Official statistics and new sources of data OECD Workshop Paris, - - PowerPoint PPT Presentation

Harnessing new social data for effective social policy and service delivery

Official statistics and “new” sources of data

OECD Workshop – Paris, October 16th 2019 Pascal Rivière INSEE - Head, General inspectorate

Introduction

• On one hand : considerable amount of data are now available,

and this number seems to grow exponentially ; demands also grow quickly

• On the other hand : National Statistical Institutes (NSIs) build

develop aggregated information based on … surveys (and some administrative sources)

• How to improve our data collection ?

– New types of surveys ? – Or couldn’t we simply use the available data and respond more

quickly and efficiently to requests ?

Official statistics

• Providing statistics on a regular basis, with some requirements :

–

Representing a population

–

Comparability through time, between countries

–

Quality requirements (e.g. Eurostat requirements)

–

Use of standard classifications (e.g. NACE : classification of economic activities)

• Standard approach

–

Sampling, questionnaire design, data collection, data editing, handling for nonresponses, variance calculation … and a whole theory supporting the process

• This process ...

–

Is more and more expensive (response rates decreasing) (Meyer & al 2015)

–

Gives the impression of being slow and tedious

–

But “new” ways : web surveys, mix-mode surveys, admin sources, big data ?

Administrative sources : “new” social data

• Many kinds of administrative sources

–

Registers of individuals, business registers, tax returns, social security data, health data, education data...

• Administrative data often derived from administrative declarations

–

This increases quality (compulsory declarations, …) (Rivière 2018)

• More recent uses :

–

Statistics at a more detailed level

–

Linking different themes by linking sources (e.g. health and socio-economic characteristics)

• New opportunities ...and new problems

–

Instability, statistical units (household vs person), lack of control over definitions, lack

• f exhaustiveness, operational use

Linking administrative sources : new horizons

• Need for a general framework for statistics based on administrative sources (Hand 2018)

–

Register-based statistics (Wallgren 2007)

• Record linkage

–

With or without ID : obviously far more complex if no common ID

–

Need for preservation of privacy if ID

–

Dynamic field of research (many papers in the last 5 years)

–

Reference books : (Herzog & al 2007, Christen 2012, Winkler 2015)

–

Requires an infrastructure and a dedicated environment → no super database

• Statistical offices, but also

–

administrative world (for example unified administrative declarations in France or Belgium),

–

academic world (US, UK, Italy, Netherlands, German Record Linkage Center, CASD in France, ...)

• In some NSIs, global vision with a general data strategy on social data based on administrative

sources and surveys :

–

Typically Australia, Canada, Netherlands (Bakker & al 2014)

Examples in the french case

• Recent evolutions post-2010
• Permanent demographic sample : links demographic data, economic data, social

data on 4% of the population (sample drawn according to day of birth)

–

Sources : civil registry data, tax data, data from social nominative declaration

–

Examples of papers : differential mortality due to social class, standard of living

–

Coming soon (2020), health data will be linked

• Secured Data Access Center (CASD) (Perignon & al, 2019, Science) :

–

Remote access to individual data

–

High level of security

–

Record linkage tools

–

Many uses by researchers throughout Europe (more recently US and Canada)

–

International Data Access Network

Evolution of data collection modes in National Statistical Institutes

• Face-to-face surveys : gold standard – high quality

–

Expensive but essential in some cases (e.g. long and complex questionnaires)

• Telephone surveys :

–

development in the 90s, varying a lot from a country to another

• Web surveys, beginning in the 00s

–

Advantages / drawbacks compared to phone surveys

• Mixed-mode surveys (combining different modes) (Dillman & al 2014)
• Administrative data, for different purposes :

–

Replacing a survey, checking data quality, combining with other modes, calibrating

• Big data, main cases

–

Satellite data (agriculture), mobile phone data, cash register data (price indexes)

–

Big data can be used in some cases as a complement but not as the main source of data

Particular surveys

• Some surveys on sensitive topics require particular methodology
• Homeless (Yaouancq & al 2014)

–

Surveys with two sampling steps : sample of services (shelters, social assistance services), and then sample of homeless people who use these services

–

Types of surveys depend on the objective : social characteristics of the population, or simply counting homeless people

–

The analysis shows a significant proportion of people who received social assistance before reaching the age of majority

• Living environment and safety survey (Cadre de vie et sécurité)
• Face-to-face interview - Survey to find out about the acts of delinquency to which households and their

members may have been victims

• The survey also includes people's opinions on security, particularly in their living environment, and in

particular to measure their "feeling of insecurity" → separate module on serious violence was converted into a self-administered computerized questionnaire

As a conclusion

• The NSIs are very familiar with the survey methodology, which makes it

possible to control the quality of statistics with proven but sometimes slow and/or expensive professional techniques.

• Their collection methods have evolved over time to meet these new

challenges and have been able to adapt to specific situations

• New data sources are significantly changing the context
• The possibility of linking an ever-increasing number of administrative

sources with each other or with surveys is the most promising possibility

• Big data in general is more a complementary possibility than a real new

collection method

Thank you