Home > Commercial Buildings Home > Table of Contents > Appendix B

All of the statistics published in this report are estimates of population values, such as the total floorspace of commercial buildings in the United States. These estimates are based on reported data from representatives of a randomly chosen subset of the entire population of commercial buildings. As a result, the estimates always differ from the true population values.

The differences between the estimated values and the actual population values are due to two types of errors:

1. Sampling Errors: These errors are random differences between the survey estimate and the population value that occur because the survey estimate is calculated from a randomly chosen subset of the entire population. The sampling error, averaged over all possible samples, would be zero, but since there is only one sample for the 1995 CBECS, the sampling error is nonzero and unknown for the particular sample chosen. However, the sample design permits sampling errors to be estimated. Estimation of Standard Errors describes how the sampling error is estimated and presented for statistics given in this report and the Building Characteristics 1995 report.

2. Nonsampling errors, which are errors related to sources of variability that originate apart from the sampling process and would be expected to occur in all possible samples or in the average of all estimates from all possible samples.

The sections titled "Annual Consumption and Expenditures" and "Annual Peak Electricity Demand," describe some of the sources of nonsampling error in the Energy Suppliers Survey and how the survey is designed and conducted to minimize such errors.

Nonsampling errors can result from: (1) inaccuracy in the data collection due to questionnaire design errors, interviewer error, respondent misunderstanding, and data processing errors; (2) nonresponse for an entire sampled building (unit nonresponse); and (3) nonresponse on a particular question from a responding building (item nonresponse). ( See Appendix B in the Building Characteristics 1995 internet release for a discussion of nonsampling and sampling errors associated with the Building Characteristics Survey portion of CBECS.)

Most unit nonresponse cases occurred because an appropriate respondent was unavailable or declined to participate in the survey. Item nonresponse resulted when the respondent did not know, or, less frequently, refused to give the answer to a particular question. Unlike the sampling error, the magnitude of nonsampling error cannot easily be estimated from the sample data. For this reason, avoiding biases at the outset is a primary objective of all stages of survey design and field procedures. The wording and format of survey questionnaires, the procedures used to select and train interviewers, and the quality control built into the data collection, receipt, and processing operations were all designed to minimize these sources of error.

The consumption and expenditures featured in this report were based on monthly billing records submitted by the buildings' energy suppliers. The section, Annual Consumption and Expenditures provides a detailed explanation of the procedures used to obtain annual consumption and expenditure estimates from the bills, as well as the procedures used to handle partial or completely missing data. The peak electricity demand estimates in this report were also based on the monthly billing data, as described in the section, Annual Peak Electricity Demand

The section titled "Additional Data Notes discusses special problems encountered when reconciling building and supplier reports on the types of energy sources used, and on gas transported for the account of others. This section also briefly discusses the estimation of energy end use intensities.

Nonresponse

Unit Nonresponse: The response rate for the 1995 Building Characteristics Survey portion of CBECS, 87.5 percent. That is, of the 6,590 buildings eligible for interview, 12.5 percent did not respond at all to the Building Characteristics Survey. The unit response rate for the Energy Suppliers Survey was 84.9 percent. This response rate for that portion of the CBECS varies by energy source. See Appendix A, Data Collection Procedures for more discussion on the nonresponse rate by energy sources.

Weight adjustment was the method used to reduce unit nonresponse bias in the survey statistics. The CBECS sample was designed so that survey responses could be used to estimate characteristics of the entire population of commercial buildings in the United States. Weight adjustment calculated basic sampling weights (base weights) that related the sampled buildings to the entire stock of commercial buildings. In statistical terms, a base weight is the reciprocal of the probability of selecting a building into the sample. A base weight can be understood as the number of actual buildings represented by a sampled building: a sampled building that has a base weight of 1,000 represents itself and 999 similar (but unsampled) buildings in the total stock of buildings.

