Improving the Design of UK Business Surveys Gareth James Methodology Directorate UK Office for National Statistics
2 Overview of presentation Background to recent redesigns at ONS –Outline of previous and new designs –Drivers for project Approach taken and work done Implementation and results in 2010
3 Short-term surveys Set of surveys, cover different economic sectors: –services (MIDSS) –production (MPI) –retail (RSI) Similar aims and systems: Collect similar variables (turnover & employment) Surveys developed independently Differ in detail
4 Short-term surveys: before MIDSS & Gaps: separate survey for employment-only Q employment sub-sample MPI: no separate survey for employment-only M employment to all sample RSI: Q employment to all sample
5 Short-term surveys: after QBS: separate survey for employment-only MBS: Q employment to sub-sample RSI: Q employment to sub-sample Looks like MBS
6 Drivers Change in NACE implies change in design Opportunity taken to review & redesign surveys: all parts of Statistical Value Chain Aim to combine surveys, improve design, standardise and streamline processes Other projects in office: –workforce jobs review –approach to editing –data collection methods
7 Redesign principles Remove unnecessary differences; standardise where possible. One new name adopted for all surveys reduce potential confusion in respondents Frequencies standardised M for turnover; Q for employment Sub-sampling for employment reduce burden Continue with same processing systems No change to total sample size
8 Redesign: sample design & estimation Scope of survey –assess user needs –in-scope / out-of-scope? –industry groups for sampling / publication? Number of industry strata reduced: t/o & emp: 330 → 180 emp-only: 40 → 30 Employment size bands,and estimator type: best choice made for each industry
9 Redesign: sample allocation Overall sample size constraint Neyman allocation principle CV targets for publication groups Estimation of s h 2 –weighted –robust –modelled?
10 Redesign: data collection Review of questions and questionnaires: appropriate questions asked reduced questionnaire types fits with Telephone Data Entry project Cognitive testing for ‘new’ industries
11 Redesign: editing and imputation rules Consistent rules introduced across all industries same approach to non-response same method for imputation Testing undertaken to determine optimum methods and thresholds
12 Backcasting Historical estimates required on new NACE Mix of: –domain estimation, calibration to new NACE groups for recent/current periods –conversion matrices for earlier periods Linking applied to join sections New seasonal adjustment models
13 Backcasting - example
14 Backcasting - example
15 Backcasting - example
16 Implementation in practice (1) Long project: central project management –working groups –early consultation with stakeholders –resources and constraints New systems required: –training and support –change in working practices
17 Implementation in practice (2) Checking quality of results –aggregates first –investigate anomalies: correct population definition? all inputs present? change in assigned classifications? Unforeseen issues –data collection –changed classifications
18 Implementation in practice (3) Customer support –communication –changes to series / website Pragmatic approach Aim to review in year’s time
19 Conclusions Taken opportunity of change in NACE to improve survey designs Redesign work in tandem with other projects: telephone data entry, editing review, workforce jobs Compromises required: not everything can be achieved. Judgement required: not always one obviously ‘right’ solution Many successes; survey quality maintained or improved.