If you would like a downloadable version, in pdf format click here

 

SARS NEWSLETTER
NO. 15 – FEBRUARY, 2001

Samples of Anonymised Records from the 1991 Census

SARs 2001

Funding of SARs 2001

As with the 1991 SARs, SARs for 2001 are being requested by the ESRC under section 4.2 of the 1920 Census Act – i.e. as a special commission and not part of the standard output that is required by the Act. The ESRC Research and Resources Board have set aside an appropriate sum to cover the cost of both an Individual and a Household SAR. They have also allocated funding for Small Area Microdata if the Office for National Statistics accept the case for this.

Support for SARs 2001

ESRC and JISC issued an open tender in October 2000 asking for bids to support and disseminate outputs from the 2001 Census. The results of this competitive procedure will be announced in May 2001 for an August 1 start.

Specification for SARs 2001

We now have a detailed specification for the Individual and Household SARs that is the result of assessing user requests against confidentiality. We have worked on the basis that the level of disclosure risk should not be significantly greater than that which was accepted for the 1991 SARs. ONS have been asked to increase the size of the Individual SAR from 2% to 3% and to reduce the population threshold to 60-70K to enable the separate identification of most LAs.

STOP PRESS
OPPORTUNITIES FOR POST-GRADUATE RESEARCH
THREE ESRC CASE STUDENTSHIPS AVAILABLE WITH CCSR FOR OCTOBER 2001

http://www.ccsr.ac.uk/case/intro.htm

 

Earlier issues of the newsletter have reported proposals for changes to the SARs and work on confidentiality. Below we have summarised the overall parameters proposed for the 2001 SARs. The full details of the proposed specification are available on the CCSR web site http://www.ccsr.ac.uk/sars/2001/.

It is important to stress that there is still time to make minor modifications to these proposals. We urge all users and potential users to examine the proposals carefully and notify us immediately if they have concerns. Often, additional detail requested has not been included in the proposal because it would take the specification far beyond the 1991 confidentiality bounds. However, it is vitally important that no requests have been overlooked and therefore we welcome comments. Please email sars@man.ac.uk.

Summary of main changes

For both SARs we require the addition of variables reflecting the new questions asked on the 2001 Census. Specifically these are:

Lowest floor level of living accommodation
Religion –

Scotland: two questions: religion belongs to; religion brought up in (9 tick boxes + write-in)
England and Wales: 7 tick boxes + write in
Northern Ireland: Do you belong to any religion?; If so, what denomination? If not, what religion were you brought up in?

General health
Year last worked
Number of people at place of work
Provision of care on a voluntary basis
Revised ethnic group question – differences between countries
Revised qualifications question
Travel to study question – Scotland only

Household SAR
It is recommended that we should retain the size and structure of the 1991 file. Within ONS confidentiality constraints, there is no prospect of raising the sample size above 1%. To do so would require losing all geography from the SAR, for which there has been little expressed support. Neither is there any prospect for a move to SAR areas at district rather than regional level.

It is recommended that the Standard Government Office Regions (GORS) should replace the Standard Regions used in 1991. However, retaining the 1991 population threshold of 1.2 million should allow disaggregation of some GORS - for example, distinguishing metropolitan from non-metropolitan areas or, for Scotland, making a division between lowlands and highlands and, for Wales, between rural and South Wales.

Individual SAR
For the Individual SAR we have proposed an increase in the sample from 2% to 3% and a reduction in the population threshold for SAR areas from 120,000 (as used for the 1991 SARS) to around 60-70,000. This is in response to user requests for less aggregation of Local Authority areas and a sample large enough to support analysis at LA level.

Under the 1991 specifications, almost two thirds of Local/Unitary Authorities in 2001 would (using 1991 population counts) have populations below the 120,000 threshold and would consequently have to be grouped with other authorities. At a 90,000 threshold this falls to less than 25% and at 60,000 to just over 10%. Among the larger metropolitan districts there is a strong case for output areas at sub-district level.

Confidentiality work reported in earlier newsletters shows that the disclosure risk of the changes outlined increases only slightly by comparison with the 1991 specification and the 1991 Individual SAR was, in fact, rather safer that than the original estimates made by Marsh, Skinner & colleagues for the 1991 SARs. Full details are available in Occasional Paper 16, downloadable from the CCSR web site (http://www.ccsr.ac.uk/).