To reduce the bias from unit nonresponse in the survey statistics, the base weights of respondent buildings were adjusted upward, so that the respondent buildings would represent not only unsampled buildings but also nonrespondent buildings. The base weights of respondent buildings were multiplied by the adjustment factor "A," defined as the sum of the base weights over all buildings selected for the sample divided by the corresponding sum over all respondent buildings. Respondent weights remained nonzero after weight adjustment. Nonrespondent weights were set to zero, because they were accounted for by the upward adjustment of respondent weights.

Item Nonresponse: Item nonresponse is the type of nonresponse that occurs when an item (or several items) is missing in an otherwise completed questionnaire. The Energy Suppliers Survey consisted of four distinct data collections (electricity, natural gas, fuel oil, and district heating surveys) to obtain the 1995 consumption information for buildings in the Building Characteristics Survey. Partial and complete nonresponse in the Energy Suppliers Survey are discussed in the following section, "Annual Consumption and Expenditures."

This report presents estimates of energy consumption and expenditures in commercial buildings during calendar year 1995. These estimates were computed from the annual consumption and expenditures determined for each building in the CBECS sample. However, these "annual" values were not obtained directly from the suppliers for the buildings. Rather, energy suppliers provided monthly billing data, which were used to calculate calendar year consumption and expenditures for each building, according to the procedures described in this section. Also described in this section are the imputation procedures used in cases where the energy supplier survey data were unavailable or inadequate.

To assure that calendar year 1995 consumption would be completely accounted for, the data requested from suppliers were bills covering the period from December 1994 through January 1996. These bills formed the basis for the annual energy consumption and expenditures estimates published in this report.

Billing Data: Ideal and Reality

The basic consumption and expenditures data for were reported for each building by billing period. Ideally, the data for each continuous-delivery energy source (electricity, natural gas, and district heating) used in each sampled building should have been in the form of complete records for every billing period that fell within calendar year 1995, providing complete coverage for 1995 and covering exactly the energy consumed within the sampled building. The data for the discrete-delivery energy source (fuel oil) should have been in the form of complete data records for all deliveries during 1995. For both continuous- and discrete-delivery energy sources, the delivered energy source should have been used entirely within the sampled building.

In practice, though, the billing data often covered more or less square footage than just the sampled building's square footage, or did not match the target time frame, calendar year 1995. There were several common types of discrepancy between the bill coverage and the ideal of a single building and fixed time frame.

• Bill coverage included days in 1994 and 1996 as well as calendar year 1995. This was the typical situation for a complete billing record. Rarely would one billing period begin on January 1 and another end on December 31, 1995.

• Bill coverage spanned at least a 1-year period, but did not include all of 1995. In most such cases, the time frame covered by the bills extended from the middle of 1995 into the middle of 1996. Many energy suppliers maintain accessible billing records only for the most recent 13 months. Thus, at the time of reporting, the data available did not cover the beginning of 1995.

• Bill coverage spanned less than a 1-year period.

• Bill coverage was for several sampled buildings combined. This occurred when no authorization form was obtained to authorize the supplier to provide data for individual buildings. In such cases, the supplier reported only annual totals for a group of sampled buildings summed together.

• Bill coverage included nonsampled buildings or equipment outside the sampled buildings, as well as the one sampled building.

• Bill coverage excluded some of the building's occupants or tenants. This under coverage occurred when the energy supplier had several customers in a sampled building and was unable to identify all of them on the basis of the information provided by the Building Characteristics Survey respondent. In a few cases, energy suppliers were unwilling to release information on all customers in a building, even in aggregate form, without having a separate authorization from each.

• The problem of determining bill coverage was compounded by incomplete dates. In the most common case, the billing period date included a month and year, but not the day of the month.

To reconcile the discrepancies between the ideal billing data and what could actually be obtained, the following seven processing steps were taken:

1. Classify each set of bills, from a particular energy supplier for a particular building, as to coverage in terms of both building and time frame.

2. Complete the billing dates for all bills

3. Annualize bills with full-year time frame coverage

4. Annualize bills with part-year time frame coverage

5. Adjust annualized bills, for building over and undercoverage

6. Impute annual consumption and expenditures for buildings with completely missing data

Each of these processing steps is explained below.

