Prioritizing Follow-up of Non-Respondents Using Scores for the Canadian Quarterly Survey of Financial Statistics for Enterprises

1
Prioritizing Follow-up of Non-Respondents Using
Scores for the Canadian Quarterly Survey of
Financial Statistics for Enterprises

• Pierre Daoust
• Statistics Canada
• CES, Bonn (Germany)

Statistique Statistics Canada Canada
2
Description of the Quarterly Survey of Financial
Statistics for Enterprises (QFS)

• Provides information on corporate income
  statements (elements of revenues, expenses and
  profits) and balance sheets (assets, liabilities
  and equity)
• Selected for Statistics Canada 2003 Strategic
  Streamlining Initiative (SSI)

3
Description of QFS

• Preliminary estimates available 58 days after the
  end of the reference quarter
• Revisions made
• the following quarter
• at the end of a calendar year
• once QFS estimates can be benchmarked to annual
  data compiled from the Canada Revenue Agency
  taxation statements.
• Emphasis in producing quality estimates for
  preliminary release and first revision to
  minimize magnitude of revisions.

4
Description of QFS

• Sample survey of 5000 large businesses from
  eighty industry groups
• Must units (400)
• Reports by mail, electronically (EDR), or through
  special arrangements (e.g. banks)
• Administrative data used for small businesses

5
Description of QFS

• BLAISE is used to monitor response status and to
  schedule follow-ups
• Critical two-week window where most follow-ups
  occur
• Follow-up of Must units done by a special team
• Out-of-scope, chronic non-respondents, special
  cases not scheduled for follow-up in BLAISE
• BLAISE uses scores to prioritize follow-ups
• Imputing for non-response prior to producing
  estimates

6
QFS Score Function

• Scores produced dynamically in production
• Static version to set up BLAISE
• Dynamic runs prior to each of the critical two
  weeks for follow-ups
• Each dynamic runs takes into account the
  up-to-date collection status of sampled units
• Inspired by the score function used for
  Statistics Canadas Unified Enterprise Survey
  (UES)

7
QFS Score Function

• UES scores derived based on coverage targets for
  total revenue
• QFS scores derived based on
• coverage targets for total revenue and total
  assets for non-financial industries
• coverage targets for total assets for financial
  industries

8
QFS Score Function

• Study was conducted in 2004 to set coverage
  targets
• Units allocated to the three priority groups
• 1 (Must units)
• 3 (Other follow-ups)
• 5 (No follow-up)
• Ranks are calculated for each priority group
  using a multi-step approach
• Progress toward achieving coverage targets needs
  to occur uniformly across industry groups.

9
QFS Score Function

• Multi-step strategy to derive ranks
• In each industry group
• First derive independently the ranks for total
  revenue and total assets
• Combine information from the two independent
  sources (sum of ranks)
• Across industry groups to prioritize units
• Categorize enterprises based on their importance
  to coverage targets
• Integrate the industry and category groups in the
  ranks by properly ordering and scaling units

10
QFS Score Function Rank and high priority
calculations for total revenue in an industry
group (Revenue coverage target 80).
11
QFS Score Function Rank and high priority
calculations for total assets in an industry
group (Assets coverage target 70).
12
QFS Score Function Category groupings within
industry/priority groups
13
QFS Score Function Categorizing and ordering
units in an industry group
14
QFS Score Function Adjusting the
industry/priority/category groups Revenue
coverage target 80,Assets coverage target 70
15
QFS Score Function Scaling the ranks

• Two issues
• Only priority groups (1,3,5) and ranks are
  provided to BLAISE
• Number of units in a priority/category group
  varies by industry
• Need to integrate the industry and category
  groups in the ranks
• Scale the ranks in each industry/priority/category
  group
• Same range of ranks for a priority/category group
  across industry groups
• Non-overlapping sequential ranges of ranks for
  priority/category groups
• Ranks are distributed uniformly in each
  industry/priority/category group

16
QFS Score Function Scaling variables for each
priority/category group
17
QFS Score Function Calculating final ranks for
an industry group
18
QFS Score Function Range of final ranks
19
QFS Score Function

• A small number of units are reassigned to not get
  called for follow-up
• Chosen amongst the highest ranks (lowest
  priorities) from those not needed to achieve the
  coverage targets
• For the dynamic run, cumulative percentages are
  based on only the enterprises that should still
  be followed-up

20
Unweighted Responses Rates (current quarter)
21
Weighted Coverage Rates Total Revenue - (current
quarter)
22
Weighted Coverage Rates Total Assets - (current
quarter)
23
Unweighted Response Rates (Revision next quarter)
24
Weighted Coverage Rates Total Revenue -(Revision
next quarter)
25
Weighted Coverage Rates Total Assets -(Revision
next quarter)
26
Future Plans

• Review current approach for follow-ups, could
  extend follow-up period and have more runs of
  dynamic scores
• Limit to be imposed automatically on number of
  attempts
• Extend scores to editing process
• Continue to improve scores as we get more results
  and experience

27
Prioritizing Follow-up of Non-Respondents Using
Scores for the Canadian Quarterly Survey of
Financial Statistics for Enterprises

• For more information, please contact
• Pour de plus amples informations veuillez
  contacter

Statistique Statistics Canada Canada
Pierre Daoust
Courriel / Email pierre.daoust_at_statcan.ca