2021 Coding Plans. Paul Waruszynski Office for National Statistics

Transcription

1 2021 Coding Plans Paul Waruszynski Office for National Statistics

2 Outline Census Transformation Programme Coding Occupation & Industry o From 1801 to 2011 o Experiences from the 2011 Census o So why change? o Front-end coding (autocomplete) Next steps for 2021 Census

3 Beyond 2011 programme ONS consulted on two options in autumn 2013 Census once a decade Census based on administrative data and large annual surveys National Statistician s recommendation An online census of all households and communal establishments in 2021 AND Increased use of administrative data and surveys

4 Beyond 2011 programme (2) Way forward agreed by Government The Government welcomes the recommendation for a predominantly online census in 2021 supplemented by further use of administrative and survey data. Government recognises the value of the census and its history as a bedrock of statistical infrastructure. The census provides information on the population that is of fundamental importance to society... Our ambition is that censuses after 2021 will be conducted using other sources of data and providing more timely statistical information... dependent on the dual running sufficiently validating the perceived feasibility of that approach. Minister for the Cabinet Office, July 2014

5 Census Transformation Programme (CTP) 2021 online Census operation Developing and implementing 2021 online Census Enhanced outputs Producing and enhancing census outputs with admin and survey data Research outputs from 2015 onwards Beyond 2021 Acquire new administrative data Develop new methods using admin data and surveys

6 A (brief) history of occupation & industry Occupation first asked in 1801 Industry followed in 1911

7 170 years of industrial change

8 Experience from the 2011 Census

9 How the data were coded Automatic Probabilistic or learned matching Occupation = 74.5% Industry = 59.8% Frontline First stage of manual coding where automatic coding fails Occupation = 22.8% Industry = 32.0% Expert Difficult to code responses and those in Welsh language Occupation = 2.7% Industry = 8.2%

10 Why should we improve? Money o Drive to save money across government Quicker processing time o Aim to release all outputs quicker than in 2011 Better quality data Improved user experience o How will expectations be different from 2011?

11 2011 Census - Time and costs Total cost of capture and coding about 9 million with 465,000 hours worked

12 2011 Census Data accuracy Census Quality Survey allowed ONS to generate agreement rates for each question Census question 2011 Census agreement rates Country of Birth 99% Ethnic Group 95% Religion 90% Industry 74% Occupation 68% National Identity 60%

13 2011 Census Item non-response Unintentional or intentional errors Missing data / partially completed responses Out of range values Not knowing Not wanting to respond Census question Overall Respondent in employment Respondent worked in the past Occupation 4% 2% 7% Industry 12% 7% 17% 17% is largest rate for any question

14 An online first census Principle to guide development and design of statistical and operational processes We believe that just under 10% of the adult population may never be able to gain basic digital capabilities, because of disabilities or basic literacy skills. Government Digital Inclusion Strategy Cabinet Office Increase in online response rate o From 16.4% in 2011 o Aiming towards 75% for 2021

15 Benefits from being online first 1. Reducing handwriting / scanning errors 2. Wider availability of help features 3. Online validation Lower levels of non-response: Census question Item nonresponse Overall Item nonresponse Internet Item nonresponse Paper Occupation 4% 1% 4% Industry 12% 3% 13%

16 How can we improve coding?

17

18 Approaches used elsewhere Vague & unclear responses are difficult to code. For example, manager Australian Bureau of Statistics asks follow-up question when trigger response is given

19 Considerations 1. Comparability 2. Quality of the data collected o How does the quality of self-coded data compare to auto or manually coded data?

20 More considerations 3. What is the burden on the respondent? o Impact of taking a processing task and giving it to the respondent 4. Effort required to create an index o o Can t assume same index can be used, will we need to create something bespoke? Can we use other sources to inform the index, commercial or administrative data? 5. How much money / time is actually saved o If all those who self-code would have been auto-coded how much would that actually save?

21 Where next for 2021? First iteration 2017 Test 100,000 households test sexual identity question Test some design assumptions Second iteration 2019 Rehearsal Large scale rehearsal of systems & services Finalise 2021 Census design Third iteration 2021 Census outputs Conduct field & processing operations Collect & process admin data Disseminate census results, enhanced with admin data Recommendation on future population statistics in 2023

22 Any questions Census Customer Services