Step 1. Classifying Coverage of Building and Time Frame

This classification was performed by the CBECS contractor as part of the data collection record keeping. To track responses to the mailed Energy Suppliers Survey, determination had to be made whether a response received represented complete data for a building. In many cases, follow-up letters converted initial responses from partial to complete, or more nearly complete. In other cases, the incomplete response was all that could be obtained.

Determining Time Frame: An important aspect of the time-frame classification was determining why data were missing for part of calendar year 1995. The main question was whether consumption had actually taken place during the entire year or was actually zero during the unreported time.

If consumption occurred through the entire year, data might be missing for several reasons. For example, the supplier's active records might not go back far enough or the data may simply have been lost from the supplier's record, even though in general these records did go back to the beginning of 1995.

A more complicated situation occurred when a new customer occupied a building in the middle of the target year. The data provided for this customer, for which the authorization form was signed, would be complete, but the data for the previous occupant, who consumed energy in the first part of the year, would be missing. In any case where part of the year's consumption data were missing, annual consumption would be understated if the reported 1995 data were treated as complete, rather than being inflated to account for the missing period.

The opposite situation could occur if a customer first occupied the building in the middle of the year, with no previous customer occupying the building. In this case, with no consumption during the first part of the year, annual consumption would be overstated if the reported data were annualized as if consumption occurred year round.

A special set of questions on the Energy Suppliers Survey forms was designed to determine if any change in customers had occurred during the target year, and if so how these customers were covered in the reported data. However, most suppliers did not answer these questions. As a general rule, data were treated as complete if they covered a full year, whether calendar 1995 or not. Part-year data were treated as incomplete, unless the supplier specifically indicated otherwise.

Particularly complicated were some electricity and natural gas cases where individual records were provided for each customer in a building with several customers. In most such cases, bills for all the customers covered the same time frame. Sometimes, though, different customers' records covered different time frames. In these cases, it was assumed that the data were complete for each customer, but the customers began or ended service at different times during the year. Aggregate consumption and expenditures were therefore computed for each time period by summing whichever customers had consumption during that period. If consumption was present for a particular customer in a particular period but expenditures were missing (or vice versa) aggregate expenditures (or consumption) were left as missing.

Determining Building Coverage: Building coverage was determined from information obtained from both the Building Characteristics Survey respondent and the energy suppliers. Two types of problems could arise: (1) the energy bills covered more buildings than just the sampled building or (2) the energy bills omitted some of the building's occupants. In the first case, if the Building Characteristics Survey respondent indicated that a particular supplier's bill covered several buildings, the total square footage of buildings on that bill was requested. Then a disaggregation factor was computed as the ratio of the sampled building's square footage to this total square footage. This factor allowed the total reported consumption to be adjusted downward to cover only the sampled buildings. In the second case, when the billing data omitted some customers in a building, an aggregation factor was computed. This factor was usually the ratio of the number of customers in the building to the number reported. Where more detailed information was available, the aggregation factor was the ratio of the total building floorspace to the floorspace occupied by the reported customers. For those cases, the reported consumption of only a portion of the building was adjusted upward to represent consumption in the building as a whole.

Step 2. Complete Billing Dates

Virtually all missing billing dates were one of two types. The first type of dates that were incomplete had the month and year entered, but the day was missing for the beginning and ending dates of all billing periods on a record. These cases were imputed by assigning "16" to each beginning date and "15" to each ending date.

The second type of incomplete dates were missing the day of the month for some, but not all, billing periods. For each case of this second type, the billing periods affected were either bounded (surrounded by billing periods with known beginning and ending dates), or unbounded (either at the beginning or end of the set of billing periods). Any set of consecutive bounded billing periods with missing dates was assigned billing dates that would make all billing periods in the set have as close to the same number of days as possible. Unbounded billing periods were assigned beginning and/or ending dates as needed so that the number of days in each unbounded period was the same as the median number of days in billing periods of known length.

