Hands on the Census: Microdata from the 1991 Census of

Population in Britain

by Catherine Marsh '

Department of Sociology and
Department ofEonometrics

University ofManchester

The information dilemma
A basic dilemma confronts the producers of data in the

public sector. On the one hand, they face users demand-
ing that information collected at public expense be made
available to the research community in ever increasing

detail. On the other hand they face those concerned with

privacy and confidentiality worrying about the possible

release of identifiable information about individuals. In

short, they have to reconcile competing moral claims:

they are caught in the middle between one group of

citizens asserting a right to information and another

group asserting a right to privacy.

The audience at this lASSIST meeting, as professionals

in the business of information acquisition and dissemina-

tion, will have rehearsed the arguments many times about

the rights to information. It is important to remember
that the arguments for a right to privacy and confidential-

ity are also strong, but have changed their character

somewhat during the last decade, and have had a pro-

found influence on the preparedness of state authorities

and census agencies to release microdata to the research

community.

The stale is no longer the only, and perhaps no longer

even the major, collector of systematic personal data on

individuals. The expansion of information technology

has led to a new private industry of information collec-

tion and management. Many companies take as their

base the publicly available state collected information

such as electoral registration lists. Some then may link to

this other published records e.g. on bankruptcies and
criminal records. Others specialise in the collation of

information from pull-out questionnaires in magazines
and so-called guarantee registration cards, others bring

together personal financial information from major credit

card companies and chain stores. Sometimes the aim of

this data collection is to assess the credit-worthiness of

individuals. Sometimes it is to facilitate modem direct

marketing techniques, targeting prime areas for particular

mail-shots. Combined with advances in telephone

technology, these techniques have become big business.

In Britain, several companies have datafiles containing

information on over 90% of households. Most of this

information is referenced by a combination of name,
address and post-code.

There is widespread public concern about the activities of

these commercial companies. The British public is very

hostile, for example, to the sale of election registration

lists to outsiders (Campbell 1987). In the lead up to the

1991 census, there were media programmes and newspa-

per articles which expressed worry about the potential of

census information about small areas to be linked to

other private information databases both of these private

companies. When the recent Census (Confidentiality)

Bill was being debated in Parliament, certain members
suggested amendments which would have made the

linkage of census information to fine-grain post-coded

information illegal (Computer Weekly 31 Jan 1991).

The attempts did not succeed, although plans to release

statistics for post-coded areas in England and Wales were

modified.

Both the political right and left have turned to the state to

protect citizen liberties and rights to privacy. Through-

out the 1980s, most European countries, unlike the US
and Canada, have enacted legislation to give citizens

rights with respect to databanks of information which

may be held about them. The Data Protection Act of

1984 gives individuals in the UK rights to find out what

computerised personal information any organisation

might hold about them, to challenge it if wrong, and

claim compensation if they suffer as a result Owners of

machine-readable lists are obliged to register any file of

identifiable personal information they hold, and to say

for what purposes it is held; the register is open for

public inspection.

The Data Protection Registrar attempts lo police the

activities of the information industry; he has, for ex-

ample, recently attempted to curb the unrestricted use of

address-based information for credit-redlining. How-
ever, in some ways, the individualistic focus of the

existing data protection legislation weakens it as a tool

for those seeking to ensure that information released

could not be linked to private databanks. Rights to know
are restricted to rights granted to individuals lo find out

what is held about themselves; there is no right for

someone wanting to establish how much census informa-

tion it is safe to release to obtain answers lo such critical

questions as how many individuals are covered in the

databank and which census variables are held which

lASSIST Quarterly



would be available for matching purposes. The format of

the Data Protection register does not enable it to be used

as a source for answering such questions. Nonetheless,

in the Office of the Data Protection Registrar there exists

a team of individuals whose job it is to know precisely

who has what computerised information about whom,
and to police the workings of the Act; in practice they

know in broad outline the information gathering activi-

ties of all the major data collection companies.