A number of other changes have been proposed in response to user requests:

Migration data: In 1991 the Individual SAR provided only the region of former residence for an individual who had moved in the previous 12 months and a banded variable giving 13 categories of distance moved. A set of alternatives have been proposed to ONS:

• the individual SAR area of origin;

• an area classification as used for migrant 'destination'

• a summary classification of area type (e.g. metropolitan/ non-metropolitan areas)

• a categorisation of the type below:
                  Move within a district
                  Move between districts within a county
                  Move between counties within a region
                  Move between counties but within country
                  Move between countries but within UK
                  Move from outside UK

In addition we have asked for more bands on distance moved, especially at the near-distance end, accounting for the majority of movers. We are grateful to Professor Paul Boyle, St Andrews University, for compiling these proposals.

Occupational coding: the 2001 Census schedule asks for occupation and work-place information from all respondents aged 16 and over who have ever worked. However, for financial reasons ONS have decided to restrict coding on occupation, industry and socio-economic class to respondents:
        aged 16-65 who have worked in last 5 years; or 65-74 and currently working

This means there will be no information on occupation, industry or social class for:
        16-65s who last worked more than 5 years ago
        65-74s not currently working
as well as those over 74 who were not asked for this information.

This has important implications for the research value of the SARs. We know from analysis of the 1991 SARs that restricting occupation to those with a last occupation 10 years ago leads to serious bias. For example, about 25% of households, nationally, had a head of household with no recorded occupation and thus no social class. In Liverpool this went up to 37%. If coding is further restricted in the 2001 Census to those who held a job in the previous five years these figures will be considerably higher.

The percentage of those without an occupation is highest for older people, for women, students, the unemployed and those on Government schemes. These groups will be disproportionately omitted from any analyses that require a measure of social class. Social class based on last occupation continues to be of considerable relevance in predicting outcomes such as ill health. With no income question in the census there are few alternative measures of life-style and life-chances. Table 1 shows the profile of those who recorded no occupation in the previous 10 years in the 1991 Census. We expect that, when occupational coding is further restricted, these percentages will be considerably higher – particularly for those aged 65–74 who will only have a coded occupation if they are currently working.

Table 1: Percentage of individuals aged 16 and over with no social class
2% GB SAR 1991

*This refers to people who were not asked for an occupation in the 1991 census.
A further 1.4% of people with a job in the last 10 years failed to give a usable response.

Whilst the 2001 Census will provide a new question which allows us to calculate how long ago someone last worked, we will not be able to relate this to the job they last held – unless it happens to be less than five years ago and they are under 65!

Imputation: as part of the One Number Census individuals and households will be imputed so that the 2001 Census database will represent a complete population count, corrected for under-enumeration. The SARs will be drawn from this database.

For some SAR users it will be important to be able to identify imputed individuals or households and for this reason ONS have agreed to flag the records for individuals that have been imputed. However, adding flags to the standard database will result in a file that is more complex to analyse and many users will not know how they should deal with imputed records. Therefore we propose to have an additional research dataset that records the flags for imputed individuals and households. This would be made available to users on request and could be merged into the main dataset for those who wished to use the information.

Area-level classification: the area-level classification added to the 1991 SARs has proved very successful and we have requested ONS to add a similar classification to the 2001 SARs. No decisions have yet been made over which classification should be added and views of users would be most welcome.

Timetable for delivery of SARs: ONS do not yet have a firm timetable for the delivery of census outputs. However, we should expect the SARs to be released by mid-2003 at the latest. The 1991 SARs were first released in June 1993 and we expect the 2001 outputs to be delivered rather more quickly.

 

Small Area Microdata Samples (SAMS)

In previous issues we have reported on an ESRC-funded project to develop small area microdata that could provide greater geographical information than the SARs although less individual detail. This work is now complete and a copy of a paper setting out the full results can be downloaded from the CCSR web site. The work has resulted in an initial specification that provides levels of confidentiality roughly comparable with the 1991 Individual SAR. This is shown below and is based on 1991 variables.