Step 3. Annualizing Full-Year Data

One of the main reasons that the CBECS requested energy supplier data from December 1994 through January 1996 was to assure that 1995 consumption would be completely accounted for in the case of a complete response. However, unless a billing period happened to end on December 31, 1994, or December 31, 1995, consumption as reported by the energy suppliers ran over from the target period of calendar 1995, forward into 1996 and backward into 1994. In general, then, procedures were required to trim away these excess data. For this trimming, different procedures were used for continuous- and discrete-delivery energy sources.

Continuous-Delivery Energy Sources (electricity, natural gas, and district sources): Consumption and expenditures for a billing period extending into 1996 were adjusted by splitting the overlapping period into two subperiods, one running from the beginning date through December 31, the other from January 1 through the billing or meter reading date. Consumption and expenditures were prorated according to the number of days in each subperiod, and the consumption and expenditures for the subperiod that fell in 1995 were included in the total expenditures and consumption for 1995. An analogous procedure was used for a billing period extending into 1994. The assumption that the use of continuous-delivery energy sources took place at a constant rate throughout the billing period may be incorrect for any particular building. However, the procedure should yield approximately unbiased overall estimates.

Discrete-delivery Energy Sources (fuel oil): Billing periods extending outside 1995 did not affect the discrete-delivery energy source (fuel oil) because, for this energy source, all deliveries during 1995 were accumulated. For fuel oil, the ending dates on the bills were used to determine which bills were for deliveries during 1995. No attempt was made to prorate bills, since there was no necessary connection between billing dates and consumption, as was the case for continuous-delivery energy sources.

For both continuous- and discrete-delivery cases where the billing time frame covered a full year but was shifted so that either the beginning or the end of 1995 was not included, a similar procedure was used. In these cases, the data were annualized to a 1-year period within the reported time frame, overlapping as much as possible with 1995.

Step 4. Annualizing Part-Year Data

The annualization procedures for cases that had partial billing data, but less than a full year, were also different for continuous- and discrete-delivery energy sources.

Continuous-Delivery Energy Sources: The number of reported days of consumption was at least as large as the number of reported days of expenditures for almost all sets of bill. Expenditures were annualized using the partial expenditures data and the annualized consumption data.

The part-year annualization method for the consumption of continuous-delivery energy sources depended on the number of days of reported consumption. If at least 331 days were reported, then consumption for the missing portion of the year was imputed by computing the average consumption per day for the adjacent billing period(s), then multiplying by the number of days of missing data. In certain cases, at least 331 days of consumption were reported, but 365 days of expenditures were reported. In these cases, the missing consumption was computed using the average price for billing periods in which both consumption and expenditures were reported. Summing all reported and imputed consumption then yielded the total annual consumption.

Expenditure imputations were performed after completion of all imputations for partially missing consumption since (1) consumption data were usually more complete than expenditures data; and (2) given a value for consumption, the expenditures could be estimated without a great deal of difficulty.

As was true for consumption, the imputation procedure for missing continuous-delivery expenditures was determined by the number of days of reported data. If 30 or fewer days of expenditures were reported, then the expenditures were treated as completely missing. Otherwise, expenditures were imputed based on average prices within the set of bills for a given building. Using bills where both consumption and expenditures were reported, the consumption and the expenditures were summed. The average price was then calculated as the sum of the expenditures divided by the sum of the consumption. This average price was multiplied by the reported (or imputed) consumption to obtain the estimated expenditures.

Discrete-Delivery Energy Source: The billing dates for fuel oil, a discrete-delivery energy source, are not linked to the time of consumption. Thus, the annualized data represent the total deliveries of fuel oil during the year. Furthermore, unlike continuous-delivery bills, discrete-delivery bills tend to be irregularly spaced. Gaps between bills could represent either missing data or periods during which no deliveries were required. The completeness of a set of bills was determined by relying on reports of suppliers. A set of bills was treated as complete if the supplier stated that the bills were complete for the year, and treated as missing otherwise, even if a partial set of bills was available.