As well as worry that information may be passed on to

the commercial sector, the other major public concern

about census and survey data relates the use to which the

government itself might put the information. Worries

[lave been expressed in Britain at the time of previous

:ensuses about the passing of identifiable census infor-

mation to immigration officials or to social security

officers. A similar set of concerns were expressed in

1991 over the possibility that census information might

be passed to Community Charge officers, responsible for

compiling lists of the adult population in order to collect

a new and very unpopular flat-rate local tax based on a

head count (the "poll lax"). Some compaigners against

the tax (e.g. in Tottenham in London) exphcitly called

for a boycou of the census on these grounds. (It is

important to emphasise that the census authorities would
never in fact pass on census data to any outside the

census office, even to other government departments.)

Taking a comparative perspective, it does appear to be

the case that the British public is more sensitive to

confidentiality issues and less willing to trust the census

authorities than in other countries. The British public is

more concerned about privacy than other Anglophone
countries; Goyder and Leiper (1985), for example, did a

content analysis of letters to the press about the censuses

in 1980 and 1981, and found much more concern over

issues of privacy and confidentiality in Britain than in US
or in Canada.

Given the concern about confidentiality, it is worrying to

learn that disbehef in the absolute confidentiality guaran-

tees given by census authorities is widespread. This is

illustrated by a recent Gallup survey undertaken in both

GB and USA. It documents that the level of trust in the

census authorities is low in both countries, but lower in

Britain: the survey shows that the British public is much
less likely to believe the confidentiality pledges given by

the census offices than the American public, as Table 1

shows.

In short, the general public has a range of worries that

census information, gathered ostensibly for assisting

government in planning purjxjses of various kinds, will

be circulated to others for purposes which were not

declared at the lime when the information was collected.

In lights of these worries, it is not surprising that census

offices do not just hand out microdata on request. It is

also not surprising that the census offices that made the

decision to release microdata earlier on were more liberal

about what they were prepared to release than those

trying to make the same decisions more recently.

The resolution of the dilemma with respect to census

data

Faced with the dilemma between rights to information on

the one hand and rights to privacy on the other, the

census bureaus in different countries have made different

responses. In general, the English-speaking countries

have tended to give primacy to rights to information, and

have made census microdata available in various an-

onymised and sanitised forms, whereas European

Ta±)le 1 : Perception of census confidentiality in Britain and
US

Question: "How confident are you that the Census Office will not release an individual's

census information to other government agencies: are you ...

GB USA
% %

very confident 17 23

somewhat confident 27 44
or not at all confident?" 42 28
(don't know) 14 5

Fieldwork dates: March 1991 for GB: March 1990for US:
Source: Gallup Political and Economic Index, No. 368, April 1991 (GB) and Gallup Reportfor USA

Reproduced with the kind permission ofUK Gallup.

Summer 1991



countries have been much more exercised about rights to

privacy and have generally not gone down this public

road. Britain, standing as it does with a foot in either

camp, has taken a long time to decide which way to go.

The United States of America was the first country to

release such public use files in the 1960s. From the time

when the population census was first computerised in the

US, discussion was initiated about releasing forms of

microdata to the research community. Partly because the

administrative culture was open to research dissemina-

tion, and partly because of the existence of energetic

individuals pushing from within the census bureau,

"public use files", as they are termed in the US, were

released retrospectively for the 1961 census and have

become a routine part of census output. They were

followed by Canada in the 1970s and Australia in the

1980s. However, Canada and Australia never released as

much information, either in terms of sample size, detail

of file structure or fine-grain detail of coding schemes as

the US public use files (see Marsh et al. 1991a for more
details). The Australian census microdata in particular

contains only the state as a geographic identifier.

Despite similar requests for microdata, many European

countries have declined to release samples into the public

domain unfettered by limitations placed on uses or users.

Academic researchers in Denmark and Sweden may only

receive microdata for specified and delimited purposes.

Some countries allow local state authorities access to

census data but deny similar access to academics: local

government researchers at all levels in Italy can have

access to microdata, for example, as can regional govern-

ments in Spain and some government departments in

Luxembourg. In Germany, anonymised census records

may only be released to the communities^ And some
countries only release a small subset of the available

information in the form of microdata; in France, for

example, only a restricted subset of census variables is

released publicly as microdata in sampling fractions

varying from 0.1% to 25%. (Redfem 1987 briefly

outlines the rules in each European country.)