The project to develop small area microdata covered the following tasks:

1. established the importance of small area effects in a modelling context
2. demonstrated the value of combining small area microdata with aggregate census tables
3. assessed the disclosure risk of a range of different microdata specifications
4. consulted potential users on the specification of a small area microdata file
5. established the methods for building a SAM geography
6. produced a recommended specification for a SAM

The Office for National Statistics made available to us, under strictly controlled conditions, population data from the 1991 Census for seven local authorities^1. This provided the basis for extracting prototype SAMs with different sampling fractions, different population thresholds and different variable specifications. In addition, Professor David Martin used partly synthetic data for Cardiff and Hampshire to assess alternative geographical bases for building SAMs.

i. The importance of small area effects in a modelling context
Modelling unemployment with area distinctions
We used a 5% sample from the 1991 census test data for seven local authorities to test a multilevel model to predict unemployment. Using a full set of individual-level controls, the decomposition of variance showed considerable variation at the two levels of geography used – SAM areas of 30K population and the local authority district.

A second analysis using a hierarchical model with four different levels within one LAD showed substantially more variation at the lowest level of geography (7.5K threshold) than at the higher (30K threshold). From this we concluded that, at least for this application, there were important sub-district area-level effects to be included.

Establishing the role of area in obtaining a properly specified model
Further multivariate analysis of the probability of unemployment using a breakdown of four ethnic groups demonstrated the unreliability of estimates that did not take geography into effect. Models without geography produced significant estimates for the Indian group by comparison with the white reference group but in all other models these estimates were not significant.

ii. The value of combining small area microdata with aggregate census tables
a) to improve the precision of estimates based on the microdata and
b) to add extra dimensions to 100% Standard Tables and Small Area Statistics by synthetic estimation.
One of the disadvantages of census microdata identified by Local Authority users is the lack of precision in estimates based on small samples. However, knowledge of the univariate or bivariate distributions of key variables based on the 100 % SAS or LBS tables (termed CAS in 2001) can be used to increase precision. Conversely, microdata can be used in tandem with the 100% tabular data to provide synthetic estimates for cross-tabulations not available in the 100% data. Five different models were compared:

1. Using only SAM and simple logistic regression
2. Using SAM only in a multilevel logistic framework
3. Combining SAM with marginal information from the SAS/LBS in a multilevel model. In this model, information from the 100% aggregate tables for each SAM area is included as an explanatory variable
4. Combining SAM with marginal information from the SAS using a logistic model
5. Combining SAM with information from SAS for two margins in a multilevel model.

Each model was assessed with SAM for 30 small areas in one LA District to estimate the unemployment rates for non-whites in SAM areas of 15K population with a 5% sample. These results show that imprecise estimates with large confidence intervals that arise from the SAM can be greatly reduced by combining the SAM with other census data under a model based approach. They also indicate how aggregate statistics can be extended using microdata.

iii. Assessment of the disclosure risk of a range of different microdata specifications
The immediate concern with microdata at small area level is the risk to confidentiality. An extensive set of risk analyses were conducted using a range of different file specifications. The final analyses used a combination of sampling fraction, population thresholds and individual detail that were the result of earlier work to establish the parameters which produced a level of risk comparable with that for the 1991 Individual SAR.

Two different measures of assessment were used, designed to reflect the motivations of an intruder trying to identify a respondent by matching against an external file. The measures were: 1) the probability of a correct match given a unique match and 2) the probability of a unique correct match. Both were assessed under several disclosure risk scenarios. The SAM files represented a 5% sample with population threshold of 5K and 10K, respectively. The 1991 Individual SAR specification formed the standard against which the disclosure risk was assessed. Both SAM showed risk probabilities of the same order as the SARs, although that the risk was consistently higher with a 5K threshold than with a 10K threshold.

iv. Methods for building SAM geography
The new Census Output Areas (OAs) and wards were both considered as the building blocks for a new SAM geography. Both have strengths and weaknesses. Advantages of using OAs relate to their small population size providing greater flexibility in zone design; their close association with postal geography facilitating data matching and the ease of aggregation to census wards, districts and other statutory areas. However, aggregations of OAs within local authority districts may result in SAM areas which do not equate, nor aggregate neatly, to any other geographical units (including wards).

The main advantages of using wards are that the ward boundaries will be known in good time for SAM area creation, and the much smaller number of wards (10,000) compared to OAs will enormously reduce the computational burden. Some large wards will be retained as SAM areas in their own right. A SAM based on wards would ensure straightforward data matching with ward tables based on 100% data from CAS – an important condition for some of the multilevel modelling applications that use 100% data to improve precision of sample estimates.

