end-use survey microdata

End-use survey microdata are the unit-level records behind government energy-consumption surveys: each row is one surveyed household (RECS) or one commercial building (CBECS), carrying hundreds of columns that describe the building, its equipment and every measured end use - space heating, cooling, water heating, lighting, appliances.

The cataloged source is EIA Consumption & Efficiency Surveys (RECS, CBECS, MECS), a directory to EIA's three flagship surveys. RECS 2020 delivers roughly 18,500 households across 799 columns; CBECS 2018 delivers 6,436 buildings across 1,249 columns (~20 MB CSV); MECS 2018 ships about 100 pre-aggregated summary tables instead of unit rows. Files come as CSV microdata plus zipped SAS transport files with codebooks in XLSX and user guides in PDF.

A sample CBECS row shows the shape: PUBID 00001, region 3, census division 5, building-attribute code PBA 2, 210,000 square feet, 350 workers, ELBTU 18,708,970 BTU of electricity - plus the statistical weight (FINALWT 2.17) that turns the sample into national estimates.

Survey cycles are sparse by design: RECS 2020 follows cycles from 2009 and 2015, CBECS has run since 1979 on a roughly biennial-to-quadrennial rhythm, and MECS runs every four years since 1994 (2018 and 2022 editions).

Microdata let you build your own cuts - by climate zone, building type, fuel or appliance mix - which no pre-aggregated table can match. But they punish teams that skip the fine print:

• Weights decide everything. Every file carries survey weights (FINALWT); unweighted averages describe the sample, not the building stock.
• Vintage gaps are structural. RECS ran 2009, 2015, then 2020; trend analysis across those gaps needs explicit methodology notes, not interpolation.
• MECS is not microdata. Manufacturing energy arrives as ~100 summary tables, so per-facility modeling cannot be done from this source at all.

1. Count rows against the official release. RECS 2020 should be near 18,500 households and CBECS 2018 exactly 6,436 buildings; materially fewer means an extract, not the survey.
2. Check column depth. 799 columns for RECS and 1,249 for CBECS are the published widths - thin mirrors often ship only the public-use subset of variables.
3. Demand the weight variables. A file without FINALWT-style weights cannot produce national estimates.

Where to see it in context: multi-sector-holdings data, multi-utilities data.

• eia-api-v2-key - the integration key mechanism for EIA's series-level data; the microdata files themselves are bulk downloads rather than API responses.