EIA Consumption & Efficiency Surveys (RECS, CBECS, MECS)

The demand side of the American energy system, measured unit by unit instead of modeled. EIA runs three flagship end-use surveys under one Consumption & Efficiency roof: RECS, the Residential Energy Consumption Survey, whose 2020 wave put 18,496 households under observation on behalf of the country's 123.5 million housing units; CBECS, the Commercial Buildings Energy Consumption Survey, whose 2018 cycle profiled 6,436 sampled buildings standing in for an estimated 5.9 million; and MECS, the Manufacturing Energy Consumption Survey, publishing roughly 100 tables of fuel and electricity use across NAICS industries and regions. Together they answer the questions billing data cannot: what the energy was used for - space heating, water heating, cooling, motors, process use - and what the consuming units look like.

Each survey arrives as end-use survey microdata or pre-aggregated tables with the statistical machinery attached: analysis weights, replicate weights for variance estimation, and flags marking which values were imputed. Within Datadory's multi-sector holdings shelf this is the real-asset layer - the record holding groups with utility or real-estate exposure benchmark against. It scores 9/10 on our rubric against a catalog average of 7.81 across the 1,744 datasets we cataloged. Get a sample of this dataset cut to your sectors before you build on it.

One record per surveyed unit. Three real rows, exactly as they arrive:

# one row per surveyed unit - verbatim variable names, codebook values

# CBECS 2018 building record 1 of 6,436:
PUBID   : 00001        REGION : 3        CENDIV : 5
PBA     : 2 (office)   SQFT  : 210000   NWKER  : 350
ELBTU   : 18708970     ELEXP : 775800
DHBTU   : 10968747     FINALWT: 2.17

# CBECS 2018 building record 2 of 6,436:
PUBID   : 00002        REGION : 4        PBA    : 2 (office)
SQFT    : 28000        NWKER : 12
ELBTU   : 1528667      NGCNS : 1946
MFBTU   : 1730655      MFEXP : 82030

# RECS 2020 household record 1 of 18,496:
DOEID        : 100001       REGIONC      : WEST
state_postal : NM           TYPEHUQ      : 2
TOTSQFT_EN   : 2100         KWH          : 12521.48
DOLLAREL     : 1955.06      TOTALBTU     : 144647.71
MONEYPY      : 13           NWEIGHT      : 3284.1036678

# ... repeats across 18,496 households and 6,436 buildings,
#     plus ~100 pre-aggregated MECS tables crossing NAICS industry with region

Three things worth reading off them. First, consumption rides beside cost: ELBTU and ELEXP on the building rows, KWH and DOLLAREL on the household row, so implied unit prices fall out of a division rather than a procurement exercise. Second, DHBTU and NGCNS show the fuel mix arriving fuel by fuel - district heat and natural gas are separate columns, not a blended total. Third, the weight columns (FINALWT, NWEIGHT) are the multiplier that turns 6,436 buildings into 5.9 million and 18,496 households into 123.5 million; ignore them and you have described the sample, not the country.

Seventeen field families, defined below with examples taken from real rows.

Identity and geography - DOEID / PUBID anonymize the respondent; REGIONC / REGION and DIVISION / CENDIV place each unit at census region and division, the finest geography the building records carry.

Physical and activity descriptors - TYPEHUQ codes the housing unit from mobile home upward; PBA types the commercial building from vacant through office, warehouse and food sales; SQFT / TOTSQFT_EN size the unit and NWKER counts the shift that works in it.

Consumption and expenditure - ELBTU / KWH, NGBTU / NGCNS and FKBTU carry electricity, natural gas and fuel oil separately; MFBTU / MFEXP derive the major-fuels total; TOTALBTU / TOTALDOL sum all fuels for households with per-end-use breakdowns underneath.

Climate and statistics machinery - PUBCLIM assigns ASHRAE climate zones, HDD65 / CDD65 bring degree days for weather normalization, FINALWT / NWEIGHT scale the samples to national populations with replicate weights alongside, and the Z* flags mark every imputed value.

Bulk-delivery variants of these families - replicate weights as their own table, value-label references joined onto coded columns, cross-cycle stitched frames - sit under additional fields on request: named at sampling, delivered as typed columns.

• Geography: United States national estimates with deliberately bounded detail - state-level consumption and expenditure tables for RECS, a nine-census-division floor for CBECS building records, NAICS industry crossed with region for MECS. No addresses survive anywhere; the masking is the price of publishing unit-level records at all.
• Temporal: fixed vintages rather than a ticker. RECS 2020 stands on waves from 2009 and 2015; CBECS 2018 extends a lineage reaching back to 1992 in the microdata era, with program history since 1979; MECS 2018 tables carry a 2022 collection cycle behind them. Trend work stitches snapshots on consistent definitions instead of reading a continuous series.
• Granularity: one record per surveyed household or commercial building, pre-aggregated cells for manufacturing - each row carrying the weight that scales it to the population it represents.

Set it against the rest of the shelf in our best multi-sector holdings datasets ranking: the LEI estate covers who owns whom, this record covers what the owned assets consume.

API, files, or your warehouse. Daily, weekly, or hourly.

Survey-shaped records reward whole-vintage handling. Take complete RECS or CBECS extracts sized for overnight warehouse loads, cut slices by sector, division or building activity when only one peer group matters, or stand up a feed keyed on survey plus vintage so a new cycle lands as fresh rows beside the old ones instead of overwriting them. Cadence is yours to set and change: hourly feeds suit dashboards waiting on a fresh vintage, nightly loads suit enrichment pipelines, weekly pulls suit research snapshots.

Every delivery ships the complete field dictionary above, value-label reference tables for the coded columns, and validation rows so you can reconcile your tabulations against the published national estimates before anyone downstream asks.

Benchmark a building portfolio against the nation. CBECS intensity distributions by principal activity, size band and division give property exposure an honest peer group. (ESG emissions analysis)

Size retrofit and equipment markets first. Building counts by activity, era and region plus RECS appliance tables are the baseline every heat-pump and controls market model starts from. (market sizing)

Model residential load from behavior, not assumptions. Appliance holdings, dwelling traits and measured kilowatthours arrive on the same row, ready for feature work. (ML model training)

Anchor industrial demand scenarios. MECS end-use tables and fuel-switching capability say which fuels each industry can run on and what it consumed when it did. (demand forecasting)

Cite the measurement, not a copy. Documented questionnaires and stated sample sizes make every figure quotable. (citation-grade research)

Investors & Quants benchmark energy intensity across property and utility exposures before sizing positions.

Market Researchers & Consultants segment with official denominators - buildings, floorspace, workers, households - already counted.

Data Scientists & ML Engineers train on a weighted, weather-normalized panel of real consuming units with hundreds of labeled variables.

Developers & Builders build sustainability dashboards on a codebook-documented vocabulary that survives between cycles.

Sales & Growth Teams size territories from the same counts the regulators publish.

• Weights make the estimates - FINALWT and NWEIGHT turn samples into national pictures, and the replicate series turn standard errors into a column operation. Unweighted averages quietly answer a question nobody asked.
• Division is the floor - disclosure-protected microdata means no addresses and no state identifiers below census division on building records; plan around the boundary, not through it.
• Imputation is flagged, not hidden - the Z* variables separate operator-reported values from statistically filled ones; models should weight accordingly.
• Vintages, not tickers - these are snapshots. Multi-cycle trend work joins consistent definitions across waves instead of expecting a continuous series. (what end-use survey microdata means)
• Where to go next - the rail below collects the identity siblings, the disclosed-holdings sources and the head-to-head comparison. The EIA source profile covers the wider catalog behind this record.