Replication package for "Estimating residential energy efficiency through energy use intensity: A multilevel analysis of the 2020 Residential Energy Consumption Survey"

Full replication package for the article "Estimating residential energy efficiency through energy use intensity: A multilevel analysis of the 2020 Residential Energy Consumption Survey" (Energy and Buildings). It reproduces every result, table, and figure from the raw inputs. Data. The analysis uses the public RECS 2020 microdata (U.S. Energy Information Administration; 18,243 households across 50 states and 17 IECC climate zones) merged with state-level contextual variables (electricity price, urbanization rate, and GDP per capita) compiled from EIA, U.S. Census Bureau, and BEA sources, with release versions and access dates documented. Methods. The dependent variable is energy use intensity (site energy per unit of conditioned floor area). A survey-weighted two-level model (households nested in IECC climate zones) is estimated in Stata, with average marginal effects and the intraclass correlation. Interaction structure is selected in two stages: a stability-frequency screening in R (inspired by stability selection, without PFER control) over 250 canonical interaction blocks (190 household x household and 60 household x state-context), which yields 24 stable candidate blocks; a Stata refit then retains 17 blocks (three involving state-level context, including the all-electric x electricity-price interaction). Key results. The largest fitted associations are dwelling vintage, envelope quality, heating technology, education, and household size; state electricity price and income operate mainly through moderation. The conditional intraclass correlation is approximately 0.21. Contents. Stata do-file and logs; R stability-selection script and outputs; processed contextual datasets; estimation files (.ster); all figures (Figure 1 climate-zone map, Figures 2-3, Supplementary Figure S1, and the graphical abstract); and the manuscript-aligned CSV tables. README.txt documents the pipeline, CHANGELOG.md records the validated clean run (27 June 2026), and SHA256SUMS.txt provides checksums for all 66 files. requirements.txt and requirements-lock.txt pin the Python environment used for the figures; R sessionInfo and the Stata environment record are included. Reproducibility. run_replication.sh executes the full pipeline (Stata to R to Stata) and regenerates all outputs; Figure 1 is regenerated separately via fig1/make_fig1.py. The deposited results correspond to a clean re-run on R 4.6.0 and Stata BE 18 (arm64 macOS).