Introduction and summary

Warning View the most recent version.

Archived Content

Information identified as archived is provided for reference, research or recordkeeping purposes. It is not subject to the Government of Canada Web Standards and has not been altered or updated since it was archived. Please "contact us" to request a format other than those available.

[an error occurred while processing this directive]13-604-m[an error occurred while processing this directive] [an error occurred while processing this directive]

Reasons for revisions
Reliability versus accuracy?
Why study revisions?

The Canadian System of National Accounts (CSNA) provides a comprehensive overview of developments in the economy for a wide range of users. To remain relevant, the main economic indicators must be accurate and reliable, in addition to being timely. "Statistics Canada defines the quality of information in terms of its fitness for use. This is a multidimensional concept embracing both the relevance of information to users' needs, and characteristics of the information such as accuracy, timeliness, accessibility, interpretability and coherence that affect how it can be used."1

The first quarterly estimate of income-based and expenditure-based gross domestic product (GDP), released approximately sixty days after the reference period, provides a timely estimate of broad economic activity. Although this schedule results in timely and relevant information, this timeliness has a trade-off with accuracy. The initial estimates are based on data available at the time of the release; however, this information is incomplete. As a result of the eventual availability of more complete data, the data undergo a series of revisions that give rise to different vintages of the datasets. Revisions can also occur as a result of new compilation methods or data sources, conceptual or methodological improvements, and, less frequently, the adoption of new international standards. Statistics Canada's Quality Assurance Framework 2002 notes that the tracking of the size and direction of revisions can serve to assess the reliability of early estimates. Revisions analysis also provides a basis for recognizing any biases in preliminary data that could be removed through estimation.

Revisions of the Income and Expenditure Accounts normally occur over a four-year period, in accordance with the current revisions policy.2 Partial revisions are sometimes carried out with respect to periods further back than four years, and historical revisions are conducted periodically, once every 10 to 15 years. Historical revisions are large-scale revisions that include a comprehensive update of concepts, methods, classifications, and statistical innovation, and as such are not a reflection of the underlying quality of the data used in the GDP estimates.

This paper provides background information on revisions within the Income and Expenditure Accounts as well as a detailed revisions analysis of the quarterly growth rate of real GDP. The analysis of revisions seeks to ascertain whether preliminary estimates have been significantly different from final estimates. Analysis of revisions has been carried out on an occasional basis over the last 40 years within the CSNA. Corrective action was taken to improve compilation methods, making estimates more reliable, when earlier revisions analyzes have shown small biases. Earlier studies include the following: An Analysis of the Revisions of the Canadian National Accounts, which examined the impact on revisions of the timeliness initiative (under this initiative, the release was moved up from 90 days to 60 days after the reference period); National Income and Expenditure Accounts: Revised Estimates for the period from 1989 to 1992, which included an historical perspective for 1971 to 1992 and concluded that the downward bias that existed in the 1970s and early 1980s was eliminated and that the absolute size of revisions was reduced by half; and National Income and Expenditure Accounts:revised estimates for the period from 1990 to 1993, which included an analysis of selected GDP aggregates for 1980 to 1993.

The revisions analysis presented here looks at the behaviour of the revisions to quarterly real GDP growth rates for the period 1981 to 2007 with the objective of determining whether a bias exists in the initial estimate of GDP growth. Revisions are examined in regard to measures of bias and dispersion, different vintages, statistical inference, and the economic cycle. The analysis finds that:

  • revisions to the real GDP growth rate have become smaller over time;
  • the largest revision occurs with the second annual revision;
  • there is no statistically significant bias in the estimates;
  • there is a tendency to revise GDP up when GDP growth is increasing  and to revise GDP down when GDP growth is slowing; and
  • there may be a slight time lag around economic turning points.

There is significant interrelatedness between the components of income-based and expenditure-based GDP. A future revisions article will examine revisions to these components.

Reasons for revisions

The GDP estimates are revised for any of a number of reasons. Revisions to income-based and expenditure-based GDP can be classified into four groups: 1) vintage of source data, which are routine revisions that occur as more complete and more comprehensive information becomes available including updates to seasonal factors; 2) changes to the statistical system, such as survey redesigns; 3) conceptual, classification, and definitional changes; and 4) methodological changes, such as improvements to estimation methods.

