Methodology
Overview
Annual estimates for the missing years (1982 through 1984, 1986, 1987, 1989, 1990,
1992, and 1993) are derived separately for each SIC. The derivation has two major stages
for each estimated year. In the first stage, total consumption of offsite-produced energy
is computed and aggregated across all nonelectric energy sources. This total is derived
separately for each SIC, then summed over SIC's to give the manufacturing sector total. In
the second stage, the SIC and manufacturing sector totals are allocated among individual
fuels.
For purposes of this analysis, a missing year is a year for which the
offsite-produced energy amounts are unknown. A known year is one for
which offsite-produced energy amounts are known, either from the MECS or from the ASM.
The earlier and later of the two known years are referred to as the starting
year and the ending year, respectively. The starting and ending
years are also referred to as endpoint years.
From the MECS, there are currently only four known years, 1985, 1988, 1991, and 1994.
Consumption data for these starting and ending years are adjusted, using production,
expenditure, and price data, to derive estimates for the intervening missing years. The
same procedures are applied to fill in the missing years between the starting year of 1981
(the last ASM known year) and the ending year of 1985 (the first MECS known year). Similar
methods can easily be used in the future to fill in between subsequent MECS.
The first stage, deriving total offsite-produced energy use, relies on a combination of
linear interpolation and indexing. The second stage, allocating the total among individual
fuels, uses fuel shares linearly interpolated between the known endpoint years.
Basic Interpolation Methods
Linear interpolation means drawing a line from the starting year to
the ending year and using that line to estimate missing years in between (Figure 1a). A
value obtained by linear interpolation is a weighted average of the values at the starting
and ending years, with higher weight given to the closer year.
An adjustment index is a ratio used to adjust quantities known only for a
base year, to estimate those quantities for missing years. In its simplest form, the
adjustment index is the ratio of a particular statistic or function for two different
years, the missing year and the base year. The function is referred to here as the
adjustment index basis. The Consumer Price Index (CPI) is a widely known example, commonly
used to adjust costs and prices to their base-year equivalents. The set of goods and
services and the methodology for combining them to compute the CPI form the basis of this
index.
In this report, forward indexing means using an adjustment index with the
starting year as the base. Backward indexing means using an index with
the ending year as the base. Two-way indexing means using linear
interpolation between the forward-index and backward-index estimates. That is, the two-way
indexed estimate is the weighted average of the estimates obtained by forward and backward
indexing, with higher weight given to the closer endpoint year (Figure 1b).
Derivation Stage I: Interpolating Total
Offsite-Produced Consumption
The ASM provides an estimate of offsite-produced electricity consumption by two-digit SIC
for every year of interest, 1974 through the current year. Thus, for the missing years, it
is only nonelectric offsite-produced energy sources that are missing. Total
offsite-produced nonelectric energy for missing years are estimated within each SIC by
two-way indexing, using the FRB production index for that SIC as the adjustment index
basis. The known electric consumption is added to the derived nonelectric total to derive
total offsite-produced energy consumption for the SIC. The totals are summed over all
SIC's to give the manufacturing sector total.
Formal Specification:
For each missing year y, total nonelectric consumption CNys for SIC s is derived from the known starting-year (y = 0) consumption CN0s and known ending-year (y = L) consumption CNLs as
CNys = (L - y)/L (Ays/A0s) CN0s + y/L (Ays/ALs)CNLs,
where Ays is the FRB production index for year y, and AOs and ALs are the production indices for the starting and ending years respectively. The manufacturing sector total CNyM is then obtained by summing over SIC:
CNyM = S CNys.
Finally, total consumption including electricity is computed for each SIC or the total manufacturing sector as:
CTys = CNys + CEys,
and
CTyM = S CTys,
where CEys is the electric consumption known from
the ASM.
Derivation Stage II: Allocating The Nonelectric
Total
The second stage of the derivation consists of allocating the total nonelectric
consumption estimates derived in Stage I among individual fuels. The allocation procedure
starts by estimating the fuel shares, defined as consumption of each
nonelectric fuel expressed as a fraction of the total nonelectric offsite-produced energy.
The set of shares is also referred to as the (nonelectric) fuel mix.
Fuel-specific consumption is then determined by multiplying these shares by the estimated
total nonelectric consumption from Stage I.
The nonelectric fuel shares for the missing years are obtained simply by linear
interpolation between the known years. Data from the known ASM years indicate that fuel
shares for two-digit SIC's do not change much from year to year. Thus, the linear
interpolation estimate should be roughly correct in most cases.
The derived fuel shares include some fuels for which data are suppressed in the known
years. Therefore, cells were filled with values of NA as appropriate. Because of these
suppressions, the interim year SIC-level estimates could not be summed to arrive at the
fuel-specific totals for the manufacturing sector as a whole. Accordingly, it was
necessary to derive manufacturing sector totals using the same methods that were used to
derive the fuel-specific totals for each individual SIC.
Formal Specification:
The (nonelectric) fuel shares for SIC s in year y are represented by the vector bys. For each missing year y, the fuel-share vector is derived by linear interpolation from the fuel-share vectors of the known endpoint-years (y = 0 and y = L):
bys = L-y/L (b0s) + y/L (bLs).
The fuel shares are restricted as follows for individual fuels f:
0 <= bfys <= 1,
and
Sbfys = UTbys = 1.
The vector Cys of fuel-specific consumption amounts Cfys is then obtained by multiplying the fuel-share vector by the total nonelectric consumption estimated in Stage I:
Cys = CNys bys.
To obtain the overall manufacturing sector consumption vector CyM, the same formulas are applied to the manufacturing sector fuel-shares vectors byM. For individual fuels f, total manufacturing consumption CfyM cannot be obtained by summing over SIC's s, because the consumption amounts Cfys are missing for some fuels in some SIC's.
Some of the change in a SIC's fuel mix between two known years (3 to 4 years apart) is
the result of long-term shifts in production practices. Linear interpolation should give a
reasonable estimate of how far these long-term trends have progressed in each missing
year. The limitation of the interpolation is that it will not capture any short-term fuel
shifts that might occur in response to short-term price fluctuations, or other outside
factors.
Missing Consumption Estimates for
Specific Cases
Even for the known years, for which consumption estimates are available from the ASM,
CM, or MECS, there are some missing items. Estimates were not published for some energy
sources in some SIC's, either to avoid disclosing data for individual establishments, or
because the RSE was greater than 50 percent. These gaps in the published data occurred
only for individual energy sources, not for the totals across all energy sources. For this
reason the manufacturing sector consumption totals could be derived by summing over SIC's
in Stage I (total consumption), but not in Stage II (allocating the nonelectric total).
In any case where the consumption estimate was withheld for a particular known year,
SIC, and fuel, the allocation procedure that depended on that information also had to
leave a gap for that fuel. Thus, a gap in the 1981 published estimates resulted in
corresponding missing items for the derived estimates for 1982 through 1984. A gap in 1985
left gaps from 1982 through 1987. A gap in 1988 left gaps from 1986 through 1990. A
gap in 1991 MECS left gaps for 1989 through 1993. A gap in 1994 left gaps for 1992
and 1993.
File Last Modified: 09/08/98
If you are having any technical problems with this site, please contact the EIA Webmaster at wmaster@eia.doe.gov
Return to Derived Estimate Data Choices