I am delighted to tell you that for the 1991 census in

Britain, agreement has been reached for samples of

anonymised records to be released. At present the plans

only include England, Wales and Scotland, but represen-

tations are also being made to the Census Office in

Northern Ireland to grant a similar request with respect to

the 1991 Northern Irish census.

Requests from British academics for census microdata go
back at least fifteen years. Spurred by interest in the

release of public use samples in North America, a

committee of interested academics was convened in the

mid-1970s to discuss the possibility of obtaining similiar

microdata in Britain for the 1971 census. One problem

which emerged early on was the geographers' demand for

microdata was for very large samples (10% or more of

census records) with very fine grain geography. During

the 1970s, it seems that the needs of geographers domi-

nated the requests, and the gulf between what academics

seemed to require and what the census offices felt able to

release while retaining confidence in the confidentiality

of the records was wide.

Furthermore, the early discussions about microdata never

progressed very far since the legality of releasing micro-

data under the terms of the 1920 Census Act was never

resolved. There were those inside OPCS who argued that

release of microdata was permissible under the terms of

the Act (Redfem 1976), but the legality of this move was

contested by others at the time.

More concerted efforts were made to obtain microdata

from the 1981 census. The White Paper outlining plans

for the 1981 census made it clear that the census authori-

ties were prepared to consider reasonable requests for

samples of anonymised records. Supfwrt for release of

microdata was also available from other sources, some of

them somewhat unexpected: from the British Computer
Society team reviewing security provisions for the 1981

census (HMG 1981a), and from these advocating cuts in

the Government Statistical Service who argued that

microdata could provide a cheap and flexible substitute

for tables (HMG 1981b). There was also interest was
shown by census office research staff (Denham 1986).

And requests from academics (e.g. Norris 1983) for

microdata persisted.

A further committee was therefore convened under the

auspices of the Environment and Planning Committee of

the Economic and Social Research Council between

1984 and 1985, to try to co-ordinate a request to be put to

the census offices to obtain retrospective samples from

the 1981 census. Despite receiving evidence from

several academics about the value of such data, the

committee never reached the stage of putting a formal

proposal to the census offices. There appear to have

been several reasons for this. First, the demands of the

geographers who wanted fine grain areal information to

the detriment of detail elsewhere and the demands of

social and policy researchers who wanted full household

information, if necessary sacrificing geographical detail

were never reconciled. Second, it proved very hard to

get agreement from others in the commercial and public

sectors to form a purchasing consortium to buy the

proposed data. Third, in 1985 it seemed likely that there

might be a major 10% household survey undertaken in

1986 which might meet needs more effectively; (in fact

this survey was never undertaken). As lime furthermore,

as time went by, the value of data relating to 1981

38 ASSIST Quarteriy



seemed to decline^ The request for data from 1981

therefore lapsed.

However, user demand did not (Marsh el at. 1988). The
Economic and Social Research Council therefore re-

newed its efforts in good time for the 1991 census, and
set up a working party to negotiate with the census

offices in Great Britain and to present a formal request.

This working parly undertook some systematic work on
quantifying the risks of disclosure from releasing census

data, and concluded that the risks were minimal. On the

basis of that work, it proposed that the census offices

release two different files of microdata, one to meet the

needs of those who wanted the maximum geographical

detail and one to service those whose prime interests

were in household structure. The request was presented

in a lengthy report in 1989 (published in Marsh et al.

1991) which was favourably received by the census

offices. The committee also seciu-ed the agreement of the

ESRC to shoulder the total costs of the purchase if

necessary.

The census offices sought advice from their solicitors

about whether microdata could legally be released under

the terms of the 1920 Census Act They were advised

that anonymised microdata came under the general

heading of a 'statistical abstract', and could be released

without changing the law. The Office of the Data

Protection Registrar and Liberty (the National Council

for Civil Liberties as was) were consulted, and neither

had any major objections. The proposal to release

microdata was therefore mentioned in the White Paper

outlining plans for the 1991 census (Cm 430, 1988),

subject to the overriding need to preserve census confi-

dentiahty. It was also commented upon by the British

Computer Society team who reviewed security arrange-

ments for the 1991 census (Her Majesty's Government
1991); since they had supported the idea for the 1981

census, they gave the idea their blessing. In July 1990,

the agreement in principle of the Registrars General to

the ESRC request was announced in a written Parliamen-

tary answer.

