François Brisebois, Statistics Canada International Total Survey Error Workshop June 15, 2010 Improvements to Economic Survey Methodologies to Reduce Revisions.

François Brisebois, Statistics CanadaInternational Total Survey Error Workshop

June 15, 2010

Improvements to Economic Survey Methodologies to Reduce Revisions in Published Estimates

Outline

Revisions Context Revisions vs. Total Survey Error My stories about revisions

1. Revisions in tax data

2. Study about the sources of revisions

3. Quality indicator incorporating revisions

Points for discussion

Revisions

Release of preliminary, revised and final figures• Timeliness-accuracy trade-off

• Users want both

Tracking of the size and direction of revisions helps assessing the trade-off• Coherence of signals from one vintage to another

Context

Sub-annual business surveys* Monthly Survey of Manufactures

• Monthly Food Services Survey

* Monthly Wholesale and Retail Trade Survey

• Quarterly Industry Revenue Indices (Services)

Used by the System of National Accounts

Context

Publication about 50 days after the reference period for monthly surveys, about 90 days for the quarterly survey

Typical revision scheme:• Preliminary, revised 1, revised 2, annual revision

Quality indicators• Sampling variability

• Nonresponse treatment

Context

All use two sources of data1. Surveyed portion (census or survey)

2. Administrative portion Goods and Services Tax (GST) database

Includes sales of businesses Annually, quarterly or monthly

Calendarized to monthly data Data for a given reference month reprocessed every

month

Revisions vs. Total Survey Error

Sources of TSE• Frame• Sampling• Measurement• Nonresponse

Focus on the last two• Measurement: Reported value revised• Nonresponse: Late reporting

“Longitudinal” dimension of total survey error

My stories about revisions

1. Revisions in tax data• GST; same reference month reprocessed every month

• Is the quality of imputed data improving through processing (until we actually receive tax data)?

June YY July YY Aug YY Sept YY Oct YY ...Reference Month - Tax data received

- 1st processing -2nd processing - 3rd processing etc.

Example for a monthly remitter

My stories about revisions2. Study about the sources of revisions

• Monthly Survey of Manufactures

• Systematic downward revisions?

• Examined main contributors to revisions

Survey/Admin * Reported/Imputed Main findings:

No significant trend in revisions

« Survey - Reported » category showed highly unexpected revisions

Small downward trend in administrative data. Why?

Look for improvements (operational, methodological)

My stories about revisions

3. Quality indicator incorporating revisions• Quarterly Industry Revenue Indices

• Design = Census Complex units = Collected data Simple units = Administrative data

• Quality indicator? No sampling error Nonresponse dealt with imputation Non-negligible revision rates for some industries

My stories about revisions

3. Quality indicator incorporating revisions• Quality indicator combining three criteria

a) Combined reported rate of the survey and administrative data portions

b) Variance due to imputation

c) Revision rate

• Approach to be examined to expend to other designs

Points for discussion

Magnitude of revisions: • Should revisions be monitored more closely?

Users’ #1 tool to evaluate/challenge quality

• How are revisions dealt with in your organisation?

Overall indicator of quality• How can we incorporate revisions into our measures of

quality?Typically sampling error and nonresponse/imputation rates