However, large numbers of wards close to or just below the SAM population thresholds will significantly reduce the available range of workable SAM area configurations, resulting in wide ranges in SAM area population. Combining two wards with just below threshold populations will result in a SAM area with almost twice the threshold population. Ward boundaries are also subject to frequent changes.

Empirical assessment of tests using wards and Output Areas as the basis for SAM areas was conducted by David Martin for Cardiff and Southampton. Extensive user consultation took place through a series of workshops and the SARs Newsletter. The conclusion of this evaluation was that, on balance, wards provided the most advantages. However, we recommend that wards with populations at least double the threshold (7-10K) (‘super-wards’) should be split into separate SAM areas. This subdivision would reduce the variance in the mean SAM population, thereby producing more homogeneous SAM areas. More, smaller SAMS would also give greater flexibility in building to higher level geography and in multilevel modelling applications.

We therefore suggest a two stage procedure:

Stage 1: the simple ward SAM would be produced using Automated Zoning Procedure (AZP) to group sub-threshold wards and delivered in the first wave of 2001 Census output.
Stage 2: AZP would be targeted on those districts containing ‘super-wards’ - identified during stage 1 – and the disaggregation of ‘super-wards’ would take place after release of the initial SAM. The additional detail would be supplied as an extra variable and merged into the main database.

vi. A recommended specification for small area microdata
A SAM specification has been produced with a 5% sampling fraction and a population threshold between 5-10K, based on wards. The confidentiality risk from such a file is assessed as broadly comparable to the risk of the 1991 Individual SAR. The specification is based on 1991 variables and should be seen as indicative only. If these proposals are accepted by ONS, the details would need to be re-worked based on the 2001 Census.

A table showing the specification for SAM based on 1991 Census appears below.

 

ALTERNATIVE HOUSEHOLD CLASSIFICATIONS

Earlier newsletters have reported the work under this project which has been developing additional household classifications to those planned for the 2001 Census output. The project is now drawing to a close and a summary of the results so far is enclosed with this newsletter.

 

Specification for Small Area Microdata based on 1991 Census

 

What have we learned from the SARs?
MANDEC Centre, University of Manchester
Monday 25 June 2001
10.00 - 4.00

A one-day conference to showcase some of the most important and innovative research findings based on the 1991 SARs.

We invite papers under the following streams:

1. Methodological innovations using the 1991 SARs. We particularly welcome papers linking the SARs with other census outputs
2. International comparisons
3. Research papers which exploit the strengths of the SARs, for example:
           household and family composition
           ethnic differences
           area-level classification

Each paper will have a 30-minute session for presentation and questions. We expect electronic versions of papers to be available at least one week before the conference and will put them on our web site.

Deadline for submission of abstracts: 30 March 2001; please email to angela.dale@man.ac.uk
We will let you know if your paper has been accepted by 12 April.

Cost: £30 for academics; £50 for non-academics; includes refreshments and papers
There will be no charge for those giving a paper (two presenters per paper, maximum)

If you need to stay overnight, please ask for our accommodation list.

For further details and on-line booking form, click here

 

PUBTRAWL

Publications based on the SARs

We are enclosing a list of all the publications based on the SARs that have been notified to us. This is important evidence of the research value of the SARs and provides the ammunition by which ESRC can justify the funding for 2001 SARs. This list is also available on our web site. If your publication is not here, please send us the details or complete the form on the web site (email: sars@man.ac.uk or http://www.ccsr.ac.uk/cmu/index.htm).

 

USING THE SARS FOR TEACHING AND LEARNING

http://www.chcc.ac.uk/

CCSR is a partner in an £800k project funded by Joint Information Systems Committee (JISC) to deliver census-based learning and teaching to the UK Higher Education sector.

A working knowledge of how to access and analyse data from the Census of Population statistics is of enormous benefit to students from a wide variety of academic disciplines. The census can provide answers to many questions of relevance to undergraduate study as well as equipping students with a marketable set of skill that they can use after graduation.