Buildings rarely had more than one supplier for a continuous-delivery energy source, such as electricity, but multiple suppliers for fuel oil occurred frequently. If data for one or more of several suppliers were missing, even though responding suppliers had reported all their 1995 deliveries, these buildings were also treated as if no data were available.

Imputations for both deliveries and expenditures made use of the observed price(s). An average price, P_x, for each set of bills, was computed using the data from billing periods in which both consumption and expenditures were reported. If expenditures were missing, the expenditures were imputed as P_xtimes the quantity delivered on date x. For missing deliveries, the reported expenditures were divided by P_xto impute the amount delivered.

Step 5. Adjusting for Building Over and Undercoverage

The annualization procedures for full- and part-year data adjusted for inconsistent time-frame coverage. After the nonmissing consumption and expenditures data were annualized, the annual values were adjusted for building coverage. Where data were requested from the supplier for a single sampled building, but were provided only for a group of buildings including the sampled one, or were provided only for a portion of the building, the coverage adjustment was a simple multiplication of the annualized consumption and expenditures by the disaggregation or aggregation factor. As described above under Step 1 , this factor was computed by the survey contractor directly on the basis of information received on the Building or Suppliers Survey.

Step 6. Imputing for Completely Missing Consumption and Expenditures

In a significant fraction of cases, the energy supplier did not provide the consumption or expenditures data for some or all billing periods or deliveries in 1995. Reasons for missing data included energy supplier refusal; archived, lost, or destroyed billing records; and authorization form refusal on the part of the building respondent. In other cases, the energy supplier provided data, but either the building data were combined with those of nonsampled buildings and could not be disaggregated, or the consumption and/or expenditures were incomplete enough to be treated as missing.

The general approach taken to the problem of imputing annual consumption or expenditures was to annualize the complete or partial sets of bills first, then to use these annualized bills in regression equations to develop imputed values for the data that were totally missing. The regression imputation approach was chosen because data from the Building Characteristics Survey were already available for all of the buildings lacking energy supplier data. The first step was the estimation of missing consumption based on characteristics of buildings. After the consumption had been imputed, missing expenditures were estimated based on the reported or imputed consumption.

Completely Missing Consumption: Each of the energy sources presented in this report was imputed separately, although the overall methodology was similar for all. The consumption imputation method is, therefore, described in general terms, referring to individual energy sources only where necessary. The regression equations were developed primarily to serve as adequate predictors of building consumption based on building characteristics.

The data used to specify regression equations and estimate the regression parameters used for consumption imputation had to meet several criteria. Only cases with essentially complete consumption data were used. For continuous-delivery energy sources, "essentially complete data" included buildings with 331 to 365 days of reported consumption; for discrete-delivery energy sources, only buildings with completely reported deliveries were included. In addition, cases were not used to estimate regression parameters if the information received from the energy supplier included too much data from unsampled buildings (before disaggregation), or the datareported from the building respondent was missing key regressor variables.

The development of regression equations began by examining the distributions of the dependent variable, consumption. Previous experience showed that the error term associated with the reqression procedure is highly skewed in the positive direction. Consequently, the regression procedures used for the 1995 CBECS minimized the sum of squares of the difference between the log of the actual consumption and the log of the predicted consumption rather than sum of squares of the difference between the actual consumption and the predicted consumption Accordingly, the imputed consumption values were calculated using parameter values estimated in two stages: the initial regression of consumption on building characteristics, and a bias correction. The bias correction coefficient was estimated by (1) summing the total actual consumption of cases used to estimate the regression parameters, (2) summing the total of the predicted values for these same cases, and (3) dividing the sum of the actual values (1) by the sum of the predicted values (2).

Completely Missing Expenditures: Similar to consumption imputations, expenditure imputations were performed separately for each of the four major fuels, although the overall methodology was similar. Again, the imputations are described in general terms, referring to individual energy sources only where necessary.

