Transcript Guest Speaker - Barry Leventhal - Office for National Statistics

Views of Commercial Users
Barry Leventhal
MRS Census & Geodemographics Group
ONS Summer Workshop
24th July 2013
Agenda
• Introduction – the MRS CGG
• How will commercial users be analysing the 2011
Census data?
• What kinds of analysis should ONS be producing?
• Conclusions
2
The MRS
Census & Geodemographics Group (CGG)
• An advisory board of The Market Research Society
• CGG represents the interests of researchers in census and population statistics
• Formed in 1989 – has helped to shape the last three censuses, and is now
looking beyond 2011
• 20 members from market research agencies, census distributors, targeting
consultancies and other commercial users
• Websites:
– www.mrs.org.uk
– www.geodemographics.org.uk
– Linkedin Group: MRS Census and Geodemographics Group Network
How will commercial users be analysing
2011 Census data?
• Continuous use of census data throughout the
commercial sector
• Applications in research & marketing include:
– Geodemographics
– Targeting local markets
– Targeting customers
– Designing market & social research surveys
Unashamed plug:
CGG Seminar Wednesday 6th November on early use of
Census and other open data - further details on MRS website
Geodemographics Applications
Developed by value added resellers such as...
Geodemographic Analysis
• Geodemographic classifications, e.g. ACORN, MOSAIC
– Applying cluster analysis to small area data
• Profiling catchment areas (and customers)
– Comparing profiles of target area vs. national profile
• Estimating Retail potential
– Applying market research estimates to area profiles
• Input to further products & services, e.g. site location analysis
– regression and gravity models
• And other applications of geographical information
Targeting Local Markets
• Retailers use census data to help make multi-million pound
investment decisions, e.g.
–
–
–
–
Where should new stores be opened?
Which existing outlets should be closed, refurbished or rebranded?
What range of products and services to offer, to meet local needs?
Which local media to employ? – newspapers, posters, radio, leaflets?
• Census analysis includes:
– Measuring size of catchment area population
– Profiling catchment area population by demographics, such as age, social
grade, ethnicity, religion, and by geodemographics
– Estimating market size - by applying market purchasing rates to catchment
area profile
Example: Growth in Sainsbury’s store network
since release of 2001 Census data
Sainsburys Stores
Main Stores
Convenience Stores
March 2003
March 2010
Between 2003 and 2011, Sainsbury’s opened
over 450 new stores and extended over 100
Targeting Customers
• All types of companies with domestic customers use
geodemographics as an input to targeting, e.g.
–
–
–
–
Which customers to target for cross-sell or up-sell?
Which customers are likely to churn?
Which customers should be sent catalogues?
Where to place advertising for gaining new customers?
• Geodemographic analysis includes:
– Profiling target groups of customers, e.g. churners vs. non-churners
– Building and deploying propensity models
– Analysing media profiles
Designing market & social research surveys
• Market researchers use the Census to design, control and
enhance sample surveys:
–
–
–
–
Sample design & execution
Custom analysis, e.g. using Census Microdata
Profiling research data by geodemographics, e.g. products and media
Modelling/integrating Census data with research and customer data
Sample Design & Execution
• A unique benefit of the Census is availability of
demographic information at small area level
• Allows analysis/understanding of areas before
researching them
• Researchers use census data for:
–
–
–
–
–
Profiling areas and planning fieldwork
Estimating penetration rates
Stratifying samples
Setting quota targets
Survey weighting
Social Grade Approximation for the Census
• “ABC1” Social Grade used to be a key omission from the Census for market
researchers and other commercial users
• Government classifications such as NS-SEC have not been adequate
substitutes
• MRS CGG has developed a model to derive approximate Social Grade
based on Census variables
• Model was first built for 2001 Census, redeveloped for 2011
•
Social Grade model was implemented by ONS as derived variable
• Tables on Approximated Social Grade are included in Census outputs from
Release 2 onwards
2011 Social Grade Approximation was
built and tested on market research data
• Original and Approximation
profiles match well for all markets
examined
• Similar targeting decisions would
be made using Approximated
Social Grade
Source: National Readership Survey
13
Example Results – Approximated Social Grade
for Inner London Boroughs
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
AB
C1
C2
DE
2011 Census: Approximated social grade for local authorities in England and Wales
Base: All household reference persons aged 16 to 64 in households
Source: Office for National Statistics
Integration of Census data with
other sources
• The value of Census data increases if it can be integrated
with other sources
• For example, new insights can be created by integrating...
Census small area
statistics
Census
microdata
Insights
Market Research
survey data
Customer database
information
Example - Generating small area
estimates of demand for “eating out”
Market research
measurement of
eating-out markets
Census small area
statistics
Model demand by
demographics
Census microdata:
estimate counts
required for models
demand estimates
for all areas
This method was applied to estimate and map the
demand for 7 eating-out markets, at small area level
What kinds of Census analysis
should ONS be producing?
• Data mining
– Squeeze out every last drop of value from the Census
• Things that users cannot easily do
– e.g. Applying user-generated models to Census database
• More data visualisation
My personal wishlist (1)
Modelling further variables onto the Census
e.g. Income estimates
• Build a model to estimate an individual’s income, on a suitable
survey, using demographics available in Census, e.g. age,
occupation, region, hours worked
• Apply model algorithm to Census database – generate income
estimates for individuals and households
• Summarise income results down to OA level
– Average income levels
– Distributions and cross-tabs by income ranges
• Benefits:
– Approximated income variable, as if collected on Census
– Can then examine relationships with other variables
• Other census users will require different modelled variables...
My personal wishlist (2)
Household Segmentation
• Develop a household segmentation using either market
research data or Census Microdata
• Apply segmentation algorithm to Census database
• Produce outputs down to OA level
– Number of households belonging to each household
• Make segmentation algorithm available to users – for
application to their own surveys and data sources
• Benefits:
– Strengthens link between research data and Census
– More accurate method of demand estimation than using
geodemographics
My personal wishlist (3)
Profile and model Internet Completion
• 16% of census forms were
completed online – distribution
seems to follow a geographical
pattern
• Profile Internet vs. non-Internet
forms by all standard demographics
– develop description of online
completers
• Build model to predict propensity of
online completion
• Apply model to other ONS and
external surveys
• Benefit:
– Enables surveys to be targeted or
stratified by likelihood of online
completion
More data visualisation, e.g...
• Recent release of
postcode-level
Census headcounts
enables dot mapping
within OA’s
• Example shows
Oxford postcodes
coloured by
population size
• Dot maps would be
useful method for
visualising
population
dispersion
• See also US Census
Dotmap
Source: http://opendatareview.wordpress.com
Conclusions
• Commercial users are continuously analysing Census to
underpin research and support investment decisions
• ONS should focus on the kinds of analyses that users
cannot easily do for themselves, including data mining
on the Census database and applying user-generated
analytical models
• Squeeze out every last drop of value from the Census!
Thank you!
Dr Barry Leventhal
BarryAnalytics Ltd
www.barryanalytics.com
[email protected]
Tel: 020 8905 2634