build-meta.json.

Weather check (48 contiguous states; blank for AK, HI and DC, which have no NOAA statewide degree-day series):
hdd_prior, hdd_latest, cdd_prior, cdd_latest   NOAA statewide heating / cooling degree days (base 65 F), summed over each period.
n_months_fit                          Months in the fit (120: Sep 2016-Aug 2026).
kwh_per_hdd, kwh_per_hdd_se           b and its standard error in kWh_m = a + b HDD_m + c CDD_m + d (t_m - mean t) + e_m,
kwh_per_cdd, kwh_per_cdd_se           c and its standard error. kWh_m = monthly kWh per customer; t = year x 12 + month.
trend_kwh_per_month_per_year          d x 12.
fit_r2                                R-squared of the fit.
max_cooks_distance                    Largest Cook's distance of any month.
influential_months                    Number of months with Cook's distance above 4/n (dropped in the refit).
most_influential_month                Month with the largest Cook's distance (YYYYMM).
weather_kwh_per_month                 [b x (hdd_latest - hdd_prior) + c x (cdd_latest - cdd_prior)] / 12.
weather_kwh_per_month_without_influential   Same, refit without the influential months.
weather_adjusted_usage_change_pct     (kwh_per_month_latest - weather_kwh_per_month) / kwh_per_month_prior - 1, percent.
weather_adjusted_usage_change_pct_without_influential   Same, from the refit.
weather_model_customer_share          US row only: share of customers in states with a weather fit (AK, HI, DC count as zero effect).

value_of_1000_kwh_saved_prior, _latest   1,000 kWh x average price of that period ($). Average price includes fixed charges, so the
                                      saving on the margin is somewhat lower.
recs_space_heating_share, recs_water_heating_share, recs_air_conditioning_share, recs_refrigerator_share, recs_other_share
                                      Share of household site electricity by end use, EIA RECS 2020 table CE4.1.EL.ST.
est_space_heating_cost_per_year_latest, est_air_conditioning_cost_per_year_latest, est_water_heating_cost_per_year_latest
                                      annual_bill_latest x the RECS 2020 share. ESTIMATES, not measured 2026 costs.
rank_bill_change, rank_price_change, rank_annual_bill_latest   1 = highest, among the 50 states and DC.

----------------------------------------------------------------------------------------------------------
build-meta.json
----------------------------------------------------------------------------------------------------------
latest_month, window_latest, window_prior, window_4yr_prior   The periods used.
preliminary_months_in_latest_window   Months EIA marks preliminary.
weather_fit_from                      First month of the weather fit.
cpi                                   BLS CPI-U all items (CPIAUCNS) and CPI electricity (CUUR0000SEHF01), not seasonally
                                      adjusted, via FRED. BLS did not publish October 2025, so both periods average the same
                                      11 calendar months (months_used, months_dropped). *_pct = percent change of those
                                      averages; *_pct_interpolated_12m = check using all 12 months with October 2025 filled
                                      as the September/November average.

Sources: EIA Form 861M (sales_revenue.xlsx, through August 2026); EIA RECS 2020 CE4.1.EL.ST; NOAA NCEI climate
divisional data (climdiv HDD/CDD, release 5 Nov 2026); BLS CPI via FRED.