Background to the British Census
In the United Kingdom, censuses of population are the

responsibility of three separate offices: the Office of

Population Censuses and Surveys under the supervision

of the Registrar General for England and Wales, the

General Register Office for Scotland under the supervi-

sion of the Registrar General for Scotland, and the

Census Office in the Northern Irish Department of Health

and Social Services under the supervision of the Regis-

trar General for Northern Ireland. The Office of Popula-

tion Censuses and Surveys (OPCS) plays a coordinating

role in the work of the three offices.

The legal firamework for censuses in Great Britain is the

1920 Census Act as amended by the Census (Confidenti-

ality) Act 1991. The 1920 Act makes provision for

population censuses to be taken at no greater frequency

than every five years. It is enabling legislation which

requires there to be a new census order outlining arrange-

ments and plans for content each time. The Registrar-

General is given the responsibility for organising the

census, the power to take on the necessary staff, and the

right to present the final accounts to Parliament

Under the terms of the 1920 Census Act, filling a census

return is compulsory for householders. Many in census

offices feel that the compulsory natiu'e of the census puts

it in a different moral category when it comes to the

release of microdata. The logic of this position is not

entirely clear, however, since similar confidentiality

guarantees are given to those who take part in voluntary

government surveys*.

The Registrar General is given the duty to ensure that

summary reports of census data re prepared.Furthemiore,

"The Registrar-General may, if he so thinks fit, at the

request and cost of any local authority or person,

cause abstracts to be prepared containing any such

statistical information, being information which is not

contained in the reports made by him under this

section and which, in his opinion it is reasonable for

that authority or person to require, as can be derived

from the census returns." [Section 4(2)].

The interpretation of this section of the Act was critical

for the release of microdata; it turned on whether micro-

data could be deemed a "statistical abstract".

Both by international standards and by comparison with

previous British censuses, the censuses of 1981 and 1991

were fairly slim. In 1991 there were 8 questions about

housing and 19 questions about individuals. There has

never been an income question nor, since 1851, any

question on religion in Great Britain, although there is

such a (voluntary) question in Northern Ireland. Enu-

meration is done on the basis of presence on census

night; information is also obtained about the usual place

of residence of visitors and about absent usual residents

(with a voluntary return if the whole household is

absent).

There are no long and short forms on the British census.

Sampling is only undertaken at the processing stage. The

answers to those questions which are laborious to code

(such as occupation, industry and qualifications) are only

fully coded for a 10% sample.

When it comes to census output, the principle of dissemi-

Summer 1991 39



nation is best expressed in the motto of the London
Statistical Society, first expressed over 150 years ago:

aliis exterendum^. At the time it was first enunciated, it

was an unattainable ideal, as the technology for taking

censuses and surveys had not progressed very far, and the

information was usually published in the form of verbal

commentaries, often of a very opinionated form (Cullen

1975). However, the trend in census output has continu-

ously been towards making more and more of the

detailed information available, aiming for the user to be
free to rework it and interpret it in any way that he or she

chooses. First this was achieved by providing more
tables. There will be 20 volumes of special topic statis-

tics from the 1991 census and monitors for each county

and parliamentary constituency. More recently, in-

creased volume of information have been supplied in

machine-readable form; the census offices estimate that

98% of census output nowadays is in electronic form. In

particular, the small area statistics are particularly

detailed; information is provided down to aggregates of
around 200 households in England and Wales (70
households in Scotland) and around 9,000 pieces of

information are available at this level. There are also

special workplace and migration statistics released in

machine-readable format for small areas.

Thus, up to 1991, the output from the British census in

forms other than microdata was exceptionally detailed,

especially in the amount of machine-readable data made
available for small areas. This may also be part of the

explanation why the demand for microdata in the past

never became overwhelming.

The information to be released

At the time of writing, agreement in principle has been
given to a plan to release two different samples of data

from the 1991 census. Their broad structure has been
agreed, but negotiations are still continuing about the

precise details. Any of the details mentioned here could

therefore change before release of the data.

The first file proposed is a two percent sample of indi-

viduals with full housing information and some limited

information about household structure attached. The
census offices have been guarded about releasing too

much information about other household members for

fear of effectively releasing what amounts to a hierarchi-

cal file under a different guise. This file will show a

geographical scheme identifying large local authority

districts or groupings of smaller ones.

The second file proposed is a one percent hierarchical

file of households, containing housing information and
full information about all the individual household
members. For confidentiality reasons this file will

classify data only to the 10 Standard Regions of Great

Britain.

Both files are to be drawn from the 10% of records which
are fully coded. The sample will be drawn systematically

from this 10% file which is ordered by county, then by
enumeration district and street. The two samples will be

drawn without replacement so that there will be no
overlapping subsamples.

Measures to protect conndentiality of the information

Great effort is put into ensuring that the data is safe from

identification and disclosure. Five different devices are

being used.

(i) Sampling of records

Sampling itself is one of the most effective ways of

reducing the risk of disclosure, provided that users

cannot identify which individuals are selected. The
sampling fractions are small (.01 and .02), and, because

the geographical identifiers are different on the two files,

it will not be possible to combine them.

(II) Supresslon of variables

Names and addresses are not entered onto the census

computer, and therefore obviously not even available for

suppression. Precise data of birth is to be suppressed;

age will only be available in yearly bands, and will be

top-coded (see below). There will also be no information

on the imputed missing household', since these are

excluded from the 10% coding operation.

(III) Limiting geographical detail

There were competing claims about the basis of SAR
geography. The main choices were:

local government administrative geography

• health service administrative geography

political boundaries

local labour market areas

postcode sector defined boundaries

grid square geography

It was not possible to have more than one system, as the

small overlaps between many of these different schemes

would have led to unacccptably small areas being

identified. On general utilitarian grounds of maximum
benefit to maximum numbers, local government adminis-

trative geography was chosen.

The geographical scheme in the individual level file will

identify only those local districts with populations of

40 lASSIST Quarterly



120,000 or more; the rest will be grouped into areas of at

least 120,000 population. However, all bar one of the

metropolitan local authority districts will be identified.

On the hierarchical file, the geography will be even

coarser. Only standard regions will be identified, with

London subdivided into inner and outer areas; this

amounts to twelve areas in all, the smallest of which East

Anglia with an estimated population of 1.9 million.

Other geographical information is also collected on the

census - the usual address of visitors to the household,

workplace address, students' term-time addresses and
address one year ago. These are to be very heavily

restricted; at the time of writing, the proposal is to

identify standard region, a same district/different district

identifier and perhaps a distance measure as well.

The order of records in the microdata file are to be

scrambled to ensure that locality information cannot be

obtained from the proximity of one records to another.

(iv) Grouping of categories

A rule is being used to restrict the fineness of the coding

scheme for each variable:

Each category of each variable to be identified on
the SAR must have an expected sample count of at

least 1 in the smallest geographical area permitted

on either file.

Expectations are formed by scaling down the distribution

of that variable for Great Britain as a whole. To illus-

trate, in the individual file, the smallest geographical area

identified will have a population of 120,000. The
sampling fraction will be 1/50. Thus the cut-off on this

file is 50/120,000 times the population of Great Britain

of 56 miUion, yielding 23,3000; this is then rounded up

to 25,000. Accordingly, any category of any variable

which in Great Britain is expected to have less than

25,000 people at the census will be grouped in with

another category in the SAR. In the household file, the

smallest region identified is East Anglia, with a popula-

tion of around 1.9 million people, and the sampling

fraction will be 1/100, so the cut-off being used for this

file is 2,700 people.

Thus only the univariate distributions are used to identify

categories at risk. Some have been worried that it is

unusual combinations of categories that cause problems:

female banisters, small householders in accommodation
with a lot of rooms, and so on. However, some recent

work suggests that the strategy of worrying only about

those categories that are small in the univariate distribu-

tion will predict people who have unique combinations

of variables extremely efficiently (Marsh et al. 1991b).

The variables most affected by the grouping rule are

occupation (where the 350 categories in the full coding

scheme will probably be reduced to around 250), indus-

try (270 categories reduced to around 2(X)), detailed

educational qualifications (103 reduced to 53) and

country of birth (reduced to 47 groups). The variables to

be top-coded are age (to be grouped into two year bands

between 91 and 94 and top-coded 95 and above), hours

worked last week (top-coded after 70), and number of

rooms in the accommodation (to be top-coded after 14).

Top-coding does not solve the problems of households

with very large numbers of individuals; the number of

people in a household affects the very structure of a

hierarchical file rather than the categories of one variable

in it. Several suggestions are currently being explored to

solve this problem, including removing all geographic

identifiers from households with more than 12 members.

(v) Perbwbation

The small area statistics from the British census have

always been subject to a degree of random perturbation

('Bamardisation') whereby a random -i-l , or -1 is added

all cells. The comprehensive application of this tech-

nique has always been unpopular with users (Marsh et al.

1988) and the cumulative effects of the small errors

introduced can pose quite serious analytic problems

(Senior and Cole 1991). Wholesale Bamardisation is of

course not possible with microdata, and the prospect of

deliberately adding more noise to many different vari-

ables on top of the natural levels of error already existing

in the data was resisted by ESRC team negotiating the

release of the data. Instead, the census offices have

suggested that a small number of individuals in each area

be switched with others in a nearby area (Griffin et al.

1989), a technique which amounts to adding a small

degree of random noise to the geographical identifier, but

preserving the rest of the household data intact.

In the USA and Canada, the census offices appoint some
sort of panel to oversee the arrangements for protecting

confidentiality (Gales 1988). Such a Microdata Review

Panel was considered for GB, but, on the recommenda-

tion of the Royal Statistical Society, the census offices

have appointed one technical adviser instead, to oversee

the specification of the SAR files and the general ar-

rangements made for confidentiahty. This advisor, a

senior academic statistician, will make independent

recommendations to the Government Minister respon-

sible for the census offices.

Contractual arrangements

The funders of the project are the Computer Board',

<footnote text> who are paying for the data, and the

Economic and Social Research Council, who are putting

up the money for a research and distribution centre to be

Summer 1991 41



established to house the data. data.

The contract to buy the data from the census offices is

regulated in part by legislation; under the terms of the

1920 Census Act, the census offices are obliged to

recoup the cost of producing extra tables (which includes

the SARs), but are prohibited from making profits on the

data which they supply. The ESRC team negotiating the

release of SARs attempted to find co-purchasers to enter

into a consortium to share the costs, but none came
forward. Thus now, the academic purchasers are bearing

the full developmental costs of the SARs. The census

offices have done their costings on the basis of passing

on the full marginal costs of producing SARs; at present

the cost of the data seems likely to be around £200,000,

but this sum will only be finalised when the file specifi-

cations have finally been approved.

The contract to must safeguard the interests of the

academic purchasers in their product. Although final

contractual details are still being sorted out, we hope that,

in return for our payment, we will get full exploitation

rights of the dataset, which will be sole rights for a

limited period. Since the academic purchasers of the data

are bearing the full cost of the production of the files, the

census offices will not receive any further royalties when
value-added products of the census are passed on to third

parties.

The contract to purchase from the census offices will

specify that all end-users of this data must give various

undertakings about respecting census confidentiality.

People will be expressly forbidden to try to identify

individuals in the SAR, link them with other sources of

data or to claim to have done so. Any breach of this

agreement would lead to withdrawal of the SARs. While
not undoing the harm caused by a breach, but its threat

would probably act as a deterrent since academics would
subsequently be unable to publish any information based
on the SARs. Various methods are under discussion to

register users and keep tract of copies. It is possible that

heads of departments will be required to be the data

holders rather than individual researchers.