Energy supplier rate schedules are usually structured so that the price per unit of energy is lower as consumption increases. The rate schedule is usually a step function with the definition of steps and rates varying by energy supplier and by rate class. For the CBECS, rate schedules were not collected for the sampled buildings but many suppliers did supply an overall rate schedule for their commercial customers. This was useful in estimating expenditures. In cases where rate schedules were not supplied a statistical procedure was needed to relate the expenditures to the consumption for imputation purposes.

As with the consumption imputations, the data used to specify the form and estimate the parameters of the expenditure imputation equations had to meet two criteria. First, only cases with essentially complete consumption and expenditures were used. For continuous-delivery energy sources, "essentially complete data" included buildings with 331 to 365 days of reported data for both consumption and expenditures; for discrete-delivery energy sources, only buildings with completely reported deliveries and expenditures were included. In addition, cases were not used to estimate regressor parameters if the information received from the supplier included too much data from unsampled buildings before disaggregation.

Once cases with complete expenditures data were chosen, they were used to develop an ordinary least squares regression equation to relate expenditures to consumption and to the fuel price for commercial customers. The independent variables were chosen to mimic a decreasing block rate structure. The resulting fitted equation was used to impute for cases where expenditures were missing.

Peak electricity demand data were requested for the same billing periods for which electricity consumption and expenditures were reported. Ideally, the metered demand represented the maximum consumption rate (in kW) during the billing period. However, two special data problems affect the availability of peak electricity demand data.

First, although virtually all electricity consumption is metered, peak electricity demand is metered where it is economical to do so. In general, peak demand meters are only installed for larger consumers of electricity. Second, in multicustomer buildings, each customer with a demand meter has its own peak demand. Since these peaks would rarely be coincident, the building peak cannot be taken as the sum of individual peaks. However, the overall building peak must be greater than or equal to the maximum customer peak.

Following Step 2 of "Annual Consumption and Expenditures," the peak electricity demand data was processed in three additional steps:

1. Using the billing data, each building was classified as either demand-metered or not demand-metered: For the 1995 CBECS, a building was considered to be demand-metered if the billing data for any account within the building showed metered peak demand. (The 1989 CBECS obtained demand-metered information from both the building respondent and the energy supplier. However, there was considerable discrepancy between the two sources of data. As a result of the building respondent to adequately provide demand-metered data, subsequent CBECS only obtained this information from the energy supplier.)

2. The annual peak demand, the season of the peak, and the annual load factor were determined for each building: For single-account buildings that were determined to be demand-metered, the annual peak demand was taken as the maximum of the billing period peaks. For the few buildings that had part-year electricity billing data, the annual peak was taken as the maximum of the peaks in the reported billing periods. This approach results in a slight understatement of the annual peak, because the actual peak may have occurred during one of the unreported periods. However, since the number of buildings involved was relatively small, the difference between the part-year and full-year maxima would be small in most cases.

   In multicustomer buildings, the overall building peak demand was not available. However, the overall peak had to be at least as high as the highest peak reported for any single customer. In buildings where one customer's peak was substantially larger than that of any other customer, that customer's peak would have been close to the overall peak. Therefore, in processing bills from multicustomer buildings, the peak demand for any single customer was designated as a "partial peak" (associated with part of the building electricity consumption), although the overall building peak was still treated as missing.

   Before assigning the peak to a season, the month of the peak was found. Since the exact time of the billing period peak was unknown, the peak was taken to have occurred in whichever month contained the most days in the billing period during which the peak occurred. Peaks occurring November through April were classified as winter peaks, while those occurring May through October were classified as summer peaks.

   The annual load factor was then calculated, using previously calculated annual electricity consumption, as follows:

   

   Additional Data Notes

   
   

   Energy Sources Used -- Building and Supplier Survey Estimates

   As explained in Appendix A, "How the Survey Was Conducted," the CBECS was conducted in two stages. During the first stage, the building representative was asked which energy sources were used in the building during 1995. In the second stage, the energy suppliers, identified by the building representative during the first stage, were asked to provide consumption and expenditures data. In some cases, contacts with the energy suppliers revealed inaccuracies in the Building Characteristics Survey response as to which energy sources had been used in the building. All statistics in this report on energy sources used are based on the final determination made during the Energy Suppliers Survey.