Project Objectives

The main objective of the project is to considerably increase use of historical and contemporary census data (CHCC) in learning and teaching. These comprise the 1991 Census Area Statistics, the 1991 Samples of Anonymised Records and the Historical Data Collection (comprising data from 1851 and 1881 Censuses). This will be achieved by:

• improving accessibility to the primary data resources.
• developing an integrated set of learning and teaching materials.
• improving awareness about the contexts in which Census data can be used in learning and teaching.
• integrating contextual materials; providing access to web-based data exploration/visualisation tools and developing resource discovery tools.

Teaching and Learning Materials

The learning and teaching materials will be developed through close consultation with the user community. This process was launched with a highly successful consultation workshop in London on 31 Jan 2001, which identified a clear demand for flexible materials that could be adapted to a range of different learning and teaching settings. A catalogue of teaching and learning ‘units’ will be developed and organised by theme. They can be slotted into an existing course, assembled to provide a complete course, or used as a self-study learning pathway. The teaching materials and data will be provided in a range of different delivery formats (downloadable and on-line) to suit the varying needs of users working in different learning and teaching environments.

CCSR will develop learning and teaching materials specifically for the SARs as well as contributing to a set of inter-disciplinary teaching resources that will draw on the full spectrum of census data. The materials based on the SARs will include units providing methodological training in the use of microdata, including exploratory data analysis, basic statistical analysis and working with hierarchical data. These will build on CCSR’s experience in delivering training via short courses. It will also include topic-based units that provide exemplars of using SARs to address a range of substantive issues across a range of different disciplines. Candidates for these include an exploration of ethnic differences, employment and unemployment and gender differences. We will hope to involve subject experts to help write these materials.

A common web portal to the census

The Data Archive will develop a Census portal to provide direct web access to all the learning and teaching materials and the various data extraction, exploration and visualisation tools developed.

A web-based data exploration interface to the SARs

As part of this development, there will be major developments in web-based access to the SARs. This will include a web interface to small subsets of the SARs (mini-SARs), providing unregistered users with immediate hands on experience of working with the SARs. The mini-SAR datasets will comprise a set of carefully defined multi-way cross-tabulations related to particular topics.

The project started in September 2000 and runs for three years. Other project partners are: MIMAS and the Census Dissemination Unit, the University of Manchester; The School of Geography, University of Leeds; The History Data Service, The Data Archive, University of Essex; The LTSN Centre for History, Archaeology and Classical Studies and University of Glasgow.

CCSR staff involved in the project include Mark Brown and Mark Elliot who will, respectively, provide oversight of the development of teaching materials and the web-based materials. Celia Russell has been appointed as web development officer and Jo Wathan will work on the development of teaching materials.

If you would like to ensure that the materials we develop will be of use to you, please contact any of the team. We would like to hear how you might use the SARs in your teaching and, particularly, if you have suggestions for the topics areas or if you would like to test out the materials.

 

Interdisciplinary Perspectives on Analysing the Life Course

A series of six ESRC-funded seminars focusing on the theory, methods and practice which bring together perspectives across the range of relevant disciplines

Seminar 2 - 'Theorising across disciplines'
10.00-4.30
Thursday 5 July 2001
Beves Room, King's College, Cambridge
Convenor: Jackie Scott, University of Cambridge

Programme to include:

Early experiences and the life course
Dale Hay, University of Cardiff
Life course determinants of adult and old age mortality 1600-1900
Richard Smith, University of Cambridge
Life course change and social change: insights from cohort comparisons,
John Bynner, Institute of Education
The contextual challenge: life course, life span and history
Speaker to be arranged

Further details and booking form:
http://www.ccsr.ac.uk/rschproj/iplc/iplchome.htm

The seminar series addresses questions and issues which have arisen as part of the current and recent research of members of the organising group and which would benefit from extended cross-disciplinary discussion. In particular, there is a need for informed methodological discussion that takes cognisance of recent advances in qualitative methods and statistical modelling.

