Catalog

Appendix: Sources of statistical discrepancy

Statistical discrepancy (SD) can result from the selection of data sources, the timing of data, or the methods used in estimating the components.

Data sources

The table lists the basic sources of data for the different components of gross domestic product (GDP). The sources can be broadly categorized into three: administrative and financial records/reports from the public and private sectors, sample surveys, and censuses. Data from these sources can either be used directly in the estimation of the GDP components or indirectly as indicators to be fed into a mathematical or statistical model. The expenditure components of GDP rely primarily on administrative/financial data while the production components are mostly drawn from sector- or establishment-based surveys. Recording errors and deficiencies commonly plague administrative-based data while sampling and nonsampling errors are inherent in survey data.

Timing of data

Discrepancies can also be attributed to the difference in the timing of data release from the different sources. Administrative-based data are more frequently released and with shorter lags than survey data. The timing issue not only affects the use of survey results (on which estimates are based) but also the recording of actual transactions. Production, especially in the agriculture sector, may span several accounting periods, while sales are recorded continuously throughout the year and can be confined to the particular year being measured.

Methodological issues

Surveys and censuses are primarily intended to obtain trends and reveal structures, and are not originally designed for national accounts purposes. Since these data may not conform exactly to national accounts definitions nor cover all the items to be measured in the accounts, measurement errors are expected. Such would be the usual case for the measurement of gross capital formation, for which information is culled from administrative and financial reports, as well as from industry surveys.

In many cases, extrapolations from sample survey data are used to approximate population data that are based on a benchmark census and valueadded data that are based on a benchmark year. These extrapolations are then adopted directly in the estimation procedure or as parameters that are fed into a statistical or mathematical model. For example, data on retail trade sales and turnover are often used to estimate personal consumption expenditures. In some cases, technical coefficients from the input-output tables, which give ratios of value added to output or of intermediate consumption to value added, are also applied, especially when information is very limited. Necessarily, these methods of using survey data will have measurement errors. The input-output tables, which are revised every 5 or 10 years or so, may likewise fail to capture changes in productivity and technology.

Conversion of current price estimates to constant price estimates also contributes to differences in GDP between the production and expenditure sides. Deflators are not used consistently. For example, the wholesale price index is used to deflate manufacturing output, the consumer price index—personal consumption expenditures, and the producers' price index—agricultural output.