   When a supplier reported that a particular building was not a customer during 1995, calls were made to the building respondent to determine the reason for the discrepancy. In some cases, a different supplier was identified for the same energy source. In others, it turned out that the energy source had not actually been used; in some of these cases, a different energy source was identified instead. For example, natural gas may have been reported originally, but the callback determined that natural gas was consumed only in a central plant outside the sampled building, while the building itself used district steam, which had not been reported originally. In this case, natural gas would be coded as "not used in the building," and district steam would be added as "used in the building." The net discrepancies between the Building Characteristics Survey and Energy Suppliers Survey estimates for the use of each energy source were small for both the building counts and the floorspace totals.

   The Energy Suppliers Survey was able to correct the energy sources used, only in cases where a supplier had been misreported as supplying a particular building with an energy source. If the Building Characteristics Survey respondent happened to omit an energy supplier, but reported all the other supplier data correctly, the omitted supplier would not have been discovered. However, the number of such cases was probably quite small.

   In some cases, a supplier reported that a particular building had been a customer for a given energy source, but not during calendar year 1995. For continuous-delivery energy sources (electricity, natural gas, and district heating ), the building was classified as not using the energy source. For the discrete-delivery energy source fuel oil, though, the building was classed as using the energy source, but with zero consumption and expenditures for 1995. Thus, for example, those buildings whose respondents reported that fuel oil was used during 1995, but which received no fuel oil deliveries in that year, were included in the count of buildings and floorspace using fuel oil, though they did not contribute to the fuel oil delivery total.

   The revised information on the type of energy sources that were used in the building had an impact on the energy end-use data also. The Building Characteristics Survey data on the type of energy sources that were used for a particular end use were collected in concert with the data on energy sources used. Edit checks on the Building Characteristics Survey data assured consistency between energy sources reported for end uses and energy sources reported at all. However, when the information on energy sources used "at all" was revised during the Energy Suppliers Survey, no new information was obtained on energy sources used for particular end uses. As a result, some energy sources were dropped from a building's list of energy sources used, even though these energy sources had end uses reported. Conversely, no associated end uses were coded for energy sources that were added for a building. For any energy source whose use was changed from "yes" to "no" for a particular building, the use of that energy source for any given end use was also changed to "no." However, the end use was still treated as having been performed in the building. That is, it was assumed that the building respondent correctly reported, which end uses were performed, even if the energy source used for the end use had been incorrectly reported. This approach left some buildings identified as having a particular end use, but with no energy source indicated for that use.

   All Building Characteristics 1995 tables on the Internet, as well as the Public Use Micro-Data have been updated to reflect the latest supplier information on the types of fuels used.

   
   

   Gas Transported for The Account of Others

   The 1995 CBECS collected data on natural gas transported for the account of others (also referred to as "direct purchase gas," "spot market gas," or "transportation gas") from both the building respondent and the natural gas suppliers--both utility suppliers and non-utility suppliers. Gas transported for the account of others is a type of purchasing arrangement where large natural gas users purchase their natural gas directly from a source other than the local distributing company (LDC) or utility. The LDC then delivers the gas to the building via the local pipelines.

   The natural gas survey form requested: (1) the volume of natural gas and expenditures for that gas purchased from the LDC; (2) the volume of natural gas purchased from a source other than the LDC; (3) delivery charges for gas purchased from other than the LDC; and (4) total charges for this gas.

   Since local distribution companies know the total volume of natural gas delivered, the total consumption data seem complete. (If natural gas consumption was completely missing, then the volume was imputed as described in Step 6 of "Annual Consumption and Expenditures"). The allocation of consumption between transported gas and local utility-owned gas was then imputed by hot-decking the proportion of gas that was transported gas. This method allowed imputed buildings to have both transported and local utility gas, as might happen if (1) building demand exceeded the direct purchase contract amount or (2) the building switched to or from a direct purchase contract during the year.