Revisions due to vintage of source data occur over the four-year revision period as more complete source data are incorporated. The estimates for GDP and its components are compiled on the basis of a vast array of data sources, including survey results and administrative data, often with various lags in the availability of these data sources. Table 1 provides an example of the major data sources used in compiling the series wages, salaries, and supplementary labour income, computing the annual growth rate at each vintage, and determining the timeline in which the source data are incorporated.3

Table 1 Revisions due to different vintages of source data, example using labour income, reference year 2003

The remaining three types of revisions (changes to the statistical system; conceptual, classification, and definitional changes; and methodological changes) can be considered exceptional revisions. Table 2 classifies some past revisions on the basis of these three categories. Changes to the statistical system occurred with the Business Survey Redesign Project in the late 1980s and the Project to Improve Provincial Economic Statistics (PIPES) in the late 1990s. PIPES brought about major changes to data sources used in the compilation of GDP, and these changes were progressively incorporated over a number of years. Conceptual, classification, and definitional changes include the implementation of some System of National Accounts 1993 (SNA 1993) recommendations, including the capitalization of software, and the adoption of North American Industry  Classification System (NAICS) 2002. A major methodological change occurred in 2001 with the adoption of the chain Fisher formula as the official measure of real GDP in terms of expenditures.

Table 2 Classification of past exceptional revisions

Reliability versus accuracy?

Data revisions are a normal part of the statistical compilation process, and ongoing revisions analysis is a component of good monitoring of data quality. The analysis of revisions facilitates transparency and provides users and producers with important information. Users will have the information necessary to assess the reliability of the first published estimates, to determine the 'fitness for use' of the dataset, and to gain a better understanding of the statistical compilation process. Producers (National Accounts statisticians or compilers) will be able to detect measurement issues and to identify areas for improvement.

It is important to note that revisions analysis provides an assessment of data reliability of the initial estimates, not an assessment of accuracy. According to the International Monetary Fund's Data Quality Assessment Framework,4 accuracy refers to the closeness of the estimated value to the (unknown) true value that the statistic is intended to measure. In practical terms, there is no single overall measure of accuracy; accuracy is evaluated in terms of the potential sources of error. One measure of accuracy may be the statistical discrepancy of the GDP income-based and expenditure-based calculations. The size, sign, and variability of the discrepancy may shed some light on the level of accuracy of the estimate. Comparisons with like estimates or with partner-country data may also provide some information on the accuracy of the estimate.

Reliability of the estimate refers to the closeness of the initial estimated value to the subsequent estimated values. Assessing reliability involves comparing estimates over time, in other words, performing revisions analysis. The analysis of revisions can reveal biases in the early estimates. It should be kept in mind that the reliability and the accuracy of the data may not be dependent on the extent of revisions of the data. Data that are not revised are not necessarily accurate and reliable; the converse is also true. Multiple revisions do not guarantee accuracy and reliability.

Why study revisions?

Economic indicators that are released on a timely basis are often used in forecasts and analytical databases. Policy decisions are based on the most recent data; consequently, revisions can have implications for policy makers. Revisions analysis will assist users in making informed judgements about the reliability of the initial estimate and about the likelihood and magnitude of further revisions.5 Ongoing analysis will enable data producers to monitor the reliability of the estimation process. Any bias that is found might indicate that the process needs to be improved. Analysis will also help determine whether revisions are larger at certain points in the economic cycle, indicating that there may be a cyclical bias to the estimates.

Statistics for revisions analysis

Performing regular revisions analysis with a standard set of summary statistics allows users to quickly see the impact of revisions.

Table 3 Suggested revision summary statistics for revisions analysis

The revision interval and analysis period should be clearly stated at the outset of analysis so that users are aware of the scope. Basic summary statistics should include measures of bias and dispersion, as well as the expected revision range. The mean and median revisions provide information on the possible bias of first estimates, whereas the mean absolute revision and the median absolute revision provide information on the dispersion of the estimates. More advanced summary statistics could include significance tests such as the t-test. Table 3 outlines some of the recommended statistics from the Organization for Economic Cooperation and Development (OECD).6

The following sections will present the results of a revisions analysis of quarterly real GDP growth rates for the period 1981 to 2007. This is an update of the Statistics Canada System of National Accounts revisions project by Doris de Zilva (April 2004) with expanded content and analysis. The summary statistics mentioned above will be used for the analysis. As of the date of this paper, the GDP growth rate estimate for 2007 was final in that it was no longer part of the regular revision cycle.


  1. Statistics Canada's Quality Assurance Framework 2002. 2002. Statistics Canada catalogue no. 12-586-XIE.
  2. More information on the Income and Expenditure Accounts revision policy can be found in Appendix A.
  3. For details on other income-based and expenditure-based GDP components, please see Appendix B.
  4. Carson, C., and L. Laliberté. 2002. Assessing Accuracy and Reliability: A Note Based on Approaches Used in National Accounts and Balance of Payments Statistics. IMF Working Paper 02/24.
  5. Revisions analysis does not predict future revisions, but can provide users with an idea of 'normal' and 'acceptable' revisions.
  6. McKenzie, R., and M. Gamba. 2008. Interpreting the Results of Revision Analyses: Recommended Summary Statistics. Contribution to the OECD/Eurostat Task Force on "Performing Revisions Analysis for Sub-Annual Economic Statistics." OECD.
Report a problem on this page

Is something not working? Is there information outdated? Can't find what you're looking for?

Please contact us and let us know how we can help you.

Privacy notice

Date modified: