What are revisions, and why do we revise

1
What are revisions, and why do we revise?

• Robin Youll
• Director
• Short Term Output Indicators Division
• Office for National Statistics
• United Kingdom
• email robin.youll_at_ons.gov.uk

2
What are revisions, and why do we revise? This
presentation will cover...

• Some basic terminology
• Revisions - a necessary evil
• Why do we need timely estimates?
• Causes of 'error'
• Causes of revisions
• Measurement challenges

3
What are revisions, and why do we revise? Some
basic terminology

• Accuracy
• the extent to which a estimate is close to the
  'true' population value (typically measured by
  its standard error)
• Reliability
• the likelihood of the estimates remaining the
  same/similar
• Relevance
• the extent to which an estimate represents the
  concept which it is trying to measure
• Timeliness
• how quickly we publish our first estimate, or
• how quickly our estimate tends towards the final
  estimate

4
What are revisions, and why do we revise?
Revisions - a necessary evil

• Tension between 'timeliness' and 'reliability'
• Why timely? users demand early estimates of
  key variables
• Why unreliable? limited information at early
  stage
• Therefore..
• Revisions necessary part of the statistical
  process of producing early estimates

5
What are revisions, and why do we revise? Why do
we need timely estimates?

• Demand for short term and 'flash' estimates
• understand position in business cycle
• inform market response to shocks and trends
• aid to regulatory bodies (central banks etc.) in
  setting policies (e.g. interests rates, taxes)

6
What are revisions, and why do we revise? Causes
of error

• Statistical error
• sampling errors
• forecasting or model errors
• Non-sampling errors
• non-response
• coverage issues
• respondent error
• Errors as in 'mistakes'
• rare and generally insignificant

7
What are revisions, and why do we revise? Causes
of revisions

• Late data
• Re-seasonal adjustment
• affects all periods

• Methodological changes
• why ?
• Need to maintain relevance of indicators in
  changing economic conditions
• deficiencies/absence in previous methods
• How ?
• Conceptual changes (e.g. ESA 95)
• Improvements in sampling/estimation, chainlinking
  etc.
• new sources
• Benchmarking/Balancing (national accounts)

8
What are revisions, and why do we revise? Causes
of revisions

• Late data (reducing sampling error)
• Late data replacing forecasts and imputations
  (increased variance, reduced bias)
• Re-seasonal adjustment
• affects all periods
• impact largest where seasonality is
• rapidly evolving
• uncertain
• very marked

• Methodological changes
• why ?
• Need to maintain relevance of indicators in
  changing economic conditions
• deficiencies/absence in previous methods
• How ?
• Conceptual changes (e.g. ESA 95)
• Improvements in sampling/estimation, Chainlinking
  etc.
• new sources
• Benchmarking/Balancing (national accounts)

9
What are revisions, and why do we revise?
Measurement challenges

• Difficulty in measuring the concept of interest
• relevance of sources
• proxy measures
• Limitations of methods
• stability of sampling frame
• impact of rotations
• accuracy - choice of stratification variable
• growth versus levels
• choice of target variables (month-month,
  headline rates, trends, etc.)
• Practical constraints
• data availability
• cost of annual versus short term estimates
• survey respondent compliance

10
What are revisions, and why do we revise? Summary

• Users want early estimates
• The statistical process required for early
  measurement inevitably leads to revisions
• We have control over much of the process
• Revisions are not errors in the sense of
  mistakes