   Estimating consumption and expenditures could become complicated because frequently the LDC filled out the gas transported for the account of others portion of the supplier form since they knew that the gas being provided was transportation gas. Conversely, transportation gas companies, which provide only transported gas did not always fill in the form correctly. They often filled in the first available space, which was intended for utility gas only. Similar confusion occurred when filling in transported gas expenditures, the LDC would be expected to fill out the transport charges column but because this was the only expense collected by the LDC, they sometimes filled it in the "total" column. Finally, since the same volume of gas was reported by the LDC and the transportation gas company, double reporting of volumes sometimes occurred. All these problems were identified by visual inspection of the appropriate records.

   
   

   Energy End-Use Intensities

   The 1995 energy end-use tables provide estimates of the amount of natural gas and electricity used specifically for nine end uses: space heating; cooling; ventilation, water heating, lighting cooking, refrigeration, office equipment and other.

   The end-use estimates had two main sources of data: (1) survey data collected by the Commercial Buildings Energy Consumption Survey (CBECS) and (2) building energy simulations provided by the Facility Energy Decision Screening (FEDS) system. The CBECS provided data on building characteristics and total energy consumption (i.e., for all end uses) for a national sample of commercial buildings. Using data collected by the CBECS, the FEDS engineering modules were used to produce estimates of energy consumption by end use. The FEDS engineering estimates were then statistically adjusted to match the CBECS total energy consumption.

   Click Here for a discussion of the FEDS load estimation methodology, the statistical adjustment procedure, and the remaining steps necessary to produce the final end-use estimates.

   
   

   Estimation of Standard Errors

   Sampling error, as described in the introduction to this appendix, is the difference between the survey estimate and the true population value due to using a random sample to estimate for a population. This difference arises because a random subset, rather than the whole population, is observed. The typical magnitude of the sampling error is measured by the standard error of the estimate. The standard error is the root-mean-square difference between the estimate based on a particular sample and the value that would be obtained by averaging estimates over all possible samples.
   
   

   If the estimates are unbiased, meaning there is no systematic error, this average over all possible samples is the true population value. In this case, the standard error is simply the root-mean-square difference between the survey estimate and the true population value. If systematic error is present, however, this bias is not included in the error measured by the standard error. Thus, the standard error tends to understate the total estimation error if there are non-negligible biases.

   In principle, random errors can be contributed to the estimate by sources other than the sampling process. Such additional sources of random error include random errors by respondents and data entry staff and random unit nonresponse. To recognize these additional sources of variation, the definition of the sampling process can be expanded to include not just the selection of buildings but all steps required to obtain a set of responses. Under this expanded definition, all random errors can be regarded as sampling errors. The procedures designed to estimate the sampling error for CBECS incorporate all random components of the estimation process. Estimation of Standard Errors describes how the sampling error is estimated and presented for statistics given in this report and the Building Characteristics 1995 report.

   Click Here for instructions on how to calculate the Relative Standard Error (RSE) for estimates given in the Detailed Tables. There are no RSE's for the estimates in the Energy End Use Intensities tables because these estimates are modeled.

   
   
   

   Go to Appendix C

   Return to "Table of Contents"

   
   File last modified: February 19, 1998

   Contacts:

   jay.olsen@eia.doe.gov

   Jay Olsen

   Appendix B

   Phone: 202-586-1137

   alan.swenson@eia.doe.gov

   Alan Swenson

   Principal Author

   Phone: 202-586-1129

   Joelle Michaels

   joelle.michaels@eia.doe.gov

   CBECS Manager

   Phone: (202) 586-8952

   FAX: (202) 586-0018

   
   URL: http://www.eia.doe.gov/emeu/cbecs/ceapp-b.html
   
   If you are having any technical problems with this site, please contact the EIA Webmaster at wmaster@eia.doe.gov