Jump Menu

• Sample Weights
• Confidentiality
• Imputation Flags

Public Use Microdata Files

The Residential Energy Consumption Survey (RECS) is a national area-probability sample survey that collects energy-related data for occupied primary housing units. First conducted in 1978, the 2005 version is the 12th RECS. The survey collected data from 4,382 households sampled at random using a complex multistage, area-probability design to represent 111.1 million U.S. households, the Census Bureau’s statistical estimate for all occupied housing units in 2005. Data were obtained from residential energy suppliers for each unit in the sample to produce the Consumption & Expenditures data.

The 2005 RECS microdata are available in 12 comma-delimited files for ease of use with multiple software applications. The files are organized according to RECS Household questionnaire section, with additional files containing energy supplier, consumption, and expenditure data. All files contain nine selected variables that are frequently used in RECS data analysis. Each data file is accompanied by a corresponding “Layout File” containing labels and formats for each variable. Users should also refer to the “2005 RECS Variable Response Code Labels” file to reference descriptions of variable codes.

Sample Weights

The RECS sample was designed so that survey responses could be used to estimate characteristics of the national stock of occupied housing units. In order to arrive at national estimates from the RECS sample, base sampling weights for each housing unit, which were the reciprocal of the probability of that building being selected into the sample, were calculated. Therefore, a housing unit with a base weight of 10,000 represents itself and 9,999 similar, but unsampled housing units in the total stock of occupied residential housing units. The base weight is further adjusted to account for nonresponse bias. Finally, ratio adjustments were used to ensure that the RECS weights add up to Current Population Survey estimates of the number of households. The variable NWEIGHT in the data file is the final weight.

EXAMPLE 1: Single Response

The respondent with DOEID = 4373 has NWEIGHT = 14,541. Hence this respondent represents a total of 14,541 households. The respondent used 1007 gallons (GALLONFO = 1007) of fuel oil, thus contributing 14,642,787 (1007 x 14,541) gallons to the estimated national total fuel oil consumption.

EXAMPLE 2: Using NWEIGHT to estimate number of households

There were 413 that used fuel oil in their homes (USEFO = 1). By adding the NWEIGHT data for these 413 cases, the estimated number of households that use fuel oil is approximately 8,402,012.

EXAMPLE 3: Using NWEIGHT to estimate percentage of households

The sum of NWEIGHT over all cases is 111,090,617. This is also an estimate of the total number of households as of July 2005. Hence, the estimated percent of households that use fuel oil (for any use in the home) is (8,402,012/111,090,617) times 100 equals 7.6 percent.

EXAMPLE 4: Using NWEIGHT to estimate total consumption

To estimate the total fuel oil consumption, multiply NWEIGHT times GALLONFO for the 413 cases where fuel oil is used in the home (USEFO = 1), then sum the product over the cases where USEFO = 1. The resulting estimate is 6,236,635,349 gallons. This should be rounded to 6.2 billion gallons.

EXAMPLE 5: Using NWEIGHT to estimate average consumption

The sum of NWEIGHT over cases where USEFO =1 is 8,402,012. Hence the estimated average fuel oil consumption, in homes that use fuel oil, is 8,402,012/6,236,635,349 = 742 gallons.

Confidentiality

These data were collected under the authority of the Confidential Information and Statistical Policy Efficiency Act (CIPSEA), as such EIA, project staff and its contractors and agents are personally accountable for protecting the identity of individual respondents. Thus, local geographic identifiers and National Oceanic and Atmospheric Administration Weather Division identifiers are not included in the public use data files. Heating and cooling degree day values were altered slightly to mask the exact geographic location of the housing unit. As in past years, two variables were also “top-coded”, NUMFLRS and NUMAPTS, to prevent identification of large multiunit residential buildings sampled in 2005. Top-coding resets the value of a variable by truncating a number of cases in the distribution tails to a maximum (or minimum) value.

Imputation Flags

The "Z variables" are also referred to as "imputation flags." Imputation is a statistical procedure used to fill in missing values for respondents that are otherwise considered to be complete. Missing values for many, but not all, of the variables were imputed. The imputation flag indicates whether the corresponding non-Z variable was based upon reported data (Z variable = 0) or was imputed (Z variable = 1 or 2). There are no corresponding "Z variables" for variables from the RECS questionnaire that were not imputed, variables where there was no missing data, and variables that are not from the questionnaire. The missing data codes for the consumption and expenditure data are contained in the "Characteristics of Energy Supplier Data" file.

Specific questions on this product may be directed to:

Consumption and efficiency data