The contractual arrangements between the purchasers of
the data and the end users with respect to have yet to be
discussed in detail. Academic users wanting the data for

research purposes will have free access to the data, but

means will be sought to regulate the other ways in which
the data may be used, to safeguard the interest of the

public fundcrs. A graduated scale of charges seem
likely, charging commercial users what the market will

bear, perhaps with lesser charges to the rest of the public

sector and the voluntary sector. University and college

faculty who contract their services as consultants outside

the academic sector will be charged for their use of the

Disseminating the data to users

The Economic and Social Research Council is funding a

centre at a university location to house, disseminate and
act as a research focus for this dataset Invitations to

house this centre were put out to several institutions, and

three teams submitted tenders. The decision about which
will be successful is expected in late May 1991.

Four types of usage are envisaged:

(i) On-line access over the academic network

The Joint Academic Network (Janet) connects all

universities and many polytechnics and other institutions

in Britain. It is funded by the Computer Board, and free

at the point of use to all universities. One important

method of giving access to the data will be to mount
versions of the data at cenffal locations on the network

for easy access by any academic users.

The decision about which software to use for the data has

not yet been made. SPSSX seems the most popular

candidate for the individual file. The hierarchical file of

households could also be set up in SPSS, but this might

prove cumbersome, especially since there are two

different hierarchies within the dataset: individuals can

be grouped into either households of families. SAS, SIR
and Oracle are other candidates which might be consid-

ered. It is likely that eventually the same information

may be held in different formats at different locations;

while some may view this as inefficient, from the user's

point of view there are great gains in terms of familiarity

and ease of use.

(ii) Tables service

The census is a benchmark source of social data. It

provides denominators for many different researchers'

numerators. However, it is not the prime data source for

more than a few researchers. It is therefore important for

there to be a service to academics which would provide a

service of this kind to other academics, although the

demand for this may lead to the need for rationing.

(Hi) Customised subsets

Another central task that the service centre will undertake

is to extract subsets of variables and cases for different

users. This will need to have regard to the media most

likely to be demanded; a PC/workstation platform is the

most likely here, although demand for data on CD Rom
and other media will need to be monitored carefully.

There will doubtless be demand for teaching datasets for

use in schools and colleges.

(iv) Passing on whole dataset

For efficiency reasons, other academic users would be

lASSIST Quarterly



encouraged to use the service provided over the academic

network. But the whole philosophy of providing samples

of anonymised records is that users will be free to port

them into their own hardware and software environment

and not be restricted by earlier decisions. No restrictions

will therefore be placed on others wanting entire copies

of the dataset.

Public and commercial users do not in general have

access to the academic network. They also tend to be

familiar with somewhat different software to that used in

universities; in Britain, for example, the most popular

tabulation software used by market researchers is a

product marketed by Quantime called Quanvert. One of

the things that the academic purchasers of the data may
want to explore is licensing a commercial agency to

provide an on-line service for the commercial sector.

The centre housing the microdata will have responsibility

for documenting the datasets, and for computing the

sampling errors and documenting these. It will also need

to establish a user group, and disseminate information

about the database to users, both in electronic media and

by hard copy newsletters.

Conclusion

The existence of microdata from the 1991 Census opens

up a valuable new resource to British social researchers.

Our social research community has grown used to

making do with small area aggregates, and drawing

inferences from these; the existence of microdata should

lead to some interesting work on ecological fallacies.

Survey and market researchers will be in a position to

design much more efficient samples to locate specific

subgroups of the population. We will have a source of

microdata on a badly neglected part of the population,

namely those who do not live in private households.

Research into different means of classifying families and

households should blossom given the richness of the

hierarchical information available.

Denham, CJ. (1986) 'Census microdata in Great Britain:

the possibilities', Nutzung von anonymisierten Einzekan-

gaben aus Daten der amtlichen Stalistik: Bedingungen

und Moglichkeiten, Verlag W. Kohlhammer.

Gallup (1991) Gallup Political & Economic Index,

Report No. 368, April.

Gates, G.W. (1988) Census bureau microdata: providing

useful research data while protecting the anonymity of

respondents, Proceedings of the Social Statistics Section

of the American Statistical Association Annual Meeting,

New Orleans, Louisiana, August 22-25.

