The best source of energy consumption data for commercial buildings traditionally has
been the Energy Information Administration’s (EIA) Commercial Building Energy Consumption
Survey that has been conducted every 4-5 years since 1989, with some notable exceptions
[CBECS]. The last such survey was completed in 2003 and the EIA is now finalizing its 2012
survey. The 2003 survey contains 5,215 records, each record corresponding to data gathered
from one sampled building. Collectively these data represent an estimated 4.9 million buildings
and 72 billion gsf in the U.S. commercial building stock. Hundreds of pieces of information are
gathered for each sampled building including data for energy use, occupancy, equipment, and
function.
The simplest form of benchmarking involves comparing a particular building’s EUI with
gross EUI for the entire U.S. building stock.
2
Of course a hospital is expected to use more
energy than an office building, so such broad comparisons have limited value. CBECS identifies
about 20 different building types in its sampling, allowing one to determine national gross EUI
for a specific peer group – say just hospitals. CBECS does not identify the location for each of its
samples but it does identify five climate zones and nine census divisions. To obtain a more
relevant peer group one can extract CBECS records filtered on such criteria. But with only 5,215
sampled buildings in CBECS, filtering on more than building type and census division may yield
only a handful of buildings in the selected peer group – with correspondingly large uncertainties
in their gross EUI and other statistics.
In the late 90’s the EPA borrowed methodology then being developed at DOE labs to
apply regression analysis to CBECS data for benchmarking purposes [Sharp, 1996; Sharp 1998].
The idea is this. Suppose one intends to benchmark, say, an office building in Topeka, KS. There
may not be a single sampled office building in CBECS that matches many of the characteristics
of the one to be benchmarked. But CBECS does include office buildings that are both larger and
smaller, ones that are older and newer, and so on. Multivariate linear regression analysis is
applied to all CBECS office buildings to see how their energy use depends on such variables.
The resulting regression coefficients may then be used to predict the energy use of a hypothetical
building with characteristics similar to those of the one to be benchmarked. Regression analysis
has the potential to yield the best of both worlds – the specificity of a narrowly defined CBECS
query with the statistics of the larger sampled dataset.
The EPA introduced its ENERGY STAR benchmarking system and score nearly 15 years
ago. The ENERGY STAR score is an index from 1-100 which is supposed to represent a
building’s percentile energy efficiency ranking with respect to similar buildings nationally. Over
the years the EPA has developed ENERGY STAR models (i.e., scoring systems) for 11
conventional building types, listed in Table 1. The earliest ENERGY STAR building models
were based on 1995 CBECS data. Models were revised as newer data became available.
From the outset the EPA’s ENERGY STAR scoring system was a voluntary
benchmarking tool intended to encourage energy efficiency [Janda and Brodsky, 2000]. Building
data submitted to Portfolio Manager (the EPA’s web-based tool for calculating scores) and
ENERGY STAR scores issued by the EPA are confidential – unless a building seeks and
receives ENERGY STAR certification, in which case its score is 75 or higher and public
disclosure holds little risk of embarrassment. The EPA has not defended or justified its
methodologies in any peer-reviewed venues. And, while, for marketing purposes, the EPA
2
Gross EUI for a set of buildings is defined to be their total annual energy use divided by their total gsf.
2683-©2014 ACEEE Summer Study on Energy Efficiency in Buildings