Data from the RECS can be used to assess potential
bias that can occur in engineering studies which are based on
small samples that may not be representative of the larger population
of interest. The RECS sample covers all types of households and
housing units in the United States, and the sample was selected
in such a way that the sampling error of its estimates are measured
and provided to users along with the estimates themselves.
For example, an engineering study of 53 homes, where
two light loggers had been used in each home, produced an annual
estimate of 2,930 kWh for lighting. [22]
The authors noted that it might be dangerous to estimate
for all lights in the house from a sample of two and they noted
that their estimates were higher than estimates commonly used
by utilities. The study suggested an annual lighting budget for
a home of 1,200 to 1,500 kWh. The RECS estimated that only 27
percent of households used 1,200 or more kWh for lighting. If
the engineering study represented single-family households only,
the lighting budget still applied to about 25 percent of single-family
homes (25 percent of single-family homes used between 953 and
1,397 kWh per annum).
It is not certain in this study whether the unusually
high estimate, compared to RECS, came from sampling lights that
were used more often (the authors thought their sample of lights
were used less often than other lights in the house) or because
the sampled of households were high users of lighting. Whichever
the source, the unusually high estimates could be detected and
an estimate derived from RECS data that would quantify how unusual
the estimates were. The assumption that these high estimates are
typical of national usage would lead to bias in any analysis that
uses the engineering data. Therefore, researchers who conduct
small-scale studies purporting to represent a customer base that
is a cross section of the general population of homes should compare
their data to the RECS data.
A recent survey of utility companies inquiring about
the prevalence of end-use metering reported that 68 percent did
end-use monitoring. The return rate was small, but it is an indication
that a number of utilities conduct these studies for use in their
planning and evaluation. The difference between the RECS lighting
estimates and the engineering studies is a major issue for them.
[23]
Bias in RECS Estimates of Lighting Consumption
Data on lighting usage can be collected in several
ways. The 1993 RECS used self-reported data gathered during in-person
household interviews. Self-reported data can also be collected
via phone interviews and diaries. Other methods that do not rely
on self-reported data include mechanical devices, such as light
loggers. To what extent do self-reported data, especially in-person
self-reported data, give an accurate picture of the amount of
time a light is turned on? Respondents may have some difficulty
knowing how many hours each light in the household is used. They
may not be home during all of the times that lights are turned
on and, even if home, may not know how others in the household
are using lights.
A study comparing various types of self-reported
data to light logger data found that self-reported data gave higher
estimates of light usage than data from the light loggers. [24]
Light logger data were 81 percent of the on-site estimates and
72 percent of the diary estimates. Estimates over the telephone
were inflated compared to on-site estimates (on site estimates
were 72 percent of the telephone estimates). A regression analysis
of the RECS electricity consumption data yielded a similar result
(see Appendix B).
These findings may lead one to believe that RECS overestimates the amount of electricity used for lighting. While it is true that the RECS relies on selfreport usage data, those data are not used directly to model electricity consumption for lighting. The RECS estimates of the amount of electricity used for lighting are regression-based estimates (see Appendix B). They are not based upon light loggers or submetering. For each RECS respondent, the regression equation, whose coefficients were estimated by a nonlinear regression procedure, was used to estimate the proportion of the total annual electricity consumption that was used for each of 10 enduse categories. Lighting is one of the categories. The estimated amount of electricity for each enduse category is the total annual electricity consumption (derived from utility bills) times the proportion for the end use. As a consequence, the enduse consumption estimates for the 10 categories will add up to the actual total annual electricity consumption. The tendency of respondents to overestimate how long lights are on is reflected in the coefficients estimated by the regression procedure.