Goyder, J., and Leiper, J. McK. (1985) 'The decline in

survey response: a social values interpretation'. Sociol-

ogy, vol. 19, 1, pp. 55-71.

Griffin, R.A., Navarro, A., and Florez-Baez, L. (1989)

'Disclosure avoidance for the 19(X) census', U.S. Bureau

of the Census, paper prepared for presentation at the

1989 Joint Statistical Meetings, Washington, D.C.,

August

Her Majesty's Government (1981a) 1981 Census of

Population: Confidentiality and Computing, presented to

parliament by the Secretary of State for Health and the

Secretary of State for Scotland, Cmmd 8201, London:

HMSO.

Her Majesty's Government (1981b) 'Initial study of the

Office of Population Censuses and Surveys', Annex to

the Review of the government statistical services, Cmmd
8236, London: HMSO.

Her Majesty's Government (1991) 1991 Census of

Population: Confidentiality and Computing, presented to

parliament by the Secretary of State for Health and the

Secretary of State for Scotland, February, Cmmd 1447,

London: HMSO.

Perhaps also we may hope that effort will be made to

construct internationally comparable files of census

microdata relating to the 1990 and 1991 censuses in

different countries. One spin-off of presenting this

information to an international audience at this lASSIST
meeting might be to stimulate discussion in this direc-

tion.

References

Campbell, D. (1987) 'The databank dossier'. New
Statesman, April 24.

CuUen, M. (1975) The Early Victorian Statistical

Societies, Brighton: Wheatsheaf.

Marsh, C, Arber, S., Wrigley, N., Rhind, D., and

Bulmer, M. (1988) "The view of academic social scien-

tists on the 1991 UK Census of Population: a report of

the Economic and Social Research Council Working

Group", Environment and Planning A, vol. 20:851-889.

Marsh, C, Skinner, C, Arber, S., Penhale, B., Open-

shaw, S., Hobcraft, J., Lievesley, D., and Walford, N.

(1991a) 'The case for samples of anonymised records

from the 1991 census'. Journal of the Royal Statistical

Society (A), vol. 154(2), 1991.

Marsh, C, Dale, A., and Skinner, C. (1991b) "Safe data

or safe settings: disseminating information from the

British Census of Population", paper to be presented to

Summer 1991 43



the ISI meetings in Cairo, September.

Norris, P. (1983) 'Microdata from the British Census', in

David Rhind ed. A Census User's Handbook, London

and New York: Methuen.

Redfem, P. (1976) Releasing statistics as aggregates

(tables) or tapes of anonymised individual data (id).

OPCS, London: mimeo, available from the author at 17

Fulwith Close, Harrogate, HG2 8HP.

Senior, M., and Cole, K. (1991) '1981 Census costs

Department of Health £190,(XX)!', Manchester Comput-

ing Centre Newsletter, no. 183, p. 2.

' Presented at the 1ASSIST 91 Conference held in Ed-

monton, Alberta, Canada. May 14 - 17, 1991.

^The Federal Republic of Germany has a policy against

the general release of microdata collected by the state. It

will not allow its Labour Force Survey to be released for

secondary analysis, for example. Since the Statistical

Office of the European Commission takes the most

restrictive poUcies of its member states as guidelines on

release of microdata, this means that Europe-wide release

of the Labour Force Survey is precluded.

' In fact, the value of census data to academic researchers

does not seem to decline as the census data becomes less

timely; if usage of small area statistics is a guide, the

usage of these increased throughout the 1980s.

* Perhaps this is the reason why two other data sources

collected compulsorily by law (the New Earnings Survey

and Census of Employment) are only released in aggre-

gates, albeit of very small units.

' (Trans.) For others to thresh, or, more colloquially, let

someone else work out what it all means! It is still the

motto of the Royal Statistical Society.

' Households with all members absent on census night

and which fail to return a voluntary return to the census

offices once they get back home.

' The body which funds purchases and support of main-

frame computing in British universities. It is soon to

become a subcommittee of the main Universities Fund-

ing Council.

ASSIST Quarterly