Organisers:
Professors Angela Dale, Andrew Pickles and Mike Savage, University of Manchester; Dr Jackie Scott, University of Cambridge; Dr Barbara Maughan, Institute of Psychiatry, University of London; Ms Jane Elliott, University of Liverpool

 

Further seminars include:

October 2001 London
Developmental pathways - perspectives on delinquency and criminal careers
December 2001 Manchester
Different methodological perspectives on the lifecourse
July 2002 Cambridge
Comparative perspectives
October 2002 London
Title to be confirmed

There is no charge for attending any of the seminars in this series

 

CCSR SEMINAR PROGRAMME
SPRING 2001

Monday 26 February
'Methods of Ecological Inference'
Dr Eric Beh, School of Mathematics and Applied Statistics, University of Wollongong

Monday 12 March
‘Employee compared with employer responses about entitlements to family-friendly working arrangements: Analysis of the 1998 Workplace Employer Relations Survey (WERS)’
Professor Shirley Dex, Judge Institute of Management, University of Cambridge

Monday 19 March
‘Some theoretical and practical issues in coverting and merging data from different geographies to estimate new datasets’
Dr Ludi Simpson, CCSR, University of Manchester

Wednesday 21 March
‘Recent histories of intimacy: Family, Kinship and Childhood’
Lynne Jameison, University of Edinburgh
(Sociology Department Seminar - 3.30 - 5.00, Roscoe Building, Room 4.9)

Tuesday 27 March
‘The effects of parental employment on children’s educational attainment’
Marco Francesconi, University of Essex

Monday 2 April
‘Job insecurity and gender: the case of the UK television industry’
Valerie Antcliff, CCSR, University of Manchester

Monday 23 April
‘Strength in numbers’
Paul Meszaros, Bradford Community Statistics Project

Monday 30 April
‘Equal oportunities: the role of legislation and public policies in women’s employment in China’
Fang Lee Cooke, Manchester School of Management

Monday 14 May
‘Cultural capital: an alternative basis for a social classification’
Tom Chippendale, CCSR, University of Manchester

Monday 21 May
‘Power and privacy in a modern liberal democracy.’
Cate Heeney, CCSR, University of Manchester

Seminars start at 4.00pm
in Room G29, Ground Floor, New Wing
Faculty of Social Sciences and Law, Dover Street
Wine/soft drinks provided
All are welcome (though visitors from outside Manchester may wish to contact us to confirm the seminar details before travelling)
tel: 0161 275 4721  email: ccsr@man.ac.uk

 

Courses and Workshops

Research Design and Data Analysis
Our new, enhanced Short Course Programme provides a range of courses in research design and analysis, all with a practical emphasis and applied focus. The programme is structured so that participants may either select an individual course which meets their needs, or build up their expertise through a portfolio of courses. Places on each course are limited to a maximum of 20.

The courses on data analysis are all PC-based and provide participants with the opportunity to complete detailed practical exercises on their own PC. Staff from CCSR or MIMAS are available to provide help and advice throughout the practical sessions. Each course will be supported by full documentation.

Level 1
        SPSS for Social Scientists Sep 2001; Jan 2002; Mar 2002
        Surveys & Sampling Oct 2001; Jan 2002
        Exploratory Data Analysis Oct 2001; Jan 2002
        Basic Data Analysis Nov 2001; Feb 2002

Level 2
        Multiple Regression Jan 2002; Jun 2002
        Logistic Regression Feb 2002, Jul 2002
        Analysing Hierarchical Surveys Nov 2001

Level 3
        Factor Analysis Mar 2002
        Cluster Analysis Apr 2002
        Multilevel Modelling May 2002
        Design & Analysis of Complex Surveys Jan 2002
        Longitudinal Data Analysis Apr 2001; Apr 2002

For full details of courses, including on-line booking and cost, please refer to our Short Course Programme web page - http://www.ccsr.ac.uk/courses/shorthome.htm
or contact Nasira Asghar on 0161 275 4736.

 

MSc SOCIAL RESEARCH METHODS AND STATISTICS

CCSR runs an MSc Programme in Social Research Methods and Statistics (SRMS). The programme provides a firm grounding in advanced quantitative methods within a social science context through lectures, seminars, project work and practical workshops. For more information about the content and structure of the SRMS programme, please see our programme website - http://www.ccsr.ac.uk/postgrad.htm, or contact Margaret Martin on 0161 275 4589.

 

PhD STUDENTSHIPS

CCSR invites applications for three newly-awarded ESRC CASE studentships available for October 2001. The collaborating organisations are: Bradford City Council (Race & population - a statistical demographic approach); DfEE (Barriers to employment for South Asian women); and the Office for National Statistics (Confidentiality issues associated with neighbourhood statistics). Further details can be found on http://www.ccsr.ac.uk/.

 

Created by Margaret Martin, 20/3/01